The neocortical neuropil has a strong vertical (orthogonal to pia) orientation, constraining the intracortical flow of information and forming the basis for the functional parcellation of the cortex into semi-independent vertical columns or ‘modules’. Apical dendrites of excitatory pyramidal neurons are a major component of this vertical neuropil, but the extent to which inhibitory, GABAergic neurons conform to this structural and functional design is less well documented. We used a gene gun to transfect organotypic slice cultures of mouse and rat neocortex with the enhanced green fluorescent protein (eGFP) gene driven by the promoter for glutamic acid decarboxylase 67 (GAD67), an enzyme expressed exclusively in GABAergic cells. Many GAD67–GFP expressing cells were highly fluorescent, and their dendritic morphologies and axonal patterns, revealed in minute detail, were characteristic of GABAergic neurons. We traced 150 GFP-expressing neurons from confocal image stacks, and estimated the degree of vertical bias in their dendritic trees using a novel computational metric. Over 70% of the neurons in our sample had dendritic trees with a highly significant vertical bias. We conclude that GABAergic neurons make an important contribution to the vertical neocortical neuropil, and are likely to integrate synaptic inputs from axons terminating within their own module.
Neurons are inherently polarized, the result of having one process (the axon) structurally and functionally different from all their other processes (the dendrites) (Black and Baas, 1989; Mattson, 1999). While the morphological polarity between axon and dendrites is common to virtually all neurons, a significant subset of neurons exhibit morphological polarity also within their dendritic tree. For example, pyramidal neurons of the cerebral cortex extend a single, thick apical dendrite in one direction, and a multitude of smaller, basal dendrites in other directions. The apical dendrite is preferentially oriented upwards, towards the pial surface. The vertically oriented apical dendrites of pyramidal neurons (Escobar et al., 1986; Peters and Sethares, 1991; White and Peters, 1993), together with vertically oriented afferent, efferent and intrinsic axons (Lorente de Nó, 1949), are the major contributors to the predominantly vertical grain of the cortical neuropil and impart a strong vertical bias to the flow of information in the neocortex, forming the structural basis for the well-documented parcellation of the cerebral cortex into semi-autonomous anatomical and functional units, called ‘columns’ or ‘modules’ (Colonnier, 1966; Hubel and Wiesel, 1969; Szentagothai, 1975; Mountcastle, 1997).
Pyramidal neurons are the dominant cell type of the neocortex, and function to excite postsynaptic neurons, both locally and in other cortical areas and subcortical targets, by secreting glutamate as a neurotransmitter (Ottersen and Storm-Mathisen, 1984; Carder and Hendry, 1994). In contrast, ~15–25% of all neocortical neurons (Fitzpatrick et al., 1987; Ren et al., 1992; Micheva and Beaulieu, 1995; Gabbott et al., 1997) are inhibitory, non-pyramidal neurons, which utilize GABA as a neurotransmitter. Dendrites of GABAergic cortical neurons usually lack morphological polarity (i.e. all their dendrites appear morphologically equivalent), but may be spatially polarized. Indeed, some GABAergic neurons have bipolar or bitufted dendritic trees, in which dendrites emanate mainly from two opposite poles of the cell body. The majority of GABAergic neurons, however, have multipolar, or ‘stellate’, morphology with dendrites emanating in all directions (Houser et al., 1983; Kisvarday et al., 1990; Prieto et al., 1994). Although many multipolar neurons appear to have vertically elongated dendritic trees (Jones, 1975; Feldman and Peters, 1978), this vertical bias has rarely been studied quantitatively (McMullen et al., 1984).
All GABAergic neurons express GAD (glutamic acid decarboxylase), the major biosynthetic enzyme for GABA (Ribak, 1978; Houser et al., 1983; Martin and Rimvall, 1993). There are two known GAD isoforms, GAD65 and GAD67, which are the products of two different genes (Erlander et al., 1991; Kaufman et al., 1991; Esclapez et al., 1994). Neuron-specific and region-specific expression of the GAD67 gene is conferred by its complex 5′ flanking region, which includes several neuron-restrictive silencer sequences and putative region-specific enhancers (Szabo et al., 1996; Katarova et al., 1998). To visualize GABAergic neurons in living tissue, and to study their axonal and dendritic morphologies, we linked the cloned mouse GAD67 promoter region to an enhanced green fluorescent protein (eGFP) reporter gene, transfected neurons in organotypic cortical slices with the resulting construct, and used a novel metric to quantify the degree of vertical bias in dendrites of transfected neurons expressing GFP. Some of these results have been previously reported in abstract form (Jin et al., 1999).
Materials and Methods
GAD67–GFP DNA Preparation
A 9 kb segment of the GAD67 gene's 5′ regulatory region, including the first intron, drives expression of a lacZ reporter transgene in hippo-campus and neocortex of transgenic mice in a partially cell-type specific pattern (Katarova et al., 1998). To improve specificity, we used a slightly longer segment, and to allow visualization of the reporter gene product in living tissue, we fused it in-frame to eGFP. We cloned a 10.3 kb segment of the murine GAD67 genomic DNA, including the 5′ upstream regulatory region, the first (untranslated) exon and intron and a portion of exon 2, from the HindIII site (at the 5′ end) to the BamHI site (at the 3′ end) (Katarova et al., 1998), and inserted it in-frame between the HindIII and the BamHI insertion sites of the pEGFP-1 vector (Clontech, Palo Alto, CA). The resulting construct encodes a fusion protein consisting of the first 17 amino acids of GAD67, four amino acids of the vector and the full-length eGFP.
Organotypic Culture Preparation
Organotypic cortical slice cultures were prepared using a previously described method (Stoppini et al., 1991). Postnatal day 2–7 mouse or rat pups (postnatal day 0 being the first 24 h after birth) were deeply anesthetized by Metofane inhalation, decapitated, and the brains rapidly removed and immersed in ice-cold artificial cerebrospinal fluid (ACSF, containing 126 mM NaCl, 3.0 mM KCl, 1.3 mM MgSO4, 2.5 mM CaCl2, 1.2 mM NaH2PO4, 26 mM NaHCO3, and 20 mM dextrose, saturated with 95%:5% O2:CO2). The following procedures were performed under sterile conditions. Coronal brain slices of parietal cortex, 350 μm thick, were cut with a Vibraslicer (WPI, Sarasota, FL) into ice-cold ACSF. Slices were then placed on transparent Millicell membrane inserts (Millipore, Bedford, MA), usually two slices/insert, in 30 mm Petri dishes containing 1 ml of culture medium, composed of (per 100 ml) 46 ml Eagle's basal medium, 25 ml Earle's balanced salt solution, 25 ml horse serum, 1 ml penicillin– streptomycin–glutamine (all from Life Technologies, Rockville, MD), and 3 ml of 20% glucose solution. The slices were kept in a humidified incubator at 37°C with a 5% CO2-enriched atmosphere, and the medium was changed twice a week.
Gene Gun Mediated Transfection
Cultured slices were transfected using a Helios gene gun (Bio-Rad, Hercules, CA). Cartridges were prepared according to the manufacturer's manual, using manufacturer's supplies. The full, supercoiled plasmid DNA (total size 14.6 kb), prepared using an endotoxin-free kit (Qiagen, Valencia, CA), was used for transfection. Gold particles (1.6 μm diameter) were coated with plasmid DNA at a ratio of 1 mg gold to 2 μg plasmid DNA, precipitated onto the inner wall of Tefzel tubing, and the tubing cut into individual cartridges. Each cartridge contained ~1 μg plasmid DNA. Assuming (arbitrarily) a 50% coating efficiency, each gold particle carried on average ~2700 copies of the DNA construct. After 1–2 days in culture, the slices were bombarded with one cartridge per insert under 100 psi helium pressure. After 2–35 more days in culture, slices were removed for imaging.
Explants were cut out together with the membrane and transferred to an imaging chamber filled with fresh ACSF. Well-separated neurons with bright GFP expression and no signs of truncation were selected for imaging. An inverted Zeiss LSM510 laser-scanning confocal microscope was used to capture images. Images were generally taken with a ×20, 0.7NA objective, using the 488 nm argon laser line and a 505–550 nm bandpass emission filter. Image stacks were collected at ~1.5–2 μm Z-axis steps to cover the full depth of the dendritic tree. Stacks were superimposed digitally, but in most cases the full 3-D dataset was saved for further analysis. At the end of each session, images were taken at a lower magnification (×10 objective) in fluorescence and transmission modes, to record the locations of the GFP-expressing neurons relative to the pia and white matter.
Confocal image stacks of individual neurons were traced digitally using Neurolucida software (Microbrightfield, Colchester, VT); all subsequent analysis was done on the traced data. For each neuron, a line was drawn on a low-power image from the cell body to the nearest point on the pial surface, and the coordinates of the two endpoints measured; these were later used in the analysis to calculate angle of growth relative to the vertical axis.
The Neuroexplorer software (Microbrightfield) was used to calculate numerical parameters of dendritic morphology, including soma area, total dendritic length and number of dendritic end points, and to measure the maximal vertical and horizontal extent of the dendritic tree. Dendritic coverage area was defined as the area of the smallest convex polygon that could encompass all the dendritic branches. Dendritic tree aspect ratio was defined as the ratio of the maximal vertical extent to the maximal horizontal extent of the dendritic tree. To analyze dendritic growth directions, Neurolucida data in ASCII format was imported into HiQ (National Instruments, Austin, TX) and calculations were carried out using custom-programmed HiQ scripts. Since organotypic slices grown in serum-supplemented medium tend to become quite thin, coordinates along the depth (Z) dimension varied very little, compared with coordinates in the plane of the slice; we therefore ignored the depth dimension in this analysis and treated the neurons as planar projections. The program first converted all dendritic chords (i.e. the segments defined by two sequential data points along the dendrite) into vectors anchored at the origin of axes, then calculated the sum of the absolute values of the dot product of all vectors with a normalized vector representing the vertical direction (equivalent to the total length of the projections of the vectors on the vertical axis), and divided by the sum of all the vector norms (equivalent to the total dendritic length), resulting in the weighted average of the absolute value of the cosine of the angle of growth relative to the vertical axis (WACAV). The average WACAV of neurons with randomly oriented dendrites should be 2/π or ~0.64; this applies both to a two-dimensional neuron with dendrites oriented randomly in the plane, and to a three-dimensional neuron with dendrites oriented randomly in space, as long as the WACAV is calculated on the two-dimensional projection of the neuron on a vertical plane (see Appendix).
Morphology of Neurons Expressing GAD67–GFP in Organotypic Cortical Slices
The data presented here were collected from 143 organotypic slice cultures prepared from 40 mice and 8 rats. To visualize GABAergic, non-pyramidal cortical neurons, we biolistically transfected organotypic cortical slices with the eGFP gene controlled by the mouse GAD67 promoter (see Materials and Methods). The number of GFP-expressing neurons observed in successfully transfected slices varied from ∼10 (or occasionally less) to more than 100, possibly reflecting fluctuations in the number of gold particles deposited on the wall of each gene gun cartridge. Invariably, any GFP-expressing neuron examined was found to contain a gold particle in its nucleus. Levels of GFP expression varied between neurons: in many, mostly small neurons, GFP fluorescence was low and mostly restricted to the cell body, but in many medium and large transfected neurons, fluorescence level was high and allowed clear visualization of the full dendritic and axonal morphology. GFP-expressing neurons occurred in all cortical layers, although in younger slices they had a tendency to be more concentrated in the deep layers.
With almost no exceptions, GAD67–GFP expressing neurons were non-pyramidal, as typical of GABAergic neurons. Their dendritic and axonal morphologies were varied; Figures 1 and 2 illustrate confocal images of neurons representative of the range of morphologies observed. The great majority of GFP-expressing neurons were either bitufted (Figs 1A, 2C, 3B) or multipolar (Figs 1B,D, 2A,D, 3C–G). While most of the GFP-expressing neurons had non-spiny or sparsely spiny dendrites typical of mature inhibitory cells in the cortex, some were more densely spiny (Fig. 1D), possibly reflecting an immature stage of development (Lund et al., 1977; McMullen et al., 1988a; Seay-Lowe and Claiborne, 1992). The GFP fluorescence also revealed highly detailed axonal arbors, rivaling in complexity those visualized by the Golgi or the intracellular dye-filling methods (Figs 2 and 3). Some GFP-containing axons exhibited varicosities at regular intervals, which were potentially en passant synaptic boutons (Fig. 2B).
Many of the GFP-expressing neurons had dendritic and axonal morphologies comparable to previously described putative or confirmed GABAergic neurons of the cerebral cortex. For example, upper layers multipolar or bitufted neurons with large cell bodies (>20 μm in diameter) and with axons giving rise to horizontally directed branches (Fig. 1B, leftmost neuron; Fig. 2C,D), resembled Golgi-stained large basket cells in monkey somatosensory cortex [type 1 of Jones (Jones 1975)] and parvalbumin-containing, axosomatic ‘fast spiking’ cells in rat frontal cortex [Fig. 6A in Kawaguchi and Kubota (Kawaguchi and Kubota, 1998)]. Among neurons in our sample with less common morphologies were lower layer VI cells such as the fan-shaped neuron of Figure 1A, which was similar to Golgi-impregnated neurons described by Ferrer et al. (their fig. 3C) in the rat sensorimotor cortex (Ferrer et al. 1986), and the inverted pyramidals in Figure 1C (upper two neurons), which were comparable with the GAD-immunostained neurons described by Prieto et al. (their figs 13:4 and 14:4) in cat auditory cortex (Prieto et al., 1994), and to the somatostatin-immunopositive neurons in rabbit visual cortex described by Ramon y Cajal-Agueras et al. — see their figures 5 and 7C (Ramon y Cajal-Agueras et al., 1989). The multipolar GFP-expressing neuron in layer I (Fig. 2D), with a horizontal axonal spread in layers I and II/III, resembled some intracellularly stained layer I neurons in the rat (Zhou and Hablitz, 1996) (their fig. 13).
Computer-assisted reconstruction of neurons with well-visualized axonal arbors (Fig. 3) allowed comparison with additional recognized classes of non-pyramidal cortical neurons. Among upper-layers cells expressing GFP were small neurons with mostly ascending dendrites and many horizontal axonal branches (Fig. 3A), similar to small basket cells in monkey somatosensory cortex [type 6 of Jones (Jones, 1975); compare also with Prieto et al.'s fig. 8:3 (Prieto et al., 1994), and with Meyer's fig. 2C (Meyer, 1983)]. Other neurons in the upper layers had initially ascending axons with downward recurving ‘arcades’ giving rise to a loose bundle of parallel, vertically oriented descending axonal branches below the level of the cell body (Fig. 3B–D), similar to type 2 of Jones (Jones, 1975) [compare also with the ‘regular spiking non-pyramidal’ neurons in Kawaguchi and Kubota's figs 3:B4 and 8:7,8 (Kawaguchi and Kubota, 1998)]. GFP-expressing neurons in the deep layers often had ascending axons (Fig. 3E,F,G), and of these some had axonal tufts ascending to and extending in layer I, typical of Martinotti cells [Fig. 3G; compare with Wahle's fig. 1 (Wahle, 1993), with Kawaguchi and Kubota's fig. 2 (Kawaguchi and Kubota, 1996) and with Klostermann and Wahle's fig. 11A (Klostermann and Wahle, 1999)].
Occasionally observed in our slices, but not included in the present report, were neurogliaform cells [type 5 of Jones (Jones, 1975)] and bipolar neurons, while apparently missing from our sample were axo-axonic (‘chandelier’) cells [type 4 of Jones (Jones, 1975)], possibly because of the relative immaturity of our cultures, since the characteristic chandelier-shaped terminals appear late in development (Meyer and Ferres-Torres, 1984). We were also not able to identify so-called double bouquet cells [(type 3 of Jones (Jones, 1975)], which to our knowledge have not been clearly documented in rodents. Very rarely (four cells, not shown), neurons with classical pyramidal morphology and spiny dendrites were observed to express GFP, possibly reflecting a low-level expression of GAD mRNA in pyramidal neurons (Cao et al., 1996). Finally, we consistently observed a small number per slice of GFP expressing cells with astroglia-like morphology; these were most often found near the pial or white matter boundaries (not shown).
Quantitative Analysis of GFP-expressing Multipolar Cortical Neurons
For quantitative analysis, 150 multipolar neurons were selected, solely based on their being brightly fluorescent and well separated from other neurons in their vicinity. (We defined as ‘multipolar’ any non-pyramidal neuron with three or more primary dendrites, a definition that may include neurons traditionally classified as bitufted.) Initial analysis included computation of the following morphological features: soma cross-sectional area, number of primary dendrites, number of dendritic branch points, total dendritic length, vertical and horizontal extent of the dendritic tree and total area covered by the dendritic tree (see Materials and Methods for definitions). Figure 4 shows a matrix of pairwise scatterplots of the main four parameters — soma cross-sectional area, number of dendritic ends (primary dendrites + branch points), total dendritic length and area of dendritic coverage. All plots showed some degree of linear correlation, but the strongest correlations were found between total dendritic length and the three other parameters (Fig. 4, bottom row), with correlation coefficients between 0.57 and 0.70. Thus, total dendritic length increased nearly linearly with soma cross-sectional area, with number of dendritic end points and with dendritic coverage area.
Directional Dendritic Outgrowth in Multipolar Cortical Neurons
To determine if dendrites of multipolar neurons have a preferred direction of growth relative to the vertical axis, we calculated for each neuron an Index of Polarization (IOP, see Methods). A bitufted neuron with all dendrites oriented directly towards or away from the pia will have an IOP very close to 1, while a horizontal neuron with all dendrites growing parallel to the pia will have an IOP close to –1.75. A multipolar neuron with dendrites distributed in purely random orientations relative to the vertical axis will have, on average, an IOP of 0. Figure 5A (solid line) shows a smoothed histogram of all IOP values in our sample of 150 GFP-expressing multipolar neurons. The histogram was multimodal, with the lowest major mode centered on 0, suggesting random orientations, and several additional peaks at more positive values, suggesting vertically biased orientations. As expected, the computed IOP values were correlated linearly with the measured aspect ratios of the dendritic trees (r = 0.50, P < 0.0001, data not shown).
To determine whether the IOP values we calculated were statistically different from those of randomly oriented dendrites, we conducted a Monte-Carlo simulation of 10 000 neurons, each consisting of 1000 randomly oriented dendritic segments, and calculated their IOP values. The histogram of all simulated IOP values (Fig. 5A, dashed line) had a peak at zero, and declined steeply on both sides of zero, so there were virtually no simulated neurons with IOP values >0.1, and <5% of the simulated neurons had IOP values >0.05 (in other words, P < 0.05 for IOP > 0.05). Since the IOP of a typical real neuron in our sample was calculated from 1500–2000 dendritic segments, and since the spread in the IOP of randomly oriented dendrites should decrease with the number of dendritic segments, the simulation results represented an upper limit on the possible statistical spread of IOP values in our sample of real neurons, if they were all taken from a population with purely random dendritic trees. Figure 5A demonstrates that the lowest major mode in the histogram of the real neurons (solid line) overlapped precisely the histogram of the simulated neurons when scaled to the same height, indicating that dendritic trees of this group of neurons were not significantly different than random. The majority of neurons, however, had IOP values considerably higher than could be explained by statistical spread, indicating a highly significant non-random bias of their dendrites towards the vertical axis.
Based on the histogram in Figure 5A, we classified all the neurons in our sample as either non-polarized (NP), moderately polarized (MP) or highly polarized (HP), with the cut-off IOP values between these three groups being 0.05 and 0.3, respectively. A small number of neurons had significantly non-random negative IOPs (<–0.05), and these were designated tangentially polarized (TP). About half of the neurons in our sample were classified as MP, while NP and HP comprised ~20% each, and ~10% were classified as TP (Fig. 5B). Thus, ~70% of the neurons in our sample had dendritic trees with varying degrees of vertical bias, but a well-separated subset, ~20% of the total, had randomly oriented dendritic trees.
Figure 6 shows representative dendritic tracings of neurons from the four groups defined above. The HP group included many neurons that would otherwise be classified as bitufted, but some of the HP neurons, and most of the MP neurons, would fall under the multipolar category. Some of the neurons in the MP group (e.g. the leftmost neuron) had dendritic trees that would not be considered polarized had they been classified by more traditional metrics such as dendritic aspect ratio, illustrating the increased sensitivity of our method of analysis.
We describe here a novel method for visualizing the detailed morphology of non-pyramidal, presumed GABAergic neurons in living neocortical slice cultures by GAD67–GFP transfection. Using a novel computational metric of dendritic polarization, we demonstrated that dendrites of ~70% of the medium and large GAD67–GFP expressing neurons are not randomly oriented, but instead show a highly significant tendency to extend along the vertical axis of the cortex.
Novel Method for Visualizing GABAergic Neuronal Morphology in Living Slice Cultures
In this study we used particle-mediated transfection with the eGFP cDNA linked to the GAD67 gene's 5′ regulatory region, to visualize GABAergic neurons in living cortical cultures. A similar DNA construct, using a much shorter segment of the GAD67 promoter, was inserted as a transgene into transgenic mice, but GFP expression in these animals was reported to be restricted to a small subset of all GABAergic neurons (Oliva et al., 2000). Our construct appeared to label a wide variety of non-pyramidal neurons, but virtually never pyramidal neurons. With the exception of spiny stellate neurons in layer IV, which were also never observed in our cultures, all non-pyramidal neurons in the cortex are GABAergic (Houser et al., 1983; Prieto et al., 1994), so there is little doubt that our construct labeled specifically GABAergic neurons, but with no apparent subtype specificity (see Results and below in the Discussion for morphological identification of the neurons in our sample vis-à-vis accepted classifications of GABAergic cortical neurons).
In its level of morphological detail, GFP transfection appears to equal or rival the Golgi and intracellular staining methods, revealing even fine dendritic and axonal processes. Like the Golgi method, gene-gun mediated transfection labels neurons in a spatially random and scattered pattern, which in favorable cases allows one to follow dendrites and even axons of isolated neurons for their full length. Unlike the Golgi technique, however, our transfection method is cell-type specific, in that it labels selectively GABAergic neurons, and allows visualization of neuronal morphology in the living state. These features of the method have important potential applications. For example, this method could be used to follow changes in morphology of live neurons during development in vitro, and to test whether such development is dependent on neuronal activity or on various growth factors. Another application would be electrophysiological recordings of synaptic responses in postsynaptic neurons, upon minimal stimulation of GFP-tagged axons (Jin et al., 2000). Finally, future studies could take advantage of the well-characterized activity-dependent regulation of GAD expression in the cortex (Hendry and Jones, 1988; Akhtar and Land, 1991; Martin and Rimvall, 1993; Benson et al., 1994) and use dynamic changes in GAD67–GFP expression to identify activation patterns in the cortex at the single-cell level.
Novel Metric for Dendritic Polarization
The observation that dendrites of multipolar cortical neurons have a vertical orientation bias was made by numerous previous studies (Jones, 1975; McMullen and Glaser, 1982; Winer, 1984a; DeFelipe et al., 1986a; Lund et al., 1988; McMullen et al., 1988a), but has rarely been examined quantitatively [a notable exception is a study by McMullen et al. (McMullen et al., 1984)]. Polarization of dendritic trees has been quantified previously using a variety of metrics [see Uylings et al.'s review (Uylings et al., 1986)], most of which are adequate for describing and comparing the spatial distribution of dendritic branches, but provide little insight into the mechanisms guiding dendritic growth. In this study, we followed a small number of previous studies (Glaser et al., 1979; McMullen et al., 1984) by measuring the local orientation of each dendritic chord in the brain coordinate system, rather than in a coordinate system centered on its parent cell body. Assuming that the trajectory of the mature dendrite closely resembles the trajectory taken by the dendritic growth cone during development, such analysis can provide important clues on the factors influencing dendritic growth. We expanded upon previous studies by employing a novel metric, the WACAV (or its normalized equivalent, the IOP), as a single quantity indicating the degree of directional bias in the dendritic tree of a given neuron. This novel approach allowed us to distinguish clearly between vertically biased and randomly oriented dendritic trees, the latter having IOP values scattered closely around zero. A major advantage of this method of analysis is that it applies equally well to the two-dimensional projection of neurons growing in three-dimensional space, and to neurons confined to the nearly planar organotypic slice (see Appendix).
Three potential caveats need to be considered before accepting our interpretation. The first is that these data were acquired from neurons whose morphological development occurred mostly in cultured slices, since at the age of culturing (first postnatal week) non-pyramidal neurons are only poorly developed morphologically (Parnavelas and Uylings, 1980; Miller, 1986; McMullen et al., 1988a). This implies that the neurons in our sample developed, to a large degree, without sensory input, and may therefore have abnormal properties. Previous studies have documented similarities in many of the morphological and electrophysiological properties of neurons that developed in organotypic slice cultures to those of neurons that developed in vivo, suggesting that proper development of these properties does not require intact sensory input (Bolz et al., 1990, 1992; Behan et al., 1991; Annis et al., 1993; Novak and Bolz, 1993; Massengill et al., 1997; Sieg et al., 1998). Similar conclusions were reached in a recent study that focused specifically on morphological and physiological properties of inhibitory cortical interneurons (Klostermann and Wahle, 1999). On the other hand, dendritic orientation does depend on afferent input (Pinto Lord and Caviness, 1979). A classic example of this dependence is the morphology of both excitatory and inhibitory layer IV neurons in rodent somatosensory cortex, which display strong dendritic (and to a lesser degree axonal) polarization towards the center of their ‘barrel’ (Woolsey and Van der Loos, 1970; Lorente de Nó, 1992). Development of this polarity requires intact thalamocortical afferents during a ‘critical period’ in early development (Steffen and Van der Loos, 1980; Harris and Woolsey, 1981). Similar (though less dramatic) dependence of dendritic orientation on patterned sensory input has been described in the visual cortex (Borges and Berry, 1978; Coleman et al., 1981; Tieman and Hirsch, 1982; Kossel et al., 1995), and elimination of all auditory input during early development causes a shift of the dendritic orientation of non-pyramidal neurons in auditory cortex towards the horizontal (tangential) plane (McMullen et al., 1988b). Thus, vertical orientation in our cultures may have been reduced by lack of sensory input, compared with vertical orientation in vivo, and our estimate of vertical dendritic bias in vitro should be regarded as a lower estimate of vertical bias in the intact brain.
A second caveat is that biolistically transfecting neurons with GFP may have altered their normal development and shape. Preliminary studies (X. Jin, P.H. Mathers and A. Agmon, unpublished observations) have revealed no apparent differences in electrophysiological and morphological properties between GAD67–GFP expressing and non-expressing GABAergic neurons in cultured slices. In addition, previous studies of pyramidal cells biolistically transfected with GFP in vitro (Lo et al., 1994; McAllister et al., 1995) have revealed seemingly normal morphologies. We therefore believe that biolistic transfection and GAD67–GFP expression did not alter the morphology of non-pyramidal cells.
A third concern is that the dendritic orientation of the neurons in our sample was altered by the slicing procedure, since some dendrites would be truncated during the slicing. However the effect of truncation on our cultured slices was most likely negligible, since the cultures were prepared from animals less than 1 week old, in which the diameter (in the horizontal plane) of the dendritic tree of non-pyramidal cells is less than 100 μm (Parnavelas and Uylings, 1980; McMullen et al., 1988a). Therefore, only neurons with cell bodies 50 μm or less from either cut surface, or less than 30% of all neurons in a 350 μm thick slice, suffered any truncation at all, and these would have been truncated on one side only and most of them by a negligible amount. Moreover, it is quite likely that the more severely truncated neurons did not survive in the long-term culture, leaving only intact or mildly truncated neurons. Most importantly, truncation had absolutely no effect on the calculated IOP, since the truncation occurred in a plane parallel to the plane of the projection, and therefore both dendrites oriented vertically and dendrites oriented horizontally were truncated to the same extent.
Classification Schemes of Non-pyramidal Neocortical Neurons and Dendritic Orientation
Neocortical non-pyramidal, GABAergic neurons have been classified in a multitude of ways, based on a variety of morphological, electrophysiological and immunocytochemical criteria. Although morphological cell types can be defined on the basis of cell body shape (Cobas et al., 1987) or dendritic arbors (Feldman and Peters, 1978; Peters and Regidor, 1981; Prieto et al., 1994), most modern classification schemes follow the lead of Cajal (DeFelipe and Jones, 1988) in classifying non-pyramidal neurons by relying heavily on their axonal arborization patterns (Jones, 1975; Szentagothai, 1978; Meyer, 1983). Axonal arborization patterns are predictive of the type of target favored by each inhibitory neuron (Somogyi et al., 1998), and most classifications now recognize at least three major types: basket cells, favoring somata and proximal dendritic shafts (Somogyi et al., 1983; DeFelipe et al., 1986a); double bouquet cells (described most often in primates and carnivores, rarely in rodents), favoring distal dendritic shafts and spines (Somogyi and Cowey, 1981; DeFelipe and Fairén, 1988); and chandelier cells, making synapses exclusively on initial segments of axons (Fairén and Valverde, 1980; Peters et al., 1982; Somogyi et al., 1982). These classes are also distinguishable by molecular markers: basket and chandelier cells most often contain parvalbumin, while double bouquet cells contain calbindin (DeFelipe et al., 1989; Hendry et al., 1989; DeFelipe et al., 1990; Williams et al., 1992; Peters and Sethares, 1997; Kawaguchi and Kubota, 1998). Falling outside this classification are several classes of neuropeptide-containing cortical interneurons, e.g. Martinotti cells, characterized by an ascending axon forming an extended arbor in layer I, and containing somatostatin, Substance P and/or Neuropeptide Y (Kuljis and Rakic, 1989; Wahle, 1993; Kawaguchi and Kubota, 1996), and bipolar cells, containing vasoactive intestinal peptide (Connor and Peters, 1984; Morrison et al., 1984; Bayraktar et al., 2000; Cauli et al., 2000). Finally, some morphological and/or immunocytochemical types also show characteristic firing patterns (Kawaguchi and Kubota, 1997). A definitive identification of a non-pyramidal neuron as belonging to one of these classes requires immunocytochemical identification, electron microscopical examination of its postsynaptic targets and/or electrophysiological characterization, in addition to knowledge of its dendritic and axonal morphologies, and was therefore outside the scope of the present study. Nevertheless, based on dendritic and axonal morphology, our sample of neurons expressing GAD67–GFP appeared to include representatives of many of the putative or confirmed GABAergic cell classes observed previously with the Golgi and intracellular staining methods, including some types with specific immunocytochemical identities. For example, our sample contained representatives of at least four of the six types of smooth non-pyramidal cells described in monkey somatosensory cortex by Jones (Jones, 1975) (see Results for specific comparisons with previous studies). Our characterization of GAD67–GFP expressing neurons according to the level of vertical bias in their dendritic tree is not intended to add yet another system of classification, and we have not attempted to equate a specific level of vertical orientation bias with specific classes of nonpyramidal cells. Indeed, it appears from our own and previous studies that a vertical dendritic orientation bias is a property superimposed, to varying degrees, on all classes of nonpyramidal GABAergic neurons.
What Dendritic Measurements Tell Us About Dendritic Development
Our analysis shows that dendritic length of non-pyramidal, putative GABAergic neurons is strongly correlated with soma cross-sectional area, with number of dendritic ends and with total dendritic coverage area (Fig. 4), confirming and expanding previous findings (McMullen et al., 1984). The first of these correlations, the correlation between soma size and dendritic length, most likely reflects the need for a larger cell body to support a larger dendrite. The other correlations may be rooted in basic rules of dendritic development. For example, the linear relationship between number of end points and dendritic length suggests that the average length that a dendrite grows between branch points is similar between different neurons, implying a common underlying mechanism involved in determining dendritic bifurcation. The linear relationship between total dendritic length and dendritic coverage area suggests that dendrites branch so that a unit of cortical area (in the nearly planar organotypic slice) is always sampled by the same total dendritic length. The three-dimensional analogue of this last correlation remains to be investigated.
Possible Mechanisms of Vertically Biased Dendritic Outgrowth
How dendrites acquire a vertical orientation bias is not known. An attractive hypothesis accounting for vertically biased growth is that developing neurons sense a pia-to-white matter gradient of a diffusible or surface-attached molecule. If a single extracellular gradient is to account for extension of processes both towards and away from the pia, one needs to assume that the receptive or signal transducing mechanism in the neuron is asymmetrically distributed, causing opposite poles of the neuron to extend neurites in opposite directions, in response to the same gradient. Evidence for precisely such a mechanism has recently been reported, indicating that the same diffusible morphogen, Semaphorin 3A, can simultaneously behave as an attractant for apical dendrites of cortical pyramidal neurons and as a repellent for their axons, by virtue of an asymmetric distribution of the enzyme guanylate cyclase (Polleux et al., 2000). The organotypic slice preparation could potentially be used to test such models. Interestingly, our analysis demonstrates that multipolar neurons did not form a continuum with respect to their Index of Polarization; rather, there was a well-separated population (NP cells, ~20% of neurons in our sample) with a statistical spread of IOP values around 0, indicating purely random dendritic orientations. An intriguing and testable possibility is that this population lacked the molecular machinery to detect or to transduce the postulated extracellular gradient.
Functional Significance of Directionally Biased Dendritic Trees
Our results imply that a vertical orientation bias, which is characteristic of dendrites of pyramidal cortical cells, is also a feature of non-pyramidal neurons — not only of the bipolar and bitufted types, but also of those with ‘stellate’ or multipolar morphologies. What advantage(s) may be conferred on the information processing capabilities of the neocortex by vertically elongated dendritic trees? The major afferents to primary sensory and motor cortical areas are thalamocortical axons, which are organized along the horizontal (tangential) dimensions in a precise topological order (Hohl-Abrahao and Creutzfeldt, 1991; Agmon et al., 1995), and terminate in narrow vertical (radial) columns and in specific laminar patterns (LeVay and Gilbert, 1976; Jensen and Killackey, 1987; Landry et al., 1987; Agmon et al., 1993). A columnar pattern of terminations, although with a fractured topology and with a different laminar distribution, is also formed by long-range corticocortical axons (Jones et al., 1975; Goldman and Nauta, 1977; Code and Winer, 1986; DeFelipe et al., 1986b; Chapin et al., 1987; Ojima et al., 1991; Saleem et al., 1993; Pucak et al., 1996). In turn, both excitatory and inhibitory cortical neurons of all layers extend much of their local axonal projections along the vertical dimension (Lorente de Nó, 1949; Somogyi et al., 1981; Bernardo et al., 1990; Kritzer and Goldman-Rakic, 1995; Fujita and Fujita, 1996; Zhang and Deschenes, 1997; Lubke et al., 2000; Porter et al., 2001), again with specific laminar distributions (Lund and Boothe, 1975). These organizational principles ensure that afferent information pertaining to specific features of sensory space will be distributed and processed within relatively narrow radial modules, whether in primary (Hubel and Wiesel, 1977; Mountcastle, 1997) or in high-order (Fujita et al., 1992; Britten, 1998) cortical areas. By extending their dendritic trees vertically, pyramidal and non-pyramidal cortical neurons can potentially sample and integrate afferent and reentrant inputs terminating at different laminar positions and representing different stages of information processing, while retaining the specificity of location, modality and/or features unique to their own module. A similar organizational rule, manifested as dendritic arbors elongated in parallel to a topologically ordered system of afferent fibers and orthogonal to the plane of the mapping, is found also in other brain areas, such as the dorsal cochlear nucleus (Blackstad et al., 1984; Berrebi and Mugnaini, 1991), the inferior and superior colliculi (Rockel and Jones, 1973; Langer and Lund, 1974), the reticular thalamic nucleus (Montero et al., 1977; Liu et al., 1995; Crabtree, 1996) and the cerebellar cortex (Voogd and Bigare, 1980; Mulle et al., 1987; Wiklund et al., 1990).
This study was supported by National Institutes of Health grants HD33463 (A.A.) and EY12152 (P.H.M.) and by OTKA grant T-029369 (G.S.). We thank Drs Albert Berrebi, Paul Brown and Ron Millecchia for critical comments on the manuscript, Dr Arie Gingold for helpful discussions on mathematical methods, and Cary Johnson, Colette Ramsburg and Jubin Ryu for excellent technical support.
Address correspondence to A. Agmon at Department of Neurobiology and Anatomy, West Virginia University, Morgantown, WV 26506, USA. Email: email@example.com.
As explained in Materials and Methods, to calculate the WACAV one takes the sum of the lengths of the projections of all dendritic segments on the vertical axis, and divides it by the sum of all dendritic lengths. The calculation can be done either in two dimensions (using two-dimensional coordinates) or, if a suitable dataset is available, in three dimensions. We show here that the two-dimensional and the three-dimensional WACAV values will be different, but that when calculated in two dimensions, the WACAV of neurons with randomly oriented dendrites is 0.64, regardless of whether the neurons have developed in a plane (e.g. in the organotypic slice) or in three-dimensional space (in vivo).
We start by calculating the average WACAV of a planar neuron with randomly oriented dendrites. We simulate the neuron by dividing it into a large number of small dendritic segments, and represent each segment as a unit vector forming a random angle ϑ with the vertical axis [this is equivalent to Glaser et al.'s ‘stick analysis’ (Glaser et al., 1979)]. Since the cosine is symmetric and since we consider the upward and downward directions equivalent, we can restrict the discussion to a quarter-circle between 0 and π/2. The normalized probability that a random angle will fall within the infinitesimal arc segment (ϑ, ϑ + dϑ) is the ratio of the length of the arc segment to the quarter circle, or (2/π)•dϑ. The average of the cosines of all angles (the WACAV of a randomly oriented dendritic tree) would therefore be 0.64, given by the integral
Often, however, only a two-dimensional dataset is available, describing the projection of the neuron on the plane of the drawing (for example, when tracing a neuron using a camera lucida device). In this case, calculating the WACAV in two dimensions would give the same sum of projections as in the three-dimensional case, since the projection of a vector on the vertical axis is not altered by first projecting it on a vertical plane, but the (apparent) lengths of the projected dendrites would be smaller than their real three-dimensional lengths. The length of the projection of a unit vector on the vertical (X–Z) plane is given by (cos2θ + sin2θ•cos2φ)1/2, so for a set of unit vectors randomly oriented in space, the average projected length of a vector is given by
Since this quantity appears in the denominator when calculating the WACAV, one needs to divide the three-dimensional WACAV (which is 0.5) by π/4, to account for the distortion due to the projection. The WACAV of a neuron with randomly oriented dendrites would therefore be, on average, 2/π, the same as in the planar case.