2.1.

Experimental Procedures

The effects of direct stimulation with the excitatory neurotransmitter L-glutamate on layer V pyramidal cells were investigated experimentally combining whole-cell patch-clamp recording, IR video microscopy, and UV-flash-induced release of the excitatory neurotransmitter glutamate from an inert “caged” form.^{6 10}

Coronal slices from rat somatosensory cortex containing the barrel cortex (S1BF, Ref. 14) were prepared from male Wistar rats (postnatal days 18 to 22). Animals were anesthetized, and the brain was quickly removed from the skull. From blocks of tissue containing the relevant region, slices of 300-μm nominal thickness were cut on a vibratome and incised along the midline to separate the hemispheres. The sectioning was performed in ice-cold artificial cerebrospinal fluid (ACSF) oxygenated with carbogen (95% O ₂/5% CO ₂). Normal ACSF consisted of (in mM): 124 NaCl, 1.25 NaH ₂ PO ₄, 26 NaHCO ₃, 1.6 CaCl ₂, 1.8 MgCl ₂, 3 KCl, 10 glucose, at pH 7.4 (all chemicals purchased from Merck, Darmstadt, Germany).

Subsequently, the slices were equilibrated in a storage container for at least 1 h in ACSF at 32°C. Individual slices were transferred to the recording chamber and submerged in ACSF at a flow rate of about 1 ml/min at 32°C. To block synaptic transmission modified ACSF containing (in mM) 0.2 CaCl ₂, 4 MgSO ₄ (low Ca ²⁺ /high Mg ²⁺ ACSF) was used for at least 30 min before the start of the experiment. During the application of caged glutamate, a total amount of approximately 5 ml ACSF containing the caged compound was continuously oxygenated and recirculated. The caged glutamate (L-glutamic acid, γ-[α-carboxy-2-nitrobenzyl] ester; Molecular Probes, Eugene, Oregon) has a broad absorption spectrum of 280 to 380 nm with a maximum around 360 nm. It was dissolved in ACSF and added to the circulating ACSF, resulting in a 1 mM concentration. Photolysis of the caged compound releases free glutamate with a quantum yield^{15 16} of 0.14 (see Fig. 1).

Figure 1

Molecular structure of L-glutamic acid, γ-[α-carboxy-2-nitrobenzyl]-ester and UV-induced hydrolysis of the ester bond. The carboxy-nitrobenzyl caging group shown in the hatched box is sensitive to UV light at ∼360 nm. The photoreaction releases active glutamate (modified from Ref. 6).

Slices were placed in a fixed-stage submerged chamber under an upright microscope (Axioskop FS, Carl Zeiss, Go¨ttingen, Germany) fitted with a 2.5× and a 40× water-immersed objective (40×/0.75 W, Olympus, Hamburg, Germany) (see Fig. 2). Individual layer V pyramidal cells in the unstained barrel cortex were visually identified using the 40× objective in combination with an IR light filter (KMZ 50-2, λ=750 nm, width 57 nm, max. transmission 0.56, Schott, Mainz, Germany), quarter-field illumination for contrast enhancement (DGC, Luigs & Neumann, Ratingen, Germany), and an IR sensitive CCD camera (C5405-01, Hamamatsu Photonics, Herrsching, Germany). Whole-cell patch-clamp recordings from the selected pyramidal cells were performed in current-clamp mode using patch pipettes (4 to 6 MΩ) pulled from borosilicate glass capillaries (1.5 mm o.d., 1.16 mm i.d., Science Products, Hofheim, Germany) on a Narishige PP-830 puller (Narishige, Tokyo, Japan). The patch pipettes were filled with (in mM): 13 KCl, 117 K-gluconate, 10 K-HEPES, 2 Na ₂ ATP, 0.5 NaGTP, 1 CaCl ₂, 2 MgCl ₂, 11 EGTA, and saccharose to achieve an osmolarity of 306 mOsm, as well as 1% biocytin for later histological processing of the intracellularly recorded cell (all chemicals purchased from Sigma-Aldrich Chemie, Steinheim, Germany). The neurons were electrophysiologically characterized by recording their resting membrane potential and intrinsic membrane properties under current-clamp conditions by injecting de- and hyperpolarizing current pulses. Membrane potential values were not corrected for the junction potential. To exclude possible influences of caged glutamate on the properties of the recorded cells electrophysiological characterization and long-time recordings were done in ACSF alone as well as in ACSF containing caged glutamate. They were virtually identical in result.

Figure 2

Schematic illustration of combined whole-cell recording and caged glutamate photolysis. Background: Schematic view of a coronal slice containing the barrel cortex; superimposed grid indicates stimulated fields with the position of the recorded neuron (filled triangle) and the stimulated field (filled square) also shown in the foreground. Foreground: Enlarged view of the investigated cortical area with the recorded neuron (left) receiving synaptic input from a presynaptic neuron within the stimulated field (right) in standard ACSF.

The setup used for the photolysis of caged glutamate was modified for single cell recordings from the one previously described in detail⁶ (Fig. 3). The microscope was placed on a mobile motor-controlled platform (Infrapatch, Luigs & Neumann, Ratingen, Germany) and could be moved under computer control relative to the fixed recording chamber. A pulsed xenon arc lamp (TILL Photonics, Planegg, Germany) delivered computer-triggered flashes of 0.5-ms duration that were guided via a glass fiber cable to a circular linear-wedge neutral-density filter (D_r=0.0 to 2.0, Melles Griot, Irvine, California) for the exact calibration of the illumination intensity. Illumination intensity in standard experiments was set to carefully avoid inducing action potential activity at the resting membrane potential when focusing at sites other than the soma of the recorded cell; for the purpose of the current experiments, the intensity was increased to facilitate flash-induced activity when targeting perisomatic sites. The light pulses (∼500 μs) were short-pass filtered (λ=340 to 390 nm) and limited in size by a variable rectangular aperture. Entering through the epifluorescence port and reflected by a dichroic mirror (405 nm, long pass), the flash was focused retrogradely through the objective onto a field of 50×50 μm in the focus plane. The standard focus plane was 50 to 100 μm below the surface of the slice and about equal to the depth of the recorded cell soma. In a set of experiments, we increased the focus depth systematically from the upper surface of the slice in 10-μm steps to a maximal depth of 150 to 200 μm. While adjusting the focus, stimulation of (peri-)somatic, basal and apical dendritic fields was repeated at 10-s intervals. The position of the flashed field could be selected either manually or under computer control for systematic scanning of a rectangular region of up to 1×2 mm.

Figure 3

Schematic illustration of the experimental setup combining IR video microscopy, UV flash photolysis of caged glutamate and computerized flash position control. The light paths for specimen observation and flash photolysis are indicated: UV, ultraviolet; IR, infrared; CCD, charge-coupled device (modified after Ref. 6).

The signals were amplified (SEC-05L, npi-electronics, Tamm, Germany), filtered at 3 kHz and digitized using an ITC-16 interface (Instrutech, Great Neck, New York). Data were recorded, stored and analyzed with PC-based software (TIDA 4.1 for Windows, Heka Electronik, Lambrecht, Germany). After recording, the slices were photographed in the bath chamber to document the topography of barrel related columns and layers as well as the respective position of the patch electrode. Slices were then fixed in 4% buffered paraformaldehyde and stored at 4°C. For visualization of the biocytin-filled neurons, the slices were processed according to a standard protocol¹⁷ and finally embedded in resin. Reconstruction and morphological analysis of the biocytin-labeled neurons were made using a Nikon Eclipse 800 (Nikon, Ratingen, Germany) attached to a computer system (Neurolucida; Microbrightfield Europe, Magdeburg, Germany). The shrinkage factors caused by the histological procedure were carefully determined for each individual slice. The shrinkage-corrected neuron morphology was superimposed onto the photomicrograph of the native slice using standard graphics software.

Different maps of glutamate-induced activity obtained in low Ca ²⁺ /high Mg ²⁺ containing ACSF were constructed using (1) flash evoked peak amplitudes and (2) delays between stimulus and onset of activity. For maps in normal ACSF, the following properties of stimulus-induced activity were analyzed within a time window of 150 ms poststimulus for each single trace: (1) delay to onset of stimulus-evoked activity, (2) occurrence of inhibitory postsynaptic potentials (IPSPs), (3) maximal peak amplitude, and (4) integral of all excitatory postsynaptic potentials (EPSPs). To distinguish between flash-induced activity and spontaneous activity, for each cell integral values of all spontaneous events within a time window of 150 ms were determined. For each cell, the highest integral value of spontaneous activity obtained in 40 control recordings was set as the cell-specific activity threshold. These control recordings were performed directly before mapping in ACSF containing caged glutamate without preceding photostimulation. Since inhibitory spontaneous activity was very rare, only integrals of excitatory spontaneous events were calculated. To identify glutamate-induced activity, integral values of all excitatory events following photostimulation in the same time window were calculated. Only activity that exceeded the cell-specific activity threshold was accepted as a glutamate-induced response. All integral values of glutamate-induced responses were corrected by the value of the cell-specific activity threshold. The analyzed response properties were transformed into pseudo-colored values using the software Origin 6.0 (Microcal Software Inc., Northampton, Massachusetts), and the resulting maps were superimposed on the respective sites on the micrographs also containing the reconstruction of the recorded neuron. Statistical analysis was performed using standard software (Systat 10, SPSS Inc., Chicago, Illinois).

2.2.

Compartmental Modeling

A layer Vb intrinsically bursting (IB) pyramidal neuron characterized in a previous publication¹⁰ and filled with biocytin was reconstructed morphologically from a whole-slice preparation using the NeuroLucida software package (MicroBrightfield, Williston, Vermont). The morphology was reconstructed from the unresectioned slice and corrected for shrinkage using a scaling factor of 1.15 ascertained for the x and y dimensions by matching it to a high-resolution photomicrograph of the recorded neuron in the original slice. Although tissue shrinkage in the z direction (depth of the slice) does not necessarily result in a corresponding linear shrinkage of the dendrites in this direction,¹⁸ for modeling purposes, the same scaling factor was also applied in the z direction to avoid an artificial elongation of the dendrites. Finally, the reconstructed morphology was aligned to a photomicrograph of the biocytin-filled original neuron by rotating it to a match within 0.5 deg (rotation around the soma coordinate).

The original reconstruction of the somatic outline was replaced by a single spherical compartment with a diameter of 20 μm, corresponding to a somatic surface of 1256 μm², similar to sizes reported from 3-D reconstructions of large layer V pyramidal neurons.¹⁹ To satisfy the criterion of isopotentiality the lengths of the dendritic compartments were edited using the meshing function of the CVAPP software package²⁰ with standard parameters (R _m =5 Ωm ², R_i=1 Ωm ⁻¹, maximal compartment length=0.01λ) resulting in a total of 2056 compartments. This edited morphology was exported to GENESIS cell descriptor-file format, where a correction for the area of missing dendritic spines was included by adding 0.83-μm² surface area per linear-micrometer of length to dendritic compartments.²¹

Information retrieved from the CoCoDat database (www.cocomac.org/cocodat) was used to equip the compartment model with passive parameters and active conductances (for details see Ref. 22). Briefly, the model comprised both fast and persistent Na ⁺ channels, several K ⁺ channels (KDR,KA,K2,KM,KC,KAHP), a transient low-threshold calcium channel, a high-threshold calcium channel and an H current. Only one type of high-threshold calcium conductance was included, as the different subtypes have been shown to have very similar biophysical properties.²³ Since most reported data were obtained either in the whole-cell configuration or using patches of somatic or apical dendritic membrane, we initially assumed that all dendritic domains contain most of the active conductances²⁴ and adjusted their conductance densities during the fitting procedure. Intracellular calcium ion concentrations were modeled in a thin shell directly under the neuronal membrane following a scheme consistent with the calcium-dependent conductances.²⁴ Since the axon was not included in the reconstruction, we located action-potential initiation in the neuron soma. The responses of the IB neuron model to current injections of 200-ms duration were fitted by adjusting the density or kinetics of the voltage-gated conductances, and the passive membrane parameters along the apical dendrites.

All voltage-gated channels were implemented as “tabchannel objects” within the GENESIS simulation system (V 2.2, Ref. 25), and simulations were run in Hines-solver mode using the Crank-Nicholson implicit method of numerical integration with a simulation time-step of 50 μs (for details see, e.g., Refs. 26 and 27). The conductance densities of the model were fitted to match the experimentally recorded responses to current injection in the same neuron. The model was further validated by comparison to published data showing an increased bursting tendency after treatment with an inhibitor of sodium current inactivation and repetitive burst firing following an increased depolarizing somatic current injection, as well as velocity measurements of back-propagating action potentials.

Maps and analyses based on the computer simulations were generated with Matlab 6.5 Rel. 13 (The Mathworks, Natick, Massachusetts).

3.1.

Direct Activation Versus Synaptic Input of the Recorded Neuron

The principal characteristics of the optical stimulation experiments are illustrated in Fig. 4 for a previously reported IB pyramidal neuron from layer Vb of the rat barrel cortex (see Ref. 10). The colored fields of the input maps code latency to the first response obtained for the same neuron using low Ca ²⁺ /high Mg ²⁺ ACSF [Fig. 4(a)] and standard ACSF [Fig. 4(b)], respectively. Low Ca ²⁺ /high Mg ²⁺ ACSF blocks synaptic transmission, which eliminates the synaptic inputs from stimulated presynaptic neurons. Synaptic inputs are clearly distinguished by the longer latencies compared to inputs from direct activation of the recorded neuron [see Fig. 4(c) and Ref. 10]. Under the low Ca ²⁺ /high Mg ²⁺ condition, somatic depolarizations were recorded only when the stimulating field covered a part of the somatodendritic region of the recorded neuron; this very precise restriction of responses to fields containing somatodendritic segments attests to the restriction of released glutamate to the focus area in the x and y dimensions. However, not all stimulated fields containing dendrites yielded a somatic response.

Figure 4

Topographical maps of activity induced by flash photolysis of caged glutamate for an IB pyramidal neuron from layer Vb (cf. Ref. 10). (a) Latencies to first activity induced by photolysis in low Ca ²⁺ /high Mg ²⁺ ACSF at a resting membrane potential of −69 mV. Latencies are always <8 ms. (b) Latencies to first activity in standard ACSF at a holding potential of −60 mV. Responses to stimulations of sites overlapping with somatodendritic structures have latencies <8 ms, whereas more remote stimulation sites (synaptically mediated activity) show latencies >9 ms. (c) Traces showing activity induced by flash stimulation of fields as indicated in the maps: traces 1 to 4, directly induced activity in low Ca ²⁺ /high Mg ²⁺; trace 5, synaptic inputs; and trace 6, directly induced activity (^*), which is followed by synaptic inputs (^**) recorded in normal ACSF. Scale bar in (a) is 200 μm.

3.2.

Factors Influencing Response Amplitude on Direct Activation

To investigate the dependence of the response amplitudes on the identity and depth of the somatodendritic structure, we analyzed response amplitude maps for this IB neuron [Figs. 5(a) and 5(b)] as well as a regular spiking (RS) neuron from layer V (not shown). APs were evoked only at perisomatic stimulation sites, and the response amplitude usually decreased drastically within ∼100 μm distance from the soma. The strongest responses were generally recorded from areas containing proximal dendritic branches, whereas weaker responses were recorded after stimulation of distal dendrites.

Figure 5

Morphology, peak response amplitudes, depth of most superficial neuronal structure, and relationship between response amplitude and depth for a layer Vb IB cell (see also Ref. 22). (a) Morphology of an intracellularly recorded and biocytin-filled cell reconstructed using NeuroLucida. (b) Somatically recorded voltage responses in low Ca ²⁺ /high Mg ²⁺ ACSF are color-coded to indicate the peak amplitude evoked after releasing glutamate in the respective square. Action potential (AP) amplitude was truncated at 13 mV. Positions where AP firing was induced are indicated with an asterisk (^*). (c) Color-coded map indicates the depth of the most superficial dendrite of the layer V pyramidal neuron at each stimulated square. (d) Relationship between the response amplitude and the depth of the most superficial dendritic segments. The response amplitude elicited by flash stimulation in a square quickly diminished below the depth of the focal plane. The dashed line in (d) shows the amplitude limit used for the model scaling function.

Maps showing the depth of the most superficial dendrite for each site in the stimulation grid were constructed using the coordinates of the shrinkage-corrected reconstructed morphologies [Fig. 5(c)]. These maps showed that neighboring stimulus sites contained most superficial dendrites at depths differing by up to 150 μm, providing a possible explanation for the differences in recorded response amplitudes between neighboring stimulus sites shown in the amplitude maps [cf. Fig. 5(b)]. The IB neuron also showed a prominent difference in response failures (i.e., stimulated fields containing dendrites but showing no response) between structures located above and below the depth of the focal plane.

When plotting the distributions of response amplitudes against the depth of the most superficial dendritic segment in the flashed field [Fig. 5(d)], a steep drop in maximal response amplitude following photolysis was seen when the most superficial dendrites were located deeper in the slice than the focal plane, which was located at the soma depth [Fig. 5(d)]. However, for individual dendritic segments there was no monotonic relationship between depth relative to the soma (=plane of focus) and response amplitude. Note that the effects of focus and of depth-dependent light absorption are synergistic below the plane of focus and antagonistic above it. Although we would expect the focus effect to lead to a steeper gradient for activation than tissue absorption, these data were insufficient to disentangle the two effects.

3.3.

Adjustment of a Computer Model to Simulate Responses to Photolysis

To model the apparent correlation between dendritic depth in the cortical slice and the elicited response amplitude, a simulation of the direct stimulation experiment was performed using the compartmental neuron model of the layer V IB cell, as described in Sec. 2.2. Each compartment of the model was equipped with a glutamate-activated synaptic conductance receiving simulated flash input as the simulated slice was scanned in a 10×25 grid identical to the experimental situation. Lacking information on the spatial profiles of flash-released glutamate in tissue slices, we used the upper boundary of the experimentally observed distribution of response amplitudes as a function of dendritic depth [Fig. 5(d)] to scale the activation strength for a given simulated synapse [dashed line in Fig. 5(d), excluding the truncated APs]. Furthermore, the model required an estimate of the time course of glutamate receptor activation in relation to photolysis of the caged compound. Release of glutamate following the UV flash stimulation has been reported^{15 28} to occur almost immediately with a time constant of ∼21 μs, and the development of the concentration profile after release has been estimated using Gaussian approximation of the focal volume of two photon excitation.²⁹ To approximate the experimentally recorded responses to the release of glutamate, the time constants of the model synaptic conductance were set to τ₁=0.1 ms and τ₂=100 ms with a simulated activation by a square pulse with a duration of 5 ms. The time constants of the synaptic conductance are not taken to reflect physiological properties of the actual channel, but they model the effect of the flash-released glutamate under the conditions of our specific setup. Since the lengths of the individual compartments in the model neuron depend on the reconstruction technique and on the isopotentiality requirement to be maximally 0.01 to 0.05λ the parts of the morphology with the smallest diameter dendritic branches (e.g., the apical tuft) contain a larger number of compartments per unit length than parts with relatively large diameter branches, such as the proximal apical dendrite. To avoid a bias toward activating distal compartments, the probability p _act of activating a model synaptic conductance was made proportional to the compartment length divided by 10 μm with a base probability p _base set to 1. Each model synaptic conductance was initially given a maximal conductance g _max,syn =25 nS and a reversal potential E _syn =0.0 mV. Activation was determined randomly for all dendritic compartments contained in a stimulated field using the probability already described, and the maximal activation amplitude was scaled according the depth in the slice for each individual synaptic conductance using the function given in Fig. 5(d).

Results from initial simulations proved it necessary to increase the maximal conductance of model synapses located on apical dendrites beyond the main apical dendritic shaft to induce measurable somatic responses. This was accomplished using a gradual increase of 5 nS/100 μm for model synaptic conductances located in apical dendritic compartments more than 75 μm away from the soma.

3.4.

Simulation of Direct Flash-Induced Activation of a Layer V IB Cell

Figure 6 shows the results obtained from three simulations of the flash experiment with different initialization seeds for the random number generator determining the choice of the individual model synapses to be activated in a stimulated field. As can be seen by comparing the simulated maps [insets in Figs. 6(a) to 6(c)] with the original experimentally obtained amplitude map from the layer V IB neuron in Fig. 5(b), activation of model synaptic conductances as a function of their depth in the slice produced maps that preserve the features reported from the experiments. Activations leading to AP firing were located close to the neuronal soma, and the response amplitudes decreased rapidly outside of the most perisomatically located stimulus sites. Individual larger responses were obtained at sites containing basal and apical oblique dendrites. Additionally, the more patchy activation patterns found in the medial and distal apical dendrite domains were also reproduced by the simulation.

Figure 6

Response amplitude distributions as topographic maps and as a function of depth from three simulations with different random seeds. The simulated distribution (triangles) is overlaid on the experimental distribution (circles). A minimal response threshold was set at 0.2 mV according to the smallest experimentally identifiable response amplitudes.

Although the response amplitude distributions from the simulations [Figs. 6(a) to 6(c)] followed the general shape of the experimental amplitude distribution, the topographic maps revealed that the model failed to produce responses for stimulation sites containing dendrites only at depths >100 μm [compare Figs. 5(b) and 5(c)] with amplitudes close to the set threshold for minimal responses of 0.2 mV.

The model did not capture the exact shape of the experimentally obtained map or the exact position and amplitude of individual recorded responses. This limitation could stem, for example, from discrepancies between dendritic positions in the experiment and the model, due to either differential shrinkage during the histological procedures or small inaccuracies in the alignment of the reconstructed morphology used in the simulations.¹⁸ Further factors to consider are tissue inhomogeneities that could influence the spatial extent of diffusion or elimination of uncaged glutamate.

3.5.

Predicted Effects of Varying Focus Depth

Prompted by the steep decline of flash-induced amplitudes at sites below the focal plane, the effect of variations of the focus setting on the map of response amplitudes was investigated using the computer model. Increasing the focus depth by roughly 10 μm into the slice increased the number of stimulation sites inducing AP firing in the simulation from two to four [compare Fig. 7(a) to Figs. 6(a) to 6(c)].

Figure 7

Response amplitude distributions as topographic maps and as a function of depth from two simulations with varying depth of the focal plane: (a) increasing depth by about 10 μm led to increased AP firing in the model and (b) moving the focal plane about 10 μm closer to the surface of the simulated slice abolished AP firing. The simulated distribution (triangles) is overlaid on the experimental distribution (circles).

Simulating a focal plane about 10 μm more superficial abolished AP firing in the model [Fig. 7(b)]. In both cases, the distribution of response amplitudes at larger dendritic depths remained unchanged by the change in focus depth. These results enabled us to predict that the extent and shape of the direct flash-evoked input maps would be highly stable, whereas the size of the response amplitudes and the occurrence of AP firing are strongly influenced by the focus depth.

During our usual experiments we keep the focal plane constant at a level of about 50 to 100 μm below the upper surface of the slice corresponding to the depth of the recorded soma. While this standard level has proven useful for obtaining reproducible input maps with optical release of caged glutamate, we decided to investigate more thoroughly the effects of focus level with a specifically designed experiment.

3.6.

Direct Experimental Test of the Predictions Concerning Focus Depth

A total of seven layer V neurons were recorded intracellularly in the whole-cell configuration and their responses noted to photolysis of caged glutamate at 41 somatic and proximal dendritic fields, while the focus depth was increased from 0 (surface of the slice) in 10-μm steps to a maximum depth of 190 μm. Altogether 522 stimuli were evaluated, of which 62 elicited APs in the recorded neuron.

As a first step, after exclusion of truncated APs, intracellular response amplitudes were normalized to the maximum observed at any focus position in the depth stack belonging to a single somatodendritic field. Figure 8(a) shows that the mean normalized amplitudes were large and relatively unchanged to a focus depth of at least 120 μm with a maximum at 40 to 70 μm. In a single field even with a focus depth of 190 μm, the amplitude still elicited a half-maximal response. This distribution indicates that optically induced glutamate release can be achieved with our setup at any focus setting between 0 and 200 μm from the surface of the slice.

Figure 8

Mean normalized response amplitudes+standard error of the mean (SEM) obtained with different focus depths from the surface of the slice (0 μm) to the deepest plane that still elicited unambiguous responses (190 μm) during the experiments. Response amplitudes were normalized to the maximum obtained at any focus depth in a given field. (a) Means calculated across all responses obtained with stimulations that were focused at the respective depth irrespective of the position of the maximum and the presence of a specific somatodendritic structure. (b) Means calculated after alignment of stacks to the focus depth that evoked the maximum response (0 μm). To the left are values for positions superficial to the maximum (negative values), to the right are values for positions deep to the maximum.

Further information about the focal depth dependence of flash-induced response amplitudes was obtained by aligning all stacks relative to the depth at which the maximal response amplitude was elicited in each of them. Figure 8(b) documents that maximal response amplitudes were not preferentially obtained with particularly shallow or deep focal planes. Although the normalized response amplitudes declined a little faster for focus depths deeper than the optimal depth, overall the falloff in response amplitude was fairly symmetric for focus planes above and below the optimum, particularly within the next 50 μm. Maxima were found up to a focus depth of 140 μm [Fig. 8(b), left end]. These results very clearly demonstrate that focus is an important factor determining the effect of optical stimulation in addition to simple light absorption with increasing tissue depth.

More detailed investigations required the reconstruction of the dendritic trees of the recorded neurons so that the induced response amplitudes could be compared specifically with the spatial distribution of somatodendritic structures. To this end we carefully determined shrinkage for each slice individually and corrected the reconstructions to match the depth during the experiments.

In the example shown in Fig. 9, the stimulated fields overlapped the soma as well as basal dendrites and the first 250 μm of the apical dendrite of a layer Va pyramidal neuron (Fig. 9, left and middle). Response amplitudes were highest at and near the soma (fields 2 to 4 and 8). Nevertheless, amplitudes were generally higher at field 4 than at field 3, although the former was 50 μm farther away from the soma than the latter. In a focus stack at a given field, the general impression is that evoked amplitudes decreased with depth; there were, however, several conspicuous exceptions where increased amplitudes occurred with deeper focus planes in positions that were characterized by the presence of dendritic structures (see stacks at fields 3, 5, 6, and 10). Thus, this example reveals both light absorption with tissue depth and focus effects, which appear to be superimposed.

Figure 9

Analysis of the correlation between depth of flash-stimulation focus and neuronal response amplitudes. Eleven fields were repetitively stimulated with increasing focal depth of the flash stimulus (10-μm steps downward) while simultaneously recording a single neuron in low Ca ²⁺ /high Mg ²⁺ ACSF. Left: photomicrograph of the living coronal slice with positioned recording electrode and, superimposed, the shrinkage-corrected somatodendritic reconstruction of the recorded layer Va pyramidal cell (light yellow). The x/y positions of the stimulated fields (50×50 μm) are marked by black squares. Middle: y-z axis views of the reconstructed neuron illustrating the dendritic organizations in the depth of the tissue. Superimposed columns 1 to 11 (position in the slice as marked to the left) illustrate the response amplitudes (color coded: blue to red) for each focal depth. Within these columns, only color coded depths have been stimulated. In the range of each column, only those cellular structures that pass through the field’s dimension are given. For positions 8 to 11 the yellow triangle marks the x position of the soma. Right: Recordings of the membrane potential for flash-stimulations at position 3 for all focal depths (increasing from top to bottom).

The population data are shown qualitatively as scatter plots in Fig. 10 separated in the results for apical and basal dendritic fields. The leftmost column shows the relationship between focus depth and (from top to bottom, ignoring the diagonal) response amplitude, z distance from the next dendritic structure, z distance from the next primary dendrite, and approximate x-y distance from the soma. The amplitude versus depth plot impressively shows the occurrence of every amplitude level at almost any focus depth (Pearson correlation coefficients r=−0.151 and r=0.027 for apical and basal dendrites, respectively). This finding speaks against the introductory scenario that superficial focus levels induce the biggest amplitudes, but on its own this is not sufficient to disprove the possibility that irrespective of the focus level, only the most superficial structures were activated. The distance to dendritic branches and primary dendrites increased with positions toward the surface or the depth of the slice, as most reconstructed dendrites were in the middle of it. The second column from the left, showing the amplitude relationships, provides the important information: response amplitudes in apical dendrites decreased visibly with focus distance from the next dendritic structure (r=−0.333); from the next primary dendrite (r=−0.348); and, of course, with the x-y distance from the soma (r=−0.548). All plots of soma distance in the bottom row show that apical dendrites were investigated to much larger distances than basal dendrites, of which the latter are largely confined to a radius of 150 μm around the soma.

Figure 10

Scatter plot matrix of all pairs of measures extracted for all optical stimulation sites that did not induce action potentials separated for apical (light, circles) and basal (dark, crosses) dendrites. Markers are jittered around the exact location to improve visibility of overlapping points. For each stimulation site, the following measures were noted: depth, response amplitude in low Ca ²⁺ /high Mg ²⁺ ACSF, z distance from next somatodendritic structure, z distance from next dendritic stem, and x-y distance from soma. Pearson correlation coefficients (r) are shown separately for apical and basal dendrites. Confidence ellipses are centered on the sample means of the two variables. The unbiased sample standard deviations of the two variables determine the major axes of the ellipse, and the sample covariance between the variables its orientation using a probability value of 0.6827. The diagonal shows box plots of the variable distributions with the box encompassing the central 50% of data points and the median shown.

Another important aspect of the experimental investigation was the analysis of factors that influence the occurrence of APs. Therefore, we extracted the 62 flash stimulations that evoked APs and performed analyses similar to the preceding, taking into account that the absolute number of occurrences must be seen in relation to the number of comparable tests performed. Figure 11 shows the proportion of stimulations that induced APs under several relevant conditions: AP firing was a rare event and did not depend on a particular focus depth [Fig. 11(a)]. It was, however, extremely important that the focus depth matched precisely the position of the dendritic structures or was in the vicinity of a primary dendrite [Fig. 11(b)]. If the focus missed a dendritic structure by only 10 μm, the likelihood of AP induction was reduced to less than 50%. Not surprisingly, the likelihood of evoking APs was higher in the vicinity of primary dendrites than of distal branches. When we split up the AP induction probability by the type of primary dendrite (basal or apical), we found that the focus distance effect was exponential for basal dendrites but not for apical dendrites [Fig. 11(c)]. It is difficult to explain, however, why a focus of 20 to 40 μm above or below the primary dendrite should be more effective than a focus discrepancy of 10 μm; a problem of alignment between experimental recording and reconstructed morphology can be excluded since this would reduce the maximum at 0 μm. Finally, the x-y (horizontal) distance from the soma played a major role [Fig. 11(d)]. Whereas flash stimulation at the root of the apical dendrite always induced an AP, the adjacent field was only very rarely effective. AP generation more likely was at 50 and 75 μm along the apical dendrite. Note that we increased the light intensity in these experiments compared to our usual experimental standard. Thus, the probability that about every fifth stimulus at this distance (and occasionally also at more distal apical compartments) induced an AP is much larger compared to what we routinely see. It seems, however, that an inhomogeneous glutamate sensitivity with “hot spots” responding strongly to photolytically released caged glutamate exists on proximal apical dendrites of layer V pyramidal cells.

Figure 11

Histograms of the proportion of sites whose flash stimulation induced APs in low Ca ²⁺ /high Mg ²⁺ ACSF. The four plots show the proportion of AP-inducing sites as a function of (a) focus depth; (b) z distance from the next primary dendrite (stem) and from the next dendritic structure, at all; (c) z distance from the next dendritic stem separately for basal and apical dendrites; and (d) x-y distance measured to the nearest 25 μm from the soma separately for basal and apical dendritic structures.

4.1.

Specificity of Optical Stimulation Techniques

The results of the reported investigations lead to the following conclusions concerning the questions raised in Sec. 1. Using our equipment, photolysis targeted at any depth from the surface of the slice to a depth of ∼200 μm activates local neurons [Fig. 8(a)]. The average amplitude of induced responses remains fairly constant to focus planes of 100 μm and only declines noticeably at greater depths [Fig. 8(a)]. In contrast to the relative focus independence of average responses, the response amplitudes obtained from individual dendritic stems and branches decrease with their distance from the plane of focus (Figs. 9 and 10). Therefore, both tissue depth and focus plane influence the induced response amplitudes. The focus dependence has very strong effects on the likelihood of AP induction, which decreases exponentially with focus distance leading to a more than 50% reduction within 10 μm [Figs. 11(b) and 11(c)]. As an additional finding, apical dendrites of layer V pyramidal cells show locally increased sensitivity to glutamate release with a maximum at 50 to 75 μm away from the soma. Locally increased sensitivity is not seen in basal dendrites at any distance [Figs. 9 and 11(d)].

These conclusions have important implications for the use of optical stimulation techniques in brain slices. First, the construction of synaptic input maps from recordings in standard ACSF relies on the perisomatically restricted induction of APs in neurons located in the illuminated region. To ensure that only stimulation of strictly perisomatic regions evokes AP activity, the light intensity must be carefully adjusted. The apical dendrites of layer V pyramidal cells provide a particular challenge since these are known to possess circumscribed regions of increased sensitivity³⁰ (see also Fig. 4 in Ref. 13). For the purpose of the presented experiments, we increased light intensity above our standard to increase the proportion of fields with stimulus-induced activity. Thus, we can exclude the possibility that flash stimulation of apical dendritic regions more than 75 μm away from the soma induces APs, leading to false localization of contributing cell bodies under our standard mapping conditions. For the same reason, we obtain a strong dependence of AP generation on the plane of focus, which can be used to advantage if structures at a particular depth are to be targeted as in direct activation experiments, or—as further options—in tangential slices or even in vivo through layer I. The downside of the high specificity is the limited sensitivity of our method as only a minority of somata in a region are stimulated sufficiently to produce APs, and only a fraction of functional inputs to a cell will be detected [see also Fig. 5(b)]. From a methodological point of view, it will be interesting to compare the results of our photolysis technique that uses an objective with a high (0.75) numerical aperture (NA) for precise targeting of the plane of focus to other methods with lower NAs (see Fig. 12). For example, a small laser beam that is coherent at the back plane of the objective does not use the full NA of the lens. Also, small optical fibers with a fixed focus and a low NA have been employed (e.g., NA=0.22, Ref. 31). A low-NA configuration might be good for testing circuitry due to stimulation of action potentials in more cells but without loss of spatial resolution. It would probably not be suitable for testing the spatial distribution of glutamate sensitivity of a dendritic arbor.

Figure 12

Illustration of the light intensity distributions (a) with high-NA and (b) low-NA optics. The light intensity is highest at the focus and diminishes with the axial (z) distance from the focal plane. This gradient is stronger with high-NA optics (our setup, NA=0.75) than with low-NA optics (e.g., laser, optical fibers). Superimposed on the focus effect is the absorption and scatter of light by the penetrated tissue.

Several other parameters can be manipulated to tailor the properties of photolysis to the requirements of specific experimental studies. Among these are the concentration of the caged compound and the light intensity for its photolysis. Pettit et al.¹³ report that increasing response amplitude by increasing caged compound concentration had no effect on axial resolution with single- or double-caged glutamate, whereas doubling the light energy by increasing flash duration significantly increased mean half-widths for double-caged but not for single-caged glutamate. In general, one would want to keep the caged glutamate low both to reduce the costs and to minimize effects of possible spontaneous glutamate release (see Ref. 32).

4.2.

Role of Computer Simulation

Although compartmental modeling of the biophysical properties of single neurons is a well-established technique, we reported here the first neuronal simulation of an optical stimulation experiment. This simulation included a model of the time constants of synaptic activation of glutamate receptors by flash-released glutamate, a model attributing the effects of glutamate release to the neuronal compartments contained in the field, and a model of the depth dependence of the synaptic peak conductance evoked in those compartments. The simulations were based on an independently tuned neuronal model where the additional parameters for optical stimulation were adjusted to reproduce the input map constructed from somatic response amplitudes to direct activation.

To evoke realistic responses to simulated glutamate release from sites on the apical dendrites more distal to the soma than ∼75 μm, a gradual increase in maximal synaptic conductance was incorporated into the model. This is in agreement with reports of increased densities of the AMPA (α-amino-3-hydroxy-5-methyl-4-isooxalone propionic acid) glutamate receptor subtype along the apical dendrite of CA1 pyramidal neurons from rat hippocampus and layer V pyramidal neurons from the rat neocortex.^{30 33} Since the direct flash stimulation experiments were carried out in low Ca ²⁺ /high Mg ²⁺ ACSF, contributions from activation of the NMDA (N-methyl-D-aspartate) glutamate receptor subtype are likely negligible due to a reduced driving force in the lower calcium environment along with a strong enhancement of the voltage-dependent magnesium ion block of this channel. The model distribution of response amplitudes as a function of the depth below the focal plane in a stimulated square showed a good correspondence to the experimental distributions, but failed to produce near-threshold responses at distances exceeding 100 μm. Clearly, to address this detail or the locally increased glutamate sensitivity along the apical stem a more sophisticated model will be required.

In addition, the simulation could be made more flexible in the future by explicitly modeling a variable plane of focus and the photolytic efficiency as a function of distance from this plane. Other possible extensions include the diffusion and elimination of uncaged glutamate. These factors have been considered in elegant calculations for hippocampal neurons and slices,^{13 29} but it is unclear how these apply to the more complex geometry of neocortical tissue.

Despite their simplicity our simulations correctly predicted that the spatial distribution of the input maps constructed from direct stimulation would be rather insensitive to the simulated focus depth, whereas the proportion of sites inducing AP activity was strongly dependent on the match between the plane of focusing and the level of activated somatodendritic structures. The dependence of induced responses from model synaptic conductances close to the focal plane was found to be very strong when using the scaling function based on the experimental distribution, with ∼10-μm displacements sufficient to double or abolish AP firing in the simulation. Nevertheless, it was necessary to be cautious concerning the validity of these predictions since they relied on crude estimates of some of the unknown model parameters and could have been affected by minor variations in the relative position of the reconstructed morphology and the stimulation grid. The nonlinear distortions of reconstructed morphologies caused by differential shrinkage during morphological processing¹⁸ may easily result in an inability of the model to capture the exact locations and amplitudes of individual responses in the maps obtained experimentally. The possible deviations are largest along the z axis where differential shrinkage during postexperimental tissue processing, as well as compression during the production of the slices, can significantly distort the shape of the slice.³⁴ Since the responses obtained in simulations using the reconstructed morphology succeeded in producing realistic maps and amplitude distributions, these effects were unlikely to be very large in our case, but possibly large enough to influence the position and strength of individual responses, as well as the near-threshold responses from dendrites deep in the slice. The fact that the model predictions held true in the subsequent experiments specifically designed to resolve these issues, argues for the utility of computer simulations even in cases where only rough estimates are available for some of the model parameters.

4.3.

Future Directions

The presented study demonstrates that optical stimulation of neurons through release of caged glutamate has a high spatial resolution even using a comparatively simple device such as a xenon flash lamp as the UV light source. The functional mapping of responsive properties of single cells by direct stimulation under elimination of synaptic transmission potentially could be much improved by chemical or optical two-photon uncaging of a variety of selective agonists and antagonists (e.g., Refs. 13, 29, and 30). The usual purpose of optical stimulation methods, however, is the mapping of the intricate microcircuitry in cortical slices based on monosynaptic transmission from systematic scanning of hundreds of stimulation sites (e.g., Refs. 2, 6, 35, and 36). On the stimulation side, the limitations are mainly in the static nature of the stimuli employed, the necessity of background activity to evaluate the effects of other than the fast excitatory glutamate transmitters, and the small proportion of cells activated beyond threshold to maintain the spatial selectivity of the stimulus. On the recording side, the obvious way to go is to record from several cells simultaneously with the aim of reconstructing the modes of information processing performed in local circuits. As an example, we show the simultaneous recording of two layer Va pyramidal cells, which were both characterized electrophysiologically as RS cells and whose inputs were mapped in a single scanning sequence with optical release of caged glutamate (Fig. 13). Although the two cells had very similar morphological features and were located in almost the same position, their input maps differed conspicuously both in their shapes and in the distribution of excitatory and inhibitory inputs. It is obvious that multiple simultaneous intracellular recording not only speeds up the process of data gathering but also enables direct comparisons of the input maps of individual cells.

Figure 13

Functional connectivity of two simultaneously recorded RS pyramidal cells in layer Va in standard ACSF. (a) Bottom, photomicrograph of a native coronal slice with recording electrodes positioned and, superimposed, somatodendritic reconstructions of the two recorded neurons (black, RS cell1; yellow, RS cell2); black dashed frame, investigated region; top, schematic illustration of layers, barrels and columns based on the photomicrograph shown below. Superimposed are the topographic maps separately for the RS pyramidal cell 1 (left) and the RS pyramidal cell 2 (right) that illustrate integrals of recorded EPSPs within 150 ms poststimulus (green to red) and fields of origin for IPSPs (blue). Note that for simplification, in fields where stimulation evoked EPSPs as well as IPSPs, only the IPSP is represented (blue). Fields given in sand color were excluded from analysis due to a strong temporal interaction between direct postsynaptic activation, AP generation, and synaptic events. Stimulated fields that did not induce any response are transparent. Black dashed frames mark the investigated cortical region. (b) Sample recordings of flash induced activity for positions as marked in (a). Responses of RS cell 1 are given in black, those of cell 2 in red. V _rmp , resting membrane potential; V _hold , holding potential.

Although intracellular recordings combined with the recovery of the cellular morphology provide a firm base for the reconstruction of the functional microcircuitry from intracellular recordings, this will, nevertheless, remain a formidable task given the number of cells types and variables to be investigated. These investigations should be complemented by multisite extracellular recordings, which are capable of capturing the population dynamics induced by a large number of different stimuli. Multisite recordings can be obtained very efficiently with electrode arrays conveniently arranged on prefabricated chips (cf. Ref. 37), and the long-term goal would be the development of optical stimulation techniques compatible with simultaneous optical recording. Also, computational approaches will continue to play an important role in these endeavors as they provide means to tackle the hard task of relating the properties of single cells to the observed population dynamics.