{"paper_id":"085af901-4d24-407c-9271-67dcf3f09ece","body_text":"1 \n \nTitle \n• Coordinated pre- and postsynaptic protein dynamics underlie rapid Sema4D-mediated \ninhibitory synapse assembly \nAuthors \n• Zachary Pranske and Suzanne Paradis* \nAffiliations \n• Department of Biology, Brandeis University, Waltham, MA 02454, USA \n• *Corresponding Author: Suzanne Paradis, Department of Biology, Brandeis University, \nWaltham, MA 02454, USA. Email: paradis@brandeis.edu \n \nAbstract \nIn the mammalian hippocampus, synapses are either excitatory or inhibitory as defined by \nthe presynaptic neurotransmitter (glutamate or GABA, respectively) and the specific ligand-gated \nion channel receptors localized to the postsynaptic specialization. While numerous studies explore \nthe formation of excitatory synapses, the process of inhibitory synapse formation is less \nunderstood. Using both loss - and gain-of-function approaches, our lab previously identified the \nclass 4 Semaphorin Sema4D as a key regula tor of inhibitory synaptogenesis. Here, u sing \nrecombinant Sema4D protein as a tool to rapidly induce GABAergic synapse formation in cultured \nhippocampal neurons, we employ two -channel live imaging to identify changes to pre- and \npostsynaptic protein dynamics during inhibitory synapse formation. We f ind that Sema4D \ntreatment promotes the mobility of presynaptic GAD65 protein assemblies while having a \nnegligible effect on the behavior of the postsynaptic gephyrin scaffold, leading to increased \ncolocalization of these proteins. In addition, Sema4D  treatment promotes the recruitment of \nGABAARγ2 subunits to immature gephyrin scaffolds , suggesting that Sema4D primes these \nscaffolds for receptor recruitment . Surprisingly, we observe new colocalization events between \nexisting gephyrin and GABAAR puncta, suggesting that clustering of either the gephyrin scaffold \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n2 \n \nor the GABAAR is sufficient to nucleate assembly  of the postsynaptic specialization. Overall our \nresults support a model in which Sema4D signaling coordinates dynamic changes in both pre- and \npostsynaptic compartments to assemble inhibitory synapses on rapid timescales.  \nSignificance Statement \nThe assembly of new synaptic contacts requires precise coordination of specialized \nproteins in pre- and postsynaptic neurons. Inhibitory synapses, which suppress neuronal activity \nand are essential for circuit stability, contain distinct molecular components, yet the mechanisms \ngoverning their assembly remain poorly understood. We used Sema4D, a protein that rapidly \ninduces inhibitory synapse formation, as a molecular tool to dissect how synaptic proteins on \neither side of the synaptic cleft are coordinated in space and time. Using live imaging we show \nthat Sema4D acts on both pre- and postsynaptic compartments to recruit synaptic proteins with \nspatiotemporal precision. Together, these findings define the sequence of molecular events \nunderlying inhibitory synapse assembly and have implications for neurodevelopmental disorders \nin which inhibition is disrupted. \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n3 \n \nIntroduction \nSynapses are the core unit of cell–cell communication in the nervous system. Early studies \nof excitatory, glutamatergic synapse assembly in hippocampus revealed that synapse formation  \nbegins with the establishment of a  transient contact between an axon and a dendrite (Scheiffele, \n2003; Ziv and Garner, 2004) . Next, mobile, pre -clustered protein assemblies are localized to the \npresynaptic compartment (Ziv and Garner, 2004; McAllister, 2007), while postsynaptic maturation \nis marked by accumulation of  neurotransmitter receptors  and scaffolding proteins (Prange and \nMurphy, 2001; Bresler et al., 2004; Ziv and Garner, 2004). The action of signaling molecules such \nas LRRTMs, Teneurins, Neuroligin/Neurexins, and Semaphorin/Plexins is thought to be essential \nfor both initial contact formation and subsequent recruitment of synaptic proteins (Kuzirian et al., \n2013; McDermott et al., 2018; Südhof, 2018, 2021), although it remains broadly unclear at which \nstage(s) of synapse development these molecules act. \nProper regulation of inhibitory , GABAergic synapse assembly is essential for circuit \nfunction, but compared to glutamatergic synapse formation, only a handful of studies have directly \naddressed the steps involved using live  imaging approaches (Wierenga et al., 2008; Dobie and \nCraig, 2011; Kuriu et al., 2012; Villa et al., 2016; Frias et al., 2019) . Collectively these studies \nrevealed coordinated mobility and gradual accumulation of proteins associated with GABAergic \nsynapses at axon–dendrite crossings over several hours, but did not resolve molecular recruitment \nevents on the order of seconds to minutes . This limitation is significant, as the earliest stages of \nsynapse assembly are likely governed by rapid, transient molecular events. Thus, there remains a \nsubstantial gap in understanding how molecular signals acutely regulate cellular processes that \ntransform nascent contacts into mature GABAergic synapses. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n4 \n \nA major barrier to addressing this problem has been the lack of tools to manipulate \ninhibitory synapse formation with temporal precision. We discovered a rapid molecular trigger for \ninhibitory synapse formation: the extracellular domain of the transsynaptic signaling protein \nSemaphorin 4D (Sema4D) is sufficient to induce the formation of new inhibitory synapses between \nhippocampal neurons within 30 minutes  (Kuzirian et al., 2013; McDermott et al., 2018; Adel et \nal., 2023). These synapses become functional within 2 hours as revealed by in vitro and ex vivo \nphysiology and in vivo mouse models (Kuzirian et al., 2013; Acker et al., 2018; Adel et al., 2023). \nBoth loss -of-function (Paradis et al., 2007; Raissi et al., 2013)  and gain -of-function studies \n(Kuzirian et al., 2013; McDermott et al., 2018; Adel et al., 2023)  demonstrate that Sema4D  \nspecifically regulates GABAergic synapse formation without affecting glutamatergic synapses, \nmaking it a uniquely selective and temporally precise tool for studying inhibitory synaptogenesis. \nBeyond their developmental  roles in axon guidance, neuronal migration, and tissue \nmorphogenesis, Semaphorins and their receptors  have emerged as  critical mediators of \nsynaptogenesis (Paradis et al., 2007; Joo et al., 2013; Duan et al., 2014; Koropouli and Kolodkin, \n2014; Uesaka et al., 2014). Sema4D is a transmembrane protein that can be cleaved from the pre- \nor postsynaptic membrane, allowing its extracellular domain to signal either in cis or in trans to \nPlexin-B1 receptors (Raissi et al., 2013) . Plexin-B1 is required in both  the presynaptic and \npostsynaptic neuron for rapid Sema4D-induced synapse formation (McDermott et al., 2018; Adel \net al., 2024), suggesting that Sema4D/Plexin-B1 signaling may synchronously coordinate pre- and \npostsynaptic changes. \nIn this study, we identify synapse formation events by leveraging the unique ability of \nSema4D to drive GABAergic synapse formation with temporal precision combined with two-\nchannel live imaging of fluorescently-labeled GABAergic synaptic proteins. By exploiting \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n5 \n \nSema4D as a rapid, selective inducer of inhibitory synapse formation, we overcome the inherent \nasynchrony of synaptogenesis and directly dissect mechanisms that are otherwise difficult to \nresolve. This work is explicitly focused on the critical time window preceding the emergence of \nsynaptic functionality, thus providing a unique cell biological view of the molecular and \nstructural events that give rise to functional inhibitory synapses. Our results establish a model \nframework for the cellular processes which underlie inhibitory synapse assembly.   \nResults \nSema4D treatment alters dynamic behavior and promotes stabilization of GAD65 -GFP \npuncta \nWe first sought to characterize the dynamics of GABAergic presynaptic boutons in \nresponse to Sema4D treatment. To begin, we generated primary hippocampal neuronal cultures \nfrom P0-1 GAD65-GFP mice, in which a subset of interneurons express a transgenic GAD65-GFP \nfusion protein (López-Bendito et al., 2004). In these animals axons are identified by diffuse, low-\nintensity GFP expression along neuronal processes while presynaptic boutons are distinguishable \nas bright  GFP-positive puncta (Fig. 1A, S1, 2A). Previous characterization of this mouse line \ndemonstrated that GAD65 -GFP puncta represent genuine presynaptic GABAergic boutons \n(Wierenga et al., 2008; Poulopoulos et al., 2009; Schuemann et al., 2013; Frias et al., 2019) , and \nwe independently confirmed that these GFP-labeled puncta contain GAD65 by immunostaining \nwith a GAD65-specific antibody (Fig. S1).  \nDissociated GAD65-GFP hippocampal neurons were plated atop an astrocyte feeder layer \nand cultured for 10 days in vitro (DIV). At DIV10–11, we treated cultures acutely with 2 nM \ncontrol protein (Fc domain of human IgG1 recombinant protein) or 2 nM Sema4D-Fc (recombinant \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n6 \n \nprotein containing the soluble extracellular domain of Sema4D fused to the Fc domain of human \nIgG1). Immediately following addition of the recombinant protein s, we acquired images of \nGAD65-GFP positive neurons using a Nikon AX-R resonant scanning confocal microscope to \nobtain z-stacks of 5 Nyquist-sampled planes at 10-second intervals for 1 hour. \nGAD65-GFP puncta and the ir associated axons exhibited a range of dynamic behaviors \nincluding splitting events, merging events, rapid local protein cluster mobility, complex \nsplit/merge events, nascent axonal branching, and active growth cones (Fig. 1A). We manually \ncharacterized the behaviors of GAD65-GFP labeled presynaptic protein clusters by identifying and \nquantifying these events of interest (Fig. 1B; for complete definitions see Table S1).  Sema4D \ntreatment significantly decreased the frequency of complex split/merge events, defined as multiple \nor repeated splitting and merging behavior of one or more GAD65 -GFP puncta, compared to \ncontrol treatment (Fig. 1B). Sema4D treatment also decreased the number of active axonal growth \ncones observed, consistent with its previously demonstrated role in growth cone collapse (Oinuma \net al., 2004; Ito et al., 2006), thus confirming that Sema4D-Fc protein is active in our cultures. By \ncontrast the rates of single split, single merge, and locally restricted mobility events (mobile puncta \nmoving within a small radius without splitting or merging) of GAD65-GFP puncta, as well as the \nrate of nascent axonal branch formation, were unaffected. This suggests that Sema4D signaling \nmay have a role in stabilizing a specific subset of immature, mobile presynaptic protein assemblies \nthat undergo repeated split/merge events. \nTo track the mobility of GAD65 -GFP puncta quantitatively, we identified puncta at each \ntimepoint using a custom-trained machine learning segmentation model in the Imaris for Tracking \nsuite (Oxford Instruments)  and empirically determined that this rigorous approach produced the \nhighest fidelity puncta identification over time  (Fig. 2A). We then performed automated particle \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n7 \n \ntracking using Imaris and a custom analysis suite in MATLAB. We quantified GAD65-GFP \npuncta stability by calculating the percentage of imaging frames in which each individual puncta \nwas present, similar to prior studies of GAD65 -GFP boutons in organotypic slice (Frias et al., \n2019). Sema4D treatment shifted the distribution of puncta duration rightward (Fig. 2B), indicating \nthat GAD65 -GFP puncta were present at a greater proportion of timepoints. Thus, Sema4D \ntreatment appears to stabilize existing GAD65-GFP puncta.  \nGAD65-GFP puncta show altered mobility in response to Sema4D treatment \nTo characterize the mobility of individual GAD65-GFP puncta in response to Sema4D we \nfocused on GAD65-GFP puncta that were consistently tracked between t = 10 min. and t = 60 min. \nof the imaging session. Beginning the analysis at t  = 10 min . was necessary due to an initial, \ntransient rise in mean displacement distance for all GAD65 -GFP puncta within this time window \nin both treatment conditions, presumably due to addition of the protein to the culture media or \nimaging-induced cell motility . We first characterized overall GAD65 -GFP puncta mobility by \ntracking displacement from origin of every identified GAD65 -GFP puncta in the field of view \nstarting at t = 10 min. We found that Sema4D treatment led to an overall increase in mean GAD65-\nGFP puncta mobility beginning about 20 minutes after addition of Sema4D protein, as shown by \na significant increase in mean displacement in Sema4D-treated cultures compared to control from \napproximately 20–60 min. (Fig. 2C). We next compared the overall distributions of GAD65-GFP \npuncta mobility across the population at 30, 45, and 60 min. after Sema4D or control treatment. \nAcross both treatment conditions, the majority (approximately 90%) of GAD65-GFP puncta were \nrelatively immobile, displacing less than 2 µm from their initial position, consistent with the \npresence of stable protein clusters localized to presynaptic boutons (Fig. 2D). We observed that \noverall mean displacement was significantly increased in Sema4D -treated cultures at these \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n8 \n \ntimepoints (Fig. 2D). However,  the shape of the  cumulative distribution was similar between \ntreatment conditions, suggesting that Sema4D induces a modest, population -wide shift toward \ngreater puncta mobility rather than actuating a distinct subpopulation of highly mobile puncta.  \nDespite the similarity in overall population distribution, the most mobile GAD65 -GFP \npuncta moved a greater distance  in Sema4D -treated neurons compared to control  across all \ntimepoints (Fig. 2D). We therefore asked whether Sema4D-dependent changes in GAD65 -GFP \npuncta mobility were partly due to increased directionality (i.e. track straightness) of mobile \nGAD65-GFP puncta (Fig. S2A) , as mean and maximum GAD65-GFP puncta velocity did not \ndiffer between conditions (Fig. S2B, C). Analysis of track straightness revealed that while Sema4D \ntreatment did not affect the average track straightness of the entire population of puncta (Fig. 2Ei), \nthe most mobile subset of GAD65 -GFP puncta (top 5% by displacement within each condition) \nfollowed significantly straighter tracks in the Sema4D -treated cultures (Fig. 2E ii). By contrast, \ntrack straightness of the top 5% of GAD65 -GFP puncta by mean velocity did not differ between \ntreatment conditions  (Fig. 2Eiii) . Interestingly, the autoregressive parameter AR(1), which \nrepresents persistence in track directionality, was increased in the top 5% of GAD65-GFP puncta \nby mean velocity in Sema4D-treated cultures, suggesting that faster-moving GAD65-GFP puncta \nmoved more freely under Sema4D treatment (Fig. S2D). Overall, these data suggest that Sema4D \ntreatment drives more directed movement among a mobile subset of GAD65-GFP puncta, and that \nincreased GAD65 puncta displacement may in part be driven by more directional movement of a \nsubset of mobile presynaptic protein clusters.  \nStabilization of presynaptic protein clusters at new synaptic sites may be mediated by \naltered mobility of existing protein clusters as well as maturation of unstable, newly -formed \nclusters. A prior live imaging study with a 10 -minute temporal resolution r eported that Sema4D \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n9 \n \nstabilizes a subset of non -persistent boutons which were intermittently present before treatment \n(Frias et al., 2019) . Thus, while  our current analysis was confined to GAD65-GFP boutons that \nwere tracked stably across the entire 60-minute imaging session, we also considered the possibility \nthat Sema4D may act on the newly-formed GAD65-GFP puncta that appear ed after the onset of  \nimaging. To test whether Sema4D affects the mobility of this subset, we focused on GAD65-GFP \npuncta that emerged between t = 3 and t = 20 min. after imaging onset and remained stably present \nthereafter. While these newly-tracked puncta were generally more mobile than existing puncta \n(Fig. S2E), we did not observe a Sema4D -dependent effect on mean displacement of newly -\ntracked puncta at any timepoint (Fig. S2F).  \nOverall, the results from this imaging experiment indicate that a subset of GAD65 -GFP \nprotein clusters are mobile and display dynamic behaviors during the 1 hr . imaging session. \nSema4D-dependent changes to presynaptic GAD65 -GFP puncta mobility are time -dependent, \nwith an overall increase in mean puncta mobility beginning about 20 min. after Sema4D addition. \nIncreased mobility is accompanied by increased puncta stability and a decrease in the frequency \nof complex split/merge events, suggesting that Sema4D pr omotes both increased exploratory \ndynamics and maturation of presynaptic sites. \nSema4D does not affect overall mobility of postsynaptic gephyrin \nWe next asked whether Sema4D similarly affects the mobility of postsynaptic proteins \nassociated with GABAergic synapses. We generated primary hippocampal cultures from P0 –1 \nGAD65-GFP mice as described above. At DIV2 these cultures were infected with an AAV9 virus \nexpressing HaloTag-Gephyrin under the control of the neuronal hSyn promoter (Halo-Gephyrin), \nallowing for visualization of postsynaptic scaffold assemblies (Fig. 3A; Fig. S3A). We confirmed \nthat virally expressed Halo-Gephyrin localizes to synapses as revealed by co-immunostaining with \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n10 \n \na GAD65-specific antibody (Fig. S3A), and while gephyrin puncta density is marginally increased \nin these cultures, baseline GABAergic synapse density was unaffected (Fig. S4). At DIV10–11 we \nlabeled cultures with cell-permeable Janelia Fluor 646 (JF646) HaloTag ligand to visualize Halo-\nGephyrin, then treated with 2 nM Sema4D or control protein and acquired images at 15s intervals \nfor up to 1h using a resonant scanning confocal. As before, we utilized a custom -trained machine \nlearning segmentation network, automated particle tracking in Imaris, and custom MATLAB -\nbased analysis software to analyze the mobility of Halo-Gephyrin puncta (Fig. 3A). In contrast to \nGAD65-GFP puncta, we observed no  effect of Sema4D treatment on Halo-Gephyrin puncta \nstability over time (Fig. 3B) . Similar to GAD65 -GFP, 90–95% of Halo-Gephyrin puncta were \nrelatively immobile, displacing less than 2 µm from their starting location (Fig. 3D). However, we \ndid not observe a Sema4D-dependent effect on the distribution of Halo-Gephyrin puncta mobility \nat any timepoint (Fig. 3D ), nor did we observe  any appreciable Sema4D -dependent changes to  \nHalo-Gephyrin puncta velocity (Fig. S3B, C) or to indicators of directed motion such as gephyrin \ntrack straightness or autoregressive  parameter (Fig. 3E, S3D). Thus, we concluded that Sema4D \ndoes not affect the overall mobility of gephyrin scaffolds within our imaging window . Taken \ntogether these data suggest that Sema4D promotes GABAergic synapse formation primarily by \naltering the behavior of the presynaptic terminal. \nSema4D treatment drives gephyrin localization to postsynaptic sites adjacent to GAD65-\nGFP labeled boutons \nOur previous studies using immunostaining in fixed cells of proteins localized to \nGABAergic synapses demonstrated that Sema4D promotes increased density of GAD65 puncta \ncolocalized with GABAARγ2 and of synapsin puncta colocalized with  gephyrin (Kuzirian et al., \n2013). Here we sought to determine the time course  of colocalization between pre - and \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n11 \n \npostsynaptic markers of GABAergic synapses that occurs in response to Sema4D. We first asked \nwhether Sema4D treatment promotes increased localization of Halo-Gephyrin to sites marked by \nGAD65-GFP puncta. For this analysis we focused primarily on stably tracked GAD65 -GFP \nboutons that were present throughout the 1 hr . imaging session. Because GAD65 and gephyrin \nshow largely overlapping point -spread functions at standard confocal resolution, we used this \noverlapping signal as an indicator of colocalization between Halo-Gephyrin postsynaptic puncta \nand GAD65-GFP labeled presynaptic boutons  (Fig. 4A). We observed an increase in the mean \nfluorescence intensity of the Halo-Gephyrin channel within the area defined by GAD65 -GFP \npuncta (hereafter referred to as “colocalized gephyrin fluorescence”) across time. At the population \nlevel, we found that Sema4D treatment significantly increased mean colocalized gephyrin \nfluorescence beginning at about 45  min. (Fig. 4Bi), consistent with our previous results in fixed \ncells. \nTo assess whether Sema4D-dependent recruitment of gephyrin to GAD65 -GFP puncta  \nvaried as a function of the initial degree of postsynaptic association , we analyzed subsets of \nGAD65-GFP puncta grouped by their baseline levels of colocalized gephyrin fluorescence.  The \ntop quintile, characterized by greater colocalized gephyrin fluorescence  at t = 0 min., showed a \ngradual decrease in colocalized gephyrin fluorescence over time in the control condition; Sema4D \ntreatment prevented the loss of gephyrin fluorescence from this subset (Fig. 4Bii). By contrast, the \nbottom quintile of GAD65 -GFP puncta , characterized by little -to-no colocalized gephyrin \nfluorescence initially, showed gradually increasing colocalized gephyrin fluorescence over time in \ncontrol and Sema4D -treated neurons. Sema4D treatment led to further increased colocalized \ngephyrin fluorescence beyond the level seen in control neurons, suggesting additional recruitment \nof gephyrin to this subset of GAD65-GFP puncta in response to Sema4D treatment  (Fig. 4B, iii). \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n12 \n \nAcross the entire dataset, we found a negative correlation between baseline gephyrin fluorescence \nand the change in gephyrin fluorescence from t = 0 to t = 60 min.; as expected, Sema4D treatment \nsignificantly reduced the strength of this correlation  (Fig. 4C). Overall, these data suggest that \nSema4D acts to 1) stabilize and prevent the loss of gephyrin from sites of colocalization with \nGAD65 and 2) promote increased colocalization of gephyrin at GAD65 -positive sites lack ing \ngephyrin. \nWe hypothesized that increased colocalization between GAD65 -GFP and Halo-Gephyrin \npuncta could occur via migration of existing GAD65 -GFP or Halo-Gephyrin puncta to new sites \n(Fig. 4Ai,ii) and/or via accumulation of Halo-Gephyrin at sites where GAD65 -GFP was already \npresent (Fig. 4A iii,iv). To examine the se possibilities, we measured the mean nearest  neighbor \ndistance of GAD65-GFP puncta, defined as the minimum distance between a GAD65-GFP puncta \nand the nearest Halo-Gephyrin puncta at any given time. Because over 90% of persistently-tracked \nGAD65-GFP and Halo -Gephyrin puncta displaced 2 µm or less (Fig s. 2,  3), we focused on  \nGAD65-GFP puncta that were initially within 2 µm of a gephyrin puncta. Sema4D treatment \ndecreased the mean nearest neighbor distance of GAD65-GFP boutons beginning at about 45 min. \ncompared to control treatment (Fig. 4D). We hypothesized that increased proximity between \nGAD65 and gephyrin resulted from new colocalization events in which previously non-colocalized \nGAD65-GFP and Halo-Gephyrin puncta became colocalized with each other during the imaging \nsession. \nWe next used our two-channel particle tracking data to identify new colocalization events \nbetween GAD65 and gephyrin. These events were considered genuine new colocalization events  \nonly if they produced stable new colocalization between a pair of GAD65-GFP and Halo-Gephyrin \npuncta that persisted for  at least 10 minutes. New colocalization events  fell into two distinct \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n13 \n \ncategories: colocalization events between existing GAD65-GFP puncta and newly-emerging Halo-\nGephyrin puncta (“new gephyrin”) and colocalization events between existing GAD65 -GFP \npuncta and  stable, pre -existing gephyrin  puncta (“existing gephyrin”).  We found that in both \ncontrol and Sema4D-treated neurons, colocalization events with new gephyrin puncta comprised \nabout 75% of the total new colocalization events (Fig. 5A). Sema4D treatment significantly \nincreased the frequency of colocalization events between GAD65 -GFP and new gephyrin puncta \nbut not between GAD65-GFP and existing gephyrin puncta (Fig. 5B). These data suggest that one \nway in which colocalized gephyrin fluorescence is increased  in response to Sema4D treatment is \nvia new colocalization events between new gephyrin puncta and existing GAD65-labeled sites.  \nGAD65-GFP puncta that undergo new colocalization events show distinctive profiles of \nmobility and proximity to gephyrin puncta \nTo better understand the spatiotemporal dynamics of GAD65 protein clusters that undergo \nnew colocalization events, we used principal component analysis (PCA) to determine which \nparticle tracking parameters most strongly characterize the variability in individual GAD65-GFP \npuncta (Table S2). We reasoned that GAD65-GFP puncta that undergo new colocalization events \ncould comprise a distinctly identifiable subset of puncta in principal component space based on a \ndistinctive profile of mobility and proximity to gephyrin puncta . W e therefore compared their \ndistribution along PC1 and PC2 , the principal axes capturing the greatest variance in puncta \nfeatures, to that of the overall population. PC1 loadings were highest for measures of GAD65-GFP \npuncta mobility, including mean velocity and acceleration, while PC2 loadings were highest for \nmeasures of GAD65 -GFP proximity to gephyrin puncta, including initial and mean nearest \nneighbor distance to gephyrin . We found that  GAD65-GFP puncta that undergo colocalization \nevents tend to be  relatively immobile and are located relatively near to existing Halo-Gephyrin \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n14 \n \npuncta (Fig. S5A). Surprisingly, GAD65-GFP puncta that colocalized with new gephyrin puncta \nshowed no difference in average velocity compared to other GAD65-GFP puncta in either control \nor Sema4D-treated cultures, suggesting that additional aspects of mobility are required to predict \nnew colocalization events (Fig. S5B). In each treatment condition, mean initial nearest  neighbor \ndistance to gephyrin was significantly lower for GAD65 puncta with a new colocalization event, \nsuggesting that even though these p uncta colocalized with gephyrin puncta  that emerged later \nduring the imaging session , local proximity to other existing gephyrin puncta at the onset of \nimaging was strongly predictive of new colocalization events (Fig. S5C). This suggests that  \nGAD65 puncta undergoing colocalization events are localized to “hot spot” regions of synapse \nassembly where more gephyrin is present to be recruited from nearby sites to form new synapses.  \nWe performed the same analysis for new colocalization events in which GAD65 -GFP \npuncta colocalized with an existing Halo -Gephyrin puncta. These events were relatively less \ncommon, only comprising about 25% of new colocalization events  (Fig. 5A); due to the smaller \nsample size, we could not separately analyze control and Sema4D treatment conditions for this \nsubset of events. Similar to GAD65-GFP puncta that colocalized with new gephyrin puncta, the \nGAD65-GFP puncta that colocalized with existing gephyrin pun cta displayed lower average \nmobility and greater initial proximity to gephyrin compared to other GAD65 -GFP puncta (Fig. \nS5D). These GAD65-GFP puncta had significantly lower mean velocity compared to puncta \nwithout a colocalization event (Fig. S5E), and initial nearest neighbor distance to existing gephyrin \nwas zero in every case in which we observed a new colocalization event (Fig. S5F), suggesting \nthat these GAD65 puncta were initially colocalized with existing gephyrin, temporarily lost \ncontact, and then colocalized again. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n15 \n \nTaken together, these results suggest that features of GAD65-GFP puncta such as mobility \nand proximity to postsynaptic sites marked by gephyrin are partially predictive of whether stable \nnew colocalization will occur. Sema4D treatment significantly increased the overall probability of \nnew colocalization events, primarily by enhancing the formation of stable associations with newly \nemerging gephyrin puncta. Overall, these analyses indicate that GAD65 boutons that form new, \nstable colocalization events are defined primarily by their initial proximity to gephyrin -rich \nregions, which strongly predicts whether they will recruit new postsynaptic scaffold. \nSema4D treatment mobilizes GAD65 -GFP puncta  prior to  colocalization with  Halo-\nGephyrin \nGiven that Sema4D increases the overall mobility of GAD65-GFP but not Halo-Gephyrin \npuncta, and that Sema4D increases the probability of new colocalization events between GAD65 \nand gephyrin, we hypothesized that presynaptic protein clusters define the sites of new GABAergic \nsynapses. To test this directly we performed an event-level analysis of paths followed by GAD65-\nGFP or Halo -Gephyrin puncta before and after new colocalization events  (Fig. 6A, E) . For this \nanalysis, all new colocalization events (including those involving new and existing gephyrin \npuncta) were combined for each treatment condition.  GAD65-GFP puncta that colocalized with \nHalo-Gephyrin puncta displaced significantly farther from their origin prior to colocalization in \nSema4D-treated cultures compared to control (Fig. 6B). However, when we examined the mean \ndistance of GAD65 -GFP puncta from the site of colocalization  specifically in the 10-minute \nwindow prior to colocalization events, there was no difference between control and Sema4D -\ntreated cultures (Fig. 6C), suggesting that Sema4D-dependent changes to GAD65 mobility occur \nin an earlier window relative to when GAD65 puncta become colocalized with gephyrin.  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n16 \n \nTo determine whether GAD65-GFP puncta are moving towards pre-determined locations \nto establish new colocalization sites, we analyzed the change in GAD65 -GFP displacement from \norigin during the 10-minute window before and the 10 -minute window after colocalization. We \nreasoned that if a new colocalization event was driven by a GAD65-GFP puncta moving to a pre-\ndetermined location along the neurite to colocalize with a gephyrin puncta (or vice versa), GAD65-\nGFP pun cta would show greater average displacement after colocalization vs. before \ncolocalization, indicating the GAD65-GFP puncta had moved farther from its origin towards the \npre-determined location. GAD65-GFP puncta d isplacement was slightly (albeit significantly) \nincreased after colocalization in control neurons, but not in Sema4D -treated neurons, confirming \nthat the Sema4D-dependent increase to GAD65 -GFP puncta  mobility occur s during an earlier \nwindow (Fig. 6D). In addition, because Sema4D treatment did not significantly change GAD65 -\nGFP displacement after colocalization vs. before, we conclude that Sema4D tre atment does not \ndirect GAD65 -GFP towards specific pre -determined locations along the axon. Notably, in the \nSema4D-treated condition, several GAD65 -GFP puncta moved more than 1 µm prior to a new \ncolocalization event, a phenomenon which was not observed in control cultures (Fig. 6A, D).  \n We next performed the same analysis for Halo -Gephyrin puncta that became stably \ncolocalized with a GAD65-GFP puncta. In contrast to GAD65 -GFP puncta, Sema4D treatment \nhad no effect on Halo-Gephyrin puncta displacement prior to colocalization (Fig . 6F), although \nHalo-Gephyrin puncta were on average farther from the colocalization site in the 10 -minute \nwindow prior to colocalization in Sema4D -treated cultures compared to control (Fig. 6 G). This \nsuggests that gephyrin puncta mobility is influenced by Sema4D signaling in the immediate \nwindow prior to colocalization, but not before , possibly because the presence of a nearby \npresynaptic bouton is required for the gephyrin puncta to mobilize. Halo-Gephyrin puncta also did \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n17 \n \nnot show greater displacement from origin after colocalizing with GAD65 -GFP vs. before \ncolocalizing in either treatment condition, suggesting that gephyrin scaffold mobility is not \nspecifically directed towards pre-determined locations along dendrites (Fig. 6H). Taken together, \nthese results suggest that GAD65-GFP mobility occurs first in the sequence of events underlying \nSema4D-dependent synapse formation, followed by later gephyrin recruitment, and that mobile \nGAD65-GFP puncta likely encounter gephyrin assemblies stochastically. \nRecruitment of GABA ARγ2 to gephyrin scaffolds is increased in response to Sema4D  \ntreatment \nRecruitment and stabilization of GABAARs at postsynaptic gephyrin scaffolds is an \nessential step in GABAergic synapse maturation , and gephyrin has been shown to  regulate \nGABAAR clustering and surface expression (Mukherjee et al., 2011; Petrini et al., 2014). Previous \nwork from our lab demonstrated that the density of GABAARγ2-positive synapses is increased in \nresponse to Sema4D treatment and that newly-formed GABAergic synapses are functional within \n2 hours (Kuzirian et al., 2013), suggesting that functional GABAARs are localized to newly-formed \nsynapses. However, gephyrin–GABAAR colocalization has not been assessed directly in response \nto Sema4D treatment , and the spatiotemporal dynamics of their association during synapse \nformation are poorly  understood. To this end, w e co-transfected plasmids expressing GFP-\nGephyrin and HaloTag-GABAARγ2 subunit (Halo-γ2) into cultured wild-type E18 rat neurons at \nDIV 4 and performed live imaging at DIV10–11 as before using cell-permeable JF646 HaloTag \nligand to label the Halo-γ2 protein (Figs. S6A, 7A). As with the virally expressed Halo-Gephyrin \nconstruct, w e observed discrete clusters of GFP -Gephyrin, whereas Halo-γ2 expression was \nsignificantly more variable. In line with prior observations (Christie et al., 2006), a subset of GFP-\nGephyrin/Halo-γ2 co -transfected neurons showed both dispersed and punctate Halo-γ2 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n18 \n \nlocalization along dendrites (Fig. S6A), with the remaining co-transfected neurons showing only \nweakly dispersed Halo-γ2 signal in dendrites, perhaps representing less mature or more weakly \ntransfected neurons . GFP-Gephyrin and Halo -γ2 puncta generally colocalized, and fixed \nimmunostaining experiments showed that overexpression of these markers did not affect baseline \nGABAergic synapse density  or interfere with synaptic localization of Halo -γ2 puncta (Fig. S7, \nS8). Thus, we proceeded with  live imaging analysis for co-transfected cells with clear punctate \nHalo-γ2 fluorescence. \nWe first analyzed the mean fluorescence intensity of  Halo-γ2 colocalized with  GFP-\nGephyrin as a measure of receptor density at postsynaptic scaffolds  (Fig. 7B), similar to our \nprevious analysis of GAD65 and gephyrin  colocalization (Fig. 4) . We found that the mean \ncolocalized Halo-γ2 fluorescence rapidly increased in response to Sema4D treatment within about \n10 minutes of application. By 1 hour, the mean increase in colocalized Halo-γ2 fluorescence puncta \nwas approximately 10–15% above baseline in Sema4D treated cultures  (Fig. 7B i). Sema4D \ntreatment did not affect the mean intensity of individual Halo-γ2 puncta or the total dendritic \nintensity in the Halo -γ2 channel (Fig. S 6B, C ); thus, the increase in colocalized Halo-γ2 \nfluorescence is likely due to increased recruitment of Halo-γ2 to gephyrin-positive postsynaptic \nscaffolds rather than increased GABAAR expression. \nBecause individual gephyrin scaffolds vary widely in their GABA AR content, we next \nasked whether the Sema4D -dependent increase in colocalized Halo -γ2 fluorescence was driven \npreferentially by recruitment to scaffolds with initially low receptor levels. We analyzed the mean \ncolocalized Halo-γ2 fluorescence in the subset of GFP-Gephyrin puncta in the top quintile of \nbaseline Halo-γ2 fluorescence vs. GFP-Gephyrin puncta in the bottom quintile of baseline Halo-\nγ2 fluorescence. Sema4D treatment significantly increased colocalized Halo-γ2 fluorescence at \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n19 \n \ngephyrin scaffolds in the lowest quintile of baseline receptor expression (Fig. 7Bii) while having \nno effect on gephyrin scaffolds in the top quintile (Fig. 7Biii), suggesting that Sema4D-dependent \nGABAAR recruitment was primarily directed toward  postsynaptic specializations  that lacked \nreceptors. In neurons treated with Fc control protein there was a slight negative correlation between \nbaseline Halo-γ2 fluorescence colocalized with GFP-Gephyrin puncta and the change in Halo-γ2 \nfluorescence over time (Fig. 7C). Sema4D treat ment enhanced this negative correlation, again \nsuggesting that Sema4D preferentially promotes recruitment of Halo-γ2 receptors to postsynaptic \nsites with low baseline GABA AR density . Additionally, Sema4D treatment reduced the mean \nnearest neighbor distance between GFP-Gephyrin and Halo-γ2 puncta relative to control (Fig. 7D). \nTogether these results suggest that Sema4D increases Halo -γ2 localization at gephyrin -labeled \nscaffolds primarily by redistributing existing receptors to scaffolds lacking GABAARs rather than \nby altering overall Halo-γ2 expression levels or receptor density at existing postsynaptic sites. \nSema4D treatment increases r ecruitment of GABA ARγ2 to gephyrin scaffolds without \ndriving new colocalization events between GABAARγ2 and GFP-Gephyrin puncta \nWhile most Halo-γ2 puncta were localized to postsynaptic specializations marked by GFP-\nGephyrin, a subset of these puncta appeared to move independently of gephyrin. Pre-clustered \nGABAAR assemblies that are not associated with gephyrin have been described previously \n(Danglot et al., 2003; Jacob et al., 2005; Christie et al., 2006) , but their role in synapse formation \nis unclear. One possibility is that these clusters may act as pre-assembled packets of receptors that \ncan be rapidly recruited to synapses when required. We hypothesized that Sema4D treatment \nwould increase colocalization events between pre-clustered GABAAR puncta and gephyrin puncta. \nSimilar to our previous analysis of GAD65 and gephyrin colocalization (Fig. 5), we identified \nGFP-Gephyrin puncta that were not initially colocalized with a Halo -γ2 puncta but became \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n20 \n \ncolocalized during the imaging session and remained stably colocalized for at least 10 minutes. \nWe found that in contrast to GAD65 and gephyrin, in which the majority of new colocalization \nevents involved emergence of a new gephyrin puncta, most new coloca lization events between \nGFP-Gephyrin and Halo-γ2 puncta (~80% of events) were between pairs of pre-existing protein \npuncta (Fig. 8A). We also found that a substantial proportion of GFP-Gephyrin puncta (20-25%) \ncolocalized with a previously independent Halo-γ2 puncta (Fig. 8Bi). Surprisingly, however, new \ncolocalization events occurred at a similar frequency in both control and Sema4D-treated neurons \n(Fig. 8Bii,iii), suggesting that Sema4D does not drive localization of independent GABA AR \nclusters to postsynaptic scaffolds. Thus, the occurrence of new colocalization events between GFP-\nGephyrin and Halo-γ2 puncta fail s to explain the Sema4D -dependent increase in total Halo -γ2 \nrecruitment.  \nSimilar to our previous analysis of GAD65 -GFP, we used principal component analysis \n(PCA) to determine which GFP-Gephyrin particle tracking parameters most strongly predict new \ncolocalization events with Halo -γ2 puncta . As before, PC1 and PC2 most strongly weighted \nparameters related to GFP -Gephyrin mobility and proximity to Halo -γ2 puncta, respectively \n(Table S3). GFP-Gephyrin puncta that colocalized with Halo -γ2 clusters showed a moderate \ntendency towards decreased mobility along PC1 and decreased nearest neighbor distance to Halo-\nγ2 puncta along PC2 (Fig. S9A,D). While mean velocity of GFP-Gephyrin puncta that colocalized \nwith Halo-γ2 puncta did not differ from that of other GFP-Gephyrin puncta (Fig. S9B,E), GFP-\nGephyrin puncta that colocalized with Halo-γ2 puncta were significantly closer to Halo-γ2 puncta \non average (Fig. S9F), suggesting they are located in receptor-rich regions of the dendrite. \nRecruitment of GABAARs that were not previously associated with GFP-Gephyrin puncta \nunderlies Sema4D-dependent postsynaptic maturation \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n21 \n \n We identified a third population of GFP-Gephyrin puncta in which colocalized Halo-γ2 \nfluorescence increased at least 1.5-fold during the imaging session regardless of whether a new \ncolocalization event occurred. We found that, compared to control, Sema4D significantly \nincreased the fraction of GFP-Gephyrin puncta that underwent a >1.5-fold increase in \ncolocalized Halo-γ2 fluorescence from about 4% to 11.5% (Fig. 8C). Next, we examined \nmobility characteristics for the population of GFP-Gephyrin that underwent a >1.5-fold increase \nin colocalized Halo-γ2 fluorescence using PCA. Unlike the previous GFP-Gephyrin population \nthat underwent new colocalization events (Fig. S9A-F), these GFP-Gephyrin puncta showed no \nclear clustering along either PC1 or PC2 (Fig. S9G) and they did not differ from other GFP-\nGephyrin puncta or between treatment conditions in their mean velocity (Fig. S9H) or proximity \nto Halo-γ2 puncta (Fig. S9I). Thus, gephyrin puncta that recruit GABAARγ2 in response to \nSema4D treatment are not necessarily localized to GABAAR-rich regions where many receptor \nclusters are already present. Taken together these results suggest two main routes to receptor \nlocalization at postsynaptic scaffolds: 1) colocalization between existing gephyrin scaffolds and \npre-formed GABAAR clusters in receptor-rich regions, and 2) recruitment of receptors that were \nnot previously clustered to sites where gephyrin is present, with only the latter being regulated by \nSema4D treatment. \nNew colocalization events between GFP-Gephyrin and Halo-GABAARγ2 may be established \nby clusters of either protein \nThe canonical view posits that relatively stable gephyrin scaffolds recruit mobile \nGABAARs to postsynaptic specializations , leading to receptor immobilization and synapse \nmaturation. Colocalization events between gephyrin and pre-clustered GABAAR puncta were \nfrequent (Fig. 8B) , and while Sema4D treatment did not specifically drive these events , we \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n22 \n \nreasoned that our live imaging approach could nonetheless give us unique insight into this process. \nWe analyzed the paths followed by GFP -Gephyrin and Halo -γ2 puncta before and after new \ncolocalization events, beginning by tracking the total displacement distance of GFP-Gephyrin and \nHalo-γ2 clusters prior to colocalizing (Fig. 9A, E). In Fc control-treated neurons we observed that, \nsurprisingly, GFP-Gephyrin and Halo -γ2 puncta moved similar distance s from their original \nlocation prior to colocalization, with few puncta moving more than 2 µm prior to colocalization in \nall conditions (Fig. 9B, F). These data suggest that either gephyrin or GABAAR cluster mobility \ncan promote a new colocalization event. We observed that Sema4D treatment led to a marginally \nsignificant increase in the mean distance traveled by GFP-Gephyrin puncta prior to colocaliz ing \n(p = 0.0545)  (Fig. 9B)  and a marginally significant decrease in the  average distance of GFP -\nGephyrin to colocalization sites in the 10-minute window prior to colocalization (p = 0.0553) (Fig. \n9C). This suggests that Sema4D treatment increases mobility of this subset of GFP-Gephyrin \npuncta earlier than 10 minutes before colocalization with pre-clustered Halo-γ2 puncta. We also \nobserved that GFP-Gephyrin was on average f arther from its origin location after colocalization \ncompared to before colocalization in control neurons, but this effect was absent in Sema4D-treated \ncultures (Fig. 9D) suggesting, in agreement with our analysis of GAD65-GFP and Halo-Gephyrin \n(Fig. 6), that Sema4D signaling does not direct mobile gephyrin puncta to specific pre-determined \nlocations on the dendrite. \n We performed similar analyses on Halo -γ2 mobility before and after colocalization with \nGFP-Gephyrin and found no difference in any of these parameters between control and Sema4D-\ntreated cultures (Fig. 9 F-H). Taken together these data suggest that, contrary to the canonical \nmodel, both gephyrin and GABAAR protein clusters are mobile and are equally capable of \ninitiating a colocalization event . These new colocalization events a ppear to occur stochastically \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n23 \n \nand do not require Halo-γ2 puncta to be recruited to pre -established sites. Overall, these \nexperiments suggest a model in which mobile preformed clusters of gephyrin and GABAARs may \nencounter each other to form new postsynaptic specializations. While Sema4D modulates gephyrin \nmobility prior to colocalization, the overall rate of colocalization events is not significantly \nenhanced; rather, enhanced recruitment or capture of individual GABAARs by relatively immature \npostsynaptic scaffolds is the primary mechanism driving Sema4D -mediated receptor recruitment \nand synapse maturation.  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n24 \n \nDiscussion \nCompared to excitatory synapse formation, inhibitory synapse formation has historically \nbeen more challenging to study due to the lack of an extensive postsynaptic density for biochemical \npurification and the relatively low abundance of inhibitory synapses. The spatiotemporal dynamics \nof inhibitory synapse assembly are poorly understood, particularly at acute timescales, as the vast \nmajority of assays employed to study the function of synaptogenic molecules have assessed \nsynapse formation by performing a ma nipulation (e.g. gene knockout) and assaying the presence \nor absence of synapses by microscopy or electrophysiology. These retrospective approaches likely \nobscure significant nuances in the spatiotemporal dynamics of synaptic protein cluster formation, \nstabilization at sites of colocalization, and maturation. Thus there remains a significant gap in \nunderstanding the process by which molecular signals and cellular processes transform nascent \ncontacts into mature inhibitory synapses. \nOver the past decade we defined a novel role for class 4 Semaphorins and Plexin -B \nreceptors in regulating inhibitory synapse formation in rodent hippocampus. Specifically, our lab \ndemonstrated that the soluble, extracellular domain of Sema4D induces inhibi tory synapse \nformation on a rapid time scale (i.e. minutes) while having no effect on excitatory synapse \nformation. Among the handful of identified transsynaptic regulators of GABAergic synapse \nassembly, which include the Neuregulin -ErbB4, Neuroligin -Neurexin, Slitrk -PTPδ, FGF, and \nDystroglycan families (Levinson et al., 2005; Krivosheya et al., 2008; Takahashi et al., 2012; Yim \net al., 2013; Dabrowski et al., 2015; Trotter et al., 2023), only Sema4D has these unique properties. \nThis rapid, selective, and inducible effect enables precise dissection of synaptogenic mechanisms \nthat are otherwise difficult to study due to the asynchronous and developmentally protracted nature \nof synaptogenesis. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n25 \n \nIn this study we leveraged the ability of Sema4D to induce GABAergic synapse formation \non the scale of minutes coupled with two -channel live imaging to characterize the dynamics of \nsynaptic proteins during Sema4D-mediated synapse assembly. Our data support a model in which \nSema4D treatment increases mobility of presynaptic GAD65-containing protein clusters, allowing \nGAD65 puncta to explore a wider radius to establish sites of putative new synapses . In contrast, \npostsynaptic gephyrin-containing protein clust ers are  mobilized only once they are i n the \nimmediate vicinity of a presynaptic bouton where GAD65 is present. Our findings agree with \nconverging evidence from multiple groups that the presynaptic compartment primarily initiates \nGABAergic synapse formation during early development (Wierenga et al., 2008; Dobie and Craig, \n2011; Kuriu et al., 2012) . These coordinated dynamics suggest Sema4D  primarily drives \nGABAergic synapse formation  by increasing the likelihood of stable colocalization of nearby \nsynaptic protein clusters that are poised to be recruited to existing contacts in the presence of a \nsynaptogenic signal.  Only a specific subset of protein clusters that are present near existing \ncontacts appears to be available to form new synapses , and synaptogenic signaling pathways  act \non these protein clusters to rapidly assemble synapses at acute timescales. \nThe canonical model of inhibitory postsynaptic assembly which emerged from early single-\nparticle tracking and FRAP experiments posits that gephyrin scaffolds are stable structures which \ncapture laterally diffusing GABAARs upon synaptic entry (see e.g. Jacob, Bogdanov et al. 2005 ). \nSubsequent biochemical work supported this anchoring model, showing that gephyrin and \ncollybistin directly bind GABAARs to stabilize receptor localization (Tretter et al., 2008; Hines et \nal., 2018; Lorenz-Guertin and Jacob, 2018). Our live imaging data support this general framework \nbut reveal a more dynamic view of postsynaptic assembly. We found that Sema4D promotes \nrecruitment of GABAARs to less mature gephyrin scaffolds with an abundance of available binding \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n26 \n \nsites rather than enhancing receptor clustering at sites where many receptors are already present . \nNotably, Sema4D did not alter the frequency of new colocalization events between pre-assembled \ngephyrin and GABAARγ2 puncta (Fig. 8), indicating that the effect of Sema4D is primarily to \nstabilize or enhance accumulation of individual receptors at postsynaptic sites lacking receptors.  \nInterestingly, we observed that both gephyrin and GABAARγ2 could establish new sites of \ncolocalization by moving toward putative new postsynaptic sites , and that this process plays an \nimportant role in postsynaptic assembly during development. Recent work suggests that gephyrin \nturnover and assembly is dynamically regulated by multiple binding (GlyR, GABARAP) and \nphosphorylation sites (via Erk1/2, GSK-3, Cdk5) (Petrini and Barberis, 2014; Choii and Ko, 2015; \nChapdelaine et al., 2021), and that gephyrin assembles into filaments that phase separate to allow \nfor flexible rearrangement of postsynaptic structures rather than forming a rigid lattice as \npreviously believed (Macha et al., 2025) . Thus, gephyrin mobility appears to be more dynamic \nthan originally appreciated. These observations challenge the traditional view that receptor \nclustering is strictly secondary to scaffold formation and raise the possibility that, in some contexts, \nGABAAR clustering may precede or even initiate recruitment of mobile gephyrin assemblies to \ndeveloping inhibitory synapses. \nThe molecular mechanisms linking Sema4D signaling to local recruitment of synaptic \nproteins remain largely unclear. Plexin-B1, the high-affinity ligand of Sema4D, is expressed both \npre- and postsynaptically and is required in both compartments for proper recruitment of synaptic \ncomponents (McDermott et al., 2018; Adel et al., 2024) . Whether Sema4D/Plexin -B1 signaling \nacts simultaneously on both sides of the synapse, or whether signaling from the pre- or postsynaptic \ncompartment acts on the other compartment indirectly, is unknown. Our findings in this study \nsuggest that Sema4D-dependent changes to presynaptic mobility precede localization of gephyrin \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n27 \n \npuncta to new contact sites. This implies that Sema4D likely exerts its presynaptic effects via direct \nbinding to Plexin -B1 expressed in the presynaptic interneuron, while postsynaptic effects are \nmediated by proximity to an eligible presynaptic bouton. Interestingly, prior work from our lab \nindicated that knockdown  of Plexin-B1 in the postsynaptic neuron prevents Sema4D -dependent \nsynapse formation (McDermott et al., 2018); thus, Plexin-B1 expression in the postsynaptic cell is \nstill required. Sema4D-dependent mobilization of gephyrin puncta once in proximity to a GAD65-\npositive bouton could presumably be mediated by paracrine interactions between Sema4D and \nPlexin-B1 or by Plexin-B1 signaling in conjunction with other signaling pathways via coreceptors \n(Giordano et al., 2002; Swiercz et al., 2004). \n Downstream signaling through the intracellular C-terminal GAP and RhoGEF domains of \nPlexin-B1 mediates cytoskeletal remodeling (Ito et al., 2006; Tran et al., 2007; Vodrazka et al., \n2009). One possibility is that Plexin -B1 directly promotes presynaptic protein mobility by \nregulating microtubule tracks, molecular motors, or force -generating cytoskeletal reorganization \n(e.g. actin branching or polymerization). A second possibility is that lo cal cytoskeletal disruption \nor disassembly effectively “releases the brake” on presynaptic bouton mobility, allowing mobile \nboutons to sample a larger dendritic area, and stabilization subsequently occurs through contact \nwith a postsynaptic specialization, preventing elimination. The latter hypothesis is supported by a \nstudy which demonstrated that local application of Sema4D to single boutons induced stabilization \nwhich could be chemically mimicked by destabilizing actin filaments (via latrunculin B treatment) \nor by inhibiting the RhoA/ROCK pathway (Frias et al., 2019). Although further work is required \nto distinguish between these possibilities, the relatively slow velocities and confined movement \nradii of protein clusters involved in new synapse formation point to modulation of local actin \nnetworks as the main structural change that promotes synapse formation downstream of Plexin -\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n28 \n \nB1 signaling. Together, our data support a model in which Sema4D/Plexin-B1 signaling facilitates \npresynaptic protein mobility by removing actin -dependent constraints on pre -assembled \npresynaptic protein clusters and increasing the probability of transient contacts with nearby \npostsynaptic specializations which are then stabilized through reciprocal adhesion. \n Overall, the findings from this study show that Sema4D signaling coordinates dynamic yet \nlocally constrained changes in both pre - and postsynaptic compartments to assemble functional \ninhibitory synapses on rapid timescales. This capacity to precisely regu late inhibitory synapse \nformation has important implications for inhibitory circuit organization in the developing and \nmature brain: inhibitory synapses regulate the timing and synchrony of network activity, and \ndisruptions to genes involved in inhibitory synapse assembly are implicated in various \nneurodevelopmental and seizure disorders (Shimojima et al., 2011; Sun et al., 2011; Lionel et al., \n2013). The ability of Sema4D to coordinate pre - and postsynaptic protein mobility to  rapidly \nassemble new synapses highlights a potential mechanism for fast, stable circuit remodeling and \npresents an intriguing yet largely unexplored therapeutic angle for disorders of excitatory -\ninhibitory balance such as epilepsy (Acker et al., 2018; Adel et al., 2023). More broadly, this model \nprovides a general framework for how cellular signaling pathways may tune inhibitory \nconnectivity and circuit balance on behaviorally and clinically relevant timescales. \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n29 \n \nAcknowledgements \n We would like to thank Dr. Avital Rodal and Dr. Andrew Stone for critical reading of this \nmanuscript, Dr. Adam Puche and Chloe Jenkins at the University of Maryland Medical School for \nproviding GAD65-GFP mice, the Brandeis Light Microscopy Core Facility and Foster Animal \nFacility for resource access and technical support, and all Paradis Lab members, in particular \nSusannah Adel, Rabia Anjum, Sarah McCallister, Yi Zhang, and Roshni Ray for critical \ndiscussions and experimental design/analysis feedback. This  work was supported by NIH grant \nR01NS065856 (S.P.), a CURE Epilepsy Catalyst Award  no. 998742 (S.P.), and NIH fellowship \naward F31NS134188 (Z.P.). \n \nMethods \nEthics statement \nAll animal procedures were approved by the Brandeis University Institutional Animal Care and \nUsage Committee, and all experiments were performed in accordance with relevant guidelines and \nregulations. \nAnimals \nMale GAD65-GFP transgenic mice (López-Bendito et al., 2004)  were obtained courtesy \nof Dr. Adam Puche ( University of Maryland School of Medicine ) and maintained in our animal \nfacility with ad libitum access to food and water on a 12 hour day/night cycle. Heterozygous \nGAD65-GFP males were crossed to female B6CBAF1/J mice (Jackson Laboratories, #100011) to \nproduce litters in which approximately half  of pups express one copy of the GAD65 -GFP \ntransgene. As this line expresses bright visible green fluorescence throughout the brain and spinal \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n30 \n \ncord, GAD65-GFP pups were identified using a handheld 488 nm laser and an orange emission \nfilter. For rat hippocampal cultures, pregnant female Sprague -Dawley rats were obtained from \nCharles River Labs when litters were approximately E14 and kept in our facility until dissection. \nPrimary mouse hippocampal cultures \nFor GAD65 -GFP and Halo -Gephyrin live imaging experiments, p rimary rat astrocytes \nwere plated onto 35 mm Petri dishes with glass -bottom 14 mm microwells (MatTek #P35G -1.5-\n14-C) that had been coated overnight  at 37ºC  with poly -d-lysine (20 μg/ml) and laminin (3.4 \nμg/ml). Before plating glia, coverslips were washed three times with sterile Ultrapure water and \nonce with DMEM  (Gibco #10313039). Glia were plated in DMEM with FBS (GenClone # 25 -\n550) and grown in a 37ºC incubator with 5% CO 2 until confluent. When glia formed a confluent \nfeeder layer, AraC (Sigma # C1768) was added at a final concentration of 5 µM in the dish to \nprevent further division. At P0 –1, GAD65-GFP mouse pups were identified as described above \nand rapidly sacrificed by decapitation. H ippocampi were harvested from GAD65-GFP pups of \nboth sexes, dissociated with papain (20 units/mL)  for 8 minutes , and gently resuspended in \nNeurobasal medium (Gibco # 21103049) with B27 supplement (Gibco # 17504044) (NB/B27) \nbefore plating atop glia within microwells at a density of 180k cells/well . After 4–24 hrs. when \nneurons were fully adhered the plating media was replaced by 1.5 mL NB/B27 with 5 µM AraC; \nthe culture media was not changed thereafter.  \nFor GFP -Gephyrin and Halo -GABAARγ2 overexpression live imaging experiments, \nastrocytes were grown on 14 mm glass-bottom microwell, 35 mm Petri dishes as described above. \nPregnant female Sprague -Dawley rats were sacrificed at embryonic day 18 (E18) by CO 2 \nasphyxiation, pups were rapidly removed and decapitated, and heads were kept in ice cold \ndissociation media prior to dissection. Hippocampi were then dissected from pups of both sexes, \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n31 \n \ndissociated, and resuspended in NB/B27 before plating atop astrocytes at a density of 120k \ncells/well. After 4–24 hrs. the plating media was replaced by 1.5 mL NB/B27 and was not changed \nthereafter, except during transfection (see below).  \nInfection/transfection and HaloTag labeling \nFor GAD65-GFP and Halo-Gephyrin live imaging experiments, neurons were infected on \nDIV2 or DIV3 with  AAV9.hSyn-HaloTag-Gephyrin virus (custom-produced by Duke Viral \nVector Core) at a final concentration of 1  × 109 GC/mL (~0.83 × 10 3 GC/neuron) in the dish. \nCulture media was not changed after addition of the virus. For labeling of HaloTag -expressing \nneurons, Janelia Fluor 646 HaloTag Ligand (Promega # GA1110) was suspended in DMSO \naccording to manufacturer recommendations to create a 200 µM stock solution whi ch was \naliquoted and stored at -20ºC for up to 1 year. To prepare the labeling solution this stock was \ndiluted 1:200 in NB/B27 to create a 1 µM 5x working stock.  At DIV10–11 cultures were live \nlabeled by aspirating all but 80 µL of growth media from each dish and adding 20 µL of 5x dye \nsolution for a final concentration of 200 nM dye in the well. Final dilution was such that DMSO \ncomprised no more than 0.1% of the total media volume during labeling. Cultures were incubated \nwith ligand solution at 37ºC for 1 5 minutes. After labeling, cells were briefly washed once with \nstandard NB/B27 media which was then replaced with phenol red -free NB/B27 media prior to \nimaging. (Note: although excess unbound JF646 HaloTag ligand is reported to be minimally \nfluorescent, we observed greatly improved signal-to-noise after a brief wash). \nFor GFP-Gephyrin and Halo-GABAARγ2 live imaging experiments, DIV4 cultures were \ntransfected with Lipofectamine 2000 using a protocol adapted from (Marwick and Hardingham, \n2017). Lipofectamine 2000 reagent (3 µL/ng DNA) was diluted to a final volume of 33 µL per \nwell using NB/B27; plasmid DNA (650 ng total split evenly between the two constructs) was \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n32 \n \nseparately diluted to a final volume of 33 µL per well with NB/B27. The L2000 solution was added \nto the DNA solution, pipetted 5 –6x, and left to incubate at RT for 20 minutes. After incubation, \nthe transfection mix was diluted to final volume of 125 µL per well using NB/B27. Working one \ndish at a time, growth media was aspirated from each dish and quickly replaced with 125 µL \ndiluted transfection mix, and cells were incubated with transfection mix for 2 –3 hrs. at 37ºC. A \nrecovery media was then made by mixing 80% saved growth media with 20% fresh NB/B27, and \ntransfection media was fully aspirated and replaced with 1.5 mL recovery media. The media was \nnot changed again prior to labeling. At DIV10–11 Janelia Fluor labeling was performed as \ndescribed above. \nLive imaging \nLive images were obtained using an inverted Nikon AX-R Resonance Scanning Confocal \nwith Ti2 body, Nikon Perfect Focus,  a piezo Z controller, and a MRD71670 Plan Apochromat \nLambda D 1.42 NA 60X Oil objective. Culture dishes were placed into a humidified environmental \nenclosure maintained at 37ºC with 5% CO 2 at a constant flow rate of 0.2 L/minute. Cells were \nallowed to habituate for at least 15 minutes prior to imaging. Single GAD65 -GFP positive cells \nwere identified and a field of view where distal axons were clearly visible was chosen. Cultures \nwere treated with either with 2 nM human IgG1-Fc control (R&D Systems #  110-HG) or 2 nM \nrecombinant Sema4D-Fc chimera (R&D Systems # 7470-S4) by pipetting directly into the dish. \nImage acquisition was started immediately after adding the protein and setting the focal plane, \ntypically within 1 -2 minutes of adding each treatment. For experiments with GAD65 -GFP and \nHalo-Gephyrin, we used 488 and 640 nm laser lines from a LUA-S4 laser unit at 2.5% and 3.0% \npower respectively, and for experiments with GFP -Gephyrin and Halo -γ2 we used 488 and 640 \nnm lines with 5.2% and 4.0% power respectively. 12-bit images were acquired at a 2048  × 2048 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n33 \n \nresolution with a pixel density of 126.6 nm/px using the resonant scanner and 8x line averaging. \nA Z-stack of 5–7 Nyquist-sampled planes encompassing a total range of 1.5-2.1 µm was acquired \nat 15 second intervals for 1 hour with optical focusing correction (Nikon Perfect Focus) to \nminimize drift in the Z direction. \nImage unwarping and registration \nTo eliminate the possibility that changes to protein cluster mobility were due to cell \nmotility, changes to dendrite/axon morphology, or image drift in the XY plane, images were \nregistered and unwarped prior to particle tracking analysis. Time lapse images were first corrected \nfor any stage XY drift during acquisition using the Linear Stack Alignment with SIFT plugin in \nImageJ with the following parameters: initial Gaussian blur 1 px, steps per octave 8, image size \n100–250 px, feature descriptor size 4, fea ture descriptor orientation bins 8, closest/next closest \nratio 0.96, maximum alignment error 5 px, inlier ratio 0.05, and rigid transform.  \nFollowing linear alignment, images were next unwarped using the BigWarp plugin in \nImageJ (Bogovic et al., 2016) to correct for non-translational drift (movement of axons/dendrites, \netc.). Briefly, time lapses were flattened into maximum intensity projections and then converted \nto a virtual Z-stack. The first frame of each image was duplicated repeatedly (once per timepoint) \nand converted to a virtual Z-stack to form a reference stack that was identical in size to the moving \nimage stack. The moving image stack and reference stack were then imported into BigWarp \nviewer, and pairs of landmarks were manually chosen in t he last plane of the moving image \n(corresponding to the last timepoint of the time lapse) and the reference stack. To ensure proper \nlinear interpolation of the transformation grid across time, the moving image was split into 30 -\nframe intervals, and landmar ks in the moving image were linearly interpolated at the \ncorresponding 30-frame intervals. The 30-frame sub-videos were then individually aligned to the \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n34 \n \ninterpolated landmarks using a thin -plate splines transform, which uses a deformable grid to \nperform exact matching between the moving image and reference image landmarks. \nTransformations were applied using BigWarp Apply with the following parameters: thin  plate \nspline transformation, bounding type FACES, samples = 5, and linear interpolation. For each 30 -\nframe sub-video, the last frame of the unwarped output served as the reference stack to which the \nnext sub-video was aligned; this prevents compounding error over time. Unwarped 30-frame sub-\nvideos were finally stitched together and converted back to a time series for particle tracking \nanalysis.  \nQualitative scoring of GAD65-GFP protein cluster behavior \nPrimary hippocampal cultures were generated from P0 GAD65 -GFP mice as described \nabove. DIV2 cultures were infected with a virus expressing HaloTag-Gephyrin under the synapsin \npromoter. Cultures were treated with 2 nM Sema4D -Fc or Fc control protein at DIV1 1 as \npreviously described and imaged at 10s intervals for 1 hour. To qualitatively assess cellular \nprocesses that may be relevant to synapse formation, the following categories of behaviors were \nchosen based on empirical observation: t rafficking, n ascent b ranching, l ocal cluster mobility, \nsplitting, merging, complex split/merge events, active growth cones, stable branch formation, and \nstable branch removal. Complete descriptions of these behaviors are found in Table S1. An \nexperimenter blinded to condition manually traced 10 axons (including branches) per cell using \nthe freehand line tool in ImageJ. Axons were selected only if the majority of the process remained \nin focus across all time points. Behaviors were manually counted by a n experimenter blinded to \ncondition and the frequency of each behavior was normalized to the total length of each axon. \nParticle tracking for live images \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n35 \n \nFor all analyses of protein cluster mobility and real-time colocalization analysis we utilized \nthe surface tracking feature in Imaris for Neuroscientists v10.2.0 (Oxford Instrument s). To \nfacilitate accurate tracking of low -intensity protein clusters, live images were denoised using \nonboard Nikon Denoise.ai in Nikon Elements prior to generating max intensity projections. Max \nintensity projections were then imported into Imaris for trac king with both the denoised and raw \nfluorescence channels. Surfaces were created in the GAD65-GFP, Halo-Gephyrin, GFP-Gephyrin, \nand/or Halo -γ2 channels using the machine learning segmentation feature in Imaris with the \ndenoised channel; we determined empirically that this led to more consistent identification of \nprotein clusters  than standard background subtraction and thresholding, particularly for bright \nGAD65-GFP clusters along axons. We next tracked surfaces across frames using the following \nparameters: autoregressive motion tracking, max frame -to-frame distance 2 µm, and max  frame \ngap 8. Tracks were automatically filtered out if their duration was less than 60 seconds and were \nmanually removed if they showed spurious linkages between adjacent neurites or if they were \nlocated on a neurite that did not appear to be an axon (for GAD65-GFP) or a dendrite (for gephyrin \nor Halo-γ2). For most analyses, only protein clusters that could be tracked for the duration of the \nlive imaging session were considered for further analysis. \nAnalysis of protein cluster mobility \nSurface and track features:  All a nalyses were performed in MATLAB R202 5a \n(MathWorks) and visualized in  Graphpad Prism 10.5.0. Surface and track statistics for each \nchannel were exported from Imaris and fed into a custom MATLAB analysis pipeline for data \nworkup. Surface features that were exported for analysis included surface area, acceleration, \ndisplacement delta length, intensity in each channel, position, overlapped area ratio, speed, and \nnearest neighbor distance to other surfaces. Track features that were exported for analysis included \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n36 \n \ntrack duration, straightness, AR(1) mean, length, and number of surfaces. Other track features were \ncalculated within our analysis pipeline, such as raw change in fluorescence (Fend - F0), fold change \nfluorescence (F end / F 0), peak fold change fluorescence (F max / F 0), and Euclidean distance ( net \ndisplacement between selected timepoints). \nFluorescence intensity analysis: All analysis of fluorescence intensity was performed using \nthe raw fluorescence (i.e., non -denoised) channel. For analysis of protein cluster fluorescence \nintensity over time, a photobleach correction was applied by fitting a one -step exponential decay \nmodel to the average particle intensity in the control condition using 𝐹(𝑡) = 𝐴 ⋅ 𝑒−𝑘𝑡 + 𝐶, where \nA is the amplitude, k is the decay rate constant , t is time, and C is the baseline offset . To correct \nphotobleaching the raw fluorescence was transformed using  𝐹𝑐𝑜𝑟𝑟𝑒𝑐𝑡𝑒𝑑(𝑡𝑛) = (𝐹𝑚𝑒𝑎𝑠𝑢𝑟𝑒𝑑(𝑡𝑛) ⋅\n𝐹(𝑡0)\n𝐹(𝑡𝑛) ). We then measured the average and total fluorescence intensity within the outline of each \nindividual puncta for each timepoint. For analysis of active GAD65 -GFP regions (Fig. S4), we \nquantified the total combined fluorescence intensity within GAD65 -GFP puncta within a band \ndrawn with a 1 µm radius around the GAD65 -GFP puncta. For figures measuring the change in \nnormalized fluorescence intensity over time, fluorescence intensity was normalized to the baseline \nmean of the first three minutes (12 frames) for all puncta within each treatment condition.  \nPrincipal component analysis: Dimensionality reduction was performed using principal \ncomponent analysis (PCA) via MATLAB’s default pca function. Track statistics for key \nparameters of puncta size, intensity, mobility, and proximity were calculated and tabulated as \nshown in Table S2–3. All track parameters were z-score normalized before PCA. All tracks were \nthen plotted along PC1 and PC2, and function outputs for component weights and percent track \nvariance explained were saved. Track parameters were considered significant contributors to a PC \nif the absolute value of their weighting for a PC exceeded 0.4. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n37 \n \nNew colocalization event analysis: New colocalization events were defined as instances in \nwhich a GAD65-GFP puncta that was not colocalized with a gephyrin puncta (or vice versa) in the \nprevious frame became colocalized, which we determined by comparing the nearest neighbor \ndistance to a Halo-Gephyrin puncta between frames. To avoid counting transient crossings, both \npuncta were required to remain colocalized in at least 95% of frames over the next 10 minutes \nfollowing the initial colocalization event (colocalization events happening in the final 10 minutes \nof the imaging session were not analyzed). To determine whether the puncta in the opposite \nchannel was newly -tracked (new puncta) or previously present  (existing puncta), we identified \npuncta in the opposite channel for which the center of mass was located within 1 µm of the \ncolocalization site at the colocalization timepoint, and used this criterion to determine the unique \ntrack ID of the opposite -channel puncta. If the opposite -channel puncta was present  at least 10 \nminutes before the colocalization event, it was considered an existing puncta; otherwise it was \nclassified as a new puncta. After classifying new colocalization events, we then quantified average \nprotein cluster velocity, fluorescence intensit y, displacement from origin, and distance from the \ncolocalization site in 1) the entire window before the new colocalization event and 2) the 10-minute \nperiods before and after the new colocalization event. \nExperimental design and statistical analysis \nStatistical analyses  for all figure  panels except time series data were performed using \nGraphpad Prism 10.5.0. For each experiment, specific statistical tests are described in the figure \nlegends. Data were tested for normality before appropriate statistical tests were applied. Where \nindicated, outliers were identified by the ROUT method with Q = 1% and removed fr om the data \nset. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n38 \n \nFor time series data, t o assess whether Sema4D treatment  interacts with  GAD65-GFP \npuncta features to regulate geph yrin accumulation in GAD65 -GFP puncta, we fit  binned time \nseries data to linear mixed-effects models (LME) using MATLAB (fitlme), which was chosen to \naccount for the effects of cell -to-cell variability on track measurements. Due to the large number \nof timepoints sampled in each image, timepoints were binned to avoid artificially inflating \nstatistical power, with bin size determined by calculating autocorrelation at a range of bin sizes for \nthe mean of the control condition across timepoints using MATLAB autocorr function. Bin size \nwas fixed as the smallest round -number bin size  at which the autocorrelation function dropped \nbelow 0.3, thus ensuring semi-independence of timepoints for the purposes of LME model fitting. \nModels tested the effects of treatment condition, time, GAD65-GFP puncta displacement length, \narea, and/or intensity, and all 2-way and 3-way interactions, on either: the nearest distance between \nGAD65 and gephyrin puncta, or mean fluorescence intensity of gephyrin within GAD65-positive \nregions. Random intercepts were used to account for variation between images (within-experiment \nvariability) and between individual tracks within images (within -puncta variability over time). \nModel fit was assessed by inspecting residuals, random effects distributions, and summary fit \nindices (AIC, BIC, LogLikelihood).  Effect estimates ar e reported as changes in the dependent \nvariable per unit change in the predictor. Time -dependent effects are interpreted as rate changes \nper minute. For main effects and interactions with p < 0.0 5, follow-up analysis was conducted \nusing subsets of puncta according to the variable of interest (e.g., a significant interaction between \ntime, treatment, and puncta size was further analyzed by grouping puncta by quintiles according \nto size and plotting mean  gephyrin fluorescence over time in each subset.) For experiments \ninvolving GFP -Gephyrin and Halo -γ2, statistical analysis of live imaging parameters was \nperformed identically to the above.  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n39 \n \nCode availability \nAll ImageJ and Python scripts used for landmark interpolation, unwarping, and stitching \nare publicly accessible via GitHub at zpranske/Bigwarp_Analysis. All MATLAB scripts and \nfunctions used for statistical analysis of particle tracking data are publicly accessible via GitHub \nat zpranske/LiveImaging_Analysis_Imaris. \n \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n40 \n \nReferences \nAcker DWM, Wong I, Kang M, Paradis S (2018) Semaphorin 4D promotes inhibitory synapse \nformation and suppresses seizures in vivo. Epilepsia 59:1257-1268. \nAdel SS, Clarke VRJ, Evans-Strong A, Maguire J, Paradis S (2023) Semaphorin 4D induced \ninhibitory synaptogenesis decreases epileptiform activity and alters progression to Status \nEpilepticus in mice. Epilepsy Res 193:107156. \nAdel SS, Pranske ZJ, Kowalski TF, Kanzler N, Ray R, Carmona C, Paradis S (2024) Plexin-B1 \nand Plexin-B2 play non-redundant roles in GABAergic synapse formation. Mol Cell \nNeurosci 128:103920. \nBogovic JA, Hanslovsky P, Wong A, Saalfeld S (2016) Robust registration of calcium images by \nlearned contrast synthesis. IEEE 13th International Symposium on Biomedical Imaging \n(ISBI):pp. 1123-1126. \nBresler T, Shapira M, Boeckers T, Dresbach T, Futter M, Garner CC, Rosenblum K, \nGundelfinger ED, Ziv NE (2004) Postsynaptic density assembly is fundamentally \ndifferent from presynaptic active zone assembly. J Neurosci 24:1507-1520. \nChapdelaine T, Hakim V, Triller A, Ranft J, Specht CG (2021) Reciprocal stabilization of \nglycine receptors and gephyrin scaffold proteins at inhibitory synapses. Biophys J \n120:805-817. \nChoii G, Ko J (2015) Gephyrin: a central GABAergic synapse organizer. Exp Mol Med 47:e158. \nChristie SB, Li RW, Miralles CP, Yang B, De Blas AL (2006) Clustered and non-clustered \nGABAA receptors in cultured hippocampal neurons. Mol Cell Neurosci 31:1-14. \nDabrowski A, Terauchi A, Strong C, Umemori H (2015) Distinct sets of FGF receptors sculpt \nexcitatory and inhibitory synaptogenesis. Development 142:1818-1830. \nDanglot L, Triller A, Bessis A (2003) Association of gephyrin with synaptic and extrasynaptic \nGABAA receptors varies during development in cultured hippocampal neurons. Mol Cell \nNeurosci 23:264-278. \nDobie FA, Craig AM (2011) Inhibitory synapse dynamics: coordinated presynaptic and \npostsynaptic mobility and the major contribution of recycled vesicles to new synapse \nformation. J Neurosci 31:10481-10493. \nDuan Y, Wang SH, Song J, Mironova Y, Ming GL, Kolodkin AL, Giger RJ (2014) Semaphorin \n5A inhibits synaptogenesis in early postnatal- and adult-born hippocampal dentate \ngranule cells. Elife 3. \nFrias CP, Liang J, Bresser T, Scheefhals L, van Kesteren M, van Dorland R, Hu HY, Bodzeta A, \nvan Bergen En Henegouwen PMP, Hoogenraad CC, Wierenga CJ (2019) Semaphorin4D \nInduces Inhibitory Synapse Formation by Rapid Stabilization of Presynaptic Boutons via \nMET Coactivation. J Neurosci 39:4221-4237. \nGiordano S, Corso S, Conrotto P, Artigiani S, Gilestro G, Barberis D, Tamagnone L, Comoglio \nPM (2002) The semaphorin 4D receptor controls invasive growth by coupling with Met. \nNat Cell Biol 4:720-724. \nHines RM, Maric HM, Hines DJ, Modgil A, Panzanelli P, Nakamura Y, Nathanson AJ, Cross A, \nDeeb T, Brandon NJ, Davies P, Fritschy JM, Schindelin H, Moss SJ (2018) \nDevelopmental seizures and mortality result from reducing GABA(A) receptor alpha2-\nsubunit interaction with collybistin. Nat Commun 9:3130. \nIto Y, Oinuma I, Katoh H, Kaibuchi K, Negishi M (2006) Sema4D/plexin-B1 activates GSK-\n3beta through R-Ras GAP activity, inducing growth cone collapse. EMBO Rep 7:704-\n709. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n41 \n \nJacob TC, Bogdanov YD, Magnus C, Saliba RS, Kittler JT, Haydon PG, Moss SJ (2005) \nGephyrin regulates the cell surface dynamics of synaptic GABAA receptors. J Neurosci \n25:10469-10478. \nJoo WJ, Sweeney LB, Liang L, Luo L (2013) Linking cell fate, trajectory choice, and target \nselection: genetic analysis of Sema-2b in olfactory axon targeting. Neuron 78:673-686. \nKoropouli E, Kolodkin AL (2014) Semaphorins and the dynamic regulation of synapse \nassembly, refinement, and function. Curr Opin Neurobiol 27:1-7. \nKrivosheya D, Tapia L, Levinson JN, Huang K, Kang Y, Hines R, Ting AK, Craig AM, Mei L, \nBamji SX, El-Husseini A (2008) ErbB4-neuregulin signaling modulates synapse \ndevelopment and dendritic arborization through distinct mechanisms. J Biol Chem \n283:32944-32956. \nKuriu T, Yanagawa Y, Konishi S (2012) Activity-dependent coordinated mobility of \nhippocampal inhibitory synapses visualized with presynaptic and postsynaptic tagged-\nmolecular markers. Mol Cell Neurosci 49:184-195. \nKuzirian MS, Moore AR, Staudenmaier EK, Friedel RH, Paradis S (2013) The class 4 \nsemaphorin Sema4D promotes the rapid assembly of GABAergic synapses in rodent \nhippocampus. J Neurosci 33:8961-8973. \nLevinson JN, Chery N, Huang K, Wong TP, Gerrow K, Kang R, Prange O, Wang YT, El-\nHusseini A (2005) Neuroligins mediate excitatory and inhibitory synapse formation: \ninvolvement of PSD-95 and neurexin-1beta in neuroligin-induced synaptic specificity. J \nBiol Chem 280:17312-17319. \nLionel AC et al. (2013) Rare exonic deletions implicate the synaptic organizer Gephyrin (GPHN) \nin risk for autism, schizophrenia and seizures. Hum Mol Genet 22:2055-2066. \nLópez-Bendito G, Sturgess K, Erdelyi F, Szabo G, Molnar Z, Paulsen O (2004) Preferential \norigin and layer destination of GAD65-GFP cortical interneurons. Cereb Cortex 14:1122-\n1133. \nLorenz-Guertin JM, Jacob TC (2018) GABA type a receptor trafficking and the architecture of \nsynaptic inhibition. Dev Neurobiol 78:238-270. \nMacha A, Liebsch F, Bruckisch EHW, Burdina N, von Stulpnagel I, Benting K, Gunkel M, \nBehrmann E, Schwarz G (2025) Gephyrin filaments represent the molecular basis of \ninhibitory postsynaptic densities. Nat Commun 16:8293. \nMarwick KFM, Hardingham GE (2017) Transfection in Primary Cultured Neuronal Cells. \nMethods Mol Biol 1677:137-144. \nMcAllister AK (2007) Dynamic aspects of CNS synapse formation. Annu Rev Neurosci 30:425-\n450. \nMcDermott JE, Goldblatt D, Paradis S (2018) Class 4 Semaphorins and Plexin-B receptors \nregulate GABAergic and glutamatergic synapse development in the mammalian \nhippocampus. Mol Cell Neurosci 92:50-66. \nMukherjee J, Kretschmannova K, Gouzer G, Maric HM, Ramsden S, Tretter V, Harvey K, \nDavies PA, Triller A, Schindelin H, Moss SJ (2011) The residence time of GABA(A)Rs \nat inhibitory synapses is determined by direct binding of the receptor alpha1 subunit to \ngephyrin. J Neurosci 31:14677-14687. \nOinuma I, Ishikawa Y, Katoh H, Negishi M (2004) The Semaphorin 4D receptor Plexin-B1 is a \nGTPase activating protein for R-Ras. Science 305:862-865. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n42 \n \nParadis S, Harrar DB, Lin Y, Koon AC, Hauser JL, Griffith EC, Zhu L, Brass LF, Chen C, \nGreenberg ME (2007) An RNAi-based approach identifies molecules required for \nglutamatergic and GABAergic synapse development. Neuron 53:217-232. \nPetrini EM, Barberis A (2014) Diffusion dynamics of synaptic molecules during inhibitory \npostsynaptic plasticity. Front Cell Neurosci 8:300. \nPetrini EM, Ravasenga T, Hausrat TJ, Iurilli G, Olcese U, Racine V, Sibarita JB, Jacob TC, \nMoss SJ, Benfenati F, Medini P, Kneussel M, Barberis A (2014) Synaptic recruitment of \ngephyrin regulates surface GABAA receptor dynamics for the expression of inhibitory \nLTP. Nat Commun 5:3921. \nPoulopoulos A, Aramuni G, Meyer G, Soykan T, Hoon M, Papadopoulos T, Zhang M, Paarmann \nI, Fuchs C, Harvey K, Jedlicka P, Schwarzacher SW, Betz H, Harvey RJ, Brose N, Zhang \nW, Varoqueaux F (2009) Neuroligin 2 drives postsynaptic assembly at perisomatic \ninhibitory synapses through gephyrin and collybistin. Neuron 63:628-642. \nPrange O, Murphy TH (2001) Modular transport of postsynaptic density-95 clusters and \nassociation with stable spine precursors during early development of cortical neurons. J \nNeurosci 21:9325-9333. \nRaissi AJ, Staudenmaier EK, David S, Hu L, Paradis S (2013) Sema4D localizes to synapses and \nregulates GABAergic synapse development as a membrane-bound molecule in the \nmammalian hippocampus. Mol Cell Neurosci 57:23-32. \nScheiffele P (2003) Cell-cell signaling during synapse formation in the CNS. Annu Rev Neurosci \n26:485-508. \nSchuemann A, Klawiter A, Bonhoeffer T, Wierenga CJ (2013) Structural plasticity of \nGABAergic axons is regulated by network activity and GABAA receptor activation. \nFront Neural Circuits 7:113. \nShimojima K, Sugawara M, Shichiji M, Mukaida S, Takayama R, Imai K, Yamamoto T (2011) \nLoss-of-function mutation of collybistin is responsible for X-linked mental retardation \nassociated with epilepsy. J Hum Genet 56:561-565. \nSüdhof TC (2018) Towards an Understanding of Synapse Formation. Neuron 100:276-293. \nSüdhof TC (2021) The cell biology of synapse formation. J Cell Biol 220. \nSun C, Cheng MC, Qin R, Liao DL, Chen TT, Koong FJ, Chen G, Chen CH (2011) \nIdentification and functional characterization of rare mutations of the neuroligin-2 gene \n(NLGN2) associated with schizophrenia. Hum Mol Genet 20:3042-3051. \nSwiercz JM, Kuner R, Offermanns S (2004) Plexin-B1/RhoGEF-mediated RhoA activation \ninvolves the receptor tyrosine kinase ErbB-2. J Cell Biol 165:869-880. \nTakahashi H, Katayama K, Sohya K, Miyamoto H, Prasad T, Matsumoto Y, Ota M, Yasuda H, \nTsumoto T, Aruga J, Craig AM (2012) Selective control of inhibitory synapse \ndevelopment by Slitrk3-PTPdelta trans-synaptic interaction. Nat Neurosci 15:389-398, \nS381-382. \nTran TS, Kolodkin AL, Bharadwaj R (2007) Semaphorin regulation of cellular morphology. \nAnnu Rev Cell Dev Biol 23:263-292. \nTretter V, Jacob TC, Mukherjee J, Fritschy JM, Pangalos MN, Moss SJ (2008) The clustering of \nGABA(A) receptor subtypes at inhibitory synapses is facilitated via the direct binding of \nreceptor alpha 2 subunits to gephyrin. J Neurosci 28:1356-1365. \nTrotter JH, Wang CY, Zhou P, Nakahara G, Sudhof TC (2023) A combinatorial code of \nneurexin-3 alternative splicing controls inhibitory synapses via a trans-synaptic \ndystroglycan signaling loop. Nat Commun 14:1771. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n43 \n \nUesaka N, Uchigashima M, Mikuni T, Nakazawa T, Nakao H, Hirai H, Aiba A, Watanabe M, \nKano M (2014) Retrograde semaphorin signaling regulates synapse elimination in the \ndeveloping mouse brain. Science 344:1020-1023. \nVilla KL, Berry KP, Subramanian J, Cha JW, Oh WC, Kwon HB, Kubota Y, So PT, Nedivi E \n(2016) Inhibitory Synapses Are Repeatedly Assembled and Removed at Persistent Sites \nIn Vivo. Neuron 89:756-769. \nVodrazka P, Korostylev A, Hirschberg A, Swiercz JM, Worzfeld T, Deng S, Fazzari P, \nTamagnone L, Offermanns S, Kuner R (2009) The semaphorin 4D-plexin-B signalling \ncomplex regulates dendritic and axonal complexity in developing neurons via diverse \npathways. Eur J Neurosci 30:1193-1208. \nWierenga CJ, Becker N, Bonhoeffer T (2008) GABAergic synapses are formed without the \ninvolvement of dendritic protrusions. Nat Neurosci 11:1044-1052. \nYim YS, Kwon Y, Nam J, Yoon HI, Lee K, Kim DG, Kim E, Kim CH, Ko J (2013) Slitrks \ncontrol excitatory and inhibitory synapse formation with LAR receptor protein tyrosine \nphosphatases. Proc Natl Acad Sci U S A 110:4057-4062. \nZiv NE, Garner CC (2004) Cellular and molecular mechanisms of presynaptic assembly. Nat \nRev Neurosci 5:385-399. \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\nLocal mobility\nSplit\nMerge\nComplex split/merge\nGrowth cone\n0.00\n0.05\n0.10\n0.15\n0.20\nEvents / μm\n✱✱\n✱\nControl\nSema4D\nB\nA\nGrowth cone\nLocal \nmobility\nSplit\nMerge\nComplex\nsplit/merge\nFig. 1.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\nFigure 1. Sema4D affects the frequency of dynamic behaviors of GAD65-GFP puncta in cultured \nmouse hippocampal neurons.\n(A) Representative images of stretches of axon from DIV11 GAD65-GFP mouse hippocampal neurons \nshowing dynamic behaviors of interest. (Top to bottom) local mobility (repeated movement of a single \ncluster along the axon), split, merge, complex split/merge, and growth cone events. Note: time scales \ndiffer between events; time points are relative to the start of the montage. Example images include cells \nfrom either treatment condition; some montages show different axonal regions from the same neuron. \nHollow arrows = single puncta; solid arrows = merged puncta. Scale bars = 2 µm. \n(B) Frequency of local mobility, split, merge, complex split/merge, and growth cones observed in control \nvs. Sema4D-treated cultures. Dots correspond to individual axons; n = 50-55 axons per condition from \n11 cells (Fc) or 13 cells (Sema4D). *p < 0.05; **p < 0.01; Mann-Whitney U-test.\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n0.00 0.05 0.10\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\nAll puncta\nTrack Straightness\nProportion of puncta\nControl\nSema4D\n0.00 0.05 0.10\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\n5% most mobile\n(displacement)\nTrack Straightness\n**\n0.00 0.05 0.10\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\n5% fastest\n(mean velocity)\nTrack Straightness\nA\nEi\nC\n0.0\n0.5\n1.0\n10 20 30 40 50 60\nTime (min.)\nDisplacement (μm) Control\nSema4D ✱✱✱\n0 1 2 3 4\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\n30 min.\nDisplacement (um)\nProportion of puncta\nControl\nSema4D\n*\n0 1 2 3 4\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\n45 min.\nDisplacement (um)\n***\n0 1 2 3 4\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\n60 min.\nDisplacement (μm)\n***\nConfocal ML Segmentation\nDi Dii Diii\nEii Eiii\nB\n0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0\n0.00\n0.05\n0.10\n0.15\n0.20\nProportion of timepoints present\nProportion of puncta\nControl\nSema4D\n✱✱✱ Fig. 2.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\nFigure 2. Sema4D treatment increases stability and mobility of presynaptic GAD65-GFP puncta. \n(A) (Left) Example stretch of axon from cultured hippocampal GAD65-GFP neurons. GAD65-GFP puncta \nappear as bright spots marked by arrows. (Right) Machine learning (ML) reconstruction of GAD65-GFP \npuncta in Imaris at the same timepoints. Scale bar = 5 µm. \n(B) GAD65-GFP puncta stability (proportion of frames in which a puncta was tracked). n = 7427 puncta (Fc), \n10213 puncta (Sema4D). The distributions of GAD65-GFP puncta are significantly different between \ntreatment conditions (Chi-square test; χ2(8) = 72.66, ***p < 0.001).\n(C) GAD65-GFP puncta mobility (mean displacement from origin) is increased beginning within 20 min. of \nSema4D treatment (binned LME: time × treatment interaction: F(1, 3651) = 11.578, p < 0.001). n = 384 \npuncta (Fc), 347 puncta (Sema4D). Note: analysis begins at t=10 min.\n(D) Cumulative frequency histogram of GAD65-GFP puncta displacement at 30, 45, and 60 min. (*p < 0.05, \n***p < 0.001; Kolmogorov-Smirnov test). n ≥ 371 puncta per timepoint from 10 cells (Fc), n ≥ 333 puncta \nper timepoint from 12 cells (Sema4D).\n(E) Cumulative frequency histogram of GAD65-GFP track straightness. Track straightness was calculated \nas displacement​ / total path length. (i) Distribution of track straightness for all puncta (n = 384 puncta (Fc), \n347 puncta (Sema4D); p = 0.2594, Kolmogorov-Smirnov test). (ii) Distribution of track straightness for the \ntop 5% most mobile puncta in each condition by displacement (n = 19 puncta (Fc), 17 puncta (Sema4D); **p \n< 0.01; Kolmogorov-Smirnov test). (iii) Distribution of track straightness for the top 5% of puncta by mean \nvelocity (n = 19 puncta (Fc), 17 puncta (Sema4D); p = 0.8075; Kolmogorov-Smirnov test).\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n0 1 2 3 4\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\n30 min.\nDisplacement (μm)\nProportion of puncta\nControl\nSema4D\n0 1 2 3 4\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\n45 min.\nDisplacement (μm)\n0 1 2 3 4\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\n60 min.\nDisplacement (μm)\n0\n0.0\n0.5\n1.0\n10 20 30 40 50 60\nTime (min.)\nDisplacement (μm) Control\nSema4D ns\nA\nDi\nEi\nConfocal ML Segmentation\n0.0 0.1 0.2 0.3 0.4 0.5\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\nAll puncta\nTrack Straightness\nProportion of puncta\nControl\nSema4D\n0.0 0.1 0.2 0.3 0.4 0.5\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\n5% fastest\n(mean velocity)\nTrack Straightness\n0.0 0.1 0.2 0.3 0.4 0.5\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\n5% most mobile\n(displacement)\nTrack Straightness\nC\n0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0\n0.0\n0.2\n0.4\n0.6\nProportion of timepoints present\nProportion of puncta\nControl\nSema4D ns\nB\nDii Diii\nEii Eiii\nFig. 3.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\nFigure 3. Sema4D treatment does not affect Halo-Gephyrin mobility at the population level.\n(A) Representative stretch of dendrite showing Halo-Gephyrin expression along dendrites in cultures from \nGAD65-GFP neurons. Halo-Gephyrin puncta appear as bright spots marked by arrows. (Right) Machine \nlearning (ML) reconstruction of Halo-Gephyrin puncta in Imaris at the same timepoints. Scale bar = 5 µm.\n(B) Halo-Gephyrin puncta stability (proportion of frames in which a puncta was tracked). n = 742 puncta (Fc), \n1026 puncta (Sema4D). The distributions of Halo-Gephyrin puncta do not significantly differ between \ntreatment conditions (Chi-square test; χ2 (7) = 10.63, p = 0.1557). \n(C) Halo-Gephyrin puncta mobility (mean displacement from origin) is not affected by Sema4D treatment \n(binned LME: time × treatment interaction: F(1, 7279) = 1.0464, p = 0.2954). Note: analysis begins at t=10 \nmin.\n(D) Cumulative frequency histogram of Halo-Gephyrin puncta displacement at 30, 45 and 60 min. There was \nno effect of Sema4D treatment on the overall distribution of Halo-Gephyrin displacement at 30 min. (p = \n0.0529; Kolmogorov-Smirnov test), at 45 min. (p = 0.3440), or at 60 min. (p = 0.7637). n ≥ 220 puncta per \ntimepoint from 8 cells (Fc), n ≥ 393 puncta per timepoint from 8 cells (Sema4D).\n(E) Cumulative frequency histogram of Halo-Gephyrin track straightness. (i) Distribution of track straightness \nfor all puncta (n = 384 puncta (Fc), 347 puncta (Sema4D); p = 0.8307, Kolmogorov-Smirnov test). (ii) \nDistribution of track straightness for the top 5% most mobile puncta in each condition by displacement (n = 23 \npuncta (Fc), 34 puncta (Sema4D); p = 0.3308; Kolmogorov-Smirnov test). (iii) Distribution of track \nstraightness for the top 5% of puncta by mean velocity (n = 23 puncta (Fc), 34 puncta (Sema4D); p = 0.6355; \nKolmogorov-Smirnov test).\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\n0 15 30 45 60\n0.9\n1.0\n1.1\n1.2\nAll GAD65-GFP puncta\nTime (min.)\nColocalized\ngephyrin fluorescence\nControl\nSema4D\n✱✱✱\n0 15 30 45 60\n0.9\n1.0\n1.1\n1.2\nTop quintile\nTime (min.)\nControl\nSema4D\n✱✱✱\n0 15 30 45 60\n0.9\n1.0\n1.1\n1.2\nBottom quintile\nTime (min.)\nControl\nSema4D\n✱✱✱\nBi\nC\n0 500 1000 1500 2000\n0\n1\n2\n3\nInitial gephyrin\nfluorescence (a.u.)\nFold change\ngephyrin fluorescence\nControl\nSema4D\nGAD65-GFP    Halo-Gephyrin\n0’ 60’50’40’30’20’10’\nML Seg.\n60’\nBii Biii\nAi\nAii\nAiii\nAiv\n0 15 30 45 600\n1\n2\n3\n4\nTime (min.)\nNearest neighbor\ndistance (μm)\nControl\nSema4D\n✱✱✱\nD\nFig. 4.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\nFigure 4. Sema4D promotes increased gephyrin colocalization with existing GAD65-GFP \npuncta.\n(A) Montages showing putative new synapse formation events. (i, ii) Existing mobile GAD65-GFP \npuncta (green arrows) localize to existing Halo-Gephyrin puncta (magenta arrows) and remain \ncolocalized (white arrows). (iii, iv) A newly-tracked Halo-Gephyrin puncta emerges and colocalizes \nwith an existing GAD65-GFP puncta (green arrows). White arrows = colocalized puncta. Scale bars \n= 2 µm.\n(B) (i) Gephyrin fluorescence colocalized with GAD65-GFP puncta is increased in Sema4D \ncondition compared to control (binned LME: time × treatment interaction: F(1, 24140) = 13.656, \n***p < 0.001). n ≥ 765 puncta per timepoint (control), 1000 puncta (Sema4D). (ii) Sema4D prevents \nthe loss of gephyrin from GAD65-GFP boutons in the top quintile of baseline gephyrin signal \n(interaction : F(1, 4820) = 17.75, ***p < 0.001). n ≥ 151 puncta per timepoint (control), 201 puncta \n(Sema4D). (iii) Sema4D treatment increases recruitment of gephyrin to boutons in the bottom \nquintile of baseline gephyrin signal compared to control treatment (interaction : F(1, 4820) = \n17.637, ***p < 0.001). n ≥ 150 puncta per timepoint (control), 197 puncta (Sema4D). Error bars = \nSEM. Data are normalized within treatment condition to the mean of the first 3 minutes.\n(C) There is a negative correlation between baseline (t=0 min.) gephyrin fluorescence colocalized \nwith GAD65-GFP puncta and change in gephyrin fluorescence from t=0 min. to 60 min; Sema4D \ntreatment reduces the strength of this correlation. Control: slope CI = (-4.568×10-4, -3.039×10-4); \nSema4D: slope CI = (-2.150×10-4, -1.241×10-4). n = 866 puncta (control), 1146 puncta (Sema4D). \nError bars = 95% CI of linear regression fit.\n(D) Sema4D treatment decreases the mean nearest neighbor distance of GAD65-GFP puncta to the \nnearest gephyrin puncta (for GAD65-GFP puncta with initial nearest neighbor distance ≤ 2 µm) \n(binned LME: time × treatment interaction : F(1,4976) = 22.334, ***p < 0.001). n ≥ 123 puncta per \ntimepoint (control), 242 puncta (Sema4D). Error bars = SEM.\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\nBi\nControl Sema4D\n0.00\n0.01\n0.02\n0.03\nNew\ngephyrin puncta\nProportion GAD65-GFP puncta\n13/\n866\n33/\n1146\n✱\nControl Sema4D\n0.00\n0.01\n0.02\n0.03\nExisting\ngephyrin puncta\nProportion GAD65-GFP puncta\n4/\n866\n12/\n1146\nns\nControl\n13\n4\nSema4D\n33\n12\nNewly-tracked\ngephyrin puncta\nExisting\ngephyrin puncta\nA\nControl Sema4D\n0.00\n0.01\n0.02\n0.03\n0.04\nAll new\ncolocalization events\nProportion GAD65-GFP puncta\n17/\n866\n45/\n1146\n✱\nBii Biii\nFig. 5.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\nFigure 5. Sema4D-dependent GAD65/gephyrin colocalization occurs via colocalization events \nbetween existing GAD65 clusters and new gephyrin clusters.\n(A) Number of GAD65-GFP puncta that colocalized with an existing Halo-Gephyrin puncta vs. a \nnewly-tracked Halo-Gephyrin puncta.\n(B) (i) Overall frequency of all new colocalization events. Sema4D treatment significantly increased \nthe frequency of new colocalization events compared to control (*p < 0.05). (ii) Sema4D treatment \nsignificantly increased the proportion of GAD65-GFP puncta that colocalized with a newly-tracked \nHalo-Gephyrin puncta (*p < 0.05, Fisher’s exact test). (iii) There was no significant difference in the \nfrequency of colocalization events with an existing Halo-Gephyrin puncta in Sema4D vs. control-\ntreated cultures (p = 0.2048). \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\nB\nControl Sema4D\n0.0\n0.2\n0.4\n0.6\n0.8\n10' window before\ncolocalization\nMean distance from\ncolocalization site (μm)\nControl Sema4D\n0.0\n0.5\n1.0\n1.5\n2.0\nMobility\npre-colocalization\nDisplacement from origin (μm) ✱\nControl Sema4D\n0.0\n0.2\n0.4\n0.6\n0.8\n10' window before\ncolocalization\nMean distance from\ncolocalization site (μm)\n✱\nControl Sema4D\n0.0\n0.5\n1.0\n1.5\n2.0\n2.5\nMobility\npre-colocalization\nDisplacement from origin (μm) F G\n50\n0\n25\nMinutes \nbefore event\nHalo-GephyrinGAD65-GFP\nC\nA E\nD H\n✱\nFig. 6.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\nFigure 6. Sema4D-dependent GAD65/gephyrin colocalization events are driven by early GAD65-GFP \ndisplacement followed by late Halo-Gephyrin mobility.\n(A) Tracks followed by GAD65-GFP puncta prior to new colocalization events with gephyrin in control (left) \nor Sema4D (right)-treated cultures. Blue dots represent relative start locations; red dots (at 0,0) represent \nnormalized location of new colocalization event. \n(B) Total displacement of GAD65-GFP puncta undergoing a new colocalization event across entire imaging \nsession (*p < 0.05, Mann-Whitney U-test). n = 17 events (control), 45 events (Sema4D). Error bars = SEM.\n(C) Mean distance of GAD65-GFP puncta from new colocalization sites in the 10-minute window preceding \nnew colocalization events. Sema4D treatment did not affect mean GAD65-GFP distance during this time \nwindow (p = 0.8513, Mann-Whitney U-test). n = 17 events (control), 45 events (Sema4D). Error bars = \nSEM.\n(D) Change in mean displacement from origin of individual GAD65-GFP puncta before vs. after \ncolocalization. GAD65-GFP puncta are significantly further from origin after colocalization events in control \nneurons (*p < 0.05, Wilcoxon matched-pairs signed rank test) but not in Sema4D-treated neurons (p = \n0.0973) . n = 17 events (control), 45 events (Sema4D). Error bars = SEM.\n(E) Tracks followed by Halo-Gephyrin puncta prior to new colocalization events with GAD65-GFP in \ncontrol (left) or Sema4D (right) treated cultures. Blue dots represent relative start locations; red dots (0,0) \nrepresent normalized location of new colocalization event. \n(F) Total displacement of Halo-Gephyrin puncta undergoing a new colocalization event across entire \nimaging session. Sema4D treatment does not affect displacement of Halo-Gephyrin puncta undergoing \ncolocalization events (p = 0.2343, Mann-Whitney U-test). n = 5 events (control), 12 events (Sema4D). Error \nbars = SEM.\n(G) In Sema4D-treated cultures Halo-Gephyrin puncta are on average significantly farther from new \ncolocalization sites during the 10-minute window preceding new colocalization events compared to control \ncultures (*p < 0.05, Mann-Whitney U-test). n = 5 events (control), 12 events (Sema4D). Error bars = SEM.\n(H) Mean displacement from origin of individual Halo-Gephyrin puncta in the 10-minute window before vs. \nafter colocalization. Halo-Gephyrin does not travel significantly further from origin following colocalization \nevents in either control (p = 0.999, Wilcoxon matched-pairs signed rank test) or Sema4D-treated neurons (p \n= 0.677). n = 5 events (control), 12 events (Sema4D). Error bars = SEM.\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\nBi\n0 10 20 30 40 50 60\n0.8\n1.0\n1.2\n1.4\n1.6\nAll puncta\nTime (min.)\nColocalized Halo-γ2\nfluorescence Control\nSema4D\n0 10 20 30 40 50 60\n0.8\n1.0\n1.2\n1.4\n1.6\nTop quintile\nTime (min.)\nControl\nSema4D\n0 10 20 30 40 50 60\n0.8\n1.0\n1.2\n1.4\n1.6\nBottom quintile\nTime (min.)\nControl\nSema4D\n✱✱\n✱\n0 10000 20000 30000\n0\n1\n2\n3\nInitial Halo-γ2\nfluorescence (a.u.)\nFold change\nHalo-γ2 fluorescence\nControl\nSema4D\nC\nBii Biii\n0’ 60’50’40’30’20’10’\nML Seg.\n60’\nGFP-Geph\nAi\nAii\n0 10 20 30 40 50 60\n0\n5\n10\n15\n20\nTime (min.)\nNearest neighbor\ndistance (μm)\nControl\nSema4D\n✱✱✱\nD\nFig. 7\nHalo-γ2\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\nFigure 7. Sema4D promotes increased localization of GABAARγ2 to gephyrin-labeled postsynaptic \nspecializations.\n (A) Representative images showing colocalization between GFP-Gephyrin (green arrows) and Halo-γ2 \n(magenta arrows). (i, ii) Examples showing a Halo-γ2 puncta moving and colocalizing with a GFP-Gephyrin \npuncta (white arrows = colocalization). Images represent different regions from the same neuron. Scale bars = 2 \nµm.\n(B) (i) Sema4D treatment increases mean Halo-γ2 intensity colocalized with GFP-Gephyrin puncta compared to \ncontrol treatment (binned LME: time × treatment interaction: F(1, 4964) = 8.5686, **p < 0.01). n ≥ 254 puncta \nper timepoint (control), 130 puncta (Sema4D). (ii) Sema4D treatment does not affect mean Halo-γ2 intensity \ncolocalized with GFP-Gephyrin puncta in the top quintile of baseline Halo-γ2 signal compared to control \ntreatment (binned LME: time × treatment interaction: F(1, 992) = 1.1088, p = 0.2926). n ≥ 48 puncta per \ntimepoint (control), 24 puncta (Sema4D). (iii) Sema4D treatment increases mean Halo-γ2 intensity colocalized \nwith GFP-Gephyrin puncta in the bottom quintile of baseline Halo-γ2 intensity compared to control treatment \n(binned LME: time × treatment interaction: F(1, 992) = 4.2854, *p < 0.05). n ≥ 49 puncta per timepoint \n(control), 24 puncta (Sema4D). Error bars = SEM. Data are normalized within treatment condition to the mean \nof the first 3 minutes.\n(C) Sema4D treatment enhances the negative correlation between baseline (t=0 min.) Halo-γ2 fluorescence \ncolocalized with GFP-Gephyrin puncta and the change in colocalized Halo-γ2 fluorescence over time. Control: \nslope CI = (-2.731×10-5, -1.203×10-5); Sema4D: slope CI = (-6.294×10-5, -3.327×10-5). n = 276 puncta (control), \n138 puncta (Sema4D). Error bars = 95% CI of linear regression fit.\n(D) Sema4D treatment decreases the mean nearest neighbor distance of GFP-Gephyrin puncta to Halo-γ2 \npuncta (for puncta with initial nearest neighbor distance ≤ 2 µm) (binned LME: time × treatment interaction : \nF(1, 2456) = 50.247, ***p < 0.001). n ≥ 131 puncta per timepoint (control), 54 puncta (Sema4D). Error bars = \nSEM.\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\nBi C\nControl Sema4D\n0.00\n0.05\n0.10\n0.15\n0.20\nNew\nHalo-γ2 puncta\nProportion GFP-Gephyrin puncta\n11/\n276\n7/\n138\nControl Sema4D\n0.00\n0.05\n0.10\n0.15\n0.20\nExisting\nHalo-γ2 puncta\nProportion GFP-Gephyrin puncta\n43/\n276\n27/\n138\nControl\n43\n11\nSema4D\n27\n7\nNewly-tracked\nHalo-γ2 puncta\nExisting\nHalo-γ2 puncta\nControl Sema4D\n0.0\n0.1\n0.2\n0.3\nIncreased colocalized\nHalo-γ2 intensity\nProportion of puncta\n11/\n276 16/\n138\n✱✱\nA\nControl Sema4D\n0.0\n0.1\n0.2\n0.3\nAll new\ncolocalization events\nProportion GFP-Gephyrin puncta\n54/\n276\n34/\n138\nBii Biii\nFig. 8.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\nFigure 8. Sema4D treatment increases recruitment of GABAARγ2 to gephyrin scaffolds without \ndriving new colocalization events between GABAARγ2 and GFP-Gephyrin puncta.\n(A) Number of new colocalization events between GFP-Gephyrin and existing Halo-γ2 puncta vs. newly-\ntracked Halo-γ2 puncta.\n(B) (i) Overall proportion of GFP-Gephyrin puncta with a new colocalization event. There was no effect of \nSema4D treatment on the proportion of GFP-Gephyrin puncta with any type of new colocalization event (p \n= 0.2525). (ii) There was no effect of Sema4D treatment on the proportion of GFP-Gephyrin puncta that \ncolocalized with a newly-tracked Halo-γ2 puncta (p = 0.6155, Fisher’s exact test). (iii) There was no effect \nof Sema4D treatment on the proportion of GFP-Gephyrin puncta that colocalized with an existing Halo-γ2 \npuncta (p = 0.3314). \n(C) Proportion of GFP-Gephyrin puncta with ≥ 1.5-fold increase in colocalized Halo-γ2 fluorescence. \nSema4D treatment significantly increased the proportion of GFP-Gephyrin puncta with increased Halo-γ2 \nfluorescence (**p <  0.01, Fisher’s exact test). \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\nControl Sema4D\n0.0\n0.5\n1.0\n1.5\n2.0\nMobility\npre-colocalization\nDisplacement from origin (μm) 0.0545\nControl Sema4D\n0.0\n0.5\n1.0\n1.5\n2.0\n10' window before\ncolocalization\nMean distance from\ncolocalization site (μm)\n0.0553\nH\nE Halo-γ2GFP-Gephyrin\n50\n0\n25\nMinutes \nbefore event\nA\nB\nControl Sema4D\n0.0\n0.5\n1.0\n1.5\n2.0\nMobility\npre-colocalization\nDisplacement from origin (μm)\n0.9191\nControl Sema4D\n0.0\n0.5\n1.0\n1.5\n2.0\n10' window before\ncolocalization\nMean distance from\ncolocalization site (μm)\n0.4235\nD\nF\n✱\nC G\nFig. 9.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint \n\nFigure 9. Both GFP-Gephyrin puncta and pre-clustered Halo-γ2 puncta are mobile prior to \ncolocalizing. \n(A) Tracks followed by GFP-Gephyrin puncta prior to new colocalization events with Halo-γ2 puncta in \ncontrol (left) or Sema4D (right) treated cultures. Blue dots represent relative start locations; red dots (0,0) \nrepresent normalized location of new colocalization event. \n(B) Displacement distance of GFP-Gephyrin puncta undergoing a new colocalization event prior to \ncolocalizing. Comparison shows displacement from origin of GFP-Gephyrin puncta that colocalized with a \nHalo-γ2 puncta in Sema4D-treated cultures compared to control (p = 0.0545, Mann-Whitney U-test). \n(C) Mean distance of GFP-Gephyrin puncta from new colocalization sites in the 10-minute window \npreceding new colocalization events. Comparison shows mean distance of GFP-Gephyrin puncta undergoing \ncolocalization events with Halo-γ2 puncta in Sema4D-treated cultures compared to control (p = 0.0553). n = \n54 events (control), 34 events (Sema4D). Error bars represent SEM.\n(D) Change in mean displacement from origin of individual GFP-Gephyrin puncta in the 10 min. window \nbefore vs. the 10 min. window after colocalization. GFP-Gephyrin puncta are significantly further from \norigin after colocalization events in control neurons (*p < 0.05, Wilcoxon matched-pairs signed rank test) \nbut not in Sema4D- treated cultures (p = 0.1126). n = 54 events (control), 34 events (Sema4D).\n(E) Tracks followed by Halo-γ2 puncta prior to new colocalization events in control (left) or Sema4D (right) \ntreated cultures. Blue dots represent relative start locations; red dots (0,0) represent normalized location of \nnew colocalization event. \n(F) Displacement distance of Halo-γ2 puncta undergoing a new colocalization event with GFP-Gephyrin \nprior to colocalization. Sema4D treatment does not affect displacement of Halo-γ2 puncta undergoing \ncolocalization events with GFP-Gephyrin puncta (p = 0.9191, Mann-Whitney U-test). \n(G) Sema4D treatment does not affect mean distance of Halo-γ2 puncta from colocalization sites during the \n10-minute window preceding new colocalization events compared to control (p = 0.4235). n = 24 events \n(control), 12 events (Sema4D). Error bars represent SEM.\n(H) Change in mean displacement from origin of individual Halo-γ2 puncta in the 10 min. window before vs. \nthe 10 min. window after colocalization. Halo-γ2 puncta are not significantly further from origin after \ncolocalization events in either control (p = 0.0738, Wilcoxon matched-pairs signed rank test) or Sema4D-\ntreated cultures (p = 0.4238). n = 24 events (control), 12 events (Sema4D).\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 21, 2026. ; https://doi.org/10.64898/2026.01.21.700908doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}