Download Eye position encoding in the macaque ventral intraparietal area (VIP)

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts
no text concepts found
Transcript
Vision, Central
NeuroReport
NeuroReport 10, 873±878 (1999)
MANY neurons in area VIP encode the location of
visual stimuli in a non-retinocentric frame of reference.
In this context the question needed to be addressed
whether the underlying coordinate transformation of
the incoming visual signals could be generated within
area VIP or whether this information would have to
arrive from other areas. We tested 74 neurons in area
VIP of two awake monkeys for an in¯uence of eye
position while animals performed a ®xation task. More
than half of the neurons (40/74) revealed an eye
position effect. At the population level, however, this
effect was balanced out. We suggest that local connections within area VIP could be used to generate an
encoding of visual information in a non-retinocentric
frame of reference. NeuroReport 10:873±878 # 1999
Lippincott Williams & Wilkins.
Key words: Area VIP; Eye position; Head-centered
encoding; Monkey
Introduction
The ventral intraparietal area (VIP) is part of the
dorsal stream of the macaque visual cortical system
[1]. Neurons in these dorsal stream areas have been
implicated in the processing of visual spatial information and in the encoding of motion signals.
Neurons in area VIP combine both response features in the sense that they have spatially congruent
visual and tactile receptive ®elds (RFs) and are also
directionally selective for moving visual and tactile
stimuli [2±4]. In addition, many neurons respond to
rotational vestibular stimulation [5]. When all three
sensory modalities are present, preferred directions
for visual, tactile and vestibular stimulation are codirectional [6,7]. This orderly arrangement of responsiveness across modalities previously led us to
investigate the reference frame used for the encoding
of sensory signals from all three modalities. We
found that a subset of cells in area VIP encodes
visual spatial information in a non-retinocentric
(such as head-centered) frame of reference [8]. With
regard to the latter observation, the question remains
concerning the necessary parameters and related
signal processing to construct such a (potentially)
craniocentric spatial encoding at the single cell level.
Several theoretical studies have shown previously
that a combination of visual signals and information
about the position of the eyes in the head can be
used to provide such a non-retinocentric encoding
[9±13]. However, the population of eye position0959-4965 # Lippincott Williams & Wilkins
Eye position encoding in
the macaque ventral
intraparietal area (VIP)
F. Bremmer,1,2,CA W. Graf,1
S. Ben Hamed1,3 and J.-R. Duhamel1,3
1
CNRS, ColleÁge de France, 11 place Marcelin
Berthelot, F-75005 Paris, France; 2 Department of
Zoology and Neurobiology, Ruhr-University
Bochum, D-44780 Bochum, Germany; 3 Institute of
Cognitive Science, CNRS UPR 9075, 67 boulevard
Pinel, F-69675 Bron, France
CA,2
Corresponding Author and Address
in¯uenced cells would need to ful®ll certain prerequisites in order to allow a non-retinocentric encoding. These conditions would be met if (i) the
preferred directions, i.e. the gaze directions accompanied by the strongest discharge of the cells, were
uniformly distributed, and if (ii) the eye position
effect, observed at the single cell level, was balanced
out at the population level (see [14]).
In the present study we tested neurons in area
VIP for an in¯uence of eye position during active
®xation in darkness. More than half of the cells
revealed a statistically signi®cant eye position effect
(distribution free ANOVA). Two dimensional regression analysis was used to quantify the effect and
proved to be a statistically signi®cant or near signi®cant model for more than half of the cells. This 2-D
regression analysis allowed de®ning for each single
cell the direction of the steepest slope of the regression plane, i.e. the preferred direction. Preferred
directions were uniformly distributed. Furthermore,
the eye position effect was at equilibrium at the
population level. We thus conclude that within area
VIP all necessary information is available for generating a non-retinocentric encoding of visual spatial
information.
Materials and Methods
Experiments were performed in two female macaque
monkeys, one rhesus monkey (M. mulatta, 4.6 kg)
and one fascicularis monkey (M. fascicularis, 3.8 kg).
Vol 10 No 4 17 March 1999
873
NeuroReport
All animal care, housing and surgical procedures
were in accordance with national French and international published guidelines on the use of animals
in research (European Communities Council Directive 86/609/ECC). Most of the experimental methods employed in this investigation have been
described previously [8].
Behavioral paradigm and recordings: During training and the recording sessions, the animals were
restrained in a primate chair (with the head ®xed
only during recordings) while performing ®xation
tasks for liquid rewards. Rewards were given for
keeping the eyes within an electronically de®ned
window of 2 3 28 centered on the ®xation target. A
PC running the REX software package (NIH)
controlled behavioral paradigms and data acquisition. At the end of the training or the experimental
sessions, the monkeys were returned to their cages.
The animal weights were monitored daily and
supplementary fruit and/or water were provided if
necessary. For extracellular recordings, tungsten-inglass electrodes (Frederick Haer, Inc., impedance 1±
2 MÙ at 1 kHz) were advanced using a hydraulic
microdrive (Narishige) which was mounted on the
recording chamber. Neuronal activity and electrode
depth were noted to establish the relative positions
of landmarks, such as gray and white matter and
neuronal response characteristics. During recording
sessions, area VIP was identi®ed by its location
within the intraparietal sulcus and by its typical
physiological response characteristics, with regard to
the neighboring areas LIP and MIP: VIP neurons
show selectivity for the direction of visual stimulus
motion and are often also directionally selective
regarding tactile responses to stimulation of the face
or the head area [8]. One animal is still used in
ongoing experiments; in the other monkey histological analysis veri®ed that recording sites had been
located in area VIP.
Animal preparation: Monkeys were prepared for
recordings by implanting a head holding device
under general anesthesia (ketamine, Propofol) and
sterile surgical conditions. Scleral search coils were
implanted in each eye for monitoring eye position
movements. The wire leads were connected to a
plug on top of the skull. A recording chamber for
microelectrode penetrations through the intact dura
was anchored ¯at to the skull centered at P 3.5, L
12. Recording chamber, eye coil, plug and head
holder were all embedded in dental acrylic which
itself was connected to the skull by self-tapping
screws. Analgesics, antibiotics, etc. were given postoperatively. Recording started 1 week after surgery.
874
Vol 10 No 4 17 March 1999
F. Bremmer et al.
Visual target presentation, data analysis, and histology:
During the experiments, the animals were viewing a
translucent screen subtending the central 80 3 708 of
the visual ®eld. A ®xation target (diameter 0.58,
luminance 0.5 cd/m2 ) was back-projected by a liquid
crystal display system in random order at nine
different locations on the screen without any further
visual stimulation. Locations were the center of the
screen plus eight concentrically located points 158
away from the center ([x, y] ˆ [08, 08], [08, 158],
[158, 08], [10.68, 10.68]). Presentation of ®xation
targets (starting at t ˆ 0 ms) lasted for 2000 ms. The
animal's eye position had to be within the electronically de®ned window not later than t ˆ 700 ms.
Trials were aborted if the animal broke ®xation after
this time (t ˆ 700 ms) and the end of the trial (t ˆ
2000 ms). Mean neuronal activity was recorded
throughout the trial and analyzed for each ®xation
location for an epoch between t ˆ 1000 ms and
t ˆ 2000 ms, thus guaranteeing that eye position had
been maintained continuously within the electronically de®ned window for > 300 ms before onset of
the analyzing epoch.
Differences in activity were tested for statistical
signi®cance with a distribution-free ANOVA. Twodimensional linear regression analysis was applied to
quantify the eye position effect. For validating the
planar model as ®t to the observed data, an F-test
was employed. Standard histological techniques
were applied to reconstruct the recording locations
[2,3].
Results
A total of 74 neurons was recorded from two left
hemispheres of two monkeys. The activity of more
than half the cells (40/74; 54%) was in¯uenced by
the differential position of the eyes in the orbit
during ®xation in darkness. With respect to this eye
position effect area VIP is structured very homogeneously since cells showing an eye position effect
were found throughout area VIP without any
obvious clustering into intra-areal sub-regions.
Furthermore, no evidence for any eye position map
could be found, i.e. neighboring cells varied their
preferences for eye positions in an unsystematic
manner.
The eye position effect: single cell level: The modulatory effect of eye position was quanti®ed using a
two-dimensional linear regression analysis. In the
illustrated example (Fig. 1), discharges were strongest for ®xation locations right and downward
(ANOVA: P , 0.0001). Activity decreased for eye
positions leftward and upward. Figure 1A shows the
mean discharges (s.d.) for the different ®xation
Eye position effects in area VIP
NeuroReport
B
40
ikes/s)
40
30
30
p
Activ. (S
Activ. (Spikes/s) 6 s.d.
A
20
10
20
10
0
C
L
LU
U
RU
R
RD
D
LD
Fixation location
20
2
10
2 g)
0 (de
al
10 rtic
Ve
0
0
20
10 g)
0 l (de
nta
izo
2
20
21
r
Ho
z(Spikes/s) 5 0.479x 2 0.196y 1 23.46
r2 5 0.875. p , 0.002
FIG. 1. Eye position effect during ®xation in darkness: single cell level. (A) Mean activity values ( s.d.) observed at nine different ®xation locations (C
ˆ center; LU ˆ left up; U ˆ up; RU ˆ Right Up; R ˆ Right; RD ˆ right down; D ˆ down; LD ˆ left down; L ˆ left). In this case, discharge rates were
strongest for ®xation locations right and downward. (ANOVA: P , 0.0001). The dotted lines indicate the ideal values, i.e. those discharges perfectly
matching the regression values. (B) The shaded plane represents the two-dimensional linear regression to the mean discharge values (r2 ˆ 0.875,
P , 0.002). The x±y plane in this plot represents the central 40 3 408 of the tangent screen. The base point of each drop line depicts the ®xation location
on the screen, and the height of each line depicts the mean activity value during ®xation at this location (same data as in A).
locations. Dotted lines indicate ideal data, i.e. values
perfectly matching the regression values. Since approximation of a regression plane with horizontal
and vertical eye position as independent variables is
equivalent to the approximation of a cosine function
with gaze angle as independent variable for a ®xed
gaze amplitude (in our case 158) [14], the ideal data
are located on a cosine-like tuning curve. The ideal
discharge value for central ®xation is given by the
intercept value of the regression plane. Figure 1B
shows the same data (needle-like drop lines) in a 3D view. The shaded plane depicts the regression
plane approximated to the mean discharge values,
with regression parameters given below the diagram.
The eye position effect: population level: A regression plane was approximated to the discharge of
each individual neuron. For 37% (27/74) of the
recorded neurons, this ®t was signi®cant at P ,
0.05. For another 16% (12/74) of the neurons, the
approximation was nearly signi®cant at P , 0.1. The
eye position effect, which could be observed at the
single cell level, was at equilibrium at the population
level (Fig. 2). Average discharge values of the
population of eye position-affected neurons for the
different ®xation locations were not signi®cantly
different (ANOVA: P . 0.8. Fig. 2A). A population
discharge plane obtained by averaging the parameters of the individual regression planes illustrates
the same result (Fig. 2B): the plane has virtually no
slope. Thus, both mathematical approaches, i.e.
average response of neuronal discharges as well as
the average response plane, show an invariance of
neuronal discharges with respect to eye position at
the population level.
The symmetry of the population response, i.e. the
¯atness of the population discharge plane, could
result, e.g. from a roughly equal distribution of eye
position effects across all parts of the oculomotor
range. Alternatively, it could result from a small
number of neurons ®ring at high rates in one part of
the oculomotor range and a larger number of
neurons ®ring at relatively low rates for the opposite
part of the oculomotor range. We, therefore, also
analysed the distribution of the gradients of the
regression planes, i.e. the amount and the direction
of the steepest increase of activity with eye position.
The analysis indicated that the directions of the
gradients were uniformly distributed (Fig. 3: central
two-dimensional graph; ÷2 test: P . 0.8). For the
population of eye position affected neurons, regression slopes tended to be normally distributed
(Fig. 3: histograms on top of and to the right of the
central scatter plot). In other words, for a number of
neurons (like the one shown in Fig. 1 with increasing activity for down and rightward eye position,
i.e. the gradient direction) there exists an almost
equivalent number of neurons with roughly the
same increasing activity in the opposite (left and
upward) gradient direction.
Discussion
Most of the recorded neurons were visually responsive. One might argue that if the target within the
de®ned control window was imperfectly ®xated,
Vol 10 No 4 17 March 1999
875
NeuroReport
F. Bremmer et al.
A
B
ikes/s)
40
30
p
Activ. (S
Activ. (Spikes/s)
30
20
20
10
0
10
Ve
rti 0
ca
l ( 21
de 0
g) 2
20
10
0
C
L
LU
U
RU
R
RD
D
LD
20
10 g)
0 l (de
nta
0
21 orizo
0
H
22
z(Spikes/s) 5 20.005x 2 0.015y 1 12.99
n 5 40
Fixation location
FIG. 2. Eye position effect during ®xation in darkness: population level. (A Mean activity ( s.d.) for nine different eye positions obtained from those
cells that showed a statistically signi®cant eye position effect (n ˆ 40). (Abbreviations as in Fig. 1). The discharge values for the different eye positions
were not signi®cantly different (ANOVA: P . 0.8), indicating that the modulatory effect of eye position is balanced out at the population level. (B) The
mean population response plane was obtained by averaging all regression planes obtained from those cells that showed a statistically signi®cant eye
position effect (n ˆ 40). Both mathematical approaches (A and B) lead to the identical result: the resulting discharge plane proved to be ¯at.
Number
10
5
Vertical slope [(Spikes/s)/deg]
0
0.5
0.0
20.5
10
5
0
20.5
0.0
0.5
Horizontal slope [(Spikes/s)/deg]
FIG. 3. Distribution of the gradients of the regression planes obtained
from those cells that showed a statistically signi®cant eye position effect.
In the central scatter plot each single dot represents the gradient of an
individual regression plane. Statistical analysis proved the directions of
the gradients to be uniformly distributed. Histograms on top and to the
right of the central scatter diagram show the distribution of slopes along
only one oculomotor axis: horizontal (top) and vertical (right). Note that
regression slopes tend to be normally distributed.
different visual signals could have been produced at
different eye positions, and that this in itself could
have generated the observed effects. Such a visual
origin of the observed effects can be excluded for
several reasons. First, many of the cells (like that
shown in Fig. 1) had visual receptive ®elds (RFs)
excluding the foveal and parafoveal region. Thus,
876
Vol 10 No 4 17 March 1999
the ®xation target would not have stimulated the
cells' RF at all. Secondly, if for the remaining cells
the animals had been systematically off the target,
these `target stimuli' still would not have modulated
neuronal activity since neurons in area VIP do not
respond to stationary visual stimuli [2,4±6,8].
Several theoretical studies have suggested that the
functional properties of neurons affected by eye
position might be used to accomplish a coordinate
transformation of the incoming visual signals into a
non-retinocentric frame of reference [9±12,14,15].
Such a non-retinocentric encoding could be centered
with respect to the head (head-centered or craniocentric), to the body (body-centered or egocentric)
or to the external world (world-centered or allocentric). The coordinate systems of all of these nonretinocentric types of encoding share the common
property that their origin is not centered on the
retina. For the sake of simplicity, any non-retinocentric frame of reference will be termed headcentered (or craniocentric) in the following.
The basic motivation of the theoretical studies
mentioned above was the observation that general
sensorimotor processing involves several parts of the
body aside from the eye. Thus, organizing appropriate motor outputs like avoiding an obstacle,
reaching for an object, etc., would require coordinate systems centered on the respective output
structure (head, body, limb etc.). The mathematical
algorithms suggested by these studies allow the
construction of such a kind of effector-centered
encoding by combining, in a ®rst-stage transformation, information about the location of a visual
Eye position effects in area VIP
stimulus on the retina with information about the
position of the eyes in the orbit. At the physiological level, however, several questions remained unanswered by these studies, in particular whether the
resulting craniocentric encoding was carried out
implicitly by a population of neurons or whether
their function could be identi®ed explicitly at the
single neuron level. In case of such an explicit
encoding, the question would arise whether it occurred within the same area where the eye positionmodulated neurons are found, or whether it occurred outside this area in a population of neurons
unaffected by eye position.
In a previous study we showed that many neurons
in area VIP encode the location of an object in space
in at least head a-centered frame of reference: visual
receptive ®elds remained spatially constant regardless of eye position [8]. Neurons with similar
properties had been found earlier in another dorsal
stream area, i.e. area V6 [16]. Clearly, there exist
visual cortical areas with an explicit head-centered
encoding at the single cell level. Interestingly, in
addition to this explicit head-centered encoding,
many neurons in both areas (VIP and V6) were
found to encode in a `classical' eye-centered frame
of reference. Finally, visual driven activity of about
half of the eye- and head-centered VIP cells, as well
as many V6 cells [17], was modulated by the
position of the eyes in the orbit. Thus, eye position
effects can also be found in areas with an explicit
head-centered encoding at the single cell level. The
present study goes one step further by showing that
the distribution of eye position effects in area VIP is
suited to generate this observed head-centered encoding: the preferred directions of the eye position
effects are uniformly distributed. This observation
indicates two different facts: (i) eye position cells
can be found across all parts of area VIP, (ii) there
exists no topographical map for eye position effects,
i.e. there do not exist any clusters of cells showing
the same preference for a speci®c eye position. In
other words, for a large enough ensemble of cells no
speci®c eye position exists that elicits a stronger or
weaker average activity compared to others. As this
study has shown, a relatively small number of eye
position affected cells (n ˆ 40) sampled throughout
area VIP is suf®cient to obtain such an unbiased
population response. These discharge characteristics
allow precise encoding of the position of the eyes in
the head, and thus indicate the capability of the
existing network to construct head-centered cells by
a simple connectivity scheme [14]. This capability,
however, does not prove that neuronal discharges of
eye-position affected neurons in area VIP are indeed
used to construct such head-centered cells. Such a
(virtually impossible) proof would require
NeuroReport
identi®cation of all inputs to a head-centered cell
and testing of all input cells for an in¯uence of eye
position on their discharge.
Head-centered encoding and the eye position effect: Head-centered cells have been described in
three cortical areas of the macaque sensorimotor
system: areas VIP and V6 [16] and the premotor
cortex [18]. In all three areas, neurons affected by
eye position were also found. As in the present
study, eye position effects suited to generate headcentered encoding within the very same area were
also shown to exist in premotor cortex [19]. Review
of the published data suggests that also the distribution of the eye position effects in area V6 allows
such a kind of encoding. Eye position effects have
also been described in other areas of the macaque
sensorimotor system: V3A, MT, MST, LIP, 7A and
SEF [20±25]. In none of these, head-centered cells
were described although the requirements, i.e. eye
position in¯uenced cells with a speci®c population
characteristic, were ful®lled. We thus consider the
modulatory in¯uence of eye position on neuronal
discharges to be a common phenomenon throughout
the monkey cortical system that probably subserves
an implicit representation of spatial information in a
head-centered frame of reference. We propose that
an explicit encoding at the single cell level, however,
seems to be present only in areas which participate
speci®cally in the control and sensory guidance of
body parts other than the eyes.
Conclusion
The activity of more than half of the cells in area
VIP is in¯uenced by eye position. The distribution
of eye position effects in area VIP is similar to that
observed previously in areas V3A, MT, MST, LIP,
7A, V6 and PMd. We thus consider the modulatory
in¯uence of eye position on neuronal discharges to
be a common phenomenon in monkey cortex,
probably subserving an implicit representation of
spatial information in a head-centered frame of
reference. Cells explicitly coding in a head-centered
frame of reference, as have been shown to exist also
in area VIP, seem to be restricted to areas speci®cally involved in the control and sensory guidance
of body parts other than the eyes.
References
1.
2.
3.
4.
Felleman DJ and Van Essen DC. Cereb Cortex 1, 1ÿ47 (1991).
Colby CL, Duhamel J-R and Goldberg ME. J Neurophysiol 69, 902ÿ914 (1993).
Duhamel J-R, Colby CC and Goldberg ME. J Neurophysiol 79, 126ÿ136 (1998).
Duhamel J-R, Colby CL and Goldberg ME. Congruent representations of visual
and somatosensory space in single neurons of monkey ventral intra-parietal
cortex (area VIP). In: Paillard J, ed. Brain and Space. Oxford, Oxford University
Press, 1991: 223±236.
Vol 10 No 4 17 March 1999
877
NeuroReport
5. Bremmer F, Duhamel J-R, Ben Hamed S et al. Soc Neurosci Abstr 21, 282
(1995).
6. Bremmer F, Duhamel J-R, Ben Hamed S et al. The representation of movement
in near extra-personal space in the macaque ventral intraparietal area (VIP). In:
Thier P and Karnath H-O, eds. Parietal Lobe Contributions to Orientation in 3D
Space. Heidelberg: Springer Verlag, 1997: 619±630.
7. Graf W, Bremmer F, Ben Hamed S et al. Soc Neurosci Abstr 22, 7 (1996).
8. Duhamel J-R, Bremmer F, Ben Hamed S et al. Nature 389, 845ÿ848 (1997).
9. Zipser D and Andersen RA. Nature 331, 679ÿ684 (1988).
10. Mazzoni P, Andersen RA and Jordan MI. Proc Natl Acad Sci USA 88,
4433ÿ4437 (1991).
11. Goodman SJ and Andersen RA. IJCNN 2, 381ÿ386 (1990).
12. Pouget A, Fisher SA and Sejnowski TJ. J Cogn Neurosci 5, 150ÿ161 (1993).
13. Pouget A and Sejnowski TJ. J Cogn Neurosci 9, 222ÿ237 (1997).
14. Bremmer F, Pouget A and Hoffmann K-P. Eur J Neurosci 10, 153ÿ160 (1998).
15. Pouget A and Sejnowski TJ. J Cogn Neurosci 9, 222ÿ237 (1997).
16. Galletti C, Battaglini PP and Fattori P. Exp Brain Res 96, 221ÿ229 (1993).
17. Galletti C, Battaglini PP and Fattori P. Eur J Neurosci 7, 2486ÿ2501 (1995).
878
Vol 10 No 4 17 March 1999
F. Bremmer et al.
18. Fogassi L, Gallese V, Di Pellegrino G et al. Exp Brain Res 89, 686ÿ690 (1992).
19. Boussaoud D, Jouffrais C and Bremmer F. J Neurophysiol 80, 1132ÿ1150
(1998).
20. Andersen RA, Bracewell RM, Barash S et al. J Neurosci 10, 1176ÿ1196 (1990).
21. Boussaoud D. J Neurophysiol 73, 886ÿ890 (1995).
22. Bremmer F, Ilg UJ, Thiele A et al. J Neurophysiol 77, 944ÿ961 (1997).
23. Bremmer F, Distler C and Hoffmann K-P. J Neurophysiol 77, 962ÿ977 (1997).
24. Schlag J, Schlag-Rey M and Pigarev I. Exp Brain Res 90, 302ÿ306 (1992).
25. Galletti C and Battaglini PP. J Neurosci 9, 1112ÿ1125 (1989).
ACKNOWLEDGEMENTS: This work was supported by grants from the European
Union (HCM: CHRXCT930267) and the Human Frontier Science Program (RG71/
96B).
Received 6 January 1999;
accepted 24 January 1999