In: Computational Neuroanatomy:
Principles and Methods. (Ascoli, G., ed).
The Humana Press (in press)
Computational anatomical
analysis of the basal forebrain corticopetal system
L. Zaborszky1,
A. Csordas1, D. Buhl1, A. Duque1, J. Somogyi2
and Z. Nadasdy3
1Center for Molecular and Behavioral Neuroscience, Rutgers University, Newark, NJ
2Department of Medicine, Flinders University, Bedford Park, SA 5042, Australia
Main Text: pp. 2-30
Acknowledgement: p. 31
Figure Legends: pp. 32-37
References: 38-43
Footnote on p.6,
1A color version of this and other figures are available in the companion CD-ROM file.
2 Voxel is a 3D pixel that is the spatial unit of our analysis. See also Appendix 10.3, 10.4 and 10.4.1. A mathematical description of the definition of voxels can be found in [41].
Corresponding author:
Laszlo Zaborszky,
Center for Molecular and Behavioral Neuroscience
Rutgers, The State University of New Jersey,
197 University Avenue, Newark, NJ 07102
Tel.: 973-353-1080/ ext. 3181
1. ABSTRACT
The basal forebrain is comprised of a neurochemically heterogeneous population of neurons, including cholinergic, GABAergic, peptidergic and possibly glutamatergic neurons that project to the cerebral cortex, thalamus, amygdala, posterior hypothalamus and brain stem. This multitude of ascending and descending pathways participate in a similarly bewildering number of functions, including cognition, motivation, emotion, and autonomic regulation. Traditional anatomical methods failed to grasp the basic organizational principles of this brain area and likened it at best to the organization of the brain stem reticular formation. Our studies, using various computational methods for analyzing the spatial distribution and numerical relations of different chemically and hodologically characterized neuronal populations as well as fully reconstructed electrophysiologically identified single neurons, began to unravel the organizational principles of the basal forebrain. According to our model, the different cell types form large scale cell sheets that are aligned to each other in a specific manner. Within each cell system, the neurons display characteristic discontinuous distributions, including high density clusters. As a result of non-homogeneity within individual cell populations and partial overlapping between different cell types, the space containing the bulk of cholinergic neurons comprises a mosaic of various size cell clusters. The composition, dendritic orientation and input-output relationships of these high density cell clusters show regional differences. It is proposed that these clusters represent specific sites (modules) where information processed in separate streams can be integrated. Via this basal forebrain mechanism a topographically organized prefrontal input could allocate attentional resources to cortical associational areas in a selective, self-regulatory fashion.
2.
INTRODUCTION
The
term basal forebrain (BF) refers to a heterogeneous collection of structures,
located close to the medial and ventral surfaces of the cerebral hemispheres.
This highly complex brain region has been implicated in attention, motivation,
and memory as well as in a number of neuropsychiatric disorders such as Alzheimer's
disease, Parkinson's disease, and schizophrenia [1-3]. Part of the difficulty
in understanding the functions of the BF, as well as the aberrant information
processing characteristic of these disease states, lies in the anatomical
complexity of the region. Basal forebrain areas, including the medial septum,
ventral pallidum, diagonal band nuclei, substantia innominata and peripallidal
regions contain cell types different in transmitter-content, morphology and
projection pattern [4-5]. Among these different neuronal populations, the
cholinergic corticopetal projection neurons have received particular attention
in numerous functional and pathological studies.
Recent interest in
BF research was prompted by discoveries showing that a specific population of
neurons in this region, namely those that use acetylcholine as their
transmitter and project to the cerebral cortex, are seriously compromised in
Alzheimer's disease [6-9]. However, cholinergic projection neurons represent
only a fraction of the total cell
population in these forebrain areas, which also contain GABAergic, peptidergic
and possibly glutamatergic neurons [10-11]. According to our unpublished
estimations in one hemisphere of the rat brain, in the cholinergic BF areas
20,000 cholinergic corticopetal cells are intermingled with other neurons,
including about 35,000 calbindin, 26,000 calretinin and 24,000
parvalbumin-containing neurons. These
calcium-binding proteins are used to characterize different non-overlapping
populations of GABAergic neurons.
A quasi 3D representation of the cholinergic
cell bodies (Fig. 1A) or the dendritic arborizations of their neurons (Fig. 1B)
does not appear to show any recognizable architectural features, confirming a
classical view in the literature that arousal is supported by a diffuse
reticular activating system, including core brain stem structures, the BF and
the so called non-specific thalamic nuclei [12-13]. On the other hand, careful
monitoring of the behavioral effects of lesions in the BF using an immunotoxin
selective for cholinergic neurons, suggests that compartments of the BF
together with their specific cortical target areas may participate in different
cognitive operations [14].
If the BF
participates in different operations, we would expect that this may be
reflected both in the local as well as in the large scale structural
organization of its constituent neuronal populations.
For example, one would expect that the BF would be constituted of
repetitive building blocks (modules) found in many other areas of the CNS,
including the cortex, striatum, hypothalamus, brain stem or the spinal cord
[15-18]. The modular structure in various brain regions is the prerequisite
structural basis for parallel, distinct operations [19]. Other structural
features like anisotropic dendritic orientation or segregation of various
afferents and efferents can also be taken as evidence for selective information
processing [20-26]. In the past several years we systematically investigated
the 3D spatial organization of the various BF neural populations, including
their dendritic organization and input-output relationship with the aim of
uncovering the organizational principles of the BF, in particular within areas
that are most heavily populated by cholinergic corticopetal neurons. This
review is an attempt to summarize how anatomical features may constrain
information processing in this brain area. The chapter is divided into several
sections, each with subheadings indicating the special methods used. Following
the main body of the text the reader can find an Appendix with a detailed
explanation of the data acquisition and analysis presented.
3. ASSOCIATION AND SEGREGATION OF DIFFERENT
HODOLOGICALLY IDENTIFIED NEURAL POPULATIONS
3.1. Overlap Analysis
Although there is considerable species variation in the precise locations of cholinergic projection neurons in the BF, the efferent projections of these cells follow basic organizational principles in all vertebrate species studied. Thus in rodents, neurons within the medial septum and nucleus of the vertical limb of the diagonal band provide the major cholinergic innervation of the hippocampus; cholinergic cells within the horizontal limb of the diagonal band project to the olfactory bulb, piriform and entorhinal cortices; cholinergic neurons located in the ventral pallidum, sublenticular substantia innominata, globus pallidus, internal capsule, and nucleus ansa lenticularis, collectively termed as nucleus basalis, project to the basolateral amygdala and innervate the entire neocortex according to a rough medio-lateral and antero-posterior topography [27-35]. Similarly, in primates, including humans, corticopetal cholinergic cells are subdivided according to the topography of their projections [36].
It is
unclear, however, what the functional equivalent of this topography is,
especially in light of a study in rat, showing that neighborhood relationships
in the BF projection neurons do not correspond to near neighbors in the
representational areas of sensorimotor cortices, thus arguing against a simple
functional organization [37]. Knowing the importance of the cholinergic BF
system in modulating cortical activity [38], we asked whether the organization
of the basalocortical system can, in any sense, be related to the distributed and
hierarchical organization of cortico-cortal connections, as proposed by Van
Essen and his associates [39]. Figures 2 and 3 display cases of overlapping and
segregated projection neurons from a study aiming at a comprehensive
reevaluation of the basalocortical projection (Csordas and Zaborszky, in
preparation). Figure 2A is from a case in which two different retrograde
tracers were injected into two cortical areas that were in the same
medio-lateral topographical register but they differed in their rostro-caudal
location. This 3D image suggests that the two neuron populations (marked by
different symbols1), projecting to two different cortical areas,
are, at least in the rostral part of the BF, intermingled. Using an overlap
analysis program described recently [21, 25, 40], Figure 2C shows that a
substantial population of the two types of projection neurons are, indeed,
located in overlapping voxels2 (the method is briefly described in
10.4.7). Figure 2B shows another case where the two retrograde tracer injections
were in different medio-laterally located cortical areas. As can be seen from
this 3D rendering, there is little overlap in the location of the neurons
projecting to these two cortical target areas.
Figure 2D, using the overlap analysis program, supports the subjective
impression, that no overlap exist between these two distinct cell populations.
3.2.
Iso-density Surface Rendering
Figure 3A shows a 3D rendering of the distribution
of four BF cell populations that project to four, arbitrarily defined
medio-lateral sectors of the neocortex reconstructed from eight individual
experiments. Since the overlap-analysis is limited to the simultaneous
comparison of only two cell populations and only in two dimensions, in order to
appreciate the overall projection pattern in the three dimensional space, we
developed an algorithm that renders a surface around voxels of similar cell
densities (Appendix 10.4.2. and [41]). Since cells are replaced by densities
and densities are rendered around by surfaces, the simultaneous 3D
visualization of multiple cell populations is feasible. Figure 3B is a 3D
composite of the iso-density maps of six different cell populations, suggesting
that the bulk of each cell population projecting to the six cortical targets is
separated in the BF. Unfortunately, when iso-surfaces of different cell types
are combined, the larger surface area may have included iso-surfaces of other
cell types. Therefore, separate renderings of the individual cell populations
and pairwise overlap analysis have to be considered [41].
A detailed overlap analysis of 9 cases each with paired injections and some 30 computer-generated combinations of these cases (Csordas and Zaborszky, in preparation) suggest that corresponding medio-laterally located frontal and posterior cortical areas receive their input from a partially overlapping area in the BF. On the other hand, topographically non-corresponding frontal and parieto-insular areas receive their projections from non-overlapping areas of the BF. Since the location of overlapping voxels in the BF is highly specific for the injection sites that represent cortical associational columns, these data suggest that the BF cholinergic input is transferred via specific cortico-cortical nodal points toward hierarchically related frontal cortical areas.
4.
INHOMOGENEOUS DISTRIBUTION OF CHEMICALLY IDENTIFIED CELL
POPULATIONS
4.1. Differential Density 3D Scatter Plot
Figure 4A, using a differential density 3D scatter plot (for a brief description of this method see: Appendix 10.4.1), shows that the density of cholinergic cells is not uniform (see also Fig. 2 in [4]). Cholinergic cells often form clusters consisting of 3-15 tightly packed cell bodies. The saliency of these clusters, nonetheless, depends on the density threshold setting. For example, when using a relatively low threshold (ł 5 cells per 250 x 250 x 50 μm voxel size) these clusters seem to be diffusely distributed. In contrast, when using a relatively high threshold (dł15 cells/voxel) the clustering of cholinergic cells seems to deviate from a random distribution. Interestingly, in primates, in comparison to rodents, a proportionally higher percentage of cholinergic cells are located in clusters [4], suggesting that increasing clustering in the phylogeny of BF cell populations might be related to the increased specialization of the cortical areas they project to. Similarly, the location of other cell populations in rat, including calretinin, calbindin and parvalbumin-containing neurons, suggests inhomogeneous distributions (Zaborszky, Buhl, Pobalashingham, Somogyi, Bjaalie and Nadasdy, in preparation). Figure 4B shows a similar type of differential density scatter plot of parvalbumin-containing neurons where dots represent the cell bodies and large filled circles represent high density spots.
4.2. Iso-relational Surface Rendering
Since the simulateneous visualization of more than two populations using differential density scatter plots is difficult, we applied another surface rendering algorithm that uses both density and spatial relational constrains (Appendix 10.4.2. and [41]). Figure 4C shows the iso-relational surfaces (dark solid) rendered around regions where the density of both cholinergic and parvalbumin cells met two criteria: (1) density is at least five for each cell type within the voxel (250 x 250 x 50 μm) and (2) the ratio of cholinergic to parvalbumin cell counts is maximum 0.5. In other words, the voxels covered by the dark surface contain at least twice as many parvalbumin as cholinergic neurons. With the iso-relational surface rendering we introduced a double, density and relational, constraints that led to a further simplification of our model. Comparing the locations covered by such iso-relational surfaces with the scatter plot distribution of the corresponding two cell populations clearly shows that these surfaces form a central core, consisting of high density cells from both cell populations that are flanked on all sides with single cell populations of gradually decreasing densities (Fig. 4C). Merging the three pairwise iso-relational surfaces (cholinergic/parvalbumin, cholinergic/calretinin and cholinergic/calbindin) into one scheme suggests that the cholinergic cell 'column' can be parcellated into several smaller clusters or larger amalgamations, where cholinergic cells are mixed with the other three cell types in a specific fashion (Fig. 4D). Using a section-by-section analysis of the overlap, as shown in Figures 2C-D, one can get a fairly good idea about the composition of the mixed clusters. The advantage of the combined 3D iso-relational surface rendering of Figure 4D is indeed in the totality of this image. Comparing similar types of renderings from different brains, a similar global pattern emerges, suggesting that the configuration of the iso-relational surfaces is not by-chance and that the high density clusters in the individual cell populations may correspond to the zones where the different cell populations overlap with each other. The location of these overlapping zones may be determined during ontogenesis.
5. CHOLINERGIC CELL GROUPS SHOW REGIONALLY SELECTIVE
DENDRITIC ORIENTATION
5.1.
Mean 3D Vector of Dendritic Processes
Since the
geometry of axons and dendrites imposes constraints on their connections, in
order to understand how information is handled in the BF, it is important to
determine how the shape of the axonal and dendritic arborizations could
influence regional connectivity patterns. Cholinergic cell bodies give rise to
2-5 primary dendrites radiating in all directions. The relatively straight
primary dendrites bifurcate in an iterative fashion, and the sum of the lengths
of the daughter branches is usually larger than that of the mother branch. The
dendrites of adjacent cholinergic neurons often constitute overlapping fields.
The dendrites are freely intermingled with passing myelinated fiber bundles
within which they are embedded. Thus, the dendritic organization of the
cholinergic BF neurons resembles that of the isodendritic type of neurons of
the reticular formation [42-44] or the so called interstitial neurons
characterized by Das and Kreutzberg [45]. The total length of the dendrites of
individual cholinergic neurons in rats is about 4 mm, arborizing in a box of
about 0.1 mm3 however, filling only a fraction of its spatial
domain. According to our estimation, one cholinergic cell dendritic domain
might share its space with 50-80 other cholinergic neurons, depending on its
location in the BF. Although a
particular orientation of cholinergic dendrites could be noticed upon
inspecting areas where the density of dendrites is low (see Fig. 1 in [46]), it
is not possible to appreciate dendritic orientation with certainty in regions
where the cell density is high as can be judged from Figure 1.
We assumed that the cholinergic cell
clusters, beyond their spatial segregation, must fit into the functional
network of their input-output connections. In other words, we assumed that
cholinergic cell clusters develop under the constraints that link together
functionally related output (neocortical) and input (brainstem and
telencephalic) pathways and this input selectivity, we reasoned, must be
reflected by the anisotropic dendritic orientation of the putative cell
clusters.
In order to correlate regional differences of
dendritic orientation to the spatially distributed population of neurons we
developed a method of representing the main dendritic trees of
individual neurons with three-dimensional vectors and embedded them into the 3D
coordinate system of the cell bodies. The origin of a vector represents the
position of the neuron, its orientation represents the dominant orientation of
the dendritic tree (for details see Appendix 10.4.4.) and the length of vector
represents the average length of the dendritic tree. The main orientation vectors of 750 individual cholinergic
cells, selected from a population of about 15,000 cholinergic cells, are shown
in Figure 5A. Rotation and navigation in the 3D plot made it possible to gain
insight into the vector orientation even in the denser cell clusters.
Comparison of Figure 5A with the differential density scatter plot of the same
dataset, shown in Figure 5B, suggests a tendency of iso-orientation of
dendrites within a given cholinergic cell cluster.
5.2. 2D Dendritic
Stick Analysis (Polar Histogram)
To obtain a quick qualitative characterization of the directional distribution of dendritic growth projected onto the plane of sectioning, the polar histogram is the method of choice (Neurolucida® software package; see also Appendix 10.4.5 and [47]). In essence, using only the x and y coordinates of the traced dendritic segments, where individual segments are composed of pairs of adjacent points, the algorithm pools together all the segments around a center but preserves their length. The total range of angles is then binned to equal sectors and the program calculates the sum of segments in each bin. The radial length of a filled sector is proportional to the total length of the dendritic branches of that specific orientation, thus the contributions of segment lengths and segment counts of that specific orientation are inseparable. In other words, a few long segments can add up to the length of many short dendrites. Depending on the choice of binning interval, the angle discrimination can be finer or broader. Analyzing BF cholinergic dendritic orientation by polar histograms suggests a regional orientation preference (Zaborszky, Nadasdy and Somogyi, in preparation).
In
contrast with the polar histogram method, the vector representation preserves
the neuronal identity of dendrites instead of pooling them together and still
provides an overall view of orientation of dendrites. The main advantage,
however, is that vectors relate the dendrites to the spatial distribution of
the neurons in a simplified and meaningful fashion. To compare the regional
dendritic orientation derived from polar histograms with the orientation
vectors calculated for individual cells, we selected a sub-space of the septal
area where a more detailed analysis of subpopulations of cholinergic neurons
was available. The Neurolucida® program allows one to outline and select cell
populations from any number of sections in a series and to construct a polar
histogram of the dendrites pooled together from the selected cell populations.
Figure 6A represents a portion of the septal area from Figure 5A viewed from a
sagittal direction. Figure 6B displays cholinergic cells whose dendrites were
traced from a series of sections cut in the sagittal plane. The selection areas
of the four polar histograms of Figure 6C-F are indicated by boxes of various
sizes in the upper right diagram. Comparing the dendritic orientation obtained from the polar
histograms of pooled dendrites with the mean 3D vector of the dendritic branches indeed suggests that
subpopulations of cholinergic cells can be delineated based upon their density
and main orientation of their dendritic arbor. Therefore, one of the key features of
cluster organization is the iso-orientation of their dendrites.
6. VARIOUS AFFERENTS IN THE BF SHOW REGIONALLY
RESTRICTED LOCALIZATION
Using
a double strategy of recording the location of putative contact sites between
identified axons and cholinergic profiles as well as identifying in
representative cases under the electron microscope the presence of synapses,
one can get a fairly good idea about the extent of potential transmitter
interactions in the BF (see for references [4, 38]). Although the noradrenergic
and dopaminergic axons contact cholinergic neurons in extensive portions of the
BF, the majority of afferents (cortical, amygdaloid, striatal, peptidergic)
appear to have a preferential distribution in the BF; thus a specific input can
contact only a subset of neurons. Figure 7 give examples of the distribution of
restricted versus more diffuse afferents.
In many cases examined, labeled terminal varicosities detected in the BF were related to both cholinergic and non-cholinergic postsynaptic elements. In fact, the detectability of synapses on cholinergic neurons was usually proportional to the density of terminals present in a given area. Thus at first approximation, the probability of synapses between cholinergic neurons and various afferents may depend on the geometry of the dendritic arbor and axonal ramifications. Since a systematic study comparing the dendritic arbor of cholinergic neurons with various afferent orientations would require a substantial time, we only briefly comment on this issue here by documenting the case of calcitonin-gene-related-peptide (CGRP)-containing axons in the internal capsule. Figure 8A shows cholinergic cells and their traced dendrites at about 1.5 mm posterior to bregma. Figure 8B is a schematic drawing from an adjacent section that was immunostained for CGRP and whose axons in the internal capsule were traced at high magnification. Comparing the polar histograms of cholinergic dendritic trees in the internal capsule (C) with that of CGRP axons (D), it is obvious that CGRP axonal ramifications have the same prevailing direction as the dendritic arbor of cholinergic neurons. Indeed, electron microscopic studies confirmed abundant presence of CGRP in axon terminals synapsing with cholinergic neurons in this region [48].
The
probability of synaptic connections can be calculated from the overlap of
axonal and dendritic domains [49]. The probability of having more than one
synapse between any given presynaptic axon and a postsynaptic cell is maximal
in the case when the terminal axon and the receiving dendrite are running in
parallel ('climbing' fiber type contact). However, only one synapse is possible
if the axon runs at right angles to the dendrite ('crossing over' type of
geometry). If the terminal axon is oriented obliquely to the receiving dendrite
the probability of synaptic contacts is a cosine function of the angle between
the axon and the dendrite [50]. Since various afferents show specific
localization, it is likely that cholinergic cells in various BF subdivisions
can sample a unique combination of afferents. It seems that in each major
subdivision of the BF along the axis of the major orientation of the cell
bodies a specific type of axon is maximally aligned with the preferred
orientation of cholinergic dendrites of that area. Different cells, or perhaps
different dendrites of the same cell, can sample the same input differently
according to the spatial organization of the dendrites and corresponding axons.
For example, the majority of dopamine-β-hydroxylase positive varicosities
(used to stain noradrenaline and adrenaline containing neurons) establish
single synapses with cholinergic dendrites, while a small population of
cholinergic neurons (at most 5%) appears to receive multiple contacts on their
dendrites in the form of climbing-type arrangements. Such climbing-type
synapses were most often detected in the substantia innominata (Fig. 3 in
[51]), but were also occasionally seen in other BF regions. It is unclear
whether cholinergic neurons with climbing-type inputs are different in other
respects, however, one can speculate that the noradrenaline released at these
climbing-type synapses must have a more powerful action on these selected
neurons as compared to the single synapses at random locations.
Our earlier assumptions about the randomness of connections [46, 52] had to be modified when we realized that prefrontal axons seem to terminate exclusively on non-cholinergic cells, including parvalbumin-containing GABAergic cells, in spite of the fact that many of these axons arborize in the immediate vicinity of cholinergic neurons [53]. It is expected that the detailed reconstruction of local axon collaterals of BF neurons may add to the specificity of the connectional scheme in the BF [38].
9. CONCLUDING REMARKS
Since the seminal paper of Schwaber et al. [56], who first used computer-aided data acquisition and 3D reconstruction of BF cholinergic neurons, the progress in understanding the organization of the BF has been very slow. It is likely that in the coming years the sophisticated use of multi-electrode recordings in awake behaving animals and the application of promising new computational tools will define how anatomical features constrain the extraction of information processed in the BF. For the time being, we can only speculate on how the BF, in particular the cholinergic neurons, process specific information despite the apparently diffuse organization of its elements.
The territory of the BF populated by cholinergic corticopetal cells can be viewed as a large interconnected network where a systematic directional variation of dendritic clouds and presynaptic axonal clouds permeate each other intimately. In spite of the lack of internal borders, within this large cholinergic assembly, smaller sub-assemblies can be delineated by differential cell densities and dendritic orientations, input-output features and numerical relations of the constituent neuronal populations. The cholinergic cell clusters with other local or projection neurons may represent special sites (modules) where information processed in separate streams can be integrated. The location and size of these modules may temporarily vary according to the prevalence of state-related diffuse brainstem modulatory and more specific telencephalic inputs. From this latter group of afferents, the prefrontal input may function as an external threshold control which allocates attentional resources via the BF to distributed cortical processes in a selective, self-regulatory fashion.
10.
APPENDIX
10.1. Animals and Tissue
Processing
The reconstructions and statistical analyses presented in this paper were prepared from data obtained from adult male Sprague-Dawley rats. All animal procedures were in compliance with the National Institutes of Health Guidelines for the Care and Use of Animals in Research and approved by the Rutgers University Institutional Review Board for the Use and Care of Animals. The anesthesia, electrophysiological recordings, perfusion of animals and tissue processing have been described earlier [55, 57-58].
Immunostained, diaminobenzidine labeled cell bodies were digitalized in BF areas with the aid of an image-combining computerized microscope system (Zeiss Axioscope, 20 x Plan-NEOFLUAR lens) using the Neurolucida® software package [59] (MicroBrightField, Colchester, VT). Outlines of the sections, contours of structures and fiducial markers were drawn with a 5 x Plan-NEOFLUAR lens. Dendritic branches were traced from the cell body by connecting tracing points by straight lines (Plan-APOCHROMAT 40 x [NA=1.0] or Plan-NEOFLUAR 63 x [NA=1.25] oil immersion lenses). Sections containing fluorescent-tagged cell bodies were mapped by using the epifluorescent setup of the Axioscope microscope equipped with appropriate filters. Fast Blue and Fluor-Gold-labeled projection neurons (exciter/barrier filter set 365/418) and the FITC-labeled (FITC exciter/barrier filter set 450-490/520) cholinergic cells could be separately visualized in the same section. Labeled cells were mapped from every 8th sections at a magnification of 20 x.
Although the outlines, and contours were drawn flat disabling z input information, dendrites were followed in the depth of the sections (50 μm or 100 μm) by changing the focus. Curvilinear dendrites were represented in the computer as a series of short straight lines giving a close fit to the original shape and length. The Neurolucida® hardware system allows a point to point discrimination of 0.3 μm in all axes. Neurons traced from each section were aligned to a common reference, e.g. the lowest midline point of the corpus callosum. Mapped sections were aligned using up to 99 alignment points for best-fit matching included in the Neurolucida® software program. The data generated by tracing the neurons using the Neurolucida® software are later referred as the Neurolucida® database. The database is composed of a stack of aligned sections.
Neurons in the Neurolucida® database were represented by the x, y and z coordinates of the cell bodies. In the database, dendritic trees originating from the same neuron at different sites were represented as separate but adjacent data blocks and were encoded independently from their cell bodies. Since cell bodies were not traced as 3D objects, the origin of the primary dendrites did not necessarily match in any dimension. Due to this independence of cell bodies from their dendrites in the encoding scheme, finding the common cell body for each dendrite was not obvious. Branching points were marked. Based on the branching points, first, second, third and higher order dendritic segments were identified as stemming from a parent node. The hierarchical encoding system (introduced by Neurolucida®) allowed us to recursively represent the complexity of any dendritic tree in the database.
10.3.
Selection of Neurons for Dendritic Tracing
In this paper dendritic data are derived from three
different datafiles. As a preliminary material all cholinergic cell bodies with
their dendritic processes were traced from seven coronal sections (50 μm).
Approximately 1,300 cells were traced in this material. Figures 1B and 8A are
from this datafile. To create a more complete database, a second brain was cut
in horizontal planes into 34 consecutive sections (100 μm thick), and
every cholinergic cell was digitized. Figures 5A and 6A are generated using
this horizontal dataset. Finally, a third brain was cut in the sagittal plane
(100 μm thick) and the data presented in Figures 6B-F are derived from the
septal region of this sagittal dataset.
In this latter brain, similarly to the horizontal set, all cholinergic
cell bodies were digitized. From the horizontal and sagittal dataset about 5%
of total cholinergic neurons were selected for dendritic reconstruction based
on a random sampling of the total population. In order to obtain a sample of
neurons representative for the inhomogenous distribution of the entire
population we used a combination of two different density criteria, one with
high and another with low resolution. For both selections, the space that
incorporated all cholinergic cells was subdivided into subspaces of identical
size (voxels or unit spaces), and on these voxels, based on the cell density,
different selection criteria were applied. As for a high density criterion, we
selected cells from voxels of 100 x 100 μm x section thickness (=100
μm). To resolve a larger scale inhomogeneity, cells were selected from
voxels of 500 x 500 μm x section thickness. The larger voxel captured the
density differences at a 500 μm resolution. The 100- μm sample size
was applied to sample local densities at a 100- μm scale. Next, the two
samples were combined. As a result, the combination sample reflected both the
global and the local distribution features of the cells. From each voxel, where
the local cell counts met the density criteria, a neuron was selected on a
random basis and marked for dendritic tracing. The number of cells s selected
for dendritic tracing from each voxel with both v’ and v” sizes was
proportional to the natural logarithm of the number of cells n within the given
space 
multiplied by a
constant c as follows:
.
For the interpretation of i,j,k voxel
indices see [41]. The purpose of the multiplication factor c was to provide
flexibility to scale up or down the number of selected cells. The value of c
was set to 0.5 with 100 μm grids and 0.4 with 500 μm grids. The edges of the two unit space (voxel)
types v’ and v” were 100 μm and 500 μm. Technically, the datafile of
traced cell bodies was exported from Neurolucida®, parsed for different objects
(cell bodies, structure outlines, etc.) and the point coordinates of cell
bodies were extracted. A custom written C++ program performed the partitioning,
cell counting and selection of target cells for dendritic tracing. With this
sampling scheme we marked 750 cells from the horizontal dataset (n=15,776) and 137 from the septal sagittal dataset (n=2,266). The generated data file with
the target cells indicated, was inserted to the original Neurolucida® datafile for subsequent dendritic tracing.
10.4.
Analysis of the Data
For data analysis as described under 10.4.1-10.4.3, we extracted the x, y, and z coordinates of the cell bodies from the Neurolucida® database and saved them in ASCII format, each cell type in different data files. Structure outlines were stored separately. The medial, lateral, dorsal and ventral extremes of the cholinergic cell distribution were taken as a 3D framework to incorporate the entire database. For expressing regional density changes the 3D framework was subdivided into virtual blocks of identical size denoted as ‘voxels’. Section thickness served as unit size for the z dimension. If cells of different types were mapped from different but adjacent sections (plots in Fig. 4 and 9C-D) then their z coordinates were collapsed into a common two-dimensional plane (master plane) by removing the within-section depth coordinates but preserving the x and y coordinate of the cells. Each different cell type was separately counted in each of these master plane lattices. For visualization, we used different thresholds. Differences in the voxel size and thresholding could significantly influence the obtained results. A more detailed methodological description and discussion is given in a recent publication [41].
10.4.1.Differential Density 3D Scatter Plot
Density differences within or between cell populations can be represented in 2D isodensity maps [60]. The obvious limitations of this method is the lack of the third dimension. Our method [41] quantifies density differences first, then plots the density descriptors in a real 3D coordinate system. The input data is provided as position of cell bodies by their locations as points in the 3D Euclidean space. In this database each row represents a single cell given by the x, y and z coordinates relative to a reference point as the origin. The entire database is placed into a framework which is partitioned into boxes of identical size (‘voxel’) as a grid system. Cells are counted within each voxel. In contrast to the parametric representation of the space this provides a 3D volumetric dataset where the dimensions are the x, y, z position of the voxel and the local cell count within the voxel space. Voxels of cell counts larger than a predefined density are considered and represented by a single marker randomly selected from the neurons in the corresponding voxel. The distribution of these markers highlights locations where high density neuronal clusters occur. Figures 4A-B and 5B represent this type of analysis.
10.4.2. Iso-density Surface Mapping
The spatial distribution of different cell types may be very complicated as neuronal populations interdigitate, intersect or overlap with one another. Instead of using scatter plots, the spatial organization of density differences is better visualized by rendering a surface around large density cell groups, especially when multiple cell types are concerned. Similar to the 'differential-density 3D scatter plot' a selected set of voxels are visualized. However, instead of representing them by single points the algorithm renders a surface around voxels of larger than certain cellular density. The procedure of subdividing the 3D database onto voxels (unit spaces) and calculating the voxel cell densities is identical to that of the 'differential-density scatter plot'. Conversion of the 3D point-coordinate database, where the entries are the cells, to a density data constructs a volumetric database. In the volumetric data the entries are voxels defined by their 3 coordinates and the associated cell densities. Then a density threshold is defined and voxels larger than the threshold are identified. The algorithm renders a 3D skeleton and determines a 2D manifold on the skeleton that is defined by interconnecting points that separate the higher density space from lover density space. The manifold is further partitioned onto triangles and surface elements are rendered to each of these triangles. These surface elements are then smoothened and reflectance property as well as light source are defined. For surface rendering the C++ program and the 3D-visualization toolbox of Matlab R11® (MathWorks, Inc.) were used. Figure 3B was generated by this method.
10.4.3. Iso-relational Surface Rendering
Similar to the 'iso-relational scatter plot' the aim of this representation is to show the co-distributive association between different cell types or other variables. In contrast to the 'iso-relational scatter plot', this plot renders a surface around the population of cells where certain density ratio is detected. Since the association of different cell types have a typically complicated spatial configuration the scatter plot of neuronal markers does not reveal the true 3D structure. In order to reduce the complexity, voxels, where a certain density ratio of two cell types is established, are rendered with a surface. This surface separates cells where certain density ratio is higher than a critical value. The unique feature of the 'iso-relational surface rendering' method is the visual representation of abstract relationships that is more important for understanding functional connections between neurons than exact locations of cell bodies. Technically, the 3D database of cell bodies is subdivided into unit spaces for both cell types. Then cell density ratios are calculated within each unit space (voxel) shared between the two cell types. The density ratios are arranged in a 3D matrix containing various ratios. Iso-density demarcation lines are calculated and rendered by a surface in such a way that cell bodies with density ratios larger than a specific number are covered by the surface. Density ratios smaller than the critical one are located outside of the surface. For visualization purposes, a range of critical density values must be applied for testing the integrity of clouds and to make sure that there are no hollow spaces covered. The algorithm of surface rendering is the same as the one described at the 'iso-density surface mapping'. Complex relationships between multiple components such as density relations of multiple cell types can be decomposed into pairwise relations and visualized as merged surfaces. Color-coding of surfaces of different cell types helps to interpret complicated arrangements. The plots in Figure 4C-D were generated according to this method. A more detailed description of the methods under 10.4.1-10.4.3 is given elsewhere [41].
10.4.4.
Mean 3D Vector of Dendritic Processes
Individual
dendritic branches may have a principal orientation adapted to making contacts
with also oriented axons, independent from the orientation of dendritic mass.
To test this, the principal orientation of dendritic branches was expressed by
the mean branch orientation and approximated by the average orientation and
average length of the dendritic tree. The orientation and length were combined
into a V(P0,P1) vector
originated from the point P(x0,y0,z0)i where the dendrite stemmed from
the cell body and pointed to the P(x1,y1,z1)i point that represented the average
length and orientation. In this
analysis we first calculate the dxi,dyi,dzi vectors as Euclidean distances
between adjacent branch points and the average of the dxi,dyi,dzi
is used as a single vector to represent the main orientation
tendency of the dendrite. Since multiple origins of dendritic processes were
possible, the orientation vector was calculated separately for each main branch
resulting in different vectors per cells originating from nearby points.
Parsing of the Neurolucida® data files and computation of vectors was all
carried out by custom written C++ codes and compiled for Silicon Graphics and
Pentium class computers using Irix and Linux operating systems, respectivelly.
For visualization purposes vectors were rescaled by a common multiplicative
factor that made it easier to appreciate the main tendency of orientation. 3D
aspects of the vector space were constructed by superposition of the vectors on
the structure outlines (such as anterior commissure). Rotation and navigation
in the database using MatlabŇ 3D
visualization interface made it possible to gain insights of the vector
orientation in the denser cell clusters. This type of analytical tool was
applied to generate Figures 5A and 6A.
The algorithm (polar
histogram) supplied in the NeruolucidaŇ software
package is similar to the analysis described by McMullen et al. [47]. The
difference is that in our case the results are collected by a computerized
system and artifacts of that collection process need to be filtered out. The
algorithm for polar histogram breaks up the dendritic processes into line
segments and determines the directions that these line segments point and the
lengths of these segments. The sum of the lengths of the small line segments is
approximately the same as the length of the original tracing. The direction of
the vector is calculated by projecting the line segment onto the plane of the
sectioning. This is accomplished with the arc-cosine function. The histogram
represents the total length by the distance from the origin and angle (theta)
that the vector makes with the x axis plotted in the radial direction. Each sector in the polar histogram is the
sum of all the dendritic growth in that particular range of angle. There is a
unique value of the polar histogram for each value of the angle theta. Some
information has been lost because the dendrites that are traced in the
sectioning plane are always going to be longer than dendrites in the plane
perpendicular to the sectioning plane. Therefore, this type of analysis is
useful primarily to characterize the orientation of dendrites that are roughly
co-planar. NeurolucidaŇ allowed us to
outline and select cell populations based on anatomical markers from single
sections or a stack of sections and to construct a polar histogram of the
dendrites pooled together from all the marked neurons.
10.4.6. Comparison of 3D location of labeled cells form different brains
After mapping labeled cell bodies, the NeurolucidaŇ files were transferred to a Silicon Graphics (Octane) workstation and formatted for further analysis and 3D visualization using the Micro3D (Oslo Research Park) program. In order to compare data from several brains with multiple retrograde tracer injections, each section was visually aligned to the corresponding map of a "master" brain with the aid of surface contours, and fiducial markers, including, the corpus callosum, anterior commisure, internal capsule, stria medullaris, stria terminalis, and the fornix. To create a maximum fit, an interactive procedure was used, including moving, rotation and shrinkage corrections along the x, y, and z axes. To avoid gaps between sections in the visualizations, individual cells were randomized in a 400 x 400 x 50 μm space (z-spread). Figures 2 A-B and 3A were generated according to this procedure.
10.4.7. Overlap
Analysis
The degree of overlap between two neuronal populations was estimated by subdividing individual sections into an array of 500 x 500 μm voxels and counting the number of digitized coordinate pairs (cell) per voxel using a custom made program similar to the one applied by Alloway et al. [40]. To avoid analyzing areas with low density of cells, only voxels containing 3 or more cells were included. Voxels containing a defined number of 'population 1' or 'population 2' cells are dark gray (red in the color version) or light gray (blue in the color version), respectively, while voxels containing a similarly defined number or more cells of both categories (at least three of each) are labeled white. The number of differently labeled voxels are counted for each section and also summed across sections and used to estimate the percentage of overlapping voxels and also the percentage of a given cell population in the overlapping voxels. The charts in Figure 2C-D are from this material.
10.4.8. Merging files containing cells of different
complexities
Figures 9 C and D were prepared merging two different datafiles: one derived from a series of sections containing four different cell populations in the BF using immunostaining for parvalbumin (PV), calretinin (CR), calbindin (CB), and choline acetyltransferase (CH) (‘four marker’ brain). The other file contained a single electrophysiologically and chemically identified neuron (NPY or CH) with their axonal ramifications and dendritic trees digitized from a series (n= 10-100) of 50 μm thick sections.
The cell mapping and anatomical landmarks were extracted from the 'four marker' brain, in which four series of alternate sections (n=48) were stained with antibodies against CH, PV, CR and CB. The distance between two consecutive sections stained with identical markers was 300 μm. Adjacent four sections containing markers for PV, CR, CB, and CH were aligned using standard anatomical landmarks (i.e., corpus collosum, lateral ventricle, fornix, thalamus, optic chiasm) and collapsed into a single section by removing the within section depth coordinates but preserving the x and y coordinates of the cells, resulting in a 3D series of 2D layers. The distance from bregma of each of this composed layers was calculated using the average of the original four sections. This way we created a set of 12 layers, each containing four different cell populations with their original x, y coordinates. The same dataset was the basis for the analysis documented in Figure 4.
Reconstruction of single identified cells was achieved by routine procedures as described in the literature (for refernces see [61]). Since the tissue sections contain only one stained neuron, all axonal and denritic processes can be followed through a series of adjacent sections. After determining the distance from bregma of the single reconstructed cell body, the corresponding section from the 'four marker’ brain was merged into the section that contained the single identified cell body. The axonal arbor fields of the single identified cells were outlined and the number and cell types that were enclosed were extracted from NeurolucidaŇ database. Data from the cell marker and fractial analysis of the axon arbor was then used to estimate the approximate numbers and types of cells that may be embedded in the axonal ramification space of the single reconstructed cell and could come into contact with it.
ACKNOWLEDGEMENT
The
research summarized in this review was supported by NIH Grant NS23945 to
L.Z. We wish to thank Mr. Jack R.
Glaser, President, MicroBrightField , Inc. with whom and other stuff members of
his company our interaction has been excellent over many years.
FIGURE LEGENDS
Figure 1. (A) 3D wireframe diagram showing the distribution of cholinergic neurons in the basal forebrain. Cholinergic cells (dots) were mapped from 12 sections, approximately 300 µm apart. The contours of the corpus callosum and the section outlines are marked. (B) Composite map illustrating the dendritic architecture of the basal forebrain cholinergic system. The location of panel (B) corresponds to the enclosed box in (A). Dendrites of approximately 1,300 cholinergic neurons were traced from 7 coronal sections. Diagonal white lines delineate the approximate location of the corresponding major forebrain areas. HDB= horizontal limb of the diagonal band; ic= internal capsule; MS/VDB= medial septum/vertical limb of the diagonal band; SI= substantia innominata. Scale bar: 1 mm (applies only to B).
Figure 2. (A) Distribution of
non-cholinergic neurons projecting to the medial prefrontal cortex (light) and
the border region between M1 and M2 region (dark). Note the substantial overlap of light and dark cells in the
rostral (right hand side of the model) basal forebrain. Fluoro-Gold was
injected into two sites in the prefrontal cortex and Fast Blue into the border
of the M1/M2 regions (upper left
insets). (B) Distribution of cholinergic neurons projecting to the
somatosensory (light) and the M1/M2 association region (dark). Note the
apparent minimal overlap between the light and dark symbols in the basal
forebrain. (C) Overlap analysis from selected sections of case shown in
(A). (D) Overlap analysis from the case depicted in (B). For (C) and (D) each section was subdivided
into 500 x 500 x 50 µm voxels and the number of cells from each of the two
populations (populations '1' and populations '2') was counted in each voxel.
Unit spaces containing at least 3 cells of either population are marked with
light gray and dark gray, respectively; those containing at least three of both
marker types are marked in white. Note the substantial overlap in (C), as
indicated by the white voxels. In (D),
no white bins are detected indicating no overlap in this case. Note that the
gray scalings (colors in the companion CD-ROM file) of the voxels here
represents population '1' and/or population '2' and does not correspond to the coding in (A) and (B). The corpus callosum
is rendered by double gray/white
surfaces around the cingulum bundle in the 3D models.
Figure 3. (A) Composite map showing the 3D distribution of cholinergic cells projecting to four, arbitrarily defined medio-lateral sectors of the neocortex. In the color version of this figure (accompanying CD-ROM) cells projecting to different regions are color-coded medial: red, intermediary sector: blue and yellow and lateral parts of the neocortex: green). Note the relatively ordered rostro-medial to caudo-lateral distribution of cells that project to medio-laterally located cortical areas. Dark (red) symbols in the lower right side of the model are rostral. Medial is right, lateral is left. (B) Iso-density surface rendering to show the major organizational features in the BF. Unit space: 400 x 400 x 50 µm, density threshold >2 cell /voxel. For appreciation of the different cell groups see the color version of this figure where dark blue surface covers unit spaces that contain cholinergic cells projecting to the posteromedial (M1/M2) cortex; yellow: medial prefrontal cortex; red: barrel cortex; green: posterior insular-perirhinal; light blue: agranular insular-lateral orbital; magenta: lateral frontal (motor) cortex. The iso-relational rendering of part B is placed into the wireframe of the section outlines and the corpus callosum to show their real position in the original brain. Note that the view in (B) is a mirror image of (A). Here and at the rest of the 3D representations the numbers along the z axis are the layers (sections) and the x, and y values correspond to the voxel indices.
Figure 4. Differential density scatter plots and iso-relational surface mapping. (A) and (B) represent the spatial distribution of cholinergic (dots in A) and parvalbumin (dots in B) cells from the same brain showed separately. Filled circles mark the high density locations where the density of cholinergic or parvalbumin cells is higher than 15 cells in the unit space (250 x 250 x 50 µm). (C) The scatter plots of both cholinergic (red in the color version of this figure) and parvalbumin (green) cells are superimposed on the iso-relational surface (dark solid area; violet in the CD-ROM file) where the density of both the cholinergic and parvalbumin cells is > 5 and the ratio of cholinergic/parvalbumin cells is at least 0.5 or higher. (D) Merging the cholinergic/parvalbumin, cholinergic/calbindin and cholinergic/calretinin iso-relational surfaces (using cell density ł5 in the unit space) into one scheme reveals that the cholinergic 'column' can be parcellated into clusters of different sizes. Different shading of surfaces cover the spaces where the relationship of cholinergic cells to parvalbumin (green in the color version), calretinin (yellow) and calbindin (blue) neurons is similar. (0.5 or higher).
Figure 5. (A) Mean orientation of dendritic branches. The initial segments of dendrites are represented by dots. The outlines of the anterior commissure (ac) are indicated by small dots. (B) Differential density scatter plot of the same database. Dots represent cholinergic cells (n= 15,700), filled circles mark the high density locations where the density of cells is ł 20 per unit space (250 x 250 x 100 µm). Flakes are due to the section steps along the z axis. Cells and their dendrites were mapped from 34 consecutive horizontal sections stained for choline acetyl- transferase. The comparison of (A) and (B) suggests the iso-orientation of dendrites in the high density cell cluster.
Figure 6. Comparison of the dendritic orientation derived from polar histograms with the orientation vectors. (A) Part of the septal area from Fig. 5A as viewed from the sagittal direction. C, D, E with arrows point to regions that may correspond to the same cell populations as selected for the polar histograms from sagittal sections of a different brain as shown in (B). (B) 137 cholinergic cells (filled circles) were selected from a stack of sagittal sections comprising the septal region (n= 2,266 cholinergic cells, dots) for dendritic tracing. Letters C, D, E and F mark boxes that were used to select dendrites for the orientation analysis. In both figures ac indicates the location of the anterior commissure. Despite slightly different orientation of the sagittal sections, one can recognize the same cell groups as seen in the 3D rendering. (C) - (F) Polar histograms representing dendritic orientation from indicated areas. Numbers at upper right indicate the number of dendritic segments in the sample. Numbers along the circles within the polar histograms mark distances in μm from the origin (see APPENDIX for explanation).
Figure 7. Differential distribution of various afferents in the cholinergic forebrain. (A) – (B)-(C), (G), (I) Composite maps illustrating putative zones of contacts between afferent fibers and cholinergic neuronal elements following PHA-L injections into the (A) far-lateral hypothalamus, (B) mid-lateral hypothalamus, (C) medial hypothalamus, (I) locus coeruleus. (G) shows the distribution of putative contact sites from a material stained for dopamine-β-hydroxylase and choline acetyltransferase. (H): PHA-L labeled terminal varicosities (arrow) in close apposition to a proximal dendrite of a cholinergic neuron. The grid simulates the ocular reticle used to screen sections for high magnification (63x) light microscopic analysis. One division of the grid = 16 µm. Cholinergic neurons are represented by dots. Zones of putative contacts between cholinergic elements and terminal varicosities are depicted as solid squares (corresponding to 80 x 80 µm areas in the section). (D), (E), (F) location of labeled cells at the PHA-L injection sites from cases depicted in (A), (B), (C). Panels (A), (B), (C), (H) are modified from Cullinan and Zaborszky [57] with kind permission from Wiley-Liss. Panels (G) and (I) are modified from Zaborszky et al. [51], with permission from Elsevier Science.
Figure 8.
Comparison of the 2D orientation of cholinergic dendritic
segments (A) and calcitonin-gene-related-peptide (CGRP)-containing axons
(B) in the internal capsule (ic). (C) and (D) polar
histograms of cholinergic dendrites (C)
and CGRP axonal ramifications (D) from
the same general area. Note that the majority of dendrites and axons occupy the
same sector of the polar histograms. Upper right numbers indicate the number of
segments in the analysis. The outer circle of the polar histogram correspond to
a 1,200 μm diameter around the origin.
Figure 9. Distribution of different cell types in the neighborhood of identified NPY (A, C, E) and cholinergic (B, D, F) neurons. (A) and (B) Schematic drawings illustrating the location of the electrophysiologically and chemically identified neurons. (C) and (D) Coronal view of the identified neurons embedded into the same general region of the basal forebrain derived from a different database that contains four different cell populations. Filled circles: parvalbumin; up triangles: calretinin; down triangles: calbindin; squares: cholinergic neurons. The approximate number of cell bodies from each cell populations that can be found in the 3D volume of the single cell axonal arbor is indicated below. (E), (F) Enlarged view of the boxes from (C) and (D). Thicker, black indicates dendritic processes and thinner lines indicate axonal ramifications. Note the small varicosities along the axonal collaterals correspond to putative synaptic boutons.
11. REFERENCES
[1] Dunnett
S.B. and Fibiger H.C. (1993) Role of forebrain cholinergic system in learning
and memory: relevance to the cognitive deficits of aging and Alzheimer’s
dementia. Progr. Brain Res. 98, 413-420.
[2] Everitt, B.J. and Robbins T.W. (1997) Central cholinergic systems and cognition. Annu. Rev. Psychol. 48: 649-684.
[3] Heimer, L., de Olmos, J., Alheid, G.F. and Zaborszky L. (1991) "Perestroika" in the basal forebrain; Opening the border between Neurology and Psychiatry. Progr. Brain Res. 87: 109-165.
[4] Zaborszky L., Pang, K., Somogyi, J., Nadasdy, Z., Kallo, I. (1999) The basal forebrain corticopetal system revisited. Ann NY Acad Sci 877:339-367
[5] Jones, B.E., Muhlethaler, M. (1999) Cholinergic and GABAergic neurons of the basal forebrain: role in cortical activation, in Handbook of Behavioral State Control - Cellular and Molecular Mechanisms (Lydic, R., Baghdoyan, H.A., eds.) CRC Press, New York, pp. 213-234.
[6] de Lacalle, S., Saper, C.B. (1997) The cholinergic system in the primate brain: Basal forebrain and pontine-tegmental cell groups, in Handbook of Chemical Neuroanatomy. The Primate Nervous System, Part I, (Bloom, F.E., Bjorklund, A. and Hokfelt T., eds.) Elsevier, New York, pp. 217-252.
[7] Geula, C., Mesulam, M.M. (1994) Cholinergic systems and related neuropathological predilection patterns in Alzheimer disease, in Alzheimer Disease (Terry, R.D., Katzman, R., Bick, K.L., eds.) Raven Press, New York, pp. 263-291.
[8] Swaab, D.F. (1997) Neurobiology and neurophatology of the human hypothalamus, in The Primate Nervous System, Part I. Handbook of Chemical Neuroanatomy, vol. 13 (Bloom, F.E., Bjorklund, A., Hokfelt, T., eds.) Elsevier, New York, pp. 39-118.
[9] Whitehouse, P.J., Price, D.L., Struble, R.G., Clark, A.W., Coyle, J.T., Delong, M.R. (1982) Alzheimer's disease and senile dementia: loss of neurons in the basal forebrain. Science 215, 1237-1239.
[10] Gritti, I., Mainville, L., Mancia, M., and Jones, B. (1997) GABAergic and other noncholinergic basal forebrain neurons, together with cholinergic neurons, project to the mesocortex and isocortex in the rat. J. Comp. Neurol. 383: 163-177.
[11] Zaborszky, L. (1989) Afferent connections of the forebrain cholinergic projection neurons, with special reference to monoaminergic and peptidergic fibers in Central Cholinergic Synaptic Transmission (Frotscher, M. and Misgeld, U. eds.). Birkhauser, Basel, pp.12-32,
[12] Heilman, K.H., Watson, R.T., and Valenstein, E. (1993) Neglect and related disorders, in Clinical Neuropsychology, (Heilman, K.M. and Valenstein, E., eds.) Oxford University Press, New York, NY. Pp. 279-336.
[13] Mesulam, M.M. (2000) Attentional networks, confusional states, and neglect syndromes, in Principles of Behavioral and Cognitive Neurology (Mesulam, M.M., ed.), Oxford, NY, pp.174-293.
[14] McGaughy, J., Everitt, B.J., Robbins, T.W., and Sarter, M. (2000) The role of cortical cholinergic afferent projections in cognition: impact of new selective immunotoxins. Beh. Brain Res. 115, 251-263.
[15] Szentagothai, J. (1983) The modular architectonic principle of neural centers. Rev. Physiol. Biochem. Pharmacol. 98, 11-61.
[16] Gerfen, C.R. (1985) The neostriatal mosaic. I. Compartmental organization of projections from the striatum to the substantia nigra in the rat. J. Comp. Neurology 236, 454-476.
[17] Bjaalie, J.G., Diggle, P.J., Nikundiwe,
A., Karagulle, T., Brodal, P. (1991) Spatial segregation between populations of
ponto-cerebellar neurons: Statistical analysis of multivariate interactions. Anat.
Rec. 231, 510-523.
[18] Graybiel, A.M., Penney, J.B. (1999) Chemical architecture of the basal ganglia, in The Primate Nervous System, Part III, Handbook of Chemical Neuroanatomy, Vol 15, (Bloom, F.E., Bjorklund, A., and Hokfelt, T., eds.) Elsevier, Amsterdam, pp.227-284.
[19] Mountcastle, V.B. (1998) Perceptual Neuroscience. The Cerebral Cortex. Harvard University Press, Cambridge, MA.
[20] Malach, R. (1992) Dendritic sampling across processing streams in monkey striate cortex. J. Comp. Neurol. 315,303-312.
[21] He, S-Q., Dum, R.P., and Strick, P.L. (1993) Topographic organization of corticospinal projections from the frontal lobe: Motor areas on the lateral surface of the hemisphere. J. Neurosci. 13, 952-980.
[22] Malach, R. (1994) Cortical columns as devices for maximizing neuronal diversity. Trends Neurosci. 17, 101-104.
[23] Malmierca, M.S., Blackstad, T.W., Osen, K.K., Karagulle, T., and Molowny, R.L. (1993) The central nucleus of the inferior colliculus in rat: A Golgi and computer reconstruction study of neuronal and laminar structure. J. Comp. Neurol. 333, 1-27.
[24] Malmierca, M.S., Leergard, T.B., Bajo, V.M., Bjaalie, J.G., and Merchan, M.A. (1998) Anatomic evidence of a three-dimensional mosaic pattern of tonotopic organization in the ventral complex of the lateral lemniscus in cat. J. Neurosci. 18:10603-10618
[25] Leergaard, B.T., Alloway, K.D., Mutic, J.J., Bjaalie, J.G. (2000) Three-dimensional topography of corticopontine projections from rat barrel cortex: correlations with corticostriatal organization. J. Neurosci. 20, 8474-8484
[26] Jacobs, G.A. and Theunissen, F.E. (2000) Extraction of sensory parameters from a neural map by primary sensory interneurons. J. Neurosci. 20, 2934-2943
[27] Sofroniew, M.V., Eckenstein, F., Thoenen, H., and Cuello, A.C. (1982) Topography of choline acetyltransferase-containing neurons in the forebrain of the rat. Neurosci. Lett. 33, 7-12.
[28] Armstrong, D.M., Saper, C.B., Levey, A.I., Wainer, B.H., and Terry, R.D. (1983) Distribution of cholinergic neurons in the rat brain demonstrated by immunohistochemical localization of choline acetyltransferase. J. Comp. Neurol. 216:53-68.
[29] Mesulam, M.M., Mufson, E.J., Wainer, B.H., and Levey, A.I., (1983) Central cholinergic pathways in the rat: an overview based on an alternative nomenclature (Ch1-Ch6). Neurosci. 10,1185-1201.
[30] Rye, D.B., Wainer, B.H., Mesulam, M.-M., Mufson, E.J., and Saper, C.B. (1984) Cortical projections arising from the basal forebrain: a study of cholinergic and noncholinergic components combining retrograde tracing and immunohistochemical localization of choline acetyltransferase, Neurosci. 13,627-643.
[31] Amaral, D.G. and Kurz, J. (1985) An analysis of the origins of the cholinergic and non-cholinergic septal projections to the hippocampal formation in the rat. J. Comp. Neurol. 240, 37-59.
[32] Carlsen, J., Záborszky, L., and Heimer, L. (1985) Cholinergic projections from the basal forebrain to the basolateral amygdaloid complex: a combined retrograde fluorescent and immunohistochemical study. J. Comp. Neurol. 234,155-167.
[33] Záborszky, L., Carlsen, J., Brashear, H.R., and Heimer, L. (1986) Cholinergic and GABAergic afferents to the olfactory bulb in the rat with special emphasis on the projection neurons in the nucleus of the horizontal limb of the diagonal band. J. Comp. Neurol. 243, 488-509.
[34] Woolf, N.J. (1991) Cholinergic system in mammalian brain and spinal cord. Progr. Neurobiol. 37, 475-524.
[35] Wainer, B.H., Steininger, T.L., Roback,
J.D., Burke-Watson, M.A., Mufson, E.J., Kordower, J. (1993) Ascending
cholinergic pathways: functional organization and implications for disease
models. Progr. Brain Res. 98, 9-30.
[36] Mesulam, M.M., and Geula, C., (1988) Nucleus basalis (Ch4) and cortical cholinergic innervation in the human brain: observations based on the distribution of acetylcholinesterase and choline acetyltransferase. J. Comp. Neurol. 275, 216-240.
[37] Baskerville, K.A., Chang, H.T., and Herron, P. (1993) Topography of cholinergic afferents from the nucleus basalis of Meynert to representational areas of sensorimotor cortices in the rat. J. Comp. Neurol. 335, 552-562.
[38] Zaborszky, L., Duque, A. (2000) Local synaptic connections of basal forebrain neurons. Behav. Brain Res. 15, 143-158.
[39] Felleman, D.J. and Van Essen, D.C. (1991) Distributed hierarchical processing in the primate cerebral cortex. Cerebral Cortex 1, 1-47.
[41] Nadasdy, Z. and Zaborszky, L. (2001) Compuattional analysis of spatial organization of large scale neural networks. Anat Embryol. (in press)
[42] Valverde, F. (1961) Reticular formation of the pons and medulla oblongata. A Golgi study. J. Comp. Neurol. 116, 71-99.
[43] Leontovich, T.A. and Zhukova, G.P. (1963) The specificity of the neuronal structure and topography of the reticular formation in the brain and spinal cord of carnivora. J. Comp. Neurol. 121, 347-381.
[44] Ramon-Moliner, E. and Nauta, W.J.H. (1966) The isodendritic core of the brain stem. J. Comp. Neurol. 126, 311-336.
[45] Das, G.D. and Kreutzberg, G.W. (1968) Evaluation of interstitial nerve cells in the central nervous system. Adv. Anat. Embryol. 41, 1-58.
[46] Zaborszky, L. (1992) Synaptic organization of basal forebrain cholinergic projection neurons, in Neurotransmitter Interactions and Cognitive Functions. (Levin, E., Decker, M., and Butcher, L. eds.) Birkhauser, Boston, pp. 27-65.
[47] McMullen, N.T., Goldberger, B., Suter, C.M., and Glaser, E.M. (1988) Neonatal deafening alters nonpyramidal dendrite orientation in auditory cortex: A computer microscope study in the rabbit. J. Comp. Neurol. 267, 92-106.
[48] Csillik,
B., Rakic, P., and Knyihar-Csillik, E. (1998) Peptidergic innervation and the
nicotinic acetylcholine receptor in the primate basal nucleus. Eur. J.
Neurosci. 10: 573-585.
[49] Braitenberg, V. and Schutz, A. (1998) Cortex: Statistics and geometry of neuronal connectivity. Springer, Berlin.
[50] Szentagothai, J. (1990) “Specificity versus (quasi-) randomness” revisited. Acta Morph. Hung. 38, 159-167.
[51] Zaborszky, L., Cullinan, W.E., and Luine,
V.N. (1993) Catecholaminergic-cholinergic interaction in the basal forebrain. Prog.
Brain Res. 98: 31-49.
[52] Zaborszky, L., Cullinan, W.E., and Braun, A. (1991) Afferents to basal forebrain cholinergic projection neurons: An update, in Basal Forebrain: Anatomy to Function. (Napier T. C., Kaliwas P.W., and Hanin, I., eds.). Plenum Press, New York, NY, pp. 43-100
[53] Zaborszky, L., Gaykema, R.P., Swanson, D.J., and Cullinan, W.E. (1997) Cortical input to the basal forebrain. Neuroscience 79, 1051-1078.
[54] Pang, K., Tepper, J.M., and Zaborszky, L. (1998) Morphological and electrophysiological characteristics of non-cholinergic basal forebrain neurons. J. Comp. Neurol. 394, 186-204
[55] Duque, A., Balatoni, B., Detari, L., Zaborszky, L. (2000) EEG correlation of the discharge properties of identified neurons in the basal forebrain. J. Neurophysiol. 84, 1627-1635.
[56] Schwaber, J.S., Rogers, W.T., Satoh, K., and Fibiger, H.C. (1987) Distribution and organization of cholinergic neurons in the rat forebrain demonstrated by computer-aided data acquisition and three-dimensional reconstruction. J. Comp. Neurol. 263, 309-325.
[58] Gaykema, R.P.A. and Zaborszky, L. (1996)
Direct catecholaminergic-cholinergic interactions in the basal forebrain: II.
Substantia nigra and ventral tegmental area projections to cholinergic neurons.
J. Comp. Neurol. 374, 555-577.
[59] Glaser, J.R. and Glaser, E.M. (1990) Neuron imaging with Neurolucida – a PC-based system for image combining microscopy. Comput. Med. Imaging Graph. 14, 307-317.
[61] Heimer, L. and Zaborszky, L. (eds.) (1989) Neuroanatomical Tract-Tracing Methods 2. Recent Progress. Plenum Press, New York.