Perineuronal Net Formation during the Critical Period for Neuronal Maturation in the Hypothalamic Arcuate Nucleus

Summary In leptin-deficient ob/ob mice, obesity and diabetes are associated with abnormal development of neurocircuits in the hypothalamic arcuate nucleus (ARC)1, a critical brain area for energy and glucose homeostasis2,3. As this developmental defect can be remedied by systemic leptin administration, but only if given before postnatal day 28, a critical period (CP) for leptin-dependent development of ARC neurocircuits has been proposed4. In other brain areas, CP closure coincides with the appearance of perineuronal nets (PNNs), extracellular matrix specializations that restrict the plasticity of neurons that they enmesh5. Here we report that in humans as well as rodents, subsets of neurons in the mediobasal aspect of the ARC are enmeshed by PNN-like structures. In mice, these neurons are densely-packed into a continuous ring that encircles the junction of the ARC and median eminence, which facilitates exposure of ARC neurons to the circulation. Most of the enmeshed neurons are both GABAergic and leptin receptor-positive, including a majority of Agrp neurons. Postnatal formation of the PNN-like structures coincides precisely with closure of the CP for Agrp neuron maturation and is dependent on input from circulating leptin, as postnatal ob/ob mice have reduced ARC PNN-like material that is restored by leptin administration during the CP. We conclude that neurons crucial to metabolic homeostasis are enmeshed by PNN-like structures and organized into a densely packed cluster situated circumferentially at the ARC-ME junction, where metabolically-relevant humoral signals are sensed.

circulation. Most of the enmeshed neurons are both GABAergic and leptin receptor-positive, including a majority of Agrp neurons. Postnatal formation of the PNN-like structures coincides precisely with closure of the CP for Agrp neuron maturation and is dependent on input from circulating leptin, as postnatal ob/ob mice have reduced ARC PNN-like material that is restored by leptin administration during the CP. We conclude that neurons crucial to metabolic homeostasis are enmeshed by PNN-like structures and organized into a densely packed cluster situated circumferentially at the ARC-ME junction, where metabolically-relevant humoral signals are sensed.
During critical periods (CPs) of early postnatal life, developing neurocircuits are exquisitely sensitive to and shaped by external cues from the environment 6 . The CP for ocular dominance plasticity in primary visual cortex (V1) is a well-studied example: during the CP, but not before or after, visual deprivation of one eye induces a strong shift of neuronal responses to the non-deprived eye 7 . The associated loss of acuity in the deprived eye, clinically referred to as amblyopia, is difficult to remedy after CP closure.
PNNs are a specialized, condensed form of extracellular matrix (ECM), composed largely of hyaluronan and chondroitin sulfate proteoglycans, arranged in lattice-like structures that enmesh the soma and proximal dendrites of subsets of neurons, primarily inhibitory interneurons 9,15 . Remarkably, experimentally disrupting PNNs in adult animals is capable of reactivating CP plasticity in these diverse brain areas. In V1, for example, PNN digestion enables restoration of vision to a previously deprived eye 5 . Comparable restorative effects of PNN disruption are reported across neurological and psychiatric disorders as diverse as PTSD, depression, drug addiction and spinal cord injury 9,15 .
Recently, compelling evidence was provided in support of the existence of a CP for the maturation of Agouti-related peptide (Agrp) neurons in the hypothalamic arcuate nucleus (ARC) 4 . As key regulators of feeding behavior, Agrp neurons are among the best-studied of all hypothalamic neurons. These GABAergic neurons co-express the potent orexigen neuropeptide Y (NPY), and they project to downstream targets in homeostatic circuits governing energy balance and glucose homeostasis. Maturation of these projections occurs during the lactation period, concomitant with a naturally-occurring leptin surge (P4-P14) 1,16 , and this maturation process appears to be dependent on leptin since these projections fail to develop properly in leptin-deficient ob/ob mice (a genetic model of obesity and type 2 diabetes (T2D)). Moreover, this defect in Agrp neuron development can be rescued by treatment with exogenous leptin, but only if administered before P28 1,4 . A CP for the trophic action of leptin on Agrp neuron maturation therefore exists, and its closure coincides with both the transition to independent feeding and maturation of the cellular response of Agrp neurons to input from leptin 17 . This CP is also a uniquely sensitive time when both over-and undernutrition predispose to obesity and glucose intolerance in adulthood [18][19][20][21][22] .
Here we show that key subpopulations of ARC neurons are enmeshed by PNN-like material in humans as well as rodents. These neurons are predominantly GABAergic and many are leptin-receptor positive, including a majority of Agrp neurons. These enmeshed neurons are organized into a densely packed ring that delimits the junction of the ARC and median eminence (ME), a circumventricular organ (CVO) that provides hypothalamic neurons with access to hormones and nutrients in the circulation. Moreover, the postnatal appearance of these PNNs both coincides with closure of the CP for Agrp neuron development and is dependent upon input from leptin.
Since PNN-enmeshed neurons in other brain areas are not tightly packed, nor are they concentrated near CVOs or regulated by circulating signals, the anatomical arrangement we report here appears to be unique. Nevertheless, based on our findings that the condensed ECM material that enmeshes neurons at the ARC-ME junction is both biochemically and ultrastructurally indistinguishable from PNNs in other brain areas, we refer to them as "PNNs" (rather than "PNN-like structures").
We performed immunohistochemistry (IHC) on serial coronal sections of mouse hypothalamus using Wisteria floribunda agglutinin, a lectin that selectively labels the Nacetylgalactosamine residue on chondroitin sulfate (CS) chains in PNNs. Our initial survey (Supplementary Figure 1) revealed numerous PNN-enmeshed cells localized to the junction of the ARC and median eminence (ME) (Figure 1a-d). Compared to the rather sparse distribution of PNN-enmeshed neurons in most brain areas (e.g., V1), enmeshed cells in the ARC-ME area are densely packed (Figure 1e, g). To better characterize the structural features of these cells, we used high-resolution confocal microscopy combined with Imaris image analysis of individually-labelled, PNN-enmeshed neurons located at the periphery of the dense cluster at the ARC-ME junction. Analysis of these cells (arrow f in Figure 1c) allowed us to better define the anatomical relationship between PNNs and the neurons they enmesh without the confounding influence of labeling on an adjacent, closely apposed cell. As expected, PNNs enwrap both cell soma and proximal processes (Figure 1f), reminiscent of PNN structures described in the visual cortex (Figure 1g,h). Wholemount preparations of the mediobasal hypothalamus, stained with WFA and imaged from either the ventricular surface ( Figure 1i) or the ventral pial brain surface (Figure 1j), revealed the presence of a continuous "collar" of PNN-enmeshed cells at the junction of the ARC and ME.
As a first step to determine if these ARC WFA-labeled structures are bone fide PNNs, we micro-injected 10 milliunits of Chondroitinase ABC (ChABC), an enzyme that digests PNNs, stereotactically into the ARC of wild-type mice. Subsequent histochemical analysis of these animals (n=3) revealed ChABC-mediated digestion of ARC-ME PNNs on the injected side ( Figure 1k). Then, following a previously validated protocol 23 , we performed pre-embedding WFA-DAB labeling of mouse brain sections for study by electron microscopy. This ultrastructural analysis revealed DAB electron-dense deposits surrounding the soma (Figure 1m white arrowheads) and neurites (Figure 1m white arrows) of ARC neurons in a distribution that closely matches the pattern observed in PNN-enmeshed cells in the cortex and hippocampus (Supplementary Figure 2).
In addition to chondroitin sulfate (CS) carbohydrate chains labeled by WFA, PNNs are comprised of two other major components: the CS proteoglycan (CSPG) core proteins to which the CS chains covalently bind, and hyaluronic acid (HA), a long carbohydrate polymer to which CSPG core proteins noncovalently bind. Using biotinylated-HA-binding protein (HABP) 24 , we histochemically stained HA in the hypothalamus. Although HABP lightly stains the extracellular matrix throughout brain parenchyma, we observed an increased abundance of HA that colocalizes with WFA at the junction of the ARC and ME (Figure 1o and Supplementary Figure 3), confirming that WFA (+) structures in this brain area contain HA. To identify the relevant CSPG(s) present in these PNNs, we next stained for each of the 5 major CSPG species found in PNNs elsewhere in the brain: aggrecan, brevican, neurocan, versican and phosphacan. Interestingly, although each of these CSPGs was detected in various mouse brain areas (data not shown), only phosphacan (PCAN) immunoreactivity was co-localized with WFA in mouse ARC-ME (Figure 1p and Supplementary Figure 3).
We also immunostained for CD44, the cell surface receptor for HA, and found that its expression in this brain area is largely limited to tanycytes, with high expression localized to E3/β-tanycytes piercing through the PNN domain and little to no expression in dorsal E2/αtanycytes that circumvented the PNN domain ( Supplementary Figure 4b, c). An IHC survey of CD44 expression in serial coronal sections through the hypothalamus revealed a striking colocalization with the PNN domain in the ventromedial ARC (Supplementary Figure 4a). These findings raise the possibility that via activation of CD44, signal transduction in tanycytes is evoked by HA present in PNNs.
The ARC is uniquely specialized to sense and transduce input from nutritionally-relevant hormones (such as leptin and insulin) and nutrients (e.g., glucose, free fatty acids and amino acids) into adaptive changes of food intake and energy metabolism 2 . Among key neuronal subsets implicated in this homeostatic circuitry are Agrp and pro-opiomelanocortin (Pomc) neurons, with the former being GABAergic 25 . To identify distinct ARC neuronal subtypes that are among those enmeshed by PNNs (= # neurons of particular subtype / total # PNNenmeshed cells), we performed IHC on coronal sections from transgenic mouse lines that provided whole cell GFP-filling. Using GAD67-GFP heterozygous knock-in mice, we confirmed that a vast majority of ARC neurons enmeshed by PNNs are GABAergic (81.8 ± 0.7% of ARC PNN-enmeshed cells; n=3 GAD67-GFP mice; note that dots in dot plots throughout this work represent data from independent animals) ( A separate but related question pertains to the fraction of Agrp or POMC neurons in the ARC that are enmeshed by PNNs (= # PNN-enmeshed neurons of particular subtype / total # neurons of that subtype). As predicted, the fraction of Agrp neurons enmeshed by PNNs (78.3 ± 1.7% NPY-GFP neurons) is greater than the fraction of POMC neurons (43.6 ± 3.3% POMC-GFP neurons). A third peptidergic cell type found in the ARC expresses somatostatin (SST), which is challenging to detect histochemically because the SST peptide is rapidly exported from soma into axons. To address this issue, we administered a single intracerebroventricular (ICV) injection of colchicine to wild-type mice, which prevents transport of SST out of the soma and thus enables somatic labeling with SST antibodies. We report that while this approach allowed us to identify ample SST+ soma, few of these were PNN-enmeshed (1.2 ± 0.6%; n=3) (Figure 2n-o). As confirmation of the efficacy of ICV colchicine to prevent neuropeptide export from the soma, we note that many Agrp+ PNNenmeshed cell bodies were also observed, corroborating our findings in NPY-GFP mice ( Figure 2o). Together, these findings indicate that only a subset of ARC neuronsspecifically, those neurons most closely linked to control of energy balance and metabolic homeostasis -are enmeshed by PNNs.
A key question raised by these observations is whether PNNs contribute to closure of the CP for ARC neurocircuit development, as is true of V1 and other brain areas. As a first step to address this question, we performed a developmental time-series analysis of PNN formation over the lactation and periweaning period from postnatal day 10 (P10) to P30 in wildtype mice ( Figure 3). This approach was based on evidence that leptin-dependent maturation of ARC-ME neurocircuits occurs during a CP that closes ~P28 1,4 , so PNNs must appear during this period if they are to contribute to closure of this CP. WFA labeling revealed only a very faint signal at P10 (10.0 ± 0.8 intensity units; n=4 C57B/6 mice) that lacked the typical PNN honeycomb configuration (Figure 3a). By P21, WFA labeling intensity had increased by more than 2-fold (27.4 ± 1.7 intensity units; n=5 mice) and structural features of PNNs were evident (Figure 3b). WFA intensity was further increased at P30 (46.2 ± 2.6 intensity units; n=3 mice), by which time PNN structures appeared fully formed (Figure 3c).
The time course of ARC PNN formation during postnatal development in wildtype mice was closely paralleled by the maturation of Agrp neuron projections, quantified as an increase in Agrp fiber density in the ARC (0.23 ± 0.07% at P10; 0.87 ± 0.11% at P21; 1.48 ± 0.14% at P30) (Figure 3d). We note that at CP closure (~P28), both PNNs and Agrp fiber density in the ARC were transiently increased over values characteristic of adult mice (P90: WFA 34.1 ± 2.5 intensity units; Agrp fiber density 1.05 ± 0.09%). During postnatal development, the domain of CD44 expressing tanycytes also expanded from medial (where it covered only the ME β-tanycytes at P21) to lateral and dorsal along the ventricular lining, paralleling the progressive appearance of PNNs (Supplementary Figure 4d,e; compare yellow and white arrows and arrowheads).
Based on these results, we hypothesized that like Agrp neuron maturation, formation of PNNs in the ARC is dependent on input from circulating leptin. To test this hypothesis, we administered either leptin or vehicle to ob/ob pups according to a schedule reported to mimic the postnatal leptin surge that was used to define the CP for Agrp neuron maturation 4 . Specifically, ob/ob pups received a daily intraperitoneal injection of either leptin (10 mg/kg) or vehicle from P10 to P30, and were euthanized one day later for IHC analysis. As predicted, ARC WFA intensity was significantly increased in leptin-compared to vehicletreated ob/ob pups (P30 rescue: 1.15 ± 0.02, n=3 ob/ob-leptin vs. 1.00 ± 0.01, n=2 ob/obvehicle, p=0.02) (Figure 4e, f, h). There was no significant difference in weight gain between the leptin and vehicle-treated groups. Interestingly, the presence of Agrp+ soma in the ARC of vehicle-treated pups, but not leptin-treated pups (insets in Figure 4e, f) appears to offer confirmatory evidence of leptin action, which is known to reduce Agrp expression 26 . A similar pattern was also observed when Agrp staining between ob/+ and ob/ob pups was compared at P30 (insets in Figure 4c, d).
To determine how ARC PNNs are affected by persistent leptin deficiency in adulthood, we performed WFA labeling of 12 week-old ob/ob mice and age-matched C57B/6 mice followed by high-resolution confocal microscopy and Imaris 3-dimensional image analysis (Supplementary Figure 6a, b). Paradoxically, in the ARC-ME area, WFA intensity in ob/ob mice (normalized to the mean value of age-matched controls) was significantly higher than in controls (1.  Supplementary Figure 7). To investigate whether this increase of ARC PNN intensity in ob/ob mice was the result of leptin deficiency or was instead secondary to obesity, we compared WFA staining intensity between cohorts of wild-type C57B/6 mice that were either made obese through high-fat diet (HFD) feeding for 12 weeks (beginning at age 12 weeks) or were fed standard chow and sacrificed at the same age. We report that whereas no difference in ARC WFA intensity was detected between HFD and chow-fed cohorts (1.00±0.00, n=5 hfd vs. 1.00±0.00, n=3 chow), tanycytic CD44 expression was increased in the former group (1.11±0.03, n=5 hfd vs. 1.00±0.01, n=5 chow; p=0.007) ( Supplementary Figure 6c-e). In contrast, no difference in tanycytic CD44 expression was observed between ob/ob and chow-fed control mice (1.01±0.03, n=5 ob/ob vs. 1.00±0.03, n=5 wt; p=0.8). Together, these data suggest that increased PNN intensity in the ARC of adult ob/ob mice is secondary to leptin deficiency rather than to obesity per se, whereas the reverse is true for the increase of tanycytic CD44 expression in HFD-fed (but not ob/ob) mice.
To determine if PNNs are present in the ARC of other mammalian species, we performed WFA staining in the hypothalamus of both rats and humans. We report that PNNs are present in the ARC-ME of all 3 species (Supplementary Figures 8 and 9). As in the mouse, a majority of cells enmeshed by PNNs in the human ARC are NPY/Agrp neurons, with both the soma and proximal dendrites of these neurons wrapped by WFA+ material (Supplementary Figure 8 and Supplementary Movie 4). Interestingly, PNN-enmeshed cells in humans are more sparsely distributed than those in mice, which enhances the ability to see how the mesh structure associates with the neuronal contours.
In summary, we report that PNNs enmesh Agrp and other leptin receptor-expressing neurons in the ARC (most of which are GABAergic, similar to neurons enmeshed by PNNs in visual cortex and other brain areas), and that their appearance during postnatal development coincides closely with both maturation of Agrp neuron projections and closure of the CP for leptin-mediated regulation of Agrp neuron development 4 . Situated at the junction of the ARC and ME, these PNNs are detected in humans as well as rodents, and their formation appears to be sensitive to input from leptin, being deficient in the ARC of postnatal ob/ob mice, and restored by leptin administration to ob/ob mice during the CP. Unlike PNNenmeshed neurons in other brain areas that tend to be sparsely distributed, these neurons form a densely packed ring that circumscribes the ME-a CVO that facilitates processing of metabolically-relevant humoral input-at its junction with the ARC. Together, these findings describe an apparently unique brain structure that is implicated in closure of the CP for the development of neurocircuits crucial to control of energy balance and glucose metabolism in adulthood. Our findings also suggest that leptin regulation of ARC neuron development during the CP involves an action on PNN formation.
In sharp contrast to the deficiency of PNNs we observed in the ARC during postnatal development, adult ob/ob mice exhibit an overabundance of ARC PNNs. To explain this paradoxical finding, we draw upon both in vivo 8,27,28 and in vitro 29 evidence that PNN formation is driven by activity of the enmeshed neuron. Thus, leptin induces depolarization and increases the excitability of Agrp neurons during the CP (prior to ~p21-23) 17 , so leptin action on these neurons may constitute a stimulus to PNN formation during this time. During subsequent development, however, a progressive increase in the expression of ATPsensitive potassium channels by Agrp neurons leads to a "phenotype switch", whereby leptin now exerts the hyperpolarizing effect characteristically observed in adult Agrp neurons 17 . Consequently, Agrp neuron activity is predicted to be reduced in leptin-deficient mice during the CP, but increased in adulthood, and PNN formation parallels these changes.
Available evidence suggests both that ARC neurocircuits are highly plastic during development and that this plasticity is markedly reduced upon CP closure. Thus, whereas Agrp neuron ablation induces life-threatening starvation when it occurs in adult mice, it has little or no phenotypic impact when induced shortly after birth 30 . Similarly, CRISPRmediated deletion of leptin receptors selectively from Agrp neurons in adults recapitulates most of the phenotype of whole body leptin deficiency (hyperphagia, obesity and diabetes), whereas the same deletion has little detectable effect when induced during development 31 . Combined with evidence that over the course of the CP, when Agrp neurons innervate their downstream targets 1,4 , their response to leptin switches from excitatory to inhibitory 17 , these data collectively point to the existence of a mechanism whereby Agrp-linked circuit plasticity during postnatal development is sharply constrained in adulthood. PNNs are likely candidates to mediate this effect, as they can limit plasticity both through direct signaling effects of CSPGs on the enmeshed neuron and by providing a scaffold for binding regulatory molecules such as Otx2 13 (which suppresses plasticity) and the chemorepulsive molecule Semaphorin 3a 32 (which promotes growth of projections away from the enmeshed neuron) 5,[8][9][10][11][12][13][14]32 .
Such a role for PNNs has important implications for understanding how nutritional excess during postnatal development affects ARC neurocircuits in ways that can predispose to obesity and T2D in adulthood 2 . Specifically, over-nutrition during lactation, due either to maternal HFD consumption 18,[20][21][22] or to culled litter size 19 , 1) reduces numbers of ARC neurons expressing leptin receptors. 2) decreases leptin responsiveness of these neurons 18,19 , 3) impairs formation of ARC projections within hypothalamic feeding circuits 18,20 , and 4) predisposes to excess body adiposity and metabolic dysregulation in adulthood [20][21][22] . Similarly, epidemiological evidence suggests that in humans, late gestation and early childhood are particularly sensitive periods when environmental exposures can shape a predisposition to metabolic disease in adulthood 33 . Interestingly, early gestational undernutrition is also reliably associated with an increased risk of adult obesity 34,35 , and this effect can be partially reversed by leptin treatment during the lactation period 36 . Since PNNs sharply limit the plasticity of neurons that they enmesh, and since elsewhere in the brain, the developmental appearance of PNNs heralds CP closure 37 , we interpret our discovery that ARC neurons become enmeshed by PNNs at a time corresponding to CP closure as being of potential relevance to experience-dependent plasticity in neurocircuits for energy balance and glucose homeostasis. It is possible, for example, that experimental re-activation of ARC plasticity in adults will ameliorate metabolic dysfunction, as has been observed following reinduction of plasticity in the visual cortex in models of amblyopia 5 .
In addition to limiting plasticity during development, PNNs can be altered in adulthood in ways that impact the function of the neurons they enmesh, and the variable effect of different CSPGs, hyaluronan and other PNN components to influence plasticity is welldocumented 37 . PNN composition can also change in response to injury, inflammation, or neurodegeneration in ways that further restrict the plasticity of enmeshed neurons 9 . Thus, either a deficiency 38,39 or an overabundance 9 of PNNs or individual PNN components can have deleterious effects on circuit behavior and neurological function. As one example, functional recovery after spinal cord injury is limited by "strengthening" of PNNs owing to inflammation and reactive gliosis, such that recovery is improved by PNN disruption 37 .
In this context, it is notable that diet-induced obesity (DIO) induced by HFD feeding is associated with reactive gliosis involving activation of both microglia and astrocytes in the same ARC area where PNNs are found 40 , and recent work suggests that these glial responses are both necessary and sufficient for obesity in this setting 41,42 . In this context, our finding of increased tanycyte expression of the hyaluronan receptor CD44 in mice with DIO is of interest because it raises the possibility that environmental exposures in adulthood (e.g., consuming a HFD) can influence signaling between PNN constituents (e.g., hyaluronan) and adjacent cells (e.g., tanycytes) in the ARC-ME. Reports that tanycytes transport circulating leptin into the mediobasal hypothalamus 43 heighten the potential importance of such interactions. That this obesity-associated increase of tanycyte CD44 content is not observed in ob/ob mice despite their severe obesity phenotype is consistent with evidence that hypothalamic gliosis is also not observed in these animals 44 , presumably reflecting the requirement for an intact leptin signal in this response.
In conclusion, we report that the ARC-ME junction, a brain area that specializes in processing humoral input relevant to fuel homeostasis, is marked by a cluster of tightly packed, PNN-enmeshed neurons, including a majority of Agrp neurons. The formation of these PNNs coincides with closure of the CP for ARC neuron development and, like postnatal maturation of Agrp neurons, is dependent on input from the adipocyte hormone leptin. Future investigation is needed to clarify the extent to which both PNN composition and ARC neuronal plasticity are altered in models of obesity, diabetes and related metabolic disorders, and whether PNN formation in this region can be altered pharmacologically.

Animals
GAD67-GFP knock-in 45 , LepRb-Cre;Ai14 reporter 46,47 , Npy-GFP 48 , POMC-GFP 49 , and wild-type C57B/6 mice (Jackson Labs), age P60 to P120, were used to characterize neuronal and glial subtypes associated with ARC PNNs. Wild-type C57B/6 mice, age P0 to P90 were used for developmental time-series studies to characterize PNN formation and Agrp neuron maturation. To study the effects of leptin deficiency on PNN formation, we used ob/ob mice and their ob/+ littermate controls (Jackson Labs). To characterize PNNs in rats, we used the Wistar strain (Harlan). Both sexes were used for all studies. Mice were on 12h:12h lightdark cycle in 5/cage group housing. Animals were perfusion-fixed with saline and 4% paraformaldehyde (PFA), the brains were extracted, post-fixed overnight at 4°C, then sectioned using a vibratome (50 um) or cryostat (12 um). The Institutional Animal Care and Use Committees at St. Joseph's Hospital and Medical Center and the University of Washington approved all animal procedures. All animal studies were conducted in compliance with the relevant ethical regulations.

Human Specimens
Three brains (ages 23, 64, and 71 years with postmortem intervals 8, 9, and 12 hours, respectively) were collected at autopsy. Multiple wholemount blocks were dissected from the hypothalamus along the 3rd ventricle from each brain, their positions along the 3 rd ventricle wall were documented, and the tissue was immersion-fixed in 4% PFA at 4°C for 24 hours. All specimens were collected with informed consent and in ethical compliance with the St. Joseph's Hospital and Medical Center Committee on Human Research (IRB approval no. 10BN159).

Stereotactic Intracranial Injections
Animals were head-fixed with ear bars in a custom digital stereotactic rig used to perform unilateral intra-ARC injections of Chondroitinase ABC (10 milliunits in 100 nanoliters sterile PBS) at the coordinates (0.3 mm lat, 1.1 mm posterior, 5.6 mm depth) relative to bregma, and sacrificed 48h later. For ICV colchicine administration, 1 ul of colchicine (10 ug/ul in sterile PBS) was injected at (1.3, 0, 1.5) relative to bregma and animals were sacrificed 24h later.

Postnatal leptin administration
We crossed ob/+ mice to generate litters containing ob/ob pups. Litters were genotyped at P1-2 and litter sizes were adjusted to 6-8 pups to standardize nutrition during lactation period. Leptin (10 mg/kg i.p.; made available via Dr. A.F. Parlow; National Hormone & Peptide Program, CA) or vehicle was administered to ob/ob littermates on a daily basis at noon from P10-P30 and the animals were euthanized 24 hours after the last injection.

Wholemount Dissection
After cervical dislocation, the mouse brain was removed from the skull and 3V wholemounts were freshly dissected using principles similar to those described for LV wholemounts 50 . Briefly, 3V wholemounts were dissected by performing a ventriculotomy of the 3V from ventral to dorsal and rostral to caudal. The exposed ventricle walls were immersion-fixed in 4% PFA at 4°C overnight prior to immunostaining.

Electron Microscopy
Adult C57B/6 mice (Jackson Labs) were transcardially perfused with 4% PFA and 0.5% glutaraldehyde (EMS) in 100 mM phosphate buffer (PB). Brains were post-fixed at 4°C overnight, and 50 μm coronal sections were cut on a Leica VT1000 S vibratome. Pre-embedding IHC was performed using WFA (Sigma L1516), amplified with Vectastain Elite ABC kit (Vector Laboratories), and developed with DAB 23 . Sections were postfixed in 1% osmium tetroxide for 30 min and then embedded in Durcupan ACM epoxy resin (Fluka, Sigma-Aldrich) 51 . For reconstruction of WFA-labeled neurons, we cut ~200 serial semithin (1.5 μm) sections on an Ultracut UC-6 ultramictrotome. Selected semithin sections were glued to resin blocks and detached from glass slides by repeated freeze-thaw. Ultrathin sections (60-80 nm) were then cut and placed on Formvar-coated single-slot grids, stained with lead citrate, and examined at 80 kV on a FEI Tecnai G2 Spirit transmission electron microscope equipped with a Morada CCD digital camera (Olympus).

Image Analysis and Quantification
To quantify the proportion of various ARC neuronal subtypes enmeshed by PNNs, we used high resolution confocal z-stacks to reconstruct the ventromedial arcuate region containing PNN structures in two coronal sections containing the median eminence (Bregma −1.7 and −2.0). Only reporter-labelled cells that were completely enmeshed (360°) by WFA-labelled PNN matrix were counted as positive. To analyze the intensity of WFA or CD44 labeling, high resolution confocal z-stacks were volumetrically analyzed in Imaris image analysis software (Bitplane) and voxel intensities throughout the region of interest (e.g. Arc) were determined (intensity unit range, 0-255) and compared across study groups. Raw images were then rendered as 3-dimensional spectral images corresponding to individual voxel intensities. For Agrp fiber density measurements, we used Imaris software to determine the volume of Agrp-labeled positive voxels, above a set intensity threshold, as a fraction of the total volume in the measured region of interest.

Statistical Analysis
Descriptive statistics and two-tailed t-tests were computed in GraphPad Prism 7. P-value less than 0.05 was considered statistically significant. Unless otherwise stated, dot plots show dots representing data from independent animals with mean and standard error of the mean shown as bars.

Supplementary Material
Refer to Web version on PubMed Central for supplementary material. Wisteria Floribunda agglutinin (WFA)-labeling in the ventromedial ARC forms a "collar" around the ME. Diagrams at top show mid-sagittal view (left) and ventral view (right) of the mouse brain with insets showing the location and orientation of panel images. (a-d) WFA-labeled (red) coronal sections through the Arc, starting just rostral to and progressing through the ME, show a concentration of WFA-labeled cells located in the ARC at its junction with the ME. Note that the very intense staining below the ME does not correspond to labeling around neurons, but to the pia around the ME.   PNN formation in the ARC occurs during the lactation and periweaning period, corresponding with the maturation of Agrp neurons. (a-c) Confocal images of coronal sections stained with WFA (red), Agrp (green), and dapi (blue) from postnatal wild-type mice at age P10 (a), P21 (b), and P30 (c). PNN staining intensity and ARC Agrp fiber density increase in parallel over this time period.
(d) Dot plot shows correlated increase in WFA intensity and Agrp fiber density in the ARC from P10 to P30, as well as at P90. Dots (WFA intensity in red, Agrp density in black) represent values from independent animals (n=4 (P10); 5 (P21); 3 (P30); and 3 (P90)), and horizontal bars represent the mean. WFA intensity is represented by the average over all voxels in the ARC region of interest, with range 0-255. Agrp fiber density is measured as the volume of Agrp+ voxels divided by the total volume of the ARC region of interest. Scale bar: 100 um (a-c). Images in (a), (b), and (c) are representative of data from 4, 5, and 3 animals, respectively. Leptin-deficient ob/ob mice have impaired PNN formation during postnatal development that can be rescued by leptin administration during the critical period. (a-d) Confocal images of ARC sections from ob/ob (b,d) and ob/+ (a,c) control littermates at P15 (a,b) and P30 (c,d), stained with WFA (red) and Agrp (green), show reduced WFA labeling and apparent disruption of PNN architecture in the ARC. Arrows in (a,b) indicate the ARC region where the earliest PNN formation is seen at P15 in ob/+ mice, but not in ob/ob littermates. Images in (a), (b), (c), and (d) are representative of data from 2, 3, 5, and 5 animals, respectively.
(e-f) Confocal images of ARC sections from ob/ob pups that received daily i.p. injections of leptin (f) or vehicle (e) from P10 to P30 before being euthanized for analysis with WFA