SGIP1 binding to the α-helical H9 domain of cannabinoid receptor 1 promotes axonal surface expression

Endocannabinoid signalling mediated by cannabinoid type 1 receptors (CB1Rs) is critical for homeostatic neuromodulation of both excitatory and inhibitory synapses. This requires highly polarised axonal surface expression of CB1R, but how this is achieved remains unclear. We previously reported that the H9 domain in the intracellular C-terminus of CB1R contributes to polarised surface expression by an unknown mechanism. Here we show the H9 domain binds to the endocytic adaptor protein SGIP1 to promote CB1R expression in the axonal membrane. Overexpression of SGIP1 increases CB1R axonal surface localisation but has no effect on CB1R lacking the H9 domain (CB1RΔH9). Conversely, SGIP1 knockdown reduces axonal surface expression of CB1R but does not affect CB1RΔH9. Furthermore, SGIP1 knockdown diminishes CB1R-mediated inhibition of presynaptic Ca2+ influx in response to neuronal activity. Together, these data advance mechanistic understanding of endocannabinoid signalling by demonstrating that SGIP1 interaction with H9 underpins axonal CB1R surface expression to regulate presynaptic responsiveness.


Introduction
The endocannabinoid system (ECS) is a negative feedback system that homeostatically controls neurotransmission in the brain. By mediating activity-dependent suppression of presynaptic release, the ECS modulates synaptic strength and plasticity, which are fundamental for many brain processes including cognition, appetite/energy expenditure, and learning and memory (1). Moreover, the ECS plays key roles in attenuating stress-induced glutamate release and is implicated in a wide range of neurological and neurodegenerative diseases (2,3).
Because ECS pharmacology is complex and pleotropic, drugs that act directly on the system often result in unwanted neurological and psychoactive side effects (4). Given these limitations, increased understanding of the biochemistry and cell biology of the ECS could provide new avenues for therapeutic intervention.
We have reported previously that CB1R is preferentially and directly axonally targeted through the secretory pathway and that polarity is maintained, at least in part, by the more rapid endocytosis of CB1Rs from the somatodendritic than the axonal membrane (13).
Furthermore, we showed that the 21-residue putative helical H9 domain in the intracellular C-terminal domain of CB1R (ctCB1R) contributes to the delivery and stabilisation of axonal CB1R (13). However, despite this progress, exactly how H9 promotes the axonal surface distribution of CB1R remains to be determined.
Here we report that SGIP1 binds to the CB1R -helical H9 domain and acts to stabilise CB1R at the presynaptic membrane. We show that overexpression of SGIP1 increases CB1R at the axonal plasma membrane, whereas SGIP1 knockdown phenocopies the decreased surface expression observed upon H9 deletion (CB1R ΔH9 ) and impairs CB1R-mediated modulation of synaptic transmission. These data provide mechanistic understanding of how CB1R polarity is established and maintained by identifying SGIP1 as an important mediator of CB1 axonal surface expression. Moreover, these findings open the possibility that manipulating this interaction could be used to regulate the availability of presynaptic CB1R for potential therapeutic benefit.

Cloning of SGIP1 from rat cortical neuronal cultures.
Deletion of H9 reduces CB1R surface expression and increases CB1R endocytosis in primary neurons (13), whereas co-expression of SGIP1 enhances CB1R surface expression in HEK293 cells (23). Based on these observations we wondered if SGIP1 interacts with H9 to regulate CB1R surface expression. To investigate this possibility, we amplified rat SGIP1 from cDNA derived from mRNA extracted from DIV21 primary cortical neurons and subcloned it into a modified pcDNA3.1 to incorporate an N-terminal FLAG tag.
The isolated sequence corresponded to predicted SGIP1 transcript variant X19 (NCBI Reference Sequence: XM_017593774.2; Figure 1A). This 660-amino acid variant differs from the full-length canonical Uniprot entry (i.e., transcript variant X9 NCBI Reference Sequence: XM_017593764.2) by two deletions: a single residue deletion in the membrane phospholipid binding domain (MP; Q34) and a 165-residue deletion in the proline-rich domain (PRD).
Importantly, this variant, which we refer to as SGIP1, does not contain the additional sequence found in the longer isoform SGIP1α (NCBI Reference sequences NM_001376936.1, transcript variant X1 NCBI Reference Sequence: XM_039109919.1, and X2: NCBI Reference Sequence: XM_039109920.1) that is necessary for membrane tubulation (31). Furthermore, the 99 C-terminal residues D708-N806 in both mouse SGIP1 and rat SGIP1, which have 100% sequence identity and contain the CB1R binding domain (23), are unchanged.
Overexpression of SGIP1 increases surface expression of CB1R WT , but not CB1R ΔH9 .
To determine the role of SGIP1 on CB1R axonal surface expression, we co-transfected DIV12 neurons with full-length EGFP-CB1R WT or EGFP-CB1R ΔH9 , and either SBP control or an SBPtagged SGIP1. Following transfection, neurons were incubated for a further 2 days and then stained for surface CB1R using anti-GFP antibody (Figure 2A). Co-expression of SBP-SGIP1, but not SBP control, significantly increased axonal surface levels of EGFP-CB1R WT (Figure 2B), similar to what occurs in HEK293 cells (23). Importantly, no such increase was observed for EGFP-CB1R ΔH9 , suggesting that H9 is necessary for this effect to occur ( Figure 2B). SGIP1 knockdown reduces surface expression of CB1R WT , but not CB1R ΔH9 .
Next, we transfected DIV9 hippocampal neurons with a scrambled shRNA (SCR29), or an shRNA knockdown construct that targets all known isoforms of SGIP1 (25), and either EGFP-CB1R WT or EGFP-CB1R ΔH9 (Supplementary Figure 1). Following transfection, neurons were incubated for a further 5 days to ensure complete knockdown, and then live stained for surface CB1R using anti-GFP antibody ( Figure 3A). Consistent with a role for SGIP1 in promoting CB1R axonal surface expression, SGIP1 knockdown reduced surface EGFP-CB1R WT in axons to levels equivalent to EGFP-CB1R ΔH9 (Figure 3B). Importantly, SGIP1 knockdown did not further reduce surface expression of EGFP-CB1R ΔH9 ( Figure 3B). These data demonstrate that CB1R ΔH9 is insensitive to regulation by SGIP1 and strongly suggest that the reduced surface expression phenotype of EGFP-CB1R ΔH9 is due to its inability to bind SGIP1.
We, and others, have shown that although CB1R is delivered to dendritic plasma membrane, it is rapidly internalised (6,7,11,13,34). While SGIP1 has been reported to preferentially localise to axons and presynaptic terminals (22,23,35), SBP-SGIP1 appears to be present throughout the neuron. However, while no effect of SBP-SGIP1 overexpression on CB1R surface expression was observed in dendrites ( Supplementary Figure 2A), pan-SGIP1 knockdown reduced dendritic surface levels of CB1R WT , but not CB1R ΔH9 (Supplementary Figure 2B). These results raise the possibility that an isoform other than SGIP1 might affect CB1R surface expression in dendrites. SGIP1 knockdown reduces the surface/total ratio of endogenous CB1R.
To determine how SGIP1 effects surface expression of endogenous CB1R, we transduced DIV7/8 primary cortical neurons with lentivirus expressing either a scrambled shRNA control (SCR29) or SGIP1 shRNA. Surface and total levels of endogenous CB1R were examined at DIV14/15 by surface biotinylation followed by streptavidin pulldown and Western blotting Surprisingly, however, in these experiments, total levels of endogenous CB1R were increased by SGIP1 knockdown compared to the SCR29 control ( Figure 4D). In contrast, neither total ( Figure 4E) nor surface ( Figure 4F) levels of epidermal growth factor receptor (EGFR) were affected by SGIP1 knockdown, suggesting the changes observed could not be attributed to global alterations in membrane protein expression or surface expression. Therefore, to assess if the increase in total CB1R protein levels could be due to increased transcription, we analysed transcript levels of CB1R in DIV14/15 cortical cells transduced with SCR29 or SGIP1 shRNA by RT-qPCR ( Figure 4G-H and Supplementary figure 3). The relative mRNA level of CB1R was significantly increased when SGIP1 was knocked down, suggesting that enhanced CB1R levels likely constitute a homeostatic mechanism in response to reduced surface expression.
As expected, in control neurons the peak Ca 2+ signal decreased after 2-AG incubation by ~45% ( Figure 5B), consistent with the presynaptic inhibitory action of CB1R signalling (37,38). The magnitude of this decrease was markedly reduced in SGIP1 knockdown neurons ( Figure 5B).
Furthermore, 2-AG significantly decreased the baseline Ca 2+ signal compared to pre-2-AG incubation in control cells, but not after SGIP1 knockdown ( Figure 5C). Lastly, the reduction in the area under the curve of the Ca 2+ response was significantly less in SGIP1 knockdown cells compared to control cells ( Figure 5D). Together, these data suggest that SGIP1 knockdown suppresses the endocannabinoid-mediated reduction in Ca 2+ influx, indicative of reduced presynaptic CB1R signalling in the absence of SGIP1.

Discussion
The context of this study was that the amphipathic -helical H9 domain in the intracellular Cterminal of CB1R contributes to the polarised presynaptic surface expression of CB1R (13). SGIP1 binds to CB1R, increasing its surface expression and modulating its signalling in HEK293 cells (23,32). The distributions of SGIP1 and CB1R overlap in mouse brain (39), and they co-localise at the presynapse (23). We therefore hypothesised that SGIP1 might interact with H9 to modulate synaptic CB1R availability at the presynaptic membrane.
We show that SGIP1 binds to the CB1R H9 domain to promote axonal surface expression. A FLAG-tagged isoform of SGIP1, which we refer to as SGIP1, co-IPs with ctCB1R WT , but not with ctCB1R ΔH9 in HEK293T cells (Figure 1A-C). Moreover, overexpression of SGIP1 increases CB1R axonal plasma membrane expression (Figure 2), while knockdown of SGIP1 reduces axonal surface levels ( Figure 3). Importantly, neither overexpression ( voltage-gated Ca 2+ channels via Gi βγ-subunit mobilisation (37,38), but the magnitude of the effect is significantly reduced with SGIP1 knockdown ( Figure 5). Notably, this effect on signalling remains even though total CB1R levels increase with SGIP1 knockdown ( Figure 4D) -a phenomenon that we speculate may represent a homeostatic mechanism in response to decreased CB1R surface expression, since CB1R transcript levels were also increased ( Figure   4G-H). Taken together, these data provide strong evidence that SGIP1 promotes axonal surface localisation of CB1R through interaction with H9.
SGIP1 is a member of the muniscin family of cargo adapters due to its similarity with FCHo1/2 proteins and interacts with AP-2 (40), intersectin (28), and Eps15 (25). Since SGIP1 is a component of the clathrin-mediated endocytosis complex, a key question is why does SGIP1 overexpression enhance, and knockdown reduce, CB1R surface expression?
We speculate that key to untangling this conundrum is the observation that the actions of SGIP1 are highly isoform-dependent (31). Both SGIP1 and FCHo1/2 proteins contain an Nterminal membrane phospholipid (MP) binding domain, an AP-2 activator domain, a proline rich domain, and a C-terminal µ homology domain (28, 40) ( Figure 1A). However, the membrane binding domain of FCHo1/2 is an F-BAR domain that deforms the plasma membrane to facilitate clathrin-coated pit formation. The corresponding region of SGIP1, however, has no F-BAR sequence similarity.
Nonetheless, a recent report has identified a 28-residue, positively charged sequence present in the membrane binding domain that is necessary for homo-oligomerisation of SGIP1 and membrane tubulation (31). Of the 32 different predicted rat transcript variants available on the NCBI database, 13 contain the membrane tubulating sequence in its entirety. Importantly, SGIP1α contains this sequence, while SGIP1 and SGIP1 do not, and previous experiments indicating that SGIP1/CB1R co-expression in HEK293 cells increases CB1R surface levels were performed using the non-membrane tubulating form of SGIP1 (23). Consistent with these findings, our data show that overexpression of SGIP1, which does not contain the tubulating sequence, increases CB1R surface levels in axons (Figure 2).
We note, however, that different SGIP1 isoforms may be selectively recruited to different cargo to mediate opposing effects. Differences between the SGIP1 isoforms are in the homooligomerisation and membrane-tubulating sequences, whereas the C-terminal 99 residue domain of SGIP1 that interacts with CB1R is present in all three SGIP1 isoforms. One possibility could be that SGIP1 and SGIP1 act as endogenous 'dominant negatives' to SGIP1α and FCHo1/2 proteins, preventing them from binding cargo by taking up the binding site.
However, further work will be required examine this possibility directly. In conclusion, our findings indicate that SGIP1 promotes CB1R surface expression via interaction with H9.

Plasmids and reagents
To N-terminally tag rat CB1R, the first 25 N-terminal amino acids were removed and an exogenous signal peptide corresponding to interleukin-2 (SP Il2 ) was added before the tags as previously characterised (8,13).
The following overexpression plasmids were used: The following double overexpression plasmids (one cassette with a CMV promoter, one cassette with an Sffv promotor) were used: The following shRNA knockdown (H1 promoter) + overexpression (Sffv promoter) plasmids were used: The non-targeting 29-mer shRNA (SCR29) sequence was: The SGIP1 shRNA target sequence was: The lentiviral helper vectors p8.91 and pMD2.G were used.
The following primary antibodies were used: The following secondary antibodies were used: All fluorescent secondary antibodies for ICC were used at 1:400 and all HRP secondary antibodies for WB were used at 1:10,000. by centrifugation at 16k RCF for 20 min. A proportion of the lysate was kept aside ("input"), and the rest was incubated with GFP-Trap beads on a rotating wheel for 1 hour. The beads were pelleted for 2 minutes at 1.5k RCF and washed 3x in wash buffer (lysis buffer minus protease inhibitors). 2x Laemmli sample buffer was added to the beads and the inputs and the beads were boiled at 95°C for 5 minutes.

Lentivirus production and transduction of cortical neurons
Lentivirus was produced in HEK293T cells following standard protocols (42). Biotinylated surface proteins were isolated using streptavidin-coated agarose beads (Merck).
The beads were washed 2 times in lysis buffer by centrifugation at low speed (<1.5k RCF). 50 μL of clarified lysate was set aside (total) and 100 µL of clarified lysate was added to 30 µL of beads along with 500 µL of lysis buffer (surface). The beads were incubated on a rotating wheel at 4°C for 1.5 hours, then washed in wash buffer (lysis buffer without protease inhibitors) 3 times, by pelleting the beads for 2 minutes at 1k RCF and discarding the supernatant between washes. 2x Laemmli sample buffer was added to both the surface and total samples. The samples were vortexed, spun down, and incubated overnight at RT (to prevent CB1R aggregation).

Western Blotting
Samples in Laemmli sample buffer were resolved on 10% acrylamide gels by SDS-PAGE, transferred onto methanol-activated PVDF membrane, and immunoblotted according to standard protocols. Chemiluminescence was detected using a LI-COR Odyssey Fc and quantified using Image Studio.

RT-qPCR
RNA was extracted from DIV14/15 cultured cortical cells transduced with lentivirus expressing a 29-mer non-targeting shRNA (SCR29) or SGIP1-targeting shRNA using the Qiagen RNeasy mini kit following manufacturer's instructions. RNA concentration was measured using a nanodrop and 1 µg of RNA was converted to cDNA using the RevertAid First Strand cDNA Synthesis Kit (ThermoFisher Scientific) according to manufacturer's instructions.
qPCR was performed using PowerUp™ SYBR™ Green Master Mix (ThermoFisher Scientific) mixed with 2 µL of each sample and gene-specific primers at 0.25 µM each and run on a qPCR machine with MxPro software and SYBR green with dissociation curve setup.
The following primers were used: Unknown samples were run in triplicate and no-RT (-RT), no template (NTC), and SYBR neg (10 µL SYBR master mix + 10 µL H2O) controls were included with each qPCR.

Relative gene expression was quantified using the ΔΔCt method. Mean cycle threshold (Ct)
values of the gene of interest were normalised first to GAPDH mean Ct values (ΔCt) and then to the SCR29 control (ΔΔCt). Fold change of gene expression was plotted as 2 -ΔΔCt and a one sample t-test was performed to determine whether the SGIP1 KD condition was significantly different from 1.

Live surface staining
To measure surface expression, DIV14 cultured hippocampal neurons grown on 25mm glass coverslips were incubated live in the appropriate antibody raised against an extracellular epitope. Briefly, cells were removed from the incubator and allowed to cool to RT for 5 minutes. Cells were then incubated in chicken anti-GFP antibody (1:1,000) in 90 μL conditioned media for 10 minutes at RT. The antibody mix was dotted onto parafilm, and the coverslips were incubated upside down to ensure even coating. Cells were washed 3 times in 1xPBS to remove excess antibody and fixed.

Fixation and fixed immunostaining
Cells were fixed in pre-warmed 4% PFA and 5% sucrose in 1xPBS for 12 minutes. Following 3 washes in PBS, residual PFA was quenched with a wash in 100mM glycine in 1xPBS, and the cells were washed 3 more times in PBS.
Cells were blocked and permeabilised in 3% BSA in 1xPBS with 0.1% Triton-X for 20 minutes.
Cells were then incubated in secondary antibody to label surface. Cells were then stained for total levels and Ankyrin-G (axon initial segment marker). Primary and secondary antibodies were diluted in 3% BSA in 1xPBS. 90 μL of the antibody mix was dotted onto parafilm and the coverslips were incubated upside down for 1 hour at RT. The cells were washed 3x in PBS between incubations. Coverslips were dipped in distilled H20 and mounted onto slides using Fluoromount G (ThermoFisher Scientific) mounting media with or without DAPI.

Fixed image acquisition and analysis
Images were acquired using a Leica SP8 confocal laser scanning microscope (Wolfson Bioimaging Facility, University of Bristol). All settings were kept the same within experiments.
To avoid bias, neurons were selected for data acquisition based only on their total staining and surface was stained in far red.
Fiji (ImageJ) was used to quantify fluorescence. Images were max projected, and regions of interest (ROIs) of approximately similar lengths were drawn around axons based on the total channel only. Axons were defined as processes whose initial segment was positive for Ankyrin-G. Surface fluorescence was normalised to total fluorescence to control for differences in expression levels and expressed as a percentage of the control condition.

SyGCaMP assay and analysis
SyGCaMP assay was performed as previously described (36). Hippocampal neurons grown on 25mm coverslips were transfected at DIV8/9 with pXlg3-SCR29-SyGCaMP3 or pXlg3-SGIP1 KD-  Area was calculated using Prism with the following parameters: baseline = mean of 8-9s and 18-19s, ignoring peaks less than 10% distance from minimum to maximum Y, ignoring any peaks defined by fewer than 8 adjacent points. Post-2-AG AUC was presented as a percentage of pre-2-AG AUC).

Statistics
All statistics were performed using GraphPad Prism (version 9). Outliers were removed using GraphPad Prism's ROUT method (Q = 1%). T-tests were used to determine statistical significance between two groups. For more than two groups, One-or Two-way ANOVAs with Tukey's or Sidak's post hoc tests were used to determine statistical significance, depending on the comparisons required.
For image analysis, 'n' denotes the total number of neurons that were analysed, as is convention in the field (6-8, 11, 13). However, the number of separate neuronal cultures prepared from litters of pups from separate dams is also noted for each experiment. For For all data, *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001. Data are presented as mean ± SEM.  (B) Representative immunoblots showing that FLAG-SGIP1 co-immunoprecipitates with GFP-ctCB1R WT , but not GFP-ctCB1R ΔH9 or GFP control, in HEK293T cells.

Figures and Legends
(C) Quantification of data represented in (B). Significantly more FLAG-SGIP1 coimmunoprecipitates with GFP-ctCB1R WT than with a GFP control (GFP vs. ctCB1R WT : mean ± SEM, 4.04 ± 1.95 vs. 100 ± 3.23; ****p < 0.0001) or with GFP-ctCB1R ΔH9 (ctCB1R WT vs.       130kDa (red arrow) that was reduced in intensity in SGIP1 shRNA samples compared to the 29mer scrambled control (C-D), suggesting this band is SGIP1 and/or SGIP1α. Biorbyt orb40255 also recognises a non-specific smear from around 100kDa to 120kDa and a band around 50kDa. Sigma HPA017963 also faintly recognised a band around 50kDa. This could represent a non-specific band or possibly a short isoform of SGIP1 or closely related protein that is insensitive to our shRNA.
SGIP1 was cloned out of our cortical neurons, suggesting that it may be one of the principal isoforms found in these cells, and FLAG-SGIP1 runs around 100kDa (see E). Sigma HPA017963 does not recognise a band at that molecular weight. However, this is likely because the antigen used to generate this antibody (residues K254-D345) almost completely overlaps with the deletion found in SGIP1 (N273-E438). The epitope used to generate Biorbyt orb40255 (M1-R30) is present in both SGIP1 and SGIP1, but the non-specific smear from 100kDa to 120kDa likely obscures this band.
For clarity, all experiments in Fig. 3 used the Sigma HPA017983 as it produced fewer nonspecific bands, with the assumption that all isoforms of SGIP1 would be knocked down in a similar ratio.
Blots were stripped and reprobed for GAPDH as a loading control.  Fig. 2A. Overexpression of SGIP1 isoform has no effect on dendritic surface levels of either CB1R WT or CB1R ΔH9 . Surface fluorescence was normalised to total fluorescence and shown as a percentage of axonal CB1R WT /SBP control (see Fig. 3B). Two-way ANOVA with Sidak's post hoc test; n = 19-27 neurons per condition from four independent neuronal cultures.
(B) Quantification of data represented in Fig. 3A. Knockdown of SGIP1 reduces dendritic surface levels of CB1R WT , but not CB1R ΔH9 , suggesting that an isoform other than SGIP1 may affect CB1R surface expression in dendrites. Surface fluorescence was normalised to total fluorescence and shown as a percentage of axonal CB1R WT /SCR29 (see Specificity of CB1R qPCR primer set used in Fig. 4G was tested against DIV14 cortical neuronal samples lentivirally transduced with CB1R shRNA (target sequence from (43) or a 19mer scrambled control (target sequence from (44)). There was a 97% reduction in transcript levels in CB1R KD neurons compared to SCR19 control, suggesting that the primers used are specific for CB1R mRNA.