Subcellular Topography of Visually Driven Dendritic Activity in the Vertebrate Visual System

SUMMARY

Neural pathways projecting from sensory organs to higher brain centers form topographic maps in which neighbor relationships are preserved from a sending to a receiving neural population. Sensory input can generate compartmentalized electrical and biochemical activity in the dendrites of a receiving neuron. Here, we show that in the developing retinotectal projection of young Xenopus tadpoles, visually driven Ca²⁺ signals are topographically organized at the subcellular, dendritic scale. Functional in vivo two-photon Ca²⁺ imaging revealed that the sensitivity of dendritic Ca²⁺ signals to stimulus location in visual space is correlated with their anatomical position within the dendritic tree of individual neurons. This topographic distribution was dependent on NMDAR activation, whereas global Ca²⁺ signals were mediated by Ca²⁺ influx through dendritic, voltage-dependent Ca²⁺ channels. These findings suggest a framework for plasticity models that invoke local dendritic Ca²⁺ signaling in the elaboration of neural connectivity and dendrite-specific information storage.

INTRODUCTION

The formation of a topographic map requires both molecular guidance cues and activity-dependent mechanisms that organize the map on different spatial scales (McLaughlin and O’Leary, 2005; Luo and Flanagan, 2007). Molecular gradients coarsely guide afferent axons to appropriate target areas within a cell population. The precise connectivity between axons and dendrites, however, is thought to emerge on a local scale, wherein the formation, stabilization, and elimination of synaptic contacts as well as axonal and dendritic branch dynamics are regulated by correlated activity in pre- and postsynaptic elements (Wong and Ghosh, 2002; Ruthazer and Cline, 2004). Support for a dendritic component of activity-dependent map refinement comes from anatomical data in vertebrate sensory systems, in which the pattern of dendritic growth is directed toward subregions of the afferent input map where presynaptic activity is thought to be correlated (Harris and Woolsey, 1981; Katz and Constantine-Paton, 1988; Sorensen and Rubel, 2006). Several pathways controlling local dendritic growth and structural maturation depend on local postsynaptic Ca²⁺ signals (Wong and Ghosh, 2002; Konur and Ghosh, 2005). This raises the possibility that Ca²⁺-dependent dendritic patterning and the refinement of receptive fields may be under the control of topographically organized afferent activity, which would require that sensory-driven Ca²⁺ signals are heterogeneously distributed and topographically biased across a developing dendritic tree.

In spite of the universal role of Ca²⁺ as a second messenger, little is known about the spatial distribution of Ca²⁺ signals in a dendritic tree during sensory stimulation. Optical recordings in higher brain regions have revealed that sensory stimulation triggers global and local dendritic Ca²⁺ signals (Svoboda et al., 1997; Helmchen et al., 1999; Charpak et al., 2001). It is not known, however, whether these Ca²⁺ signals follow topographic principles that specify the location of dendritic signals in relation to stimulus space. A possible topography of dendritic Ca²⁺ signaling may reflect the anatomical map of afferent inputs (McLaughlin and O’Leary, 2005; Luo and Flanagan, 2007). This has been found at early processing stages in the vertebrate visual system in which afferent inputs are dominated by the columnar organization of the retina (Euler et al., 2002). In higher stages of visual processing, however, afferent axons overlap extensively with recurrent and feedback connections, which may scramble dendritic input activity across the dendritic tree. Furthermore, visually evoked dendritic activity may be dominated by global dendritic spikes mediated by the excitable properties of the dendritic arbor, which could lead to a homogenous, cell-wide Ca²⁺ signal without spatial discrimination.

To distinguish between these possibilities, we measured visually driven Ca²⁺ elevations in individual dendritic trees in vivo in the retinotectal projection of Xenopus tadpoles. The optic tectum is the major retinorecipient brain center in lower vertebrates and offers favorable conditions for examining functional topography at the subcellular, dendritic scale for several reasons. First, a simple topographic order of retinal inputs is already established at early developmental stages. Specifically, ventrally and dorsally derived retinal ganglion cell (RGC) axons segregate in the tectal neuropil and target medial and lateral tectal neuropil, respectively (Figure 1A) (Holt and Harris, 1983; Sakaguchi and Murphey, 1985; Mann et al., 2002). Second, tectal neurons are accessible to electrophysiological recordings in vivo, which have been performed to study neural excitability (Aizenman et al., 2003), synaptic plasticity (Zhang et al., 2000), and activity-dependent modifications of receptive fields at the cellular level (Mu and Poo, 2006; Vislay-Meltzer et al., 2006). Third, the retinotectal system is a prominent model system for structural plasticity in visual system development, in which both presynaptic axonal and postsynaptic dendritic structure can be modified by visual activity and the influence of neurotrophic factors (Sin et al., 2002; Ruthazer and Cline, 2004; Cohen-Cory and Lom, 2004). Therefore, topographically organized differences in dendritic Ca²⁺ signals would be likely to impact the controlled growth of pre- and postsynaptic elements, synapse formation, and the maturation of visual circuitry.

Figure 1. In Vivo Imaging of Visually Evoked Ca²⁺ Signals in Proximal and Distal Tectal Cell Dendrites.

(A) Schematic of the retinotectal projection and visual stimulation. Flashing spots are presented in the visual field (left). Retinal ganglion cells (RGCs) project contralaterally to the tectal neuropil. Dorsal (D) and ventral (V) RGCs target lateral (L) and medial (M) tectal neuropil, respectively. The left half is a lateral view of the eye; the right half is a dorsal view of the left tectal lobe.

(B) Experimental set-up. A two-photon laser scanning microscope is used to measure dendritic Ca²⁺ signals, while the retina is stimulated with flashing spots by using a projector and an image conduit. Infrared laser light detected by a photodiode in trans-illumination mode generates contrast images used for patch-clamp and single-cell electroporation.

(C and D) Simultaneously acquired (C) negative-stain fluorescence image and (D) infrared contrast image of a patched tectal neuron in vivo.

(E) Z-projection of a filled tectal neuron, overlaid with the IR contrast image of the tectum. Note different granularity in the periventricular cell body layer (upper left) and in the tectal neuropil (lower right).

(F) ΔF/F transients recorded in proximal and distal dendritic regions and in the soma in response to 12 dimming spots (same as in [A]). The timing of stimuli is indicated by tick marks above traces. The location of dendritic regions is indicated by numbers in (E). Brackets and colors indicate groups of traces acquired in the same frame scan.

Here, we use in vivo two-photon microscopy to measure visually driven Ca²⁺ elevations in subcellular, dendritic compartments and compare their response tuning curves to the relative anatomical position within the afferent, retinotopic input map. Our findings show that, in a higher brain center in the vertebrate visual system, the topographic map defined by the afferent retinal network extends to the postsynaptic level at the subcellular, dendritic scale.

RESULTS

Visually Driven Dendritic Ca²⁺ Signals In Vivo

To examine the distribution of evoked dendritic activity, we imaged visually driven Ca²⁺ elevations in individual dendritic trees by using functional two-photon Ca²⁺ imaging, which uses infrared light to excite fluorescent Ca²⁺ indicators and therefore does not interfere with visual stimulation of the retina (Denk et al., 1990; Euler et al., 2002) (Figure 1B). Single tectal neurons within three cell-body diameters from the ventricular boundary of the tectum were filled with the Ca²⁺ indicator OGB-1 by using the single-cell electroporation technique or during very brief whole-cell recordings under visual control (Figures 1C and 1D). Three-dimensional stacks were acquired to determine the dendritic structure of individual neurons (Figure 1E), which exhibit a repertoire of local branch dynamics at this developmental stage (Sin et al., 2002). Neurons used for functional imaging had a total dendritic branch length of 628 ± 78 µm (mean ± SEM; range 317–1063 µm, n = 12 reconstructed neurons drawn from the entire data set). They extended multiple higher-order branches into the superficial tectal neuropil, indicating that they were in an advanced stage of continuous dendritic elaboration and received glutamatergic retinal inputs mediated by both NMDA-type and AMPA-type receptor channels (Wu et al., 1996).

Labeling individual neurons allowed us to examine the responsiveness of different dendritic compartments to patterned visual stimulation. When dimming spots were presented in different locations of the visual field (Figure 1A, left), we frequently observed stimulus-locked fluorescence transients (ΔF/F) in the primary dendrite (region-of-interest [ROI] 1) (Figures 1E and 1F), and in the soma (ROI 15). When higher-order branches were scanned, we found that dimming spot stimulation could trigger robust Ca²⁺ signals also in the more distal dendrites, which are embedded in the superficial tectal neuropil (ROIs 2–14, Figures 1E and 1F). In other neurons, dimming squares evoked more restricted Ca²⁺ signals in some dendrites, whereas other regions of the dendritic tree remained silent (Figure S1, available online). This indicates that visual stimulation can drive both global and local dendritic Ca²⁺ signaling at a developmental stage at which the retinotectal circuitry is formed and refined.

Subcellular Topography of Visually Driven Dendritic Activity

RGC axons terminate topographically in the Xenopus tectum even at early stages of afferent innervation, forming a coarse afferent input map (Holt and Harris, 1983; Sakaguchi and Murphey, 1985). Therefore, we asked (1) whether different higher-order branches within a single dendritic tree are tuned to different locations in the visual field and (2) whether the tuning and the anatomical position of these branches are related in a topographic manner (Figures 2A and 2B; Movie S1). Because dorsal and ventral RGC axons segregate in the tectum earlier during development than nasal and temporal RGC axons (O’Rourke and Fraser, 1986), we designed the visual stimulus specifically to test topography along the dorsoventral retinal axis, by using a sequence of horizontal bars flashed in five vertical positions of the visual field (Figure 2B, left). Horizontal bar stimulation evoked ΔF/F transients in higher-order dendrites with different relative amplitudes (Figure 2C). Response tuning curves were obtained from the normalized ΔF/F amplitudes (Figure 2D), and their center of mass was calculated for each dendritic region. Then, differences in the tuning curve centers (ΔR) were compared with the relative position of the dendritic regions within the dendritic tree. We noticed that tuning curve centers depended markedly on dendritic location along the dorsomedial-to-ventrolateral axis (u′) of the tectum (blue arrow in Figure 2A). When (ΔR, Δu′) pairs were pooled across experiments, ΔR and Δu′ were significantly correlated (r = −0.51; p = 4.4 × 10⁻⁵, n = 8 cells) (Figure 2E).

Figure 2. Visually Evoked Dendritic Ca²⁺ Signals Exhibit Topographic Bias.

(A) Three-dimensional reconstruction of a tectal neuron (red) filled with OGB-1. Same orientation as in Figure 1E (the x axis is medial-to-lateral, the y axis is caudal-to-rostral, and the z axis is dorsal-to-ventral of the animal). The blue arrow indicates the dorsomedial-to-ventrolateral axis of the tectum. Ca²⁺ signals were measured in dendrites distal to the first branch point (black rectangles). Regions of interest (ROIs) are located in different z-planes.

(B) Horizontal bar stimulation in five vertical positions to test dorsoventral topography in visual space.

(C) ΔF/F transients in dendritic regions (1–4 in [A]) in response to horizontal bar stimulation. The stimulus consisted of five bars shown for 0.5 s every 5 s in pseudorandomized order at the positions indicated above traces. ΔF/F transients (black traces) averaged from five individual sweeps (gray traces) for each ROI. Individual sweeps were peak scaled before averaging. Approximate tuning curves for the average ΔF/F transient of dendritic regions 1–4 are indicated by dashed lines. The timing of flashing bars is shown above traces.

(D) Normalized ΔF/F tuning curves of dendritic regions 1–4 calculated from average ΔF/F transients shown in (C) (mean ± SEM). Center-of-mass values (arrows) were calculated from these tuning curves by summing over stimulus positions, which were weighted by the amplitude of corresponding ΔF/F transients.

(E) Mean-subtracted center of mass of tuning curves in different dendritic regions versus their position, Δu′, along the dorsomedial-to-ventrolateral axis in the tectum (blue arrow in [A]). Data are pooled from eight cells. The solid line is a straight line fit.

Line Scan Analysis of Dorsoventral Dendritic Topography

Because of the three-dimensional structure of the dendritic tree, dendritic regions located in different z-planes had to be scanned in separate trials, introducing trial-by-trial variability in Ca²⁺ signals (e.g., Figure 2C, gray traces). To remove this source of variability and to measure individual ΔF/F transients at high time resolution, we performed line scans across multiple distal branches of a single dendritic arbor (Figures 3A and 3B). This allowed us to compare ΔF/F transients in dendrites located in the same z-plane and acquired simultaneously within the same stimulus trial (Figure 3C). Horizontal bar stimulation evoked ΔF/F transients in distal dendrites with different relative amplitudes (Figure 3D). We determined response tuning curves from the normalized ΔF/F amplitudes (Figure 3D) and calculated their center of mass for each dendritic region and individual trial. To compare differences in the tuning curve centers (ΔR) with position of the dendritic region within a trial, we measured the coordinate (Δu) of each region along the medial-to-lateral axis of the tectum (arrow in Figure 3B). Then, ΔR and Δu values from multiple line scans in the dendritic tree were pooled, and the correlation coefficient was determined for the individual neuron (Figure 3E). ΔR and Δu were negatively correlated in 14 out of 15 neurons (Figure 3F), with correlation coefficients ranging between 0.003 and −0.83 (also see Figure S2 for data on a cell-by-cell basis). This suggests that Ca²⁺ signals in medial dendrites are more sensitive to dorsal visual field stimulation than lateral dendrites in the same neuron. When the (ΔR,Δu) pairs were pooled over all 143 line scans in 15 neurons, the correlation coefficient between ΔR and Δu was r = −0.305 (p < 10⁻¹¹). The topographic bias did not correlate with the distance of the scanned dendritic regions from the first branch point toward the distal branch tips (Figure S3). These findings show that postsynaptic dendritic arbors at this stage are sufficiently compartmentalized to support differential Ca²⁺ signals in medial and lateral dendritic branches, whose tuning reflects the topographic input map in the developing optic tectum.

Figure 3. Line Scan Analysis of Dorsoventral Topographic Bias of Dendritic Ca²⁺ Signals.

(A) Z-projection of a filled tectal neuron.

(B) Reconstruction of boxed dendritic region in (A) with scan line (blue). Circles mark intersections of scan line with dendrites. Blue dashed lines indicate coordinates of dendritic locations (1–4) projected onto the medial-to-lateral tectal direction (u, arrow).

(C) Individual ΔF/F transients acquired simultaneously in dendritic regions (1–4) in response to one sequence of horizontal bar stimulation. The sequence consisted of five bars shown for 0.5 s every 5 s in pseudorandomized order at the positions indicated above traces.

(D) Normalized ΔF/F tuning curves from dendritic regions 1–4 measured in a single trial (same as [C]). Center-of-mass values (arrows) were calculated from these tuning curves by summing over stimulus positions, which were weighted by the amplitude of corresponding ΔF/F transients.

(E) Mean-subtracted center of mass (ΔR) of tuning curves in different dendritic regions versus their Δu coordinate measured in a single cell. Data are pooled from 12 line scans in one cell (same cell as in [A]–[D]). The solid line is a straight line fit (r = −0.48, p = 0.0005).

(F) Correlation coefficients determined as in (E) from line scan measurements in 15 cells. The mean correlation coefficient is r = −0.37 ± 0.05 (mean ± SEM).

Subcellular Dendritic Topography Depends on NMDAR Activation

Excitatory synaptic transmission in tectal neurons at this stage is mediated by both NMDAR and AMPAR channels (Wuet al., 1996), which are probably clustered in postsynaptic puncta scattered across the dendritic tree (Sanchez et al., 2006). Therefore, visually driven Ca²⁺ signals in distal dendrites could be mediated by Ca²⁺ influx directly through NMDAR channels or through depolarization-induced opening of voltage-dependent Ca²⁺ channels (VDCCs).We examined whether mechanisms other than NMDAR activation are sufficient to drive topographically biased dendritic Ca²⁺ signaling by pharmacologically blocking NMDARs. Line scans through distal dendritic compartments (similar to those shown in Figures 3B and 3C) were performed during bath (n = 4) or local (n = 4, see also Figure S4) application of APV to examine the dependence of tuning curves on dendritic position. Visually evoked dendritic Ca²⁺ signals persisted when NMDARs were blocked (Figure 4A), indicating that AMPAR-mediated excitation is sufficient to drive dendritic Ca²⁺ signaling. However, the average ΔF/F amplitude was reduced by 14% (APV: ΔF/F = 34.4% ± 1.2% [mean ± SEM; n = 1230 Ca²⁺ transients] versus control: ΔF/F = 39.9% ± 0.7% [n = 2425 Ca²⁺ transients]; p = 8.9 × 10⁻⁵, two-tailed t test), and the topographic bias of tuning curve centers was abolished (Figure 4B). The pooled data showed no significant correlation between ΔR and Δu (r = 0.10, p = 0.106, pooled from 69 scans in 8 neurons, Figure S2). Furthermore, when comparing the correlation coefficients for all cells in the control group (same as Figure 3F) with those in the ”APV” group, we found that their medians were significantly different (Figure 4C). This suggests that activation of NMDAR channels can confer topographic input bias onto dendritic Ca²⁺ signals.

Figure 4. NMDAR Activation Is Required for Topographic Bias of Dendritic Ca²⁺ Signals.

(A) Individual ΔF/F transients acquired simultaneously in three dendritic regions (1–3) in response to horizontal bar stimulation during inhibition of NMDAR by APV (100 µM).

(B) Mean-subtracted center of mass of dendritic tuning curves is not correlated with dendritic position Δu in the presence of APV (r = 0.04, p = 0.81). Data are from ten line scans in a single dendritic tree (same as in [A]).

(C) Histogramand cumulative distribution of correlation coefficients for control cells (blue, n = 15) and cells when NMDARs were blocked by APV (red, n = 8). Medians of the distributions are significantly different (−0.406 [control] and +0.047 [APV], p = 0.0004, Wilcoxon’s rank sum test)

Contribution of VDCCs to Dendritic Ca²⁺ Signaling

Tectal neurons around this stage can respond to dimming stimuli with short spike bursts (Zhang et al., 2000), which may be a trigger for VDCC-mediated dendritic Ca²⁺ signals. To examine the relationship between tectal spiking activity and dendritic Ca²⁺ signaling, we combined dendritic imaging in the primary dendrite with somatic spike recordings during visual stimulation (Figure 5A). ΔF/F transients coincided with tectal spike bursts, and the ΔF/F amplitude covaried with the number of spikes per burst (Figure 5A). The spike-burst-associated ΔF/F amplitude was normalized to the averaged ΔF/F amplitude associated with a single spike for each neuron (30.5% ± 4.5%, n = 6) and compared with the number of spikes per burst (Figure 5B). Proximal dendritic ΔF/F signals were approximately proportional to the number of spikes in a short burst for up to seven spikes. This suggests that the proximal dendritic Ca²⁺ signal encodes the number of spikes per burst during sensory stimulation, similar to pyramidal neuron apical dendrites (Svoboda et al., 1997). However, the density of VDCCs in distal dendrites may be low at early developmental stages (Tao et al., 2001), and attenuation and failure of action potential propagation in the dendritic tree may suppress Ca²⁺ signals in some branches (Hausser et al., 2000). Therefore, we performed whole-cell patch-clamp recordings in tectal neurons while all excitatory and inhibitory synaptic inputs were blocked pharmacologically (Figures 5C–5E). When action potentials were triggered by postsynaptic current injection, a series of increasing ΔF/F signals was measured in proximal and distal dendrites (Figure 5D). ΔF/F transients were largest in the primary dendrite and the first branch region, but they were not substantially reduced in distal dendrites (Figure 5E), which supports the notion that distal dendritic VDCCs contribute to the NMDAR-independent, nontopographic component of distal dendritic Ca²⁺ signals, while an NMDAR-mediated synaptic component biases the summed dendritic Ca²⁺ signal in a topographic way.

Figure 5. Voltage-Dependent Ca²⁺ Channels Mediate a Global Dendritic Ca²⁺ Signal during Tectal Cell Spiking.

(A) Simultaneous recording of proximal dendritic ΔF/Ftransients and somatic cell spiking (I_{cell-attached}) in the cell-attached configuration. Lower traces: two examples of tectal spike bursts triggered by a dimming flash.

(B) Relationship between ΔF/F transients in primary dendrite and somatic spike number (n = 6 cells). For each cell, the ΔF/F amplitude was normalized to the ΔF/F amplitude during a single spike (gray lines). Dots, individual ΔF/F amplitudes; diamonds, average ± SEM. The dashed curve is an exponential fit to average data.

(C) Z-projection of a reconstructed tectal neuron. Dendritic regions in which ΔF/F transients were measured are indicated by rectangles 1–6.

(D) Upper trace: action potentials evoked by somatic current injection (1, 2, 4, 8, and 16 action potentials). Lower traces 1–6: ΔF/F transients measured during somatic current injection in the rectangular regions shown in (C). Synaptic transmission was blocked by APV, CNQX, SR95531, and strychnine in the bath solution.

(E) ΔF/F transient amplitudes measured in proximal and distal dendritic regions, plotted versus their intracellular distance from soma (n = 5 cells). Individual ΔF/F amplitudes from the same cell are indicated by identical symbols. ΔF/F amplitudes were normalized to the maximal amplitude measured within that cell, which could occur either distally or proximally. After normalization, data points from all cells were binned and averaged (diamonds, average ± SEM).

Dorsoventral Topographic Mapping of Presynaptic Axons

We sought to examine whether the direction of significant topographic bias in dendritic Ca²⁺ signals (Figures 2 and 3) is consistent with the anatomical map of ventrally and dorsally derived RGC axons reported previously (Holt and Harris, 1983; Sakaguchi and Murphey, 1985). Therefore, we traced the retinotectal projection by injecting spectrally different, dextran-conjugated dyes into dorsal and ventral retina, respectively, and imaged the tectal target areas under our experimental conditions (Figure 6A). Ventral RGC axons exhibited a strong preference for medial tectal neuropil (81% ± 3% fractional intensity, n = 4), whereas dorsal RGC axons showed a somewhat weaker preference for lateral neuropil (59% ± 7% fractional intensity, n = 4, Figure 6B). Furthermore, we found that the different degrees of mediolateral segregation of dorsal RGC and ventral RGC axons in the tectum was reflected in the relative amplitudes of ΔF/F signals in medial and lateral dendrites, respectively (Figure S5). Thus, the dendritic topographic bias in Ca²⁺ signaling is consistent with the relative distribution of ventrally and dorsally derived RGC axons in the tectum.

Figure 6. Dorsoventral Topographic Mapping of Presynaptic Axons.

(A) Overlay of an IR image of a tectal lobe with fluorescence images of RGC axonal termination zones in the same lobe. Dashed curves demarcate the medial and lateral half of tectal neuropil (lower left and upper right, respectively). A pipette (upper left) was used to stabilize tectal preparation during the experiment. Inset: schematic of retinal injection sites.

(B) Fractional intensity in tectal neuropil of axons from dorsal retina (red) and ventral retina (green). Individual experiments (circles) and average (thick line, error bars: SEM, n = 4 for each injection site)

Dorsoventral Axis of Visual Space Is Functionally Mapped onto the Tectal Cell Population

Finally, we tested whether the confirmed anatomical topography of RGC arbors is preserved at the level of postsynaptic population activity. This is not known a priori because functional topography may be confounded within the postsynaptic cell population since (1) RGC axons overlap considerably in the tectal neuropil (Holt and Harris, 1983; Sakaguchi and Murphey, 1985) and (2) polysynaptic pathways within the intratectal circuitry may diminish topographic order in the postsynaptic tectal cell population (Pratt et al., 2008). To test postsynaptic topography, tectal cell position was measured as the relative distance from the medial pole along the periventricular cell body layer (PVL)-ventricle boundary (red curve in Figure 7A) toward the lateral end. Using a sequence of 12 dimming squares (Figure 7B), ΔF/F transients were measured in the proximal dendrite, and, from their average response (Figures 7C and 7D), the receptive field center was estimated (white cross in Figure 7C). Both the horizontal and the vertical coordinate of the receptive field center were compared with tectal cell position along the medial-to-lateral dimension (Figures 7E and 7F). Whereas the horizontal coordinate did not correlate with tectal cell position, the vertical coordinate exhibited significant correlation with the position of the neuron. The slope of the regression analysis was about three-fold larger than the average slope of regression lines measuring the topography within individual dendritic trees (Figures S2 and S6). Thus, the map of dorsoventral visual space onto the medial-to-lateral dimension of the tectum is preserved anatomically as well as functionally at the population scale (Niell and Smith, 2005) and the subcellular, dendritic scale.

Figure 7. Dorsoventral Topographic Mapping of Postsynaptic Population Activity.

(A) Single dye-filled neuron (yellow), overlaid with a simultaneously acquired IR image of the entire tectal lobe. Cell position is measured from the medial pole as the fractional distance (blue curve, arrow) relative to total PVL length (red curve). The image plane is ~100 µm from the dorsal edge of the tectum. (M, C, L, R represent approximate medial, caudal, lateral, and rostral, respectively, in the tectum.)

(B) Schematic diagram of dimming square positions, flashed for 0.5 s every 5 s. The sequence of spots is indicated by numbers (1–12).

(C) Color-coded ΔF/F amplitudes (from average trace in [D]) mapped onto stimulus position. Center of mass (cross) of the receptive field was calculated as the weighted vector average of the four stimulus positions that evoked the largest responses in the primary dendrite.

(D) ΔF/F transients measured in the primary dendrite of the neuron in (A) in response to dimming spots, as indicated in (B). Six individual traces (gray) and the average trace (black) are shown.

(E and F) Coordinates of receptive field centers versus tectal cell position (0 is the most medial, 1 is the most lateral along the red line in [A]). The horizontal coordinate is not correlated with cell position, as indicated by a linear fit (E); correlation coefficient, r = −0.04; p = 0.83). (F) The vertical RF position is correlated with cell position along the PVL boundary. The solid line is a linear fit (r = −0.60; p = 0.0004). Data are from 31 cells.

DISCUSSION

In summary, our results show that the anatomical, dorsoventral map of RGC axons onto the tectal neuropil extends functionally to the population level of postsynaptic neurons (~200 µm) and, remarkably, to the subcellular scale within single dendritic trees (~50 µm). Whereas non-NMDAR-mediated synaptic input was sufficient to trigger global Ca²⁺ signals in tectal dendritic trees, NMDAR activation led to larger ΔF/F transients in distal dendrites and was necessary for topographically biased differences in dendritic Ca²⁺ signals. By contrast, voltage-dependent Ca²⁺ influx could be triggered throughout the dendritic tree by somatic current injection in the absence of synaptic transmission, suggesting that dendritic depolarization mediates a global dendritic Ca²⁺ signal through VDCCs.

Mechanism of Topographically Organized Dendritic Ca²⁺ Signaling

Potential pathways for Ca²⁺ entry into dendritic compartments during visually driven activity are through VDCCs and glutamate receptor channels. Our experiments with somatic current injection indicate that functional VDCCs are inserted in both proximal and distal dendrites in developing neurons at this stage. On the other hand, tectal neurons of comparable size (measured as total dendritic length) contain functional AMPAR- and NMDAR-type glutamate channels (Wu et al., 1996), which are probably concentrated in postsynaptic densities. Expression of a GFP-tagged version of PSD-95 in these neurons shows that postsynaptic densities are distributed across the dendritic tree and form anatomical synapses with presynaptic elements (Sanchez et al., 2006). Furthermore, the decay time course of NMDAR-mediated currents in stage-47 tectal neurons is sensitive to block by ifenprodil, suggesting that NMDARs contain both NR2A and NR2B subunits (Ewald et al., 2008). Together, this suggests that Ca²⁺ signals are probably mediated by a mixture of VDCCs and glutamate receptor channels in the majority of dendritic compartments during visual activity. In contrast, release of Ca²⁺ from internal stores is less likely to contribute (Tao et al., 2001).

The following mechanistic model could explain the dependence of topographic bias in dendritic Ca²⁺ signals on NMDAR activation and the more homogeneous distribution of dendritic Ca²⁺ signals when NMDARs were blocked. First, concerted activation of neighboring RGCs leads to glutamate release from topographically distributed RGC presynaptic terminals clustered in a corresponding region of the postsynaptic dendritic tree.

Second, rapid AMPAR-mediated transmission is summed at a central integration site, possibly near the first branch point of the tree, and triggers Na⁺ and Ca²⁺ action potentials that travel throughout the tree and evoke global, uniform Ca²⁺ transients. The observation that topographic bias in dendritic Ca²⁺ signals was abolished during application of APV suggests that (1) the direct entry of Ca²⁺ influx through Ca²⁺-permeable AMPAR channels is too small to contribute to measurable differences in Ca²⁺ signals. This is consistent with the observation that fractional Ca²⁺ currents (P_f) mediated by recombinant NMDAR channels are larger (8%–11%) than those mediated by recombinant AMPAR channels, even if the latter do not contain the GluR-2 subunit (P_f < 4%) (Burnashev et al., 1995), which makes them Ca²⁺ permeable. Furthermore, dendritic depolarization mediated by AMPARs alone appears to be insufficient to trigger significant voltage-dependent Ca²⁺ influx that remains restricted to subregions of the dendritic tree.

Third, the topographic activation of NMDARs within the dendritic tree may give rise to an additional local component of Ca²⁺ signals that mirrors the topographic input bias. This local component could be mediated directly through NMDAR channels owing to their large Ca²⁺ permeability (Schneggenburger et al., 1993; Bollmann et al., 1998). Also, long-lasting NMDAR-mediated postsynaptic potentials may be more efficient than those mediated by AMPARs in generating local dendritic spikes that may boost Ca²⁺ influx through VDCCs on a local scale (Waters et al., 2003; Losonczy and Magee, 2006). In addition, strong activation of nearby NMDARs can trigger NMDA spikes restricted to individual dendritic branches in neocortical pyramidal neurons (Schiller et al., 2000), which is also a possibility in developing tectal dendrites. In summary, a combination of these mechanisms is likely to contribute to the topographic component in dendritic Ca²⁺ signaling observed in the Xenopus tectum during visual stimulation.

Function of Dendritic Topography

The present study shows that the dendritic tree of relatively complex tectal neurons supports global and local Ca²⁺ signals in response to visual subfield stimulation. Dendritic Ca²⁺ signals serve many functions in dendritic structural and synaptic plasticity, and likely serve a role in dendritic computation (Segev and London, 2000; Hausser and Mel, 2003). First, dendritic Ca²⁺ signals may result from the combined action of local inputs in dendritic subregions and serve as a trigger for nearby dendritic transmitter release (Rall et al., 1966; Isaacson and Strowbridge, 1998; Euler et al., 2002). Second, postsynaptic Ca²⁺ signals are an integral part of several pathways that control dendritic branch dynamics and filopodial growth, including NMDAR activation, AMPAR insertion, BDNF signaling, CaMKII activity, Rho-GTPases, and MAPK cascades (Sin et al., 2002; Wong and Ghosh, 2002; Konur and Ghosh, 2005; Lohmann et al., 2005). Hence, the complex structure of a mature dendritic tree is probably the result of directed growth mechanisms that are partly determined by the distribution of Ca²⁺ signals in the developing tree. Third, postsynaptic dendritic Ca²⁺ signals are necessary for induction of long-term potentiation and long-term depression (Zucker, 1999), and their spatial extent may be a restricting factor for the spread of these forms of synaptic plasticity (Tao et al., 2001). As a corollary, the mapping of synchronously active synaptic inputs onto the same dendritic neighborhood of a neuron is expected to be more efficient in driving (dendritic) spike-timing-dependent plasticity mechanisms (Kampa et al., 2007; Sjostrom et al., 2008) than when the same inputs are distributed randomly across a complex dendritic tree (Poirazi and Mel, 2001; Mehta, 2004; Govindarajan et al., 2006; Larkum and Nevian, 2008). For example, if clustered within one dendritic region, coactive inputs may cooperatively drive long-lasting changes in synaptic strength and dendritic excitability (Frick et al., 2004; Harvey and Svoboda, 2007; Losonczy et al., 2008). These observations of nonlinear dendritic signal integration have inspired several plasticity models that invoke local dendritic Ca²⁺ signaling in the elaboration of neural connectivity and in dendrite-specific information storage (Poirazi and Mel, 2001; Mehta, 2004; Konur and Ghosh, 2005; Govindarajan et al., 2006).

A topographic organization of dendritic Ca²⁺ signals in vivo has been observed previously, e.g., in large motion-sensitive neurons in the fly visual system (Borst and Egelhaaf, 1992), in cricket auditory neurons (Baden and Hedwig, 2006), and in non-spiking starburst amacrine cells in the rabbit retina (Euler et al., 2002). The preserved input topography in these cells, together with the passive and active properties of the dendritic tree (Hausselt et al., 2007), have been implicated in computational tasks performed by the neuron, such as motion detection and dendrite-specific direction selectivity. We hypothesize that, in the developing optic tectum, the topographic distribution of dendritic Ca²⁺ signals most likely serves a role in the structural elaboration of pre- and postsynaptic elements, which leads to the refinement of the retinotopic map. The local, topographic component of the dendritic Ca²⁺ signal probably reflects correlated activity in axon terminals from neighboring RGCs, forming synapses with the same dendritic region, and activating it cooperatively. Coactive axon branches are thought to stabilize their connections when they form synapses on a common target (Ruthazer and Cline, 2004), which involves retrograde signaling and contributes to activity-dependent map refinement (Schmidt, 2004). The local, NMDA-dependent component of dendritic Ca²⁺ signals may be well positioned to trigger a stabilizing retrograde growth signal to those presynaptic elements that contributed to the local dendritic event. Alternatively, topographically distributed Ca²⁺ signals could trigger a negative retrograde signal that destabilizes connections that did not contribute to evoking the dendritic Ca²⁺ elevation. Thus, connections with topographically inappropriate RGC axons would be destabilized. Both mechanisms would lead to a more specific segregation of afferent input channels onto different dendritic branches, while the retinotopic map is refined by correlated visual input activity. Thus, our finding that topography is preserved and encoded in local dendritic Ca²⁺ signals may help explain asymmetrical and topographical rearrangements of dendritic trees in the frog optic tectum (Katz and Constantine-Paton, 1988), chick auditory system (Sorensen and Rubel, 2006), and in other, cortical projections (e.g., barrel cortex [Harris and Woolsey, 1981]) under normal and activity-deprived conditions.

EXPERIMENTAL PROCEDURES

Preparation

Albino Xenopus tadpoles (stage 46–48) (Nieuwkoop and Faber, 1967) were anaesthetized (0.02%, MS222) and mounted dorsal side up in a Sylgard dish containing external solution (in mM: 115 NaCl, 2 KCl, 2.5 CaCl₂, 1 MgCl₂, 10 HEPES, 10 glucose, 0.01 glycine [pH 7.3], osmolality 260 mosmo/kg). d-tubocurarine (0.1 mM) or α-bungarotoxin (2 µg/ml) was added to prevent occasional twitching of muscle fibers. The optic tectum was cut along the dorsal midline, and one tectal lobe was dissected out to obtain access to the periventricular aspect of the remaining lobe for electroporation and patch-clamp recordings (Zhang et al., 1998). In some cases, a glass micropipette was micropositioned rostrally in the tectal lobe to minimize vertical drift of the preparation during recordings. All experiments were approved by Harvard University’s Standing Committee on the Use of Animals in Research and Training.

Single-Cell Dye Fills

Individual tectal neurons were filled with Oregon-Green-Bapta-1 (OGB-1, Invitrogen, USA) by using the single-cell electroporation technique (Haas et al., 2001; Nevian and Helmchen, 2007; Kitamura et al., 2008). Glass micropipettes (open tip diameter ca 1–2 µm) were filled with an internal solution containing (in mM): 110 K-gluconate, 10 KCl, 5 NaCl, 1.5 MgCl₂, 20 HEPES, 2 Mg-ATP, 0.3 Na-GTP, and OGB-1 (2–4 mM) ([pH 7.3], osmolality 255 mosmo/kg). Under infrared-visual control (see below), the pipette was brought into contact with a tectal soma. A 200 Hz train of voltage pulses (amplitude 2–10 V, pulse width of 4 ms) was applied for 250 ms to electroporate the somatic membrane. Alternatively, brief whole-cell patch-clamp recordings were used to fill individual neurons with OGB-1 and Alexa 594 (Invitrogen, USA). Patch pipettes were filled with an internal solution as described above, but with different dye concentrations (OGB-1: 0.625 mM; Alexa 594: 1.25 mM), and were used to dialyze the indicators into the tectal cell. The pipette was retracted ~30 s after rupturing the tight-sealed membrane patch, and an outside-out membrane patch was formed, suggesting that the somatic membrane resealed as well. With both techniques, dye was allowed to diffuse throughout the dendritic tree for ~1 hr before dendritic imaging was started. In one set of experiments, NMDAR channels were blocked by applying APV (0.1–0.2 mM, Tocris), either in the bath or locally through a micropipette positioned in the tectal neuropil (Figure S4). In a different set of experiments after electroporation with OGB-1, somatic loose-patch recordings were performed to record visually driven action potential firing of electroporated cells, which was used to calibrate ΔF/F signals in the proximal dendritic tree to the number of action potentials per burst (Figure 5). After recording, the membrane patch was ruptured and the neuron was filled with Alexa 594 from the recording pipette to verify that the dendritic Ca²⁺ recording and the somatic spike recording were from the same neuron.

Whole-Cell Patch-Clamp Recordings

Recordings were performed by using micropipettes pulled from borosilicate glass capillaries (Kimax) with an open tip resistance of 6–10 MΩ and an Axo-patch 200B amplifier (Molecular Devices). Pipettes were filled with the same internal solution as described above, but with reduced dye concentrations (OGB-1: 0.1–0.2 mM; Alexa 594: 0.2–0.4 mM). When measuring dendritic Ca²⁺ signals in response to somatic current injection (Figure 5), d-APV (0.1–0.2 mM), CNQX (0.05 mM), SR95531 (0.01 mM, all from Tocris), and strychnine (0.06 mM, Sigma) were added to the external solution to block glutamatergic, GABAergic, and glycinergic transmission, respectively. The cell was held in current clamp at −60 to −70 mV, and action potential trains were evoked by repetitive current injection into the soma (4 ms duration, 50 Hz).

Functional Two-Photon Ca²⁺ Imaging In Vivo

A custom-built multiphoton confocal microscope was used to record dendritic Ca²⁺ signals. A Ti:Sapphire laser (MaiTai, Newport Corp) tuned to 940–950 nm and focused through a water-immersion objective (20×, NA 0.95) (Olympus, Japan) was used to image the morphology and record dendritic Ca²⁺ concentration changes in tectal neurons (Denk et al., 1990; Euler et al., 2002; Zelles et al., 2006). The detection pathway consisted of a dichroic mirror (690 dcxxr; Chroma) and a bandpass (e700sp-2p; Chroma) to separate fluorescence emission from the infrared excitation. Fluorescence emission was further split into two channels by a dichroic beam splitter (585 dcxr; Chroma) and a green (HQ522/40 m; Chroma) and a red (628/40; Semrock, USA) bandpass, respectively, and recorded by two photomultiplier tubes (R3896, Hamamatsu, Japan). Furthermore, infrared laser light was recorded with a fast photodiode (DET100A, Thorlabs, USA) in trans-illumination mode, which can be used to generate an infrared contrast image of the tectum (e.g., Figures 6A and 7A) to distinguish between the periventricular cell body layer and the neuropil. Similarly, positioning of micropipettes during electroporation and patch-clamp recordings of tectal cells were aided by simultaneously recording (resolution 256 × 256, 2.5 Hz) the infrared contrast image and the negative stain image, which is transiently created by a fluorescent indicator ejected into the extracellular space while the micropipette approaches the cell body (Figures 1C and 1D). During experiments, the three-dimensional dendritic structure of one or a few dye-filled neurons was first visualized in image stacks at a resolution of 512 × 512 pixels. Fast area scans (typically 64 × 64 pixels at 14 Hz frame rate) or line scans (256 pixels, at 864 Hz) were subsequently performed at selected dendritic regions in the primary dendritic branch and at more distal dendritic elements in the tectal neuropil during visual stimulation.

Visual Stimulation

Two-dimensional visual stimulus patterns were displayed on the exit face of a rigid image conduit (3 mm diameter, Edmund Optics), micropositioned to a fixed distance (1.5 mm) in front of the eye. Orientation of the eye and alignment with the visual stimulator were carefully adjusted under visual control through the eye piece of the microscope. Stimulus patterns were generated in LabView (National Instruments, USA), and projected with a DLP projector (Optoma, Taiwan) and an air objective lens (10×, NA 0.3) onto a flexible plastic fiberoptic bundle (Nanoptics, USA), which was connected to the image conduit. Stimulus light was in the yellow range, filtered with a bandpass (575/10 nm; Omega Optical, USA). Dimming squares (subtending ~10° visual angle) or bars (10° × 70°) were flashed on a bright background. Topographic bias in the dorsoventral axis of the visual field was tested by using five horizontal bars in different vertical positions (−30°, −15°, 0°, 15°, 30°, Figures 2–4). Dimming bars were shown in a pseudorandomized order, with a dimming bar presented for 0.5 s every 5 s. Contrast was set in the range between 30% and 90%, typically 88% for dimming bars. The angular distribution of light intensity emitted from the image conduit was measured and used to implement a radially symmetric increase in background light and stimulus intensities toward the periphery of the image conduit in order to keep the effective stimulus contrast ratio and background illumination constant across the visual field.

Data Acquisition and Analysis

Two-photon Ca²⁺ imaging data and electrophysiological recordings were acquired and analyzed offline by using custom-written software in LabView (National Instruments, USA), Igor Pro (Wavemetrics, USA), and Matlab (Math-works, USA). Functionally imaged neurons used to analyze dendritic topography (Figures 2–4) exhibited a robust Ca²⁺ signal in the primary dendrite and most parts of the distal dendritic tree in response to visual subfield stimulation. Visually evoked ΔF/F transients were measured as peak amplitudes averaged in a 150 ms window surrounding the maximum fluorescence value after the stimulus and subtracted with the baseline value averaged in a 1 s interval preceding the stimulus. Subcellular dendritic topography was determined by comparing the mean-subtracted center of mass of tuning curves (ΔR) with the relative position of the dendritic region (Δu′ in Figure 2; Δu in Figures 3 and 4) within the dendritic tree of a neuron. The center of mass of a tuning curve was calculated by taking the sum of vertical stimulus positions weighted by the ΔF/F amplitude measured in response to that stimulus. This was done either from the ΔF/F responses averaged across multiple trials in the same dendritic region (Figure 2D) or from the individual ΔF/F responses measured within a line scan (Figures 3 and 4). Specifically in frame scan mode, where dendritic regions in different z-planes were sequentially scanned, Δu′ was determined by projecting the ROI coordinates of scanned regions against the u′ axis (blue arrow in Figure 2A). The u′ coordinate of the midpoint between the maximum and minimum u′ coordinate of all scanned dendritic regions within a neuron was taken as the origin. Specifically, in line scan mode (Figures 3 and 4), where different dendritic regions in the same z-plane were scanned simultaneously, the coordinate of each dendritic region was projected against the medial-to-lateral tectal axis (u; arrow in Figure 3B) and measured relative to the midpoint between the most medial and lateral dendritic region in each scan (”0” in Figure 3B). Three-dimensional image stacks were filtered with an anisotropic diffusion filter (Broser et al., 2004) for display purposes. For reconstruction and measurement of dendritic length, the software tool ”Neuromantic” by D. Myatt, University of Reading was used.

Retinal Dye Injections

RGC axonal arbors were labeled by injection of Alexa Fluor 488-Dextran (10,000 MW) and Alexa Fluor 594-Dextran (10,000 MW) into the ventral and dorsal retina of stage-46/47 tadpoles. The tip of a glass micropipette was broken to yield a 10–20 µm tip opening, and the pipette was filled with a 50% (w/v) solution of the indicator dissolved in Xenopus external solution. A total of 18–24 hr after dye injection, the contralateral tectal lobe was screened for labeled axons in a fluorescence dissecting microscope. Tadpoles with labeled retinotectal projections were dissected as in the physiological Ca²⁺ imaging experiments such that the location and orientation of the tectal neuropil could be compared between functional and anatomical imaging data. To quantify differences in the tectal location of dorsally and ventrally derived RGC axons, the extent of the tectal neuropil was outlined and divided into a medial and a lateral half (Figure 6A), and the fractional intensity of the green and red channel was measured for the two hemifields.