To test the hypothesis that AMPAR-mediated glutamatergic neurotransmission is required for dendritic arbor growth we expressed GluR1Ct or GluR2Ct peptides that interfere with AMPAR synaptic expression (18–20) in optic tectal neurons in the Xenopus retinotectal system. We tested the effect of GluR1Ct and GluR2Ct expression on AMPA transmission in optic tectal neurons by whole-cell patch-clamp recordings of AMPAR spontaneous miniature excitatory postsynaptic currents (mEPSC) 2 and 3 days after transfection. Expression of GluR1Ct or GluR2Ct significantly reduced mEPSC amplitude (Fig. 1), but had no significant effects on AMPAR mEPSC rise-time or decay kinetics (data not shown). Our finding that both GluR1Ct and GluR2Ct reduced AMPAR transmission under normal rearing conditions suggest that both activity-dependent, GluR1 subunit-mediated, and constitutive synaptic receptor recycling by GluR2 subunit-containing receptors, are required for AMPAR insertion associated with maturation of immature synapses (8, 9, 21).

Fig. 1. GluR C-terminal domain peptides reduce AMPAR synaptic currents. Representative traces of AMPAR-mediated mEPSCs from tectal neurons expressing GluR1Ct (A) or GluR2Ct (B) and their untransfected neighbors 2–3 days after transfection. (C) Cumulative (more ...)

In vivo single-cell electroporation (22) targeted coexpression of GFP and GluR1Ct or GluR2Ct to individual tectal neurons in the intact tadpole. This selectively reduced AMPA transmission in transfected neurons, while leaving the remainder of visual system circuitry intact. We detected long-term effects of reduced AMPAR transmission in individual neurons by in vivo time-lapse two-photon microscopy of single developing neurons at 24-h intervals over 5 days (Fig. 2). Dendritic arbors of control tectal neurons grew by extending primary dendrites and concurrently adding higher-order interstitial branches (Fig. 2A). Over subsequent days, arbor complexity increased as higher-order branches extended and supported further addition of interstitial branches. Three-dimensional neuron reconstructions demonstrate that total dendritic branch length and dendritic branch number increased over the first 4 days of imaging, after which growth plateaued with no further net increase in branch length or number (Fig. 2 B and C).

Fig. 2. GluR C-terminal peptides alter patterns of long-term dendritic arbor growth. (A) In vivo time-lapse images acquired at 24-h intervals over 4 days starting 24 h after transfection. Control tectal neurons elaborate complex dendritic arbors by iterative (more ...)

Although dendrites of tectal neurons expressing GluR1Ct or GluR2Ct were similar to controls 1 day after electroporation, their growth patterns quickly diverged from controls. Both peptides produced the same pattern of dendrite growth, characterized by less complex arbors with longer major dendrites than controls (Fig. 2 A–C). Total dendritic branch length and the number of branches were significantly reduced in neurons expressing GluR1Ct or GluR2Ct from 2 to 5 days after transfection, and no further net dendrite growth was observed in GluRCt-expressing neurons after 3 days. These data indicate that neurons expressing GluR1Ct and GluR2Ct were able to grow, but at a reduced rate compared to controls.

We used three-dimensional Sholl analysis to quantify dendritic arbor complexity (Fig. 2 D and E). Over the first 4 days after transfection, control neurons dramatically increased dendritic arbor complexity within a circumscribed area, typically between 50 and 80 μm from the soma. By contrast, tectal neurons expressing GluR1Ct or GluR2Ct failed to substantially increase dendritic arbor complexity in this region after the second day of imaging. In addition, GluR1Ct and GluR2Ct neurons extended dendrites farther from the soma than controls on days 4 and 5. On average, 4 days after transfection, control neurons extended dendrites 100.6 ± 4.4 μm from the soma, compared to 120.8 ± 8.1 μm for GluR1Ct, and 121.0 ± 10.4 μm for GluR2Ct neurons (P < 0.05).

If AMPAR trafficking into developing synapses is required to stabilize the new dendritic branches, expression of the GluR C-terminal peptides is predicted to affect branch dynamics. To test this hypothesis, neurons expressing GFP with or without coexpression of the GluRCt were imaged at 2-h intervals over 6 h starting 3 days after transfection. This protocol allowed the dynamic behaviors of dendritic branches to be followed between subsequent 2-h time points (Fig. 3). Dendritic branches of developing neurons were highly labile: 31.5 ± 3.4% of branches were retracted within 2 h and 53.4 ± 4.0% were retracted within 6 h (Fig. 3 B and C). Neurons expressing GluR1Ct and GluR2Ct were significantly more dynamic: 49.4 ± 4.6% and 48.5 ± 4.3% of branches retracted within 2 h, whereas 66.3 ± 4.1% and 66.5 ± 2.5% of branches retracted over 6 h, respectively. Importantly, neurons expressing GluR1Ct or GluR2Ct had significantly more transient branches with lifetimes <4 h, the shortest lifetime we can estimate from images collected at 2-h intervals (Fig. 3D). Therefore, dendritic branches of neurons with reduced AMPAR transmission were less stable resulting in higher retraction rates and shorter branch lifetimes. It is interesting to note that GluRCt-expressing neurons also show increased rates of branch additions compared to controls (Fig. 3B), suggesting that branch retractions, stability, and branch additions may be coordinated.

Fig. 3. GluR C-terminal peptides reduce dendritic branch stability. Two-hour time-lapse images demonstrate the dynamic behavior of individual dendritic branches. (A) Color-coded three-dimensional renderings of one control and one GluR1Ct tectal neuron imaged (more ...)

The data from the above experiments predict that GluRCt-expressing neurons have more immature synapses than control neurons. To test this possibility, we collected electron microscopic images of synapses in control and GluRCt-expressing neurons. A distinctive feature of developing synapses is an increase of synaptic vesicles per axon terminal profile (23, 24). To quantify synapse maturity, we measured the area of the presynaptic element filled with clustered synaptic vesicles (25) and normalized this value to the area of the presynaptic terminal (Fig. 6, which is published as supporting information on the PNAS web site). The relative area of the presynaptic terminal occupied by clustered synaptic vesicles increases almost two-fold during early larval stages of Xenopus development (Fig. 4 A–C). Presynaptic contacts onto GluRCt-expressing neurons appeared less mature than synapses on control GFP-expressing or nontransfected neurons and had less than half as much area occupied by synaptic vesicle clusters than GFP-expressing controls (16.3% vs. 38.8%, P < 0.001, Mann–Whitney, n = 28 and 45 synapses, respectively; Fig. 4 A–C). These data are consistent with the idea that presynaptic elements were affected by postsynaptic expression of GluR C-terminal peptides. In addition, synaptic contacts onto GluRCt-expressing neurons are preferentially located on large-caliber dendrites (Fig. 4D, P < 0.001, Kolmogorov–Smirnov test), suggesting that postsynaptic expression of GluRC terminal peptides resulted in a shift in the distribution of synapses from dynamic fine-caliber dendrites to larger-caliber, more stable dendrites.

Fig. 4. GluR C-terminal peptides prevent synapse maturation. (A) Electron micrographs show ultrastructural morphology of synaptic terminals that contact nontransfected (Left, stage 39; Center, stage 47) and GluR1Ct-transfected (Right, stage 47) dendrites. Arrows (more ...)

Previous studies indicate that activity-dependent mechanisms, such as long-term potentiation, enhance synaptic strength through the delivery of AMPAR to glutamatergic synapses after sensory experience (19) and associative learning (20). Other experiments indicate that sensory experience affects dendritic arbor structure (2, 26–28), but they do not necessarily demonstrate a corresponding change in synaptic strength. Finally, changes in spine size correlate with changes in its synaptic strength (11, 12, 29). Here we tested whether experience-dependent structural plasticity requires increased synaptic expression of AMPAR and might therefore share a mechanistic basis with experience-dependent changes in synaptic strength.

We imaged individual optic tectal neuron dendritic arbors before and after exposure of freely swimming tadpoles to 4 h of darkness, during which retinal ganglion cells are thought to exhibit low levels of spontaneous uncorrelated activity, followed by 4 h of moving-bar visual stimulation (2), during which cells show increased activity (30). Tectal neurons were electroporated with either GFP alone, or with GluR1 or GluR2 C-terminal peptides and imaged 3 days later when neurons were in their dynamic growth phase. Control tectal neurons grew little or retracted during the initial 4 h without visual stimulation, but dramatically increased their dendrite branch numbers and growth rates in response to 4 h of visual stimulation (Fig. 5 A and B). Expression of GluR C-terminal peptides blocked the increase in branch length and branch tip numbers normally seen with visual stimulation. These data indicate that AMPAR transmission is required for experience-dependent structural plasticity.

Fig. 5. GluR C-terminal peptides block experience-dependent dendrite arbor plasticity. Retinal input activity to the optic tectum was increased by exposure of freely swimming tadpoles to moving-bar visual stimulation for 4 h after 4 h in the absence of visual (more ...)

It is interesting to note that GluR C-terminal peptide-expressing neurons showed net branch retraction after visual stimulation, whereas global application of AMPAR antagonists slowed dendritic arbor growth but did not result in net retraction of branches (2, 14, 15). The results are consistent with a competition-based mechanism in which convergent afferent activity strengthens synapses by a mechanism based on AMPAR trafficking into synapse, but low levels of AMPA transmission in GluR C-terminal peptide-expressing neurons prevent these neurons from responding to afferent activity and this in turn results in synaptic depression and branch retraction. A comparable competitive mechanism was recently found to regulate retinal axon arbor growth (17), suggesting that presynaptic and postsynaptic arbor developmental mechanisms are coordinately controlled by differential activity levels.

These data support the following model of dendritic arbor development in which newly added branches sample the local environment for potential presynaptic contacts. Initial contacts with retinal axon branches may be mediated by adhesive mechanisms (31, 32), which then form immature glutamatergic synapses, mediated by NMDAR. New branches receiving sufficient glutamatergic transmission mobilize intracellular signaling cascades that culminate in addition of AMPAR to developing synapses as well as local cytoskeleton stabilization and branch maintenance. These cellular events may involve calcium-dependent signaling pathways (33), CaMKII and the Rho GTPases, which have been shown to increase synaptic maturation and dendritic arbor stability (1, 2, 16, 34–38). This model is consistent with studies demonstrating that synapses are present on dendritic filopodia (3, 6, 39, 40), that morphogenesis and synaptogenesis share intracellular signaling cascades (1, 2, 16, 34–36, 41), and that synapse formation promotes filopodial stabilization (3). The “stabilized” new short branches then extend to form longer, persistent components of the dendritic arbor. Our findings also suggest that AMPAR transmission restricts the further growth of the major dendritic branch. Growth and stabilization of new branches coordinated with stabilization of the parent branches would restrict dendritic arborization to regions with convergent coactive innervation. The incremental increase in branch stability and synaptic strength gradually lead to significant changes in neuronal structure and circuit formation over a period of days to weeks.

Neuronal Transfection and Constructs. Individual optic tectal neurons from stage 46 to stage 47 Xenopus laevis tadpoles were transfected by using single-cell electroporation (22), in which a glass pipette (tip diameter, 0.5 μm) containing plasmid DNA (2 μg/μl in dH₂O) is inserted into the brain and brief electrical stimulation is delivered (0.5-s trains of 1-ms² waves, 1–2 μA, at 200 Hz, generated by an Axoporator 800A; Axon Instruments, Union City, CA). Bulk tectal neuron transfection involved injecting plasmid DNA into brain ventricles, followed by brief stimulation (five exponential decay pulses, 200 V/cm at peak, τ = 70 msec, 1 Hz, SD9 Stimulator; Grass Instruments, Quincy, MA) between external platinum plate electrodes (42). The C-terminal domains of the Xenopus or rat AMPAR subunits GluR1 (809–889) and GluR2 (813–862), termed GluR1Ct and GluR2Ct, respectively, were fused to EGFP and expressed from the Clontech (Mountain View, CA) pEGFP vector (18). The X. laevis and rat sequences share 80% homology and have conserved protein binding motifs (Fig. 7, which is published as supporting information on the PNAS web site). No significant differences were found between the effects of expression of the Xenopus and rat peptides on electrophysiological measures or dendrite growth, so data were pooled.

Electrophysiology. AMPAR mEPSCs were recorded from tectal neurons in an isolated whole-brain preparation with whole-cell patch-clamp recording at a holding potential (V_h) of −60 mV, with 1 μM tetrodotoxin (8). Recordings were made 2 and 3 days after bulk electroporation and analyzed as described (1). Fusion proteins of GluR1Ct or GluR2Ct with GFP allowed visual detection of transfected cells within 24 h after transfection. Control neurons for each transfected neuron were chosen as direct neighbors not expressing GFP. For kinetics measurements, nonoverlapping mEPSCs from each neuron were normalized to amplitude and averaged, rise-time was calculated at 35–90% of peak, and decay slopes were fit with single exponentials. Results between pairs of cells were tested for significance by using paired t tests.

In Vivo Imaging. Tadpoles were anesthetized with 0.02% MS222, and individual optic tectal neurons were imaged in vivo by using a custom designed two-photon microscope consisting of a modified Olympus Fluoview confocal scan box mounted on an Olympus BX50WI microscope with a Tsunami femtosecond-pulsed Ti:Sapphire laser. An Olympus LUMPlanFl/IR 60×, water-immersion objective (0.9 NA) was used, and stacks of images were taken by using a z-axis step size of 1.5 μm. One day after single-cell electroporation, tadpoles were screened for individually transfected immature optic tectal neurons at early stages of dendritic arbor elaboration (total arbor length <500 μm).

Visual Stimulation. Tadpoles were reared in ambient light with a 12-h light/dark cycle until stage 46 or 47. For controlled visual stimulation experiments, freely swimming tadpoles were exposed to 4 h of darkness followed by 4 h of visual stimulation (2, 43, 44).

Electron Microscopy. Three days after electroporation with GluR1 C-terminal peptides fused to GFP, or GFP alone, transfected and untransfected age-matched tadpoles were fixed in 3.5% paraformaldehyde and 0.2% glutaraldehyde for 1–2 h. Brains were cut into 50-μm sections and reacted with an antibody against GFP (overnight at 4°C in 1:500, rabbit polyclonal; Chemicon, Temecula, CA), visualized with HRP-tagged secondary antibody. Tadpoles from stages 39 and 47 were fixed in the same fixative for >3 days. Tissue was processed for EM as described (45). To avoid oversampling and variation caused by measurements on the boundary of synapses, we collected 70- to 90-nm serial sections and selected the section containing with the longest postsynaptic density and most synaptic vesicles for morphometric analysis. For statistical analysis, the raw data were transformed by using a Box Cox method rendering error variances near normal and homogeneous. Significance between groups was determined with ANOVA.

Neuronal Reconstruction and Data Analysis. Three-dimensional reconstructions of dendritic arbors were created by using Object Image software and were used to calculate the total dendritic branch length (TDBL) and branch tip number (BTN). Renderings were also used for three-dimensional Sholl analysis, in which multiple concentric spheres with incrementally increasing radii (1.5-μm step size) were placed around the soma. The number of dendritic branches that intersects each sphere was summed in 10-μm bins for plotting and statistical analysis using two-way ANOVA and Bonferroni posttests. All morphometric data except Sholl analysis were tested for significance by using unpaired Student's t tests.

We thank Robert Malinow (Cold Spring Harbor Laboratory, Cold Spring Harbor, NY) for the GluR C-terminal peptide constructs, Shu-Ling Chiu (Cold Spring Harbor Laboratory) for sequencing the GluR1 and GluR2 C-terminal domains from X. laevis, and the members of the Cline laboratory for helpful discussions. This research was supported by the National Eye Institute and the National Institutes of Health Director's Pioneer Award (to H.T.C.), the American Epilepsy Society, and the National Alliance for Research on Schizophrenia and Depression (K.H.).