The electrophysiological and morphological properties of layer I neurons were studied in visual cortex slices from 7- to 19-d-old rats using whole-cell recording and biocytin labeling. A heterogeneous population of small, nonpyramidal neurons was found. Approximately one third of the cells we recorded were neurogliaform cells; another third were multipolar neurons with axons descending out of layer I. The remaining cells were heterogeneous and were not classified. In slices from 7- to 10-d-old animals only, we identified Cajal-Retzius cells.
Neurogliaform neurons had a very dense local axonal field, which was largely contained within layer I. Cells with descending axons had a relatively sparse local axonal arbor and projected at least to layer II and sometimes deeper. Spiking in neurogliaform neurons was followed by an afterdepolarizing potential, whereas spiking in cells with descending axons was followed by a slow after-hyperpolarizing potential (AHP). In addition, neurogliaform cells exhibited less spike broadening and a larger fast AHP after single spikes than did cells with descending axons. Generally, cells in layer I received synaptic inputs characterized as either GABA- or glutamate-mediated, suggesting the presence of excitatory and inhibitory inputs.
With their output largely limited to layer I, neurogliaform cells could synapse with other layer I neurons, the most distal dendritic branches of pyramidal cells, or the dendrites of layer II/III interneurons, which invade layer I. Cells with descending axons could contact a wide variety of cortical cells throughout their vertical projection.
Layer I of the neocortex is relatively cell-sparse and contains mostly nonpyramidal neurons (Prieto et al., 1994). This low density and the location immediately beneath the pial surface has hampered detailed studies of the cellular elements in layer I. In particular, there have been only few studies of layer I neurons using intracellular recordings (Martin et al., 1989; Zhou and Hablitz, 1996).
Ramon y Cajal (1911) described a horizontal cell type, later named the Cajal–Retzius (CR) cell, and several types of “short-axon” cells in layer I. The CR cells are observed only rarely in adult tissue (but see Condé et al., 1994), whereas the short-axon cells of layer I are maintained in adult cortex. The majority of layer I cells (90–95%) stain for glutamate decarboxylase or GABA and therefore may be GABAergic (Gabbott and Somogyi, 1986; Winer and Larue, 1989; Li and Schwark, 1994; Prieto et al., 1994). Moreover, most of the cells in layer I have smooth dendrites or only a few spines, suggesting that these may be similar to interneurons or nonpyramidal cells in layers II–VI. Most of the information reported has been obtained using Golgi techniques, some of which may not reveal fine axonal projections. We used patch-clamp techniques (Edwards et al., 1989) to record from visually identified layer I neurons and obtained physiological and morphological characterization of these cells.
MATERIALS AND METHODS
Slice preparation. Parasagittal slices (300 μm thick) from the visual cortex of 7- to 19-d-old rats (Wistar) were obtained using a vibroslicer (Campden). Ice-cold recording solution (see below) was used during slicing. The slices were maintained at 35°C for 1 hr. Slices were kept at room temperature during the recording. The solutions were bubbled with a gas mixture of 95% O2/5% CO2.
Recording. Patch pipettes (3–5 MΩ) were made from thin-wall (1.5 mm outer diameter, 1.17 mm inner diameter) borosilicate glass (Clark) using a vertical electrode puller (PP83, Narishige). Whole-cell recordings (Edwards et al., 1989) were made from layer I neurons selected under visual control using an upright microscope (Zeiss, standard 16, fixed stage) with Nomarski differential interference contrast optics using a water immersion lens (40× 0.75 NA). Whole-cell recordings in current-clamp or voltage-clamp mode were obtained using a patch-clamp amplifier (List EPC-7). No correction is made for the pipette junction potential (approximately −10 mV). The voltage and current output were filtered at 1 or 2 kHz (Frequency Devices, Haverhill, MA) and digitized at 12-bit resolution (TL1, Axon Instruments, Foster City, CA). After electrophysiological characterization, the pipette was withdrawn from the cell, and the slice was processed for histology (see below). Miniature excitatory or inhibitory synaptic currents were recorded in the presence of the sodium channel blocker tetrodotoxin (TTX) (Sigma, St. Louis, MO) and detected off-line using a tape storage device (Vetter, Rebersburg, PA). Data were digitized at 10 or 20 KHz and transferred to a disk. Miniature detection was based on threshold crossing of a set amplitude compared with a baseline amplitude. Miniatures were aligned at the 50% amplitude. An exponential function was fitted to each event. The rise times (20–80%), amplitudes, and decay time constants were stored in a separate ascii file used for the construction of histograms.
Solutions. The recording solution contained (in mm): 126 NaCl, 2.5 KCl, 1.25 KH2PO4, 1 MgSO4, 2 CaCl2, 26 NaHCO3, and 10 glucose, pH 7.4 (305 mOsm). Pipettes were filled with a solution containing (in mm): 144 K-gluconate, 3 MgCl2, 10 HEPES, 0.2 EGTA, 4 MgATP, and 0.3 NaGTP, pH 7.2 (295 mOsm). For labeling neurons, biocytin (0.1–0.3%; Sigma) was added to the pipette internal solution. To record miniature inhibitory postsynaptic currents (mIPSCs), we used a chloride-rich internal solution that contained (in mm): 80 K-gluconate, 40 KCl, 10 HEPES, 4 MgATP, 20 creatine phosphate (Na), 0.3 GTP, and 10 EGTA, pH 7.2 (295 mOsm). The AMPA receptor antagonist 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX) (RBI) and the GABA antagonist picrotoxin (RBI) were dissolved in aqueous solution.
Histology. Slices were fixed overnight in a solution of 4% paraformaldehyde, 0.2% picric acid in 0.15 m sodium phosphate buffer (pH 7.2–7.4) at 4°C. After thorough rinsing in PBS, the slices were incubated for 6–18 hr in avidin-biotin-horseradish peroxidase complex (ABC; Vector Labs, Burlingame, CA) in PBS containing 0.5% Triton X-100. The slices were rinsed four times in PBS and then reacted in a solution containing 0.06% 3–3′ diaminobenzidine, 0.03% H202, and 0.01–0.05% nickel ammonium sulfate for ∼30 min. The slices were then cleared and mounted in a 50–50% solution of glycerol-PBS. The coverslips were sealed with fingernail polish for storage. Filled neurons were drawn with a Nikon Optiphot microscope equipped with a drawing tube. Figures of drawn neurons were created after the drawings were scanned at 300 dpi (Hewlett Packard ScanJet IIcx). Digital micrographs were acquired with a cooled CCD camera (Photometrics, Tucson, AZ) with a frame resolution of 1200 × 1500. Photomontages of individual neurons were created from several focal planes. Final figures were constructed in Adobe Photoshop on a Power Macintosh, where contrast adjustments were made and paste marks removed. Micrographs were printed to a Tektronix Phaser 440 printer at 300 dpi.
Statistics. All error terms listed are SD. Between-group comparisons were made with a two-tailed Mann–Whitney Utest.
We were able to recover histologically ∼50% of all recorded cells. This article is based on 32 cells that were characterized both morphologically and electrophysiologically. All of the recorded cells could be classified as nonpyramidal neurons, as described earlier (Ramon y Cajal, 1911; Marin-Padilla, 1984). We characterized the pattern of action-potential trains in response to current injection, and in some cases we also studied the cells under voltage clamp. We found that neurons in layer I are heterogeneous in morphology. Most cells could be classified as neurogliaform cells, cells with descending axons, or early in development, CR cells. Photomicrographs of representative cells belonging to these three classes are shown in Figure 1. The remaining cells (about one third) were mostly multipolar, but were not classified further because the axon was not stained intensely enough to determine whether it arborized within layer I or descended to deeper layers.
CR cells are thought to be the first postmitotic cells that appear during embryonic development (Marin-Padilla, 1984; Bayer and Altman, 1990; Derer and Derer, 1990; Huntley and Jones, 1990). Biocytin labeling from cells in slices obtained from 7- to 9-d-old rats revealed CR-like cells (n = 6; Fig. 1,C1,C2). These neurons were characterized by an orientation largely horizontal to the one to three primary processes that arose from a round-to-ovoid soma (16.3 ± 4.5 × 10.3 ± 2.5 μm). Four of the six cells were bipolar in appearance, with two processes extending laterally from each pole of the soma. One process was thick, with a roughly contoured appearance resulting from a high density of appendages. These appendages varied from small spine-like protrusions to longer filiform processes. Often the thicker process expanded distally. From the opposite somatic pole, the opposing process was thin and axon-like, as in Figure 2. Both thick and thin processes could give rise to thin, vertical branchlets, which in turn could branch and project horizontally (Fig. 2). These thin processes extended anteroposteriorly several hundred micrometers.
The remaining two CR cells had processes with a similar morphology except that the bipolar orientation was not evident. Regardless, a thin process arose directly from either the soma or a thick process and projected several hundred micrometers beneath the pial surface, branching frequently. All of the CR cells were located in the upper half of layer I. In older rats (11–19 d), we recovered no cells with the unique CR morphology.
The average resting potential of CR cells was −64.4 ± 8 mV, and all CR cells were electrically active, as action potentials could be initiated with depolarizing current (Fig. 2, inset). These spikes had a long duration (half-width: 3.9 ± 0.7 msec). Trains of spikes were not always induced and when present were characterized by pronounced spike broadening and amplitude reduction. The input resistance of CR cells was high (610 ± 163 MΩ).
Eight of the 26 cells we recorded in slices from 11- to 19-d-old rats had a very dense axonal arbor that was contained almost entirely within layer I, and these were characterized as neurogliaform cells (Figs. 1 A1,A2, 3). In addition to the dense local axonal arbor, neurogliaform cells had five to six smooth and short primary dendrites that were restricted to layer I, and a small smooth soma (15.7 ± 4.3 × 9.9 ± 1.9 μm). Distally, all of the processes, but particularly the axon, were very thin and required high-magnification, oil-immersion optics to be followed. Neurogliaform cells were typically located in the middle to upper half of layer I. The axonal projection field could extend throughout the dorsoventral extent of the layer and anteroposteriorly 200–300 μm and could fill the thickness of the slice. The density of the arborization and the thinness of the processes often made distinguishing axons from dendrites difficult.
The average resting potential of neurogliaform cells was −62.0 ± 6.8 mV, and their average input resistance and membrane time constant were 500 ± 166 MΩ and 71 ± 42 msec (n = 8), respectively. Current injections from resting potential elicited single spikes at threshold level (Fig.4 A). Action potentials had a short half-width (1.2 ± 0.2 msec) and a large fast afterhyperpolarization potential (fAHP), measured as the difference between spike threshold and the peak hyperpolarization immediately after the spike (−19.9 ± 2.3 mV). In response to suprathreshold current injection, spike trains were often interrupted by quiescent periods (Fig. 4 B). Larger current injection resulted in an uninterrupted train of spikes showing some spike frequency adaptation in all but one neuron, which responded with only one spike. In addition to these characteristics, we also found that seven of the eight neurogliaform neurons tested exhibited a slow membrane depolarization after action potentials (+5.8 ± 1.7 mV) (Figs. 3,inset, 4 B–D). This afterdepolarization (ADP) peaked at 60.1 ± 28.7 msec after the termination of a single spike or a short train of action potentials and repolarized slowly with a time constant of 203 ± 60.2 msec (Figs. 3, inset,4 B–D). The ADP was observed after a train of spikes (Fig.4 C,D) or could be initiated by a single spike (Fig. 3,inset; see Fig. 8 D). The ability of cells to generate an ADP was not strongly dependent on the resting membrane potential. Under membrane hyperpolarization to −80 mV or more negative, however, the rising phase of the ADP was obscured by the spike repolarization. As will be shown below, the current underlying the ADP is inward at the resting membrane potential. Also shown in Figure 4 A is a depolarizing sag that developed with stronger membrane hyperpolarization, which was found to some degree both in neurogliaform cells and in cells with descending axons.
Cells with a descending axon
Seven of the 26 cells from 11- to 19-d-old rats had an axon that descended to layer II or deeper (Figs. 1 B1,B2,5). Like neurogliaform cells, the somata were small (14.9 ± 3.4 × 7.9 ± 1.7 μm) but were typically found in the middle to lower half of layer I. These cells had four to five sparsely branching dendrites. Two of the seven neurons had spiny dendrites, whereas the remaining five were relatively aspiny. For most of these neurons, the dendritic tree extended from the top to the bottom of layer I; additional dendritic projections were commonly found in the upper half of layer II. The anteroposterior extent of dendritic spread, however, was limited to <200 μm from either side of the soma. The distal portions of the dendrites were thin, as were secondary and tertiary branches. The single axon arose from either the soma or a primary dendrite. Although the axon sometimes branched within layer I, the local arborization was sparse when compared with that of neurogliaform cells. Scattered axonal branches were also found in deeper layers, and varicosities were often visible along collateral branches. The axon of two neurons reached layer IV and that of a third reached layer V. The remaining neurons had axons restricted to layers I and II. Collateral branches in deeper layers were not observed to extend anteroposteriorly much beyond ∼100 μm from the cell body. The axons were thin distally, however, and the staining was faint, and we cannot rule out a more extensive projection.
The resting potential (−63.8 ± 5.1 mV), input resistance (756 ± 308 MΩ), and membrane time constant (79 ± 27 msec) of cells with a descending axon were not significantly different from those of neurogliaform cells. These cells responded with a single spike to threshold stimulation and under suprathreshold current injection fired a steady train of action potentials. Action potentials had a half-width of 1.7 ± 0.3 msec and an fAHP of −12.2 ± 1.7 mV. Compared with neurogliaform cells, the fAHP was smaller (p ≤ 0.02), whereas the action-potential half-width was longer (p ≤ 0.02). During a spike train, there was spike frequency adaptation (Fig.6 A), and action potentials were broadened (Fig. 7). Cells with a descending axon exhibited more frequency adaptation than did neurogliaform cells. The ratio of the spike frequency of the fourth to the first interspike interval from briefly evoked spike trains was 0.85 ± 0.1 versus 0.70 ± 0.14 in neurogliaform cells and cells with descending axons, respectively (p ≤ 0.05). After a train of spikes, cells with a descending axon exhibited an AHP (Fig.6 B2). An AHP was also induced by a single spike (Fig.6 B1). The average amplitude and decay time constant of the AHP were −5.1 ± 2.7 mV and 211.6 ± 44.0 msec, respectively. Because of the more extensive spike broadening exhibited by cells with descending axons, the differences in the half-width and fAHP between these and neurogliaform cells were accentuated when the second spike was examined, and together these two features clearly distinguished the two cell types (Fig. 7). A single cell with a descending axon displayed a notch after the action potential, and an action potential could be triggered after a rebound from hyperpolarization (data not shown).
Three neurons were recovered that had an axon apparently restricted to layer I but did not locally arborize with the density of neurogliaform cells and thus were not classified as such. In addition, some of the dendrites extended into layer II. None of these three cells exhibited an ADP, and only one exhibited the slow AHP characteristic of cells with descending axons. Soma size averaged 16.8 ± 1.4 × 9.1 ± 0.5 μm.
The remaining neurons from which we recorded physiological data and recovered a stained neuron were left unclassified because the axon was not well stained. These cells were morphologically and electrophysiologically diverse; two exhibited an ADP and had short dendrites restricted to layer I, reminiscent of neurogliaform cells; five had a slow AHP and longer dendrites that often extended into layer II. Soma size ranged from 15.2 × 2.6 μm for the largest to 10.4 × 7.2 μm for the smallest neuron.
The slow time course of the ADP and AHP displayed by the neurogliaform cells and cells with a descending axon, respectively, may reflect slowly relaxing voltage-dependent currents or may be driven by calcium or other second messengers. Under current-clamp conditions, the time course of the ADP or AHP depends on voltage-dependent conductances and the membrane time constant and therefore may not represent the kinetics of the underlying conductance waveform. Figure8 illustrates AHPs and ADPs in cells with a descending axon (Fig. 8 A,C) and in neurogliaform cells (Fig.8 B,D), respectively. Under voltage clamp, a brief membrane depolarization induced a slowly developing outward current in neurons with an AHP (Fig. 8 C,C1) and an inward current in cells with an ADP (Fig. 8 D,D1). The decays of the I-AHP and I-ADP had a time course of several hundred milliseconds, similar to that of the voltage traces after action potentials. Furthermore, the I-ADP had a clear rising phase, suggesting that this current does not represent a tail current but reflects a slow response to events triggered by brief membrane depolarization.
The AHP has been observed in several cell types and is generated by an increase in K-conductance (for review, see Sah, 1996). The mechanisms underlying the ADP, however, have not been established. It has been suggested that in pyramidal and nonpyramidal neurons the ADP may reflect a calcium tail current, a decrease of potassium conductance (Constanti et al., 1993), an electrogenic ion pump (Friedman et al., 1992), or a nonselective cationic conductance (Caeser et al., 1993). We measured the voltage dependency of the conductance underlying the ADP by stepping the membrane potential to various voltages at the peak of the I-ADP (Fig. 9 A). The tail-current amplitude, defined as the difference between the steady-state current at the end of the trace and the current just after the voltage step plotted against the membrane voltage, indicated a reversal potential near −50 mV (Fig. 9 B). These data suggest that the conductance mechanism is most likely an increase in mixed cationic conductance rather than a decrease in potassium conductance or calcium tail current.
Excitatory and inhibitory synaptic inputs
Synaptic contacts in layer I of both symmetrical and asymmetrical types have been found (Beaulieu and Colonnier, 1985; Beaulieu et al., 1994). The functions of these synapses in relation to layer I neurons have been noted only recently (Hablitz and Zhou, 1995). To characterize these inputs further, synaptic currents were recorded, under voltage clamp, in the presence of TTX (0.5 μm) to prevent asynchronous release of neurotransmitter. In these recordings, biocytin was not included in the pipette solution. Synaptic currents recorded under these conditions are quantal events thought to originate from single synaptic contacts. We used pharmacological agents to isolate either the AMPA receptor-mediated excitatory postsynaptic currents (EPSCs) or GABA receptor-mediated IPSCs.
When the GABAergic blocker picrotoxin (100 μm) was present, the synaptic currents observed at −70 mV were excitatory inward currents (Fig. 10 A1). Application of the AMPA receptor antagonist CNQX (10 μm) blocked these currents, indicating that these are AMPA receptor-mediated miniature EPSCs (mEPSCs). To minimize the possible attenuation of synaptic currents by dendritic filtering, we selected miniature currents with a rise time faster than 0.3 msec. Under these conditions, the distribution of mEPSCs (Fig. 10 A2) was skewed to the right. The mean amplitude of the mEPSCs was 18.6 ± 4.6 pA (n = 7). Assuming that the AMPA receptor-activated conductance has a reversal potential near 0 mV, the conductance of the mEPSCs is 265.7 pS. Individual mEPSCs were well fitted with a single exponential function. The decay time constants were narrowly distributed about the mean (Fig. 10 A3). The average decay time constant was 2.2 ± 0.5 msec (n = 7).
To record IPSCs, we filled the patch pipettes with chloride-rich internal solution (see Materials and Methods). The estimated chloride reversal potential under these conditions is −20 mV. Therefore, the IPSCs recorded at −70 mV should produce an inward current. Inhibitory synaptic currents were recorded in the presence of TTX (0.5 μm) and the AMPA receptor antagonist CNQX (10 μm). Application of picrotoxin blocked these synaptic currents, indicating that these were mIPSCs. For analysis, we selected mIPSCs with a rise time of <0.6 msec. The inward currents recorded under these conditions had slower decay kinetics compared with that of the mEPSCs (compare Fig. 10, A1 and B1; note the different time scale). The peak amplitudes of the mIPSCs exhibited a skewed distribution (Fig. 10 B2) similar to that of the mEPSCs. The average mIPSC was 29.9 ± 4.9 pA (n = 6). Assuming that the reversal potential of the IPSC is −20 mV, we estimate that the quantal conductance of the IPSCs is 598 pS. When fitted with single exponential function, the mIPSCs were significantly slower compared with the mEPSCs, and the decay time constants were broadly distributed (Fig. 10 B3). We found, however, that the mIPSCs were better fitted with a dual exponential function (see Materials and Methods). The average fast time constant was 5.7 ± 2.2 msec, and the slow component was 30.1 ± 15.5 msec (n = 6). The average relative amplitudes of the fast and slow component were 62.3% and 37.7%.
Our main finding is that within layer I, there are distinct groups of small neurons that can be differentiated by their axonal projection and electrophysiological properties. In particular, we identified a group of neurogliaform cells with a very dense axonal field contained within layer I and a second group of neurons with an axon descending to the lower cortical layers. In addition, these two cell types exhibit different electrophysiological properties. Neurogliaform cells and cells with a descending axon are morphologically distinct from the CR cells observed in younger rats.
Neurogliaform cells have been described by several authors (Jones, 1984). Ramon y Cajal (1911) reported neurogliaform cells in layer I that he also called spiderweb or dwarf cells. Neurogliaform cells have been described in layer I of human newborn infants (Marin-Padilla, 1984) and in the rat visual cortex (Hedlich and Werner, 1987). Martin et al. (1989) used intracellular recording and HRP injection to obtain the receptive field properties and morphology of a single, layer I neuron. The morphology of that cell is similar to that of the neurogliaform cells we describe here (compare Fig. 2 of Martin et al., 1989, and Figs. 1 and 2 in this paper). Sousa-Pinto et al. (1975),Winer and Larue (1989), and Anderson et al. (1992) have not identified neurogliaform cells in layer I, but this may reflect incomplete staining of the fine axons. The robust filling of neurogliaform cells we have obtained in some cells revealed a very dense axonal arbor that rarely extended beyond the border of layer I. These data suggest that neurogliaform cells synapse primarily onto targets within layer I. These targets could include the dendrites of pyramidal neurons, dendrites from nonpyramidal neurons of layer II/III (Kawaguchi, 1995), and other layer I neurons. Kawaguchi (1995) described neurogliaform cells in layer II/III of frontal cortex after biocytin filling, with morphological characteristics remarkably similar to those reported herein.
It has been shown previously that spiking parameters can be used to differentiate pyramidal from nonpyramidal cortical neurons (McCormick et al., 1985; Connors and Gutnick, 1990). Further differentiation among cortical nonpyramidal neurons in layers II–VI has also been observed (Foehring et al., 1991; Kawaguchi, 1993, 1995, 1996; Kawaguchi and Kubota, 1993). The electrophysiological properties of neurogliaform cells have been studied only by Kawaguchi (1995), who found that these cells exhibit a delayed spike occurring at the end of a depolarizing current pulse. We have not observed the late spiking behavior in layer I neurogliaform cells, which possibly reflects different conditions of our experiments or perhaps a difference between layer I and layer II/III neurogliaform cells. We found, however, that layer I neurogliaform cells exhibit a characteristic ADP after spikes, which was not reported by Kawaguchi (1995). Cortical pyramidal neurons also exhibit a form of ADP, the time course of which is typically faster than the ADP in neurogliaform cells (Friedman and Gutnick, 1987;Schwindt et al., 1988; Foehring and Waters, 1991). Under voltage clamp, brief depolarization of neurogliaform cells generated a slow inward current (I-ADP) with kinetics similar to those of the ADP, suggesting that the time course of the ADP is not reflecting the membrane time constant but rather a slow conductance. The reversal potential of approximately −50 mV suggests a mixed cationic conductance. A cationic conductance is thought to underlie the ADP of hippocampal pyramidal cells induced by activation of glutamate metabotropic and cholinergic muscarinic receptors (Caeser et al., 1993).
In response to a steady current injection, layer I neurogliaform cells fire action potentials characterized by a large fAHP and less frequency adaptation and spike broadening when compared to cells with descending axons. Spike frequency adaptation and spike broadening are prominent in most pyramidal neurons and are found to a lesser degree in some nonpyramidal cells (Kawaguchi, 1995, 1996).
Cells with descending axons
We have identified a group of cells that had a prominent descending axon in addition to their local collateral projections. The characteristics of cells with a descending axon have not been studied extensively in the literature; however, Ramon y Cajal (Fig. 54 inDeFelipe and Jones, 1988) described layer I cells, which he called cells with descending axons, that had been discovered previously by Schaffer in 1897 (cited in DeFelipe and Jones, 1988). He described them as being located in the lower part of layer I, and they seem to correspond to the cells with descending axons that we describe here.
It is not possible to determine from our material whether the axons extend to the white matter, because the axonal staining became faint distally. Martı́nez-Garcı́a et al. (1994) found that some cells in the inner half of layer I project to the contralateral visual cortex. Therefore, it may be that some of the cells we classify as cells with descending axons provide input to distant regions in addition to the collateral inputs seen in deeper cortical layers.
Cells with a descending axon are clearly differentiated from neurogliaform cells both morphologically and electrophysiologically. Compared with neurogliaform cells, those with a descending axon exhibit a smaller fAHP and a larger spike width. The slow AHP and frequency adaptation seen in cells with a descending axon recall the regular spiking nonpyramidal cells described by Kawaguchi (1995; 1996) in layers II/III and V. Cells with a descending axon are morphologically similar to the subclass of nonpyramidal neurons that are immunoreactive for vasoactive intestinal polypeptide (Kawaguchi, 1996).
CR cells are among the first postmitotic cells. Recent findings suggest that they are important developmentally (Ogawa et al., 1995); however, their fate in the adult cortex has not been resolved. Only a few authors have reported on CR cells in the adult brain (Condé et al., 1994; Martı́nez-Garcı́a et al., 1994). Indeed, we have identified CR cells only in relatively young animals (7- to 9-d-old). Presently it is not known whether CR cells die at early postnatal age, change their morphology, or diminish in proportion to other proliferating cells (Marin-Padilla, 1984; Bayer and Altman, 1990; Derer and Derer, 1990; Huntley and Jones, 1990; del Rio et al., 1995). Electrophysiologically, CR cells are active, as has been shown recently by Zhou and Hablitz (1996) (Fig. 2, inset), and their spike parameters are similar to immature cortical neurons (Kriegstein et al., 1987; McCormick and Prince, 1987; Lorenzon and Foehring, 1993; Kim et al., 1995; Zhou and Hablitz, 1996).
Our reticence to classify approximately one third of the neurons reflects what we consider to be weak or incomplete staining of axonal arbors. In addition, the morphologies of a few well filled neurons simply could not be generalized to other groups of neurons and stood as isolated examples. It is likely that layer I contains other distinct groups of neurons that we have not identified simply because they were insufficiently represented in our study.
The overwhelming majority (90–95%) of neurons in layer I are GABAergic (Gabbott and Somogyi, 1986; Winer and Larue, 1989; Li and Schwark, 1994; Prieto et al., 1994). Lambolez et al. (1996) found that the AMPA receptors in layer I neurons have a GluR1–4 subunit composition that is characteristic of layers II–VI nonpyramidal neurons (Geiger et al., 1995). Moreover, the response to rapid application of glutamate in layer I neurons (Lambolez et al., 1996) is similar to that of other nonpyramidal neurons (Hestrin, 1993; Jonas et al., 1994). Thus, most layer I neurons can be classified, morphologically as well as physiologically, as being similar to inhibitory neurons found elsewhere in the cortex.
We found that layer I neurons receive both excitatory and inhibitory synaptic inputs. The excitatory input may be derived from collaterals of local pyramidal neurons and/or extracortical projection, including thalamic fibers. The source of inhibition is probably other neurons in layer I, but axons of inhibitory neurons from the lower layers may also contribute.
The predominant postsynaptic elements within layer I are the distal portions of apical dendrites from pyramidal neurons in layers II–V. Cortical inputs that selectively target layer I may undergo significant electrotonic attenuation (Cauller and Connors, 1994). Pyramidal cell dendrites, however, may generate sodium and/or calcium spikes that could boost synaptic inputs (Huguenard et al., 1989; Pockberger, 1991;Amitai et al., 1993; Kim and Connors, 1993; Magee et al., 1995;Schwindt and Crill, 1995; Stuart and Sakmann, 1995). Under these conditions, the distal dendrites could generate responses that are independent of more proximal cellular regions (Cauller and Connors, 1994). Inhibitory inputs originating from layer I neurons could selectively target distal dendrites and therefore may play an important role in local integration. Both apical dendrites and axon collaterals from pyramidal neurons reach layer I. Thus, it is possible that layer I neurons serve in a feedback inhibitory circuit. The dense but confined axonal projection of neurogliaform cells is particularly intriguing in that regard, suggesting that these cells function as local inhibitory neurons within specific local domains in layer I. In contrast, cells with descending axons are positioned to make contacts throughout the thickness of cortex. Whether the targets include the proximal portions of apical pyramidal cell dendrites as well as nonpyramidal cell elements remains to be determined.
This work was supported by National Eye Institute Grant EY-09120 (S.H.) and National Institutes of Health Grant NS-23941 (W.E.A.). We thank Mario Galarreta and Charlie Wilson for their help, Bob Foehring for comments on this manuscript, and Emin Kuliyev for his excellent technical assistance.
Correspondence should be addressed to Shaul Hestrin, Department of Anatomy and Neurobiology, College of Medicine, University of Tennessee, 855 Monroe Avenue, Memphis, TN 38163.