A transgenic zebrafish model for in vivo long-term imaging of retinotectal synaptogenesis

The retinotectal synapse in larval zebrafish, combined with live time-lapse imaging, provides an advantageous model for study of the development and remodelling of central synapses in vivo. In previous studies, these synapses were labelled by transient expression of fluorescence-tagged synaptic proteins, which resulted in the dramatic variation of labelling patterns in each larva. Here, using GAL4-Upstream Activating Sequence (GAL4-UAS) methodology, we generated stable transgenic lines, which express EGFP-tagged synaptophysin (a presynaptic protein) in retinal ganglion cells (RGCs), to reliably label the pre-synaptic site of retinotectal synapses. This tool avoids the variable labelling of RGCs that occurs in transient transgenic larvae. We obtained several stable transgenic lines that differ consistently in the number of labelled RGCs. Using stable lines that consistently had a single labelled RGC, we could trace synaptogenic dynamics on an individual RGC axonal arbor across different developmental stages. In the stable lines that consistently had multiple labelled RGCs, we could simultaneously monitor both pre- and post-synaptic compartments by combining transient labelling of post-synaptic sites on individual tectal neurons. These tools allowed us to investigate molecular events underlying synaptogenesis and found that the microRNA-132 (miR-132) is required for developmental synaptogenesis. Thus, these transgenic zebrafish stable lines provide appropriate tools for studying central synaptogenesis and underlying molecular mechanisms in intact vertebrate brain.

at different scales. We further applied this model to monitor both pre-and post-synaptic markers simultaneously and examined the potential role of a specific microRNA (miRNA) in developmental synaptogenesis.
MicroRNAs are small, non-coding RNAs that mediate post-transcriptional gene regulation [14][15][16] . Several in vitro studies revealed that miR-132, a CREB-(cAMP-response element binding) and neuronal activity-regulated miRNA, is involved in dendritogenesis and spinogenesis by targeting the GTPase-activating protein p250GAP [17][18][19] . It is also found that miR-132 plays regulatory roles in dendritic spine plasticity 20,21 and experience-dependent ocular dominance plasticity 22,23 . However, whether miR-132 affects synaptogenesis during development remains unclear. Using in vivo long-term imaging on stable transgenic zebrafish lines, in which retinotectal synapses were fluorescently labelled, we provide in vivo evidence for the role of miR-132 in regulating developmental synaptogenesis.

Results
Visualizing retinotectal synapses in double transgenic zebrafish larvae. To generate GAL4-UAS double transgenic zebrafish for in vivo imaging of retinotectal synaptogenesis, we first created the GAL4 and UAS transgenic lines, which carry the activator and effector constructs, respectively. In activator lines, the pou4f3 (also known as brn3c) promotor drives the transcriptional activator GAL4-VP16 (Fig. 1a, top). And in effector lines, the DNA-binding motif of GAL4, 14 × UAS-E1b, controls the zebrafish synaptic vesicle protein synaptophysin b (sypb, also known as Syp in mammals) fused with enhanced green fluorescent protein (EGFP), and the whole construct 10 is flanked by Tol2 cis sequences for transposition 24 (Fig. 1a, bottom) (see also Methods). Crossing the activator and effector lines gave rise to the double transgenic fish Tg(pou4f3:   ion6d ;Tg(14 × UAS-E1b:sypb-EGFP) ion7d (PGUSG), in which the effector gene is expressed in a spatial pattern as that of the activator controlled by pou4f3 promotor 25 , including a subset of RGCs, and mechanosensory hair cells in the inner ear and lateral line (Fig. 1b,c). The Sypb-EGFP signal on RGC axonal arbors in the tectal neuropil showed a punctate structure (Fig. 1c, inset).
Synaptophysin fused to EGFP is a validated pre-synaptic marker for live imaging of synapses in animals with transient expression of the fused protein 10,13,26,27 . To validate whether Sypb-EGFP puncta in stable transgenic lines well mark pre-synaptic sites, we evaluated its colocalization with the synaptic vesicle protein 2 (SV2) by whole-mount double immunohistochemistry (IHC) with antibodies against EGFP and SV2. We examined the degree of colocalization in the developing zebrafish spinal cord where SV2 immunostaining was less dense. To achieve this, we generated another double transgenic stable line Tg(elavl3:   ion8d ; Tg(14 × UAS-E1b:sypb-EGFP) ion7d (EGUSG), in which the spinal motor neurons could be sparsely labelled by Sypb-EGFP. In 2.5-dpf EGUSG larvae, we scored the percentage of anti-EGFP puncta (Fig. 1d) on the individual spinal motor neuron that colocalized with anti-SV2 puncta (Fig. 1e) and found a high colocalization probability  Fig. 1f; 87.6 ± 2.5%, n = 11; see also Methods). These data show that Sypb-EGFP puncta in transgenic larvae largely correspond to pre-synaptic sites on neuronal axons in zebrafish.
Diverse labelling patterns of RGCs in double transgenic zebrafish. Interestingly, due to the position-effect variegation of GAL4-UAS transgenes 28 , the labelling pattern of RGC axonal arbors within the tectum is strikingly diverse in different PGUSG stable lines, embodied by the number and the topographic position of axonal arbors and the expression level of Sypb-EGFP. In one tectal hemisphere of PGUSG lines, the number of labelled RGC axonal arbors ranged from only one to ~ 30% of pou4f3-positive RGCs ( Supplementary Fig. S1). Even with the same number of labelled RGCs, the topographic position of the axonal arbors in the tectum was variable cross cells and cross fish. Fig. 2 shows examples of punctate Sypb-EGFP expression on an individual (Fig. 2a), two topographically separate (Fig. 2b), and multiple RGC axonal arbors that cover the entire tectal neuropil in the maximum projection view (Fig. 2c). This variability could be found among larvae from different lines, and each stable double transgenic line had its characteristic labelling number of RGCs.
To investigate whether the labelling pattern is preserved throughout development, we selected PGUSG larvae with one or several labelled RGCs' axonal arbor in one tectal hemisphere and performed two-photon time-lapse imaging for consecutive days from 3 to 11 dpf. We found no increase in the number of labelled RGC axonal arbors during development, indicating that once the expression pattern is determined at the time (70-72 hpf) when retinal axons reach the tectal neuropil 29 , it persists through development. Thus we could follow the development of pre-synaptic sites in transgenic fish with a single labelled RGC axonal arbor (Fig. 2d-l) as well as those with a population of labelled RGC axonal arbors (Fig. 2m-p). Therefore, PGUSG offers a reliable model for studying central synaptogenesis in vivo.
Long-term imaging of retinotectal synapse remodelling in larval zebrafish. In vivo time-lapse imaging reveals that the construction of neuronal connections is a dynamic process consisting of the concurrent formation and elimination of arbors and synapses [10][11][12][13]30 . To determine whether PGUSG is a good model for studying the synaptogenic dynamics, we performed two-photon time-lapse imaging at a 10-min interval on the same individual RGC axonal arbor expressing Sypb-EGFP ( Fig. 3a-j). We found that during 4-5 dpf, RGC axonal arbors added 182 ± 8.6% and lose 175 ± 7.6% of the initial number of Sypb-EGFP puncta within 4 h, yielding a synaptic turnover rate (N formation + N elimination /2N total ) as 16 ± 0.3% per hour (n = 20). The higher rate of synapse formation leads to a gradual increase of synapse number during development. These results further demonstrate that central synaptogenesis is a highly dynamic process consisting of rapid and nearly balanced synapse formation and elimination. Thus the PGUSG larvae with a single labelled RGC provide a suitable model for quantitatively analysing synaptogenic dynamics on individual central neurons in the developing brain. The zebrafish visual system develops rapidly 8,9 . After 7 dpf, the total length of RGC axonal arbors and tectal cell dendrites as well as the number of synaptic puncta on them remain relatively stable 10,11 . Thus the visual system of larvae older than one week reaches a relatively mature state. In the mature brain, the central synapse is known to undergo structural remodelling to support functional changes. Furthermore, as teleost fish grow continuously throughout their lifespan 31 , the retinotectal synapse must continuously undergo structural changes to coordinate growth between the retina and optic tectum. To observe structural synaptic remodelling on individual RGC axonal arbors in older larvae, we generated the PGUSG; casper line, which is transparent throughout lifespan 32 . We selected PGUSG fish that showed bright labelling on a single RGC axonal arbor at 7 dpf and performed time-lapse imaging with a 10-min interval at ~14 dpf. We observed the formation and elimination of Sypb-EGFP puncta on branches at all orders of the axonal arbor without changes in gross morphology or synapse number ( Fig. 3k-t). This optically transparent PGUSG; casper can be used as a model for studying retinotectal structural plasticity and remodelling in the intact mature brain.

Simultaneous imaging of both pre-and post-synaptic compartments in vivo.
Synaptogenesis is a complex process which requires spatially and temporally coordinated development and interaction between pre-and post-synaptic compartments 1,2 . Unlike neuromuscular junctions, neuron-neuron synaptic contacts in the central nervous system have been far less accessible due to the technical challenge in simultaneous labelling of both pre-and post-synaptic partners of single synapses. The PGUSG larvae with the Sypb-EGFP expression on a large number of RGC axonal arbors may provide us with an opportunity to image interneuronal synaptogenesis in vivo.
To visualize retinotectal synaptic contacts in the PGUSG larvae with multiple RGCs labelled, we sparsely labelled single tectal neurons with the fluorescence-tagged post-synaptic marker protein Psd95-DsRedEx which is driven by the pan-neuronal elavl3 (also known as HuC) promotor 33 (Fig. 4a). We first validated that the Psd95-DsRedEx puncta indeed mark the post-synaptic sites by showing the overlapping juxtaposition of DsRed and SV2 immuno-labelled puncta in the tectal neuropil (Supplementary Fig. S2 and Methods). Transient expression of Psd95-DsRedEx in PGUSG larvae allows us to simultaneously visualize both pre-and post-synaptic structures of retinotectal synapses. Fig. 4 shows an example of simultaneous time-lapse imaging of pre-synaptic Sypb-EGFP puncta and post-synaptic Psd95-DsRedEx puncta with a 20-min interval for 2 h at 4 dpf (see also Supplementary Fig. S3). Analysis of the time-lapse sequences revealed synaptic dynamics for both preand post-synaptic sites (Fig. 4c,e-j, coloured arrow). We observed stably maintained associations of pre-and post-synaptic puncta ( Essential role of miR-132 in developmental retinotectal synaptogenesis. In addition to the observation of synaptogenesis by live optical imaging, it is also important to investigate molecular mechanisms underlying synaptogenesis through forward and reverse genetic manipulations. Using PGUSG larvae carrying a single labelled RGC, we examined whether miR-132, a small non-coding RNA, plays roles in developmental synaptogenesis. MiR-132 is a brain-enriched miRNA and has been found to be involved in dendritogenesis and spinogenesis in vitro [17][18][19] . Whole-mount in situ of miR-132 showed that it expresses in both the optic tectum and the RGC layer (Fig. 5a). To manipulate the expression of miR-132, we microinjected antisense morpholino oligomers (MO) specific for mature miR-132 34,35 into PGUSG embryos at the one-cell stage to reduce the miR-132 level. To determine the knockdown efficiency and specificity of miR-132 MO 36 , we quantified the amount of miR-132 and another brain-enriched miR-219 37 by quantitative real-time PCR (qRT-PCR). In miR-132 morphants, the expression level of miR-132 was significantly reduced (61.2 ± 2.4%, n = 6, P < 0.001) in comparison with control fish (Fig. 5b), but the expression of miR-219 was not affected (Supplementary Fig. S4). We then performed time-lapse imaging on single RGC axonal arbors at 96 hpf and 114 hpf (Fig. 5c) and measured the change of the total number of Sypb-EGFP puncta on the same axonal arbors. We found that during 96-114 hpf, the net growth rate of puncta was significantly slower in miR-132 morphants than in control fish ( Fig. 5d; P < 0.01), indicating miR-132 is involved in the development of central synaptogenesis in vivo.

Discussion
In this study, we created the double transgenic zebrafish PGUSG to visualize retinotectal synapses in intact animals. We found that different PGUSG stable lines show different labelling patterns of RGCs, while in individual fish the pattern keeps consistent throughout development. Using in vivo two-photon time-lapse imaging of PGUSG larvae, we were able to monitor the synapse behaviour on developing RGC axonal arbors over hours, days or even weeks. We then developed methods to simultaneously image both pre-and post-synaptic compartments at putative synaptic connections in vivo. Finally, using reverse genetic manipulation on PGUSG larvae, we showed that miR-132 plays a role in developmental synaptogenesis.
Our study showed that the number and topographic position of labelled RGC axonal arbors in PGUSG larvae are highly diverse, a phenomenon similar to a previous study in which the GAL4-UAS binary system was used to transgenically express mGFP in RGCs 38 . This diversity may be derived from two sources: (1) the transgene of the activator line. We found that even crossed with the same UAS reporter line, different GAL4-VP16 lines allowed us to label different numbers of RGCs. This variegated (mosaic) expression of UAS-reporters seems to be an intrinsic property in transgenic lines expressing GAL4-VP16 driven by specific promotors that were created by conventional transgenesis 28 ; (2) the transgene of the effector line. Although the effector lines were created by Tol2-mediated transgenesis, which can lead to consistent transgenic labelling among siblings for GAL4 transgenic lines 28,39 , we found different UAS lines show different labelling patterns when crossed with the same ubiquitously expressed GAL4 line, further supporting the fact that the UAS-transgenes are subjected to position effects 28 . Importantly, despite the variability cross fish and lines, the labelling patterns were consistent throughout development for individual PGUSG fish.
As the GAL4-UAS system is a binary system, it brings the flexibility by combining the USG line with other tissue-specific GAL4 lines to study synaptogenesis of other types of neurons at single cell level, or vice versa, by combining the GAL4 line with UAS transgenic lines carrying other reporter genes and effector genes. For instance, the expression of the reporter gene like calcium indicators fused to Sypb would allow simultaneous monitoring of both the structure and function of retinotectal synapses 40 , while the expression of effector gene like the fluorescent protein-tagged tetanus toxin light-chain (TeNT-Lc) 41 , which can block synaptic transmission, would allow us to examine how synaptic activity regulates synaptogenic dynamics on single retinal axons and neighbouring axons under competition 42,43 .
Although synaptic remodelling is a hall mark of early development, synapse formation and elimination are still prevailing in the mature brain to fine-tune neural circuits to adapt to changing environments. To visualize retinotectal synapses in more mature brains of zebrafish, we created PGUSG;casper. Taking advantage of its  lifelong transparency, we could perform in vivo time-lapse imaging of retinotectal synapses on individual RGC axonal arbors in fish of two-weeks old. We found that, at this stage, the axon branching pattern and punctum number are stable, but the punctum turnover still occurs frequently at the timescale of minutes. This is comparable to dynamic dendritic spines and synaptic boutons observed in the cerebral cortex of adult mammals 44,45 . Besides basal dynamics, in vivo studies provide strong evidence for supporting the structural reorganization of synaptic connections during learning and memory 44,[46][47][48][49] . As it is possible to image synaptic remodelling at high temporal resolution (e.g. 10-min interval) on older PGUSG; casper larvae, we can combine behavioural assays 50 to study how retinotectal synapses change with visual perception-based and behaviour-related experience alteration, learning, and memory in intact larvae.
Synaptogenesis is the formation of cellular junctions between neurons and their targets. We cannot understand the wiring of neuronal circuits without concurrent observing of both pre-and post-synaptic elements. At the mammalian peripheral neuromuscular junction, in vivo concurrent imaging of multiple synaptic elements has revealed how multiply innervated junctions transform to singly innervated ones during development 51 . However, due to the technical challenge for in vivo imaging of pre-and post-synaptic compartments of identified connections over time during development, details of the formation, maintenance and elimination of central synapses remain largely unknown. One study has attempted to image pre-and post-synaptic sites in vivo in the submandibular ganglion in the peripheral nervous system of mice 52 . Here, we found a way of simultaneous imaging of pre and post-synaptic elements of central retinotectal synapses in vivo by expressing Psd95-DsRedEx in a single tectal neuron in PGUSG larvae with a large population of RGCs labelled by Sypb-EGFP. Although the pre-and post-synaptic markers we used might not be the earliest protein involved in synapse assembly and disassembly, this method allows us to study the dynamic interactions between pre-and post-synaptic elements during synapse formation, stabilization and elimination. Using MO-mediated gene-specific knockdown, we demonstrated that miR-132 is important for developmental synaptogenesis. Knockdown of mature miR-132 significantly reduced the synapse growth rate. As synapse formation and elimination occur concurrently during development, miR-132 may regulate the rate at which new synapse form and/or the stability of existing synapse. In the future, using time-lapse imaging with high temporal resolution, it is of interest to examine which dynamic processes of synaptogenesis are regulated by miR-132.
In mammals, miR-132 regulates dendrite development and spine formation by inhibiting synthesis of p250GAP 17,18 , which specifically inactivates Rac1 activity in cultured hippocampal neurons 18 . Therefore, it was hypothesized that miR-132 increases the activity of Rac1, which is a crucial regulator of the actin cytoskeleton, and thus promotes spine growth by suppressing p250GAP 18,53 . In our study, miR-132 might also regulate retinotectal synaptogenesis by targeting p250GAP. Notably, in our experiments, miR-132 regulated the growth of pre-synaptic terminals, while p250GAP has been shown to be enriched in the post-synaptic density 54 . So miR-132 might regulate post-synaptic development and then retrogradely influence pre-synaptic terminals. Alternatively, miR-132 might directly regulate pre-synaptic terminals by targeting the same or different substrates locating in RGCs.
In sum, the double transgenic zebrafish PGUSG is a versatile and reliable model for in vivo long-term imaging of retinotectal synapses. We demonstrated the applicability of this model in the long-term imaging of structural synaptic remodelling in both developing and relatively mature brain at the single-cell level, the long-term simultaneous imaging of pre-and post-synaptic elements, and the investigation of molecular mechanisms underlying synapse growth. Therefore, the PGUSG provides a transgenic stable model for studying central synaptogenesis in intact vertebrate brains.

Generation of DNA constructs and transgenic lines.
To generate Tg(pou4f3:GAL4-VP16);Tg(1 4 × UAS-E1b:sypb-EGFP) (PGUSG), the PG and USG lines were generated independently. For PG, pou4f3:-GAL4-VP16 plasmid (gift from Dr. Rachel Wong, University of Washington, Seattle, USA) was linearized with AflII, purified by QiaexII gel extraction kit (Qiagen) and microinjected into embryos at 1-2 cell stage at a concentration of 40 ng/μl. For USG, the Tol2 transposon system 24 was used. The USG fragment was excised with DraIII-XhoI from UAS:sypb-EGFP plasmid 10 and subcloned into the BglII-XhoI sites in Tol2-MCS-EGFP plasmid (gift from Bo Zhang, Peking University, Beijing, China) to replace MCS-EGFP. Purified Tol2-USG plasmid was microinjected together with Tol2 transposase mRNA synthesized in vitro into embryos at 1-2 cell stage at a concentration of 50 (25 + 25) ng/μl. Injected PG and USG embryos were raised to sexual maturity and crossed with available UAS and GAL4 line, respectively, to identify founder fish. Founders of PG and USG were crossed and the embryos from these crosses were scored for their EGFP expression and raised. This led to the production of several PGUSG stable lines.
Validation of Sypb-EGFP as a pre-synaptic marker. We first validated transgenic Sypb-EGFP as a pre-synaptic marker by examining the percentage of colocalization of Sypb-EGFP with a known pre-synaptic marker SV2 in developing spinal motor neuron. In transgenic EGUSG larvae, the spinal motor neurons can be sparsely labelled by Sypb-EGFP. We performed whole-mount double-IHC on these fish at 2.5 dpf with EGFP and SV2 antibodies. Then we scored the colocalization of anti-EGFP punctum with anti-SV2 punctum by evaluating their overlap (yellow signal) through each optical section in the composite image stack using ImageJ software. And we found 87.6 ± 2.5% of anti-EGFP puncta colocalize with anti-SV2 puncta. Furthermore, based on time-series projections of RGC axons expressing Sypb-EGFP with high temporal resolution (10 min interval), more than 90% of Sypb-EGFP puncta could be distinguished as transporting cluster or not according to their kinetics and we found about 85% were not. Collectively, transgenically expressed Sypb-EGFP puncta on axonal arbors represent pre-synaptic sites in zebrafish.
Validation of Psd95-DsRedEx as a post-synaptic marker. We also validate Psd95-DsRedEx as a post-synaptic marker by examining its colocalization with the pre-synaptic marker SV2. We fixed fish expressing Psd95-DsRedEx in single or several tectal cells and immuno-labelled them with DsRed and SV2 antibodies on 20-30 μm horizontal cryostat sections at 4 dpf. Due to the high density of anti-SV2 puncta in tectal neuropil, we performed high-resolution confocal imaging of the sections with oil immersion lens (60×) and applied image SCIENtIfIC RepoRts | (2018) 8:14077 | DOI:10.1038/s41598-018-32409-y deconvolution (Huygens Essential) afterwards to further enhance the resolution. We could not find the full overlap of the anti-DsRed and anti-SV2 puncta, but the immediate juxtaposition of the two elements. The juxtaposition is defined as the immunofluorescence intensity overlap >50%, and we found 85.1 ± 1.5% of anti-DsRed puncta juxtapose anti-SV2 puncta. Furthermore, in the tectal neuron, the density of Psd95-DsRedEx puncta along dendrites did not change systematically with distance from the soma. Taken together, the Psd95-DsRedEx puncta are acting as a post-synaptic marker.
In vivo imaging and data analysis. Zebrafish larvae at 3-14 dpf were anesthetized with 0.02% tricaine methanesulfonate (Sigma) or paralyzed with pancuronium dibromide (PCD, 1 mM, Tocris) and then mounted in 1.5% low-melting agarose (Sigma) for imaging, after which the fish were released immediately and reared normally. Two-photon and confocal imaging were performed under 40× (N.A., 0.8) or 20× (N.A., 0.95) water-immersion objective on an upright microscope equipped with a two-photon laser (900 nm) and single-photon lasers at 488 nm and 559 nm, respectively. The step size for z-stack imaging is 1-2 μm.
Punctum counting was performed on maximum intensity projections of image stacks using Image-Pro Plus software. Puncta were identified as discrete local accumulations in Sypb-EGFP of more than 2 native pixels (0.4 μm) in diameter.

Study approval.
All experiments were carried out in accordance with Chinese ethical guidelines for the care and use of laboratory animals and approved by the Institute of Neuroscience, Chinese Academy of Sciences.

Data Availability
The datasets generated in the current study are available from the corresponding author on reasonable request.