Neuronal growth regulator 1 (NEGR1) promotes synaptic targeting of glutamic acid decarboxylase 65 (GAD65)

Neuronal growth regulator 1 (NEGR1) is a glycosylphosphatidylinositol-anchored cell adhesion molecule encoded by an obesity susceptibility gene. We demonstrate that NEGR1 accumulates in GABAergic inhibitory synapses in hypothalamic neurons, a GABA-synthesizing enzyme GAD65 attaches to the plasma membrane, and NEGR1 promotes clustering of GAD65 at the synaptic plasma membrane. GAD65 is removed from the plasma membrane with newly formed vesicles. The association of GAD65 with vesicles results in increased GABA synthesis. In NEGR1 deficient mice, the synaptic targeting of GAD65 is decreased, the GABAergic synapse densities are reduced, and the reinforcing effects of food rewards are blunted. In mice fed a high fat diet, levels of NEGR1 are increased and GAD65 abnormally accumulates at the synaptic plasma membrane. Our results indicate that NEGR1 regulates a previously unknown step required for synaptic targeting and functioning of GAD65, which can be affected by bidirectional changes in NEGR1 levels causing disruptions in the GABAergic signaling controlling feeding behavior.

Increased susceptibility to pentylenetetrazol-induced seizures of these animals (Singh et al., 2018) suggests that NEGR1 is involved in control of neuronal activity via currently unknown mechanisms.
Inhibitory γ-amino butyric acid (GABA)ergic signaling controls the excitability of neuronal networks (Roth and Draguhn, 2012). Deficits in GABAergic neurotransmission are associated with 4 neurodevelopmental disorders, including attention deficit / hyperactivity disorder (Edden et al., 2012), developmental co-ordination disorder (Umesawa et al., 2020), and depressive disorders (Luscher et al., 2011). GABAergic neurotransmission also plays a key role in regulating food intake (Meng et al., 2016;Xu et al., 2012). GABA is synthesized from glutamate by glutamic acid decarboxylase (GAD), an enzyme encoded by GAD1 and GAD2 genes coding for GAD67 and GAD65 isoforms. GAD67 produces the majority of GABA and is required for development of inhibitory neuronal circuits and maintenance of basal inhibitory firing. GAD65 synthesizes GABA to fine-tune and maintain GABAergic synapse function during neuronal activity (Baekkeskov and Kanaani, 2009;Patel et al., 2006). GAD65 is a soluble cytosolic protein, which attaches to membranes via hydrophobic modifications, with subsequent palmitoylation required for its targeting to synaptic vesicles (Baekkeskov and Kanaani, 2009). Glutamate decarboxylating activity of purified GAD65 is increased on purified synaptic vesicles (Hsu et al., 2000) and is coupled to the transport of GAD65-synthesized GABA into synaptic vesicles by the vesicular GABA transporter (VGAT) (Jin et al., 2003). Synaptic vesicles fuse with the PM to release GABA and are then reformed via the retrieval of synaptic vesicle membranes from the PM. Many peripheral proteins involved in this vesicle recycling process are recruited to synapses and then to synaptic vesicle membranes in a highly regulated manner and at different stages of the recycling process. GAD65 is co-purified with vesicles in the presence of pyridoxal 5'-phosphate, a co-factor necessary for GAD65-mediated synthesis of GABA (Jin et al., 2003), but not in its absence (Reetz et al., 1991;Takamori et al., 2006;Takamori et al., 2000), suggesting that the attachment of GAD65 to vesicles is regulated. Regulation of the synaptic targeting of GAD65 is, however, poorly understood.
In this work we show that GAD65 attaches to the PM and is removed from the PM with newly formed vesicles. NEGR1 targets GAD65 to synapses by promoting clustering of GAD65 at the presynaptic PM. NEGR1 deficiency causes GABAergic synapse loss indicating that clustering of GAD65 at the presynaptic PM is required for the maintenance of inhibitory synapses.

NEGR1 promotes synaptic targeting of GAD65
Since NEGR1 is involved in regulating feeding behavior, we studied its distribution in cultures of hypothalamic neurons, which play a key role in regulating food intake. Confocal microscopy showed that NEGR1 was present along dendrites and axons (Fig. 1A) identified as thick tapering MAP2-positive protrusions and thin tau-enriched protrusions of a uniform diameter, respectively (Fig.   1B). The majority of synapses in these neurons (76.6 ± 2.03%, n = 20 neurons) were positive for GAD65. NEGR1 clusters co-localized with accumulations of VGAT and GAD65 indicating that NEGR1 accumulates in inhibitory synapses (Fig. 1A) and suggesting that it plays a role in regulating their formation or function.
Western blot analysis demonstrated that the levels of an inhibitory synapse marker protein VGAT were similar in brain homogenates of 3-month-old NEGR1+/+ and NEGR1-/-mice (Fig. 1C), while the levels of GAD65 tended to be higher in NEGR1-/-brain homogenates (Fig. 1E) (mean ± SEM fold change in NEGR1-/-vs NGER1+/+ was 1.17 ± 0.08 for VGAT and 1.67 ± 0.31 for GAD65 (n = 5, p = 0.06, Wilcoxon signed rank test)). NEGR1 was highly enriched in synaptosomes vs brain homogenates of NEGR1+/+ mice (Fig. 1C). Hence, we asked whether NEGR1 deficiency affects targeting of GAD65 to synapses. To take into account changes in total protein levels, synaptic targeting (ST) was estimated by calculating the ratio of protein levels in synaptosomes and brain homogenates in the same animal. STGAD65 was reduced in NEGR1-/-vs NEGR1+/+ mice (Fig. 1E, F), whereas STVGAT was not changed (Fig. 1C, D). To determine whether NEGR1 promotes synaptic targeting of GAD65 by recruiting it to membranes from the soluble protein pool, the enrichment of GAD65 in soluble protein fractions relative to its total levels in brain homogenates was analyzed. Surprisingly, GAD65 enrichment in the soluble protein fraction was reduced in NEGR1-/-vs NEGR1+/+ mice (Fig.   1E, F), indicating that NEGR1 deficiency causes accumulation of GAD65 in non-synaptic membranes. 6 GAD65 distribution was then analyzed in cultured NEGR1+/+ and NEGR1-/-hypothalamic neurons by confocal microscopy. In NEGR1+/+ neurons, GAD65 clusters predominantly co-localized with synaptophysin accumulations (Fig. 1G), however, GAD65 clusters which did not co-localize with synaptophysin were also observed (Fig. 1G). The percentage of these non-synaptic GAD65 clusters was increased in NEGR1-/-neurons (Fig. 1G). To exclude a possibility that this effect was solely due to increased levels of GAD65 in NEGR1-/-vs NEGR1+/+ neurons, the synaptic targeting of GAD65-GFP was compared in NEGR1-/-hypothalamic neurons co-transfected with NEGR1 or empty pcDNA3 vector. In NEGR1-co-transfected neurons, clusters of GAD65-GFP co-localized with NEGR1 accumulations, and the percentage of synaptophysin negative GAD65-GFP clusters was reduced ( Fig. 1H). Our combined observations thus indicate that NEGR1 promotes synaptic targeting of GAD65.

Trans-interactions of NEGR1 promote synaptic targeting of GAD65
To determine whether GAD65 is targeted to synapses by trans-interactions of NEGR1, NEGR1+/+ cultured hypothalamic neurons were treated with recombinant soluble NEGR1 (sNEGR1), which binds to cell surface NEGR1 (Kim et al., 2014). Remarkably, sNEGR1 induced a strong increase in synaptic GAD65 levels ( Fig. 2A). The effect was observed at 15 min after application of sNEGR1 and lasted for 24 h (the last time point analyzed) ( Fig. 2A). A similar effect was found in slices of NEGR1+/+ hypothalamic tissue treated for 24 h with sNEGR1. Western blot analysis of synaptosomes from these slices showed an increase in synaptic levels of GAD65 when compared to GAD65 levels in synaptosomes from control tissues treated with bovine serum albumin (Fig. 2B). Interestingly, the overall levels of GAD65 were also increased in sNEGR1-treated slices (Fig. 2B). To analyze whether the NEGR1-dependent increase in synaptic GAD65 levels results in enhanced GABA synthesis, GABA levels in the synaptosomes were determined by mass spectrometry. Surprisingly, this analysis 7 showed that GABA levels were reduced in synaptosomes isolated from sNEGR1-treated tissues despite the increase in GAD65 levels, indicating that NEGR1-mediated trans-interactions restrain GAD activity (Fig. 2C).
NEGR1 restrains the retrieval of synaptic vesicle membranes from the PM GAD65 protein and mRNA levels are reduced in response to prolonged stimulation of synaptic vesicle recycling (Buddhala et al., 2012), suggesting that the regulation of GAD65 levels is coupled to synaptic vesicle recycling. Since sNEGR1 induced an overall increase in GAD65 levels, we asked whether NEGR1 regulates synaptic vesicle recycling. Activity-dependent synaptic vesicle recycling was visualized in cultured NEGR1+/+ and NEGR1-/-hypothalamic neurons by loading synaptic vesicles for 2 min with the lipophilic FM4-64 dye applied in buffer containing 47 mM K + , which causes PM depolarization and Ca 2+ influx initiating synaptic vesicle fusion with the synaptic PM followed by vesicle reformation. The amount of FM4-64 taken up into synaptic boutons was increased in NEGR1-/-neurons (Fig. 3A). FM4-64 release in response to electric field stimulation was then analyzed as a measure of synaptic vesicle fusion with the PM and was found to be faster in NEGR1-/neurons ( Fig. 3A).
NEGR1 was reported to be present post-synaptically (Hashimoto et al., 2008) but was also detected in pre-synaptic compartments (Takamori et al., 2006) and was proposed to mediate transsynaptic adhesion (Ranaivoson et al., 2019;Venkannagari et al., 2020). Hence, we investigated whether pre-or post-synaptic NEGR1 regulates synaptic vesicle recycling by determining the effects of NEGR1 overexpression in axons and dendrites of cultured NEGR1+/+ hypothalamic neurons.
Control neurons were transfected with the empty pcDNA3 vector. Neurons were co-transfected with GFP, and their proximal dendrites were identified as thick tapering protrusions, while axons were identified as thin protrusions of a uniform diameter. Synapses formed by GFP-positive axons of 8 transfected neurons on GFP-negative dendrites of non-transfected neurons were analyzed to determine the effect of presynaptic overexpression. Synapses formed by GFP-negative axons of non-transfected neurons on GFP-positive dendrites of transfected neurons were analyzed to determine the effect of postsynaptic NEGR1 overexpression. The FM4-64 uptake was reduced in synaptic boutons formed by axons of NEGR1 overexpressing vs control neurons (Fig. 3B) and was also reduced in synapses formed on dendrites of NEGR1 overexpressing vs control neurons (Fig. 3C). FM4-64 release in response to the electric field stimulation was slower in axons of NEGR1 overexpressing vs control neurons ( To exclude the possibility that these changes in synaptic vesicle recycling reflect solely developmental effects of prolonged NEGR1 loss or overexpression, NEGR1+/+ cultured hypothalamic neurons were acutely treated with antibodies against NEGR1 used as a specific NEGR1 ligand, a method used to cluster and mimic trans-interactions of cell adhesion molecules at the cell surface (Sheng et al., 2015;Sheng et al., 2019). Control neurons were incubated with non-specific immunoglobulins (Ig). The FM4-64 uptake was strongly reduced in NEGR1 antibody-vs Ig-treated neurons (Fig. 3D). Unexpectedly, FM4-64 release in response to the electrical field stimulation was slightly faster in NEGR1 antibody-treated neurons (Fig. 3D).
Altogether, our data indicate that both pre-and post-synaptic NEGR1 restrain the retrieval of synaptic vesicle membranes from the synaptic PM, while only presynaptic NEGR1 also plays a role in regulating fusion of synaptic vesicles with the synaptic PM.
Inhibition of synaptic vesicle membrane retrieval promotes GAD65 clustering at the synaptic PM 9 To determine whether a restraint of synaptic vesicle membrane retrieval influences the synaptic localization of GAD65, the association of GAD65 clusters with synaptophysin accumulations was analyzed in NEGR1+/+ cultured hypothalamic neurons treated for 15 min with dynasore. This inhibitor of dynamin selectively suppresses synaptic vesicle membrane retrieval after action potential-induced neurotransmitter release but does not affect spontaneous synaptic vesicle trafficking (Chung et al., 2010). Control neurons were treated with vehicle (DMSO). The numbers of synaptophysin-negative GAD65 clusters were reduced in dynasore-treated vs control neurons (Fig. 4A). sNEGR1 induced a similar effect in a non-additive manner with dynasore (Fig. 4A). Interestingly, tetanus toxin, which blocks fusion of synaptic vesicles with the PM, also reduced the percentage of synaptophysin-negative GAD65 clusters in a non-additive manner with sNEGR1 ( Fig. 4B).
Synaptic vesicle membrane retrieval is coupled to vesicle fusion with the PM (Haucke et al., 2011). To determine whether sNEGR1 inhibits fusion of synaptic vesicles with the PM or retrieval of their membranes from the PM in GABAergic synapses, live cultured hypothalamic neurons pretreated with sNEGR1 or culture medium for control were incubated for 30 min with antibodies against the lumenal domain of VGAT. The antibodies bind to the lumenal domain of VGAT when it is exposed at the cell surface after fusion of synaptic vesicles with the PM. The antibodies then either remain at the cell surface or are taken up into vesicles reformed via membrane retrieval from the PM. The pool of antibodies remaining at the cell surface (VGATsurface_remaining), which is inversely proportional to the retrieval rate, was visualized with secondary antibodies applied to neurons which were not permeabilized with detergent. The total pool of antibodies at the cell surface and in newly formed vesicles (VGATsurface_delivered), which is proportional to the rate of synaptic vesicle fusion with the PM, was visualized with secondary antibodies applied to fixed and detergent-permeabilized neurons (Fig.   4C, D). The total synaptic pool of VGAT (VGATsynaptic) was visualized by co-labeling detergentpermeabilized neurons with antibodies against the cytoplasmic domain of VGAT. VGATsurface_delivered and VGATsynaptic were similar in sNEGR1-treated and control neurons (Fig. 4C), whereas 10 VGATsurface_remaining was increased in sNEGR1-treated neurons (Fig. 4D), indicating that sNEGR1 did not affect the fusion of vesicles with the PM but inhibited the retrieval of synaptic vesicle membranes from the PM.
Inhibition of synaptic vesicle membrane retrieval causes clustering of synaptic vesicle proteins in the presynaptic PM (Dason et al., 2014). To test whether sNEGR1 induces accumulation of GAD65 in the presynaptic PM, synaptosomes isolated from slices of hypothalamic tissue were used to obtain synaptic PMs isolated after synaptosome lysis with osmotic shock, which releases synaptosomal contents including synaptic vesicles. Western blot analysis demonstrated that GAD65 was co-isolated with the synaptic PMs (Fig. 4E). In slices treated with sNEGR1 for 24 h, the levels of GAD65 coisolated with the synaptic PMs were ~2-fold higher than in membranes from slices treated with bovine serum albumin for control (Fig. 4E). Dynasore induced a similar increase in GAD65 levels in the synaptic PMs in a non-additive manner with sNEGR1 ( Fig. 4E). An increase in GAD65 levels coisolated with the synaptic PMs was also found in slices treated with tetanus toxin (Fig. 4E), indicating that fusion of vesicles with the PM is not required for clustering of GAD65 at the synaptic PM and that GAD65 is not delivered to the synaptic PM by synaptic vesicles.

GAD65 attaches to the PM and NEGR1 promotes its recruitment to the PM-derived vesicles in CHO cells
Our finding that GAD65 is co-isolated with the synaptic PM contrasts with extensive microscopy data in different cell types showing that GAD65 does not accumulate at the PM. It is, however, inherently difficult to visualize transient interactions of peripheral proteins such as GAD65 with the PM in intact cells. We attempted to achieve this by using bimolecular fluorescence complementation (BiFC) to simultaneously capture and visualize GAD65 in close proximity to the PM via irreversible reconstitution of the Venus fluorescent protein. As a membrane localized sensor, we engineered an LCK-VN protein consisting of the N-terminal fragment of Venus fused to the first 26 amino acids of human LCK containing sites for myristoylation and palmitoylation which target it to the PM (Fig. 5A). The sensor was used to detect the PM attachment of GAD65 fused to the complementary C-terminal fragment of Venus (GAD65-VC). Reconstitution of Venus from the VC and VN fragments brought in close proximity to each other by the attachment of GAD65 to the PM captures GAD65 at the site of PM attachment and simultaneously visualizes it by producing fluorescence (Fig. 5A). The crowded environment of synaptic boutons does not allow unambiguous determination of whether proteins are bound to the PM rather than synaptic vesicles. Therefore, we used cultured CHO cells as a model system to visualize GAD65 interactions with the PM in vesicle free PM areas. Labelling of LCK-VN transfected CHO cells with antibodies against GFP, which detect the VN fragment, showed that this sensor was localized at the PM and intracellular accumulations partially overlapping with the Golgi marker GM130 (Fig. 5B). Co-transfection of cells with LCK-VN and non-mutated GAD65WT-VC produced BiFC fluorescence, which accumulated in the perinuclear region ( Fig. 5C) where GAD65 tightly attaches to the Golgi (Kanaani et al., 2008). In addition, clusters of BiFC fluorescence were found at the PM and in small accumulations in the cytosol (Fig. 5C). A closer analysis of cross-sections of the 3D reconstructed transfected cells showed that BiFC signals at the PM were found in small clusters or vesicle-like structures (Fig. 5D). A similar level of BiFC fluorescence was observed in CHO cells transfected with GAD65(C30,45A)-VC mutant (Fig. 5C), in which cysteines responsible for GAD65 palmitoylation (Kanaani et al., 2008) were mutated to alanine.
The BiFC signal produced by the GAD65(C30,45A)-VC mutant was also found at the PM (Fig. 5C), indicating that palmitoylation is not required for the attachment of GAD65 to the PM. The BiFC signal was dramatically reduced in cells co-transfected with GAD65(24-31A)-VC (Fig. 5C), in which amino acids 24-31 responsible for binding to membranes (Shi et al., 1994) were substituted for alanine. In cells co-transfected with NEGR1, cell surface clusters of NEGR1 were found at sites of BiFC fluorescence accumulations (Fig. 5E). A subpopulation of PM-localized BiFC fluorescence clusters 12 co-localized with clathrin accumulations suggesting that the attachment of GAD65 to the PM triggers its recruitment to the clathrin-coated pits (Fig. 5F). To determine whether NEGR1 plays a role in recruiting GAD65 to the vesicles derived from the PM, CHO cells were incubated live with FM4-64 to load the dye in and visualize newly formed vesicles. Since stabilization of the PM attachment via BiFC may influence endocytosis, GAD65-GFP was used in these experiments to visualize GAD65 on the vesicles. The levels of FM4-64 uptake were reduced, however, the percentages of FM4-64-loaded vesicles co-localizing with GAD65 accumulations were increased in NEGR1-transfected vs control pcDNA3-transfected cells (Fig. 5G). Altogether, these observations indicate that GAD65 attaches to the PM and NEGR1 promotes its targeting to the PM-derived vesicles. Our data also further suggests that this sorting requires a restraint of the retrieval of membranes from the cell surface.

NEGR1 promotes the assembly of lipid rafts at the PM and in synapses
Disruption of lipid rafts blocks synaptic clustering of GAD65 resulting in formation of nonsynaptic GAD65 clusters (Kanaani et al., 2002) resembling those found in NEGR1-/-neurons ( Fig.   1G). NEGR1 associates with lipid rafts via its GPI-anchor and is involved in cholesterol transport (Kim et al., 2017). We therefore asked whether NEGR1 regulates synaptic clustering of lipid microdomains. The targeting of cholesterol and ganglioside GM4 to the PM was increased in NEGR1transfected vs control pcDNA3 vector-transfected CHO cells (Fig. 6A, B) indicating that NEGR1 promotes the assembly of lipid rafts at the PM. A dot blot analysis showed that the enrichment of lipid raft-localized gangliosides and phosphatidylinositol 4,5-bisphosphate in synaptosomes vs homogenates was strongly reduced in NEGR1-/-vs NEGR1+/+ mice (Fig. 6C), indicating that NEGR1 can target GAD65 to synapses by inducing synaptic clustering of lipid rafts, which can also restrain the retrieval of synaptic vesicle membranes since synaptic vesicle proteins synaptophysin and VAMP2 bind to the PM localized cholesterol (Thiele et al., 2000).

13
The association of GAD65 with vesicles promotes GABA synthesis The trafficking of GAD65 to post-Golgi vesicular membranes is controlled by its palmitoylation cycle (Kanaani et al., 2008). We assessed whether a palmitoylation-deficient mutant of GAD65, which is capable of firmly anchoring to ER/cis-Golgi but incapable of anterograde trafficking from cis-Golgi to TGN and post-Golgi vesicles (Phelps et al., 2016), would exhibit altered enzyme activity. Pancreatic beta cells synthesize and secrete large quantities of GABA, making them excellent candidates for studying GAD enzyme activity (Menegaz et al., 2019). INS-1 and MIN6 beta cell lines have lost the expression of endogenous GAD65 and GAD67 (Cianciaruso et al., 2017). We thus used INS-1 beta cells to study the GAD enzyme activity of cells transfected with GAD65-GFP relative to the palmitoylation deficient mutant GAD65(C30,45A)-GFP in the absence of a background of endogenous GABA biosynthesis. As previously observed (Kanaani et al., 2008;Phelps et al., 2016), GAD65-GFP distributed between cytosol, ER, Golgi, and post-Golgi vesicles, while GAD65(C30,45A)-GFP was distributed only to the cytosol, ER, and cis-Golgi (Fig. 7A). HPLC analysis of amino acids from INS-1E cell lysate showed a 50% decrease in the cellular GABA content for GAD65(C30,45A)-GFP and a significant increase in glutamate, the precursor to GABA synthesis . These data demonstrate that palmitoylation-dependent vesicular trafficking of GAD65 is correlated with higher levels of GABA biosynthesis.

The density of GABAergic synapses is reduced in 7-8-month-old NEGR1-/-mice
In cultures of disassociated hypothalamic neurons, NEGR1 was particularly strongly expressed in axons and synaptic boutons of neuropeptide Y (NPY) / GAD65 positive neurons when compared to other neurons present in these cultures (Fig. 8A). NPY positive neurons and their projections are present at high density in the ARC (Suyama and Yada, 2018). Analysis of NEGR1+/+ and NEGR1-/-14 brain sections by confocal microscopy showed that the density of VGAT/GAD65 positive inhibitory synapses was strongly reduced in the ARC of NEGR1-/-vs NEGR1+/+ mice (Fig. 8B). In NEGR1-/mice, the density of VGAT/GAD65 positive synapses was also reduced in the CA3 region of the hippocampus (Fig. 8C), which also contains NPY-positive interneurons (Kuruba et al., 2011).
Altogether, our data indicate that NEGR1 deficiency causes a loss of inhibitory synapses.

NEGR1-/-mice show reduced motivation for a highly palatable food reward
GABAergic neurotransmission is involved in regulation of the motivational aspects of food reinforced behaviors and is sensitized by intake of palatable food (Newman et al., 2013). To determine whether NEGR1 deficiency affects motivation for food, 7-8-month-old NEGR1+/+, NEGR1-/-and NEGR1+/-mice were tested using an instrumental conditioning task. The mice were initially allowed to eat sucrose pellets released into a magazine at a rate of 1 pellet / min over 30 min. Analysis of these initial magazine training sessions performed over 6 days showed that the mice of all genotypes initially consumed similar numbers of pellets per session (Fig. 9A). Over time, the numbers of pellets consumed per session increased for NEGR1-/-but not NEGR1+/+ mice with an intermediate effect observed for NEGR1+/-mice (Fig. 9A). The mice were then trained to press a lever to obtain a sucrose pellet. In this fixed ratio task, mice of all genotypes pressed the lever with approximately the same frequency and consumed similar numbers of pellets per session (Fig. 9B). Finally, the mice were tested using a progressive ratio schedule of reinforcement, where the number of lever presses required to obtain the pellet progressively increased across the session. When the number of lever presses required to obtain a single sucrose pellet exceeded motivation for the pellet, the mice stopped responding and this was considered the break point. This analysis showed that breakpoint values were similar in all genotypes on the first day of tests (Fig. 9C), and progressively increased in NEGR1+/+ mice over three days of tests (Fig. 9C, D). Accordingly, the number of pellets delivered per session and the number of 15 active lever presses per session also increased in NEGR1+/+ group (Fig. 9E). This sensitization was inhibited in NEGR1+/-and NEGR1-/-mice ( Fig. 9C-E). Hence, NEGR1 deficiency does not negatively impact on the palatability of the food pellet or learning around the instrumental task, but rather specifically affects motivation for highly palatable food rewards.

A high fat diet causes an increase in NEGR1 levels affecting subsynaptic localization of GAD65 and synapse maintenance
A high fat diet (HFD) causes a reduction in GABA levels at least in some brain regions (Hassan et al., 2018;Sandoval-Salazar et al., 2016). Hence, we asked whether levels of NEGR1 are influenced by HFD. Western blot analysis demonstrated that levels of NEGR1 were increased in brain homogenates and synaptosomes of HFD-vs chow-fed mice (Fig. 10A, B). NEGR1 can be shed by ADAM10, which releases its ~50 kDa cleavage product (Pischedda and Piccoli, 2015). The levels of this product in the soluble protein fraction were, however, below the detection limit indicating low levels of the ADAM10-mediated cleavage of NEGR1 in the mature brain (  (Bocarsly et al., 2015). Selective loss of inhibitory but not 16 excitatory synapse markers is found in dynamin 1 and 3 double knock-out neurons (Raimondi et al., 2011), indicating that defects in the retrieval of synaptic vesicle membranes affect the maintenance of inhibitory synapses. To determine whether the NEGR1 overexpression-induced decrease in synaptic vesicle membrane retrieval (Fig. 3B, C) causes changes in synapse numbers, synaptophysin levels were measured along dendrites and axons of NEGR1-and control pcDNA3-transfected cultured hypothalamic neurons. Synaptophysin levels were strongly reduced along axons and dendrites of NEGR1-overexpressing neurons (Fig. 10H). Altogether, our data indicate that NEGR1 is overexpressed in brains of HFD-fed mice, and an increase in levels of this protein correlates with reduced GABA synthesis and can cause a loss of inhibitory synapses.

Discussion
We report that NEGR1 promotes the synaptic targeting of GAD65, a GABA synthesizing enzyme. We demonstrate that GAD65 attaches to the synaptic PM and show that trans-interactions of the synaptic PM localized NEGR1 result in an increase in this pool (Fig. 11). This increase correlates with a reduction in numbers of non-synaptic GAD65 clusters, indicating that NEGR1 promotes the recruitment of non-synaptic GAD65 to the synaptic PM. Fusion of synaptic vesicles with the PM may also result in the PM targeting of GAD65 attached to the synaptic vesicle membranes. We show, however, that the pool of GAD65 at the synaptic PM is not reduced and is even increased in tissues treated with tetanus toxin, which blocks fusion of synaptic vesicles with the PM. These experiments, electron microscopy data from previous studies showing GAD65 on some but not all synaptic vesicles, and biochemical studies showing that GAD65 does not co-purify with synaptic vesicles (Reetz et al., 1991;Takamori et al., 2006;Takamori et al., 2000) collectively indicate that GAD65 disassociates from vesicles before they fuse with the PM. Interestingly, synaptic vesicles contain high levels of palmitoyl protein thioesterase-1, an enzyme which de-palmitoylates GAD65 (Kim et al., 2008) and can therefore reduce the palmitoylation-dependent association of GAD65 with the vesicles (Kanaani et al., 2008). Deficiency in palmitoyl protein thioesterase-1 leads to a reduction in GAD65 levels in the brain (Kim et al., 2008) indicating that GAD65 de-palmitoylation is required for the maintenance of inhibitory synapses and GAD65 pool.
Cell adhesion molecules are involved in organizing the presynaptic nanoarchitecture required for highly regulated neurotransmitter release (Tang et al., 2016). NEGR1 is attached to the PM via a GPI anchor and accumulates in lipid microdomains, where palmitoylated proteins, including GAD65 (Kanaani et al., 2002) are also targeted. Trans-synaptic adhesive bonds formed by NEGR1 can promote the synaptic recruitment of GAD65 by limiting diffusion of lipid microdomains that GAD65 binds to. This scenario is supported by our observations showing that the synaptic clustering of lipid rafts is reduced in NEGR1-/-neurons. NEGR1-dependent synaptic clustering of lipid rafts can also constrain the retrieval of synaptic vesicle membranes, since synaptic vesicle proteins interact with PM-localized cholesterol (Thiele et al., 2000). Clustering of lipid rafts by GPI-anchored proteins triggers the assembly of the submembrane spectrin cytoskeleton (Leshchyns'ka et al., 2003) which also inhibits the retrieval of vesicle membranes (Puchkov et al., 2011). Our experiments showing that inhibition of synaptic endocytosis triggers clustering of GAD65 at the synaptic PM indicate that GAD65 is removed from the PM with membranes of reformed synaptic vesicles. An NEGR1-dependent restraint of vesicle reformation from the PM may be necessary for "loading" of GAD65 on the membranes of reforming synaptic vesicles, which may be facilitated within the cholesterol-enriched environment required for synaptic vesicle reformation (Martin, 2000), created by interactions between cholesterol enriched NEGR1-containing lipid microdomains and cholesterol-enriched synaptic vesicle membranes (Fig.   11). Palmitoylation plays a critical role in synaptic sorting of GAD65 (Kanaani et al., 2004;Kanaani et al., 2008). The GAD65-palmitoylating enzyme huntingtin-interacting protein HIP14 (Huang et al., 2004) is found at the neuronal PM and in Golgi and sorting/recycling and late endosomal structures in neurons (Huang et al., 2004), but is not detectable in synaptic vesicle preparations (Takamori et al., 2006). Slower reformation of vesicles may therefore facilitate the HIP14-mediated palmitoylationdependent attachment of GAD65 to the membranes of vesicles during their retrieval from the PM.
We demonstrate that NEGR1 is highly expressed in NPY positive hypothalamic neurons and accumulates in the inhibitory GABAergic synapses formed by these neurons in culture. NPY/AgRPexpressing neurons play a prominent role in promoting food intake by inhibiting anorexigenic proopiomelanocortin (POMC) cells (Dietrich and Horvath, 2013;Suyama and Yada, 2018). We show that NEGR1 deficiency causes a reduction in numbers of inhibitory synapses in the ARC, where NPY/AgRP positive neurons form synapses on the POMC neurons (Dietrich and Horvath, 2013;Suyama and Yada, 2018). This synapse loss can be caused by impaired synaptic targeting of GAD65 in NEGR1-/-mice, because GAD65 deficiency causes a reduction in the size of the synaptic vesicle pool (Tian et al., 1999). We cannot exclude, however, that NEGR1-mediated adhesive bonds are also 19 required for formation or maintenance of the GABAergic synapses. It is noteworthy that NEGR1 deficiency causes a reduction in the density of spines in cortical neurons (Pischedda et al., 2014;Szczurkowska et al., 2018). In cultured hippocampal neurons, NEGR1 expression peaks at 14 days in vitro, and NEGR1 overexpression in twenty-one-day-old hippocampal neurons, i.e., when the levels of endogenous NEGR1 decline, results in an increase in synapse density (Hashimoto et al., 2008), also suggesting that NEGR1 promotes synapse stabilization. A reduction in the efficiency of the inhibitory synaptic input into the anorexigenic POMC cells could cause a reduction in food intake observed in mice with inactivated NEGR1  and could also cause a reduction in body mass , observed also in our colony: at 2-3 months of age, mean ± SEM: 26.38 ± 0.54 g NEGR1+/+ males vs 23.63 ± 0.71 NEGR1-/-males, *p = 0.006 Mann-Whitney test; 20.43 ± 0.27 g NEGR1+/+ females vs 19.02 ± 0.39 NEGR1-/-females, *p = 0.004 Mann-Whitney test). A similar mechanism may also contribute to a reduction in the body mass index in humans with mutations affecting NEGR1 expression (Antunez-Ortiz et al., 2017).
NPY/AgRP neurons also play a key role in coordinating the activity of hypothalamic hunger circuits with the activity of midbrain reward circuits (Alhadeff et al., 2019). Our results, demonstrating that NEGR1 deficiency results in blunting of the reinforcing effects of food, indicate that NEGR1 also functions in the brain reward pathways. Dysregulated brain reward pathways may be contributing to increased intake of palatable foods and ultimately obesity (Berthoud et al., 2011). Mutations in noncoding regulatory elements upstream of NEGR1 containing binding sites for transcriptional repressors are found in people with severe early-onset obesity (Wheeler et al., 2013;Willer et al., 2009), suggesting that NEGR1 overexpression causes changes in feeding behavior, which ultimately lead to obesity. We report that the levels of NEGR1 and its synaptic enrichment are increased in brains of mice with high fat diet induced obesity confirming a previous study, which found increased NEGR1 mRNA levels in the hypothalamus of 15-week-old female mice fed a high fat diet (Lee et al., 2010).
High fat diet induces the loss of synapses in multiple brain regions including the hypothalamus 20 (Dietrich and Horvath, 2013;Horvath et al., 2010;Lizarbe et al., 2018), hippocampus (Lizarbe et al., 2018;Stranahan et al., 2008;Valladolid-Acebes et al., 2012), and cortex (Bocarsly et al., 2015;Lizarbe et al., 2018). Overexpression of NEGR1 in fourteen-day old hippocampal neurons, i.e. at the peak of endogenous NEGR1 expression in these neurons, leads to a reduction in synapse density (Hashimoto et al., 2008). We demonstrate that NEGR1 overexpression causes a reduction in the density of synapses formed on dendrites and axons of NEGR1-overexpressing hypothalamic neurons, indicating that NEGR1 overexpression can directly contribute to the high fat diet induced synapse loss, including the loss of inhibitory synapses reported previously (Valladolid-Acebes et al., 2012). NEGR1 overexpression causes a strong inhibition of synaptic vesicle membrane retrieval and can therefore affect synapse formation or maintenance by inhibiting synaptic vesicle biogenesis, which is required for synapse formation during neuronal development (Hannah et al., 1999) and maintenance of neurotransmitter release in mature synapses (Milosevic, 2018). Similarly, the loss of inhibitory synapses was found in dynamin 1, 3 double knock-out mice (Raimondi et al., 2011). Furthermore, we demonstrate that the high fat diet and trans-interactions of NEGR1 cause similar reductions in synaptic GABA concentrations and induce similar increases of GAD65 levels at the PM, suggesting that the high fat diet caused reduction in the concentration of GABA, observed at least in some brain regions including the frontal cortex and hippocampus of rats (Sandoval-Salazar et al., 2016) and the prefrontal cortex of mice (Hassan et al., 2018), can also be associated with NEGR1 overexpression.
NEGR1 downregulation in mice also results in impaired core behaviors related to autism spectrum disorders (Szczurkowska et al., 2018). In humans, deletions of the NEGR1 gene have been found in patients with developmental co-ordination disorder, attention deficit / hyperactivity disorder, learning disability, delayed speech and language development, and dyslexia (Genovese et al., 2015;Tassano et al., 2015;Veerappa et al., 2013). The results of our work suggest that NEGR1-dependent dysregulation of the GABAergic system can also be a factor contributing to the development of these disorders.

22
DNA coding for full length HA-tagged human NEGR1 (NM_173808.2) was synthesized using the GeneArt® Gene Synthesis service (Life Technologies) and subcloned into the pcDNA3 vector.

Mice
C57Bl/6NJ mice for high fat diet experiments were obtained from the Australian Phenomics Network (APN) based at the Monash University, Melbourne, Australia, and housed at 22 ± 1 °C with a controlled 12-h light/dark cycle and had ad libitum access to water. Mice (9-11-week-old) were fed with either chow (8% calories from fat) or high-fat diets (HFD) (45% calories from fat) (Turner et al., 2009)

Analysis of the GABA levels in synaptosomes
Serially diluted GABA standards (0 to 5 µM) were used to construct a standard calibration curve.
All standards and samples contained a fixed amount of deuterium-labeled GABA internal standard (d6_GABA) to correct for variations during extraction and instrumental analysis. Synaptosomes were pelleted down by centrifuging them at 100000 g for 45 min at +4ºC. The pellets were immediately frozen and stored at -80ºC. Synaptosomes were extracted by adding 50 µl of water to tubes with synaptosomes followed by brief sonication and addition of 100 µl acetonitrile with vigorous vortexing.
Internal standard (100 µl in acetonitrile) was added and the tubes were vortexed and incubated at -20°C for 10 min. Tubes were centrifuged at 13000 rpm for 10 min and pellets containing proteins, buffer salts and sugars were discarded. Supernatants were dried, reconstituted in 100 µl acetonitrile: water (4:1) and filtered through 4 mm x 0.2 µm disks (Phenomenex, Australia) into LC vials. For highly specific and sensitive measurement of GABA, analysis was carried out on a LC-MS/MS system consisting of Accela Open AS autosampler, Accela UPLC pump coupled to a Quantum Access triple quadrupole mass spectrometer equipped with a heated electrospray (HESI) probe (Thermo Scientific, San Jose, CA, USA). The capillary and spray temperatures were both set to 300°C and the electrospray capillary voltage to 4 kV. The dwell time for each transition was set to 250 ms. Argon was used as the collision gas at a pressure of 1 torr. Standards and synaptosome extracts (20 µl each) were injected onto a Luna NH₂ column (3 µm, 2.0 mm × 150 mm) (Phenomenex, Australia) maintained at room temperature. The flow rate was set at 300 µl / min. Mobile phase A was a mixture of 10% aqueous 5 mM ammonium acetate at pH 9 and 90% neat acetonitrile. Mobile phase B was 100% aqueous 5 mM ammonium acetate at pH 9. A gradient program was used for the optimum retention of GABA followed ECD with FAO3DS column) prior to electrochemical detection (Zandy et al., 2017). Serial dilutions of GABA and glutamate were analyzed to generate a linear calibration curve using area under the curve analysis from chromatograms.

Isolation of synaptic PM
Synaptic PM was isolated from synaptosomes as described (Leshchyns'ka et al., 2006). Unless otherwise stated, all steps were performed at 4ºC. Briefly, synaptosomes were lysed by diluting them in 9 volumes of ice-cold H2O and then immediately adjusted by 1 M HEPES, pH 7.4 to a final 27 concentration of 7.5 mM HEPES. After incubation on ice for 30 min, the mixture was centrifuged at 100000 g for 20 min and the pellet containing synaptic PM was collected.

SDS-PAGE electrophoresis and Western blot analysis
Proteins were separated on a 4-12% Bis-Tris bolt mini gel (Life Technologies) and electroblotted

Culture, transfection and treatment of primary hypothalamic neurons
Primary hypothalamic neurons were prepared by dissecting hypothalami from brains of 1-3-dayold mice and disassociating neurons as described (Sheng et al., 2015;Sheng et al., 2019). Neurons were maintained in Neurobasal A medium supplemented with 2% B-27, GlutaMAX and FGF-2 (2 ng/ml) (all reagents from Thermo Fisher Scientific) in 24-well plates for 2 weeks on coverslips coated with poly-D-lysine (100 µg/ml). Neurons were transfected before plating by electroporation using a Neon transfection system (Thermo Fisher Scientific). Alternatively, neurons were transfected using the calcium phosphate method essentially as described (Jiang and Chen, 2006 or dynasore (80 µM prepared using a stock solution in DMSO) were applied to neurons in the culture medium. Control neurons were treated with the culture medium containing the same concentration of the vehicle used to prepare the stock solutions.

Labelling of live hypothalamic neurons with antibodies against the lumenal domain of VGAT
Neurons were pre-incubated for 10 min with sNEGR1 (10 µg/ml) in the culture medium or mock treated with the culture medium at 37°C and 5% CO2. Rabbit polyclonal antibodies against the lumenal domain of VGAT were then added to live neurons and incubated for 30 min at 37°C and 5% CO2.

Immunofluorescence labelling and analysis of fixed neurons
The labelling was performed essentially as described previously (Bliim et al., 2019;Sytnyk et al., 2002). intensities of GAD65 in synaptophysin accumulations, they were outlined using a threshold function of ImageJ, and GAD65 labelling intensities were measured within the outlines. To calculate percentages of synaptic and non-synaptic GAD65 clusters, all GAD65 accumulations were outlined using a threshold function in ImageJ, and the presence of synaptophysin labelling within the outlines was then manually analyzed.

Analysis of the activity-dependent FM4-64 dye uptake and release in neurons
The analysis was performed essentially as described previously (Andreyeva et al., 2010;Leshchyns'ka et al., 2006;Shetty et al., 2013). When indicated, neurons were incubated for 10 min at room temperature with rabbit polyclonal antibodies against NEGR1 or non-immune rabbit immunoglobulins diluted in 4 mM K + buffer ( Sigma) supplemented with 5% foetal bovine serum (Sigma)) in an incubator at 37°C and 5% CO2.
Cells were transfected using calcium phosphate method. Thirty min prior to transfection, growth medium was replaced with DMEM/F12. For each coverslip, 1 µg of plasmid DNA and 3.1 µl of 2 M CaCl2 were mixed with water to a final volume of 25 µl. DNA-Ca 2+ -phosphate precipitate was prepared by adding DNA/CaCl2 solution to 25 µl of 2x HEPES-buffered saline (HBS, 280 mM NaCl, 1.5 mM Na2HPO4, 50 mM HEPES, pH 7.06) (1/8 th at a time and mixing briefly between each addition) and incubating for 10 min at 37°C. The resulting suspension was applied dropwise to the coverslips and incubated at 37°C and 5% CO2 for 3 h. Cells were then treated with 15% glycerol in DMEM/F12 for 1-2 min in the incubator, washed with PBS, placed in growth medium, and maintained for 24 h at 37°C and 5% CO2, and then used for analysis. When indicated, CHO cells were loaded with FM4-64FX dye (15 μM, Thermo Fisher Scientific) applied for 15 min at 37°C and 5% CO2 in 4 mM K + buffer, washed 3 times with PBS, fixed and embedded in the ProLong Gold Antifade mounting medium (Thermo Fisher Scientific).

Operant Conditioning Task
Prior to start of the experiment, mice were put on diet restriction for 7-15 days to achieve the weight loss to 80-85% of their original weight. They were kept on diet restriction throughout the experiment to maintain the weight of mice at 80-85% of their original weight. During the experiments, the mice were placed into an operant chamber containing two levers. The left "active" lever was connected to sucrose pellet dispenser releasing sucrose pellets into a food-magazine. The right lever was inactive. An operant conditioning test was conducted in three stages consisting of magazine training, fixed ratio training and progressive ratio test. The magazine training lasted for 30 min during which palatable sucrose pellets were automatically delivered from the food-dispenser to the foodmagazine at a rate of 1 pellet / min. The numbers of times mice entered the magazine and numbers of pellets eaten by mice were counted. The magazine training was conducted for 6 to 10 days until the mouse left 13 or less pellets in the magazine by the end of the training session. The mice were then trained to press the lever in the fixed ratio training. During this training, active lever press but not inactive lever press resulted in the delivery of a single pellet into the food magazine. Each fixed ratio training session lasted for ≤ 30 min or until 30 pellets were delivered. Fixed ratio training was performed for ten days until the mouse left <15 pellets in the food-magazine by the end of the session.
The mice were then subjected to the progressive ratio test essentially as described (June and Gilpin, 2010) during which the number of active lever presses was progressively increased to activate the release of the pellet. The number of presses required to release each consecutive pellet was calculated as described by Richardson and Roberts (1996) (Sharma et al., 2012) using the following formula: [5e ] -5, where R denotes the number of pellets rewarded already plus 1. Each PR session lasted for 1 h or until the break point at which mice refused to press the lever the number of times required to obtain the next pellet. The total number of times the active lever was pressed and the number of magazine entries were also counted as described (June and Gilpin, 2010). The progressive ratio sessions were performed on three consecutive days.

Immunohistochemical analysis
Mice were deeply anaesthetized with intraperitoneally injected pentobarbitone sodium (100 mg / kg) and transcardially perfused with ice-cold PBS for 2 min followed by ice-cold 4%             (A) Mean ± SEM percentages of pellets eaten over 30 min during magazine training sessions. One pellet was released into the magazine each minute.
(B) Mean ± SEM numbers of active lever presses and pellets eaten during the fixed ratio task. Each active lever press led to the release of a pellet.
(C) Individual progressive ratio break point values measured over three consecutive days in the progressive ratio task. Lines connect data for individual mice. Graphs showing data normalized to the values on day 1 set to 100% are included to illustrate an increase in the progressive ratio break point in NEGR1+/+ mice.
(D) Mean ± SEM of the normalized values shown in C.
(E) Mean ± SEM numbers of pellets delivered and active lever presses in the progressive ratio task.
The numbers of active lever presses were normalized to the number on day 1 set to 100%.
In A-E, n = 9 NEGR1+/+, 8 NEGR1+/-and 12 NEGR1-/-mice were analyzed. *p, repeated measures two-way ANOVA and Tukey's multiple comparisons test. Graphs show levels of GAD65 in brain homogenates and synaptosomes of HFD fed mice relative to the levels in chow fed mice set to 100%. n = 3 mice per group.
(H) Axons and dendrites of NEGR1+/+ cultured hypothalamic neurons co-transfected with cherry and control pcDNA3 vector or NEGR1. Neurons were immunolabelled for synaptophysin and NEGR1.
In  Contains uncropped Western blot source data for Figure 10. (A) Synaptic vesicle localized GAD65 uses glutamate to synthesize GABA, which is transported to the vesicle lumen by VGAT (a). GAD65 is de-palmitoylated and removed to the cytosol from refilled vesicles (b), which then fuse with the PM and release GABA. GAD65 released to the cytosol can travel to the Golgi in the neuronal cell body (red arrow, (Kanaani et al., 2008)) or attach to the PM in the synapse (c). Palmitoylation of GAD65 at the PM targets it to the NEGR1-containing cholesterolenriched microdomains, which are anchored in synapses by NEGR1-containing adhesive bonds (d).
NEGR1-dependent synaptic clustering of lipid rafts promotes 'loading' of palmitoylated GAD65 on the newly formed synaptic vesicles (e). This loading is facilitated by interactions between NEGR1 containing lipid rafts and components of synaptic vesicles constraining the retrieval of synaptic vesicle membranes. Synaptic recycling of vesicles and GAD65 is shown with black and pink arrows, respectively.
(C) In NEGR1-/-neurons, synaptic clustering of lipid rafts is reduced, synaptic targeting of GAD65 is inhibited, and non-synaptic GAD65 clusters are formed.