THE JOURNAL OF COMPARATIVE NEUROLOGY 334:19-46 (1993)
Cortical Hierarchy Reflected in the
Organization of Intrinsic Connections in
Macaque Monkey Visual Cortex
Y. AMIR, M. HAREL AND R. MALACH
Weizmann Institute of Science, Rehovot, Israel 76100
ABSTRACT
Neuronal response properties vary markedly at increasing levels of the cortical hierarchy.
At present it is unclear how these variations are reflected in the organization of the intrinsic
cortical circuitry. Here we analyze patterns of intrinsic horizontal connections at different
hierarchical levels in the visual cortex of the macaque monkey. The connections were studied in
tangential sections of flattened cortices, which were injected with the anterograde tracer
biocytin. We directly compared the organization of connections in four cortical areas representing four different levels in the cortical hierarchy. The areas were visual areas 1, 2, 4 and
Brodman's area 7a (Vl, V2, V4 and 7a, respectively). In all areas studied, injections labeled
numerous horizontally coursing axons that formed dense halos around the injection sites.
Further away, the fibers tended to form separate clusters. Many fibers could be traced along the
way from the injection sites to the target clusters.
At progressively higher order areas, there was a striking increase in the spread of intrinsic
connections: from a measured distance of 2.1 mm in area V1 to 9.0 mm in area 7a. Average
interpatch distance also increased from 0.61 mm in area V1 to 1.56 mm in area 7a. In contrast,
patch size changed far less at higher order areas, from an average width of 230 pm in areaV1 to
310 pm in area 7a. Analysis of synaptic bouton distribution along axons revealed that average
interbouton distance remained constant at 6.4 p,m (median)in and out of the clusters and in the
different cortical areas. Larger injections resulted in a marked increase in the number of labeled
patches but only a minor increase in the spread of connections or in patch size.
Thus, in line with the more global computational roles proposed for the higher order visual
areas, the spread of intrinsic connections is increased with the hierarchy level. On the other
hand, the clustered organization of the connections is preserved at higher order areas. These
clusters may reflect the existence of cortical modules having blob-like dimensions throughout
macaque monkey visual cortex. D 1993 Wiley-Liss, Inc.
Key words: visual system, cerebral cortex, cortical areas, biocytin, tract-tracing
Visual information undergoes major transformations as
it progresses from lower to higher order cortical areas. In
primary visual cortex, area V1, the most notable transformation is the generation of orientation selectivity (Hubel and
Wiesel, '62, '68) but many other transformations were
discovered over the years (for review see Van Essen, '85).At
higher order visual areas, the transformations become even
more involved, ranging from sensitivity to illusory contours
in area V2 (van der Heydt and Peterhans, '89) to "face"
cells in inferotemporal cortex (IT) (e.g., Gross, '72; Perret et
al., '82) and cells with integrative visuomotor function in
area 7a (Andersen et al., '85).
In all these transformations, the interactions between
neighboring points in the visual field play a crucial role. In
o 1993 WILEY-LISS. INC.
area V1, these neighborhood interactions are quite local, as
reflected by the small receptive fields of neurons in this
area. At higher order areas, such interactions extend over
longer distances, producing receptive fields of increasing
size and culminating in neurons that respond to most of the
visual field in areas 7a and the inferotemporal cortex (e.g.,
Andersen et al., '90; Boussaoud et al., '91).
In searching for the anatomical circuitry underlying
these neighborhood interactions, pathways that connect
neighboring cortical visuotopic sites are natural candidates.
Accepted March 8, 1993.
Address reprint requests to Dr. Rafael Malach, Department of Neurobiology, Weizmann Inst. of Science, Rehovot, Israel 76100.
Y. AMIR ET AL.
20
TABLE 1 List of Tracer Injedionsi
Corticalarea
Injection
Monkey
Hemisphere
Figure
Size(pm)
v2
V1
1'
2
MB1
L
B
190
MB1
L
B
130
3
MB1
L
B
220
MB2
L
5
MB5
L
300
D
230
4*
6
MB6
I*
L
MB1
L
720
B
230
v4
2
MB1
L
B
190
3
MB7
4
I*
IdB7
MI31
la
2
ME31
3*
MB4
4
MB7
1
MBI
R
R
R
L
R
E
200
E
A
E
A
180
B
220
C
190
900
210
190
R
R
2
MB5
L
D
230
3*
MB7
IT
I*
MBI
L
L
F
250
F
210
'The first row speufies the cortical area iwected (abbreviations as in Fig 1) The second row provides an identification number for each iniection Asterisk indxates injections whose
pattern of labeling ISshown in one of the figures The third and fourth rows show themonkey and hemsphere inpcted, respectively The fifth row inhcates in which part of Figure 1the
location of the injection is shown The last row shows the mameter of tracer uptake zone in pm
There are several cortical pathways that could perform
such horizontal integration, among them, diverging thalamic afferents (Hubel and Wiesel, '62; Chapman et al., '91)
and extrinsic corticocortical connections (e.g., Zeki and
Shipp, '89). However, the system of connections that most
directly relates neighboring cortical sites is the set of
intrinsic horizontal connections. These connections run
mostly parallel to the cortical surface, are confined to one
cortical area, and in primate V1 are most prominent in
layers 2-3 and 4b (Rockland and Lund, '83). The horizontal
connections originate mainly from pyramidal cells, although some short range contribution from inhibitory
interneurons is likely as well (Kisvarday et al., '86). Their
target includes mainly other pyramidal cells, with about
20% of the output on inhibitory interneurons (McGuire et
al., '91).
In primary visual cortex, the horizontal intrinsic connections have been intensively studied both anatomically
(Rockland and Lund, '82, '83; Gilbert and Wiesel, '83;
Livingstone and Hubel, '84b) and physiologically, by crosscorrelation and inactivation techniques (Ts'o et al., '86;
Ts'o and Gilbert, '88; Bolz and Gilbert, '86). These studies
have indicated that the intrinsic connections emanating
from a localized cortical site are not distributed uniformly
but tend to form clusters. The clusters were found to be
closely related to the underlying cortical columnar architecture (Malach et al., '92). Thus, in cat visual cortex, orientation columns of similar preference were shown to be
connected to each other (Gilbert and Wiesel, '89, but see
Matsubara et al., '87) while in primate V1, intrinsic fibers
connect cytochrome oxidase blobs to neighboring blobs and
interblobs to interblobs (Livingstoneand Hubel, '84b; Ts'o
and Gilbert, '88).
Inactivation and cross-correlation studies have suggested
that the intrinsic connections can contribute directly to the
construction of cortical receptive fields (Bolz and Gilbert,
'86). Thus, the organization of intrinsic connections in area
V1 appears to be tightly related to its columnar architecture
and to the physiological properties of its neurons.
Given this tight link, we were interested in exploringhow
the architecture of intrinsic connections is organized in
different cortical areas belonging to different levels of the
cortical hierarchy (for review, see Felleman and Van Essen,
'91). Two questions are of particular interest here: first,
how does the spread of intrinsic connections relate to the
reported increase in receptive field size at high order areas?
Second, do the connections remain patchy and are the
shape and dimensions of the patches conserved across
areas?
The present study is aimed at addressing these questions
by quantitative analysis of intrinsic connections in several
visual areas representing four levels in the cortical hierarchy. The results show that intrinsic connections spread to a
greater extent at higher order cortical areas. In contrast,
patch dimension changes relatively little, suggesting that
similar columnar building blocks may underlie cortical
architecture at different levels of the cortical hierarchy.
MATERIALS AND METHODS
Intrinsic connections were revealed by iontophoretic
injections of the tracer biocytin (King et al., '89) into
superficial cortical layers. Tracer injections were performed
in eight hemispheres in 1adult Macaca mulatta and 5 adult
Macaca fascicularis. Table 1 lists the areas injected and
numbers of injection sites. The locations of most of these
injections are shown in Figure 1 as seen from a lateral view
of the injected hemispheres. To obtain control over the size
and depth of the injection sites, injections were restricted to
the exposed crowns of cortical gyri, thus preventing direct
spread of the tracer along the injection track. Note that the
high order areas (V4, 7a, and IT) never received more than
a single injection per area.
All animals except MB1 were injected following a 1-2 day
experiment involving optical imaging of intrinsic signals in
area V1 whose results will be reported elsewhere. Animals
were presurgically injected intramuscularly (I.M.) with
0.01 mg/kg atropine sulfate, and anesthetized with 20-30
mg/kg ketamine (1.M.) followed by either intravenous (I.V.)
injection of sodium pentobarbital (as needed in steps of 10
mg/kg) or with sodium penthotal(20 mg/kg, supplemented
as needed). Paralysis was induced with vercuronium bromide (Norcuron) and the animals were artificially respirated. The electroencephalogram, electrocardiogram, endtidal CO2 and temperature were continuously monitored.
Mannitol(20%, 10 ml/kg) was injected i.v. Alarge, bilateral
craniotomy was performed, overlying the occipital, parietal
and part of the temporal lobes, and the dura was reflected.
Biocytin (2% in Tris buffer 0.05 M pH 7.6) was injected
through glass pipettes (10-12 pm tip diameter) that were
lowered to a depth of 0.2-0.6 mm into the cortex at the
target locations in an attempt to label only supragranular
layer cells. In areas V4 and 7a, the sites were chosen to be at
the approximate center of the injected areas according to
sulcal landmarks. Care was taken to maintain the injection
pipettes at a direction exactly normal to the cortical surface.
The tracer was injected by iontophoresis (4 )LA,7 seconds
on- 7 seconds off, DC current for 3-10 minutes, pipette
positive).
After the injections were completed, the dura was pulled
over the cortical surface, supplemented with artificial dura
(Dow Corning) and thin layers of Gelfoam, and the bone
was replaced. In monkey MB1, the bone was anchored with
screws and dental cement to the rest of the skull, and the
skin pulled and sutured back. The animal was allowed to
recover for 24 hours. All other animals were kept anesthetized for 10-12 hours until perfusion.
21
INTRINSIC CONNECTIONS AND CORTICAL HIERARCHY
A
3
7
D
P
1 cm
H
Fig. 1. A-F: Locations of injection sites in this study. Schematic
drawings, based on photographs of the injected hemispheres, showing
lateral views of six cortical hemispheres in which high order areas were
successfully injected. Arrows point to the sites of injections. Centers of
injections are indicated by stars. Major sulci are indicated hy solid lines.
CeS, central sulcus; LS, h a t e sulcus; IOS, inferior occipital sulcus;
IPS, intraparietal sulcus; SF,sylvian fissure; STS, superior temporal
sulcus. Cortical areas injected are V1, visual area 1: V2, visual area 2;
V4, visual area 4;7a, Brodmann’s area 7a; IT, inferotemporal cortex.
22
Y. AMIR ET AL.
Fig. 2. Intrinsic horizontal connections in macaque monkey area
V1. Darkfield, low power image of a tangential section from a flattened
area V1. Intrinsic fibers were labeled by a small (190 pm diameter,
injection no. 1 in Table 1)injection of the anterograde tracer biocytin
(curved arrow points to the injection site). Note that fibers produce a
dense halo around the injection site and tend to form clusters further
away (white arrows). Note also many fibers coursing between patches
or running individually. The areas in rectangles a and b are shown
enlarged in Figure 3. Scale bar = 200 km.
Fig. 3. A,B: Examples of clustered fibers in area V1. Enlarged images of the areas demarcated by
rectangles a and b in Figure 2. Note the clear tendency of fibers to form patches (open arrows). Scale bars =
100 pm.
23
INTRINSIC CONNECTIONS AND CORTICAL HIERARCHY
0
Fig. 4. Schematic drawing of the main clusters of intrinsic fibers
produced by biocytin injection in area V1. This drawing is from the
same injection shown in Figure 2. Curved arrow marks the injection
site. Stippled patches indicate clusters of fibers chosen for quantitative
analysis. Darker stippling around the injection site demarcates the halo
of fibers. Gray patches (demarcated by thick lines) are cytochrome
oxidase (co) dense blobs drawn from a neighboring section. For clarity,
only the blobs situated near the injection site are drawn in. Straight
arrows point to the same patches indicated by arrows in Figure 2. Note
that the injection site resides outside a CO blob, and that the patches of
axons also appear to avoid the CO blobs. A, anterior; M, medial.
For perfusion fixation, animals were deeply anesthetized
with Nembutal, received 20,000 Units of heparin i.v., and
were perfused transcardially with phosphate-buffered saline (PBS) followed by light fixation with 2.5%glutaraldehyde and 0.5% paraformaldehyde. In one case (MB7), the
fixative included 4% paraformaldehyde in PBS containing
5% sucrose. The brains were removed, photographed, and
the gyri containing the injection sites dissected along the
fundi of neighboring sulci. The gyri were flattened according to the method of Tootell and Silverman (’85) and
postfixed in the same fixative solution containing 15%
sucrose for 24 hours. After postfixation, the flattened
cortices were frozen in the flattened position.
The frozen blocks were cut tangentially on a freezing
microtome at 70-100 pm thickness. In the blocks containing V1 and V2 injections, alternate sections were processed
for cytochrome oxidase histochemistry (Silverman and
Tootell, ’87). In two cases (MB5, MB7), every third section
was stained for myelin (Gallyas, ’79).
Biocytin was revealed with a modified Avidin Biotin
horseradish peroxidase (HRP) Complex (ABC,Vector Labs)
protocol including the following main steps. Sections were
rinsed thoroughly and incubated overnight in 0.5% Triton-X. The following day they were rinsed and incubated for
3 hours in Vector Lab’s Elite ABC Kit (PK 6100). Staining
was done with diaminobenzidene (DAB)as the chromogen
for 25 minutes. Sections were mounted on gelatinized slides
and coverslipped with Permount.
Stained fibers were inspected under darkfield and brightfield illumination at low power using Wild M-5 microscope.
High power viewing was done with a Universal Zeiss
microscope. For quantitative analysis, a special computer
mapping system was used, which allowed registration and
quantification of labeled axons and cell bodies (Paperna and
Malach, ’91). Finally, detailed 3-dimensional structure of
axonal networks and terminal boutons was revealed by
scanning sections with the confocal laser scanning microscope, using reflected light illumination (Malach, ’92a).
Methodological problems
A problem inherent in any tracer is the possibility of
“false negatives” i.e., the underestimation of the spread of
intrinsic connections that may be caused by a failure in
transport of the tracer biocytin over longer distances. While
it is difficult to rule out such a possibility, it is of less
importance in the context of the current project, which is
aimed at comparing labeled patterns across different areas
using identical injection parameters. To produce an erroneous relationship between intrinsic connections and cortical
hierarchy, this failure of transport needs to be selective for
lower order areas. Such selective failure was never reported, and thus seems unlikely.
RESULTS
In the first part of the Results, we will show examples of
biocytin-labeled intrinsic connections in each of the areas
studied. These are meant to show the detailed distribution
of labeled fibers, and to illustrate the criteria by which
clusters of connections were chosen for quantitative analysis. Generally, axons were determined to constitute a
24
Y. AMIR ET AL.
Fig. 5. A radially symmetric form (“rosette”) of intrinsic connections in area V1. Same symbols as in Figure 2 (injection no. 4 in Table
1).Curved arrow marks injection site. Open arrows show where fibers
form clusters. Note the striking difference in overall shape between this
round pattern and the highly elongated one shown in Figure 2. The
rosette shapes appeared less frequently than the oval ones. The source
for this diversity of shapes is unclear at present. Scale bar = 200 pm.
cluster if it was possible to observe a marked, unequivocal
increase in their density at particular sites. This was most
often associated with occurrence of fibers curving from
different directions and targeting the cluster region, It
should be emphasized that the main features seen in these
examples were consistently seen in the several cases injected in each area.
s o n s coursing in between the clusters. Several s o n s
extended further, and formed additional clusters that were
of much lower axonal density (Fig. 2).
Generally, we observed in this set of injections and in
many additional ones in V1, which were part of another
study (Malach, ’92a,b),that clusters tend to overlap similar
CO tissue compartments. Thus, in the present example,
both the injection site and the clusters avoid the CO dense
blobs (Fig. 4). Interestingly, cluster dimension appeared
qualitatively similar in blob-injected or interblob-injected
Area V1 (area 17)
Figures 2-4 show the labeling of intrinsic horizontal
connections in area V1. The labeling pattern was produced
by biocytin injection in area V1, which was localized in an
interblob territory. Figure 2 shows a low power montage of
the entire fiber distribution. Figure 3 shows higher power
view of selected clusters from this pattern. Figure 4 is a
schematic drawing of the labeled pattern demarcating the
clusters of fibers chosen for analysis and their relation to
the cytochrome oxidase (CO) dense blobs. The injection was
190 km in diameter, confined t o supragranular layers, and
straddled the edge of a blob (Fig. 4). The labeled fibers form
intense halo around the injection site (Fig. 2, curved arrow).
Further away, they clearly tend to cluster (Figs. 2, 3,
straight arrows), although many examples can be found of
Fig. 6. Intrinsic horizontal connections in area V2. Darkfield,
low-power image of a tangential section from flattened area V2.
Intrinsic fibers were labeled by a small (230 pm diameter) injection of
biocytin (curved arrow). Note the elaborate network of fibers produced
by this injection. In this and subsequent low-power photographs, the
reader is advised to use a magnifying glass to appreciate the full
richness of the fiber projections. Similar to area V1, the fibers form
clusters (open arrows). Unlike area V1, these clusters extend over
longer distances and tend to be somewhat looser. There was an
occasional tendency for the patches to be aligned in stripe-like formations (rectangle b). The areas in rectangles a and b are shown enlarged
in Figure 7. Scale bar = 200 pm.
Figure 6
Y. AMIR ET AL.
26
Fig. 7. A,B: Examples of clusters in areaV2. Enlarged images of areas demarcated by rectangles a and b
in Figure 6. Note the clear clustering of fibers (arrows) including the stripe-like formation in B. Some
individual s o n s can be seen connecting different patches, indicating that a single neuron may innervate
more than one patch. Scale bars = 100 km.
cases; however, we did not perform a quantitative analysis
to verify this point. This suggests that the interblob
territory may itself contain subdivisions similar to the blobs.
Figure 4 illustrates the criteria by which patches of fibers
were chosen for quantitative analysis. Note that only
clear-cut patches were considered. Isolated axons, and the
more peripheral patches that contained too few axons to be
unequivocally defined, were excluded. Thus, our quantitative analysis focused only on the major components of the
intrinsic connectivity. An important point to emphasize is
that this analysis will lead to a lower estimate of the spread
of connectivity as compared to studies that ignore the
density of labeled patches in their measurements.
Analysis of many sections throughout the cortical thickness indicated that the clusters of fibers are aligned and
extend vertically through the supragranular layers. The
intrinsic connections labeled by our superficial layers injections are mostly confined to supragranular layers, although
isolated fibers could occasionally be observed in infragranular layers as well, but these were too scant to allow
comprehensive quantitative analysis. Consequently, our
study concentrated on analyzing the connections of the
more robustly labeled superficial layers.
In most cases, the overall distribution of labeled fibers is
elongated, extending further along one (tangential t o the
cortical surface) axis. The injection shown in Figure 2
provides an example of such ovoid distribution. However,
occasionally the pattern of labeling was markedly different,
having a radially symmetric arrangement around the injection site. An example of such “rosette-like” organization is
shown in Figure 5. Note that here too, the fibers are clearly
clustered (arrows), but in this case the clusters are radially
distributed, located at roughly equal distances from the
injection site. Inspection of many such injections suggested
that, in fact, there was a continuum of shapes ranging from
extreme ovals to round rosettes.
Area V2 (area 18)
This area is one hierarchical level above area V1. All
injections in area V2 were confined to the vicinity of the
17/18 border, along the anterior edge of the operculum.
The border was accurately defined by a sharp transition in
the GO staining. An example of injection in area V2 is
shown in Figures 6-8. The intrinsic connections are clearly
organized in clusters (Figs. 6, 7, white arrows), which are
quite similar to those found in area V1. This result is
INTRINSIC CONNECTIONS AND CORTICAL HIERARCHY
27
rM
J.
0
\
A
\
Fig. 8. Schematic drawing of the main clusters of intrinsic fibers produced by biocytin injection in area
V2. This drawing is from the same injection shown in Figure 6. Fine lines indicate examples of individual
fibers. Dashed line shows the border between area V1 and V2. Other symbols and conventions same as in
Figure 4. Here too, the larger spread of intrinsic projections, relative to area V1, is evident (note scale in
both figures).
interesting, since the stripe-like CO compartmental organization would suggest an altogether different columnar
structure (see below).
In comparison to area V1 intrinsic connections, the
patches in area V2 could be found at a greater distance from
the injection site (e.g., compare Figs. 6 and 2) and the fibers
appeared to be somewhat more loosely distributed. Follow,ing the connections at different cortical depths revealed
that, as in area V1, the patches form vertical columns,
which could be found mainly in supragranular layers.
Relationship to the CO stripes. The intrinsically connected patches sometimes tended to be organized in stripelike aggregates whose shape was similar to the GO stripes
(Figs. 6 , 8). Although the GO staining of V2 stripes in our
material was not sufficiently sharp for a quantitative study,
we observed that often the injection site and some of the
intrinsic connections were localized within a single GO
stripe, indicating that intrinsic connections can link subcompartments within individual GO stripes.
Area V4
Area V4 is at a higher level in the cortical hierarchy than
area V2 (Felleman and Van Essen, '91). An example of an
injection in this area is shown in Figures 9-11. Several
aspects should be noted about the organization of intrinsic
connections in area V4 as compared to areas V1 and V2.
Figure 9
INTRINSIC CONNECTIONS AND CORTICAL HIERARCHY
29
Fig. 10. A,B: Examples of fiber clusters in areaV4. Enlarged images of areas demarcated by rectangles a
and b in Figure 9. Note the clear tendencies of axons to ramify at particularly preferred sites (arrows). The
gap in the patch in A is due to imperfect flattening of the cortex around the STS. Scale bars = 100 pm.
Fig. 9. Intrinsic horizontal connections in area V4. Darkfield, low
power image of a tangential section from flattened area V4. The
injection of biocytin (no. 1in Table 1) is indicated by curved arrow. Note
the widely distributed nature of labeled fibers and the prominent halo
around the injection site. Nevertheless, close examination of the
photograph reveals several sites in which axons are clearly aggregated
into patches (open arrows). The areas in rectangles a and b are shown
enlarged in Figure 10. Scale bar = 200 pm.
First, the intrinsic connections form a looser network with
a clear tendency to spread over longer distances (Fig. 9).
The halo around the injection site is also increased in size.
Further away from the halo, clusters of intrinsic connections could be discerned (Figs. 9, 10, arrows). Thus, despite
the increased spread of the connections in area V4, their
patchy organization is maintained.
Similar to lower order areas, the clusters in V4 are
organized in vertical columns confined mostly to the super-
30
Y. AMIR ET AL.
M
t
A
\7'
I
\"-
Fig. 11. Schematic drawing of the main clusters of intrinsic fibers produced by a biocytin injection in
area V4. Same symbols and conventions as in Figures 4 and 8. Note the additional increase in spread of
intrinsic connections in this area. The asymmetric shape of the overall fiber distribution can also be
discerned here.
INTRINSIC CONNECTIONS AND CORTICAL HIERARCHY
ficial layers. The overall distribution of the connections is
often asymmetric with respect to the injection site.
Areas 7a and IT
Finally, the highest levels of cortical hierarchy in our
study were represented by areas 7a in the inferior parietal
lobule, and the inferotemporal cortex (IT).Area 7a is part of
the occipitoparietal stream, while IT cortex is part of the
occipitotemporal stream (Ungerleider and Mishkin, ’82).
Figures 12-14 show the fiber labeling caused by an injection in area 7a. The trend for increased size halo around the
injection site, and a more diffuse, looser pattern of connections, is accentuated further in area 7a (Fig. 12). One can
see fibers emanating in all directions, travelling over long
distances without terminating in any clear cluster of connections. Nevertheless, despite the diffuse nature of the connections, clear instances of fiber clustering can be observed.
These clusters are quite similar in size and shape to those
found in lower order areas, but fiber ramification within
these patches is less dense (Figs. 12,13, mows). Compared
to lower order areas, patches can be found at a longer
distances from the center of injection (Figs. 12, 14).
Here too, the clusters were found to be organized in
vertical columns concentrated mostly in the supragranular
layers. Another point of similarity of area 7a to all previous
areas is in the overall distribution of the connections, which
tends to be a asymmetrically elongated, ovoid pattern (Fig.
14).
IT cortex
Only one injection was placed unambiguously in IT
cortex. By its relationship to the sulcal pattern, it was
determined to be within posterior inferotemporal cortex
(Felleman and Van Essen, ’911, corresponding to area TEO
(Boussaoud et al., ’91). The injection site in this case was
small and the fiber labeling was rather weak. Consequently
we show the distribution of the fibers in a camera lucida,
computer-assisted drawing (Fig. 15).The overall trend seen
in area 7a could be observed here as well, i.e., a tendency for
looser, more diffuse connections. Examination of the stained
fibers in several sections revealed some clustering at a
distance from the injection site (see Fig. 15).
Comparison of injection site halos
in different areas
Together with the spread of intrinsic connections, the
halos of fibers surrounding the injection sites are larger at
higher order areas. This is illustrated in Figure 16, which
shows the halos produced by injections of similar size in the
different areas studied. It is clear that there is a progressive
increase in halo size as we ascend in the cortical hierarchy.
The halos become more radially uniform, culminating in
“sunburst” patterns in area 7a (Fig. 16D).
As will be shown below, the halos around the injection
sites carry numerous boutons, which make them apowerful
carrier for information transfer between neighboring neurons.
Quantitative analysis of connections
in the different areas
TOobtain a more rigorous comparison of the patterns of
connections in the different hierarchical levels, several
aspects of these patterns were quantified. The analysis
included only clear-cut, unambiguously defined patches of
connections. Furthermore, only clusters that were clearly
31
recognized in three consecutive sections spanning a thickness of no less than 200 pm were included. The parameters
assigned to each patch were the average of the values
measured in the three sections in which the patch was most
prominent. We performed this analysis on 182 patches in
four areas (Table 1 lists the injection sites whose connectional patches were analyzed). We first analyzed patch
dimensions. Are the patches different in size at different
levels of cortical hierarchy? Since the patches were most
often oval shapes, we defined their long axis as length, and
short axis as width. Area was measured by a computer program based on the drawn patch borders. Figure 17 compares the patch area and width among the different cortical
areas studied. Although a consistent increase can be observed in both patch area and width as one ascends from
area V1 to 7a, this increase is relatively small. Thus, patch
length (not shown in Figure 17) increases by 30%, patch
width by 34%, and patch area by 92% going from area V1 to
area 7a.
Another interesting parameter we analyzed was the
interpatch distance. This is the average distance between
each patch and its nearest neighbors. Such a measure may
be related to the spatial dimension of periodicities in
columnar organization, analogous to the “hypercolumns”
of area V1. Relative to patch width, interpatch distance
increases more markedly with cortical hierarchy. Figure
18A,B shows the results of this analysis. Note that area 7a
interpatch distance is larger than V1 by a factor of 2.5.
The most dramatic variation between areas was reflected
in the spread of intrinsic connections from the injection
site. We used two parameters to quantify this spread: first,
the average distance, which is the mean center to center
distance from the injection site to all the patches within an
area. Second, the maximum range, which is the sum of
distances from the injection site to the two most faraway
patches in each area. Figure 18D displays both the average
distance and maximum range. Note the striking increase in
these two parameters with hierarchy level. The maximum
range of the connections increases fourfold from area V1 to
area 7a, while the average range shows a 3.4-fold increase.
A parameter that may be more functionally relevant in
this analysis is not the absolute distance of connections
within each visual area, but over what fraction of the visual
area the intrinsic connections spread following a single
injection. To obtain this measure, it is necessary to normalize the absolute distance of the connections to the size of
each cortical area. Unfortunately, the boundaries of the
higher order areas are not fully defined so only very gross
estimates of areal size can be obtained (Felleman and Van
Essen, ’91).Nevertheless, keeping this cautionary note in
mind, the results of such analysis (Fig. 18E) suggest an
even more pronounced increase with hierarchy due to the
smaller size of areas V4 and 7a.
Effect of injection size on labeling
of intrinsic connections
An interesting question concerns the spatial relationships between patterns of connections emanating from
neighboring cortical sites. A simple way to approach this
question is by injecting larger amount of tracer, thus
enlarging the uptake zone at the injection site. Comparing
patch geometry resulting from large vs. small injections can
provide some hints as to the neighborhood relationship of
connectional patterns (see Discussion). Biocytin is a particularly convenient tracer for this purpose because an unam-
Figure 12
INTRINSIC CONNECTIONS AND CORTICAL HIERARCHY
biguous demarkation of its effective uptake zone is provided
by the distribution of labeled somata.
Figure 19 shows the pattern of labeling produced by a
large injection in area V4. Similar to the cases with smaller
injections (Figs. 9-11], the intrinsic connections are clearly
patchy. However, substantially more clusters were labeled
by the larger injection (compare Fig. 19 with Figs. 9, 11).
The clusters are organized in vertical columns. Unlike the
smaller injection sites, many labeled fibers could be found
in infragranular layers as well. Another clear characteristic
of large biocytin injections is the retrograde labeling of cell
bodies away from the injection site. Interestingly, the
retrogradely labeled cells and the anterogradely labeled
axonal clusters are not always colocalized, suggesting that
not all intrinsic connectivity is reciprocally organized (Fig.
19B, dots). Similar phenomenon was reported by Boyd and
Matsubara (’91).
We further studied the effects of enlarging the injection
site, by a quantitative analysis of the intrinsic connections.
Figure 20 shows the results of this analysis comparing
small and large injections in areas V1 and V4.
Somewhat unexpectedly, the width of individual clusters
barely changes as a result of enlarging the injections (Fig.
20A). Rather, there is a substantial increase in the number
of labeled patches (Fig. 20B). Thus, in both areas V1 and
V4, the number of patches more than doubled as a result of
about a fourfold increase in injection site diameter. This
increase is mainly due to the increased density of patches
since the distance of labeled connections from the injection
site did not increase markedly following the large injections. The implication of these results to principles of
intrinsic connectivity are addressed in the Discussion.
Analysis of bouton distribution along axons
The real parameter determining the functional impact of
the intrinsic connectivity is not the length and spatial
distribution of the axon trunks, but rather the distribution
of synapses these axons make with their target neurons.
Electron microscopy studies (Somogyi and Cowey, ’81;
Schuz and Munster, ’85; Braitenberg and Schuz, ’91) have
shown that most of these synapses are located at axonal
protrusions, termed boutons, which can be seen at high
magnification along the axons and their terminations.
Examples of such boutons taken from areas V1 and 7a are
shown in Figure 21. We analyzed the distribution of 1,671
boutons in 57 axons (approximately 400 boutons per area)
in order to obtain a more realistic account of the potential
influence the intrinsic axons may impart at different cortical sites.
The first question of interest was whether bouton frequency along the axonal trunk increases as the axons
penetrate into their target patches. Figure 22A indicates
that this is not the case. Thus, the frequency histograms of
interbouton intervals taken from axons inside and outside
of patches appear remarkably similar. Thus, the increase in
axonal impact in the clusters is achieved solely through
Fig. 12. Intrinsic horizontal connections in area 7a. Darffield,
low-power image of a tangential section from flattened area 7a. Only the
lateral part of the pattern of connections is shown. Curved arrow marks
injection site. Note the strikingly large halo around the injection site.
The fibers are much more spread out and loosely distributed. However,
they form many unequivocally defined patches of connections. Open
arrows indicate fiber clusters. The areas in rectangles a 4 are shown
enlarged in Figure 13. Scale bar = 200 pm.
33
increase in branching and bending of axons and not through
changes in synapse distribution along individual axons.
Second, we investigated whether bouton density may change
at higher order areas. The results of this analysis are shown
in Figure 22B. Here too, the frequency histograms are very
similar across areas. Thus, it can be concluded that the
marked change we observed in the spread of connections is
not reflected in the synaptic density along individual axonal
trunks.
Finally, inspecting the overall mean curve (Fig. 223),
which represents the entire bouton population, reveals that
interbouton intervals less than 2-3 pm in size are quite
rare, suggesting a low bound on the spacing distance
between boutons. Note that the frequency distribution at
high interbouton intervals declines following a damped
oscillatory curve. One interpretation of such oscillation
could be that bouton frequency is quite constant (with
about 180 boutons/mm of axon) and the oscillatory curve
could have been produced by a small percentage of boutons
missed, resulting in whole number multiples of the basic
interbouton interval.
Statistical analysis of the entire data set confirmed that
bouton distribution remains invariant under all conditions.
We used two-way ANOVA with interactions on log interval
to normalize the asymmetric distribution. The main effects
were areas (4 levels) and in and out of patches ( 2 levels).
The results showed no significant interaction ( P = 0.2).
Another way to search for clustering of synaptic boutons
is to look for correlation between consecutive interbouton
intervals (Braitenberg and Schuz, ’91). This analysis is
shown in Figure 23, which is a plot of bouton intervals
against their subsequent intervals. A tendency for bouton
density to wax and wane along the axon should be reflected
in this scatter diagram as a clustering around the center
diagonal. No such effect could be observed, indicating that
boutons were distributed uniformly along the axon.
DISCUSSION
Spread of connections
The principle of cortical hierarchy (Rockland and Pandya, ’79; Van Essen and Maunsell, ’83) organizes the entire
set of cortical areas along an ascending sequence of levels.
Its basis is the unique laminar “signature” of ascending
and descending extrinsic (i.e., between areas) connections
(for a recent review see Felleman and Van Essen, ’91). By
direct comparison of intrinsic connections across several
cortical areas obtained through careful matching of tracer
type, injection size, laminar location, and axis of penetration, we were able to unequivocally establish that the
extrinsically determined hierarchy is tightly correlated to
the intrinsic circuitry of cortical areas. More specifically,
our results show that at progressively higher hierarchical
levels, the spread of intrinsic connections markedly increases. This spread of connections constitutes an independent, intrinsically determined attribute of the cortical
hierarchy and thus provides further support for the hierarchy scheme.
As for functional properties that may correlate to the
lengthening of intrinsic connectivity, the most relevant one
seems to be the size of the “point image,” i.e., the cortical
domain that can be activated by stimulation of a single
retinal point (McIlwain, ’86). The point image may, in
effect, be an approximate measure of the tangential spread
of activation in a cortical area and thus can most intuitively
34
Y. AMIR ET AL.
Fig. 13. A-D: Examples of fiber clusters (open arrows) in area 7a. Enlarged images of areas demarcated
by rectangles a 4 in Figure 12. Note the diversity between patches in the size and density of axonal
ramification. Nevertheless, the tendency for clustering is quite evident. Scale bar = 100 km.
35
INTRINSIC CONNECTIONS AND CORTICAL HIERARCHY
\
0
L
A
0
......I
. ..?A. ....
*.
..I..
.
.
I
.
.
.
.
.
.
.
**i
..v......m
. .. :
I..
\
\
0
/
0
..I
0
\
1mm
\
Fig. 14. Schematic drawing of the main clusters of intrinsic fibers
produced by biocytin injection in area 7a. This drawing is from the same
injection shown in Figure 12. Same symbols and conventions as in
Figures 4 and 8. Dotted line indicates the area shown in Figure 12. The
fundi of the STS and IPS form, respectively, the posterior and anterior
borders (thin lines) of the section. Here the spread of connections
reaches its highest level. Note also the extremely oval and asymmetric
distribution of labeled patches in this area.
Y. AMIR ET AL.
36
0
0
\
I
lmrn
-
Fig. 15. Intrinsic horizontal connections in inferotemporal cortex
(IT). Computer-assisted drawing of labeled fibers produced by biocytin
injection in the dorsal part of posterior IT. The drawing is based on a
composite of five neighboring sections. Note the highly diffuse nature of
the connections and occasional clusters.
be related to the spread of intrinsic connections. In most
reports, the point image is not directly measured, and we
estimated it by calculating the product of average receptive
field size and magnification factor at a specific eccentricity.
This calculation does not take into account receptive field
scatter and asymmetries in the visuotopic map. Consequently it should be regarded as only a crude estimate of the
point spread. Furthermore, point image estimates change
under different experimental conditions and different types
of visual stimuli (Tootell et al., ’88)and thus vary greatly in
different studies. In spite of all these drawbacks, even
tentative estimates of the point image may be instructive in
assigning plausible functional roles to the intrinsic connections under study.
The putative anatomical counterpart to the point image,
i.e., the spread of connections, will be represented here by
two parameters-the
average range and the maximum
range. The average range was defined as twice the average
patch distance and provides an estimated spread for the
bulk of the intrinsically connected patches. The maximum
range was defined as the combined distance to the two
furthermost patches in each case and gives an upper
boundary on the range of influence of intrinsic connections.
Several reports estimated V1 point image to be about
1-1.5 mm distance at 5 degrees eccentricity (Hubel and
Wiesel, ’74; Dow et al., ’81; Van Essen et al., ’84; Tootell et
al., ’88).In comparison, our estimate for the average range
of connections was 1.3 mm, which is quite close, while the
maximum range was 2.14 mm, which exceeds the point
spread (not accounting for field scatter).
In area V2, estimating from the maps of Gattas et al.
(’81), we obtained a point spread value at 5 degrees
eccentricity of 2 mm, which again matches quite reasonably
our average range value of 2.6 mm but is below the
maximum range value (4.2 mm). In areaV4, the situation is
more complex, since the visual map is quite disordered.
From the calculations of Gattas et al. (’881,we obtained an
estimate of 2.9 mm while the receptive field measures of
Boussaoud et al. (’91) imply a value of 4.4 mm at 5 degrees
eccentricity. These values should be compared to our
estimated average and maximum ranges of intrinsic connections for V4, which are 3.56 and 5.7 mm, respectively.
Finally, area 7a apparently lacks organized retinotopy and
contains neurons with exceedingly large receptive fields, up
to 100 degrees diameter (e.g., Andersen et al., ’90). These
characteristics suggest an extremely large point spread in
7a, encompassing a substantial part of the entire area,
which is about 10 mm in extent (Felleman and Van Essen,
’91). Our estimates of the average and maximum ranges of
intrinsic connections in this area are 4.4 mm and 9 mm,
respectively, ranging over roughly 44% (average range) and
90% (maximum range) of the extent of area 7a.
To conclude, there is a qualitative correlation between
the spread of intrinsic connections, and the calculated point
image at different levels of the cortical hierarchy. The
average range appears to match this physiological measure,
while the maximum range may exceed it. These results
suggest that the bulk of intrinsic connections could be
playing a direct role in constructing cortical receptive fields.
Experimental evidence that this is the case was provided by
Bolz and Gilbert (’86, ’89) for the interlaminar horizontal
connections in area V1. On the other hand, the results also
imply that the most far reaching fibers, relating quite
disparate retinotopic points, may play a role in producing
effects beyond the borders of the “classical” receptive field,
such as modulatory, “contextual” effects (Gilbert et al.,
’90).
Finally, theoretical models have suggested that the coordinate transformation from visuotopic to nonvisuotopic
mapping, which occurs at high order cortical areas, is aimed
at bringing together retinotopically disparate points so as to
shorten the length of connections needed to perform nonretinotopically related global computations (e.g., Barlow, ’81).
The fact that the intrinsic connections in the nonretinotopically organized area 7a are far longer than in the retinotopically organized areas, argues against this hypothesis. Rather,
it appears that the long range connections could be one of
the factors that produce the disorderly retinotopy and large
receptive fields in these nonretinotopic representations.
INTRINSIC CONNECTIONS AND CORTICAL HIERARCHY
Fig. 16. Examples of injection site halos in the different areas.
Darkfield views of the injection site halos produced by biocytin injections. Curved arrows mark injection sites. A: Injection in areaV1. Note
the relatively small and densely packed appearance of the halo. B:
Injection in area V2. Here the halo is somewhat increased and patches
form at a short distance from the injection site. C : Injection in area V4.
37
In this area, the size of the halo increases markedly with diffuse
projections emanating to relatively large distances. D: Injection in area
7a. Here the halo is particularly striking with fibers organized in a
“sunburst” radiation. In all these cases injection size was similar. Scale
bar = 100 p,m.
Y. AMIR ET AL.
B
0.131
0.15
cortical Area
v1
v2
El
0
v4
7a
7-
n
c1
E
E
0.1
0.10
W
0.075
Q
0.068
Q)
7-
2
5
a"
7
0.05
.c,
0.W
&o.os
C
O.OSAJ.I
ai-o.is
O.IJ-O.Z
0.2-0.25
0.25-0.3
0.34.3s 0.35-0.4
Patch Area (mm2)
v2
h
W
7a
0.31
Cortical Area
#
V4
Cortical area
D
50
h
v1
0.27
0.30
E
E
40
VJ
Q)
3
W
.c)
a"
5
L
l
O
c
0
0.20
;
sa"
30
20
0.10
.m
;
.c)
10
0.00
0
0-0.1
0.1-0.2
0.2-0.3
0.3-0.4
0.4-0.5
0.5-0.6
Patch Width (mm)
Fig. 17. Quantitative analysis of intrinsically connected patches in
cortical areas V1, V2, V4, and 7a. A: Frequency distributions of patch
area. Histogram bin size is 0.05 mm2. Different shading indicates
patches belonging to different cortical areas. B Average patch area.
Numbers above histogram bars are the average patch area. Error
The patchy nature of the connections
Invariably, in all the areas studied, the intrinsic connections tend to form patches or clusters. The fibers within a
patch are more densely packed in the primary areas V1 and
V2, and are more diffuse in the higher order areas. The
innervation of the patches is most often densest near the
injection site and declines with distance. In all the areas,
there are also many axons running in isolation without
forming clear patches. The most remarkable fact about the
clustering of the connections is how similar they appear
across areas despite relatively minor increases in patch
dimensions at higher order areas.
In area V1, it has been shown in several studies that the
patchy nature of the intrinsic connections is related to the
v1
v2
v4
7a
Cortical area
bars = 1 S.E.M. C : Frequency distribution of patch width. Same
symbols as in A. D: Average patch width in the different cortical areas.
Same symbols as in B. Note the small but consistent increase in patch
dimensions with cortical hierarchy.
Fig. 18. Quantitative analysis of separation and distances of intrinsically connected patches in cortical areas V1, V2, V4, and 7a. A:
Frequency histogram of distances between neighboring patches. B:
Average distance between neighboring patches. Numbers above histogram bars are the average interpatcb distance. C: Frequency histogram
of patch distance from the center of injection site. Same symbols as in A.
D: Comparison of patch distances from the injection sites across areas.
Two measures are shown: average distance (light gray) and maximum
range (dark gray, combined distance to the two most extreme patches;
see text for more details). E: Same histogram as D, but normalized to
the extent of cortical area (see text for details). These results indicate a
clear relationship between the spread of connections and cortical
hierarchy. Particularly striking is the normalized distance which
undergoes a 13-fold increase from V1 to 7a. This change is quite
compatible with the reported increase in receptive field size with
cortical hierarchy (see Discussion).
39
INTRINSIC CONNECTIONS AND CORTICAL HIERARCHY
A
B
1.56
7
Cortical Area
E
1.33
7
1.1-
-
Q)
u
1.1s
s
Y
m
1.0-
n
0-
2
0.61
Y
m
0.1
-
0.0
-+-
*
E
Q)
H
0-0.5
n
b
&!
W
m
0.5-1
1-1.5
1.5-1
1-2.5
2.5-3
3-3.5
3.5-4
4-43
Vl
Interpatch Distance (mm)
v1
.s
u
m
40
1
0
6.0
-
5.0
-
Mean distance from center of injec
Maximum range
5.72
.
I
E
c1
u
.",
*
0
c
0
7.0
c
Y
7a
8.98
Cortical Area
Q)
E
v4
Cortical Area
D
_i I
v2
c
4.19
4.0 -
.I
E
20
.
3
Y
u
m
20
0
&
3.0
8
2.0
-
5
cn
.
I
0-0.5
0.5-1
1-1.5
1.5-2 2-25
2.53
3.3.5
3.5-4 4-4.5 4.5-5
Q
5-5.5
0.0
v1
Distance from injection site (mm)
v2
Cortical Area
n
E
Y
Mean Distance (Normalized)
Maximum Range (Normalized)
v1
v4
v2
VQ
Cortical Area
Figure 18
83.8
7a
7a
Y. AMIR ET AL.
40
1
M
B
4
IFig. 19. Intrinsic connections labeled by a large injection in area V4.
A Lightfield, low-power image of a tangential section from flattened
areaV4, which received an injection (curved arrow) spanning 900 wm in
diameter (no. 3 in Table 1).Open arrows indicate examples of clearly
defined patches. Scale bar = 1 mm. B: Schematic drawing of the same
injection shows main clusters chosen for analysis. Same symbols and
conventions as in Figures 4 and 8. The anterior border of the section
(thin line) corresponds to the posterior hank of the STS. Dots indicate
neurons retrogradely labeled by the hiocytin injection. Their distribution is only loosely correlated to the anterogradely labeled patches.
Comparing this injection to the small one shown in Figure 11reveals a
marked increase in the number of clusters produced by the larger
injection site.
underlying functional columnar organization of cortical
neurons (Livingstone and Hubel, '84b; Gilbert and Wiesel,
'89; Malach et al., '92b; Blasdel et al., '92). Although it is
debatable how precise this relationship is (e.g., Kisvarday et
al., '92), if it does hold true at high order areas, the present
results suggest that a basic cortical module, a column about
200-300 pm in width, pervades throughout the entire
macaque visual cortex.
Our results do not rule out the possibility that larger
subdivisions also exist within the various cortical areas.
The small columnar modules suggested here may constitute building blocks for these larger aggregates. Thus, the
CO stripes of area V2 were shown by several groups to form
connectional and functional subdivisions within area V2
(Livingstoneand Hubel, '84a,b; DeYoe and Van Essen, '85;
Shipp and Zeki, '85; Tootell and Hamilton, '89). The
organization of intrinsic connections in area V2 suggests
that these CO compartments may be subdivided into
smaller, discrete patches of blob-like dimensions (e.g., see
Fig. 6). These findings are in line with several previous
reports suggesting a heterogenous organization within the
V2 CO stripes (Wong-Riley and Carrol, '84; Tootell and
Hamilton, '89; Zeki and Shipp, '89). Similarly, large scale
(2-4 mm wide) compartments were postulated in area V4
(Zeki and Shipp, '89; Van Essen et al., '90). Our results
suggest these may be aggregates of columns having bloblike dimensions.
In a previous study (Malach, '92a), it was hypothesized
that the width of CO blobs is determined by the size of
dendritic arbors of upper layer pyramidal neurons. A
consequence of such size relationship is a graded transition
in the sampling properties of blob neurons. Thus, neurons
at the blob centers have dendrites which are fully confined
to blob territory while neurons at the blob borders have
dendrites sampling equally from blob and interblob territories. The fact that throughout visual cortex, the intrinsic
connections are organized in clusters of "blob-like'' dimensions, suggests that this phenomenon may be a universal
one, producing a graded diversity of dendritic sampling
properties throughout visual cortex.
Overall organization
Our results suggest that the mapping from a cortical
region onto neighboring target sites does not constitute a
simple transformation, i.e., it is not the case that neighboring cortical points project to neighboring and overlapping
patches. Rather, nearby cortical points innervate disparate
sets of patches. These conclusions are inferred from our
41
INTRINSIC CONNECTIONS AND CORTICAL HIERARCHY
B
A
-
m
401
I
33
Small Injections
Large Injections
I1
v4
v1
Cortical area
v1
v4
Cortical Area
Fig. 20. Intrinsic connections produced by small vs. large biocytin
injections in areas V1 and V4. A: Effects of enlarging the injection site
on patch width. Note that patch width is barely affected by the
increased injection site both in area V1 (left bars) and V4 (right bars).
B: Effects on number of patches. Here increasing the size of injections
had a marked affect. Patch number both in areas V1 and V4 more than
doubled due to the increased in size of injection sites.
observation that increasing the size of the injection sites
leads mainly to an increase in the number of labeled
patches and not their size or distance from the injection.
These points are illustrated in Figure 24, which shows, in
a schematic form, two plausible organizations of intrinsic
connections. Note that enlarging the injection site (i.e.,
labeling simultaneously the connections from the cortical
sites indicated by dark and light patches in Fig. 24) will lead
to larger patches in the model shown in Figure 24A, and to
more patches in the model shown in Figure 24B. The
present results are compatible with the second alternative
(Fig. 24B); however, we should stress that this is quite
speculative at this stage and will require a double label
study to be fully confirmed. The model does not necessarily
imply a total disorder in the mapping from one cortical site
onto its surround. An attractive alternative, illustrated
here, could be that the intrinsic connections emanating
from each cortical site are organized in an overall oval shape
(e.g., Fig. 14). The orientation of this oval undergoes a
systematic rotation (in the tangential plane) as the injection
site shifts to a neighboring location (see change from light
to dark patches in Fig. 24B).
Mitchison and Crick ('82) have postulated a similar
systematic rotation, related to the orientation specificity of
intrinsically projecting neurons, in area V1. However, the
existence of radially symmetric "rosette-like'' organizations in area V1 (Fig. 5) suggests the situation in this area
may be more complex. It is not clear what is the functional
significance of these contrasting oval and rosette forms. We
are currently exploring the possible relationships of these
organizations to the underlying cortical functional architecture.
First, it suggests that the mechanism by which information
is targeted to particular cortical sites is based on increasing
axonal density at the cluster sites, through increased
branching rather than elevating the synaptic density within
each axon, Thus, it is clearly the case that the axons
distribute information to numerous neurons along their
way.
In particular, this observation is relevant to the possible
functional role of the halo of axons around the injection
site. In this dense plexus, axons appear to ramify fairly
uniformly, carrying numerous boutons. Although some of
these axons may belong to inhibitory interneurons, many of
them can be traced over longer distances and are presumably excitatory. This local halo of axons constitutes a
cortical circuit most closely resembling a truly distributed
neural network with massive interconnections among neighboring neurons. These networks are reminiscent of models
for associative learning (e.g., Hopfield, '82). It is interesting
that higher order areas are characterized by a marked
increase in the size of such halos (see Fig. 16).
Bouton distribution
Our quantitative analysis revealed that the density of
boutons distributed along the intrinsic axons is invariant
both inside and outside the patches and in different cortical
areas. These results lead to several important conclusions.
Comparison to other studies
While many studies have explored the extrinsic connections between different visual areas (for detailed review, see
Felleman and Van Essen, '91), the information regarding
intrinsic connections in primates is more limited. The most
extensive studies of these connections in primates were
performed in areaV1 (Rockland and Lund, '83;Livingstone
and Hubel, '84b; Blasdel et al., '85). Our results corroborate
those of Livingstone and Hubel ('84) in estimating the
length of intrinsic connections, and in the tendency for
segregation into blob and interblob compartments. The
present results extend their findings in showing that the
halo around the injection sites contains a massive amount
of bouton-carrying axons and thus should be considered a
powerful substrate for cross-compartmental interactions.
In area V2, several studies have revealed patchy intrinsic
connections both in macaque and squirrel monkeys (Tigges
Figure 2 1
INTRINSIC CONNECTIONS AND CORTICAL HIERARCHY
I5
-
I
43
B
In Patch
Out of Patch
-
Cortical Area
-
10
v1
v2
v4
7a
Areal Mean
5
0
.2
.4 .6 . 8
.10.12.14.16.18.20.22.24.26.28.30.32.34.36
Inter-bouton Interval (pm)
Inter-bouton Interval (pm)
Fig. 22. Quantitative analysis of bouton distribution on intrinsic
axons. A Comparison of frequency distributions of interhouton intervals inside and outside of patches. On the abscissa are shown intervals
between consecutive boutons. On the ordinate is shown the fraction of
intervals of particular length. Data is based on analysis of about 800
boutons per category. Open circles, bouton on axons within patches;
filled circles, boutons on axons outside patches. Note the close similarity in the frequency distribution of these two populations. B: Comparison of the frequency distributions of interbouton intervals in the
different cortical areas. Different areas are indicated by different
symbols. Data for each area is based on measurement of about 400
bouton intervals. Areal mean, i.e., the frequency distribution of the
entire population of boutons studied, is shown by a thicker line.
Regardless of area sampled, all the distributions look remarkably
similar. Thus, in contrast to the marked changes observed in the
distribution of intrinsic axons in different areas, the bouton density
appears to be an invariant property.
et al., '74; Wong-Riley, '79; Cusick and Kaas, '88; Livingstone and Hubel, '84a). Significantly, these studies employed a wide array of tracers and received results quite
similar to those reported here based on biocytin labeling,
further confirming the reliability and sensitivity of this new
tracer. Rockland ('85) and Rockland and Lund ('83)have
reported a "lattice-like" configuration of intrinsic connections after making very large tracer injections. Such lattice
may have been created by merging of nearby and overlapping patches due to the massive area infiltrated by the
tracer. In such injections, it is difficult to estimate the
effective uptake zone of the tracer. Nevertheless, the estimated maximal spread of connections (2.5-3 mm) is quite
close to our half range estimate of about 2 mm. While
preparing the manuscript, a report on intrinsic connectivity
within area V4 has appeared (Yoshioka et al., '92). Although these authors emphasized the analysis of coronal
sections, their report agrees with our results concerning the
patchy nature of intrinsic connections in macaque area V4
and their estimate of patch extension (6-8 mm) and patch
size (250-450 pm) agree well with our estimates of 5.7 mm
maximum range and 270 pm patch width, respectively.
Finally, drawings in studies of area 7a extrinsic connections
(e.g., Pandya and Seltzer, '82; Andersen et al., '90) also
suggest the existence of an extensive set of intrinsic connections in this area.
Fig. 21. Examples of boutons on intrinsic axons. A-D: Mediumpower, lightfield photomicrographs showing axons carrying boutons in
area V1. Arrows point to examples of boutons. Note that the halo
around the injection site contains numerous boutons (C). E,F Highpower, lightfield photomicrographs showing boutons on axons in area
V1. The houtons (arrows) can be either at the end of short branches or,
as thickening, directly on the trunk. G , H Reflected light, confocal laser
scanning microscope images of axonal boutons in area 7a. This is an
extended focus view through a depth of 32 pm. White arrows point to
examples of boutons. Open arrow points to a bouton which is enlarged
further in H. Scale bar = 50 pm for A-D; 10 p m for E,F; 2.5 pm for
G,H.
Interpretive uncertainties
Several potential problems may be relevant to the present experiments. First, it could be argued that some of the
intrinsic patches included in the quantitative analysis are
located in neighboring cortical areas. Such misclassification
may bias the estimate of the spread of connections towards
larger values. This problem could arise only in the higher
cortical areas, since the boundaries of areas V1 and V2 were
unequivocally determined by the CO staining. At higher
order areas, the situation is more problematic because the
borders cannot be so precisely defined, even in the cases in
which we obtained excellent myelin staining. Two points
confirm that the patches included in our analysis were
intrinsic. First, in most cases, the intrinsically labeled
patches form a continuous array, suggesting they all belong
to the same set (e.g., Figs. 11,14). Second, laminar analysis
of the labeled patches showed that in all cases (not including cases with massive injections), the patches are distributed almost exclusively in superficial layers. This is the
expected pattern of intrinsic projections of supragranular
cortical sites. Such a pattern rules out the possibility that
some of the patches were projections to areas of different
hierarchical level. In particular, neighboring area V4 are
areas VOT and DP; both receive forward connections from
V4, which should terminate in layer 4. Similarly, the
extrinsic descending projections from area 7a to neighboring areas DP and LIP should be confined to layers 1and 6.
Thus, the supragranular location of the labeled patches in
Y. AMIR ET AL.
44
0
0
04
0
,
.
I
.
.
I0
,
.
18
.
.
x)
.
,
m
.
.
m
.
.
?s
.
.
40
.
.
48
.
.
SJ
(1-1)t~Inter-bouton Interval (p>
Fig. 23. Relationship between consecutive interbouton intervals.
Scatter diagram showing each interval length on the abscissa, and the
length of the subsequent interval on the ordinate. If the density of
boutons along the axons shows a tendency to wax and wane, we would
expect the points in the scattergram to be concentrated along the center
diagonal. No such effect can be discerned, suggesting bouton density is
rather uniformly scattered along the axon.
Fig. 24. Neighborhood relationship of intrinsic connections. Two
hypothetical schemes illustrating possible arrangements of intrinsic
connections emanating from neighboring cortical sites. In the diagrams, the arrows represent intrinsically connected columns in superficial layers of visual cortex. Black and white patches belong to two sets of
superficial layer columns connected to two neighboring cortical sites. A
Continuous mapping. Cortical sites abutting each other are connected
to sets of sites which adjoin each other as well. I t is easy to see that in
such a case, increasing the injection site should result in increased size
of labeled patches. This was not observed in the present study. B:
Discontinuous mapping. Cortical sites abutting each other are connected to separated sets of cortical sites. In this example, the sets are
arranged in two elongated stretches having their long axis rotated
relative to each other. In the general case of discontinuous mapping,
increasing the size of injection should result in increased number of
patches with no change in patch dimension. In the specific model
suggested here, small injections should produce highly elongated
stretches of patches. Both these phenomena were observed in the
present study.
INTRINSIC CONNECTIONS AND CORTICAL HIERARCHY
the present study indicates they are part of the intrinsically
connected system.
Finally, a major unresolved concern is the possibility of
heterogeneous intrinsic connectivity of different subdivisions within each cortical area. We cannot rule out the
possibility that certain sites in each area connect over
longer or shorter distances than our results indicate.
Judging from the connectivity patterns within area V1, in
which numerous injections were made (e.g., Livingstone
and Hubel, '84b; our own unpublished observations) the
length of intrinsic connections is fairly constant regardless
of the tissue compartment injected. This situation may
change at higher order areas. However, our injections in
areas V4 and 7a gave labeling patterns, which were quite
consistent across cases, indicating that such heterogeneously connected cortical sites are likely to be rare.
Conclusions
The present study indicates that the tangential spread of
intrinsic connections changes markedly with the cortical
hierarchy. On the other hand, the columnar structure, as
reflected in the clustered nature of the connections, remains essentially invariant across areas. This invariant
clustering suggests that cortical modules of similar size
exist throughout visual cortex.
ACKNOWLEDGMENTS
We thank the photography and graphics departments for
assistance, and Dr. S. Barash for helpful comments on the
manuscript. This research was funded by a grant from the
Israel Academy of Sciences and Humanities to R. Malach.
LITERATURE CITED
Andersen, R.A., G.K. Essick, and R.M. Siegel (1985) The encoding of spatial
location by posterior parietal neurons. Science 230t456-458.
Andersen, R.A., C. Asanuma, G. Essick, and R.M. Siegel (1990) Corticocortical connections of anatomically and physiologically defined subdivisions
within the inferior parietal lobule. J. Comp. Neurol. 296t65-113.
Barlow, H.B. (1981) Critical limiting factors in the design of the eye and
visual cortex. Proc. R. SOC.
Lond. B 212:l-34.
Blasdel, G.G., J.S. Lund, and D. Fitzpatrick (1985) Intrinsic connections of
macaque striate cortex: Axonal projections of cells outside lamina 4C. J.
Neurosci. 5t3350-3369.
Blasdel, C.G., T. Yoshioka, J.B. Levitt, and J.S. Lund (1992) Correlation
between patterns of lateral connectivity and patterns of orientation
preference in monkey striate cortex. Soc. Neurosci. Abst. 18:169.4.
Bolz, J., and C.D. Gilbert (1986) Generation of end-inhihition in the visual
cortex via interlaminar connections. Nature 320:362-365.
Bolz, J., and C.D. Gilbert (1989) The role of horizontal connections in
generating long receptive fields in the cat visual cortex. Eur. J. Neurosci.
1.263-268.
Boussaoud, D., R. Desimone, and L.G. Ungerleider (1991) Visual topography
of area TEO in the macaque. J. Comp. Neurol. 306:554-575.
Boyd, J., and J. Matsubara (1991) Intrinsic connections in cat visual cortex:
A combined anterograde and retrograde tracing study. Brain Res.
560207-215.
Braitenberg, V., and A. Schuz (1991) Anatomy of the Cortex: Statistics and
Geometry. Studies of Brain Function. Vol. 18. Berlin: Springer Verlag.
Chapman, B., K.R. Zahs, and M.P. Stryker (1991) Relation of cortical cell
orientation selectivity to alignment of receptive fields of the geniculocortical afferents that arhorize within a single orientation column in ferret
visual cortex. J. Neurosci. 11~1347-1358.
Cusick, C.G., and J.H. Kaas (1988) Cortical connections of area 18 and
dorsolateral visual cortex in squirrel monkeys. Visual Neurosci. 1211237.
DeYoe, E.A., and D.C. Van Essen (1985) Segregation of efferent connections
and receptive field properties in visual area V2 of the macaque. Nature
31 7:5%61.
45
Dow, B.M., A.Z. Snyder, R.G. Vautin, and R. Bauer (1981) Magnification
factor and receptive field size in foveal striate cortex of monkey. Exp.
Brain Res. 44.213-288.
Felleman, D.J., and D.C. Van Essen (1991) Distributed hierarchical processing in primate cerebral cortex. Cerebral Cortex 1:147.
Gallyas, F. (1979) Silver staining of myelin by means of physical development. Neurol. Res. 1,203-209.
Gattas, R., C.G. Gross, and J.H. Sandell (1981) Visual topography of V2 in
the macaque. J. Comp. Neurol. 201t519-539.
Gattas, R., A.P.B. Sousa, and C.G. Gross (1988)Visuotopic organization and
extent of V3 and V4 of the Macaque. J. Neurosci. 8:1831-1845.
Gilbert, C.D., and T.N. Wiesel(1983) Clustered intrinsic connections in cat
visual cortex. J. Neurosci. 3:1116-1133.
Gilbert, C.D., and T.N. Wiesel (1989) Columnar specificity of intrinsic
horizontal and corticocortical connections in cat visual cortex. J. Neurosci. 9:2432-2442.
Gilbert, C.D., J.A. Hirsch, and T.N. Wiesel (1990) Lateral interactions in
visual cortex. Cold Spring Harbor Symp. Quant.Biol.55:663-677.
Gross, C.G. (1972)Visual functions of inferotemporal cortex. In R. Jung (ed):
Handbook of Sensory Physiology. Berlin: Springer-Verlag, pp. 451482.
Hopfield, J.J. (1982) Neuronal networks and physical systems with emergent collective computational abilities. Proc. Natl. Acad. Sci. 7925542558.
Hubel, D.H., and T.N. Wiesel(1962) Receptive fields, binocular interaction
and functional architecture in the cat's visual cortex. J. Physiol. (Lond)
160: 106-154.
Hubel, D.H., and T.N. Wiesel(1968) Receptive fields and functional architecture of monkey striate cortex. J. Physiol. (Lond) 195:215-243.
Hubel, D.H., and T.N. Wiesel (1974) Sequence regularity and geometry of
orientation columns in monkey striate cortex.J. Comp. Neurol. 158:267-294.
King, M.A., P.M. Louis, B.E. Hunter, and D.W. Walker (1989) Biocytin: A
versatile anterograde neuroanatomical tract-tracing alternative. Brain
Res. 497:361-367.
Kisvarday, Z.F., K.A.C. Martin, T.F. Freund, Z.S. Magloczky, D. Whitteridge, and P. Somogyi (1986) Synaptic targets of HRP-filled layer 111
pyramidal cells in the cat striate cortex. Exp. Brain Res. 64:541-552.
Kisvarday, Z.F., D . 3 . Kim, U.T. Eysel, and T. Bonhoffer (1992) Functional
and structural topography of lateral inhibitory connections in cat visual
Neurosci. Abst. 18:169.2.
cortex (area 18).SOC.
Livingstone, E.M., and Hubel, D.H. (1984a) Anatomy and physiology of a
color system in the primate visual cortex. J. Neurosci. 4~309-356.
Livingstone, M.S., and D.H. Hubel (198413) Specificity of intrinsic connections in primate primary visual cortex. J. Neurosci. 4 ~ 2 8 3 0 4 8 3 5 .
Lund, J.S., T. Yoshioka, and J.B. Levitt (1993) Comparison of intrinsic
connectivity in different areas of makaque monkey cerebral cortex.
Cerebral Cortex 3~148-162.
Malach, R. (1992a) Dendritic sampling actoss processing streams in monkey
striate cortex. J. Comp. Neurol. 315303-312,
Malach, R., Y.Amir, E. Bartfeld, and A. Grinvald (1992h) Biocytin injections
guided by optical imaging, reveal relationships between functional
architecture and intrinsic connections in monkey visual cortex. SOC.
Neurosci. Abst. 18~169.3.
Matsubara, J.A., M.S. Cynader, and N.V. Swindale (1987) Anatomical
properties and physiological correlates of the intrinsic connections in cat
area 18.J. Neurosci. 7t1428-1446.
McGuire, B.A., C.D. Gilbert, P.K. Rivlin, and T.N. Wiesel (1991) Targets of
horizontal connections in Macaque primary visual cortex. J. Comp.
Neurol. 305:370-392.
McIlwain, J.T. (1986) Point images in the visual system: New interest in a n
old idea. Trends Neurosci. 9~354-358.
Mitchison, G., and F. Crick (1982) Long axons within striate cortex: Their
distribution, orientating and patterns of connections. Proc. Natl. Acad.
Sci. 79,3661-65.
Pandya, D.N., and B. Seltzer (1982) Intrinsic connections and architectonics
of posterior parietal cortex in the rhesus monkey. J. Comp. Neurol.
204: 196-2 10.
Paperna, T., and R. Malach (1991) Patterns of sensory intermodality
relationships in the cerebral cortex of the rat. J. Comp. Neurol.
308t432-456.
Perret, D.I., E.T. Rolls, and W. Caan (1982) Visual neurons responsive to
faces in the monkey temporal cortex. Exp. Brain Res. 47t329-342.
Rockland, K.S. (1985) A reticular pattern of intrinsic connections in primate
area V2 (area 18).J. Comp. Neurol. 2 3 5 4 6 7 4 7 8 ,
46
Rockland, K.S., and D.N. Pandya (1979) Laminar origins and terminations
of cortical connections of the occipital lobe in the rhesus monkey. Brain
Res. 179:3-20.
Rockland, K.S., and J.S. Lund (1982) Widespread periodic intrinsic connections in tree shrew visual cortex (area 17). Science 215:1532-1534.
Rockland, K.S., and J.S. Lund (1983) Intrinsic laminar lattice connections in
primate visual cortex. J. Comp. Neurol. 216r303-318.
Schuz, A., and A. Munster (1985) Synaptic density on the axonal tree of a
pyramidal cell in the cortex of the mouse. Neuroscience 15:33-39.
Shipp, S., and M. Zeki (1985) Segregation of pathways leading from area V2
to areas V4 and V5 of macaque monkey visual cortex. Nature 315:322-325.
Silverman, M.S., and R.B.H. Tootell (1987) Modified technique for cytochrome oxidase histochemistry: Increased staining intensity and compatibility with 2-deoxyglucose autoradiography. J. Neurosci. Meth. 19:l-10.
Somogyi, P., and A. Cowey (1981) Combined Golgi and electron microscopic
study on the synapses formed by double bouquet cells in the visual cortex
of the cat and monkey. J. Comp. Neurol. 195x547466.
Tigges, J., W.B. Spatz, and M. Tigges (1974) Efferent cortico-cortical fiber
connections of area 18 in the squirrel monkey (saimiri). J. Comp. Neurol.
158.219-236.
Tootell, R.B.H., and S.L. Hamilton (1989) Functional anatomy of the second
visual area (V2)in the macaque. J. Neurosci. 92620-2644.
Tootell, R.B.H., and M.S. Silverman (1985) Two methods for flat mounting
cortical tissue. J. Neurosci. Meth. 15:177-190.
Tootell, R.B.H., E. Switkes, M.S. Silverman, and S.L. Hamilton (1988)
Functional anatomy of macaque striate cortex. I1 retinotopic organization. J. Neurosci. 8:1569-1609.
Ts'o, D.Y., and C.D. Gilbert (1988) The organization of chromatic and spatial
interactions in the primate striate cortex. J. Neurosci. 8:1712-1727.
Ts'o, D.Y., C.D. Gilbert, and T.N. Wiesel (1986) Relationships between
horizontal interactions and functional architecture in cat striate cortex
as revealed by cross-correlation analysis. J. Neurosci. 6:1160-1170.
Ungerleider, L.G., and M. Mishkin (1982) Two cortical visual systems. In
Y. AMIR ET AL.
D.J. Ingle, M.A. Goodale, and R.J.W. Mansfield (eds): Analysis ofvisual
Behavior. Cambridge Mass: MIT Press, pp. 549-586.
van der Heidt, R., and E. Peterhans (1989) Mechanisms of contour perception in monkey visual cortex. I. Lines of pattern discontinuity. J.
Neurosci. 9: 1731-1748.
Van Essen, D.C., and J.H.R. Maunsell(1983) Hierarchical organization and
functional streams in the visual cortex. Trends Neurosci. 6:370-375.
Van Essen, D.C., W.T. Newsome, and J.H.R. Maunsell (1984) The visual
field representation in striate cortex of the macaque monkey: Asymmetries, anisotropies and individual variability. Vision Res. 24:429-448.
Van Essen, D.C. (1985) Functional organization of primate visual cortex. In
A. Peters and E.G. Jones (eds): Cerebral Cortex: Vol. 3, Visual Cortex.
New York: Plenum Press, pp. 259-330.
Van Essen, D.C., D.J. Felleman, E.A. De Yoe, J. Olavarria, and J. Knierim
(1990) Modular and hierarchical organization of extrastriate visual
cortex in the macaque monkey. Cold Spring Harbor Symp. Quanta. Biol.
55:679-696.
Wong-Riley, M. (1979) Columnar cortico-cortical interconnections within
the visual system of the squirrel and macaque monkeys. Brain Res.
162:201-217.
Wong-Riley, M.T.T., and E.W. Carroll (1984) Quantitative light and electron
microscopic analysis of cytochrome oxidase-rich zones in VII prestriate
cortex of the squirrel monkey. J. Comp. Neurol. 222:18-37.
Yoshioka, T., J.B. Levitt, and J.S. Lund (1992) Intrinsic lattice connections
of macaque monkey visual cortical area V4. J. Neurosci. 122785-2802.
Zeki, S., and S. Shipp (1989) Modular connections between areas V2 and V4
of macaque monkey visual cortex. Eur. J. Neurosci. 1:494-510.
Note added in proof: Followng acceptance of this manuscript, a related
study has been published (Lund et al., '931, which confirms and extends our
suggestion (Malach '92 and the present paper) that the size of axonal patches
is related to the size of dendritic arbors of upper layer pyramidal neurons.
These findings lend further support to the notion that this size relationship
is a fundamental principle of cortical circuitry.