| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Center for Molecular and Behavioral Neuroscience, Rutgers University, Newark, New Jersey 07102
Submitted 31 December 2002; accepted in final form 18 March 2003
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
;
Head and Holmes 1911). Modern
electrophysiological measurements revealed the subdivisions and defined the
neuronal properties of the inferior parietal lobule available to construct
representations of surrounding visual space
(Andersen et al. 1985;
Heilman et al. 1993;
Siegel and Read 1997a). The
inferior parietal lobules of the macaque monkey cortices have two visual
association areas that lie on the cortical gyrii, area 7a and the dorsal
prelunate (DP) area (Siegel and Read
1997b). The receptive fields of electrically recorded single
neurons in monkeys for these regions often approach 60° in size, can be
bilateral, and, at least those of area 7a, are selective to navigational optic
flow (Motter and Mountcastle
1981; Read and Siegel
1997; Siegel and Read
1997a). The gain of the visual responses of inferior parietal
lobule neurons is modulated by the position of the eye in the orbit and the
monkey's behavioral state (Andersen et al.
1985; Bushnell et al.
1981; Read and Siegel
1997). There has been no evidence from any measurements of single
cells for a mapping of these properties across the inferior parietal lobule's
surface in the behaving monkey (Andersen et
al. 1990; Blatt et al.
1990).
Anatomical projections between the inferior parietal lobule and the frontal
and temporal lobes suggest that there may be topographies. The projections are
patterned regions of interdigitated columns and regions of overlap
(Andersen et al. 1990;
Cavada and Goldman-Rakic 1989;
Lewis and Van Essen 2000).
When retrograde tracers are injected in two projective areas (e.g., area 8 and
46), stripes of overlapping cell bodies that can diverge are found in area 7a
(Andersen et al. 1990). Such
projection patterns elsewhere [e.g., between V1 and V2
(Ts'o et al. 2001)] have been
correlated with functional architectures and could indicate the presence of
similar organizing principles in the inferior parietal lobule. Given the
relatively small surface area of the cortical regions in the inferior parietal
lobule and the large receptive and gain fields, the paucity of published
electrophysiological mapping data might simply indicate that the orbital gain
fields overlap substantially across the surface and have no topography.
Alternatively the single-unit methodology may be technically unable to unveil
a functional architecture in chronic behaving monkey studies because there are
substantial errors in the localization of electrode penetrations over the 1 or
2 yr needed for recording (Andersen et al.
1990; Siegel and Read
1997a). Another possibility is the relationship of gaze direction
to cortical topography may be dynamic in ways that require large areas to be
examined simultaneously (or nearly so) for these properties. The absence of
explicit knowledge for an inferior parietal lobule functional architecture has
substantially hindered an exploration of how the underlying circuitry can
compute a neuronal correlate for spatial cognition.
Optical imaging utilizes light to assess the oxygenation of hemoglobin (Hb)
and thus indirectly measure neuronal metabolism and activity. This technology
permits multiple measurements over an extended period of time and space in the
behaving monkey (Shtoyerman et al.
2000) and allows for a direct assessment of maps in the inferior
parietal lobule. In the current study, intrinsic optical imaging has revealed
a novel map of eye position modulating visual responses in the inferior
parietal lobule. This architecture is discussed in terms of constraints on
subsequent spatial perceptual and motor processing.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
Two monkeys were prepared for chronic behavioral studies using standard
methods (Siegel and Read
1997a). The use of the artificial dura permits long-term studies
and followed published methods (Shtoyerman
et al. 2000) with modifications as described here. During the
implant surgery performed under sterile conditions and isoflurane
(0.5–2% in O2) anesthesia, the animal was given ceftriaxone
sodium antibiotic (Rocephin, Roche, 100–150 mg ·
kg–1 · d–1 im),
mannitol (25% 1 ml/kg iv), and furosimide (1 mg/kg im) prior to opening the
dura. The latter two minimized cerebral edema. The artificial dura consisted
of a thin 50 µm silicon sheet with an embedded silicon ring
(Shtoyerman et al. 2000); it
was inserted after resecting the biological dura in an "X" shape
within the stainless steel recording chamber. The flaps of the dura were glued
to the chamber edge and the 25 mm diameter artificial dura was inserted
between the real dura and the cortex. A silicon ring (18-mm diam) prevented
movement of the artificial dura. The chamber was rinsed with body temperature
saline and sealed. Antibiotics were continued for 7–10 days; analgesics
(buprenorphorphine; 2–6 µg/kg im) were given for 
3 days.
Typically the granulation tissue in the chamber sealed against the edge of the
silicon ring providing a watertight seal within 3 days of the surgery.
Monkey 1 had recordings made from its right hemisphere from March 1999 to July 2002 and is referred to as M1R; monkey 2 had recordings from its right hemisphere from June to August 2001 and is referred to as M2R. A second chamber was implanted overlying the left hemisphere in M2 in September of 2002; recordings collected two months after the implant are described and are referred to as M2L.
The full angle-of-gaze study (see following text) for M1R was mainly collected over the first 78 days after the implant; controls and other experiments were collected subsequently. M2R had a subdural bleed (4 x 4 mm) that obstructed the cortex 2 wk after the implant that prevented electrical recording and extensive optical recording. After the bleed cleared, additional optical recordings were made for 6 wk until granulation tissue under the artificial dura obscured the cortex. On removal, the artificial dura was found to have a small tear, which probably initiated the bleeding. M1R and M2R chambers continue to be studied in additional experiments as of December 2002.
All procedures were approved by the Rutgers University Animal Institutional Review Board and were in accordance with the National Institutes of Health Guidelines on the Care and Use of Animals in Research.
Behavior
The monkey pulled back a key within an 800 ms time window of the fixation
point onset. Two seconds after the fixation point onset, the dot stimulus
would start (Fig. 1A).
During 4,000–6,000 ms after fixation onset, the stimulus would change
its structure (Fig.
1B) and the monkey had to release the key within a 150 to
800 ms reaction time window for 0.1 to 0.2 ml juice reward. Breaking of eye
fixation (>1° deviation) terminated the trial with no reward
(Siegel and Read 1997a). In
most experiments, the fixation point was placed in one of nine positions in a
3 x 3 grid, 20° on a size, and the expansion optic flow field was
placed over the fixation point (Fig.
1C). Optic flow is known to modulate the firing rate of
neurons in area 7a (Siegel and Read
1997a). As the receptive fields in area 7a are 20–40° in
size (Andersen et al. 1985;
Read and Siegel 1997), 20°
diameter flow stimuli were used.
|
In some experiments, another stimulus set was utilized to examine upper and
lower gaze field tuning. Two fixation positions [e.g., (0, 10°) and
(0,–10°)] were used; over the fixation, one of two different optic
flows (expansion, compression, clockwise and counterclockwise) was presented
in each trial. The fixation conditions for which expansion and compression
optic flow was presented were analyzed as part of the current study; the
remaining data serve as the basis for a study of optic flow
(Raffi and Siegel 2002; M.
Raffi and R. M. Siegel, unpublished data).
Optical imaging technique
The monkey's head was firmly attached to a floating Newport air table via
an implant made of Palacos R radiopaque bone cement (No. 12-0001, Smith+Nephew
Richards, Memphis, TN) over the skull held with 
20 Synthes (Paoli, PA)
titanium screws. This implant was made in a recovery surgery 1–6 mo
prior to the artificial dura implant. The implant covered the skull from the
occipital notch to the frontal bone and laterally replaced the insertion
points of the temporalis muscles. Embedded in the cement was a custom
stainless steel t-bar fixture with a 6.35 x 50 x 30 mm hardened
steel plate in the frontal plane. This combination provided exceptional
rigidity. The camera was also attached to the Newport table using
off-the-shelf components.
Intrinsic optical imaging methods were used to study the cortical
topography (Shtoyerman et al.
2000). The macroscope, somewhat based on optical principles of
Ratzlaff and Grinvald (1991),
consisted of a Nikon Nikkor AF Micro 60 mm/2.8 D lens and a 50 mm Nikon 1.2
lens (No. 385083) as the objective. Unlike the Ratzlaff/Grinvald macroscope
where the matched lenses are focused at infinity, the 60 mm Nikkor Micro lens
focused on the inverted image from the 50 mm objective lens. Adjusting the
focal plane of the 60 mm Nikkor lens permitted variation of the magnification
as well as an unusually long, 30–80 mm, working distance while
maintaining a narrow depth of field. Images were taken from two monkeys who
had 20 mm diameter chambers implanted over a trephination in the skull (as
described in the preceding text), based on magnetic resonance images
(Fig. 2, A and
B). The chamber was filled with 0.9% saline and
hydraulically sealed with a glass plate for optical imaging.
|
Typically 750 x 480 pixel images were collected at 605 nm with 17.3 µm/pixel resolution at a depth of 500 µm below surface capillaries (imaged with green light). These were resampled to provide a 34.6 µm/pixel resolution. The data were not spatially or temporally filtered other than the reduction of the spatial resolution by a factor of two to avoid inducing spatial distortions or filtering artifacts. Major veins and arteries could be distinguished based on the presence of pulsations.
Image analysis
The Optical Imaging 2001 system (Rehovot, Israel) was used to collect
optical data. Data collection was initiated by first collecting a reference
image in the interval between trials while the monkey was not in the task.
This reference image served to set the amplifiers' gains and offsets. Two
hundred and fifty six frames at 30 Hz were averaged; a reference image was
collected every 16 trials (160 s). This reference image was stored in
memory and subtracted from incoming images in real-time by the optical imaging
system. This difference image was digitized by the imaging system and stored
on disk. Off-line, the reference image and difference image were combined to
provide measurements of total reflectance with 16 bits precision. Optical
images were collected for every trial. At the same time, a behavioral control
computer kept records of the animal's performance
(Siegel and Read 1997a) and
was synchronized with the optical-imaging system via a set of digital lines.
An IBM SP2 computer and an imaging package (Khoral Research, Albuquerque, NM)
were used for subsequent analysis and display. All trials for which the monkey
incorrectly performed the trial (e.g., eye movement, incorrect lever movement)
were rejected from the data set. A typical run would result in 30–90
trials per condition or 270–1,200 trials each day.
BASELINE NORMALIZATION ANALYSIS (BNA). A regression analysis was
utilized. It normalized each trial's data by a baseline value collected at the
start of the trial. The evoked reflectance signal was quantified by
subtracting the "baseline" image (averaged for –1,000 to 0
ms relative to stimulus onset) from the signal averaged over the
2,000–3,000 ms after stimulus onset. At this time the monkey was
fixating an initial red target and holding back the manipulandum. The
resulting difference image for the ith presentation was expressed as
a percentage change from the "baseline." Thus the percentage
change in reflectance was
 | (1) |
(N is the number of frames in the interval (2000,3000) and similarly
for the mean baseline response 
. Four hundred to
1,200 images corresponding to all behaviorally correct trials were collected
per experiment.
Data from some trials needed to be rejected as outliers, either from
excessive movement of the monkey's torso, which could move the brain slightly,
or from an error in the data collection system. Failure to perform this
rejection could result in a topography that would be dominated by the
gargantuan signal from one aberrant trial. This rejection was performed
off-line by an automated algorithm. A mask was superimposed on each of these
images solely to perform off-line automated trial rejection
(Fig. 3). The mask served to
exclude large blood vessels and dimly illuminated cortex from the rejection
procedures. The masked image was computed with the following heuristic. The
mean reference image (Fig.
3A) on a pixel-by-pixel basis of the 26–52
reference images were computed. From this average image, the mean ± SD
of all its pixel values was computed. The pixels of the image were thresholded
to 1 if they were one-half of a SD greater than the mean to form the mask and
to 0 otherwise (Fig.
3B). This mask excluded the larger blood vessels. This
binary mask was then multiplied on a pixel-by-pixel basis with the individual
difference images from each trial and the mean ± SD of this masked
collection of pixels was thus computed. The distribution of the means of the
masked regions followed a reasonable approximation to a normal distribution
(Fig. 3D), and only
trials that fell within one SD of the mean were further analyzed. A plot of
the mean versus the SD yielded a parabolic-like curve, which was further
utilized for automated rejection (Fig.
3D). Points that had SDs within 0.1% of the value 0 were
rejected; such trials arose from an error in the data collection software and
were <1% of the total trials. This parabolic relationship is expected from
a normal distribution of the pixels' values within each image and could be
exploited in the future for additional higher order noise based analysis. In
short, trials for which the mean evoked signal was >1 SD from the mean of
all evoked signals were rejected to remove outliers. These varied between 10
and 20% of the behaviorally correct trials. This automated approach differs
from earlier studies for which trial rejection was performed manually
(Grinvald et al. 1991) or not
at all (Vnek et al. 1999).
|
The mean image for each stimulus condition was computed resulting in nine
average images per experiment corresponding to the nine fixation points.
Parameter maps were constructed using standard linear regression methods (PROC
GLM, SAS, Durham, NC) on individual pixel values
(Fig. 4). The nine average
images in units of percentage change in luminance (units of %) were used. For
every pixel, the equation
 | (2) |
x(I,J) and
y(I,J) are the slopes of the regression
for each pixel (%/°), 
(I,J) is the intercept for each pixel
(%), 
i(I,J) is the error values, and
Ex and Ey are the fixation point (and
stimulus center) position. This equation defines a plane with a maximum slope
of
at an angle of 
= arctan
(y/x) relative to the
x axis [indices (I,J) omitted here for clarity]. As the optical
signal is the negative of the expected rate of neuronal firing
(Shtoyerman et al. 2000), the
angle maps were constructed by multiplying each slope parameter by –1
prior to computing the quadrant for each arctan. The same parameter maps were
obtained whether the average single condition maps were used or all the
individual trials were used in the regression, presumably because the error is
approximately a normal distribution about the mean. Generally the average
single-condition images were used to reduce computational requirements.
Regions of interest (ROIs) were chosen for simple computations of means and
SDs as described in RESULTS. A Monte Carlo analysis was used to
establish the error for this entire analysis. In summary, the BNA was devised
to examine the changes in the optical signal from a defined baseline and is
analogous to electrophysiological studies where the changes in firing rate
relative to baseline are assessed (Read
and Siegel 1997).
|
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
), the cortex was imaged at a wavelength of 605 nm with a
modified macroscope (Ratzlaff and Grinvald
1991) at a depth of 500 µm.
|
Prior single-unit studies have established that both area 7a and DP neurons
have "gain fields" (Andersen et al.
1985,
1990;
Colby and Goldberg 1999;
Read and Siegel 1997). The
concept of a gain field means that the amplitude of a response to a visual
stimulus can be increased or decreased by the position of the eye in the
orbit. To determine if there was a cortical topography of the gain field, two
monkeys performed the motion detection task with the fixation point placed in
one of nine locations in a 20 x 20° grid
(Fig. 1C) while a 13
x 8 mm region of cortex was imaged. An expansion navigational optic flow
stimulus was presented 2,000 ms after fixation point onset and was always
centered over the fixation point.
Time course of optical signal
In the inferior parietal lobules of the monkeys performing the task, the
time course of the optical signal differed from that typically reported in
primary visual cortex in behaving animals. In primary visual cortex studies,
the initial event triggering the alteration in blood flow that underlies the
optical signal is the actual visual mapping stimulus
(Shtoyerman et al. 2000). In
the reaction task used here, there were multiple retinal and extra-retinal
events that could alter the neural activity of inferior parietal lobe and
hence the hemoglobin and optical signal. The initial relevant event in the
recording sequence is the onset of the fixation point closely followed
(<500 ms) by the saccadic eye movement to the fixation point and the hand
pulling the key. Both area 7a and DP neurons are sensitive to eye position,
fixation point onsets, and the planning of motor activity (Andersen et al.
1985,
1987,
1990;
Siegel and Read 1997a), so it
was not unexpected that these three events taken together were correlated with
changes in the optical signal measured at 605 nm
(Fig. 6). The ROI in the two
images is illustrated in the line drawing above each time course. Temporal
signals were computed by spatially averaging an 2 x 2 mm square region
of cortex but not averaging in time. Often, but not always, there was a
negative dip in the optical signal followed by a positive overshoot (e.g.,
dark thick line of Fig.
6A). The timing and amplitude of the initial changes over
the first 1,000 ms of the trial was variable across experiments reflecting the
uncontrolled behavioral state prior to fixation (cf.
Fig. 6, A and
B, for M1R and M2L,
respectively.) Variation was found both within animals and within cortical
areas, hence the differences in Fig. 6,
A and B, were not simply a result of
the cortical region or animal studied. The baseline period from –1,000
to 0 ms before the stimulus was used to normalize the optical signal, as
complete behavioral control was obtained just before this interval and the
time course was reasonably similar.
|
Based on gain field single-unit work
(Siegel and Read 1997a), the
optical signal should modulate as a function of the position of the eye in the
orbit and the visual stimulus. To evaluate the optical correlate of the gain
field effect, expansion or compression flow stimuli were presented in a 2
x 2 factorial design with either up or down fixations in M1R
and M2L. The flow stimulus began 2 s after the fixation point and was
centered over that location (Fig.
6). Although there is variability in the time course prior to the
(-1000,0) interval, at that time, the time courses converge indicating that
the optical signal is similar across the different fixations. For the area DP
region of Fig.
6A, at 1,000 ms after stimulus onset, the
optical time course depended on the type of optic flow for upward fixation
(cf. the heavy and thin black lines). The differential response to the optic
flows was also found for downward fixation (cf. heavy and thin gray lines.)
For the 7a region of Fig.
6B, the dependence of the optical signal on the
type of optic flow is best seen for the downward (gray lines) fixation. For
this particular patch of cortex, there is a weak dependence of the signal on
the visual stimulus.
Across experiments, maximal differences were seen in the 2,000 to 3,000 ms interval following the onset of the visual stimulus. Hence the 2,000 to 3,000 ms interval after flow onset was used as a measure of the underlying visually evoked neural activity; optical measurements were expressed as the percentage change from the baseline signal and were the basis of the baseline normalization analysis.
This type of experiment was repeated over 10 times in M1R and 6
times in M2L. The reflectance signal depended on the expansion and
compression stimulus as well as eye position. The modulation depended on the
position on the cortex. Indeed this optical tuning in area DP is the first
evidence for optical flow selectivity in DP. These results suggest a mapping
of optical flow as well as gain fields across the inferior parietal lobe,
which serves as the basis of another study under preparation
(Raffi and Siegel 2002). The
remainder of the current report only examines the effect of eye position on
the expansion optic-flow-evoked response.
Cortical topography of optical signal
To explore the dependence of the reflectance on eye position, only
expansion optic flow was used. The monkeys performed the fraction of structure
detection task for nine different fixations on a 20 x 20° grid
(Fig. 1C). In each
case, the expansion stimulus was centered over the fixation point. Two
experiments in M1R are presented in Figs.
2/4
and Fig. 7. For M1R,
averaging across all fixation conditions, the modulation in the reflected
light over the baseline was 0.5% and varied across the cortex (Figs.
2C and
7B). There was a
smaller (0.1%) amplitude light evoked response that depended on the
monkey's eye position (Figs.
2D and
7F). The reflected
light varied both as a function of the particular eye position and the
particular location on the cortex, suggesting there was a cortical topography
for the gain field. The two experiments performed 1 day apart have similar
effects in the single condition maps (cf. Figs.
2 and
7). Thus when the fixation and
the stimulus were at the upper right on the screen (10°,10°), a bright
signal was found at the right portion of area 7a. When the fixation and
stimulus were at lower left on the screen (–10°,–10°), a
bright signal was found in the right portion of DP (Figs.
2D and
7F). There was a clear
border between 7a and DP at the blood vessels that runs across the middle of
the images. By low-power microscopic examination, it was determined that the
superior temporal sulcus did not extend that far dorsally, so that the border
ran under the large vessel but across a flat cortical surface. Thus in both
experiments there appears to be a discontinuity in the representation
underneath this blood vessel.
To combine the data from these nine maps, the mean optical signal at every
pixel was linearly regressed on the orbital position using the baseline
normalization analysis (METHODS). Maps of the regression parameters
were constructed. The intercept parameter map (Figs.
4A and
7C) is the change in
measured light expected when the monkey was fixating straight ahead and the
stimulus was over the fixation point. The map of the vertical slopes
(y) of the regression (Figs.
4C and
7E for M1R)
illustrate how the optical signal depended on the vertical fixation position.
DP had predominantly negative values for the vertical regression coefficient,
whereas 7a had positive values. This means that fixation in the upper visual
field leads to smaller (i.e., more negative) optical signals in DP. Similarly,
the positive vertical coefficient values in area 7a indicate a maximal
reflected light response for upper field fixations. There is also a horizontal
eye position dependence of the reflected light (x) which can
best be seen in DP (Figs.
4B and
7D, which depict the
horizontal slope, x). For example in
Fig. 4B, most
of area 7a and DP have similar horizontal tuning except for the most lateral
part of DP, which is darker.
The horizontal and vertical coefficient parameter maps were transformed
from rectangular to polar coordinates (METHODS). In polar
coordinates, each pixel was represented by a polar vector with an amplitude
(Fig. 4D,
data not shown for Fig. 7) and
angle (Fig. 5,
A–C). The "amplitude map" was
constant across the imaged regions outside of the blood vessels and suggests
that the optical signal strength is constant across the imaged region. To
compute an "angle map" that reflected the underlying neuronal
electrical activity, the sign of the slopes was changed to account for the
negative relationship between the 605 nm signal and the neuronal signal
(Shtoyerman et al. 2000); the
angle map illustrates the gain fields' dependence on eye position and cortical
location. The angle map had two clearly demarcated gain fields within the 13
x 8 mm image. ROIs are indicated for each map; the angle corresponding
to that ROI is indicated in small type above and below each image. According
to the angle map, neurons in DP should increase firing rate when a visual
stimulus was presented over the fovea while the monkey was looking
up; similarly neurons of 7a should be best activated when looking
down and to the left. This 7a/downward and DP/upward split of the imaged
regions was modulated within each region by the horizontal eye position. As
one progresses counterclockwise from ventral DP, the direction of gain field
tuning progresses clockwise. As the strength of the horizontal modulation was
variable between the two experiments, the smoothness and completeness of the
representation varies. A more lateral image of the gain field map taken from a
third experiment in M1R is illustrated in
Fig. 4F; the
gain field extends to the most posterior portions of DP that can be imaged; in
this one lateral view, there appears to be another shift in the gain field in
the more lateral portions of DP and 7a. In each of the three examples of the
gain field map of M1R (Fig. 5),
upper and lower field gain breakdown is seen. The horizontal tuning is weaker
and more variable between these different experiments.
|
Comparison of maps across days
This main result of a division in the gain field between area 7a and DP was
reproduced within M1R by collecting 14 maps over a period of 78 days.
Two ROIs, one in area 7a and one in area DP were initially chosen for analysis
(Fig. 8). The ROIs were placed
so as to be observed in as many day's data as possible (for comparisons
between days) and to be roughly the same distance from the junction of the
intra-parietal and lunate sulci as well as equidistant from the large vessel
in the middle of the chamber. This was to make the effect of vessel induced
pulsations equivalent for the two locations. The medial-lateral position was
arbitrarily selected. A ROI was chosen to be 2 x 2-mm square. The
circular mean and standard error of the gain field for the ROIs was 283
± 11° for area 7a, and 99 ± 8° for area DP. The
locations and means for six additional ROIs were computed as summarized in
Fig. 8 to sample across the
cortex in an unbiased manner. Using the total of eight ROIs, there are a few
key observations. The area 7a and DP regions clearly had differences in terms
of upper and lower field representation at all positions. There appeared to be
a slight trend for further modulation along the horizontal direction within
each cortical field which might have been obscured by the averaging across
days.
|
The measurements for the ROIs in area 7a and DP were compared for the most lateral position with a Watson's F test for circular means and were significantly different (P < 0.01, DF = 11). The mean difference between the 7a and DP tuning on a day by day basis was 190 ± 17° and was significantly different from a uniform circular distribution. Thus DP and 7a have significantly different gain field representations for these two ROIs, with DP expected to have stronger electrophysiological responses for upper fixations and area 7a having a stronger expected gain field representation for lower field fixations. Additional paired comparisons were made for each matched medial-lateral positions and each was found to be different for the upper and lower positions.
Monte Carlo analysis
To get an independent estimate of the reproducibility of the measurements and analysis within a day, Monte Carlo methods were used (METHODS). The data from some of the full nine position gain field experiments were tested. The core idea behind the Monte Carlo study was to test the hypothesis that the maps were specifically linked to the stimulus conditions. More formally, the null hypothesis was that there was no relationship between the stimulus presentation and the resulting maps.
To test this hypothesis, two manipulations were compared. In the first manipulation, half of the data were randomly selected from the original data and the map computed. The other half of the data were used for another map. The second manipulation was to randomly assign a stimulus condition to each trial for half the data and compute the map; the remaining data were used to construct another map continuing this randomization. (Formally, one would say that the data were randomly assigned to the stimulus conditions without replacement.) Thus the specific relationship between the actual data and the stimulus used to collect it was destroyed. Maps were then constructed for the resulting "new" data set (Fig. 9).
|
Each pass through the data set then provided four new maps; two were made respecting the relationship between the data and the stimuli and two with disrupted relationships. Two populations of parameter maps were constructed in this way, and the resulting distributions were compared. The direction parameter map is reproduced from Fig. 5 as Fig. 9B. Figure 9, A and C, shows some examples of the bootstrap parameter maps respecting and disrupting the relationships between the stimuli and data respectively. If the null hypothesis of no fixed relationship between the measured data and the stimulus condition was true, then the two populations would be the same. If there was indeed a relationship between the recorded data and the stimulus conditions, then the populations would be different. The distributions of the directional tuning in the ROIs are in Fig. 9D. The distribution of the gain fields from the ROIs when randomization was not performed show a downward effect for 7a and upward effect for DP. The directional distribution is essentially uniform for the ROI for which the randomization between the stimulus and signal was performed.
For the data of Fig. 2, this procedure was repeated for 115 surrogate sets yielding 230 new maps. Additional ROIs could have been chosen; however, the Monte Carlo analysis was too computationally intensive to permit this. Maps were generated and the same ROI were sampled. Circular means and errors were computed. A comparison in the two distributions was computed using a using a circular Watson F-test analysis. (Oriana Software, Kovach Computing Services, Anglesey, Wales).
For the data of Fig. 2, the bootstrap circular mean and standard error was 132 + 2.2° for the area 7a ROI and was 42 + 1.8° for the area DP ROI. For comparison, if the optical measurements were randomly assigned to a stimulus, the circular standard error had a high value of 76 and 70° for DP and 7a, respectively. The distribution of directions was essentially flat for the randomized data. These random versus the bootstrap distributions were significantly different (P < 0.0001). Similar results were found for two other days tested a few weeks apart showing that within a single day, the maps had within day errors of <5°.
Replication in a second monkey
GAIN FIELD MAPS. The gain field map was replicated in two hemispheres of a second animal using the BNA (the data of M2R and M2L). The quality and number of the maps in M2R was few and of poorer quality than the other maps due to the subdural bleed. Nonetheless the primary result of an upper-lower gain field separation between 7a and DP could be confirmed (Fig. 10A for M2R). The reds and yellow of the gain fields in area 7a indicate lower field effects while the blues and purples in DP indicate the upper gain field. As before, the numbers at the edge of the images represent the average for the ROIs, which appropriately correspond to the upper and lower gain fields. In M2L, the chamber placement was more lateral and the image is at a higher magnification. Hence the union of the IPS and LS are beyond the left of the panel (Fig. 10B). Gain field maps were again obtained with the upper and lower gain field divisions. The periodicity seen in M2R (Fig. 10A) is perhaps reminiscent of ocular dominance periodicity in primary visual cortex. However, the present study does not provide any source for this structure and it was only seen in this one map.
|
ROI ANALYSIS. The circular mean and SE data for ROIs of M2R in 7a and DP were 240 ± 23 and 75 ± 8°, n = 4, respectively; these two ROIs were significantly different (P < 0.01, DF = 6, Watson's F test). In M2L, the ROIs were picked to be similar to that of M1R with the means for the ROIs being 310 ± 26 and 94 ± 19°, n = 5, for 7a and DP, respectively; the ROIs were significant at P < 0.01, DF = 8, Watson's F test. The upper and lower breakdown between area 7a and DP was found in all these data to agree for all three hemispheres.
MONTE CARLO ANALYSIS. The reproducibility of the maps within a
day using the Monte Carlo analysis in M2R was similar to that of
M1R (circular SE of 4°.) A Monte Carlo analysis was also
performed in M2L with an approximately doubled error. Thus we
estimate that in three hemispheres the error of our measurement within a 2
x 2-mm region is 4°.
In summary, our optical recordings have shown area 7a and area DP have gain field tuning. At least for the two regions imaged, there appears to be a division of labor with 7a representing the lower gain field and DP representing the upper gain field. There is also modulation with the horizontal eye position.
Electrophysiological confirmation of eye position maps
Typically, for optical imaging experiments, single-unit recordings are made
to "verify" the optical responses correspond to classically
defined electrophysiological responses
(Shmuel and Grinvald 1996;
Shtoyerman et al. 2000). This
approach has worked very well in V1 where there is a columnar architecture for
orientation and ocular dominance. The optical signal only indicates reflected
light from the upper layers, and it is reasonable to assume that the bulk of
the changes in reflected light arise from metabolic changes in the smallest
processes with the largest surface-area to volume ratio
(Frostig et al. 1990;
Malonek and Grinvald 1996;
Malonek et al. 1997).
Low-impedance electrodes were introduced through the artificial dura and
measurements of gain fields were made. Recordings were only made in one
animal, M1R; the number of penetrations with the artificial dura was
minimized because the electrodes caused a small pinhole in the artificial
dura. In our hands, the artificial dura often self-sealed, but at times, air
could seep in under the artificial dura. This lead to concerns about subdural
infections compromising the continuation of the studies; although this never
occurred. As well, even though the size of the bubbles were small (being
1 mm in diameter), they precluded any optical recording from 2 day to 1
wk while they were being reabsorbed.
The multiunit response to the alteration of eye position
(Fig. 11) appeared similar to
the responses obtained from single-unit recordings as reported elsewhere
(Read and Siegel 1997). Gain
fields in area 7a may be linear or humped
(Read and Siegel 1997), and so
a quadratic stepwise regression model was used; the stepwise selection only
permits coefficients significant different from 0 at P < 0.05 to
remain in the model (Read and Siegel
1997). Both the peristimulus time histograms as well as the
regression surfaces are illustrated (Fig.
11). After the approach from single-unit studies on optic flow in
the inferior parietal lobule (Phinney and
Siegel 2000; Read and Siegel
1997), baseline activity was evaluated for 1 s prior to stimulus
onset. The response of the multiunit activity was evaluated over 1 s
immediately after stimulus onset to combine phasic and tonic responses. To
reduce variability caused by any instability of the recordings, the baseline
rate was subtracted from the evoked response on a trial-by-trial basis.
|
Sample recordings from area 7a (Fig.
11, A and B) and DP
(Fig. 11, C and
D) illustrate the type of tuning found electrically. The
gain field tuning of these cells was similar to that reported elsewhere with
these stimuli (Read and Siegel
1997).
Cells were recorded from a 5 mm strip in DP (9 penetrations) and a 1.25 mm
strip in 7a (5 penetrations) in hemisphere M1R
(Fig. 12A). The
region from which the area 7a recordings are taken is indicated by 
,
while the region for which the DP recordings are taken is indicated by 
.
Twelve of 54 area 7a multiunit recordings (22%), and 14/56 (25%) of area DP
multiunit recordings were found to significantly depend on the position of the
eye in the orbit. In 7a, half of the neurons had significant linear
coefficients, while the other six had only significant quadratic components.
In DP, 11 of the cells only had a significant linear component; 3 were purely
quadratic and 2 had both a linear and quadratic components.
|
For the purely linear cells, the direction of the gain fields could be summarized by the vertical and horizontal slopes; for the purely quadratic cells, the gain field was symmetric in the vertical and horizontal and the gain field was assigned a coordinate of (0,0°). For the neurons with both a linear and quadratic component, the gain field was appropriately corrected to place the peak in the appropriate quadrant (Fig. 12B). The 7a recordings sites were mostly in the lower contralateral lower field while the DP recordings were mostly in the ipsilateral upper field.
As a population, the area 7a and DP neurons were statistically different.
The significance between the DP and the 7a population regressions were
computed three different ways. First, a Mann-Whitney test was used to
determine if each of the regression coefficients were different across areas;
all effects were significant at P < 0.05, except for the
second-order "y" coefficient (yy). Second, a
canonical discriminant analysis (SAS PROC CANDIS, Durham, NC), which considers
a linear combination of all five regression coefficients, was used. The sum
0.08+ 16.3x+ 4.16 y – 17.4
xx+ 1.44 yy was significantly different
for the two populations at P < 0.001. Last, to determine the eye
position for which the activity was maximal, the slope coefficients from each
recording were converted into polar coordinates. The electrophysiological
circular mean and standard error was 231 ± 30° (n = 5) and
27 ± 30° (n = 10), for 7a and DP, respectively. The
directional tunings for the two areas were significantly different (P
< .05; DF = 3, Watson's F test). Thus the directional tuning from
the electrical recordings was able to discriminate between area 7a and DP, a
novel finding not yet reported from electrophysiological recordings. The lack
of any report of this in earlier studies may be because of the poor spatial
electrode localization in earlier work obscured such effects.
Figure 12 illustrates both
the average signal for the optical ROI and for the electrophysiological data.
In comparison to the optical data, the multiunit data replicated the upper and
lower gain field divisions between 7a and DP; the ipsilateral-contralateral
distinctions did not match that of the optical data. This may be due to the
disparity in location of the optical and electrical recording sites or
differences in the source of the electrical and optical signals
(Logothetis et al. 2001).
Ideally, simultaneous, multisite, tangential electrode penetrations should be
made within area 7a and DP to precisely match the two types of data.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
Relationship between optical signal and underlying activity
Key to the demonstration of these maps was the utilization of intrinsic
optical imaging at 605 nm. The signal at this wavelength scales with the level
of reduced-Hb (Frostig et al.
1990; Malonek and Grinvald
1996; Malonek et al.
1997). The optical signals mostly mirror the metabolism of the
small neuronal elements such as presynaptic fibers, boutons and dendrites
(Logothetis et al. 2001).
Those elements have the smallest diameters and hence the largest
surface-area-to-volume and maximal contribution to the oxidative metabolism.
Hence the intrinsic optical signals are most similar to local field potentials
(Logothetis et al. 2001). The
question is how much of this signal is above threshold and will be impressed
onto the spiking neuronal activity. There is a substantial literature in
primary visual cortex that supports the assertion that intrinsic signals
ultimately reflect outputs from cells (i.e., spiking). The principles
underlying the source of the optical signal derived from striate studies
should also be applicable to the current studies in inferior parietal
lobe.
Our electrical measurements indeed confirm the optical results. The upper
and lower gaze field division of representation between area 7a and DP are
found with both methodologies. The optical parameter maps in one chamber were
confirmed using multiunit electrophysiological recordings. Multiunit
recordings were made as the tip impedance needed to be low (300 k
)
to permit piercing of the artificial dura. The percentage of selective cells
was lower as compared earlier studies
(Andersen et al. 1990;
Read and Siegel 1997;
Siegel and Read 1997a),
perhaps because the stimuli parameters (e.g., retinal stimulus location, type
of optic flow) used in the present study were not optimized for each recording
site. A second reason for the lower number of significantly tuned recordings
between the two studies is that multiunit data can sum the contributions of
single cells with disparate tunings resulting in a broadened and
nonsignificant response. Despite these technical limitations, there was a
reasonable match between the multiunit cell physiology and the optical
maps.
Confirmation of finer details of the maps, such as the horizontal modulation of the gain field, were not made in that it would have required repeated penetrations and damage to the artificial dura. Given that the optical methods provide a detailed map and the substantial body of evidence supporting the neural underpinnings of these maps, we were restricted and conservative in the electrical recordings; mostly serving to verify the signal in two ROIs.
Time course of the optical signal
The time course of the signal in this association cortical region shows
novel properties as compared with earlier measures of the striate cortex. In
our studies, the initiation of the task, known from electrophysiological data
to activate parietal neurons (Andersen et
al. 1990; Motter and
Mountcastle 1981; Read and
Siegel 1997; Siegel and Read
1997a) led to a biphasic wave on which the test visual responses
ride. Once behavioral control was achieved, the baseline optical signal was
similar across different fixation conditions. When the optic flow stimulus
started, there was a difference in the signal starting at 1,000 ms for
the expansion versus the compression stimuli; this difference depended on
whether the animal was looking up or down as well as the region of the cortex.
[The interaction between the optical flow and the eye position tuning is the
subject of work in progress (Raffi and
Siegel 2002)]. This later visual signal was dependent on the
position in the orbit in a manner reminiscent of gain fields described from
electrophysiological studies. Thus both visual input and eye position
contributes to the optical response.
Baseline normalization analysis model
For any single location, the strength of the optical signal depended on the
eye position in the orbit. This dependence was modeled as a linear function.
Other models might have been used, as 40% of inferior parietal lobule
neurons have peaked nonlinear gain fields
(Andersen et al. 1985;
Read and Siegel 1997). In
preliminary analysis, higher-order functions such as a quadratic were used.
The qualitative results in terms of an upper and lower field breakdown between
area DP and 7a were similar; it was difficult to justify the higher-order
model on a pixel-by-pixel basis as the signal to noise was low. Hence the data
were modeled with the linear regression, and a Monte Carlo analysis was used
to validate the model parameters.
The effect of eye position on the visual response was visualized with
parameter maps. A positive dependence of the expected neuronal responses on
eye position was found in DP, whereas the negative relation was found in 7a.
This does not mean that DP exclusively represents upper field eye positions.
Rather the results indicate that DP is modulated by both upward and downward
eye positions and that upward positions leads to an increase in neural
activity with downward leading to a decrease in activity. Similarly 7a is
modulated by both upper and lower eye positions with an opposite sign to that
of DP. Interestingly, these parameter maps indicate that when the point of
regard for the eyes is in the upper visual field, both areas should be
modulated; similarly both are active for lower fixations. Thus either
projections from DP or 7a can provide information needed by recipient cortical
zones. It is also possible that areas such as area 46 and 8a that receive
interdigitated projection from both DP and 7a
(Andersen et al. 1990) could
use these complementary signals in a push-pull fashion (cf.,
Grossberg and Kuperstein
1986). It is tempting to speculate that this differential signal
computed from both 7a and DP could provide a more reliable signal to other
cortices representing visual space or planning motor behaviors.
Nature of the topographic representation of the gain field
The gain field topography was found in three hemispheres of two animals. In all three monkeys, the upper and lower gain field representation was found distributed between 7a and DP. This was shown using the nine position gain field test in all three animals. The upper and lower representation was the strongest topography, whereas the contra/ipsi-lateral representation was markedly weaker. One possibility that could explain the weaker contra/ispi-lateral representation is that it is in the more lateral 20 mm of 7a and DP that cannot be imaged in these experiments due to the chamber placement. For example, the ipsilateral lower gain fields could extend right up to the area 7a/7b border. The scant data that were acquired suggest that this is not the case (Fig. 5B), although additional data are certainly needed to resolve this issue. In some experiments, the contra/ipsilateral modulation is clear, and there appears to be an almost complete representation of the contralateral gain field; whereas in others the contra/ipsilateral representation is scarcely evident. The hemodynamic response and/or the noise in the optical signal may preclude a repeatable measurement or there is plasticity in this portion of the representation. Imaging with voltage-sensitive sensors that more faithfully represent the signals in both space and time may resolve whether there is indeed a contra/ipsi representation.
In the animal for which extensive mappings were performed, the maps were
not precisely the same day to day as evaluated across the 3 mo of recordings.
In comparison, the fine structure of ocular dominance is exquisitely
reproducible across months of recordings in striate cortex
(Shtoyerman et al. 2000). The
experimental procedures are essentially the same in the striate study as in
the present study of the inferior parietal lobule; eye position control is
similar, the wavelength of light is the same; the species are the same.
Possible sources for the variation in the inferior parietal lobule maps appear
to be linked to the cortical areas under study; the inferior parietal lobule
areas receive both highly processed visual signals from dorsal and ventral
stream areas as well as feedback from the frontal areas. It is clearly
possible that the inferior parietal lobule maps are modulated by the state of
the animal's attentional, intentional, or vigilance systems. The possibility
that the monkey's behavioral state could cause plasticity on a day-to-day
basis needs to be considered. Specific experiments to address the effect of
these factors o
or 
indicate
fixation points. In some experiments, compression optic flow was used as well
as expansion flow (see text).
2 test (see
text).
