Representation of the visual field in the lateral intraparietal ... - CiteSeerX

2 downloads 0 Views 257KB Size Report
Jul 24, 2001 - of macaque monkeys: a quantitative receptive field analysis. Received: 30 October 2000 ... cessing of visual information at the level of the RFs.
Exp Brain Res (2001) 140:127–144 DOI 10.1007/s002210100785

R E S E A R C H A RT I C L E

S. Ben Hamed · J.-R. Duhamel · F. Bremmer W. Graf

Representation of the visual field in the lateral intraparietal area of macaque monkeys: a quantitative receptive field analysis Received: 30 October 2000 / Accepted: 27 April 2001 / Published online: 24 July 2001 © Springer-Verlag 2001

Abstract The representation of the visual field in the primate lateral intraparietal area (LIP) was examined, using a rapid, computer-driven receptive field (RF) mapping procedure. RF characteristics of single LIP neurons could thus be measured repeatedly under different behavioral conditions. Here we report data obtained using a standard ocular fixation task during which the animals were required to monitor small changes in color of the fixated target. In a first step, statistical analyses were conducted in order to establish the experimental limits of the mapping procedure on 171 LIP neurons recorded from three hemispheres of two macaque monkeys. The characteristics of the receptive fields of LIP neurons were analyzed at the single cell and at the population level. Although for many neurons the assumption of a simple two-dimensional gaussian profile with a central area of maximal excitability at the center and progressively decreasing response strength at the periphery can represent relatively accurately the spatial structure of the RF, about 19% of the cells had a markedly asymmetrical shape. At the population level, we observed, in agreement with prior studies, a systematic relation between RF size and eccentricity. However, we also found a more accentuated overrepresentation of the central visual field than had been previously reported and no marked differences between the upper and lower visual representation of space. This observation correlates with an extension of the definition of LIP from the posterior third of the lateral intraparietal sulcus to most of the middle and posterior thirds. Detailed histological analyses of the recordS. Ben Hamed · J.-R. Duhamel (✉) · F. Bremmer · W. Graf CNRS Collège de France, 11 place Marcelin Berthelot, 75005 Paris, France e-mail: [email protected] Tel.: +33-04-37911235, Fax: +33-04-37911210 F. Bremmer Department of Zoology and Neurobiology, Ruhr University, 44780 Bochum, Germany S. Ben Hamed · J.-R. Duhamel Institut de Sciences Cognitives, CNRS UPR 9075, 67 boulevard Pinel, 69675 Bron, France

ed hemispheres suggest that there exists, in this newly defined unitary functional cortical area, a coarse but systematic topographical organization in area LIP that supports the distinction between its dorsal and ventral regions, LIPd and LIPv, respectively. Paralleling the physiological data, the central visual field is mostly represented in the middle dorsal region and the visual periphery more ventral and posterior. An anteroposterior gradient from the lower to the upper visual field representations can also be identified. In conclusion, this study provides the basis for a reliable mapping method in awake monkeys and a reference for the organization of the properties of the visual space representation in an area LIP extended with respect to the previously described LIP and showing a relative emphasis of central visual field. Keywords Parietal cortex · Monkey · Electrophysiology · Receptive field · Visual representation

Introduction Higher-order areas in the occipitoparietal cortical pathway carry multiple classes of signals related to sensory, motor and cognitive parameters. One such example is the lateral intraparietal area (LIP), which, by its connectivity and by the response properties of single neurons in different behavioral conditions, is thought to be involved in predictive visual processing, visuospatial attention and saccadic eye movement programming (Lynch et al. 1977; Gnadt and Andersen 1988; Blatt et al. 1990; Duhamel et al. 1992; Colby et al. 1996). In macaques, area LIP has been described to occupy the most posterior third of the lateral bank of the intraparietal sulcus. Its major sources of input originate in the extrastriate visual cortex. It has an extensive set of connections with both dorsal and ventral stream areas, including V2, V3, V3A, MT, MST, PO, V4, MDP, DP, TEO, and TE (Seltzer and Pandya 1986; Colby et al. 1988; Blatt et al. 1990; Andersen et al. 1990a; Felleman and van Essen 1991; Bullier et al. 1996). Area LIP is re-

128

ciprocally connected with other parietal areas (VIP, 7a) (Seltzer and Pandya 1986; Blatt et al. 1990; Andersen et al. 1990a) and has outgoing connections directed to the frontal eye fields (Cavada and Goldman-Rakic 1989; Andersen et al. 1990a; Schall et al. 1995; Bullier et al. 1996), to different subdivisions of the premotor cortex (Cavada and Goldman-Rakic 1989), and to the intermediate layers of the superior colliculus (Asanuma et al. 1985). The strong visual input received by LIP predicts an equivalently strong visual responsivity, and indeed most LIP neurons display reliable and robust responses to visual stimulation (Andersen et al. 1985; Colby et al. 1996). However, other types of stimuli can activate LIP neurons or modulate their visual responsiveness. LIP visual responses have been shown to be enhanced through the manipulation of stimulus behavioral relevance or saliency (Lynch et al. 1977; Colby et al. 1996; Ben Hamed et al. 1997a; Platt and Glimcher 1997). Learned responses to auditory stimuli have also been described (Stricanne et al. 1996; Grunewald et al. 1999; Linden et al. 1999), as well as by eye and head position signals (Andersen et al. 1990b; Brotchie et al. 1995; Bremmer et al. 1997, 1998), and to fire before purposive saccadic eye movements in the absence of visual stimulation (Barash et al. 1991a, 1991b; Colby et al. 1996). Such diverse response properties have led to postulating a central role for area LIP in multimodal spatial analysis, and in attentional and oculomotor functions. Considerable effort has been made in recent years to document the effects of behavioral variables in area LIP through the use of elaborate experimental paradigms. However, with the exception of a single study conducted in anesthetized animals (Blatt et al. 1990), comparatively little attention has been given until recently to basic receptive field properties and to visual field representation in LIP (Platt and Glimcher 1998). Such information is necessary in order to better understand how visual space is represented in a single area, how visual representation varies across the different subdivisions of the cortical visual system, and how these variations relate to the achievement of specific functional goals. Furthermore, sampling visual responses over a large portion of space may help highlight the mechanisms through which a given behavioral context influences information processing across the whole visual field representation in a given cortical area and not only at the locus where the attended stimulus or the programmed saccade endpoint is positioned. Classical studies that aimed to characterize the receptive fields (RFs) of subcortical and cortical visual areas used hand-mapping procedures for which the decision about the significance of the visual response is dependent upon the experimenter. The determination of RFs is shown to be constant across two successive mapping sessions for a given experimenter but not necessarily for different experimenters. As a result, estimates of RF size vary from one study to another. However, the relationship between size and eccentricity remains fairly consistent for a given visual area. The main limit of this proce-

dure is that it is qualitative and does not allow a refined analysis of both temporal and spatial properties of RFs. The introduction of automatic mapping procedures using white noise stimulation and reverse correlation techniques has yielded a new understanding of the processing of visual information at the level of the RFs. This approach has been used to study the RFs of lower visual areas in anesthetized animals (in LGN: Cai et al. 1997; in V1: DeAngelis et al. 1993a, 1993b). It has also been extended to higher visual areas such as MT and MST (Raiguel et al. 1995, 1997). The precise spatiotemporal characteristics of RFs were analyzed in relation to the process of feature extraction from an ongoing visual scene. Independently, other studies have aimed to analyze the changes in visual responsiveness in relation to attention to stimulus attributes such as color or spatial location (Bushnell et al. 1981; Goldberg et al. 1990; Motter 1993, 1994; Steinmetz et al. 1994; Duhamel et al. 1995; Connor et al. 1996; Ben Hamed et al. 1997b), or to the preparation of an eye-orientation movement (Duhamel et al. 1992). These latter approaches, carried out in higher visual areas in awake animals, have not examined in detail the neurons’ RF characteristics or substructure. In the present study, we tested an experimenter-free method that uses a computerized stimulus presentation procedure and allows the generation of a quantitative representation of the RF structure of single neurons. The RF data presented here have been obtained during the performance of a foveal fixation task. In the first part of the present study, the reliability and limits of this computerized mapping method were investigated, and the basis for further analysis and comparisons of RF structure were laid out. In the second part, the representation of the visual field in LIP was investigated by two complementary approaches: (1) spatial properties of a single RF were studied and a critical perspective was laid on the informative parameters of the fine structure of RFs. (2) Information about all the recorded RFs was used to represent the allocation of spatial resources (in terms of cell number) as a function of visual field extent. We thus describe an overrepresentation of the central field at the cortical level with respect to other areas. We also show that the visual, saccadic and delay activities which characterize LIP neurons are not restricted to the posterior third of the lateral intraparietal bank but also extend into most of the middle third of it. This newly defined anterior part of LIP contains both an emphasis on the lower visual field in the ventral part of the bank and on the central space in the dorsal part. Thus it emerges that LIP does not hold a representation of space with an up-down asymmetry, but rather with a center-to-periphery asymmetry. The coarse but systematic topography of LIP is also confirmed and expended: the central visual field is found to be mostly represented in the dorsal region, while the visual periphery is represented more ventrally, with an anteroposterior gradient from the lower to the upper visual field representations.

129

Materials and methods Animal preparation Two female macaque monkeys (Macaca mulatta and M. fascicularis) were used for this study (3.8 kg and 4.6 kg, respectively). Before surgery, they were trained to sit in a primate chair and to fixate a spot of light to receive a liquid reward (Wurtz 1969). They were surgically prepared under general anesthesia (induced with 10 mg/kg propofol and maintained at 15 mg/h/kg) for chronic neurophysiological recording by implantation of scleral search coils (Judge et al. 1980), headholding devices, and recording chambers through which electrodes could subsequently be introduced into the cerebral cortex. Recording chambers (1.8 cm diameter) were anchored over the intraparietal sulcus at stereotaxic coordinates AP –11 and ML 16 mm for a first chamber and AP 5 and ML 11 for the two other chambers. The chambers were placed flat against the skull. Given the orientation of the intraparietal sulcus, this placement was approximately orthogonal to the skull, and allowed long, tangential electrode penetrations through either banks of the sulcus. Animals were watched closely following surgery and given analgesics as needed. During the recording period, animal weight and health status were carefully monitored. Fluid supplements were given as needed. Recording chambers were flushed with saline before and after each recording session and antibiotics were applied as needed. When necessary, granulation tissue accumulated over the exposed dura in the recording chamber was removed with the animal under ketamine anesthesia. All experimental procedures were in compliance with local and European regulations (European Communities Council Directive 86/609/EEC).

penetration, and while the monkey was performing this task, we actively searched for neurons with visual, delay-period and saccade-related activity which are characteristic of LIP neurons. However, because we were interested in the modulation of visual response properties of LIP neurons by both external and internal cues, and thus very aware of this phenomenon, we also explored the responses of the isolated neurons in a wide range of other tasks, aiming to characterize the specificity of its visual response (orientation selectivity, color selectivity, habituation properties, attentional modulations, etc.). When a neuron was isolated at the electrode tip and the signal was stabilized, its visual and saccadic properties were thoroughly explored across the visual field encompassed by the tangential screen and beyond in all directions, using static and dynamic hand-held objects and projected stimuli on the side walls and ceiling. This was done for two main reasons: (1) towards the fundus of the sulcus, LIP is adjacent to area VIP. The neurons of this area have unmistakable properties (sensitivity to stimulus direction of movement, optic flow responses, bimodal responses, etc.), which enable us to distinguish them from LIP neurons and thus to identify the border between the two areas (Colby et al. 1993). The RF of VIP neurons can, however, be very eccentric and we thus systematically explored as large an extent of the visual field as possible. (2) The focus of the present study and of the study being conducted in parallel being the visual representation in LIP, it was crucial not to miss any type of visual response in the explored cortical region. We were specifically interested in recording from neurons with definite visual responses. Neurons with purely visual responses were intermingled with visual and saccade related neurons. Typically, complete data sets of neurons with mostly presaccadic activity were not recorded.

Physiological methods Recordings were made with flexible tungsten commercial microelectrodes (Frederick Haer electrodes) introduced through stainless steel guide tubes. A nylon grid held rigidly in the recording cylinder was used to maintain the guide tubes in place and permitted highly reproducible electrode penetrations with a resolution of 0.5 mm (Crist et al. 1988). The electrode was lowered into the brain with a step-motor driven hydraulic microdrive (Narishige). Penetrations performed several months apart at the same grid location were found to yield neurons with similar response types at similar depths. During a recording session, the head-fixed monkey sat in a primate chair in a totally dark room facing a tangent screen 57 cm away. The screen was 1×1.4 m in size, and the effective projection surface subtended slightly more than 90° in width and 70° in height. Visual stimuli of adjustable size, shape, color and motion pattern were produced by computer software developed in the laboratory. The stimulations were back-projected onto the screen via a liquid-crystal projection system. Horizontal and vertical eye position signals were measured using the magnetic search coil method. Behavioral control, eye position monitoring, stimulus position and timing and unit recording were performed using a PC-based real time experimental system (REX, Hays et al. 1982). Eye position was sampled and stored at 250 Hz and discriminated units were stored at 1000 Hz. The computer program was able to display rasters online, synchronized to one of several events such as achievement of fixation, stimulus appearance or extinction, eye movement onset, or reward. Unit discharges, eye position traces, task parameters and behavioral indicators were saved on disk for offline analysis. The monkeys were trained on a series of tasks designed to differentiate sensory, attentional and motor correlates of neural activity. In particular, a memory guided saccade task was used to test for specific saccade-related discharges. In this task, the monkey had to maintain fixation on a central target, while a peripheral stimulus was flashed for 150 ms at one of several possible predefined locations. After a delay of 1600 ms, the fixation spot was extinguished, providing the cue for the monkey to make a saccade to the remembered location of the peripheral flash. In each electrode

Histological methods Histology was carried out on the Macaca mulatta, the other monkey still participating in ongoing experiments. After recording sessions were terminated in the rhesus monkey, microlesions (50 µA for 15 s) were made at specific locations in the lateral banks of the intraparietal sulci of both hemispheres. The animal was initially fixated with a buffered solution (pH 7.4) of 4% paraformaldehyde. The head was severed and introduced into the stereotaxic apparatus. Marking pins were inserted through the periphery of the recording grid with a microdrive to document the extent of the chamber as well as the orientation of the recording grid. Subsequently, the head was postfixed in paraformaldehyde solution for several days. At the end of this period, the pins were removed, and the brain was extracted. A block of tissue containing the intraparietal sulci and neighboring regions was cut from both hemispheres. These tissue blocks were immersed in 0.4 M phosphate buffer/10% sucrose solution for 2 days. The tissue was then cut on a freezing microtome (50-µm sections). In order to capture complete electrode penetrations within single brain sections, the tissue was cut along planes parallel to the marking pins. Sections were counterstained with thionine. One in ten sections were myelin stained using the Schmued method (Schmued 1990). Sections were digitized using a microscope coupled to a Neurolucida system. Twodimensional flattened reconstructions of the intraparietal sulcus were subsequently obtained using software developed in our laboratory. Receptive field mapping procedure In the present study, and after an exhaustive characterization of the properties of the neural response of the isolated unit, receptive fields were mapped while the monkey was performing a standard central fixation task in an otherwise totally dark room. The monkey gazed at a central green fixation point and was rewarded for holding its eye position within a 2° wide electronically defined window for a 2300- to 3000-ms interval and for releasing a lever

130 large as 10°, depending on the manual estimate of the RF size (Fig. 1b). Online receptive field mapping Once the spatial and temporal mapping parameters were set to the neuron’s response characteristics, stimulation cycles were repeatedly run on the screen during the pre-discrimination period of the fixation task. A single cycle stimulated the center of each subregion in random order. Typically, stimulus durations and interstimulus intervals were 100 ms and 200 ms, respectively. About six to eight stimuli were presented during a single fixation trial. Thus, a complete stimulation cycle was obtained every seven to nine trials. Depending on the neural responsiveness, 7–12 stimulation cycles were run for a given unit. Stimulus order was reshuffled every time a new cycle began. Thus, each point of the stimulation grid was stimulated at least 7 times, each time embedded in a different spatial and temporal context. Data analysis Offline reverse correlation analysis

Fig. 1A, B RF mapping procedure during a standard central fixation task. A Visual stimuli presented during fixation (fixation point and mapping stimuli); bar status and eye position are shown relative to the same time scale. The monkey had to grasp a bar in order to initiate a trial (straight line in bar status). This triggered the appearance of the fixation point (black in fixation point status), which the monkey was required to fixate. After 2300–3000 ms following the beginning of fixation, the central stimulus changed color (gray in fixation point status), which signaled to the monkey to release the bar within a maximum delay of 700 ms (dashed line in bar status) in order to obtain a liquid reward. During a single fixation trial, six to eight mapping stimuli were flashed (numbered stimuli). B Stimuli were flashed at randomly selected locations within a predefined grid, optimally centered on the RF of the cell (represented by the idealized ellipse), and not on the fixation point (solid dot) within 700 ms of the dimming or change in color of the fixation point (Fig. 1a). While the monkey was performing this task, the receptive field of the recorded neuron was mapped in two steps: Optimization of the mapping parameters to each recorded neuron (a) The receptive field of the isolated unit was first approximated manually with a hand held projector. (b) The cell was tested with several computer-controlled visual stimuli varying in color, size, shape and orientation. The mapping was carried out using a simple achromatic 1° wide spot, which was an optimum stimulus for the vast majority of neurons. (c) The neuron’s response to repeated presentations of a brief flash at a fixed location inside the RF was analyzed in order to estimate its latency and the time course of its decay. This information was used to optimize the duration of the mapping stimulus and the interstimulus interval. (d) A matrix of stimulus locations was defined in order to cover an area larger than the RF. The matrix contained 7×7-square subregions in about two-thirds of the cells. The remaining third of the cells was tested at a higher resolution (9×9). For the majority of the neurons, each subregion had a width of 6°, but it could be as small as 1° or as

Peristimulus histograms (PSTHs) were constructed for each matrix pixel using the information collected over the successive mapping cycles. The PSTHs of the ten stimuli giving the highest neuronal activity were displayed using an interactive interface. A time window was adjusted interactively around the responses on the basis of the latency and the time course of the response. The same time window was applied to all the locations of the matrix in order to determine the average frequency of discharge induced by the corresponding stimulus. LIP neurons responded remarkably well to these briefly flashed stimuli presented in rapid sequence. The reverse correlation method implies that the averaging window is adjusted to the latency and peak response, which yielded relatively high mean firing rates for cells with a sharp phasic burst of activity. The obtained rates are somewhat higher than the typical mean firing rates reported in other studies using wider averaging windows set on stimulus onset rather than burst onset. Construction of receptive field maps Visual receptive fields are often approximated by gaussian or Gabor functions. These approximation procedures make essential hypotheses on the spiking properties of neurons, and more particularly on their spatial profile. When this profile is unknown or when, as in the present study, there is a specific interest in a quantitative description of the spatial distribution of activity within receptive fields, these methods are not appropriate. However, in order to increase the size of each single recording set beyond the experimental limitations of the spatial and temporal resolutions, a two-dimensional cubic interpolation method was used. It consisted of new interleaving data points that were generated between the experimentally obtained ones to the best fitting polynomial function of the third degree. The input to this operation was a square matrix containing the average spike responses for each single stimulus. The output was a larger square matrix that necessarily contained the experimental data points. In essence, three new points were generated between each two experimental data points. Thus if the input was a 7×7 matrix of experimental data points, the output was a 25×25 matrix. Description of receptive fields Several descriptive parameters were associated with the representation of RFs for each data set. (1) The maximum discharge rate was referred to as the peak. (2) The x- and y-locations of the peak were expressed in degrees of visual angle with respect to the cen-

131 ter of the screen. (3) The x- and y-locations of the center of gravity, i.e., of the center of mass of activity for the portion of the grid for which responses exceed half the maximum response of the neuron. (4) A width (the average extent of the RF) was defined in degrees as the square root of the surface of the RF whose activity was above a critical value. This critical value was set to half the peak discharge rate.

Results A convenient way of representing neural activity over a mapped region of the visual field is to use gray-shadescoded two-dimensional maps. The procedure of how such a map is derived from the peristimulus histograms obtained for each stimulated subregion is illustrated for one representative neuron (Fig. 2). The first step in the analysis of LIP receptive field data was to assess the reliability of our mapping procedure, since to our knowledge this is the first time such a method is applied to analyze the visual properties of parietal cortex neurons. Validation of RF mapping procedure The descriptive parameters were obtained from interpolated data matrices corresponding to averaged discharge frequencies. However, standard errors around these mean values were available only for those frequencies corresponding to experimental data points. In a first step, we addressed the question of the number of matrix cycles necessary to produce consistent estimates of RF parameters, with respect to the response variability of LIP neurons. This was necessary because the number of matrix cycles is the major limiting factor of the present experimental design in terms of time cost of a single cycle. An analysis was carried out on the total set of reconstructed RFs from the three studied hemispheres (171 cells) to establish 95% confidence intervals for different RF parameters and for each stimulation cycle. For the entire cell sample, normalized confidence intervals (the parameter r, expressed as a percentage) were Fig. 2A–C Offline RF reconstruction. A For each set of flashes presented at a given location of the mapping grid, trials are cumulated. The three panels represent 200 ms of spike trains and the corresponding poststimulus histogram (PSTH) relative to stimulus onset for three different locations of the mapping grid (calibration bar on the left of the histogram = 250 spikes/s, 4-ms bins). B To compute the RF map, the highest PSTHs are displayed offline, and a time window (inverted triangles above histograms) is defined manually. Average discharge rate throughout this window is calculated for each stimulus location. This average discharge rate (and standard error) is displayed as a function of stimulus position in a cross-section of the grid running 4° below the horizontal meridian. Solid symbols correspond to the values obtained from the left panel histograms. C The discharge rate of the neuron is also displayed as a function of stimulus position represented in color code. x- and y-coordinates are indicated in degrees. White cross indicates the location of the fixation point, i.e., the intersection of the retinal meridians. White line indicates the level at which the section of the middle panel is taken. White dots indicate the mapping positions for which discharge rate was significantly above the spontaneous discharge rate of the neuron

132 Fig. 3A–C RF parameter estimates as a function of the number of stimulation cycles for a representative subset of nine neurons. Asymptotic values are well established after six stimulation cycles. A Maximum spiking frequency in spikes per second. B Width in degrees. C X-position of the spiking peak in degrees

calculated for maximum discharge rate and for RF width. The r-value for maximum discharge rate has a mean of 16.6% and an SD of 9.64%. Estimates of width are somewhat more reliable (mean = 9.57% and SD = 5.85%). These values set the limit of our quantification method of the neuronal response tuning at the single cell and population level, as well as for qualitative comparisons between two successive behavioral mapping conditions. Confidence intervals for the horizontal and vertical peaks and center of mass did not differ significantly; thus in subsequent analyses, only radial eccentricity of the peak and center of mass were considered. Eccentricity of peak location could be determined with a 95% confidence interval of 2.5° for more than half of the cells (mean = 3.15°, SD = 2.79°). Even better estimates were obtained when center of mass location was smaller (mean = 1.66°, SD = 1.08°). Confidence intervals for discharge rate, size and eccentricity were calculated on the total number of stimulation cycles obtained for each RF mapping, but an analysis carried out on all cells indicated that these parameters show almost no variation after six stimulation cycles (see examples of nine representative cells in Fig. 3). Characteristics of the population of neurons studied Visual and oculomotor-related activity During electrode penetrations visual and eye-movement related responses of the neurons were used to estimate the recording location in the intraparietal sulcus. LIP neurons are known to carry multiple signals (Colby et al. 1996). In order to establish the comparability of the population of LIP neurons studied here with previously reported data, we characterized each neuron in terms of its activity during the memory-guided saccade paradigm. Four or eight standard saccade directions were used (7.5° of eccentricity, uniformly and radially arranged around the fixation point at 90° or 45° intervals, respectively), and the direction eliciting the best responses was selected for analysis. For all directions, the presence of visual activity, delay period and saccade related activity was defined statistically by comparing activity during these

Table 1 Visual and saccadic characteristics of the recorded LIP population (N=313) during a memory-guided saccade task

Visual No visual

Presaccadic

Postsaccadic

Not saccadic

188 8

61 7

47 2

epochs with baseline activity with a t-test using a significance criterion of P