вход по аккаунту


Spatial segregation between populations of ponto-cerebellar neuronsStatistical analysis of multivariate spatial interactions.

код для вставкиСкачать
THE ANATOMICAL RECORD 231:510-523 (19911
Spatial Segregation Between Populations of Ponto-Cerebellar
Neurons: Statistical Analysis of Multivariate Spatial Interactions
Department of Anatomy, Institute of Basic Medical Sciences, Oslo, Norway (J.G.B., A.N.,
T.K., P.B.), and Mathematics Department, School of Engineering, Computing, and
Mathematical Sciences, Lancaster University, Lancaster, United Kingdom (P.J.D.)
This study applies terms and methods for describing spatial interactions between multivariate spatial point patterns, which are, to our knowledge, new in neurobiology. We consider two categories of points, type 1 and 2,
distributed within a certain reference volume (such as a nucleus of the brainstem
or a cortical area). The points may, for example, represent different categories of
labelled cells or axonal fields of termination. We say that there is spatial neutrality
between points of type 1 and 2 if the types are signed by random labelling. If a
mechanism drives the two point categories together, we say that the point patterns
are positively associated. Conversely, if a mechanism drives type 1 and 2 points
apart, we say t h a t they are segregated. By comparing two cumulative distribution
functions of distances between points, we can distinguish neutrality, positive association, and segregation. One function, H,,(t), is the cumulative distribution
function of the distance t between a pair of randomly selected points of type 1 and
2. The other, H,,(t), is the corresponding function for a pair of points randomly
selected without reference to type. Plots of the estimated difference between these
two functions give a n indication of positive association, neutrality, or segregation.
A statistical test, based on simulations of random (neutral) distributions, can be
used to see whether deviations from neutrality are significant.
We apply the analysis described above to a major pathway of the brain, namely
the ponto-cerebellar projection. Different types of cells in the pontine nuclei are
retrogradely labelled with the fluorescent tracers Rhodamine-B-isothiocyanate,
Fluoro-Gold, and Fast Blue. The tracers are injected in adjacent or more distant
folia of the cerebellar paraflocculus. The location of the somata of labelled cells are
recorded and the total distribution reconstructed in three dimensions and displayed on a dynamic graphics workstation. We ask whether different units (folia)
in the paraflocculus receive information from the same population, from two different positively associated populations, or from segregated cell populations. We
find a statistically significant tendency for cell populations projecting to adjacent
folia to be positively associated, although there are few cells containing multiple
labels. Populations of neurons projecting to folia wider apart are significantly
segregated. From inspections of the reconstructions, using real-time rotations, we
find that the swarms of labelled neurons tend to accumulate in shells or lamellae
in the pons. Within the lamellae, the cells are aggregated in clusters and bands
with empty holes (containing unlabelled ponto-cerebellar cell bodies, presumably
projecting to other cerebellar targets) in between. By determining the average
distance to a reference plane for each cell population, we find that cell populations
shift in a ventro-medial direction as the injection sites move from the medial part
of the dorsal paraflocculus toward the lateral part and into the ventral paraflocculus. We therefore conclude that there is a continuous shift in location of pontocerebellar cell populations, corresponding to specific shifts in cerebellar target
To understand the wiring patterns of the brain in
functional terms depends on comprehensive and reliable
of the topological arrangement Of neurOnS and terminal axonal fields. In traditional neuroanatomy, neuronal somata or fields of termination
Received January 30, 1991; accepted March 22, 1991.
Address reprint requests to Jan G. Bjaalie, Dept. of Anatomy, University of Oslo, P.O. BOX1105 Blindern, N-0317 Oslo 3, Norway.
(the position of which we refer to as events) are presented as points in two-dimensional drawings of tissue
sections. The shape and arrangement of individual categories of points and their spatial interactions and relationships are then described on the basis of such
drawings, usually in a qualitative manner. With the
use of computer technology and three-dimensional reconstructions, however, i t is possible to improve the
visualization of shape and arrangement. Furthermore,
on the basis of a file of data, one may even employ
formal statistical tests, e.g., to distinguish random and
clustered patterns (cf. Bjaalie and Diggle, 1990; Bjaalie
et al., 1990). Very little has been done, however, on
quantitative estimations of the spatial interactions between different categories of events.
Currently, the terms overlap and segregation (or separation) are often used to describe spatial relations
(see, e.g., Bjaalie and Brodal, 1989; Goldman-Racik
and Selemon, 1990). When events of type 1 and 2 occupy the same space in some parts of the reference
volume, we usually say t h a t overlap occurs. When they
do not occupy the same space, we say t h a t the two types
of events are segregated. This terminology, however, is
not very precise. For example, if there is spatial neutrality between two point patterns, both overlap and
segregation (in the sense used above) may nevertheless
occur. Thus “overlap-by-chance’’ is different from overlap caused by a mechanism driving two types of events
together in space. We introduce the term positive association to describe the tendency for different point patterns to be located together in space. Segregation between clustered point patterns may also occur simply
because, by chance, clusters of different categories fail
to overlap. In contrast, a mechanism during development may be responsible for segregation. Mathematical analyses are required both to distinguish “overlapby-chance’’ from positive association, and “segregationby-chance” from “real” segregation between two point
patterns. Along the same lines of reasoning, we need
measures of the degree of positive association or segregation.
The present study aims at establishing more precise
terms as well as procedures for quantitative analyses of
spatial interactions in two- and three-dimensional
space. We have initially applied such analyses on the
cortico-ponto-cerebellar system, which provides the
quantitatively dominating input to the cerebellum (for
reviews on basic organization, see Brodal 1982, 1987;
Brodal and Bjaalie, 1987). In this article, we study the
amount of positive association or segregation between
well-defined populations of ponto-cerebellar neurons.
Thus we consider the two- and three-dimensional organization of the neurons of origin of the projection to
a limited and macroscopically well-defined region of
the cerebellum-the paraflocculus (see, e.g., Larsell,
1970). The paraflocculus has a regular folial pattern,
and the dorsal and ventral paraflocculus are continuous. It is thus ideally suited for a comprehensive and
quantitative study on how the distribution of pontocerebellar cell somata varies with the location of their
cerebellar target region.
We ask whether different units of the paraflocculus
(single folia or smaller) receive information from the
same populations, from positively associated populations of ponto-cerebellar neurons, or if not, to what de-
gree they are segregated. To this end, we inject two or
three different retrograde fluorescent tracers into separate folia. A priori, we assume that a high frequency
of double- or triple-labelled neurons together with
positive association between differently labelled cells
in the pontine nuclei indicate that the same information is conveyed to the various folia injected. The converse would be a situation with few or no double- or
triple-labelled neurons, and with spatially segregated
populations, indicating the existence of separate channels for the transfer of information from the pontine
nuclei to the cerebellar cortex.
Our hope is that such quantitative information and
the use of more precise terminology to describe spatial
interactions may provide new insight that in t u r n will
improve our functional understanding of the connectivity of the brain.
This study employed 15 adults cats of both sexes,
ranging from 1.9 to 5.1 kg in weight. Under deep pentobarbital anaesthesia (30 mglkg body weight, administered intraperitoneally), supplemented with 0.2-0.5
ml of Hypnorm (Janssen), the posterolateral quadrant
of the cerebral hemisphere was removed to expose the
cerebellar paraflocculus. Aqueous solutions of fluorescent tracers, 2% Rhodamine-B-isothiocyanate (Sigma),
2% or 6% Fluoro-Gold (Fluorochrome Inc.) and 2% Fast
Blue (Sigma), were injected into different folia of the
dorsal and ventral paraflocculus with a Hamilton syringe fitted with a glass micropipette. Small amounts
of the tracers (0.03-0.10 pI) were deposited. A summary of the volumes and types of tracers used in the
individual animals is shown in Table 1. Injection sites
are presented in Figures 1 and 2.
After survival periods of 6-7 days, the cats were
anaesthetized and perfused transcardially with 1 litre
of warm physiological saline, followed by 2 litres of
warm 4% paraformaldehyde in 0.1 M phosphate buffer
(pH 7.4f, and finally with 1 litre of buffered cold 10%
sucrose. The brain was left in situ for 2-3 hours a t
room temperature to allow further fixation. It was then
dissected out and the cerebellum was photographed
and inspected macroscopically for obvious signs of the
injection site. RITC leaves a red spot at the injection
site, and thus the folium injected with this tracer was
sketched with respect to the other folia. The brains
were soaked overnight in 30%buffered sucrose at 4°C.
Some cases were prepared for three-dimensional computer reconstruction of the distribution of labelled
cells, and to secure perfect alignment of the sections,
fiducial marker holes were drilled in the tissue block
comprising the pontine nuclei. Sections were cut on a
freezing microtome at 40 pm. The pons was cut transversely and the cerebellum parasagittally.
average perpendicular distance to the ventromedial surface of the peduncle
dorsal paraflocculus
Fast Blue
ventral paraflocculus
TABLE 1. Case summaries
Volume injected (pl)
was collected in a separate well and mounted serially
for the purpose of complete reconstruction of the distribution of retrogradely labelled cells.
The unstained sections were studied with a Leitz incident light fluorescence microscope providing excitadP1
tion light of 530-560 nm for RITC labelled neurons
340-380 nm for F-G and FB labelled neurons. The
distribution of retrogradely labelled pontine cells as
well a s the boundaries of the cerebellar injection sites
were plotted at 30 x magnification with a n x-y recorder
attached to the microscope stage. All cells labelled with
a given tracer, including double labelled cells, were
For three-dimensional reconstructions, drawings of
complete series of sections (or every second section)
were aligned with the aid of the drilled holes, vessels,
surface contours. The location of each cell and a
few contours of the pontine nuclei for each case were
digitized and saved on file. The data files thus contain
x,y,z coordinates for each cell; the z coordinate was deVP
fined by the section number and thickness of the secVP
tions. Since a substantial number of labelled cells (up
to 14.078 for the RITC-500 population, cf. Fig. 1)was
plotted and subsequently digitized, a method for didP
rectly defining coordinates €or each cell without digitidP
zation was also developed. With this procedure, analog
signals from the microscope potentiometres were transdP
mitted directly to the computer via a n analog to digital
converter (for further details, cf. Bjaalie 1991).
The reconstructions were displayed on a Tektronix
XD88-34 3D graphics superworkstation, with the TekdP
tronix 4G graphics pipeline. Different populations of
labelled cells were coded with pseudo colors. Each popdP
ulation was inspected from different angles of view,
using real-time rotations in hardware, either alone or
together with the other populations reconstructed in
the same animal.
The statistical methods employed are described in
Crus I1
'Dorsal paraflocculus.
'Ventral paraflocculus.
36%concentration; otherwise, concentration for all tracers was 2%.
In most cases, the cerebellum was encased in a mixture of gelatin and 30% buffered sucrose and hardened
overnight in 4% paraformaldehyde prior to sectioning
on a freezing microtome (Switzer, 1979; Blackstad et
al., 1981). This embedding protocol prevents the cerebellar folia from falling apart during sectioning,
thereby making possible exact reconstruction of the injection sites.
In cases aimed only at providing information from
single sections (not computer reconstructed), adjacent
sections from the pons were collected in groups of five
in individual wells containing phosphate buffer. One
series was mounted on chrome-alum-coated slides, air
dried overnight, and cleared rapidly in xylene and coverslipped with Eukitt. Another series was counterstained with thionin, dehydrated in graded alcohols,
cleared also in xylene, and coverslipped with Eukitt. In
cases used for computer reconstructions (CCoL 488,
490, 495, 499, 500, and 5021, each section of the pons
Cell Distribution and Spatial Relationships Judged From
3-0 Reconstructions
Altogether, 14 populations of labelled cells in the
pontine nuclei were available as three-dimensional
computer reconstructions. Twelve of these-two in
each of cats CCoL 490,499, and 502, and three in each
of cats CCoL 495 and 500-comprise large parts (from
one-third to two-thirds) of the pontine nuclei. They
could thus be used to visualize the structural arrangement of pontine cells projecting to single folia in the
paraflocculus. Reproducible trends in spatial arrangement (clustering, banding, etc.) of the labelled cells, as
well as the numerical capacity of the projections, are to
be dealt with in detail elsewhere (A. Nikundiwe, J.G.
Bjaalie, and P. Brodal, in preparation). Quantitative
data on spatial interactions, including statistical tests
on positive association and segregation as well as attempts to measure a possible segregation process, are
considered below. Here, we give some comments on the
information directly available by inspection of the reconstructions.
Each of the reconstructed populations of labelled
neurons were studied on our graphics workstation.
They could thus be viewed from all possible angles.
Fig. 1. Location of injection sites in the paraflocculus (tracer-case number). For orientation, see the
flattened map of the cerebellum (upper right) with the paraflocculus drawn with thick lines.
Figure 3 shows a typical population of labelled neurons
viewed a t seven different angles. In Figure 3C, the
pons is seen from the top (rostral end), the plane of
sectioning is facing the viewer. The view demonstrated
in Figure 3G shows the same reconstruction rotated 90”
(compared to the situation in Fig. 3C) about the x axis,
so that the ventral surface of the pons is facing the
viewer. The labelled cells are seen forced into their
respective section planes (z coordinates were defined as
section number and thickness of the sections, see
above). Intermediate views, Figure 3D-F, strengthen
the impression that the cells are aggregated, with holes
or empty zones in between. In some places, the aggregates form snakelike patterns; in other places they are
more fractionated and occur as large clusters, often interconnected with tiny strands of cells. The shape of the
pontine nuclei and the corresponding curving of the
swarm of labelled cells make it difficult to distinguish
(on the two-dimensional pictures presented here) between real aggregation and pseudo-aggregation occurring in certain angles of view. Essential for the understanding of the arrangement of the labelled cells,
however, is the view presented in Figure 3A. Here, the
swarm of labelled cells appears as a narrow zone, curving along the mediolateral extent of the pontine nuclei.
We call this the angle of “maximum lamellarity.” The
reconstruction is seen from the rostral end of the pons,
a bit from the dorsal and lateral side. Thus in conclusion, each population of labelled cells tends to be confined to a lamella. Within a lamella, the cells are aggregated into clusters and bands with empty holes in
between (containing the somata of ponto-cerebellar
neurons presumably projecting to other cerebellar targets).
The spatial relationships between two (or more) populations of cells labelled with different tracers can to
advantage be studied with the graphics workstation.
Choosing different pseudo-colours to represent the different types of labelled cells, one gets a fairly good
visual impression of overlap and separation between
different populations in single animals. We have cho-
sen to present double-labelled cells as separate points,
one for each tracer. It should be noted that the number
of double labelled cells were generally less than 10%.’
Thus in zones of the reconstruction where overlap between different labels are seen, most of the overlap is
due to the fact that cells containing different labels are
located together. Whether or not the overlap observed
is simply “overlap-by-chance” or whether there is a
process driving driving cells of different categories together in space (positive association), cannot easily be
decided by inspection only. Interpretation of observed
segregation between differently labelled cells may also
pose problems. In all animals, there are clusters of spatially separated cells. In some cases, the separation is
so obvious and the degree of overlap so small that there
is no doubt that a segregation mechanism is responsible for the distribution patterns. In other cases, the
degree of separation is smaller, and there seems to be
more overlap. Consequently, it is difficult to decide
whether there is a segregation mechanism operating or
whether there is actually spatial neutrality between
the different cell populations.
In the color illustrations (Figs. 4,5), two typical cases
illustrating spatial relationships between multiple
populations are shown. Figure 4 illustrates a case in
which adjacent folia of the dorsal paraflocculus were
injected with different tracers. The two categories of
labelled cells are partly overlapping and partly separated. The angle of “maximum lamellarity” (defined
above) demonstrates that the two lamellae of cells are
similarly located. Figure 5 shows a case in which three
tracers were injected in the dorsal paraflocculus. Two
of the tracers were injected in adjacent folia, whereas
the third was injected many folia away from the others
(cf. Fig. 1: injection sites RITC-500, F-G-500, and
FB-500). A complicated three-dimensional mosaic of
patches or cell clusters is seen. At the angle of
’After control injections of two tracers in the same folium, more
than 80% of the labelled cells were double labelled.
C C o L 500
C C O L 490
Fig. 2. Drawings of sections through injection sites in two animals. Note that the total extent of each
injection site is always smaller than the size of a single folium, although some injections cover more than
one folium.
“maximum lamellarity,” the different categories of cells, also contains a n internal loose zone of cells close
cells appear as partly segregated. There seems to be a to the peduncle. The need for objective measures for
shift in location of the lamellae from a position fairly spatial interactions should be clear from the two examclose to the peduncle (the corticospinal and corticobul- gles demonstrated in Figures 4 and 5.
Our main impression, however, from inspection of
bar fiber tract) for the F-G-500 population (red dots),
to a site far more ventrally and medially for the the computer reconstructions and plots of single secRITC-500 population (blue dots). It seems t h a t the tions, is that there are certain reproducible trends in
third population, the FB-500, occupies a n intermedi- distribution of labelled ponto-cerebellar neurons. Popate position (yellow dots), but it is difficult to discern ulations projecting to the medial part of the dorsal
its precise relationship to the other two populations. paraf locculus are located fairly close to the peduncle,
Furthermore, the RITC-SO0 population (projecting to whereas populations projecting to more lateral parts
a folium in the “bend region” at the transition between are more externally situated (ventrally and medially).
the dorsal and ventral paraflocculus, cf. Fig. 11, in ad- The cell populations projecting to the “bend region”dition to the main ventromedially located lamella of the transition between the dorsal and ventral parafloc-
Fig. 3. Photographs of the screen of a dynamic graphics workstation
demonstrating the location of the somata of 2.864 retrogradely labelled ponto-cerebellar neurons (population projecting to injection site
F-G-490, cf. Fig. 1).The reconstruction comprises a consecutive series of 35 sections through the rostral 28%of the pontine nuclei. The
thin lines represent the outline of the pontine nuclei, medial to the
left. Seven different views (A-G) are shown. The plane of sectioning
(transverse) is indicated as a dashed line in the inset (lower right).
In C , the reconstruction is seen with the plane of sectioning facing the
viewer. In G , the plane of sectioning is seen “on edge” with the rostral
end of the pontine nuclei upwards (reconstruction is rotated 90” about
the x-axis compared to C). Intermediate views are shown in D-F. The
angle of “maximum lamellarity” is demonstrated in A. From this
angle of view, the pons is viewed from the rostral end, a bit from the
dorsal and lateral side, compared to the view in C. The view presented
in B is intermediate between A and C.
Fig. 4. Photographs of the screen of a dynamic graphics workstation
showing the location of the somata of altogether 9.948 retrogradely
labelled neurons in cat CCoL 490. F-G labelled cells (2.864) are shown
as yellow, and RITC labelled cells (7.084) as red. The reconstruction
comprises a consecutive series of 35 sections through the rostral 28%
of the pontine nuclei. Injection sites are shown in Figure 1 (F-G-490
and RITC-4901. The thin lines represent the outline of the pontine
nuclei. Upper picture: The same view as demonstrated for the F-G
population in Figure 3C (plane of sectioning facing the viewer, medial
to the left, and ventral downward). Middle picture: The same view as
for the F-G population in Figure 3E (rotated 45” about the x-axis
compared to the plane of sectioning). Lower picture: The angle of
view of “maximum lamellarity” (see also Fig. 3A).
Fig. 5. Photographs of the screen of a dynamic graphics workstation
showing the location of the soma of altogether 28.471 retrogradely
labelled neurons in cat CCoL 500. RITC labelled cells (14.048) are
shown as blue, F-G labelled cells (10.084) as red, and FB labelled cells
(4.309) as yellow. The rostral 56% of the pontine nuclei was reconstructed (35 sections, every second). The thin lines represent the outline of the pontine nuclei. The two upper pictures show the same angle
of view as demonstrated for another reconstruction in Figure 3E (see
also inset, lower right, in Fig. 31, i.e., a rostro-ventral view of the pontine nuclei (angle of view of 45” to the plane of sectioning). Medial is
to the left, and rostral upward. Upper picture: F-G population (red)
and FB population (yellow). Middle picutre: Same as upper picture
with the addition of the RITC population (blue).Lower picture: same
as middle picture but rotated so as to be seen from the angle of
“maximum lamellarity.”
culus-in addition contain an extra internal lamella of
cells close to the peduncle. When injections are placed
farther along the ventral paraflocculus, the general
impression is that labelled cell populations are even
more externally placed than those projecting to the
dorsal paraflocculus. This pattern of organization is
further tested and elucidated with statistical methods
in the next section.
Statistical Analyses: Segregation, Random Labelling, or
Positive Association ?
The limitations imposed by pure inspection of threedimensional reconstructions make it worthwhile to
search for objective criteria for positive association and
segregation. Positive association and segregation are
essentially deviations from spatial neutrality (“overlap-by-chance” and “segregation-by-chance”). Neutrality, however, is not an unambiguous term. We consider
two fundamentally different definitions of neutrality in
a multivariate spatial point pattern. Either, the component patterns are statistically independent of one another, or they are formed by random labelling (Diggle,
1986). For reasons considered below (see Discussion),
we use random labelling as a benchmark hypothesis,
and search for deviations from this.
We say that random labelling occurs if cells projecting to two different target regions are a random sample
from the totality of labelled cells, ignoring the target
region. We do not know whether the fluorescent tracers label all neurons projecting to the target region or
not, or what fraction of cells that are labelled. We assume, however, that the unlabelled cells are a random
sample of the totality of cells projecting to the target
region in question (i.e., that there is “random thinning,” see Discussion). Accordingly, we require a measure of segregation that is invariant under random
thinning of the spatial distribution pattern. Also, we
require a measure that is symmetric in the sense that
the spatial segregation between cell population 1 vs. 2
should be the same as that between population 2 vs. 1.
To achieve this, we proceed as follows.
Let H,,(t) denote the cumulative distribution function of the distance t between a pair of randomly selected cells belonging to target regions 1 and 2, and
H,,(t) the corresponding function for a pair of cells randomly selected without reference to their target regions. If there is random labelling, H12(t) = H,,(t) for
all t, whereas if the patterns are segregated there will
be a tendency for distances between cells projecting to
different target regions to be larger than distances between cells chosen without reference to the target region, leading to H,,(t) < H,,(t), a t least for small values of t. Conversely, if the patterns are positively
associated in the sense that cells from different target
regions tend to be located together, we expect that
H,,(t) > H,,(t), a t least for small values or t. In the
present context, positive association could arise
through a functional dependency between the patterns,
as, for example, if one pattern is a perturbation of the
other. One should also note that the occurrence of double-labelled cells increases the probability for finding
positive association.
The above considerations suggest that a useful diagnosticfor segregation is a plot of the estimated difference, D(t) = Al2(t) - A,,(t), against t. We calculate the
Fig. 6. System for numbering of folia in the paraflocculus, beginning with 1 in the dorsal paraflocculus, ending with 32 in the ventral
estimates H,,(t) and A,,(t) as the observed proportions
of intercell distances less than or equal to t. Examples
of such plots are given in Figures 7 and 8. For clarity,
we assign numbers to each folium of the paraflocculus,
beginning with 1 in the dorsal paraflocculus, ending
with 32 in the ventral paraflocculus, as shown in Figure 6. We first consider plots of D(t) obtained for combinations where tracers are injected in widely separated folia (6-15 folia between injection sites, cf. 5 of
the data sets in Fig. 7). Over a range of values oft, we
find H,,(t) < H,,(t), suggesting that the two cell populations compared are spatially segregated (Fig. 7).
There is considerable individual variation in the behavior of D(t). For example, compared to other cases
with widely separated injection sites, D(t) is close to
zero for cat 495 (folium 14 vs. 22, Fig. 7). In cases where
adjacent folia are injected (Fig. 81, we find both H,,(t)
< H,,(t) (suggesting segregation, in cat 499) and H,,(t)
> H,,(t) (indicating positive association, in cats 490
and 500). Furthermore, for comparisons of adjacent folia, the overall data show that the deviation from H,,(t)
= H,,(t) (random labelling) is generally smaller than
for cases where tracers are injected in widely separated
The curves presented in Figures 7 and 8, however,
only tell us whether there is a tendency for patterns to
be segregated or positively associated. They do not give
precise quantitative information on the strength of the
possible segregation or positive association. Therefore,
we canI;lot make a direct comparison between the estimates D(t) for different animals. The reason for this is
that D(t) depends on the size and shape of the threedimensional region within which the complete cell distributions are observed (the “observation region”). In
principle, we could eliminate these dependencies on the
observation region by using the reduced second moment measure or K function (cf. Bjaalie and Diggle,
1990), which is essentially the H function with a correction for edge effects. However, the edge corrections
in three dimensions are technically feasible only for
simple shapes of the observation region such as cuboidal regions.
Cat 495 ( f o l i u m 14vs.20 )
Cat 495 ( f o l i u m 14vs.22)
<I -0.010
-0 0301
4 000
- 0 020
2 000
L 000
Cat 500 ( folium 1 vs. 16 )
Cat 495 ( folium 2Ovs.22 )
<I - 0 0 1 0
<I - 0 0 2 0
<I -0.020
- 0 030
-0 030
L 000
Cat 500 ( f o l i u m lvs.15)
Cat 502 ( folium 2vs.12)
- 0 030
Fig. 7. Plots of the estimated difference, H,,(t) - Ho,,(t), against
distance t, for different animals and combinations of injection sites (cf.
Fig. 6 for numbering of folia). HJt) is the cumulative distribution
function of the distance t between a pair of randomly selected points
of type 1 and 2. H,,(t) is the corresponding function for a pair of points
randomly selected without reference to type. The curves suggest segregation between populations of labelled cells projecting to widely
separated folia, since HJt) < H,,(t) over a range of values o f t .
The next logical step in our analysis is to construct a
formal test of significance to see whether deviations
from Hl,(t) = H,,(t) (random labelling or neutrality),
presented in Figures 7 and 8, are significant. Our test
of significance is based on 99 simulations of random
distributions. Each simulation randomly gives a type 1
or type 2 label to each of the totality of cells, irrespective of the tracer-label. We compute the statistic
Cat 490 ( f o l i u m 12vs.13 )
<I -0.010
<I -0.020
Cat 499 ( folium 20 vs. 21 )
<I -0.01ot
<r -0.020
Quantitative Description of a Possible
Segregation Mechanism
Having established the existence of spatial segregation between the locations of cells projecting to separated target regions, our second objective is to give a
quantitative description of the phenomenon. Specifically, we want to develop a measure to test further the
hypothesis that with a gradual change in target region
(from dorsal to ventral paraflocculus, nos. 1-32 in Fig.
6), there is a corresponding drift of the shell-like spaces
containing labelled cells away from the ventromedial
surface of the peduncle and, further, that the extent of
this drift differs at different rostrocaudal levels of the
To investigate this hypothesis, we take sections from
up to eight different rostrocaudal levels of the pons
in each cat and, within each section, convert the cell
locations to perpendicular cell distances from the ventromedial border of the peduncle. In practice, we first
digitize a reference line, corresponding to the ventro-
Cat 500 ( f o l i u m 1 5 v s . 1 6 )
recompute values of u2, us, . . . ,ulo0 after each of the 99
independent random distributions of the totality of
cells, and rank the value of ul, amongst the ul, u2, . . .,
uloo . If u1 ranks kth largest, the p-value of the test is p
= k/100 (Bernard, 1963; Besag and Diggle, 1977).
The results obtained in this way for each of the total
of nine combinations shown in Figures 7 and 8 are p =
0.01 for all combinations. Accordingly, all deviations
from H12(t) = H,,(t) shown in Figures 7 and 8 are
statistically significant.2 We therefore conclude that
there is a segregation process that moves cell populations projecting to widely separated folia apart from
each other. For cell populations projecting to adjacent
folia (Fig. B), we have evidence in favor of both positive
association between cell populations a s well a s segregation. A possible interpretation is that a n inherent
positive association between populations projecting to
adjacent folia becomes attenuated and is eventually
reversed with a progressive increase in spatial segregation between populations projecting to more widely
.separated folia. This leads us to conclude that populations of ponto-cerebellar neurons are confined to shelllike tissue volumes, the spatial location of which gradually move in accordance with specific target region in
the paraflocculus (from folium 1 to 32, cf. Fig. 6).
Fig. 8. Plots of the estimated difference, fi,,(t) - A,,(t), against
distance t, for different animals and combinations of injection sites (cf.
Fig. 6 for numbering of folia). The curves suggest positive association
between populations of labelled cells projecting to adjacent folia, in
two of the data sets, since H,,(t) > H,,(t) over a range of values oft.
Presentation is otherwise as in Figure 7.
'In one remaining case, cat 488, only a small fraction of the pontine
nuclei (the rostra1 7%) was reconstructed. Therefore, compared to
other cases, only relatively few labelled neurons were recorded (511
F-G labelled cells; 1,179RITC labelled cells). Injections were placed in
folium no. 13 (RITC) and 14 (F-G) (cf. Fig. 6). Deviations from neutrality were not statistically significant (P = 0.10).
Cat 4 9 5
Cat 4 9 0
1500 -
- ---
500 . . . -
' ,1
500 -
pons ( l e v e l 1
pons ( level 1
Cat 4 9 9
Cat 500
pons ( l e v e l 1
pons ( l e v e l 1
Cat 502
Cat 511
\ / - - '
pons ( l e v e l 1
pons ( level 1
Fig. 9. Plots of the average perpendicular distance to the ventromedial surface of the peduncle (d,,,),
for cells in a population, against rostro-caudal level within the pontine nuclei. Level 1 is the most
rostral, level 8 the most caudal. Numbers assigned to each curve correspond to the location of injection
site (folium numbers, cf. Fig. 6).
medial border of the peduncle; we then use an algorithm to calculate the perpendicular distance from a
labelled cell to a line defined by the two closest (digitized) points on the reference line. In the computer reconstructed cases, we already have the x,y coordinates
available for the labelled cells at each section level. For
remaining animals, we digitize the x,y position of each
category of labelled cells as a basis for computing the
Plots of the average perpendicular distance to the
ventromedial surface of the peduncle (d,,,) against the
rostrocaudal level within the pontine nuclei clearly
demonstrate how the different populations of labelled
cells in each animal are shifting in location (Big. 9). As
an example, we take cat 511 (Fig. 9, lower right). This
animal received injections in three widely separated
folia (nos. 2,21, and 30 in the terminology presented in
Fig. 6; see also Fig. 1). The average perpendicular distance to the ventromedial surface of the peduncle, at
six different rostrocaudal levels within the pontine nu-
paraflocculus ( 1 - 3 0 )
Fig. 10. Trend curve for d,,, a t pontine level 2 (defined in legend to
Fig. 9) based on data pooled for all animals presented in Figure 9, and
plotted against location of injection site in the paraflocculus (cf. Fig.
6 ) . The trend curve demonstrates that with a change in target region
(from dorsal to ventral paraflocculus, folium nos. 1-32 in Fig. 6 ) ,
there is a corresponding increase in the average distance to the peduncle for cells in a population projecting to a single folium.
clei are shown (FB labelled cells were not found at level
1). At all levels, the mean distance to the peduncle
increases with the progression in target region number. The increase was strongest in rostra1 parts of the
pontine nuclei (level 2: from a mean distance of 613 pm
for cells projecting to folium no. 2, to a mean distance of
959 pm and 1,470 pm for target 21 and 30 cells, respectively). This overall trend was reproduced in several other cases (see, e.g., cats 502,500, and 499 in Fig.
9). Even the mean distances for cells projecting to adjacent targets (such as 15 vs. 16 in cat 500, and 20 vs.
21 in cat 499) showed an increase with increasing target numbers, although this was not always the case (cf.
virtually no shift in location of cells projecting to targets 12 and 13 in cat 490, Fig. 9). Also, in one case (cat
495, Fig. 9) a t some levels in the pontine nuclei, a
partly reverse trend of the one described above was
seen. Nevertheless, our material supports the hypothesis of a systematic shift in location of cells with progression in target region in the paraflocculus.
There is considerable individual variability in the
absolute values of average perpendicular distances to
the ventromedial surface of the peduncle. The conclusions about a shift in location of cell groups are therefore based upon reproducible changes from animal to
animal. Still, if results for dvmpare pooled for all animals presented in Figure 9 and plotted against location
of injection site (folium nos. 1-32) in the paraflocculus,
the overall trend is clear, despite individual variations
(Fig. 10).
The present investigation uses computer threedimensional reconstructions for studying the spatial
organization of large populations of points (such as the
point coordinates corresponding to the center of individual neuronal somata). Moreover, we present quantitative data of spatial relationships between reconstructed multivariate point patterns, such as multiple
populations of neurons within the same reference volume in the brain. Statistical methods are developed in
order t o objectively distinguish patterns created by
spatial neutrality (random labelling) from patterns
created by either a process driving different point types
together in space (positive association) or apart from
each other (segregation). Using such a quantitative approach, we have shown that following injections of two
tracers in adjacent cerebellar folia, populations of labelled cells in the pontine nuclei are significantly positively associated. In contrast, populations projecting to
target regions wider apart are significantly segregated.
We also find that with gradual change in target region
(from dorsal to ventral paraflocculus), there is a corresponding drift of the spatial distribution of pontine cell
groups in a ventromedial direction.
Methodological Considerations
The validity of the patterns observed in the computer
three-dimensional reconstructions depends on several
methodological factors. One is the alignment of the sections. The fiducial marker holes together with blood
vessels passing from section to section provide the basis
for our manual alignment of complete series of sections. Using careful microtomy a t stable temperature,
most sections were free from distortions. In a few sections, twisting or compression nevertheless occurred.
This was adjusted for by manually aligning different
parts of the section a t a time. Thus we feel certain that
the alignment of sections was sufficiently accurate for
the present analysis. The observations of well-defined
bands and clusters of cells in the reconstructions (and
not blurred distributions) also indicate successful
The accuracy of the plotting of the labelled cells is
another important methodological factor. The plotting
was performed with an accuracy with regard to x and y
coordinates of single cells of about * l O pm. The z coordinate was defined by section number and thickness.
We regard these inaccuracies as insignificant in the
present context, since we do not deal with intricate
details a t a small scale, but rather with the overall
distributions and spatial relationships. With regard to
the possibility of errors by double counts of cells, our
system for plotting enabled us to control that all cells
were included, and virtually none had been plotted
twice. It is likely, however, that fragments of cell bodies occurring in two adjacent sections could sometimes
be plotted twice. Thus we did not check for double
counts in the careful way done by Bjaalie and Diggle
(1990). However, the double counts occur with equal
probability for the different types of labelled cells.
The properties of the fluorescent tracers, in terms of
uptake and efficiency of labelling, are more likely to
influence our findings. Fast Blue is generally regarded
as a very sensitive fluorescent tracer. In our hands,
efficiency of labelling (number of labelled cells per injected amount of tracer) varied a t least with a factor of
five. The Fluoro-Gold tracer gave more stable results in
terms of number of labelled cells from experiment to
experiment. On the average, the efficiency of labelling
was higher than that of Fast Blue. The most efficient
tracer, however, was RITC. In terms of variability in
number of labelled cells, it held an intermediate position between Fast Blue and Fluor-Gold. (Further considerations on numerical capacities of the labelled neuronal populations are to dealt with in a study in
preparation by J.G. Bjaalie.) However, regardless of
individual variability in efficiency of labelling, the labelled neurons were distributed throughout comparable regions of the pontine nuclei, and the arrangement
of clusters and bands were apparently not influenced
by the number of labelled cells. This observation
strongly supports the view t h a t unlabelled neurons
projecting to the injection site were randomly distributed, i.e., that the patterns observed are the result of a
“random thinning” (for further considerations on random thinning, see Bjaalie and Diggle, 1990). Our statistical tests are designed so as not to be influenced by
this methodological phenomenon.
A few methodological remarks should also be made
concerning the statistical analyses. In our search for
deviations from spatial neutrality, we have chosen random labelling as a benchmark hypothesis. As mentioned above (see Results, section on Statistical Analyses), however, independence might also be considered
as a reasonable definition of spatial neutrality. We
must therefore consider the difference between these
two kinds of spatial neutrality. This is important because positive association and segregation in relation
to independence could be mechanistically different
from positive association and segregation in relation to
random labelling.
Statistical independence is a situation in which the
component patterns are generated by independent stochastic processes (see also Diggle, 1986). Such a situation is unlikely to occur in our material. The density
distribution of the labelled cells over the reference volume (the pontine nuclei) is inhomogeneous, and because of this there is necessarily a spatial dependence
between different categories of labelled cells, because
all types are more likely to occur within the more dense
regions. Therefore, the rejection of independence is not
very informative in our context.
Random labelling, in contrast, is a kind of neutrality
between type 1 and 2 events quite different from independence. One should note t h a t in the case of random
labelling, the component patterns are generally dependent, even if the differentiation into type 1 and type 2
events is completely random (Diggle, 1986). In a biological context, it seems reasonable to say t h a t a
“label” or sign is attached to each neuron when its
target region(s) are determined at a certain stage during development. Our statistical tests takes into consideration distances between events (the location of individual labelled cells), and therefore, deviations from
random labelling tell us something about possible
mechanisms operating on the single neuron level. It is
not meant to represent a final solution for determining
the complicated spatial interactions between different
categories of neurons. Nevertheless, the approach we
have used provides objective results understandable in
relation to subjective evaluation of the data.
Does Precise Terminology on Spatial interactions improve
Our Understanding of the Connectivity of the Brain?
This study deals with spatial relationships between
neuronal populations in a new way. We generally assume that two neurons t h a t are located close together
are more likely to be influenced by the same afferent
fibers than neurons located far apart. Does it matter,
however, whether they are located close together sim-
ply by chance or due to a mechanism driving them
together in space (positive association)? Perhaps it does
not matter for the operations of the individual neurons,
but it may nevertheless be of importance for the operations of large neuronal populations. The present approach therefore may serve to focus attention on the
connectivity of populations of neurons, rather than of
individual neurons. Thus a positive association between two neuronal populations increases the probability of similar information transfer, compared to a
situation where there is only “overlap-by-chance.’’
Along the same lines of reasoning, i t is more likely that
the different kinds of information can be transferred
through neuronal populations t h a t are segregated due
to a specific mechanism, compared to populations that
are only incidently separated. Furthermore, as briefly
considered below, the mathematical approach has
made it possible for us to suggest a new model for organization of neuronal populations in the pontine nuclei.
We therefore believe that the terminology used in
the present work provides a n advantage over the conventional terminology. We replace subjective evaluations with formal tests, and we introduce methods for
quantitative estimations of the basis for intricate wiring patterns.
The Biological Significance of a Gradual Shift in Location
of Neuronal Cell Populations
Our data suggest t h a t there are gradual shifts in
location of pontine neuronal populations projecting to
gradually changing locations of cerebellar target regions. It seems reasonable to assume that this gradual
shift in location also leads to a gradual change in the
information flow transferred by the populations. Consequently, it appears that adjacent cerebellar spots of
about the size of a folium (or smaller) receive only
slightly different inputs. This would seem to be cornpatible with a model of a continuous map of mossy fiber
input to the cerebellum, rather than a patchy of scattered map (for a review of parallel computer maps and
brain maps, see Nelson and Bower, 1990) a s suggested
by previous anatomical and physiological data (for references, see Brodal, 1982,1987; Welker 1987,1990).As
pointed out above, the statistical approach is essential
as a basis for our conclusions. Pure inspection of spatial
interactions between cell clusters indeed suggests a
mosaic segregated cell distribution, as a product of a
patchy mapping process. One could therefore speculate
that our hypothesis of continuous shifts of spatial distributions of cells would actually apply also to other
systems where segregated cell groups are described
(see, e.g., Goldman-Rakic and Selemon, 1990). This
could be tested by applying our methodological approach. In developmental terms, one could think of gradients of neuroactive substances guiding the growth or
signing of different neurons to distinct target regions,
thus creating continuous maps. Gradients would be
more parsimonious to code genetically than a tremendous number of specific abrupt changes. Our methodological approach may therefore also be a useful tool in
developmental studies, by enabling the investigator to
detect and describe changes not already revealed by
qualitative methods only.
Of specific interest to us is the relation between our
hypothesis for mapping of cerebellar mossy fiber input
and that of previous electrophysiological micromapping studies (for reviews, see Welker 1987, 1990).
Welker and collaborators have found that, within a cerebellar folium, there is a “fractured somatopic map”
(patchy map) in parts of the cerebellum investigated in
both cat and rat. (The paraflocculus has not been studied in this way, however.) It is not obvious how these
data can be reconciled with our hypothesis, unless we
assume t h a t a certain set of functionally defined
patches are repeated from folium to folium. In t h a t
case, in principle, the small shift in location of pontocerebellar neurons projecting to adjacent folia could
correspond to small changes in the exact organization
of repeated combinations of patches. Some of our observations, however, are difficult to reconcile even with
this assumption. Thus regardless of the size of cerebellar spots injected with a tracer, we always find a widespread distribution of ponto-cerebellar labelled cells.
Admittedly, our injection sites are not anywhere near
the size of the single patches of Welker. Previous investigators, however, have made extremely small injections of horseradish peroxidase-covering only a
tiny fraction of a folium-in the cerebellar paramedian
lobule (Hoddevik and Walberg, 1979). They find that
the few labelled cells, nevertheless, are widely distributed. We are therefore inclined to believe that pontocerebellar cell bodies projecting to a minor part of a
folium (a few functionally defined patches) are distributed within the same lamellae as the neurons projecting to a larger target region (an entire folium). It remains to be determined, however, whether injections of
minute amounts of multiple tracers into different parts
of a single folium would result in a positive association
between the various groups of labelled cells or, if in
fact, at such a smaller scale, there would be a n element
of segregation corresponding to differential input to
functionally distinct patches. Thus our findings raise
further questions about the principles of organization
of ponto-cerebellar pathways.
We thank Kari Ruud and Gunnar F. Lothe for expert
technical assistance. This project was supported by the
Wennergren Foundation and the Norwegian Research
Council for Science and the Humanities.
Barnard, G.A. 1963 Contribution to the discussion of Professor Bartlett’s paper. J.R. Statist. Soc., B25t294.
Besag, J., and P.J. Diggle 1977 Simple Monte Carlo tests for spatial
pattern. Appl. Statist., 26t327-333.
Bjaalie, J.G. 1991 Three-dimensional computer reconstructions in
neuroanatomy. Basic principles and methods for quantitative
analysis. In: Quantitative Methods in Neuroanatomy. John
Wiley & Sons, Inc., Chichester (in press).
Bjaalie, J.G., and P. Brodal 1989 Visual pathways to the cerebellum:
segregation in the pontine nuclei of terminal fields from different
visual cortical areas in the cat. Neuroscience, 29t95-107.
Bjaalie, J.G., and P.J. Diggle 1990 Statistical analysis of cortiopontine
neuron distribution in visual areas 17, 18, and 19, of the cat. J.
Comp. Neurol., 295t15-32.
Bjaalie, J.G., P.J. Diggle, T.W. Blackstad, A. Nikundiwe, and P.
Brodal 1990 Three-dimensional neural architectonics. J. Anat.,
I 7Ot2 19-22 1.
Blackstad, T.W., L. Neimer, and E. Mugnaini 1981 Experimental neuroanatomy: General approaches and laboratory procedures. In:
Neuroanatomical Tract-Tracing Methods. L. Heimer and M.J.
Robards, eds. Plenum Publishing, New York, pp. 1-53.
Brodal, P. 1982 The cerebropontocerebellar pathway: Salient features
of its organization. Exp. Brain Res. Suppl., 6t108-133.
Brodal, P. 1987 Organization of cerebropontocerebellar connections as
studied with anterograde and retrograde transport of HRP-WGA
in the cat. In New Concepts in Cerebellar Neurobiology. J.S.
King, ed. Alan R. Liss, New York, pp. 151-182.
Brodal, P., and J.G. Bjaalie 1987 Quantitative studies of pontine projections from visual cortical areas in the cat. In: Cerebellum and
Neuronal Plasticity. M. Glickstein, C. Yeo and J. Stein, eds. Plenum Publishing, New York, pp. 41-62.
Diggle, P.J. 1986 Displaced amacrine cells in the retina of a rabbit:
Analysis of a bivariate spatial point pattern. J . Neurosci. Methods, I8t115-125.
Goldman-Rakic, P.S., and L.D. Selemon 1990 New frontiers in basal
ganglia research. TINS, 13t241-244.
Hoddevik, G.H., and F. Walberg 1979 The pontine projection onto
longitudinal zones of the paramedian lobule in the cat. Exp.
Brain Res., 34t233-240.
Larsell, 0. 1970 The Comparative Anatomy and Histology of the Cerebellum From Monotremes Through Apes. University of Minnesota Press, Minneapolis.
Nelson, M.E., and J.M. Bower 1990 Brain maps and parallel computers. TINS, I3:403-408.
Switzer, R.C. 1979 Wrinkle-free mounting of sections from gelatinembedded brains. Neurosci. Lett., I2t103-105.
Welker, W. 1987 Spatial organization of somatosensory projections to
granule cell cerebellar cortex: functional and connectional implications of fractured somatotopy (summary of Wisconsin studies).
In: New Concepts in Cerebellar Neurobiology. J.S. King, ed. Alan
R. Liss, New York, pp. 239-280.
Welker, W.I. 1990 The significance of foliation and fissuration of cerebellar cortex. The cerebellar folium as a fundamental unit of
sensorimotor integration. Arch. Ital. Biol., 128t87-109.
Без категории
Размер файла
1 898 Кб
population, interactions, multivariate, segregation, cerebellar, analysis, spatial, neuronsstatistical, ponti
Пожаловаться на содержимое документа