Robust Type-specific Hemisynapses Induced by Artificial Dendrites

Type-specificity of synapses, excitatory and inhibitory, regulates information process in neural networks via chemical neurotransmitters. To lay a foundation of synapse-based neural interfaces, artificial dendrites are generated by covering abiotic substrata with ectodomains of type-specific synaptogenic proteins that are C-terminally tagged with biotinylated fluorescent proteins. The excitatory artificial synapses displaying engineered ectodomains of postsynaptic neuroligin-1 (NL1) induce the formation of excitatory presynapses with mixed culture of neurons in various developmental stages, while the inhibitory artificial dendrites displaying engineered NL2 and Slitrk3 induce inhibitory presynapses only with mature neurons. By contrast, if the artificial dendrites are applied to the axonal components of micropatterned neurons, correctly-matched synaptic specificity emerges regardless of the neuronal developmental stages. The hemisynapses retain their initially established type-specificity during neuronal development and maintain their synaptic strength provided live neurons, implying the possibility of durable synapse-based biointerfaces.


Results and Discussion
Construction of modularly-tagged synaptic CAM ectodomains. To generate type-specific artificial dendrites, ectodomains of NL1 (48-639), NL2 , and Slitrk3  were excerpted from their full sequences. NL1 ectodomain was determined by the minimally secreted functional domain 19 and the crystal structure that reveals Leu636 as the end of the α -helix required for NL1 dimerization 20,21 that induces presynaptic differentiation via clustering of presynaptic neurexins (Nrxs). Ectodomain of NL2  was determined by sequence comparison with NL1. The Slitrk3 ectodomain contains two leucine-rich repeat domains (LRR1 and LRR2) followed by several extra C-terminal residues.
To facilitate massive production, purification, quantification, and localization of the ectodomains, we designed a tagging module that can be generally applied to virtually all kinds of synaptic CAM ectodomains. We began by engineering the NL1 ectodomain by placing YFP or RFP either at the N-or C-terminus of the protein and a biotin acceptor peptide (AP, also called AviTag) 22 at the end of the expression cassette to utilize biotin-streptavidin (SAV) interaction for immobilization of the protein (Figs 1 and S1a,b). The in vivo biotinylation was further simplified by bicistronic expression of endoplasmic reticulum-specific bacterial biotin ligase (BirA-ER) (Fig. 1b).

C-terminal fusion is optimal for NL1 ectodomain function in vitro.
Although the previously reported excitatory hemisynapses were induced by NL1 ectodomain docked on obligate supported lipid bilayer (SLB) membrane 5,6 , obviating the lipid layer should simplify the fabrication process of the artificial dendrites and greatly extend the types of surface on which synaptic proteins are immobilized. In addition, we can minimize a hindrance to communication, if possible, between neurons and the artificial dendrites. When we compared the synaptogenic activity of the N-or C-terminally-tagged NL1 ectodomains immobilized on SAV-coated microbeads (Dynabeads M-280 Streptavidin, 2.8 μm in diameter) and silica microbeads (5 μm in diameter) covered with SAV-loaded fluidic SLB membrane (Fig. S1c), only C-terminally tagged NL1-R immobilized without lipid layer showed the most prominent synaptogenic activity by showing aggregated synapsin I, a general synapse marker protein (Fig. 2). The synaptogenic potential of NL1-R was greater than the one of poly-D-lysine (PDK) that is also known to induce the formation of synaptic boutons with mature neurons 10,23,24 . We then applied the same tagging module to the inhibitory NL2 and Slitrk3 ectodomains to establish a complementary type-specificity, generating NL2-R and SL3-R (Fig. 1a).

NL1-R forms durable glutamatergic presynaptic boutons independently of developmental stages of neurons.
The PDK-coated beads serve as a synaptogenic substrates only when they are introduced to dissociated hippocampal cells after a certain developmental stage (at least DIV7) 10 . The PDK-induced hemisynapses, instructed by F-actin reorganization and heparan sulfate proteoglycans (HSPGs), increased after 24 h incubation and could be observed up to 72 h 10 . Unfortunately, they lack synapse specificity, inducing the formation of both excitatory and inhibitory synaptic boutons. Postsynaptic NL1, on the other hand, is a negatively charged protein that specifically binds to β -Nrx expressed on outgrowing axonal surface, thereby regulating the synapse specificity. However, synaptic CAMs can induce the formation of mismatched synapses between glutamatergic and GABAergic synaptic terminals in the early developmental stages [23][24][25] . We thus examined the synaptic specificity of the NL1-coated artificial dendrites by seeding them to mixed culture of neurons in various developmental stages. When we seeded the PDK and NL1-R beads from the beginning of the neuron culture (DIV0) and incubated up to DIV8 and 15, we found that in both durations only the NL1-R-coated beads successfully induced excitatory presynaptic differentiation, showing immunoreactivity toward synapsin I and VGluT1, an excitatory synapse marker protein, but not toward vesicular GABA transporter (VGAT), an inhibitory synapse marker protein (Figs 3a,b and S2a,b). As expected, PDK could induce presynaptic differentiation when they met only with the fully developed cells (DIV14), showing slightly weaker hemisynapses than NL-1R as shown in (Figs 2c,3c and S2c). When we elongated the incubation up to 8 d, however, the majority of the synaptic boutons generated by the PDK beads diminished while the ones induced by the NL1-R beads increased (Figs 3d and S2d). When the beads were seeded on DIV7, only NL1-R exerted synaptogenic activity upon incubation for 8 d (Figs 3e-h and S2e-h), similar to the result for NL1-R seeded on DIV14 (see Figs 3c and S2c).
The polybasic molecules such as PDK, if applied to developed neurons, can satisfy the eager for the synaptogenesis of outgrowing axons by allowing hemisynapse formation. It seems, however, that the factors sustaining hemisynapses become accustomed to the PDK molecules and eventually lose their synaptogenic potential upon extended period of coincubation. In comparison, the NL1-R artificial dendrites not only induced excitatory hemisynapses with neurons at any developmental stages, but also strengthened and maintained the excitatory specificity of the synapses as long as live neurons are provided.
NL2-R and SL3-R forms inhibitory presynaptic boutons only with mature neurons. We next applied the NL2-R-and SL3-R-coated inhibitory artificial dendrites on neuron cultures in various developmental stages to compare the timing of the instruction of inhibitory hemisynapses. When we added NL2-R or SL3-R beads to immature neuronal cells (DIV0 and 7), we could observe only excitatory presynaptic differentiation regardless of the coculture duration (Figs 4a-d and S3a-d,h-k). The inhibitory synaptic specificity appeared when the coculture started on DIV9 and was fully restored on DIV14 (Figs 4e-g and S3e-g,l-n). Thus, it was the timing of the encounter with artificial dendrites that determined the inverted synaptic specificity of growing neurons.
In the developing hippocampal neurons, GABAergic synapses are formed before glutamatergic synapses 26 . In the dissociated hippocampal neurons, however, the formation of glutamatergic synapses typically precedes the one of GABAergic synapses 26,27 , where the mismatched synapses form frequently. This mismatch diminishes as neurons mature, probably promoted by the regulated expression of type-specific synaptic CAMs or by increased transmitter activity between matched synaptic machineries 26 including a signalling for proteolytic cleavage of synaptic adhesion proteins for sculpting of the mismatched synapses 28 . By contrast, our abiotic artificial dendrites are immune to neurotransmitter signals and proteolytic cleavage, not allowing correction of initially formed mismatched specificity. It is possible that the NL1-R-induced glutamatergic hemisynapses found in the immature hippocampal neurons (see Fig. 3) originate in part from a simple mismatch independent of the type-specific    The NL2-R and SL3-R beads were seeded and analysed for the synaptogenesis as in the case of NL1-R. (a-g) ICC images for NL2-R-induced hemisynapses are aligned on upper lines. (a-d) Only excitatory presynaptic differentiations appeared during the early neuronal development stages, which lasted until DIV15. Correctly matched synapses began to appear on DIV9 when the glutamatergic specificity is still dominant (e). The fully inhibitory hemisynapses could be formed only when the artificial dendritic beads are in contact with mature neurons (DIV14) (g). * * p < 0.01; others, p < 0.001 between the same marker proteins of NL2-R and SL3-R.
Scientific RepoRts | 6:24210 | DOI: 10.1038/srep24210 synaptic CAMs. Thus, in classical culture of hippocampal neurons, the synaptic specificity defined by the synaptic vesicle proteins is determined not by the synaptic CAMs but by the degree of neuronal development.
Axonal compartmentalization allows the formation of type-specific synapse networks with artificial dendrites. In the brain, neurons manifest a polarized subcellular structure and complex geometry through which the formation of synapses by the synaptic CAMs is chronologically and geometrically controlled. In order to address the functional importance of the subcellular synaptic specificity, we reconstituted spatial compartmentalization using microfluidic culture chamber that enables isolation of axons from somata in an analogous fashion to the patterned hippocampal neural networks 29,30 . The directional growth of neurons provides a homogeneous population of axons with which the postsynaptic CAM-covered solid substrates can form hemisynapses without intervening endogenous dendrito-and somato-axonic synapses (Fig. 5a-c).
About one week after seeding the neuronal cells in the left somal side of the chamber, axons escaped the microchannels to reach the right compartment of the chamber and covered about two-thirds of the field on DIV14. After brief verification of synaptogenic activity of NL1-R beads (Fig. S4), we applied artificial dendrites displaying NL1-R, NL2-R, and SL3-R to the axonal compartments and compared the developmental pattern of the synaptic specificity. As expected, in the axonal compartment of adult neurons the artificial dendrites exerted correctly matched synaptic specificity, similar to the results obtained from random-cultured mature neurons (Figs 4d and S5a; also see Figs 2 and 3). Surprisingly, when we seeded the synaptic CAM beads on DIV7 and incubated for 8 d, there was no inversion of synaptic specificity caused by inhibitory NL2-R and SL3-R beads (Figs 4e and S5b), showing similar patterns of synaptic puncta between the beads near and far from the microchannels (Fig. S6). Thus, in order to induce correctly matched hemisynapses, the artificial dendrites appear to require isolated axonal components of cultured neurons.

NL1-R-induced presynaptic boutons are glutamatergically active.
Having established the polarized artificial synaptic interfaces, we wanted to ensure the synaptic functionality of the hemisynapses to genuinely mimic the directional flow of neuronal information via neurotransmitters. We thus engineered the intensity-based glutamate-sensing fluorescent reporter (iGluSnFR) 33 to directly measure the rapid glutamate transients. The engineered iGluSnFR, dubbed iESFb, shares C-terminal tagging module with our synaptic CAM ectodomains and has the sensitivity enough for our experiments (will be reported elsewhere). We placed the NL1-R beads in the axonal side of the culture chamber on 14 DIV, allowing two days of hemisynapse formation before spreading the freshly prepared iESFb beads (Fig. 6a,b). Upon potassium shock in the somal side of the chamber, we could observe an increase in green fluorescence of iESFb near NL1-R beads in comparison to SAV beads without NL1-R, indicating that the hemisynapses possess type-specific neurotransmitter activity (Fig. 6c).
Abiotic solid substrata displaying engineered ectodomains of postsynaptic CAMs can serve as artificial dendrites, inducing the formation of durable excitatory and inhibitory hemisynapses with spatiotemporal regulation. The ectodomains of NL1, NL2, and Slitrk3 allow C-terminal tagging for optical tracking and surface immobilization that dispenses with lipid bilayers for its synaptogenic activity. The modular C-terminal tags may change ectodomains of virtually any pre-and postsynaptic CAMs that have functional extracellular domains. Notably, the artificial dendrites could instruct the formation of inhibitory hemisynapses with dissociated culture of well-developed neurons and with isolated axonal component of micropatterned neurons even in their early developing stages. Once determined, either the matched or mismatched synaptic specificity of hemisynapses remained unchanged during the extended neuronal development. Such robust hemisynapses should bode well for a fundamental study on information process of neural networks in the brain and novel type of synapse-based biointerfaces where signals are transduced via synaptic boutons with defined synaptic specificity and spatial geometry.

Stable cell lines.
For the initial screening of synaptogenic activity of NL1 ectodomains, NL1-encoding plasmid DNA (pNICE-H8-YFP-NL1-AP, pNICE-H8-RFP-NL1-AP, and pNICE-HA-H6-NL1-H8-RFP-AP) (24 μg of each) was added to the Opti-MEM I Reduced Serum Media (1.5 mL). Likewise, 1.0 mg/mL PEI (60 μl) solution was added to the Opti-MEM solution (1.5 mL). After 5 min incubation at 25 °C, the two solutions were mixed at room temperature for 30 min and added to HEK293 cells grown to ~20% confluence in a culture dish of 10 cm diameter at 37 °C. The DMEM medium was replaced after 4 h incubation. After three days, the cells were treated with G418 (final conc. 0.8 mg/mL). The G418 treatment was repeated with fresh medium after two days. After two weeks, single colonies with brightest fluorescence signals were picked and seeded on a 24-well plate. Among them, the best fluorescent colonies were repeatedly selected until only one colony was left and seeded on a culture dish of 10 cm diameter for the subsequent passage. The established stable cell line was kept in DMEM medium containing G418 (100 μg/mL).

Preparation of biotinylated synaptic CAMs. Plasmid pDisplay-BirA-ER (24 μg) in Opti-MEM solution
(1.5 mL) was mixed with Opti-MEM solution (1.56 mL containing 60 μg of PEI) for 20 min at 25 °C. The mixture was added to the established HEK293-H stable cell lines at ~20% confluence in a culture dish of 10 cm diameter. After 4 h incubation, the DMEM medium was replaced with a fresh one containing G418 (100 μg/mL) and biotin (10 μM). The in vivo biotinylated synaptic CAM was allowed to secrete into culture medium for three days at 37 °C. Then, the medium (10 mL) was saved and the whole cells were transferred to a culture dish of 15 cm diameter and filled with 30 mL of DMEM containing G418 (100 μg/mL) and biotin (10 μM). After another three days, the culture media were combined and subjected to column purification using 50% Ni-NTA resin (2 mL) according to the manufacturer's protocol. The three elution fractions (1 mL) were subjected to fluorescence analysis using Synergy Mx fluorescence microplate reader (BioTek, Seoul, Korea).
For in vitro biotinylation, the stable cell line was grown without BirA transfection. To the column elution fraction (1 mL) showing the highest fluorescence signal was added MgCl 2 (5 mM), ATP (1 mM), biotin (0.1 mM), and BirA enzyme (30 nM) as final concentration and shake incubated for 2 h at 37 °C. The level of in vivo and in vitro biotinylation was analysed via western blot using SAV-HRP or using goat or mouse anti-NL1 antibody and HRP-conjugated secondary antibody. The purity of NL1 was analysed by SDS-PAGE with silver staining.
Preparation of PDK microbeads. Silica microbeads (2.56 and 5.06 μm in diameter, 3 × 10 5 beads) were incubated with PDK solution (0.5 mg/mL in DPBS) overnight at room temperature, washed three times with DPBS, and spread on cultured neurons per condition. For the coincubation with the artificial dendrites, 0.5 × 10 5 beads were spread. Otherwise, 1.0 × 10 5 beads were used to examine the formation of the PDK hemisynapse.
Quantification of synaptic CAMs on SLB microbeads. The synaptic CAM-SLB microbeads (7.5 × 10 4 beads) solution was micro-centrifuged for 1 min and the supernatant was removed. To this was added 50 μl of TBS/2% Triton X-100 followed by brief vortexing. After 10 min incubation, the mixture was micro-centrifuged for 1 min and the supernatant was transferred to a 96-well plated to read fluorescence by Synergy Mx fluorescence microplate reader (BioTek, Seoul, Korea). The amount of the proteins was calculated by applying conversion factor, 19,000 MFI/μL for NL1-R, obtained by SDS-PAGE using BSA standard and relative molecular mass (M r ) of the protein, 135,000.
Neuronal cell culture. Primary hippocampal neurons were obtained from Sprague-Dawley rat embryos at day 18 of gestation (E18). Briefly, hippocampus dissected from E18 rat embryos were rinsed with HBSS, and then incubated with papain and DNase with 60 rpm for 30 min at 37 °C. After sequential rinsing with solution of 10% and 5% FBS in HBSS, individual single cells were mechanically isolated by trituration 10 times in 2 mL HBSS containing DNase with silanized Pasteur pipette of which tip was barely polished with fire. The cell suspension was diluted to density of 2 × 10 5 cells/mL with plating media containing MEM supplemented with 0.6% (w/v) glucose, 10 mM sodium pyruvate, 1 mg/mL FBS, and 1% penicillin-streptomycin. Then, the cell-media solution plated on the PDK-coated glasses placed in Petri dish. After 3 h the cell culture media was exchanged with B27-supplemented Neurobasal Media containing 2 mM Glutamax. Cultures were maintained in an incubator at 37 °C and 5% CO 2 atmosphere. For microfluidic neuronal cell culture, 50 μL of hippocampal neuronal cells (1.5 × 10 5 cells/mL) was seeded in the left compartment of the PDMS chamber. Immunocytochemistry (ICC). Cells were fixed using 4% formaldehyde for 25 min and rinsed three times with PBS (100 mM, pH 7.4). The cells were then incubated in blocking solution, containing 4% BSA and 0.1% Triton X-100, dissolved in PBS for 30 min, followed by incubation in primary antibodies diluted in TBS (Tris-buffered saline, pH 7.4) containing 0.5% BSA and 0.1% Triton X-100, overnight at 4 °C. The samples were then washed three times with TBS and the fluorescent secondary antibodies were applied for 1 h at room temperature in TBS containing 0.5% BSA solution. The samples were washed again three times with TBS and once with ultrapure water, and stored in VECTASHIED Mounting Medium at − 80 °C until microscopic examination. Fluorescence images were taken with a Zeiss LSM710 confocal laser scanning microscope equipped with ZEN 2009 software at the National Centre for Inter-university Research Facilities (NCIRF) of Seoul National University.
Image quantification and analysis. To reflect both the frequency of synaptogenesis and the intensity of the hemisynapses formed by the artificial dendrites, at least 30 beads were randomly chosen as regions of interest (ROI) from the brightfield (differential interference contrast, DIC) image by excluding somal regions per condition per experiment using ZEN 2009 software. The same-sized ROI with the similar neurite density and without the bead was set as a control. The ROI profile was applied to each fluorescence channel and the intensity (MFI) was collected by ZEN 2009 software. For the time-dependent experiments (Figs 3 and 4), the MFI values of each channel was separately normalized to the maximum MFI values of the respective marker proteins. For the time-and compartment-dependent experiment using micropatterned neurons (Fig. 5), the overall MFI values of VGluT1 and VGAT were first normalized to the ones of Syn I, followed by the individual normalization to the maximum MFI values of the respective marker proteins. All images were prepared for figures using Photoshop (Adobe).
Statistics. All statistics were performed using Origin 8.0 software. For comparisons of MFI between multi-group datasets, we carried out one-way ANOVA with Bonferroni's method. All data shown are mean ± SEM (standard error of the mean). In figures, statistical significance is indicated by n.s. for p > 0.05, *for 0.05 > p > 0.01, **for 0.01 > p > 0.001, and ***for p > 0.001.