Trophic factor BDNF inhibits GABAergic signaling by facilitating dendritic enrichment of SUMO E3 ligase PIAS3 and altering gephyrin scaffold

Posttranslational addition of a small ubiquitin-like modifier (SUMO) moiety (SUMOylation) has been implicated in pathologies such as brain ischemia, diabetic peripheral neuropathy, and neurodegeneration. However, nuclear enrichment of SUMO pathway proteins has made it difficult to ascertain how ion channels, proteins that are typically localized to and function at the plasma membrane, and mitochondria are SUMOylated. Here, we report that the trophic factor, brain-derived neurotrophic factor (BDNF) regulates SUMO proteins both spatially and temporally in neurons. We show that BDNF signaling via the receptor tropomyosin-related kinase B facilitates nuclear exodus of SUMO proteins and subsequent enrichment within dendrites. Of the various SUMO E3 ligases, we found that PIAS-3 dendrite enrichment in response to BDNF signaling specifically modulates subsequent ERK1/2 kinase pathway signaling. In addition, we found the PIAS-3 RING and Ser/Thr domains, albeit in opposing manners, functionally inhibit GABA-mediated inhibition. Finally, using oxygen–glucose deprivation as an in vitro model for ischemia, we show that BDNF–tropomyosin-related kinase B signaling negatively impairs clustering of the main scaffolding protein at GABAergic postsynapse, gephyrin, whereby reducing GABAergic neurotransmission postischemia. SUMOylation-defective gephyrin K148R/K724R mutant transgene expression reversed these ischemia-induced changes in gephyrin cluster density. Taken together, these data suggest that BDNF signaling facilitates the temporal relocation of nuclear-enriched SUMO proteins to dendrites to influence postsynaptic protein SUMOylation.

Posttranslational addition of a small ubiquitin-like modifier (SUMO) moiety (SUMOylation) has been implicated in pathologies such as brain ischemia, diabetic peripheral neuropathy, and neurodegeneration. However, nuclear enrichment of SUMO pathway proteins has made it difficult to ascertain how ion channels, proteins that are typically localized to and function at the plasma membrane, and mitochondria are SUMOylated. Here, we report that the trophic factor, brainderived neurotrophic factor (BDNF) regulates SUMO proteins both spatially and temporally in neurons. We show that BDNF signaling via the receptor tropomyosin-related kinase B facilitates nuclear exodus of SUMO proteins and subsequent enrichment within dendrites. Of the various SUMO E3 ligases, we found that PIAS-3 dendrite enrichment in response to BDNF signaling specifically modulates subsequent ERK1/2 kinase pathway signaling. In addition, we found the PIAS-3 RING and Ser/Thr domains, albeit in opposing manners, functionally inhibit GABA-mediated inhibition. Finally, using oxygen-glucose deprivation as an in vitro model for ischemia, we show that BDNF-tropomyosin-related kinase B signaling negatively impairs clustering of the main scaffolding protein at GABAergic postsynapse, gephyrin, whereby reducing GABAergic neurotransmission postischemia. SUMOylationdefective gephyrin K148R/K724R mutant transgene expression reversed these ischemia-induced changes in gephyrin cluster density. Taken together, these data suggest that BDNF signaling facilitates the temporal relocation of nuclearenriched SUMO proteins to dendrites to influence postsynaptic protein SUMOylation.
The family of small ubiquitin-like modifier (SUMO) proteins initially identified in Saccharomyces cerevisiae is now known to be expressed in all eukaryotes (1). SUMO conjugation on substrate proteins occurs over three-step process involving ATP and SUMO-specific enzymes. While the SUMO-1, -2, -3 proteins are expressed from three different genes in humans, only one E2 conjugating enzyme, Ubc9, has been described in eukaryotes (2). E3 ligases trigger SUMO conjugation on substrates by recruitment of Ubc9. They consist of two major classes, namely HECT-domain and RING-domain type ligases. The RING-type ligases bind both substrate and Ubc9 (3). Protein inhibitor of activated STAT (PIAS) family of RING-type SUMO E3 ligase are well described in literature for their SUMO-conjugating role in eukaryotes (4)(5)(6). The initial link between SUMOylation and nucleocytoplasmic transport was established when the import factor RanGAP1 SUMOylation was shown to localize it to the nuclear pore (7). Subsequently, numerous independent reports have shown that several cellular proteins alter their nucleocytoplasmic distribution and function upon SUMOylation (8,9). Although most SUMO conjugates described in the literature are localized within the nucleus, SUMO modification can also occur outside the nucleus as SUMOylation of membrane receptors (GluK2 and Kv2.1; (10,11)); cytosolic proteins (CASK), syntaxin1 and gephyrin (12,13); and metabolic enzymes localized within the cytoplasm have been reported (14). Even though the controversy surrounding the intracellular site for SUMO conjugation has dissipated, our understanding about the occurrence rate of protein SUMOylation and its upstream signal(s) remains limited.
In neurons, SUMO conjugation of cytoplasmic and membrane proteins influences cell physiology by allowing rapid adaptations to shifts in cellular metabolism via intermolecular and intramolecular interaction (13,14). Therefore, SUMOylation of synaptic proteins has emerged as a critical regulator of synaptic plasticity (15). For example, SUMOylation has also been shown to contribute to the GABAergic postsynapse organization through both SUMO-1 and SUMO-2 conjugation on gephyrin, the main inhibitory scaffolding protein (13). In the same study, it was reported that PIAS-3 and SENP-2 modulate gephyrin SUMOylation levels (at K148 and K724 residues) downstream of α2 GABA A Rs to facilitate scaffolding at inhibitory postsynaptic membrane (13).
While SUMO substrates and the functional consequences of SUMOylation are becoming clear, the upstream signaling that facilitates SUMO conjugation onto substrates remains less well understood. It has been reported that under conditions of cellular stress, protein SUMOylation increases (16,17), and after ischemia, brain-derived neurotrophic factor (BDNF) levels transiently increase (18). Although a functional link between the BDNF and SUMO pathway has not been established in literature, acute application of BDNF has been reported to weaken GABAergic transmission (19,20) and GABA A R surface expression in hippocampal primary (21). BDNF signaling has also been linked to ubiquitin-mediated GABA A receptor internalization and degradation in neurons (21). Furthermore, it is reported that GABA A Rs are rapidly depleted from synapses via AP2-dependent endocytosis following ischemia (22). At the molecular level, BDNF activation of its high affinity receptor, tropomyosin-related kinase B (TrkB) receptor, could influence SUMOylation of the main scaffolding protein gephyrin, whereby contributing to reduced cell surface expression of GABA A R and gephyrin clustering.
In the current study, we report that BDNF signaling regulates nucleocytoplasmic transport of SUMO-1, SUMO-2/3, and PIAS-3 proteins in neuronal cells. Specifically, PIAS-3 is the only member of the E3 ligase family whose cytoplasmic localization in neuronal cells are reversibly affected by the duration of TrkB activation. At a mechanistic level, we report functional uncoupling between PIAS-3 RING-domain and C-terminus S/T domain influences GABAergic neurotransmission changes. We identify ERK1/2 kinase pathway as downstream effector of PIAS-3 nuclear localization and function. Finally, we uncover that ischemia in hippocampal slices induces loss of gephyrin clusters and GABAergic synaptic transmission. Moreover, this gephyrin cluster loss can be rescued by transgenic expression of SUMO-defective gephyrin K148R/K724R mutant or BDNF scavenging.

Acute BDNF treatment alters subcellular localization of SUMO pathway proteins
To test whether BDNF acted as upstream signal to regulate the subcellular localization of SUMO proteins in neurons, we treated primary hippocampal neuronal cultures at 15 days in vitro (DIV 15) with BDNF (10 ng/ml, 90 min) followed by immunostaining of endogenous SUMO-1 or SUMO-2/3 ( Fig. 1,A and B' ). In untreated control neurons, endogenous SUMO-1 and SUMO-2/3 showed a strong nuclear enrichment consistent with previous published reports (Fig. 1, A and B). However, in contrast, the BDNF-treated neurons showed redistribution of SUMO-1 and SUMO-2/3 to somatic and dendritic compartments (Fig. 1,A' If the subcellular localization changes in SUMO proteins leads to differential substrate SUMO conjugation, we reasoned that PIAS family of E3 ligase might also exhibit similar somatic enrichment after BDNF treatment. In order to assess this, we transfected DIV 7 neurons with myc-PIAS (myc-PIAS-1, -2, -3 or γ), and at DIV 15, we treated the culture with BDNF (10 ng/ml, 90 min) followed by immunostaining for myc. Of the different PIAS family members tested in our assay, only myc-PIAS-3 showed nucleus to soma translocalization upon BDNF application (Figs. 1C and S1). To test whether BDNFinduced myc-PIAS-3 somatic enrichment was acting via the TrkB receptor signaling, we treated myc-PIAS-3-transfected primary neurons with BDNF for either 90 min or up to 48 h. At 90 min time point, PIAS-3 was enriched in the soma and dendrites; interestingly, at 48 h time point, we observed enrichment of myc-PIAS-3 within the nucleus (Fig. 1,D-D'). In order to test if TrkB receptor signaling was necessary to see this relocalization, we treated the primary neurons with the pharmacological TrkB antagonist (ANA-12, 400 nM) 5 min prior to BDNF application. We imaged the cells at 90 min after ANA-12 and BDNF treatment ( Fig. 1D'') and found nuclear enrichment of myc-PIAS-3. Quantification confirmed that somatic localization of myc-PIAS-3 is indeed reversible and can be successfully blocked using a pharmacological inhibitor of TrkB (chi-squared test myc-PIAS3 versus myc-PIAS3 BDNF treatment, χ 2 (1, N = 100) = 73.08 p < 0.00001). We also assessed if endogenous PIAS-3 somatic underwent relocalization at 90 min post BDNF application (Fig. 1,E-E') in order to eliminate any myc-PIAS-3 subcellular localization change after BDNF treatment as an overexpression artefact. Endogenous PIAS-3 was enriched within the nucleus in control untreated neurons. We found that scavenging BDNF using chimeric TrkB-Fc (1 μg/ml) prevented relocalization of endogenous PIAS-3 to the nucleus ( Fig. 1E''). However, upon 90 min BDNF application, endogenous PIAS-3 relocalized to soma and dendrites (one-way ANOVA, Dunnett's test F (2,9) = 303, p < 0.0001). Together, our results concur with conclusion that BDNF is a novel and specific regulator of SUMO proteins subcellular localization in neurons.

BDNF and not NT3 or NT4 influences gephyrin clustering
It is established that both BDNF and NT4 can activate TrkB signaling. Hence, we compared BDNF with NT3 which preferentially activates TrkC and BDNF with NT4 that activates TrkB (23). The activation of signaling cascade downstream of TrkB upon activation by BDNF or NT4 is distinct (24). To understand signaling crosstalk between neurotrophic factors (BDNF, NT3, and NT4) for gephyrin clustering changes, we treated the primary neurons transfected with eGFP-gephyrin with either BDNF, NT-3, or NT-4 (10 ng/ml, 90 min).

EDITORS' PICK: PIAS-3 alters GABAergic inhibition
Quantification for cluster size confirmed that the reduction of eGFP-gephyrin cluster size was specific to BDNF treatment as there were no changes after NT-3 and NT-4 application ( The results confirm that BDNF signaling specifically reduces gephyrin cluster size through SUMOylation at K148 and K724 sites respectively.

PIAS-3 harbors two gephyrin interaction sites
In order to understand the biochemical basis for PIAS-3mediated gephyrin clustering changes, we assessed for PIAS-3 ability to directly interact with gephyrin. We have previously shown that of the various PIAS family members, only PIAS-3 and PIAS-2α interact with gephyrin (13). Importantly, PIAS-3 interaction with gephyrin is determined by the phosphorylation status of gephyrin at S268 and S270 sites, respectively (13). Here, we assessed for binding domain(s) within PIAS-3 for gephyrin interaction. For this, we cotransfected the HEK293 cells with FLAG-gephyrin and myc-PIAS-3, myc-PIAS-3 RING domain catalytic inactive mutant (Rm), myc-PIAS3 PINIT domain, myc-PIAS3 RING domain, or myc-PIAS3 S/T domain. Immunoprecipitation (IP) for myc-PIAS-3, . These results suggest that gephyrin interaction with PIAS-3 can occur via more than one interaction site.
To determine the PIAS-3 interaction sites on gephyrin, we cotransfected HEK293 cells with myc-PIAS-3 and FLAGgephyrin, FLAG-G, FLAG-GC, or E domain truncation mutant of gephyrin. IP for myc-PIAS-3, followed by Western blotting for FLAG-gephyrin confirmed the previously reported interaction between full-length PIAS-3 and gephyrin ( Fig. 5B; lane 2). In addition to binding to full-length gephyrin, PIAS-3 interaction was seen with FLAG-G, FLAG-GC, and FLAG-E domain truncation mutations of gephyrin ( Fig. 6B; lanes 3-5). Our biochemical data are consistent with the earlier observation that PIAS-3 SUMOylates gephyrin at K148 and K724 sites located on the G-and E-domain, respectively (13).
Next, we investigated if there were changes in eGFPgephyrin cluster density in neurons co-expressing different myc-PIAS-3 deletion mutants. The co-expression of fulllength myc-PIAS-3 significantly reduced eGFP-gephyrin cluster density ( decreasing the gephyrin cluster density. However, PINIT domain can partially block the RING domain function.
We next quantified the effect of myc-PIAS-3 S/T expression on eGFP-gephyrin clustering. We treated the neurons with BDNF and found no change to the size or density of gephyrin clusters. Therefore, we tested TrkB-Fc in these transfected cells and analyzed for morphological changes in gephyrin

c -P I A S 3 ( S / T ) m y c -P I A S 3 ( R I N G ) m y c -P I A S 3 ( P I N I T ) m y c -P I A S 3 m y c -P I A S 3 R m m y c -P I A S 3
IgG myc myc-PIAS3 F la g -g e p h y r in F la g -E F la g -G C F la g -G F la g -g e p h y r in However, upon analysis of neurons that were treated with TrkB-Fc, we found that eGFP-gephyrin cluster size returned to base line levels ( Fig. 7I; 0.41 μm 2 ± 0.003 versus 0.79 μm 2 ± 0.15). Similarly, quantification for gephyrin cluster density in neurons co-expressing PIAS-3 S/T returned to base line level after TrkB-Fc treatment ( Fig. 7J; 3.8 ± 0.5 versus 1 ± 0.3 clusters/20 μm). These data show that S/T domain function is regulated in a mechanism opposite to RING domain function, which requires active BDNF signaling.
To understand this discrepancy in BDNF-mediated PIAS-3 regulation better, we used myc-PIAS3Rm wherein the RING domain in FL PIAS-3 has been rendered catalytically inactive by mutations (C299S/H301A) (31). Given that the PIAS-3Rm is defective for SUMO conjugation, we did not expect a phenotype change in eGFP-gephyrin clustering. However, cluster size of eGFP-gephyrin was increased in neurons transfected with myc-PIAS-3Rm and cluster density was reduced as seen with wild type myc-PIAS-3. To better understand how myc-PIAS-3Rm altered eGFP-gephyrin clustering, we treated neurons co-transfected with myc-PIAS-3Rm with BDNF (90 min). BDNF application did not change the morphology of eGFP-gephyrin in neurons coexpressing myc-PIAS-3Rm ( Together, our data show that BDNF via TrkB signaling regulates PIAS-3 RING domain function, while scavenging BDNF impacts S/T domain function.
As a next step, we overexpressed WT eGFP-PIAS-3 and treated cells with BDNF to evaluate its direct impact on PIAS-3 function and gephyrin clustering. In eGFP-PIAS-3-transfected control neurons, we saw a significant reduction of mIPSC amplitude (66.1 ± 2.4pA versus 57.4 ± 0.9pA; p < 0.05, Kolmogorov-Smirnov test) and significant increase in mIPSC interevent interval (1984.2 ± 128.5 versus 1024.1 ± 54.4; p < 0.05, Kolmogorov-Smirnov test) (Fig. 8, D and E), suggesting reduced GABA A Rs at synaptic sites and reduced density of GABAergic synapses. However, BDNF application reversed the eGFP-PIAS-3 effect on GABAergic inhibition and returned mIPSc interevent interval to baseline levels as seen in the mock-transfected control cells. BDNF translocates PIAS-3 from the nucleus to dendrites, and BDNF resets gephyrin cluster size but not cluster density in PIAS-3 overexpressing neurons. In contrast, our functional data suggest that perhaps gephyrin-independent GABA A Rs facilitate inhibitory neurotransmission when PIAS-3 SUMOylates gephyrin to prevent macroclustering (oligomerization) at synaptic sites. Similar compensation in GABAergic inhibition has been reported upon ablation of α2 GABA A R subunit containing GABA A Rs in hippocampal pyramidal neurons (32).
As a next step, we assessed the influence of PIAS-3Rm on GABAergic transmission. We compared differences in mIPSC amplitude between eGFP, eGFP-PIAS-3Rm, or eGFP-PIAS3Rm treated with TrkB-Fc (Fig. 8F). Although PIAS-3Rm mutant increases eGFP-gephyrin cluster size morphologically, at a functional level, mIPSC amplitude is not altered (56.9 ± 3.9 versus 57.2 ± 1.9). This suggests that gephyrin-independent GABA A Rs contribute to the amplitude, while the large gephyrin aggregates observed in PIAS-3Rmtransfected dendrites are perhaps cytosolic protein aggregates due to SUMOylation defect. The eGFP-PIAS-3Rm-expressing cells show shorter (45%) mIPSC interevent intervals ( Fig. 5G; 6124 ± 473 versus 3505 ± 352; p < 0.05, Kolmogorov-Smirnov test), suggesting reduced number of GABAergic synapses. The reduced interevent intervals is consistent with the morphological reduction in gephyrin cluster density in myc-PIAS-3Rm expressing neurons (Fig. 7, K-M). Scavenging BDNF using TrkB-Fc showed interevent intervals similar to eGFP control cells (Fig. 8G). The composition of the GABA A R subunits can be ascertained by analyzing the rise and decay kinetics of GABAergic mIPSCs. Analyses of rise and decay kinetics of GABAergic mIPSC showed no differences between eGFP-PIAS-3-transfected cells undergoing mock or BDNF treatment (Fig. 8H). Our analysis showed no differences in rise and decay times between eGFP-PIAS-3Rm-transfected cells undergoing mock or TrkB-Fc treatment (Fig. 8I). Overall, PIAS-3 impairs GABAergic synaptic transmission by reducing GABAergic mIPSC amplitude and synapse density. The negative effect of PIAS-3 on GABAergic transmission is reversed by BDNF signaling. In the PIAS3Rm, RING domain is rendered inactive for SUMO conjugation, whereby not impacting gephyrin clustering. However, we observe functional impact on GABAergic transmission. We concur that perhaps in the absence of the functional RING domain, another domain such as the S/T domain might influence gephyrin clustering abilities to impact GABAergic inhibition.

ERK1/2 phosphorylation of gephyrin at S268 impairs PIAS3 influence on clustering
We have reported earlier that gephyrin is a direct substrate for ERK1/2 phosphorylation and that ERK phosphorylation at S268 residue results in reduced gephyrin cluster size, causing a functional reduction in GABAergic inhibition (26). Given our data that ERK pathway directly influences PIAS-3 function, we assessed for crosstalk between gephyrin phosphorylation at S268 and PIAS-3. For this, we transfected primary neurons with eGFP-gephyrin and myc-PIAS-3, myc-PIAS3Rm, myc-PIAS3 RING, or myc-PIAS3 S/T to assess their influence on gephyrin clustering in the presence of PD98059. At a morphological level, the expression of either myc-PIAS-3, myc-PIAS3Rm, myc-PIAS3 RING, or myc-PIAS3 S/T and treatment with PD98059 reduced the intracellular gephyrin aggregates and formed smaller submembrane clusters (Fig. S2, A-H). Similarly, PD98059 treatment facilitated the formation of numerous eGFP-gephyrin clusters to increase the density significantly (Fig. S2, C-H). However, specifically in neurons expressing myc-PIAS-3, PD98059 treatment did not increase the density of eGFP-gephyrin clusters (Fig. S2A). Overall, we uncover a direct link between gephyrin phosphorylation at S268 residue and PIAS-3-mediated SUMO conjugation on gephyrin.

Oxygen-glucose deprivation induces downregulation of gephyrin scaffolding and GABAergic inhibition
Independent reports have shown BDNF and global SUMO upregulation under ischemic conditions (16,33). Rapid internationalization of GABA A Rs after ischemia has also been reported (22). We speculated that during ischemia, increased BDNF expression could reduce synaptic abundance of GABA A Rs via PIAS-3-mediated gephyrin modification at K148 and K724 residues. To test our idea, we used organotypic hippocampal slice culture, as the local neuronal network is well preserved in this in-vitro system. We focused on CA1 pyramidal neurons as they have been reported to be more susceptible to ischemia (34). We induced OGD for 4 min and analyzed for BDNF upregulation after 90 min. We performed quantitative real-time PCR (RT-qPCR) analysis to measure change in the bdnf transcript at 90 min post OGD. Analysis upon normalization using the house-keeping gene GAPDH showed a significant increase in bdnf mRNA levels ( Fig. 10A; p = 0.046). Next, we stained for gephyrin and analyzed for changes in cluster size and density at 24 h post-OGD (Fig. 10, B and C). Quantification confirmed that at 24 h post-OGD, gephyrin cluster volume is not changed. We also blocked BDNF signaling using TrkB-Fc and assessed for morphological changes in gephyrin clustering (Fig. 10, D and E). Quantification confirmed that blocking BDNF signaling using TrkB-Fc does not impact gephyrin cluster volume at 24 h post-OGD ( Fig. 10D; 0.096 μm 3 ± 0.008 versus 0.091 μm 3 ± 0.006; twotailed Mann-Whitney t test p = 0.63). However, at 24 h post-OGD, gephyrin cluster density was significantly reduced. Importantly, TrkB-Fc treatment of OGD slices could prevent the loss of gephyrin clusters (Fig. 10E; 47.44 ± 8.78 versus 331.9 ± 22.37; two-tailed Mann-Whitney t test p < 0.0001).
We next examined whether morphological loss of gephyrin cluster density at 24 h post-OGD also resulted in functional loss of GABAergic transmission. For this, we performed whole-cell patch clamp recording GABAergic mIPSC in organotypic slices that were mock treated, treated with TrkB-Fc, underwent OGD, or underwent OGD in the presence of TrkB-01Fc (Fig. 10, F-H). Consistent with the morphology which showed that gephyrin cluster volume is not changed at 24 h post-OGD, mIPSC amplitude was not altered at 24 h post-OGD ( Fig. 10G; 30.34 pA ± 1.403 versus 30.05 pA ± 1.231, two-tailed unpaired t test p = 0.88). Similarly, consistent with the morphological reduction in gephyrin cluster density, interevent intervals of GABAergic IPSCs were also significantly increased at 24 h after OGD ( Fig. 10H; 198.2 ms ± 13.34 versus 146.0 ms ± 10.25, two-tailed unpaired t test p = 0.0046). Importantly, scavenging BDNF using TrkB-Fc (1 mg/ml) prevented the loss of GABAergic inhibition at 24 h post-OGD. Analysis showed that TrkB-Fc application prior to OGD does not influence the mIPSC amplitude (Fig. 10G, 28.63 pA ± 3.02 versus 28.77 pA ± 2.6, two-tailed unpaired t test p = 0.97). Similarly, the loss of GABAergic synapses was prevented in slices treated with TrkB-Fc prior to OGD (Fig. 9H, 182.8 ms ± 19.25 versus 141.9 ms ± 15.49, two-tailed unpaired t test p = 0.114). Taken together, our data establish a direct link between BDNF signaling at 24 h post-OGD with morphological changes in gephyrin clustering and functional alteration in GABAergic transmission.
In primary hippocampal neurons, we demonstrate that gephyrin SUMO-defective mutants K148R and K724R are nonresponsive to BDNF treatment (Fig. 2, D and E). We wondered whether SUMOylation-defective gephyrin mutant    Figure 10. BDNF promotes gephyrin cluster loss and reduced GABAergic inhibition after OGD. A, qPCR of bdnf transcript from CA1 area of hippocampus after 90 min after OGD. B-C 0 , morphology of gephyrin clusters in organotypic hippocampus CA1 area in control mock-treated slices, control slices treated with TrkB-Fc, OGD slices after 24 h recovery, OGD slices treated with TrkB-Fc after 24 h recovery (Scale bar 2 μm). D, quantification of gephyrin cluster volume in control and OGD slices. E, quantification of gephyrin cluster density in control and OGD slices. F, representative current traces show pharmacologically isolated GABAergic mIPSCs in OGD slices after 24 h recovery or OGD slices treated with TrkB-Fc in presence of tetrodotoxin. G, mIPSC mean amplitude at 24 h post OGD under different conditions tested. H, mIPSC interevent interval at 24 h post OGD under different conditions tested. I-J 0 , morphology of pyramidal neurons co-transfected with td-Tomato (green) and eGFP-gephyrin (red), or eGFP-gephyrin K148R/K724R SUMO-defective mutation in mock-treated or OGD slices after 90 min recovery. Boundaries of the neuronal dendrites within the panel are indicated with dashed white line. K, quantification of eGFP-gephyrin cluster size in pyramidal neurons co-expressing td-Tomato and eGFP-gephyrin or eGFP-gephyrin K148R/K724R SUMOdefective mutation under control or in OGD slices after 90 min recovery. L, quantification of eGFP-gephyrin cluster density in pyramidal neurons co-expressing td-Tomato and eGFP-gephyrin or eGFP-gephyrin K148R/K724R SUMO-defective mutation under control or in OGD slices after 90 min recovery. Data were collected from three independent experiments. Error bars st.dev. Scale bar 5 μm. *p < 0.05, **p < 0.01, and ***p < 0.0001. BDNF, brainderived neurotrophic factor; mIPSCs, miniature inhibitory postsynaptic currents; OGD, oxygen-glucose deprivation; SUMO, small ubiquitin-like modifier; TrkB, tropomyosin-related kinase B.

Discussion
In the present study, we demonstrate that BDNF signaling shuttles SUMO-1 and SUMO-2/3 from the nucleus into the soma and dendrites in a time-dependent manner. The TrkB receptor downstream of BDNF activates ERK1/2 pathway to impinge upon gephyrin and PIAS-3, influence their cooperativity, and in turn impact GABAergic inhibition. PIAS-3 and gephyrin exhibit more than one biochemical interaction site, which allows for PIAS-3 to influence gephyrin clustering via its RING and S/T domains. This influence of PIAS-3 on gephyrin clustering is in turn regulated by ERK1/2 kinase pathway and phosphorylation at Ser268 residue on gephyrin. Using OGD as in vitro model for brain ischemia, we demonstrate that after OGD, there is increased BDNF mRNA. Using TrkB-Fc chimera to sequester BDNF signaling in our OGD model, we could prevent reduction of gephyrin cluster density and downregulation in GABAergic inhibition. At 24 h post-OGD, BDNF signaling via TrkB receptor and downstream ERK1/2 pathway converge on PIAS-3 and gephyrin to influence functional adaptation at GABAergic postsynaptic sites. We report that kinase and SUMO pathways converge on determining the outcome of BDNF signaling and PIAS-3 function. Specifically, gephyrin phosphorylation by ERK1/2 on S268 and SUMO-1/-2 conjugation on K148R/K724R renders gephyrin insensitive to PIAS-3. Our data highlight that in physiology and pathology, cellular signaling cascades crosstalk with each other to influence gephyrin posttranslational modification(s) and in turn impact GABAergic inhibitory neurotransmission.

BDNF signals for PIAS-3 and gephyrin cooperativity
Our biochemical analysis identified more than one interaction site for gephyrin on PIAS-3 and vice versa. It has been reported that gephyrin is SUMO-1 conjugated at the K148 (G domain) and SUMO-2 conjugated at K724 (E domain) residues (13). The identification of PIAS-3 binding site(s) on gephyrin indicates that this could be the basis for gephyrin SUMO conjugation. It has been reported in stem cells that PINIT domain mutation leads to both nuclear and cytosolic localization of PIAS-3 (35).
As a proof of principle, we demonstrate that gephyrin SUMO-1 and SUMO-2/3 site mutations K148R and K724R, respectively, are insensitive to BDNF signaling (Fig. 2). Several neuronal proteins have been characterized as novel SUMO1 substrate in vivo (36); however, there is little mechanistic understanding of how SUMOylation is achieved at synaptic locations. Our data offer an elegant model for nucleo-dendritic shuttling of SUMO1/2/3 and PIAS-3 in response to BDNF signaling, thereby facilitating SUMOylation of synaptic proteins. We also provide evidence showing long-term BDNF treatment (48 h) renders proteins of the SUMO pathway insensitive to BDNF, again causing these proteins to relocalize within the nucleus. It is well accepted in the field that protein SUMOylation is a labile process; however, within the neuronal context, our data offer a mechanistic underpinnings of a dynamic regulatory process.
Our results show that myc-PIAS-3, myc-PIAS-3Rm, and myc-PIAS-3 S/T domains restore gephyrin cluster size and density to base line condition upon blocking of ERK1/2 signaling. We show that BDNF treatment restores gephyrin cluster size but not density in myc-PIAS-3-overexpressing neurons (Fig. 7A). However, PD98059 treatment restores both cluster size and density in myc-PIAS-3-overexpressing cells (Fig. 4A). Importantly, PD98059 treatment restores cluster size and density in neurons overexpressing the PIAS3Rm or S/T domain (Fig. 9, A-F). We envision a scenario wherein RING domain and S/T domain control the regulation of gephyrin size and density, respectively. Given that ERK1/2 also phosphorylates gephyrin at S268 to reduce cluster size (13), BDNF treatment could reduce the gephyrin cluster size via this direct phosphorylation event. However, in parallel, BDNF activates PIAS-3 to influence its SUMOylation function. Hence, PIAS-3 effect on gephyrin clustering occurs downstream of ERK1/2 pathway involving the PIAS-3 Ring and S/T domains via mechanisms that we do not understand fully.

BDNF signaling and gephyrin modulation for brain network integrity
Our observations confirm that BDNF and not NT-4 via TrkB receptor activates ERK1/2 pathway downstream to influence PIAS-3 function and gephyrin SUMOylation. This is consistent with established literature showing BDNF-TrkB interaction but not NT-4-TrkB interaction leads to less efficient sorting of TrkB receptors and enhanced activation of downstream signaling (23). The signaling downstream of BDNF is mediated by the Shc adaptor binding site on TrkB and Ras/MAPK pathway activation. The generation of mouse line harboring the Shc binding site mutation in the trkB gene has helped to delineate that NT4-dependent signaling is EDITORS' PICK: PIAS-3 alters GABAergic inhibition independent of BDNF-dependent signaling. Also, neurons derived from trkB shc/shc mutant mice do not show any defects in BDNF-dependent signaling (24). Our results are consistent with these reports and show that BDNF and not NT-4 signaling through TrkB receptor regulates GABAergic synapse plasticity. We report that dynamic time scale of synaptic plasticity adaptations is facilitated by ERK1/2 pathway directly impinging on PIAS-3 localization and function. PIAS-3 function for gene transcription regulation in photoreceptor cells has been reported (6). Our data provide a molecular framework for PIAS-3 function at synaptic sites.
Dynamic modulation of GABAergic inhibition is especially relevant within the context of synaptic homeostasis, wherein individual neurons and/or synapse adapts to fluctuations in activity. In addition, sensory input-dependent adaptations in GABAergic inhibition and gephyrin clustering have been reported (37)(38)(39). Furthermore, during a narrow postischemic timeframe, synaptic plasticity plays an important role in the recovery process (40). Posttranslational modification like SUMOylation of cellular proteins are thought to contribute to the recovery process after ischemic insult (16). Although, elevated SUMO-conjugated proteins and BDNF levels after an ischemic stroke have been reported in literature (18), a functional link between BDNF and SUMO pathway has not been reported so far. Our study provides the first evidence linking BDNF signaling with the regulation of SUMO pathway.

Experimental procedures
All animal experiments were approved by the cantonal veterinary office of Zurich (ZH011/19). All experiments were performed in accordance with guidelines from the Swiss Veterinary office or Canadian Council on Animal Care and the National Institutes of Health in the USA. All animal procedures at McGill were approved by the Animal Resource Committee of the School of Medicine at McGill University Protocol number 5057.

Immunohistochemistry of primary cells culture
Seven days posttransfection, the cells (8 + 7DIV) were fixed in 4% paraformaldehyde for 10 min, then permeabilized for 5 min with 0.1% Triton X-100 in 10% normal goat serum (NGS, Bio-Rad, C07SA) and PBS, pH 7.4. The cells were quickly washed with PBS (pH7.4) before being labeled with the appropriate primary antibody cocktail (antibodies with 10% NGS and PBS) for 90 min. After three washes of 10 min each with PBS, the secondary detection was achieved with the secondary antibody mixture supplemented with DAPI (1:1000) for 30 min. The coverslips were mounted with Dako Fluorescence Mounting medium (Dako North America, Inc).

Image analysis and quantification
All images were acquired on confocal laser scanning microscope (LSM 710, Carl Zeiss) with objective lens of 40× (NA 1.4) with a pinhole set at 1 Airy unit and a pixel size of 0.13 μm. For each condition, images from a minimum of 9 cells from three independent batches of neuronal culture were acquired using a z-stack (3-5 steps at 0.5 μm per step size). From each cell, a dendritic segment was taken for analysis. Image analyses were performed with a custom written analysis for Image J software using maximal intensity z-projected images.
Gephyrin clustering size area and density were analyzed 7 days posttransfection in hippocampal primary neuronal culture following the protocol previously described (43,44). The generated data are then plotted using Excel software and GraphPad Prism software.

Statistical analysis
When multiple groups were compared using either two-way ANOVA or one-way ANOVA followed by a Bonferroni pairwise comparison as indicated and Mann-Whitney pair-wise comparison as indicated.

Immunoprecipitation and Western blot
Interaction between two proteins was determined using the heterologous cells HEK293. For the IP followed by Western blot (WB) assays, the cell lysates were incubated 90 min at 4 C with 1 to 2 μg purified antibody followed by incubation with protein A/G UltraLink Resin (Thermo Scientific, #53133) 45 min at 4 C. Unspecific binding to the resin was minimized by washing with EBC-based high-salt buffer (50 mM Tris, 500 mM NaCl, 1% NP-40) followed by washes with normal EBC buffer. The samples were boiled with SDS sample buffer containing 15% fresh β-mercaptoethanol at 90 C for 4 min and separated on appropriate acrylamide % SDS gel at 140 V. The proteins were transferred onto a PVDF membrane on which the WB could be performed. The membrane was blocked with 5% Western blocking reagent (Roche, #11921681001), then incubated with the primary antibody mixture for 3 h or overnight. After washing with Tris-buffered saline with Tween20 (TBS-T), the membranes were incubated with the secondary antibodies mixture containing either Donkey horse radish peroxidase antibodies (HRP 1:10,000, form Jackson ImmunoResearch: mouse #715-035-150 and rabbit #711-035-152) or fluorescent secondary's (1:30,000): mouse IR680 (#926-68022) or rabbit IR 800 (#926-32213) from Odyssey-AB/Li-COR. For loading controls, protein lysates were boiled with 5× SDS buffer before performing WB with the appropriate antibodies.
Miniature inhibitory postsynaptic currents (mIPSCs) were isolated by adding 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX) (25 μM, Merck), AP-5 (50 μM, Alomone Labs), and tetrodotoxin (1 μM, Affix Scientific). Cell were recorded at holding potential of −70 mV. Recordings were amplified by Multiclamp 700B amplifier and digitized with Digidata 1440 (Molecular Devices). Cells were recorded for a total duration of 5 min: after 3 min of establishing a stable whole-cell mode, mIPSCs were analyzed for the last 2 min. Only cells which showed a stable recording in the first 3 min, series resistance increase <30%, and signal above the noise background (5-10pA) were further analyzed. The decay time of mIPSCs was fitted with a single exponential curve and fitted between 10 and 90% of its amplitude. Events were recorded using Clampex 10.7 software (Molecular Devices) with sampling rate of 10 kHz and filtered offline using Bessel low pass filter (Clampfit 10.7) and analyzed using MiniAnalysis 6.0.7 (Synaptosoft).
Whole cell patch clamp recording in organotypic hippocampal slice culture All electrophysiological recordings were made using an Axopatch 200A amplifier (Molecular Devices). GABA A Rmediated mIPSCs were gathered from whole-cell voltageclamp recordings of CA1 pyramidal neurons obtained at 25 C using electrodes with resistances of 4 to 5 MΩ and filled with intracellular solution containing (in mM): CsCl, 140; NaCl, 4; 0.5, CaCl2; Hepes, 10; EGTA, 5; QX-314, 2; Mg-ATP, 2; Na-GTP, 0.5; and 290 mOsm, pH adjusted with CsOH to 7.36. mIPSCs were recorded at −60 mV and in the presence of 1 μM tetrodotoxin, 25 μM CPP, 5 μM CGP55845, 5 μm 6-cyano-7nitroquinoxaline-2,3-dione (CNQX), and 0.3 μm strychnine in external Tyrode's solution. Access resistance was monitored with brief test pulses at regular intervals (2-3 min) throughout the experiment. After the holding current had stabilized, data were recorded at a sampling frequency of 10 kHz and filtered at 2 kHz for 10 to 15 min. mIPSCs were detected offline using the Mini Analysis Software (Synaptosoft). The amplitude threshold for mIPSCs detection was set at four times the rootmean-square value of a visually event-free recording period.
From every experiment, 5 min of stable recording was randomly selected for blinded analysis of amplitude and interevent interval. The data obtained were then used to plot cumulative histograms with an equal contribution from every cell.

Organotypic hippocampal slice cultures
Organotypic hippocampal slices (400 μm thickness) were obtained from postnatal day 7 C57BL/6J mice or transgenic mice expressing MARCKs-enhanced GFP tagged to the CA1 neuronal membrane. Tissue slices of 400 μm thickness were prepared following the roller-tube method from Gähwiler technique (45). The slices were incubated in an antibiotic-free serum medium containing 25% heat-inactivated horse serum, 25% Hank's balanced salt solution, and 50% Basal Medium Eagle. They were maintained for 3 weeks minimum allowing maturation prior to experimentation at 36 C in a roller drum incubator.
Images were acquired on a Leica DM6000B laser scanning microscope (Leica Microsystems) with an objective lens of 63× NA 1.4 oil immersion. At least three slices from three independent batches per condition were acquired (0.3 μm z stack). Image analysis of gephyrin clustering in the hippocampal CA1 region were done, postdeconvolution with Huygens Essential software, using the Surpass and the Spot functions of Imaris 7.00 software (Biplane AG).

OGD treatment
The slices were incubated in glucose-free Tyrode (ACSF) solution supplemented with 2 mM 2-deoxyglucose, 8 mM sucrose, and 3 mM sodium azide (NaN 3 ) and bubbled with 95%N 2 /5%CO 2 . The slices were incubated during 4 min in the OGD solution or normal Tyrode solution (control conditions) and returned in normal culture medium for 90 min, 24 h before experimenting as a model for ischemic injury in vitro (46).

Immunohistochemistry of organotypic hippocampal slice cultures
Slices were fixed using 4% paraformaldehyde for 1 h and washed with 0.1 M phosphate buffer, subsequently permeabilized using 0.4% Triton x100 and blocked with 1.5% heat-inactivated horse serum overnight at 4 C. The primary antibody cocktail were incubated (in permeabilizing buffer) over 5 days at 4 C. The slices were then washed several times with 0.1 M PBS during the whole day, followed by the incubation with the secondary antibody mixture overnight at 4 C. Slices were mounted using Dako Fluorescence Mounting medium (Dako Canada).

Real-time qPCR
Areas CA1 and CA3 were microdissected from five to six slices from three independent litter and used for each experimental condition. Total mRNA was extracted using BioRad extraction kit. Subsequently, 1 μg of mRNA was reverse transcribed to cDNA following the manufacturer's protocol (Roche Diagnostic). The RT-qPCR was performed using 30 ng of cDNA in a 20 μl reaction mixture containing EVA green mastermix (Solis BioDyne #08-24-00008). All qPCR reactions were performed under those conditions: 40 cycles; denaturation at 95 C for 15 s, annealing at 62 C for 25 s, and extension at 72 C. Primers: the following primer pairs were used for each reaction: bdnf Fwd: 5 0 -TGC AGG GGC ATA GAC AAA AGG-3 0 , Rev: 5 0 -CTT ATG AAT CGC CAG CCA ATT CTC-3'; Gapdh Fwd: 5 0 -TGCCCCCATGTTTGTGA TG-3 0 Rev: 5 0 -TGTGGTCATCAGCCCTTCC-3'.

Data availability
The authors declare no restrictions on data availability.
Supporting information-This article contains supporting information.