Layer 2/3 (L2/3) pyramidal neurons are the most abundant cells of the neocortex. Despite their key position in the cortical microcircuit, synaptic integration in dendrites of L2/3 neurons is far less understood than in L5 pyramidal cell dendrites, mainly because of the difficulties in obtaining electrical recordings from thin dendrites. Here we directly measured passive and active properties of the apical dendrites of L2/3 neurons in rat brain slices using dual dendritic–somatic patch-clamp recordings and calcium imaging. Unlike L5 cells, L2/3 dendrites displayed little sag in response to long current pulses, which suggests a low density of Ih in the dendrites and soma. This was also consistent with a slight increase in input resistance with distance from the soma. Brief current injections into the apical dendrite evoked relatively short (half-width 2–4 ms) dendritic spikes that were isolated from the soma for near-threshold currents at sites beyond the middle of the apical dendrite. Regenerative dendritic potentials and large concomitant calcium transients were also elicited by trains of somatic action potentials (APs) above a critical frequency (130 Hz), which was slightly higher than in L5 neurons. Initiation of dendritic spikes was facilitated by backpropagating somatic APs and could cause an additional AP at the soma. As in L5 neurons, we found that distal dendritic calcium transients are sensitive to a long-lasting block by GABAergic inhibition. We conclude that L2/3 pyramidal neurons can generate dendritic spikes, sharing with L5 pyramidal neurons fundamental properties of dendritic excitability and control by inhibition.
Pyramidal neurons in layer 2/3 (L2/3) constitute the majority of cells in the neocortex (Zilles, 1990). They are thought to be fundamental for cortical tasks such as feature selection and perceptual grouping (Grossberg and Raizada, 2000; Binzegger et al., 2004). Their morphology is similar to L5 pyramidal neurons, but the apical dendrite is thinner and shorter (Larkman et al., 1992). The differences in dendritic length and diameter have implications for their electrical compactness and integrative properties (Larkman et al., 1992; Zador et al., 1995). L5 pyramidal neurons have a rich assembly of dendritic conductances and correspondingly rich input–output capabilities. It has been assumed that L2/3 pyramidal neurons possess a composition of dendritic conductances similar to other pyramidal neurons (Wang, 1999; Traub et al., 2003). However, little is known about the regenerative properties of the apical dendrites of L2/3 pyramidal neurons except that they sustain active backpropagation of action potentials (APs) supported by dendritic Na+ channels (Waters et al., 2003; Waters and Helmchen, 2004) and have a concomitant influx of calcium ions (Svoboda et al., 1999).
Studies of dendritic excitation have focused on the large thick-tufted L5 neurons mainly for technical reasons. Although direct patch-clamp recordings from the apical dendrite of L5 neurons have become a routine technique (Gulledge et al., 2005), similar experiments on L2/3 neurons are more difficult because of their thinner dendrites. Here, we have overcome this difficulty by combining infrared differential interference contrast (IR-DIC) with fluorescence microscopy for guidance of patch pipettes. This approach enabled us, for the first time to our knowledge, to obtain simultaneous recordings from the somata and apical dendrites of L2/3 neurons in brain slices of rat somatosensory cortex. These dual recordings permitted us to study the subthreshold attenuation of synaptic potentials generated in the tuft as they spread toward the soma as well as the ability of the apical dendrite to actively participate in suprathreshold dendritic signaling.
L2/3 neurons receive their input predominantly from other L2/3 pyramidal neurons and L4 spiny stellate neurons (Lübke et al., 2003; Binzegger et al., 2004). These inputs project almost exclusively to the basal and apical oblique dendrites. Inputs to the apical tuft are thought to be feedback connections from higher cortical areas and nonspecific thalamic nuclei (Felleman and Van Essen, 1991). These connections show synaptic depression (Walcott and Langdon, 2001), can elicit APs, and can be potentiated by theta-burst stimulation (Walcott and Langdon, 2002). In L5 neurons, distal inputs can induce AP bursts via dendritic Ca2+ spikes (Connors and Gutnick, 1990; Amitai et al., 1993; Kim and Connors, 1993; Schiller et al., 1997; Larkum et al., 1999b, 2004; Larkum and Zhu, 2002). We were therefore particularly interested to determine whether spikes can be initiated in L2/3 distal apical dendrites. Although burst firing has been observed in L2/3 pyramidal neurons in vivo (Gray and McCormick, 1996), it is unclear whether AP bursting in L2/3 pyramids can be generated by distal dendritic input. Here, we find that L2/3 neurons are capable of generating dendritic spikes and associated calcium influx but not bursts consisting of three or more APs.
Materials and Methods
Experiments were performed in somatosensory neocortical slices from postnatal day 27–43 (n = 46) Wistar rats using procedures described previously (Waters et al., 2003). Rats were deeply anesthetized with halothane and decapitated. The brain was quickly removed into cold (0–4°C), oxygenated physiological solution containing the following (in mm): 125 NaCl, 2.5 KCl, 1.25 NaH2PO4, 25 NaHCO3, 1 MgCl2, 25 glucose, and 2 CaCl2, pH 7.4. Parasagittal slices, 300 μm thick, were cut from the tissue block with a vibratome and kept at 37°C for 30 min and then at room temperature until use.
All experiments were performed at 34.0 ± 0.5°C. Single L2/3 pyramidal neurons were identified using IR-DIC optics. Somatic (5–10 MΩ) and dendritic (∼20 MΩ) recording pipettes were filled with intracellular solution containing the following: 105 mm K-gluconate, 10 mm HEPES, 2 mm MgCl2, 2 mm MgATP, 2 mm Na2ATP, 0.3 mm GTP, 30 mm KCl, and 0.2% biocytin, pH 7.3. The intracellular solution also contained 10 μm Alexa 468 (Invitrogen, Carlsbad, CA), which was used to visualize the dendritic tree by fluorescence microscopy after establishing a whole-cell recording at the cell body. Here, an overlaid, static fluorescence/DIC image was compared with a live DIC image for positioning the pipette precisely onto the dendrite of the somatically patch-clamped neuron. Whole-cell recordings were made with Axoclamp-2B amplifiers (Molecular Devices, Palo Alto, CA). Synaptic stimulation was made via a glass pipette placed in the superficial half of L1, 100–200 μm from the axis of the apical dendrite. In some experiments, 2.5 μm bicuculline (Sigma, St. Louis, MO) and 1 μm (2S)-3-[(1S)-1-(3,4-dichlorophenyl)ethyl]amino-2-hydroxypropyl)(phenylmethyl)-phosphinic acid (CGP 55845; Tocris Bioscience, Ellisville, MO) were added to the extracellular solution. After recordings, slices were fixed and stained as described previously (Schiller et al., 1997) for later reconstruction.
For optical recording of dendritic Ca2+ transients, we combined dual whole-cell recording with Ca2+ imaging. Neurons were loaded with a fluorescent Ca2+ indicator [100 μm Oregon Green 488 BAPTA-1 (OGB-1); Invitrogen], which was added to the internal solution. Fluorescence changes were detected using a 512 × 512 back-illuminated frame-transfer CCD camera (PXL or MicroMax; Roper, Tucson, AZ). Images were binned at 5 × 5 or 10 × 10 and acquired at a frame rate of ∼40 Hz. For measurements of the relative fluorescence changes from baseline (ΔF/F), two methods were used. First, regions of interest (ROIs) in the dendrites were selected off-line at different locations from the soma. For each ROI, two background ROIs of the same dimensions were placed nearby over a region not including the cell. These were averaged and subtracted from the total fluorescence before calculating ΔF/F (Helmchen, 2005). A second method of determining ΔF/F was based on a spatial profile, which we determined off-line by placing a line along the dendrite (and an adjacent line for background subtraction ∼10–20 μm from the profile line). ΔF/F values were calculated by dividing pixel-by-pixel the fluorescence values in each frame by the average of the prestimulus frames.
Pooled results are reported as mean ± SD except when noted. Regression statistics were performed using the program GBStat (Dynamic Microsystems, Silver Spring, MD).
Dual somatic–dendritic recordings from L2/3 pyramidal neurons
We examined L2/3 pyramidal neurons of the somatosensory cortex in 4- to 6-week-old rats using dual somatic and dendritic whole-cell recordings in vitro. The mean depth of cell somata below the pia was 421 ± 58 μm (range of 276–505 μm; n = 36). To determine basic cellular properties, we injected 400-ms-long current pulses at either the dendritic or the somatic electrode. We first analyzed subthreshold voltage deflections (Fig. 1). All cells displayed a steady-state voltage–current (V–I) relationship showing anomalous rectification (Fig. 1B). Mean input resistance at resting membrane potential was 30.6 ± 10.7 MΩ (n = 9) at the soma and increased with distance from the soma (Fig. 1C) (mean dendritic input resistance, 42.1 ± 11.9 MΩ for distances between 60 and 322 μm from the soma; n = 10). Membrane time constants were 8.0 ± 1.3 ms for the soma and 8.0 ± 1.8 ms for dendritic sites. Rectification was quantified using a quadratic fit to the I–V curves (Waters and Helmchen, 2006), yielding mean rectification coefficients of 37.1 ± 23.1 and 26.1 ± 14.5 MΩ/nA for soma and dendrite, respectively.
In contrast to L5 pyramidal neurons, which show prominent sag in the voltage response to long current injections (30–40% in tuft dendrites in 28-d-old rats) (Zhu, 2000), L2/3 neurons displayed little sag either at the soma or in the dendrite (1.4 ± 2.5 and 3.2 ± 4%, respectively; full dendritic range from 60 to 322 μm). This indicates that less hyperpolarization-activated (Ih) current is present in the dendrites of L2/3 neurons than in L5 pyramidal neurons.
We next examined bidirectional voltage attenuation between the soma and dendrite. Steady-state attenuation was determined by analyzing voltage deflections in response to small hyperpolarizing currents and fitting the distance dependence of amplitude attenuation with an exponential curve (Fig. 1D). In the somatopetal direction (dendrite to soma), the space constant λFit was 302 μm, and, in the somatofugal direction (soma to dendrite), λFit was 556 μm. We also estimated the attenuation of spontaneous EPSPs. Although it was not possible to determine the exact origin of EPSPs from the dual-site recordings, we could in many cases identify the direction of propagation based on the order of onset at the two electrodes and the slopes of the rising phases of the EPSPs. In the example shown in Figure 1E, the first EPSP propagated somatopetally and was attenuated to 25% of its original amplitude, whereas the second, somatofugally propagating EPSP was attenuated to 46% at the dendritic recording site. In 28 neurons, we found a mean somatopetal EPSP attenuation constant (λEPSP) of 122 μm and a mean somatofugal λEPSP of 283 μm (Fig. 1F). Thus, as expected for transient membrane changes (Agmon-Snir and Segev, 1993), EPSPs were more strongly attenuated than steady-state membrane changes and spontaneous EPSPs were attenuated differently depending on the direction of propagation (Zador et al., 1995). On average, a distal EPSP was attenuated to approximately one-third its original amplitude over 120 μm, which is approximately one-quarter of the total apical dendritic length for L2/3 pyramidal neurons.
Suprathreshold current injections at both soma and dendrite caused regularly spiking trains of APs showing spike frequency adaptation in all cells (Fig. 2A). With long current pulses, APs in nearly all cases initiated at the soma regardless of the site of current injection (Fig. 2A). Only in 3 of 36 cells was an isolated dendritic spike (i.e., no somatic AP) elicited at threshold current injection to the dendrite (Fig. 2B). The amplitude of the dendritic backpropagating APs (bAPs) showed decremental attenuation with distance from the soma as reported previously (Waters et al., 2003). Dendritic bAP amplitudes were similar regardless of the site of current injection (Fig. 2C,D) (amplitude from threshold, 46.3 ± 19.6 and 45.0 ± 16.3 mV for somatic and dendritic injection, respectively; p > 0.25). Backpropagating APs were, however, significantly broader in the dendrite when long current pulses were injected at the dendritic site (Fig. 2C,D) (half-width, 1.8 ± 0.7 and 2.6 ± 1.5 ms for somatic and dendritic injection, respectively; p < 0.05, paired t test). This suggests that additional dendritic voltage-activated channels are activated during dendritic compared with somatic stimulation. This is consistent with an increased Ca2+ influx during sustained depolarization of the dendrite (Waters and Helmchen, 2004).
With further increased current injections the interspike intervals (ISIs) of the AP trains were reduced. The ISI of the first two APs became disproportionately faster, whereas all other ISIs in the train remained more or less regular. Hence, large current injections produced an initial doublet of spikes followed by a train of evenly spaced APs (supplemental Fig. 1, available at www.jneurosci.org as supplemental material).
Critical frequency for dendritic regenerative potentials
Because only a small fraction of cells produced dendritic spikes during long current injections, we asked whether trains of APs summate in the distal dendrite to induce additional dendritic conductances, as in L5 pyramidal neurons (Larkum et al., 1999a). Using somatic–dendritic dual recordings, we examined the amplitudes of APs in the dendrite during low-frequency (10–20 Hz) trains of APs produced by trains of short current pulses injected at the soma (Fig. 3). We found that the AP amplitude in the dendrite decreased only slightly during low-frequency trains of 10 APs (∼20% accommodation at 250 μm from the soma) (Fig. 3A). Pooled results for low-frequency trains (10–20 Hz) in 25 neurons confirm that there was very little attenuation even for the most distal recordings (Fig. 3B). This contrasts with results from neocortical L5 and hippocampal CA1 pyramidal neurons, which show pronounced activity-dependent attenuation of AP backpropagation in the apical dendrite (Spruston et al., 1995; Larkum et al., 2001).
In contrast, during higher-frequency AP trains (50 and 160 Hz), the dendritic amplitude of the last APs in the train increased in a frequency-dependent manner (Fig. 4A,B). The change in AP amplitude occurred over a narrow “critical” frequency (CF) range, above which an additional regenerative component was recruited, as in L5 neurons (Larkum et al., 1999a). The difference in dendritic AP amplitude was also reflected at the soma by an increase in the afterdepolarizing potential (ADP) of ∼1.5 mV (Fig. 4B). Although small, the somatic effect occurred consistently from trial to trial and from cell to cell. In addition, amplified dendritic APs were associated with large dendritic calcium influx (Fig. 4C).
The CF was not significantly different whether it was determined at the peak of the last dendritic AP or at a later time point from the change in the somatic afterpotential (Fig. 4D) (117 ± 13 and 118 ± 13 Hz for dendritic and somatic traces, respectively; p > 0.5, paired t test). Similar CF values were also obtained from dendritic calcium signals and somatic ADPs measured in a separate set of experiments (Fig. 4E) (121 ± 22 and 126 ± 23 Hz, respectively; p > 0.05). Finally, because there was no significant difference between the CFs measured at the dendrite and the soma, we evaluated the CFs in many neurons with only somatic recordings. The average CF for 68 L2/3 pyramidal neurons using this method was 128 ± 21 Hz (Fig. 4F), and the average change in ADP amplitude was 2.9 ± 1.9 mV (Fig. 4G). We conclude that amplified dendritic potentials with concomitant large calcium influx are generated by high-frequency somatic stimulation, consistent with the supralinear calcium influx into L2/3 dendrites during bursts of APs reported previously (Waters et al., 2003).
Dendritic spikes evoked by simulated EPSPs
In L5 pyramidal neurons, short current injections are more likely to elicit dendritic spikes than longer injections (Larkum et al., 2001). Because long current injections rarely evoked dendritic spikes at threshold, we investigated whether EPSP-like waveforms would evoke dendritic spikes in L2/3 neurons (Fig. 5). We injected a double-exponential current waveform with rising and falling time constants of 2 and 8 ms, respectively, resulting in a waveform with ∼3.5 ms time-to-peak. When injected into the dendrite, this waveform elicited EPSP-like traces with a time course at the soma similar to a typical EPSP evoked by extracellular stimulation in L1 (half-widths at the soma 15.4 ± 0.5 and 14.9 ± 0.8 ms, respectively; n = 14 and n = 13; p > 0.6). We then increased the amplitude of the simulated EPSP until it evoked a dendritic spike (Fig. 5A). The threshold for eliciting a dendritic spike was 1.8 ± 0.6 nA (n = 21). In some cases, the dendritic spike was coupled with a somatic AP (Fig. 5B). Our criterion for a dendritic spike was either an isolated spike at the dendritic electrode (Fig. 5A, Isolated) or a spike that arose at the dendritic electrode before the somatic electrode (Fig. 5B, Coupled). To quantify the regenerative dendritic component, we subtracted the “subthreshold” component (estimated by extrapolation from voltage responses to the previous two current steps) from the dendritic spike at threshold (Fig. 5A, middle traces). Average values of the amplitude and half-width of the additional potential (VDiff) were 20.7 ± 7.9 mV and 2.6 ± 0.9 ms, respectively (n = 11). In four cells, we could analyze the extra somatic depolarization caused by the dendritic spike, which was 12 ± 3% of the dendritic amplitude (2.0 ± 1.1 vs 16.0 ± 5.0 mV). The amplitude of the calculated dendritic regenerative potential was not dependent on the distance of the dendritic electrode from the soma (correlation coefficient of 0.1; p > 0.05), but the width of the dendritic potential was correlated with distance (correlation coefficient of 0.56; p < 0.05) (supplemental Fig. 2, available at www.jneurosci.org as supplemental material). A small negative component was apparent in the calculated potential after the dendritic spike, suggesting a contribution from K+ channels. Isolated or coupled dendritic spikes were also observed in four experiments with synaptic stimulation in layer 1 (supplemental Fig. 3, available at www.jneurosci.org as supplemental material).
We further characterized the coupling of the dendritic spike to the soma by calculating the ratio of the dendritic currents required to reach threshold for eliciting a dendritic spike and a somatic AP, respectively (Fig. 5C). A ratio of <1 indicates an isolated dendritic spike at threshold. Ratio of 1 indicates a coupled dendritic spike at threshold. A ratio of >1 indicates that a somatic spike was evoked at threshold but a coupled dendritic spike could be evoked with increased current injections. The ratio of the threshold currents clearly depended on the distance of the dendritic recording from the soma (Fig. 5C). A regression line fit to the data points crossed the unity line at 170 μm, indicating that dendritic current injection beyond 170 μm from the soma typically results in a dendritic spike at threshold. Because L2/3 neurons have variable length, we also analyzed where an input should occur relative to the length of the dendrite to elicit a dendritic spike at threshold (Fig. 5D). To this end, we marked the positions of the dendritic injection electrodes along a schematic L2/3 pyramidal neuron, normalized to the total distance between soma and pia (Fig. 5D). From this schematic, it is clear that a dendritic spike is likely to result from current injection into much of the apical dendrite, including more than half of its total length.
In all neurons, it was possible to inject sufficiently large current at the dendritic electrode to evoke a somatic AP. In some cases, it was possible to generate a doublet of APs at the soma, but the current threshold for somatic doublets was always higher than that for a single coupled AP. This contrasts with layer 5 neurons, in which even at threshold a dendritic spike frequently generates a somatic burst of two or more APs (Larkum et al., 1999b; Schwindt and Crill, 1999). Nonetheless, the dendritic component was still complex in L2/3 neurons, involving multiple regenerative components, some of which occurred before the bAP but became mixed with it (Figs. 5B, 6). The half-width of this complex potential waveform in the dendrite was up to 50 ms.
Backpropagating APs facilitate dendritic spike generation
The threshold for inducing a dendritic spike could be reduced by pairing dendritic current injection with a bAP (Fig. 6). For isolated dendritic spikes, the current threshold was reduced by 25 ± 7% when a somatic AP was elicited by a brief current injection 10 ms before the dendritic EPSP-shaped injection (Fig. 6A,C) (n = 11). Similarly, pairing a bAP with an EPSP-shaped dendritic waveform reduced the threshold for coupled APs by 24 ± 10% (Fig. 6B,C) (n = 8). The optimal timing for this facilitatory effect occurred with the dendritic current injection after the somatic current injection by 5–10 ms (n = 7; compared with 5 ms in L5 cells) (Larkum et al., 1999b). This facilitatory effect was not dependent on the passive spread of somatic depolarization because it was not observed with current injection just below threshold for a somatic AP (supplemental Fig. 4, available at www.jneurosci.org as supplemental material). In some cases, the pairing protocol led to an additional somatic AP at dendritic currents insufficient to cause a somatic spike under control conditions (e.g., for the cell in Fig. 6, a 2 nA waveform injected at the dendrite alone caused an isolated dendritic spike, whereas it was sufficient to cause a second somatic AP when paired with a bAP). This phenomenon is therefore similar to the backpropagation-activated Ca2+ spike (BAC) firing seen in L5 pyramidal neurons (Larkum et al., 1999b) in that the bAP facilitates a dendritic spike. In L5 neurons, BAC firing typically triggers a burst of three or more somatic APs. Although a doubling of the number of APs in L2/3 pyramidal neurons is possible with simultaneous input to the soma and apical trunk, in contrast with layer 5 neurons, we never observed more than two APs at threshold. This probably results from the briefer duration of dendritic spikes in layer 2/3 neurons, which never caused AP bursting (see above).
As reported previously (Waters et al., 2003), pairing a bAP and an EPSP evoked a large Ca2+ transient in the distal dendrite (115 ± 61% ΔF/F compared with 5 ± 4% for bAP alone and 5 ± 3% for EPSP alone; n = 4). This value was not significantly different from the ΔF/F value obtained with supracritical burst stimulation at the soma of the same cells (124 ± 74% ΔF/F; p = 0.68, paired t test), indicating that in both cases a regenerative dendritic potential was triggered. In one experiment, we succeeded in measuring the distal Ca2+ transient evoked by an isolated dendritic spike, demonstrating that a large Ca2+ influx was associated with the brief dendritic spike in an all-or-none manner (Fig. 6D).
Inhibitory control of the distal apical dendrite
As in L5 pyramidal neurons (Larkum et al., 1999b), the dendritically evoked distal Ca2+ influx that accompanied BAC firing could also be elicited with extracellular stimulation of L1 fibers, but only if GABAergic transmission was partially blocked with 2.5 μm bicuculline (GABAA) and 1 μm CGP 55845 (GABAB) (Fig. 7). Higher concentrations of bicuculline led to epileptiform discharges in the pyramidal neuron. Figure 7A shows an image of a biocytin-filled L2/3 neuron indicating the regions that were imaged during the experiment after the cell had been filled with the Ca2+ indicator OGB-1. White boxes show regions of interest at different distances along the apical dendrite (115, 260, and 365 μm). A single bAP caused almost no increase in [Ca2+]i in the distal tuft (Fig. 7B) (Waters et al., 2003), and there was also very little increase with paired distal synaptic input using an extracellular stimulating electrode in L1 (Fig. 7B). However, with antagonists of inhibitory receptors in the perfusate, the Δ[Ca2+]i along the dendrite increased, and a very large Δ[Ca2+]i in the tuft region was uncovered (Fig. 7B,C). This indicates that not only are Ca2+ channels associated with dendritic spikes but that inhibition dominates excitation in blocking Ca2+ regenerative events using extracellular stimuli, as in L5 pyramidal cells (Pérez-Garci et al., 2006).
We took advantage of the fact that AP trains of supracritical frequency evoke dendritic electrogenesis equivalent to dendritic spikes to further explore the inhibition of dendritic regenerative events (and dendritic calcium influx) by GABAergic transmission. As in Figure 7, an extracellular electrode placed in L1 was used to generate a compound EPSP/IPSP using five pulses at 200 Hz. In addition, a train of bAPs above the CF was generated at various delays by current injection through the somatic patch electrode (Fig. 8A). We then analyzed distal calcium transients, measured near the principal bifurcation, as well as the ADP after the last AP in the somatic recording as a function of the delay between synaptic stimulation and the AP train. We found that dendritic electrogenesis and calcium influx were progressively blocked as the extracellular stimulus got closer in time to the AP train (Fig. 8B). This block was apparent even for long delays of >300 ms. Only at 400 ms delay between stimuli did the calcium transients and ADP reach their control amplitudes (Fig. 8C,D). Suppression of the dendritic spike was attributable to the IPSPs because addition of antagonists of GABAergic transmission to the perfusate prevented blockade of the dendritic spike by the stimulus to L1 (data not shown).
The results presented in this paper complement the findings of Waters et al. (2003) in which we demonstrated that many of the properties of L2/3 pyramidal neurons are similar in vitro and in vivo. Here we investigated in more detail the regenerative properties of L2/3 pyramidal cell dendrites using dual recordings in vitro (which we were unable to achieve in vivo). We found that L2/3 pyramidal neurons share many features with L5 pyramidal neurons but also have distinct features that will influence their role in the cortical network. Most prominent among the similarities is the existence of a spike initiation zone in the distal apical dendrite, which acts as a second functional compartment and allows L2/3 pyramidal neurons to associate distal apical input with AP activity. In addition, we found that an inhibition blocks this dendritic activity with the same long timescale found in L5 pyramidal neurons (Larkum et al., 1999b; Pérez-Garci et al., 2006).
The most noticeable difference to L5 neocortical pyramidal neurons is that L2/3 dendrites support only brief dendritic spikes, not plateau potentials. Consequently, dendritic spikes facilitated by backpropagating APs caused only a single additional AP, not bursts of three or more APs. The lower capability of L2/3 neurons to generate AP bursts is seen in vivo (de Kock et al., 2007) and has consequences for the function of L5 and L2/3 pyramidal neurons in cortical networks (Lisman, 1997; Williams and Stuart, 1999). Another notable difference is the near absence of Ih in L2/3 neurons, which in L5 pyramidal neurons shunts the tuft dendrite (Williams and Stuart, 2000; Berger et al., 2001).
Common features of L2/3 and other pyramidal neurons
We found that input to the distal apical dendrite in L2/3 cells can initiate a dendritic spike involving both Na+ and Ca2+ currents. Dendritic spikes in L2/3 pyramidal neurons resemble dendritic Na+ spikes in hippocampal CA1 neurons (Golding and Spruston, 1998; Gasparini et al., 2004) and fast dendritic spikes in L5 neocortical pyramidal neurons that were evoked by short current injection (Larkum et al., 2001). Nonetheless, dendritic spikes in L2/3 neurons were associated with large dendritic Ca2+ influx, demonstrating activation of dendritic Ca2+ channels. The likelihood of eliciting dendritic spikes was highest when stimulating the upper 60% of the apical dendrite. This effectively means that inputs to the apical dendrite in L1 and upper L2 may cause dendritic spikes. Interestingly, despite the fact that the tuft dendrite of L2/3 pyramidal neurons is substantially closer to the soma than in L5 neurons, isolated dendritic potentials (i.e., without a somatic AP) were more common in L2/3 neurons (cf. Zhu, 2000). This difference is probably attributable to the small diameter of the L2/3 dendrite, implying a higher axial resistance (Larkman et al., 1992), and the fact that the dendritic spike is much briefer than in L5 pyramidal neurons.
The threshold for dendritic spike generation was high, but near-coincident bAPs reduced this threshold, much as in L5 pyramidal neurons (Larkum et al., 1999b). Because the tuft dendrites of L2/3 and L5 neurons overlap in L1, it seems likely that they would receive similar feedback inputs from higher cortical areas (Rockland and Pandya, 1979; Felleman and Van Essen, 1991; Lavenex and Amaral, 2000) and nonspecific thalamic inputs (Diamond, 1995). The associative properties of the apical dendrites are probably not as potent in L2/3 pyramidal neurons because (1) the threshold reduction by AP activity is smaller compared with L5 pyramidal neurons and (2) the dendritic spike produces at most one extra somatic AP, not the full-blown burst of three or more APs seen in L5 neurons (Larkum et al., 1999b). At the level of a cortical circuit, this lower efficacy in terms of additional single-cell output activity could be counterbalanced, however, by the greater number of L2/3 than of L5 pyramidal neurons (Peters et al., 1985).
Critical frequency and inhibition
High-frequency trains of bAPs evoked supralinear Ca2+ influx into the distal apical dendrite (Waters et al., 2003). The critical frequency was slightly higher compared with L5 neurons (128 vs ∼100 Hz) (Larkum et al., 1999a). In addition, the transition from subcritical to supracritical responses occurred over a greater range of frequencies. Nonetheless, the same criterion for defining the critical frequency was retained, namely, the turning point of the sigmoidal fit to the ADP versus frequency curve. Our measurements were done in a large population sample by exploiting the finding that the somatic ADP robustly reflects dendritic excitation. Notably, the dendritic calcium signal amplitude observed above the critical frequency was the same as for a dendritic spike evoked by direct dendritic depolarization. This indicates that, in both cases, a regenerative dendritic event is triggered.
The Ca2+ influx that occurs during either a dendritic spike or a supracritical frequency train of bAPs can be blocked by extracellularly evoked inhibition as in L5 pyramidal neurons (Larkum et al., 1999a,b). Furthermore, the prolonged time course of this inhibition is identical to that in L5 neurons (Pérez-Garci et al., 2006), which suggests that the underlying mechanisms are the same, presumably involving metabotropic action of GABA receptors on dendritic Ca2+ channels (Kavalali et al., 1997). The specificity of this inhibitory action on dendritic Ca2+ activity implies that it is of functional importance for cortical processing. We suggest that this mechanism is important in preventing associations between feedback and feedforward pathways via pyramidal neurons. It should also prevent Ca2+-dependent synaptic plasticity occurring in the distal tuft under certain conditions.
Distinct features of L2/3 pyramidal cells
The firing properties of L2/3 pyramidal neurons in response to injection of long depolarizing current steps at the soma were similar to those reported previously (Mason and Larkman, 1990), showing regular spiking with spike frequency adaptation. Unlike in both L5 and CA1 pyramidal neurons (Spruston et al., 1995; Stuart et al., 1997; Larkum et al., 2001), there was almost no spike amplitude adaptation in a train of backpropagating APs at any distance from the soma. In L5 pyramidal neurons, long depolarizing dendritic current injection evokes pronounced repetitive burst firing characteristic of intrinsically bursting neurons (Connors and Gutnick, 1990; Larkum and Zhu, 2002), even when the same current at the soma evokes a regular spiking pattern. Our findings here indicate that L2/3 pyramidal neurons of the somatosensory cortex in rats do not respond with long bursts to dendritic current injection. We never saw the fast rhythmic bursting or “chattering” L2/3 neurons seen in cat visual cortex in vivo with sharp electrodes (Gray and McCormick, 1996; Nowak et al., 2003) and in L2/3 of ferret cortex (Brumberg et al., 2000). Brumberg et al. (2000) showed that repetitive bursting in ferret L2/3 pyramidal neurons depends critically on the age (>4 months) and extracellular calcium concentration (≤1.2 mm). It is thus possible that the same constraints apply to rat cortex, which might explain the absence of such cells in this study.
The input resistance showed an increasing relationship as a function of distance from the soma, which is different from L5 pyramidal neurons of the same age (Zhu, 2000). In L5 neurons, dendritic Ih has a counterbalancing role for input resistance (Berger et al., 2001) and leads to even more pronounced sag in the distal tuft region (Zhu, 2000; Williams and Stuart, 2000; Berger et al., 2001). We were therefore interested to test for dendritic sag with direct dendritic current injection. Although the slight sag in L2/3 pyramidal neurons indicated that some Ih current is present, it is far less pronounced than in L5 pyramidal neurons (Zhu, 2000) and does not affect the dendrite more than the cell body. Because Ih can reduce the amplitude and time window of summation of synaptic input (Williams and Stuart, 2002; Berger et al., 2003), the relative absence of Ih in L2/3 neurons suggests that the timing for association of top-down and bottom-up inputs might be different. Although the optimal separation of bAP and dendritic input appeared to be slightly longer (5–10 ms) than in L5 pyramidal neurons (Larkum et al., 1999b), this question will require a separate study.
In conclusion, we found that L2/3 pyramidal neurons can generate dendritic spikes that are accompanied by large Ca2+ influx. L2/3 neurons thus can associate feedforward and feedback inputs arriving at the proximal and distal regions of the neuron. However, the dendritic Ca2+ influx can be prevented for a prolonged period of time by inhibition. Although these properties are similar to those of L5, the tuft dendrites of L2/3 neurons appear less coupled to the soma compared with L5 tufts. The associative properties demonstrated here will have consequences for the function of L2/3 pyramidal neurons in cortical microunits.
This work was supported by a Marie Curie individual fellowship from the European Community (J.W.) (Contract Number QLGA-CT-2001-50999). We thank Paul Rhodes for helpful discussions and comments on this manuscript and Marlies Kaiser for expert technical assistance.
- Correspondence should be addressed to Matthew Larkum at his present address: Institute of Physiology, University of Bern, Bühlplatz 5, CH-3012 Bern, Switzerland.