SorCS2 dynamically interacts with TrkB and GluN2B to control neurotransmission and Huntington’s disease progression

Huntington’s disease (HD) is a fatal neurodegenerative disorder characterized by progressive motor dysfunction and loss of medium spiny neurons (MSNs) in dorsal striatum. Brain-derived neurotrophic factor (BDNF) sustains functionality and integrity of MSNs, and thus reduced BDNF signaling is integral to the disease. Here we show that SorCS2 is expressed in MSNs with reduced expression in R6/1 HD model, and that SorCS2 deficiency exacerbates the disease progression in R6/1 mice. Furthermore, we find that SorCS2 binds TrkB and the NMDA receptor subunit GluN2B, which is required to control neurotransmission in corticostriatal synapses. While BDNF stimulates SorCS2-TrkB complex formation to enable TrkB signaling, it disengages SorCS2 from GluN2B, leading to enrichment of the subunit at postsynaptic densities. Consequently, long-term potentiation (LTP) is abolished in SorCS2 deficient mice, despite increased striatal TrkB and unaltered BDNF expression. In contrast, the addition of exogenous BDNF rescues the phenotype. Finally, GluN2B, but not GluN2A, currents are also severely impaired in the SorCS2 KO mice. To conclude, we uncovered that SorCS2 dynamically targets TrkB and GluN2B to orchestrate BDNF-dependent plasticity in MSNs of dorsal striatum. We propose that SorCS2 deficiency impairs MSN function thereby increasing neuronal vulnerability and accelerating the motor deficits in Huntington’s disease.


INTRODUCTION
Huntington's disease (HD) is a devastating monogenic neurodegenerative disease characterized by advancing motor dysfunction, cognitive impairments, and psychiatric symptoms, which ultimately lead to the death of the patient (1,2). HD symptoms are caused by progressive neurodegeneration of GABAergic medium spiny neurons (MSNs) in dorsal striatum, which is followed by striatal and cortical atrophy (3,4).
Genetically, HD patients carry an unstable CAG repeat expansion in exon 1 of the gene encoding Huntingtin protein (HTT) which translates into an abnormal N-terminal polyglutamine tract resulting in mutated Huntingtin protein (mHTT). In the healthy organism, HTT plays many important roles in neurogenesis, intracellular protein trafficking, and synapse morphology and plasticity (1,2). Accordingly, the accumulation of cytotoxic mHTT results in several deleterious effects such as reduced trophic stimulation and perturbed synaptic plasticity, as well as diminished synaptic delivery of N-methyl-Daspartate (NMDA) receptors favoring their extrasynaptic accumulation. These abnormalities impair local neurotransmission while at the same time predispose MSNs to glutamate-induced excitotoxicity (2,(5)(6)(7)(8)(9)(10)(11).
Reduced levels of brain-derived neurotrophic factor (BDNF) in dorsal striatum and its impaired downstream signalling have been considered central to the altered plasticity and vulnerability of MSNs in HD (12,13). Through the binding to the receptor Tropomyosin related kinase receptor B (TrkB), BDNF triggers TrkB phosphorylation and activates manifold of intercellular signaling cascades that control diverse biological functions. In the mature nervous system, TrkB activation regulates neuronal survival and synaptic plasticity such as consolidation of long-term synaptic potentiation (LTP) (14)(15)(16). For the LTP to occur, TrkB must phosphorylate the NMDA receptor subunits GluN2B to mediate GluN2B translocation into the synapse, and to increase the proportion of NMDA receptors that contain this subunit (17)(18)(19).
Unlike neurons in many other brain areas, striatal MSNs do not produce BDNF (20,21) but instead depend on the BDNF release from their corticostriatal afferents (12,22). In mouse models of HD, reduced levels of BDNF (23) and/or TrkB (24,25) accelerate the behavioral and pathological changes, whereas cortical overexpression of BDNF ameliorates the symptoms (26,27). The impaired BDNF-TrkB signalling has been mostly attributed to the reduced delivery of BDNF to the striatum due to transcriptional repression of cortical BDNF expression (5,12), and dysfunctional anterograde axonal transport to the striatum (6). More recent studies reported that in some HD mouse models, TrkB (28) and BDNF (8) levels are normal in young symptomatic mice, but that TrkB does not properly activate the signalling pathways controlling potentiation at corticostriatal synapses. Hence, the abnormal neurotransmission is probably important at the earlier stages of HD progression, while later on, reduced trophic signalling renders the neurons vulnerable, e.g. to excitotoxicity from increased extrasynaptic glutamate stimulation.
The primary determinant of HD onset is the length of the CAG repeats on the expanded allele. Yet, this only accounts for 50-70% variations at the age of onset. These observations suggest the existence of genetic modifiers, which remain largely unknown (1,2,11,(29)(30)(31). A recent genome wide association study of human and rat HD samples identified several single nucleotide polymorphisms in the SORCS2 gene (32). SorCS2 is a member of the Vps10p-domain receptor family, so-called sortilins that are enriched in distinct neuronal populations. Receptors of this family regulate several aspects of neuronal survival, neurodevelopment, synapse remodelling, and synaptic plasticity. When dysfunctional, they are often causative to psychiatric and neurodegenerative disorders (33)(34)(35)(36). Conceptually, these receptors function by a dual mode of action; they can mediate intracellular sorting of ligands and co-receptors, and they can form complexes with signalling receptors at the plasma membrane to control their activity (36).
We recently discovered that SorCS2 enables BDNF/TrkB-dependent plasticity in hippocampus (33). We thus hypothesized that SorCS2 modulates BDNF/TrkB signalling in the dorsal striatum, and when absent could aggravate HD-related symptoms in R6/1 mice. Indeed, we found that SorCS2 deficiency severely worsens the motor symptoms present in the HD mice. Here we present a new molecular mechanism of BDNF-TrkB-SorCS2 signaling axis, and its implication in MSN plasticity, and HD progression. We show that SorCS2 enables TrkB activation and the synaptic translocation of GluN2B to MSNs synapses. While BDNF stimulates the formation of a complex between SorCS2 and TrkB to boost BDNF-TrkB signalling capacity, it uncouples SorCS2 from GluN2B resulting in enrichment of the neurotransmitter receptor in postsynaptic densities. Given that symptomatic HD mice have reduced SorCS2 levels, the attenuated activity of the receptor may be fundamental to the perturbations in synaptic plasticity that typifies HD.

SorCS2 is selectively expressed by the MSNs in dorsal striatum
SorCS2 is typically expressed by neurons in the adult mouse brain, with the highest expression in dentate gyrus of the hippocampus and piriform cortex (33,37). To explore SorCS2 expression profile in corticostriatal circuits, we first took advantage of the available in situ hybridization data (IHC) from Allen brain atlas. We observed that besides the piriform cortex, SorCS2 is highly expressed in dorsal striatum (caudoputamen), cortical layers II and III (including motor cortex and anterior cingulate cortex), induseum griseum, and intermediate lateral septal nucleus (Figure 1A-C). To determine which cell types express SorCS2 in dorsal striatum, we performed immunofluorescence using a number of known cell type-specific markers. Age-matched SorCS2 knockout (KO) brain tissue served as negative control. We found that  Figure S1A-D). As previously reported for the hippocampus, SorCS2 was absent in Glial fibrillary acidic protein positive (GFAP+) glia cells in dorsal striatum (Suppl. Figure S2). Our data suggest that SorCS2 may be involved in the biology of MSNs in the dorsal striatum.
SorCS2 deficiency accelerates HD progression in R6/1 mice Since MSNs in dorsal striatum control motor skills, we tested whether SorCS2 dysfunction worsens HDrelated motor symptoms. To address this, we generated double transgenic line by crossing SorCS2 KO mice into the R6/1 transgenic HD model, which carries a human 115 CAG repeat in exon 1. First, we subjected 12 weeks old males to rotarod behavior testing to study motor skill learning. Mice were placed on the accelerating rotarod without any prior training and received four trials per day for three consecutive days. Each mouse had a single attempt on the rotarod for each trial and we recorded the latency to fall (Figure 2A) To study the progression of HD, we next monitored the mice from 7 weeks to 22 weeks of age with respect to body weight, grip strength, and clasping. Both, SorCS2 KO and R6/1 males had lower body weight than WT (p<0.0001 and p=0.0006, respectively), which was further reduced in SorCS2 KO; R6/1 mice when compared to R6/1 (p=<0.0001) and SorCS2 KO (p=0.0028) ( Figure 2B) Figure S3A). Grip strength was tested by letting the mouse grip a horizontal metal bar connected to a newton meter, and pulling the mouse away horizontally by the tail. Grip strength was diminished in R6/1 mice in both males (p=<0.0001) ( Figure 2C) and females (p=0.0009) (Suppl. Figure S3B). However, SorCS2 KO mice did not stand out from WT, and neither did receptor deficiency aggravate the phenotype of R6/1 in this paradigm. We next tested for clasping, which is a marker of neurodegeneration commonly used to study disease progression in HD mice (40). The mice may initially present a dystonic posture that, as the disease builds up, eventually develops into overt clasping of the hind limbs when lifted by the tail. SorCS2 KO displayed minor clasping when compared to WT in both males (p=0.0059) ( Figure 2D) and females (p<0.0001) (Suppl. Figure S3C) To characterize motor function in older mice, we tested 22 weeks old males in an in-house-built catwalk apparatus to evaluate gait performance. The assay enabled us to distinguish single steps of a walking mouse, and measure the stride length and the base width of the gait (Figure 2E, Suppl. Figure S3D). The gait was recorded, with the representative videos available for WT (Suppl. Video SV1), SorCS2 KO (Suppl. Video SV2), R6/1 (Suppl. Video SV3), and SorCS2 KO; R6/1 (Suppl. Video SV4). The double transgenic line showed impaired gait pattern through the corridor (Figure 2E, Suppl. Figure S3E). Gait analysis revealed that SorCS2KO; R6/1 mice have significantly reduced front limb stride length compared to the R6/1 (p=0.0043), while the single transgenic lines performed as WT ( Figure 2F). Hind limb stride length was also significantly reduced in SorCS2 KO; R6/1 mice compared to R6/1 mice (p=0.0038) (Figure 2G), and hind limb base width was increased only in SorCS2KO; R6/1 compared to the other genotypes; to WT (p=0.0009), SorCS2 KO (p=0.0141), and R6/1 (p=0.0133) ( Figure 2H). These data indicate that the impaired gait is caused by cumulative effect of SorCS2 loss of function in the HD model. Finally, we subjected 22 weeks old males to the open field test. In this test, R6/1 as well as SorCS2KO; R6/1 mice moved less than WT (p=0.0061 and p<0.0001, respectively) and with significant worsening of the activity in SorCS2 KO; R6/1 compared to the R6/1 mice (p=0.0082) (Figure 2I). Taken together, our data demonstrate that SorCS2 loss of function accelerates and worsens some of the HD-related motor symptoms in R6/1 mice.

Striatal expression of SorCS2 and DARPP-32 during HD progression
To investigate whether HD progression affects SorCS2 expression (similarly to TrkB or BDNF), we examined RNA and protein levels in our transgenic mice using the real-time quantitative PCR (qPCR) and western blot analyses. While SorCS2 expression in the dorsal striatum of 6 weeks old asymptomatic R6/1 mice did not differ from WT mice (Suppl. Figure S4A-B), it was reduced by 50% both at protein (p=0.0001) and RNA levels (p=0.0019) in 22 weeks old symptomatic animals ( Figure 3A-C). This difference was tissue specific as SorCS2 expression in cortex was unaltered (Suppl. Figure S4D-F). Hence, a reduction of striatal SorCS2 expression may accelerate and exacerbate the HD progression. A decline in DARPP-32 is known to reflect the neurodegenerative process of MSNs in several HD mouse models (23,41). We found that DARPP-32 expression in 6 weeks old asymptomatic R6/1 mice was normal (Suppl. Figure S4A, C). However, at 22 weeks of age, DARPP-32 was significantly reduced in both R6/1 (p=0.0082) and SorCS2 KO; R6/1 mice (p=0.0044). DARPP-32 was unaltered in Sorcs2 -/mice of similar age (Figure 3D, 3G). The data suggest that a secondary reduction in SorCS2 expression may fuel the HD progression, possibly by altering MSNs function rather than by affecting their degeneration.

SorCS2 is required for robust TrkB signaling in dorsal striatum
Reduced BDNF signaling is fundamental to the neurodegenerative process and altered neurotransmission of MSNs. In some HD mouse models, the cortical transcription of BDNF, its anterograde transport and release from cortical terminals in the dorsal striatum, as well as the TrkB expression are suppressed (6,7,12,13,23,42). In other cases, BDNF and TrkB expression is preserved but TrkB signaling is compromised (8,28,43). We therefore quantified BDNF and TrkB levels in corticostriatal circuits to study the impact of SorCS2 on TrkB signaling. Despite severe motor symptoms in the R6/1 and SorCS2 KO; R6/1 mice, BDNF levels were not reduced in cortex nor in dorsal striatum, indicating that the production and trafficking of BDNF is unaffected by mHTT and SorCS2 (Suppl. Figure S4G-J). However, we observed an increase in BDNF levels in dorsal striatum of SorCS2 KO; R6/1 when compared to SorCS2 KO (Suppl. Figure S4H) which might indicate a compensation in BDNF secretion to overcome the neurodegeneration. There was also no difference in BDNF maturation, as we observed no shifts in the ratio between proBDNF and BDNF levels in dorsal striatum (Suppl. Figure S4I-J). Expression of the full-length TrkB receptor was slightly increased in the SorCS2 KO mice (p=0.0482) but this difference disappeared on the R6/1 background (Figure 3D-E).
The expression of the truncated TrkB variant (named TrkB-T.1) was independent of genotype ( Figure 3D, F). We next investigated whether SorCS2 may physically interact with TrkB to control its signalling abilities.
To this end, TrkB was immunoprecipitated from dorsal striatum and the precipitate probed for SorCS2.
Indeed, we found that the two receptors co-immunoprecipitated demonstrating that SorCS2 and TrkB can physically interact to form a heteromeric complex ( Figure 4A). Strikingly, in rat primary hippocampal cultures, this interaction was increased by approximately 200% when treated with 10 ng/ml BDNF ( Figure   4B). We then asked if SorCS2 enables BDNF-TrkB signaling in the histologically intact dorsal striatum.
BDNF-induced TrkB phosphorylation in ex vivo striatal slices is detectable in newborn pups, peaks around P7, thereafter it declines, and disappears around P12 (44, 45). Hence, we prepared acute brain slices from P5 pups that were maintained in artificial cerebral spinal fluid (aCSF). Dorsal striatum from one set of hemi-slices was stimulated with 50 ng/ml BDNF (in aCSF) for 1 hour, while the other set of matching hemislices were incubated only in aCSF serving as unstimulated control. BDNF treatment induced a 2.5-fold increased phosphorylation of TrkB when normalized to total TrkB (p=0.008) (Figure 4C-D). Strikingly, when comparing the response to BDNF treatment in WT and SorCS2 KO mice, we found that TrkB phosphorylation was reduced by 60% in slices from the knockouts (p=0.006) (Figures 4E). We thus conclude that SorCS2 binds TrkB in a BDNF-dependent manner, and that this interaction is required for robust TrkB phosphorylation in the dorsal striatum.

SorCS2 promotes BDNF-dependent LTP in dorsal striatum
Given a critical role of BDNF-TrkB signaling in ionotropic LTP, we investigated if SorCS2 modifies synaptic potentiation in the corticostriatal synapses. We first examined AMPAR activity by recording input-output (I/O) curves in the dorsal striatum in P35-45 mice. Field potentials were evoked in the white matter by stimulating the lower part of the corpus callosum directly overlying the dorsal striatum, and the synaptic response was quantified as the slope of the fEPSP. The I/O curves showed no differences in AMPAR activity between genotypes (Figure 5A-B). We then applied a theta burst stimulation (TBS) protocol for induction of LTP (46). A stable baseline of at least 20 min preceded the TBS, and LTP was subsequently quantified as the percentage increase in the slope of the fEPSP compared to the baseline. Notably, whereas we observed robust LTP in WT slices for up to 180 min post stimulation (p<0.0001), LTP was substantially impaired in both the SorCS2 KO and R6/1 mice (p<0.0001 for both conditions) ( Figure 5C). To explore whether the perturbed BDNF-TrkB signaling accounted for the blunted LTP in SorCS2 KO and R6/1 mice, we performed rescue experiments in which 10ng/ml BDNF was supplemented to the aCSF (Figure 5D-G).
In WT mice, LTP was maintained although with reduced amplitude (Figure 5E). This is in accordance with studies reporting that acute exposure to BDNF increases basal synaptic potentiation and reduces LTP in WT mice (47). Remarkably, BDNF fully restored LTP in SorCS2 KO slices as the synaptic strength was established at the same level as observed for untreated WT sections ( Figure 5F). Hence, the attenuated BDNF-TrkB signaling in Sorcs2-/-mice can be overcome by addition of exogenous BDNF. On the contrary, BDNF failed to rescue the LTP in R6/1 mice. This data suggest that either the functional deficits originate from a complete block in TrkB signaling or from defects in BDNF-independent pathways ( Figure 5G). All the measurements and statistics are summarized in (Figure 5H).

Synaptic function of GluN2B is controlled by SorCS2
Synaptic accumulation and conductance of NMDA receptors is required for the induction of BDNFdependent LTP (48). We therefore compared NMDAR currents in the dorsal striatum of 35-45 days old SorCS2 KO and R6/1 mice. The recordings were performed in 0.1 mM magnesium to partially relieve the magnesium block of the NMDARs. Once AMPAR-and NMDAR-mediated potentials were stable, we first blocked AMPARs by adding the antagonist, cyanquixaline (DNQX) ( Figure 6A). NMDAR potentials were then measured as the amplitude of the slow potential and normalized to the slope of the AMPARmediated potentials. We found that synaptic NMDAR-mediated currents were substantially attenuated in both SorCS2 KO and R6/1 mice (corresponding to approximately 60%), followed by a severe worsening of the phenotype (down to 31%) when the lines were crossbred ( Figure 6B). Next, we explored the relative contribution of GluN2A and GluN2B to the total NMDAR conductivity. To this end, neurotransmission was recorded in brain slices treated with the selective GluN2B inhibitor RO 25-6981 maleate, followed by the GluN2A inhibitor PEAQX, respectively ( Figure 6A). The GluN2B and GluN2A components were then calculated and normalized to the AMPAR currents for each experiment. Strikingly, we found that the attenuated NMDA currents in Sorcs2 -/animals were entirely accounted for by elimination of GluN2B neurotransmission ( Figure 6D) with no impact on the GluN2A activity ( Figure 6C). We measured a similar selective regulation in GluN2B activity in hippocampal CA3-CA1 synapses (Suppl. Figure S5), suggesting a more general function of SorCS2 in controlling NMDAR dependent neurotransmission. In marked contrast to SorCS2 KO, only GluN2A currents were affected in the dorsal striatum of the R6/1 mice (Figure 6C-D).
These findings show that SorCS2 deficiency and the accumulation of mHTT affect the synaptic plasticity by different modes of action, which may explain why mice harboring both impairments have a more severe disease course.

SorCS2 targets GluN2B to postsynaptic densities in a BDNF-dependent manner
To explore whether the effect of SorCS2 on synaptic plasticity was a consequence of altered expression of GluN2B or other key synaptic proteins, we performed western blotting of homogenates and synaptosomes from the dorsal striatum of WT and SorCS2 KO mice, age P35 (Suppl. Figure S6A). In the synaptosome fractions, we found no alterations in the NMDAR subunits GluN1, GluN2A, and GluN2B, neither in synaptophysin and pAkt levels (Suppl. Figure S6B-D, G, K). However, PSD95, TrkB, total Erk1/2, and pErk1/2 (but not the ratio between pErk and Erk1/2) were all increased in the homogenates (Suppl. Figure S6E, F, H-J), possibly to compensate for the reduced ability of BDNF to activate TrkB (cf. Figure 4C-D). Total Erk1/2 was increased in synaptosomes, but this did not translate into any alterations in Erk1/2 phosphorylation (Suppl. Figure S6H-J). Hence, the impaired GluN2B activity of corticostriatal synapses in the SorCS2 KO mice is not accounted for by altered expression of NMDARs or the selected synaptic proteins involved in the neurotransmission.
BDNF regulates the synaptic targeting and conductance of GluN2B in the corticostriatal pathway (18, 19, 49-51). We therefore studied GluN2B expression in synapses of the dorsal striatum from SorCS2 KO mice with or without prior BDNF stimulation. The sections were immunostained for PSD95 and GluN2B, processed by Airy-scan super-resolution microscopy. Strikingly, GluN2B co-localization with PSD95 was lower by approximately 25% in the unstimulated SorCS2 KO slices compared to WT mice ( Figure 6E-F).
Furthermore, BDNF treatment increased postsynaptic enrichment of GluN2B by approximately 50% (from 7.5% to 11.25%) but SorCS2 KO were completely refractory. Hence, it is likely the local activation of GluN2B at the synapse, rather than its recycling and recruitment from extrasynaptic sites, that accounts for the restoration of LTP by BDNF in SorCS2 deficient mice. These data suggest that SorCS2, aside from its ability to support TrkB signaling, also directly enables the synaptic targeting of GluN2B.
Finally, we asked whether SorCS2, similarly to its interaction with TrkB, might physically interact with GluN2B in a BDNF-dependent manner. To this end, we used proximity ligation assay (PLA) for SorCS2 and GluN2B interaction. This method labels proteins that are closely juxtapositioned. Immunofluorescence microscopy revealed that the receptors avidly co-localized throughout MSNs in naïve WT slices. Strikingly, stimulation with BDNF reduced the SorCS2-GluN2B interaction by 60% indicating disengagement of the SorCS2-GluN2B complex ( Figure 6G-H). Remarkably, almost 50% of the SorCS2-GluN2B PLA puncta were present in PSD95-positive domains where the BDNF treatment completely abolished their interaction ( Figure 6I). Taken together, the data demonstrate that SorCS2 binds GluN2B to enable lateral surface mobility of GluN2B into the synapse after which BDNF abolishes their heterodimerization.

DISCUSSION
In the healthy brain, BDNF-TrkB signaling serves two functions; it enables the activity-dependent regulation of synaptic structures and function, and it supports the neuronal integrity through trophic stimulation. A cornerstone in the HD neuropathology is the diminished BDNF activity as consequence of reduced BDNF production or its impaired anterograde transport from cortex to striatum, or decreased expression or signalling abilities of TrkB. At early stages of the disease, altered NMDA receptor activity caused by reduced BDNF-TrkB signaling is considered fundamental while at later stages lowered neuroprotection may propel the pathophysiological changes (52, 53). Accordingly, some HD mouse models develop overt neurodegeneration, whereas in other models synaptic disturbances such as disrupted NMDA-dependent LTP manifest but without apparent loss of neurons (8,28,54). Here we identified SorC2 as a molecular hub that regulates BDNF signaling, synaptic plasticity, and HD progression through dynamic changes in its physical interactions with TrkB and GluN2B.
Here we show that SorCS2 is selectively expressed by MSNs in dorsal striatum, and that the performance of SorCS2 KO mice on the rotarod is compromised. Most likely, this effect was not accounted for by neuronal cell loss since the surrogate parameters of neuronal degeneration DARPP-32 and clasping were not altered in the aged SorCS2 knockouts. On the other hand, SorCS2 deficiency on the R6/1 genetic background severely aggravated the motor phenotypes and accelerated the disease progression. One of the earliest alterations during HD development is a dysfunction of corticostriatal synapses (52). In HD mouse models such as BACHD, Q175, or HdhQ7/Q111, the motor deficits have been attributed to the loss of LTP (8,55). BDNF signalling is fundamental for ionotropic potentiation, and both GluN2A and GluN2B are required for LTP in corticostriatal synapses (8,56,57). In hippocampal neurons, the activation of TrkB results in phosphorylation of GluN1 (58) and GluN2B subunits (18,19), but not GluN2A (59), which is required for synaptic trafficking of NMDA receptors and LTP induction (49). Here we show that SorCS2 binds TrkB in BDNF-dependent manner, and that this interaction is required for the efficient activation of TrkB. Furthermore, the absence of SorCS2 caused a significant decrease in TrkB phosphorylation and abolishment of LTP in the corticostriatal pathway. These features were probably caused by a severe deficiency in GluN2B currents as the neurotransmission by GluN2A was preserved. Our findings contrast those of Ma et al 2017 who reported that GluN2A expression is decreased in MSNs from SorCS2 KO mice.
However, the authors did not study any functional consequences of their observation (60). Although we cannot discount the possibility that SorCS2 affects the subcellular distribution of GluN2A, it does not seem to be of major functional significance. In contrast to SorCS2 deficient mice, GluN2A currents were affected only in the R6/1 line. This is in accordance with a recent study that reported reduced LTP, impaired GluN2A but unaltered GluN2B currents in the hippocampus of R6/1 mice (61). In our SorCS2 KO; R6/1 double transgenic mice, GluN2A and GluN2B were both severely attenuated which translated into a marked reduction in total NMDA receptor currents. This observation may explain the worsening of the behavioral phenotypes in our double transgenic mice.
Aside from the function of SorCS2 in TrkB activation, which induces phosphorylation of GluN2B and GluN1 that is necessary for their synaptic enrichment, SorCS2 also modulates GluN2B function through their direct, BDNF-independent interaction. Co-immunoprecipitation experiments recently showed that SorCS2 and GluN2B form heterodimers in hippocampal extracts (62). We found that synaptic clustering of GluN2B in the dorsal striatum from SorCS2 KO was reduced by 25% compared to WT mice. Furthermore, GluN2B currents in naïve, unstimulated slices were diminished by 80% in SorCS2 KO while GluN2A currents remained unaffected. Strikingly, the interaction between SorCS2 and GluN2B is dynamic in MSNs as BDNF disrupted the complexes. SorCS2 is engaged in several aspects of protein sorting directing its cargo between Golgi, cell surface and endosomes (36). Notably, we found that SorCS2 facilitates the synaptic translocation of TrkB to postsynaptic densities in hippocampal neurons (33). Given the many trafficking pathways, SorCS2 could potentially enable the synaptic insertion of GluN2B during exocytosis, its lateral diffusion between the synapse and extrasynaptic sites, or its endocytosis and recycling. Even though the sorting pathway is currently unclear, our study identifies SorCS2 as a novel sorting mechanism to regulate the postsynaptic membrane localization of GluN2B.
Psychiatric symptoms are common in HD (1). Our previous studies discovered that SorCS2 KO mice exhibit cognitive impairments and traits typifying ADHD and schizophrenia (33,63). Similarly to the dorsal striatum, ionotropic LTP was absent in CA3-CA1 synapses but could be rescued by exogenous BDNF. Here we find that GluN2B but not GluN2A currents are substantially reduced also in CA3-CA1 synapses, suggesting a broader function of SorCS2 in modifying NMDAR-dependent plasticity. SorCS2 plays an important role for dopaminergic functionality as SorCS2 deficiency severely perturbs dopaminergic firing rates in the ventral tegmental area, striatal dopamine metabolism, and dopamine D1 receptor (D1R) sensitivity (35,63). Inquiringly, D1R cooperates with TrkB to modulate ERK signalling in striatal neurons and to increase GluN2B phosphorylation (64,65). Given that dysfunctional striatal dopamine signalling contributes to the HD pathophysiology in mouse models (66), it possible that yet another loop may be added to the many functions of SorCS2 in Huntington's disease.
SorCS2 mRNA and protein levels were reduced in the dorsal striatum of symptomatic R6/1 mice with no changes in cortex. Interestingly, the expression level was dependent on age and HD progression as it was not detected in young asymptomatic mice. A separate study reported reduced SorCS2 expression also in zQ17Z and R6/2 HD mice (60). Inversely, others found upregulation of SorCS2 mRNA in MSNs in response to deep-brain stimulation in a Parkinson's disease mouse model (67). It appears that SorCS2 expression in MSNs is highly controlled by internal and external stimuli, and can dynamically change, e.g. as a reaction to pathophysiological changes in the tissue. A recent study reported that mHTT can bind SorCS2 leading to its redistribution and aggregation in MSNs (60). Hence, in addition to attenuated gene transcription, scavenging of SorCS2 protein by mHTT may further reduce its expression in MSNs.
Taken together, we provide evidence that SorCS2 operates by two mechanisms to orchestrate GluN2Bdependent plasticity in MSNs. First, it associates with TrkB to enable its activation by BDNF, which is required to phosphorylate GluN2B and to empower GluN2B enrichment at the postsynaptic densities.
However, in parallel to this, SorCS2 may assist in targeting GluN2B to the synapse through their direct interaction. Once at the synapse, BDNF uncouples SorCS2 from GluN2B, which may condition GluN2B for the required biological activity. The dynamic interaction of SorCS2 with TrkB, and GluN2B, respectively, might thus produce an amplifying loop where SorCS2, which is liberated from GluN2B by BDNF, can be reused by TrkB to further fuel BDNF signaling, phosphorylation of GluN2B, and the induction of LTP. This proposed model is schematized in the Figure 7. Even though the primary determinant of HD onset is the length of the CAG repeats on the expanded allele (1,68), yet this only accounts for 50-70% variations in the age of onset (30,31,68,69) . Notably, a recent GWAS study identified SNPs and missense mutations in SORCS2 associated with HD (32), and another study has reported reduced levels of SorCS2 mRNA in post-mortem brains of HD patients (70). Here we have provided mechanistic support for SORCS2 as one of the long-sought disease modifying genes in Huntington's disease.

Mice
All mice were bred and group-housed with their littermates at Aarhus University Animal Facility with unlimited access to food and water and a 12/12 hour light/dark cycle. Generation of the SorCS2 KO mice is described in (63). SorCS2 KO line has been backcrossed for ten generations into C57BL/6J, as were WT mice. The R6/1 mouse model (obtained from the Jackson Laboratory, USA) was originally created on a CBA/C57BL/6 hybrid background (40), but has since been backcrossed for ten generations into C57BL/6J (71). R6/1 mice were maintained on the C57BL/6J background by crossing heterozygous males with WT females since R6/1 females are infertile. R6/1 mice were crossed into our SorCS2 KO mouse line to obtain SorCS2 KO; R6/1 line. Mice were genotyped for Sorcs2 and HDexon1 by PCR. Three of the R6/1, SorCS2 KO males (two 18 weeks old and one 21 weeks old) and one SorCS2 KO; R6/1 female died (17 weeks old). Data from these death individuals are included in the graphs until the last age they were tested. The experiments were double-blinded. Mice were tested weekly for clasping and grip strength from the age of 7 weeks and until they were 22 weeks old. Body weight was monitored weekly.

Behavioral testing
Clasping was tested by holding the mouse by its tail and lifting it from the top of its home-cage for 10 successive trials of three seconds, with three-second rest intervals. The testing was recorded on video, and later scored to evaluate the disease progression. Each three-second trial was scored as following: A) 0 points: Normal righting response/no certain abnormalities. Slight collapse of the hind limbs toward the midline was not scored as abnormal unless the hind limbs were at least parallel. Struggling to grab hind limbs or tail with forelimbs was not scored as abnormal. B) 1 point: One hind limb with abnormal retraction or clearly abnormal posture, or both hind limbs abnormally collapsing toward midline until they were at least parallel but without touching or crossing. C) 2 points: Both hind limbs clasping (touching or crossing) after a delay of more than one second or intermittently. D) 3 points: Almost immediate (within the first second) and persistent clasping of hind limbs. The scores from each of the 10 three second trials were summed to give a score from 0-30. Grip strength was tested by letting the mouse grip a horizontal metal bar connected to a newton meter and pulling the mouse away horizontally by the tail. The force recorded by the newton meter was stored digitally by MATLAB software. Grip strength was recorded as the highest value of 10 successive trials.
For rotarod testing, open field testing and gait analysis, mice were brought to the testing room in their home-cage for at least one hour before the testing started. All equipment was wiped thoroughly with ethanol between the testing of individual mice.

Electrophysiological experiments
Male mice (P35-45) of indicated genotypes were deeply anesthetized with 4% isoflurane and decapitated. The brain was quickly extracted, transferred to ice-cold artificial cerebral spinal fluid (aCSF) bubbled with 95% O 2 and 5% CO 2 and cut into 400 µm coronal slices on a vibratome (Vibratome 3000 Sectioning System).
Recording CA1 of hippocampus: described in (33). Recording dorsal striatum: Slices containing striatum rostral to and including the level of the commisura anterior were transferred to a grid in a storage chamber with aCSF bubbled with 95% O 2 and 5% CO 2, where they were allowed to recover for at least 1.

BDNF stimulation of dorsal striatum in acute brain slices from P5 pups
P5 pups were decapitated, the brain quickly extracted and transferred to ice-cold aCSF containing 126mM NaCl, 2.5 mM KCl, 1.25 mM NaH 2 PO 4 , 26 mM NaHCO 3 , 2.5 mM CaCl 2 , 1.3 mM MgCl 2 and 10 mM D-glucose bubbled with 95% O 2 and 5% CO 2 and cut into 400 µm coronal slices on a vibratome (Vibratome 3000 sectioning system). Slices containing the striatum rostral to and including the level of the commisurra anterior were transferred to a grid in a storage chamber with aCSF bubbled with 95% O 2 and 5% CO 2 where they were allowed to recover for at least 1.5 hour at RT. Thereafter the dorsal striatum was dissected from individual slices while they were still in oxygenated and carbonated aCSF under a stereo microscope.
The dorsal striatum from matching hemi-slices was then directly transferred to aCSF continuously bubbled with 95% O 2 and 5% CO 2 containing either 50 ng BDNF/ml or no BDNF for 60 min at 37˚ C with hemi-slices from individual pups (typically three per pup) kept separated in cell strainers. Thereafter, the samples were immediately processed for the subsequent analysis.

Proximity Ligation Assay (PLA) and super-resolution microscopy
The hemi-slices were handled as free-floating sections. First, they were fixed in 4% formaldehyde for 1 hour, washed 3 times with 1xPBS pH7.4 containing 0.1% TritonX-100. Then they were blocked in PBTA Dorsal striatum was imaged with Zeiss LSM800 confocal microscope using the Airyscan mode for the super-resolution imaging, x63 objective, and voxel size 40x200nm. Co-localization of GluN2B to PSD95 was quantified with the FIJI software as the object area positive for both GluN2B and PSD95 divided by the total object area positive for GluN2B. Co-localization of PSD95 to the SorCS2-GluN2B PLA signal was quantified with the FIJI software as the object area positive for both the PLA and PSD95 divided by the total object area positive for PSD95. All intensity thresholds were obtained with the Fiji autothreshold command. All pixels of the GluN2B-, PLA-and PSD95-positive objects were quantified via the AnalyzeParticles Fiji command. The 3D reconstruction was performed by Imaris 8.2 software.

Statistics
We used GraphPad Prism 9.1.2 for the data visualization and statistical analysis. All data sets were evaluated for normality and statistical outliers. The data are presented as mean ±SEM. For simplicity, the symbols * indicate p<0.05, ** indicate p<0.01, *** indicate p<0.001, and **** indicate p<0.0001. An overview of used statistical tests can be found in the Table 1.                        A. RNA In situ hybridization shows sorcs2 expression in coronal section of P56 WT mouse brain, including the corticostriatal circuits. B. Quantified expression profile of sorcs2 from the same section. Blue color indicates low expression, green and yellow stand for increasing expression. Scale-bar is 700um. C. Sorcs2 is expressed in multiple brain regions including induseum griseum (a.), interior cirgulate cortex (b.), caudoputamen (c.), lateral septum (d.), and piriform cortex (e.). A.-C. Image credit: ©2004 Allen Institute for Brain Science. Allen Mouse Brain Atlas Available from mouse.brain-map.org/gene/show/57520. D-E. Immunofluorescence followed by confocal microscopy imaging revealed that SorCS2 is expressed by NeuN (D.) and DARPP-32 (E.) positive MSNs neurons in dorsal striatum of adult mice. Blue channel corresponds to nuclei stained by Hoechst. The scale bar corresponds to 50µm, and to 20µm in magnified figures.