Cerebral Cortex Advance Access published online on July 26, 2007
Cerebral Cortex, doi:10.1093/cercor/bhm126
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Subcolumnar Dendritic and Axonal Organization of Spiny Stellate and Star Pyramid Neurons within a Barrel in Rat Somatosensory Cortex
Department of Cell Physiology, Max-Planck-Institute for Medical Research, Jahnstr. 29, 69120 Heidelberg, Germany, 1 Present address: Institut für Physiologie der Ludwig-Maximilians-Universität München, Pettenkoferstr. 12, 80336 München, Germany, 2 Present address: Institute for Physiology, Bern University, Bühlplatz 5, 3012 Bern, Switzerland
Address correspondence to Dr Veronica Egger, Institut für Physiologie der LMU, Pettenkoferstr. 12, 80336 München, Germany. Email: V.Egger{at}lmu.de.
 |
Abstract |
|---|
 TopNotes Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Excitatory neurons at the level of cortical layer 4 in the rodent somatosensory barrel field often display a strong eccentricity in comparison with layer 4 neurons in other cortical regions. In rat, dendritic symmetry of the 2 main excitatory neuronal classes, spiny stellate and star pyramid neurons (SSNs and SPNs), was quantified by an asymmetry index, the dendrite-free angle. We carefully measured shrinkage and analyzed its influence on morphological parameters. SSNs had mostly eccentric morphology, whereas SPNs were nearly radially symmetric. Most asymmetric neurons were located near the barrel border. The axonal projections, analyzed at the level of layer 4, were mostly restricted to a single barrel except for those of 3 interbarrel projection neurons. Comparing voxel representations of dendrites and axon collaterals of the same neuron revealed a close overlap of dendritic and axonal fields, more pronounced in SSNs versus SPNs and considerably stronger in spiny L4 neurons versus extragranular pyramidal cells. These observations suggest that within a barrel dendrites and axons of individual excitatory cells are organized in subcolumns that may confer receptive field properties such as directional selectivity to higher layers, whereas the interbarrel projections challenge our view of barrels as completely independent processors of thalamic input.
Key Words: cortical column • dendritic symmetry • interbarrel projection • intrabarrel confinement • minicolumn • shrinkage
|
Introduction |
|---|
|
TopNotesAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
In several primary sensory input regions, dendritic arbors exhibit a marked asymmetry, which appears to be closely linked to cortical columns. Dendritic asymmetry is observed most frequently within layer (L) 4, the main target region of thalamocortical afferents, and L2/3, the major subsequent station along the sensory pathway. In visual cortex, several cases of dendritic asymmetry related to borders of cortical columns have been reported (Katz et al. 1989
; Hübener and Bolz 1992; Kossel et al. 1995; Elston and Rosa 1997). In the barrel field of rodents, a distinct asymmetric orientation of dendrites away from the barrel borders has been observed, as illustrated in Figure 1; neurons located at barrel centers extend their dendrites in all directions (Lorente de Nò 1922; Woolsey et al. 1975; Steffen and Van der Loos 1980; Simons and Woolsey 1984; Lübke et al. 2000; Staiger et al. 2004). Thus, barrel neurons are apparently subject to intrabarrel confinement. However, all these observations in the barrel cortex were qualitative in nature; here, we provide a quantitative description.
|
The 2 main excitatory neuronal classes within L4 of somatosensory and visual cortices are spiny stellate neurons (SSNs) and star pyramid neurons (SPNs; Jones 1975
; Lund 1984; Martin and Whitteridge 1984; Simons and Woolsey 1984; Hirsch 1995). Their morphological features are similar, except for the apical dendrite of SPNs with a tapered thick trunk that is absent in SSNs. By and large, the similarity also extends to their axonal projections (Lund 1984; Lübke et al. 2000; Staiger et al. 2004). In the barrel cortex, however, SPNs and SSNs appear to differ in dendritic symmetry (Simons and Woolsey 1984; Lübke et al. 2000). Functionally, barrel cortex SSNs and SPNs have been shown to be integrated into different cortical circuits (Schubert et al. 2003), raising the question whether there is an anatomical segregation of SSNs and SPNs within barrels.
Recent work has reinforced the notion of subcolumnar structures or minicolumns (Mountcastle 1978) within a barrel at a functional level: Bruno et al. (2003) have demonstrated the existence of angular tuning domains within a barrel, incorporating neurons that respond preferentially to the same angle of whisker deflection, whereas Andermann and Moore (2006) reported that these domains are arranged in L2/3 such that the direction preference of a barrel neuron is correlated with the somatotopic map. A possible anatomical correlate of minicolumns has been revealed by cytochrome oxidase stainings, although these subbarrels are rather large structures (Land and Simons 1985; Land and Erickson 2005). Moreover, it was observed that axons of spiny L4 neurons are also confined to the respective home barrels at the level of L4, giving rise to the notion of a mostly isolated, intrabarrel excitatory network (Woolsey et al. 1975; Lübke et al. 2000; Petersen and Sakmann 2000, 2001; Brecht and Sakmann 2002a; but see Staiger et al. 2004 for a less pronounced confinement of SPN axons). If functional subbarrel columns would indeed exist, one would predict an even more localized confinement of axons. So the final aim of our analysis was to assess the overlap of axonal and dendritic fields of individual excitatory barrel neurons within the barrel.
These tasks involve several methodological issues. The aforementioned studies on barrel neuron symmetry relied on tangential slices, which allow to relate the position of a neuron within a barrel to the symmetry of dendrites and axon collaterals. The characteristic apical dendrites are, however, often cut or difficult to identify, rendering the classification into SSN and SPN impossible. Thus, we used reconstructions from cells filled in acute thalamocortical slices and from cells filled in vivo to establish symmetry properties of the 2 cell classes in conjunction with their position within the barrel. Next, the bias indices used in the studies of visual cortex neurons are inappropriate to estimate symmetry of barrel neurons in somatosensory cortex: a bias index is the ratio of the dendritic densities within 2 regions of interest. In many barrel neurons, there are simply no dendrites opposing the main dendritic field orientation and thus more meaningful parameters are required. Finally, in preparations made from acute brain slices there is considerable shrinkage perpendicular to the slice surface, distorting morphological measurements. To properly quantify shrinkage of neurons in our preparation, we compared neuronal morphologies as established with 2-photon microscopy in acute slices to the same biocytin-labeled structures postembedding. The measured degree of shrinkage was then used to estimate its influence on the critical morphological parameters. If applicable, these tools may also be put to use in other cortical areas and central nervous system preparations.
|
Materials and Methods |
|---|
|
TopNotesAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Slice Preparation and Filling of Neurons
Thalamocortical slices and tangential slices (350–400 µm; Bernardo and Woolsey 1987; Agmon and Connors 1991; Feldmeyer et al. 1999; Fleidervish et al. 1998) were prepared from the somatosensory cortex of P12–14 old rats. Slices were placed in a recording chamber and viewed with an upright microscope fitted with 2.5x plan/0.075 NA and 40x water/0.80 NA objectives and an additional 4x magnifying lens. The barrel structure of L4 was readily detectable under phase contrast at low magnification. Spiny L4 barrel neurons were selected both on morphological criteria with the aid of infrared video microscopy (Dodt et al. 1998) and their regular spiking firing pattern following depolarizing current injection (Feldmeyer et al. 1999). Whole-cell voltage recordings of spiny L4 neurons were made at physiological temperatures (34–36 °C) using pipettes of 6–9 M
with an access resistance below 30 M. The extracellular solution contained (in mM) 125 NaCl, 2.5 KCl, 25 glucose, 25 NaHCO3, 1.25 NaH2PO4, 2 CaCl2, 1 MgCl2 bubbled with 95% O2 and 5% CO2; the pipette solution contained 105 K-gluconate, 30 KCl, 10 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid, 10 phosphocreatine, 4 adenosine triphosphate-Mg, 0.3 guanosine triphosphate, adjusted to pH 7.3 with KOH. The osmolarity of this solution was 300 mOsm. Biocytin (2 mg/ml; Sigma, München, Germany) was added to the pipette filling solution for subsequent staining and further identification of the cell type. We also used 19 neurons of P24–35 (mean P29) old rats that were similarly filled in vivo and subsequently recovered from tangential sections. These neurons were published previously (Bruno and Sakmann 2006). See their Supplementary Information for methods.
Histological Procedures
After recording, slices were fixed at 4 °C for at least 16 h in fixative containing 0.1 M phosphate buffer (PB), 2–4% paraformaldehyde and in some cases an additional 1% glutaraldehyde. If slices were stored for longer before processing, the fixative was exchanged with 0.1 M PB after a few days. Biocytin stains were developed as described elsewhere (Horikawa and Armstrong 1988; Lübke et al. 2000). Finally, slices were embedded in Mowiol (Hoechst, Frankfurt, Germany) on a slide mount and coverslipped. Several slices containing biocytin-labeled neurons were double stained for cytochrome oxidase to visualize the position of spiny L4 neuron somata relative to barrel borders, including the in vivo filled cells. A modified version of the classical protocol (Wong-Riley 1979; Wong-Riley and Welt 1980) was used prior to biocytin staining, as described in Feldmeyer et al. (1999). A result of such a double stain of a tangential slice is shown in Figure 1C. Moreover, barrel borders can be distinguished to some extent in slices stained normally for biocytin just before the final embedding, although in thalamocortical slices borders may be blurred. The schematic barrel borders indicated in some of the figures were established this way.
Selection of Neurons and Morphological Analysis
Adequately stained small spiny L4 neurons in thalamocortical slices without obviously truncated dendrites were selected and classified according to the following criteria (Lübke et al. 2000): If there was a single, thick apical dendrite (trunk diameter > 2 µm) extending to at least lower L2/3 and with a tapered trunk such that the shape of the soma was slightly triangular (see Fig. 2B), the cell was classified as SPN, otherwise as SSN. The third class of excitatory L4 cells, that is, pyramidal cells that are characterized by a skirt of basal dendrites (Jones 1975; Staiger et al. 2004), was not present in our sample, perhaps because we did not label cells near the lower border of L4. For reconstruction and analysis, all SPNs within this sample were used. SSNs were selected at random from the remainder to obtain equal numbers for each class, because they are more numerous within the population (Simons and Woolsey 1984; Feldmeyer et al. 1999). For reconstruction, neurons were magnified with a 40x or 100x oil/1.3 NA objective fitted to a video microscope. The NEUROLUCIDA system (MicroBrightField, Inc., Colchester, VT) allows a 3-D reconstruction of neuronal structures except for the 2-D representation of the soma. For the computer-aided part of asymmetry and overlap analysis mentioned below, the reconstructed structures were converted to NEURON-readable format. Within NEURON (Hines and Carnevale 1997), 3-D voxel representations of the dendritic and axonal arbors of neurons (see Hellwig 2000) were generated with custom-written software. The origin of the coordinate system was placed to the soma center; a set of voxel representations with different voxel side lengths (1 µm, 5, 10, 15 ... 50 µm) was computed. If the distance between 2 subsequent NEUROLUCIDA data points on the same branch was larger than half the chosen voxel side length, additional data points and thus eventually voxels were interpolated. The voxel representation did not account for arbor diameters.
|
Measuring Asymmetry of Dendritic Trees
To estimate the extent of neuronal asymmetry due to the confinement of the dendritic tree to a barrel, 3-D reconstructions of neurons from thalamocortical slices were rotated with the aid of NEUROEXPLORER (MicroBrightField Inc., Colchester), such that their dendritic tree was projected in the plane parallel to the pia (i.e., the cortical surface), in other words into the barrel field tangential section. This procedure is illustrated in Figure 3A. To achieve a proper rotation independent of the cutting angle of the slice preparation, we made use of the fact that the initial part of the axon of SSNs and also SPNs is directed vertically downwards toward the white matter in almost all cases (Woolsey et al. 1975; Lund 1984; Feldmeyer et al. 1999). All other axon collaterals were removed from the reconstruction to avoid confusion. Neurons were rotated until the remaining axon pointed vertically out of the plane of monitor, that is, the viewer looks at the barrel field from the point of view of the white matter. The task of measuring symmetry was thus reduced from 3 to the 2 relevant dimensions. Next, a circle with radius 30 µm was centered at the middle of the soma of the projected neuron. The dendrite-free angle 
was determined as the angular section outside that inner circle that contains no dendrites or at most a single branch without further arborization (Fig. 3B). The used size of the circle accounts for the dendrites that first leave the soma in the "wrong" direction and then turn toward the inner of the barrel. Its radius corresponds roughly to one fourth of the average horizontal dendritic field span of spiny neurons (see Table 1). This choice may appear arbitrary, but in the common morphometric Sholl analysis for evaluation of dendritic branching patterns inner circles with comparable radii of 20–40 µm are used (e.g., Sholl 1953; Hübener and Bolz 1992; Elston and Rosa 1997; Mizrahi et al. 2000). In this description, neurons that appear planar in fixed thalamocortical slices due to the shrinkage in the z-direction would have an -value close to 180°, for example, L2/3 pyramidal cells situated close to the slice surface. -values of truly asymmetric neurons should be significantly larger than that, whereas radially symmetric neurons from tangential slices would usually yield values smaller than 90° (see below).
|
|
As a more general index for asymmetry of neurons in thalamocortical slices we used the length of the vector between soma center and center of gravity (similar to the vector mapping in Tailby et al. 2005
), which was calculated using a fine voxel representation of the dendritic tree. This representation was generated as described above, with a voxel side length of 1 µm. Our aim was to measure symmetry with respect to the barrel cross section. Thus, prior to the distance calculation the center of mass was projected into the plane defined by the initial axon part (representing the normal vector), that is, into the projection plane used previously for the determination of . Neither dendritic diameters nor the soma volume was taken into account. Measuring Axonal Symmetry and Overlap between Dendritic and Axonal Fields within L4
To investigate the overlap between dendritic and axonal fields of neurons with a fully stained axonal tree, this procedure was then repeated with an axonal tree pruned such that it retained all collaterals within L4. In order to ensure proper rotation, the first projection described above was used as a template, because the initial part of the axon is often no longer clearly discernible within the tree structure. The axon-free angle was thus determined for the axonal tree reduced to L4. To estimate the overlap between dendritic and axonal tree of the same spiny barrel neuron, the fraction of volume taken up by both axonal and dendritic segments was determined with the aid of coarse voxel representations (see above). Upon comparison of representations with increasing voxel side length (5, 10, 15... µm), the appropriate voxel side length for this analysis appears to be in the range of 20–25 µm (see Results section). The overlap is given as fraction of dendritic voxels.
Coordinate Systems and Shrinkage Definition
Relative to the Slice
Here, the coordinates are x, y, z with the z-axis perpendicular to the slice plane, and the x- and y-axes within the slice plane. In Figure 2A the coordinate system is shown both for the thalamocortical and tangential slice preparation. Thus, the shrinkage effect perpendicular to the slice plane is called z-shrinkage, in short Sz. It is defined as shrunk length zs divided by original length z0 and thus equivalent to the linear factor known from anatomical studies (e.g., Hellwig 2000).
Relative to the Barrel Field
Horizontal/tangential and vertical are the directions relative to the barrel field, with vertical being perpendicular to the whole whisker map.
Measurement of Shrinkage
Shrinkage within the slice plane (xy-shrinkage, Sxy) and perpendicular to the slice plane (Sz) was estimated by comparing the morphology acquired from 2-photon fluorescence stacks of 4 spiny stellate cells filled with 50 µM Alexa 594 (Invitrogen, Carlsbad, CA) and biocytin in acute slices to the morphology after biocytin labeling and postembedding. Two-photon fluorescence image stacks were acquired with a confocal scanning unit (LFS SP2RS, Leica Microsystems, Mannheim, Germany) attached to an upright microscope (DMLFS; Leica) equipped with a 63x objective and a Ti:Sa-Laser (MIRA 900F, pumped by a 5W Verdi; Coherent, Santa Clara, CA) (Nevian and Sakmann 2006). Fluorescence stacks were analyzed using ImageJ.
Estimate of Shrinkage Distortion of the Dendrite- or Axon-Free Angle
Considerable z-shrinkage Sz may distort the symmetry measurements especially in thalamocortical slices. This is illustrated schematically in Figure 8A. As a first approximation, Sz is assumed to be homogenous within the slice. Because this problem is independent of the angle 
between a dendrite and either the x- or y-axis in the xy-plane, cylindrical coordinates are used with r2 = x2 + y2, and each dendritic tip can be described by the coordinates (z, r, ). As illustrated in Figure 8C, left panel, the r-coordinate of a dendrite with initial length L is conserved because there is no xy-shrinkage (see Results). But the z-coordinate shrinks from z0 to zs, causing an overall reduced length of the dendrite Ls. The amount of shrinkage depends on the relative orientation of the dendrite with respect to the z-axis, as shown for all possible orientations in Figure 8C, right panel.
 | (1) |
The influence of Sz on the angle 
between a dendrite and the slice plane normal (Fig. 8D, left) can be estimated by introducing the shrunk dendritic length Ls into the following expression for the angle (eq. 2). In the case of the overall dendritic orientation drawn in Figure 8A, both the dendrites that enclose the largest dendrite-free angle extend toward the slice surface. In the course of shrinkage these dendrites will be bent away from the slice plane normal (i.e., the z-axis), therefore twice the squeezing of will result in an upper limit for the change 
that could be induced by shrinkage (eq. 3). Figure 8D shows the estimated dependency of on the dendritic length, given our measurements of L in tangential slices and Sz (see Results).
 | (2) |
 | (3) |
In terms of shrinkage distortion the case underlying the above estimate represents the worst-case. In any other arrangement—one or 2 of the dendrites that include being directed away from the slice surface—the distortion would be smaller because these dendrites would be bent toward the dendrite-free section and therefore counteract the increase in . Thus, one may also argue that there is a bias in shrinkage distortion with respect to cell type: Radially symmetric neuron reconstructions probably suffer from a stronger increase in than asymmetric neurons because in symmetric neurons the probability of both outermost dendrites (that enclose ) being directed toward the surface is larger.
Measurement of Field Span and Caliber of the Uppermost Dendritic Trunk
The thick bars within the schematically drawn slices in Figure 2A denote the respective horizontal and vertical field spans. In the thalamocortical preparation the vertical field span corresponds to the extension of the dendritic field in the z-direction, and the horizontal field span to its spread in the plane spanned by the x- and y-axes. To determine the caliber of the trunk of the apical/uppermost dendrite, the diameter of the uppermost dendrite emerging from the soma was measured right at the branchpoint from the soma that was approximated to be of spherical shape. Figure 2B shows an example of a SPN and SSN uppermost dendrite and the trunk diameter measurement.
Measurement of the Position of a Neuron Relative to the Barrel
In tangential slices, the distance of neuronal somata from the next barrel wall was also measured and normalized by the barrel diameter. For every cell, the barrel diameter was measured by taking the wall-to-wall-distance that would run through the soma of the cell along the direction of rows within the barrel field. Neurons were classified as barrel wall neurons if their somata were found within the outer two fifths of a barrel, that is, had a fractional distance of less than 0.40 from the barrel border to the center; otherwise they were classified as barrel center neurons (see also Fig. 4E).
|
Mean values are given ±SD. A 2-tailed, unpaired Student's t-test was applied to test for significance levels.
|
Results |
|---|
|
TopNotesAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Dendritic Symmetry of L4 Neurons in Thalamocortical Slices
The spatial extension of the dendritic field of SSNs, for example, in the visual cortex has been characterized as spherical and radially symmetric, hence the name "stellate cell" (Lorente de Nò 1922; Lund 1984). In contrast to this symmetric distribution of dendrites within all directions in space, rat barrel cortex SSNs characteristically show an asymmetric, eccentric dendritic field (Fig. 1). This asymmetric orientation appears to be due to their confinement to single barrels (Woolsey et al. 1975; Steffen and Van der Loos 1980; Simons and Woolsey 1984). However, the second class of small spiny L4 barrel neurons, the SPNs, does not show a similarly strong deviation from radial symmetry.
The distinction between SSNs and SPNs was based on the presence of a thick apical dendrite (see Methods). A few neurons with relatively long uppermost dendrites lacked a tapered thick trunk and thus were classified as SSNs, although they may represent an intermediate form that belongs to the proposed continuum between the 2 cell classes (Jones 1975; Lund 1984). However, their removal from the sample did not significantly change the main results.
To measure neuronal asymmetry, we devised a method that is adapted to the morphological features of spiny L4 neurons within barrels (described in detail in the Methods section). Figure 3A shows the rotation of 3-D reconstructions of spiny L4 neurons from thalamocortical slices, such that their dendritic tree was projected in the plane parallel to the cortical surface. This projection is independent of the slice angle because the initial part of the axon of spiny L4 neurons projects vertically to the white matter (Lorente de Nò 1922; Woolsey et al. 1975; Lund 1984; Feldmeyer et al. 1999). For clarity, all other axonal collaterals were removed from the reconstruction. The task of quantifying symmetry was thus reduced from 3 to the 2 relevant dimensions. Projections of the 2 neuronal types are shown in Figure 3B,C. In the projection, the dendrite-free angle was determined as the angular section outside an "inner circle" drawn around the soma (radius 30 µm, Fig. 3B), which contains either no dendrites or just one single branch.
In addition, the vertical and horizontal field spans and the horizontal span of the longest dendrite were measured (see Methods, Fig. 2A,B). Finally, the center of gravity within the projection plane was determined.
Differences in Dendrite-Free Angle
The data measured from SSNs and SPNs in thalamocortical slices are summarized in Table 1. The measurement of the dendrite-free angle of SSNs confirms the impression of strong asymmetry: The dendrite-free angular section of 21 SSNs within the horizontal plane ranged from 183° to 306° (mean 245 ± 42°). Conversely, the eccentricity of the dendritic field measured for 22 SPNs is far less pronounced, their mean dendrite-free angular section being = 162 ± 26°, ranging from 80° to 199° (see Fig. 3C). This difference in asymmetry between the 2 cell classes is also reflected in the horizontal field span, with SPNs exceeding SSNs by 
40%, because in a radially symmetric neuron there are opposed dendrites adding up to the field span, that are absent in asymmetric neurons. Moreover, the difference in field span is not due to a difference in the maximal horizontal span of single dendrites as these values are similar.
The difference in dendrite-free angle and thus symmetry appear not to change with further development, because the difference between SSNs and SPNs in older animals is about 88° (around P29; n = 9 vs. n = 10; P < 0.005), similar to younger ones.
SPNs have slightly triangular somata of a size similar to that of the round somata of SSNs. Unlike most of the other dendrites, their apical dendrite has a tapered trunk and an initial part twice as thick as that of dendrites emerging from the upper part of SSNs (see Figs 2B, 7A). However, it has to be emphasized that this apical dendrite is much less pronounced than its counterparts of L5 or L2/3 pyramidal neurons because it lacks a tuft and does not extend to reach L1. Its average length exceeds that of the uppermost dendrite emerging from SSNs by only 25%.
Figure 4A,B summarizes the symmetry properties of both SSN and SPN dendrites in thalamocortical slices. The dendrite-free angle is plotted against the caliber of the initial segment of either apical dendrite (SPNs) or the dendrite closest to the vertical top of the soma in the case of SSNs and against the ratio of horizontal single dendrite to field span. Figure 4A shows the codependence of symmetry and the existence of a thick apical dendrite, which served as main criterion for classifying the neuron as SPN.
Taken together, an additional criterion for distinction between SSNs and SPNs in barrel cortex emerges: Whenever a spiny L4 barrel neuron has a strongly asymmetric dendritic field, it is probably a SSN, whereas radially symmetric fields are more characteristic for SPNs.
Dendritic Symmetry of Neurons in Tangential Slices from In Vitro and Vivo Fillings
An example of a tangential slice with 4 stained neurons located at the "corners" of a barrel is shown in Figure 1C,D. Neurons were classified as barrel wall or barrel center neurons according to the relative distance of their soma to the outer barrel border (see Methods). The mean dendrite-free angular section of 18 barrel wall neurons in tangential slices from both in vitro and in vivo fillings was 175 ± 54° whereas 16 barrel center neurons had a much smaller dendrite-free section of 79 ± 50° (P < 0.001; Fig. 4E). The considerably smaller dendrite-free sections in tangential versus thalamocortical slices (see Table 1) can be explained by the influence of shrinkage (see below). These data indicate that barrel wall neurons are preferably eccentric and barrel center neurons preferably radially symmetric, which corresponds to the earlier finding of intrabarrel confinement. Given our previous observation of symmetries of SSNs and SPNs in thalamocortical slices, one might conclude that this preference is also mirrored in the cell type, with SSNs located mostly at the barrel wall and SPNs mostly at the center. However, when looking at the completely reconstructed set of neurons that was recovered from in vivo fillings, where the presence of an apical dendrite allows for further classification, this notion does not hold for all neurons (Fig. 4E; see Discussion). Nevertheless, the barrel wall is mostly populated by asymmetric SSNs and the barrel center mainly by symmetric SPNs. The latter finding is also illustrated by Figure 7A, showing a cluster of 5 SPNs within the barrel center.
All asymmetric cells in our sample including thalamocortical and tangential slices and neurons reconstructed from in vivo fillings (n = 39) had their dendrites oriented away from the barrel border, in accordance with the previously reported intrabarrel confinement—with the exception of the SSN shown in Figure 4F. Perhaps this cell is confined to a subbarrel (Land and Erickson 2005).
Symmetry of Axons and their Overlap with the Dendritic Field in L4
Although axonal structures are difficult both to recover and resolve for reconstruction from acute tangential slices, thalamocortical slices allow for their reconstruction and subsequent analysis with respect to symmetry properties and overlap with dendritic structures. The axons of spiny L4 neurons typically were found to span all cortical layers, with their collaterals most dense in L4 and lower L2/3; their orientation within these layers is mainly vertical, and at the level of L4 the axonal arbor appears to be confined to a single barrel, similar to the dendritic field (Woolsey et al. 1975; Feldmeyer et al. 1999; Lübke et al. 2000; Petersen and Sakmann 2000, 2001). Here, this observation was confirmed by measuring the asymmetry of the axonal arbor exclusively within L4. Moreover, the overlap between dendritic and axonal field was calculated (Methods). These data are summarized in Table 2.
|
An example of extensive overlap is shown in Figure 5A,B. For SSNs, the mean axon-free angle in L4 amounts to 225 ± 49°, only slightly smaller than their dendrite-free angle of 245 ± 42°. Thus, SSN axon collaterals are almost as eccentric as their dendrites. The axon-free angle of SPNs is also similar to their dendrite-free angle. The relation between the axon-free angle and the cell class shown in Figure 4C resembles the relation between dendrite-free angle and cell class described above. In the projection plane, axon- and dendrite-free angles closely overlapped in all cases and for both neuronal classes (Fig. 4D).
|
Overlap Analysis
Because the overlap between dendritic and axonal voxels within L4 increases linearly with voxel side length until 20–25 µm as illustrated in Figure 5C, a voxel side length of 25 µm represents a plausible choice for the overlap analysis. Moreover, it is on the order of magnitude of distance intervals commonly used for Sholl analysis. For SSNs, the overlap is within 70 ± 15% of dendritic voxels, whereas SPNs display this overlap to a lesser extent (47 ± 14% of dendritic voxels, Fig. 5C upper 2 panels). This difference in overlap was not due to a reduced density or extent of dendritic and axonal branches, because there were more elementary axonal and dendritic voxels with side lengths 1 and 25 µm in SPN than in SSN (see Table 2).
This strong overlap between dendrites and axons is a hallmark of L4 barrel neurons that is not displayed by cortical pyramidal neurons: Reconstructed L5 and 2/3 pyramidal neurons with well-filled axons were selected from our local database. The overlap difference between neurons from the 2 layers was insignificant, thus results were pooled. At a voxel side length of 25 µm, there was an overlap in 15 ± 6% of dendritic voxels in pyramidal neurons (n = 12, Fig. 5C bottom, P < 0.001 vs. both SSNs and SPNs). If the apical dendrite was not taken into account, overlap increased to 30 ± 8% but was still significantly smaller (P < 0.001 vs. SSNs and P < 0.05 vs. SPNs).
Interbarrel Projection Patterns
Previous studies have shown that similar to dendrites, axons of spiny L4 neurons are subject to intrabarrel confinement at the level of L4 (Woolsey et al. 1975; Lübke et al. 2000; Petersen and Sakmann 2000, 2001), although SPN axons tend to extend into the immediately adjacent walls of the surrounding barrels (Staiger et al. 2004). However, 3 exceptions to this generalization were found which had clear interbarrel projections. Figure 6 shows 2 SSNs with an axonal projection to an adjacent barrel and the region in L2/3 above it. Even these neurons are clearly confined with respect to the home-barrel border. Their interbarrel projections target roughly the same region within the adjacent barrel as the intrabarrel projection. Figure 7 shows the axonal projections of 3 SPNs clustered at the barrel center; one cell projects to the centers of 2 adjacent barrels, although the projection is not dense. A similar SPN projection pattern has been reported previously (Brecht and Sakmann 2002a).
|
|
Measurement of z-Shrinkage and its Influence on the Dendrite-Free Angle
Given the substantial degree of z-shrinkage in the slice preparation, it is important to assess its influence on symmetry measures and therewith the reliability of our conclusions. By comparing morphologies of the 4 same L4 neurons within the acute slice acquired with 2-photon microscopy and the embedded preparation, the average z-shrinkage Sz was determined to be 0.45 ± 0.10 (n = 32 distances between structures in the 4 cells; Methods), whereas the planar shrinkage Sxy was negligible (n = 21; both Fig. 8B). This low degree of Sxy is in the same order of magnitude as Sxy previously observed in comparable brain slice preparations (e.g., Hellwig 2000). Therefore, we neglected Sxy in all further considerations. Although Figure 8B shows a certain nonlinearity in Sz—in all cells Sz was stronger close to the slice surface and decreases with depth—we assume a linear Sz to estimate its effect on the dendrite- or axon-free angle: The maximal artifactual increase in due to a Sz of 0.45 would amount to on average 
33° and maximally 45° (see Methods eq. 3, Fig. 8D). In conclusion then, although there is considerable z-shrinkage, its influence on symmetry does not compromise the observation of different dendritic symmetries between SSNs and SPNs, because the average difference in dendrite-free angle of 83° between the 2 classes (Table 1) is roughly twice the maximal estimated as caused by shrinkage.
|
|
Discussion |
|---|
|
TopNotesAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Here, we have quantified dendritic symmetry of the 2 main excitatory neuronal classes within L4 of rat barrel cortex; SSNs and SPNs. SSN had mostly eccentric morphology, whereas SPNs were nearly radially symmetric. Most asymmetric neurons were located near the barrel border. The axonal projections within L4 were largely restricted to a single barrel except interestingly for those of 3 interbarrel projection neurons. Dendritic and axonal fields closely overlapped, being more pronounced in SSNs versus SPNs and considerably stronger in spiny L4 neurons versus extragranular pyramidal cells. Thus, dendrites and axons of individual excitatory barrel cells appear to be organized in subcolumns that further determine the organization of RF properties across L2/3.
Methodological Aspects: Asymmetry Measures and Shrinkage Distortion
To capture the asymmetry of spiny L4 neurons, we measured the dendrite-free angle of the projection of the neuron into the tangential plane relative to the barrel field. This simple and efficient measure could be easily extended to axonal structures. The previously mentioned measures of dendritic asymmetry used in visual cortex (i.e., bias indices; Katz et al. 1989; Hübener and Bolz 1992; Kossel et al. 1995) are not applicable to asymmetric barrel cortex neurons because of their extremely eccentric field. Polar plots that depict the number of dendritic intersections or coverage per angular section, appear to be another suitable means to characterize dendritic asymmetry (e.g., Leventhal and Schall 1983; Elston and Rosa 1997; Staiger et al. 2004). However, such plots simplify the neuronal structure only to some extent and therefore cannot be easily analyzed and/or classified. Furthermore, the comparison of neurons that are located in different corners of the barrel would require suitable normalization procedures.
In acute slice preparations, the processing and embedding causes slices to shrink mainly in the z-direction, that is, perpendicular to the slice plane. Shrinkage reduces the extension of neurons in this direction by about one half. Unfortunately, it is not possible to correct properly for this effect due to its nonlinearity: According to our direct comparison of neuronal structures pre- and postembedding, shrinkage is strongest in the region closest to the surface. This might be caused by various factors. First, cells at the slice surface die and therefore flatten before fixation. Secondly, embedding might lead to further shrinkage due to the pressure of the coverslip onto the slice, probably induced by adhesion forces between the coverslip and the embedding material. Third, in acute slices the washout of cell debris from the slice surface is likely to render the uppermost part of the slice more compressible than the deeper tissue. These effects may also cause the reduced visibility of barrel structures in thalamocortical slices after embedding. However, the assumption of linear shrinkage is still a good first-order approximation (Fig. 8B), that was used here to estimate the shrinkage-induced distortion of the dendrite-free angle. Accordingly, an increase of the dendrite-free angle of at most 50° is induced, which cannot invalidate our conclusions regarding the difference between SSNs and SPNs (see Results).
Dendritic Structure of Cell Types and their Location Within Barrels
In accordance with Golgi studies of mouse and rat barrel cortex, the dendritic field of most spiny neurons in L4 displayed a marked asymmetry, being oriented toward the respective barrel center and thus confined to the home barrel (Lorente de Nò 1922; Woolsey et al. 1975; Steffen and Van der Loos 1980; Simons and Woolsey 1984). This symmetry property was quantitatively characterized by the dendrite-free angular section of reconstructed neurons that were projected into the tangential plane of the barrel field. Considerable asymmetry was observed for SSNs, and less so for SPNs. Thus, a spiny L4 barrel neuron with a strongly asymmetric dendritic field is more likely an SSN, whereas lack of asymmetry is more characteristic of SPNs. The strong asymmetry of barrel SSNs differs from the radial symmetry of visual cortex SSNs (Lund 1984; Martin and Whitteridge 1984; Anderson et al. 1994; Hirsch 1995), although moderate asymmetries at borders of ocular dominance columns have been observed (Katz et al. 1989; Kossel et al. 1995). Staiger et al. (2004) also report symmetry properties of L4 neurons in thalamocortical slices in a qualitative manner; although these authors do not observe a cell-type–related bias in symmetry, it may well be that rotation of the neurons into the tangential plane would yield a somewhat different result.
The fairly strong correlation between spiny cell type and symmetry demonstrated here suggests that SPNs are located preferentially within barrel centers, whereas SSNs tend more toward "inhabiting" the barrel walls. The data set based on in vivo fillings clearly shows that this is more a tendency than a strict rule (Fig. 4E; see also Sun et al. 2006, their Fig. 1). In particular, we find symmetric SPNs close to the barrel border that extend part of their dendrites into the septa. There is also a subset of symmetric SSNs that was not observed in the thalamocortical slices; this difference may be related to development (P29 vs. P14). Nevertheless, dendrites of most symmetric neurons are also subject to intrabarrel confinement.
Functional Differences between SSNs and SPNs
Are SSNs and SPNs also functionally distinct within the context of the barrel circuitry? As far as known by now from paired recordings from spiny L4 neurons, their synaptic properties do not differ to a large extent (Feldmeyer et al. 1999; but see Cowan and Stricker 2004), and SPNs have been viewed as precursors of SSNs, that evolve by degeneration of the apical dendrite (e.g., Lund 1984; Elston and Rosa 1997). Clearly, our reconstructions represent snapshots at around P14 and P29, respectively, and differences might disappear or increase later on. However, in the adult the general pattern of dendritic intrabarrel confinement is conserved (Woolsey et al. 1975; Steffen and Van der Loos 1980; Greenough and Chang 1988; Tailby et al. 2005), and SPNs are still present (Simons and Woolsey 1984).
Moreover, SPNs have been shown to be integrated in different cortical circuits than the pathways that include SSNs, both in somatosensory and visual cortex. In the barrel cortex, SPNs receive more interbarrel inputs and also more input from other cortical layers within the same barrel column than SSNs (Schubert et al. 2003). In the primate visual cortex, SSNs and SPNs were also found to receive different inputs from L4 sublayers (Yabuta et al. 2001).
As to their primary sensory activation, SSNs and SPNs appear to receive the same type of direct thalamocortical input (Porter et al. 2001; Staiger et al. 2004; Bruno and Sakmann 2006); nevertheless they show different degrees of adaptation to repetitive whisker stimulation in vivo (Brecht and Sakmann 2002a). Cat visual cortex SSNs and SPNs neurons responded similarly to visual input and electrical stimulation of their afferents (Martin and Whitteridge 1984). Differences between the axonal projection patterns of SPNs and SSNs are discussed below.
Axon Arbor Confinement and Overlap with the Dendritic Arbor
The columnar axonal confinement of SSNs and SPNs to the domain of the dendritic tree and thus to the barrel at the level of L4 is rather strict in comparison with similar tendencies observed in monkey visual and somatosensory cortex (Jones 1975; Katz et al. 1989). Moreover, projection patterns of spiny neurons in L4 of visual cortices may vary considerably (e.g., Martin and Whitteridge 1984; Lund et al. 1995; Fitzpatrick 1996; Yabuta and Callaway 1998), whereas they are rather uniform in barrel cortex, possibly indicating a higher degree of complexity of visual cortex circuitry.
We found that spiny L4 neurons display a dense axonal arborization and a strong overlap between their axonal and dendritic domain compared with pyramidal neurons in supra- or infragranular layers. The dense arborization underlies the high intrabarrel connectivity of spiny neurons (30%; Feldmeyer et al. 1999; Petersen and Sakmann 2000). The overlap is significantly stronger in SSNs than in SPNs, constituting another difference between the 2 neuronal classes. Because the stronger overlap is not due to a larger axonal projection of SSNs at the level of L4 (see Table 2), it appears that SPNs project more diffusely within the barrel. This observation is in line with data from cat visual cortex, where SPNs were observed to project in a more diffuse manner than SSNs; although some SSN projections were found to cluster locally, this was also not observed for SPNs (Martin and Whitteridge 1984). Similarly, Yabuta et al. (2001) investigated SPNs and SSNs in layer 4B of primary visual cortex in the primate and suggested that the 2 cell types were part of 2 functionally separate subsystems, that is, the P and M pathway, also based on their different projection targets.
Taken together with the above observations on different inputs to barrel SSNs and SPNs, these findings suggest that throughout visual and somatosensory cortex the 2 cell types, STPs and SSNs, may be integrated into different circuits that are involved in different computational tasks. It remains to be elucidated what kind of tasks that could be within barrel cortex. Perhaps the relative localization of SSNs closer to the barrel border and their stronger local axo-dendritic overlap as described here might make them more suitable for carrying subcolumnar information (see below), whereas SPNs might be more apt to process information that pertains to the principal whisker as a whole.
Functional Implications of Strong Overlap between Axonal and Dendritic Domains
The asymmetric orientation of SSN dendrites has been suggested to maximize the contact with the segregated afferent thalamocortical projections (Killackey 1973; Arnold et al. 2001). In fact, the activity of the thalamocortical afferents maintains dendritic bias in L4 (Tailby et al. 2005). Because of the strong axonal confinement of SSN and/or SPNs, this input segregation would be further preserved at the level of L4, ensuring that each barrel primarily processes information from its principal whisker (but see below).
Our anatomical observation of a pronounced overlap with the dendrites of the same SSN/SPN promotes recurrent excitation and thus could enhance the responsiveness of barrel neurons. Obviously, the narrow axonal domain will also restrict lateral spread of excitation within a barrel column. Such locally restricted networks on a subbarrel scale strongly imply the existence of coactive subcolumns or perhaps minicolumns within a barrel, as suggested previously (Land and Simons 1985; Yuste et al. 1992; Simons 1995) that may encode for particular receptive field (RF) properties. The strictly columnar projection of SSNs to L2/3 (Lübke et al. 2000, 2003) implies that any subcolumnar patterns in L4 will be repeated at the level of L2/3.
Indeed, barrel neurons appear to be direction sensitive (e.g., Brecht and Sakmann 2002a) and in addition to exhibit clusters of neurons that are selective for certain directions of whisker deflection, also called angular tuning domains (Simons 1978; Bruno et al. 2003). These domains may be arranged in a stereotypical manner, at least within L2/3 of the D3 barrel, where they would continuously represent the direction toward the next-neighbor whisker (Andermann and Moore 2006). However, according to the in vivo single unit recordings (Bruno et al. 2003), angular tuning domains are at maximum 100 µm in width and thus smaller than the average dendritic field span of SSNs and SPNs, thereby fitting more into the canonic size of minicolumns as suggested by Mountcastle (1978, 1997). It remains to be elucidated whether L4 neurons are indeed organized in direction-selective minicolumns, whether direction selectivity is mapped in a more continuous way and how such minicolumns would be mapped onto the coarse subbarrel structure that was recently revealed (Land and Erickson 2005). Direction-selective RFs could be based on dendritic asymmetry, as suggested for the L6 Meynert cell (Livingstone 1998). However, complex RF properties are usually not found in the primary input regions, and a link between direction selectivity and dendritic asymmetry in the primary visual cortex has been questioned (Anderson et al. 1999; cells L3–6), albeit these neurons display rather weak asymmetry in comparison with spiny barrel neurons. However, dendritic asymmetry in L4 barrel neurons is unlikely to underlie directional tuning, because direction-sensitive responses are observed along the entire whisker input pathway (e.g., Lichtenstein et al. 1990; Brecht and Sakmann 2002b; Minnery and Simons 2003; Timofeeva et al. 2003; Bruno and Sakmann 2006).
Irrespective of the RF properties that may be encoded within minicolumns, the strong recurrent excitation provided by the local axonal projection may also serve to amplify cortical inputs (Feldmeyer et al. 1999), to sustain persistent cortical activity within the barrel (McCormick et al. 2003), or to render the cortical circuits sensitive to synchronous inputs (Pinto et al. 2003).
Interbarrel Connectivity
A small fraction of all neurons from our sample violated the principle of the intrabarrel confinement of axon collaterals, projecting directly into neighboring barrels. These neurons may thus provide direct input to neighboring barrels and therefore represent a substrate for transcolumnar inputs observed in vitro (Schubert et al. 2003) and multiwhisker surround RFs observed in vivo (Fox et al. 2003) that both originated from within layer 4. Due to truncation of axonal processes and insufficient staining the 5% interbarrel projection neurons reported here represent a lower limit, although thalamocortical slices are oriented along rows, the preferred direction for interbarrel connections (Bernardo et al. 1990; Keller and Carlson 1999). In addition, SPNs were observed to extend their local axonal domain into the adjacent septum as well as the barrels (Staiger et al. 2004). Although these interbarrel projections are unlikely to represent a major pathway of cortical processing within L4, they challenge the view of barrels as entirely independent processing units of thalamic input (Goldreich et al.