Abstract

Attention serves to select objects from often complex scenes for enhanced processing and perception. In particular, the perception of shape depends critically on attention for integrating the various parts of the selected object into a coherent representation of object shape. To study whether oscillatory neuronal synchrony may serve as a mechanism of attention in shape perception, we introduced a novel shape-tracking task requiring sustained attention to a morphing shape. Attention was found to strongly increase oscillatory currents underlying the recorded field potentials in the γ-frequency range, thus indicating enhanced neuronal synchrony within the population of V4 neurons representing the attended stimulus. Errors indicating a misdirection of attention to the distracter instead of the target were preceded by a corresponding shift of oscillatory activity from the target's neuronal representation to that of the distracter. No such effect was observed for errors unrelated to attention. Modulations of the attention-dependent enhancement of oscillatory activity occurred in correspondence with changing attentional demands during the course of a trial. The specificity of the effect of attentional errors together with the close coupling between attentional demand and oscillatory activity support the hypothesis that oscillatory neuronal synchrony serves as a mechanism of attention.

Introduction

Psychophysical investigations have shown that selective attention is an essential prerequisite for shape perception: subjects fail to report the shape of non-attended objects while their individual features such as color or texture may well be recognized (Rock and Gutman, 1981; Rock et al., 1992). In line with these observations, feature integration is known to be a necessary step in shape perception, and to depend on attention (Treisman, 1995, 1998; Wolfe and Bennett, 1997). Different neuronal mechanisms are expected to increase the impact of neuronal responses to attended stimuli and have been proposed to subserve attention. An increase of firing rate for neurons representing the attended stimulus and/or a decrease for non-attended stimuli has been supported by a large number of experimental (Moran and Desimone, 1985; Motter, 1993; Treue and Maunsell, 1996; Luck et al., 1997; McAdams and Maunsell, 1999; Reynolds et al., 1999) and theoretical (Olshausen et al., 1993; Hahnloser et al., 1999; Nakahara et al., 2001) studies. An alternative mechanism to enhance common impact is synchronization of neuronal responses to an attended stimulus (Niebur and Koch, 1994; Singer, 1999; Steinmetz et al., 2000; Fries et al., 2001; Keil et al., 2001) that leads to more effective dendritic summation of postsynaptic currents and increases the signal to noise ratio (Segev and Rall, 1998; Usrey et al., 2000; Azouz and Gray, 2003). Precise neuronal synchronization has also been proposed to be particularly instrumental in conjoining neuronal responses related to the same stimulus and to integrate them into a functionally coherent, temporally defined neuronal representation (Singer and Gray, 1995; Kreiter and Singer, 1996a; Singer, 1999; Freiwald et al., 2001). Thus, for a shape-perception task a particularly strong and reliable effect of selective attention on neuronal synchrony is predicted. Consequently, correct behavioral performance is expected to depend on adequate synchronization of the neuronal population processing the attended stimulus.

To test these predictions, we developed a new and demanding shape-tracking task that requires tracking the shape of a morphing object while ignoring a similar distracter. Because of the strict dependence of shape perception on attention (Rock and Gutman, 1981; Rock et al., 1992), successful performance requires allocating selective attention to the behaviorally relevant object. Recordings were taken from area V4, which is known to contain neurons selective for a large variety of form-related features (Desimone and Schein, 1987; Gallant et al., 1993; Tanaka, 1993; Pasupathy and Connor, 1999, 2001, 2002), indicating its participation in a distributed representation of shape (Pasupathy and Connor, 2002). We used an array of epidural electrodes to record field potentials from an extended part of area V4. Field potentials represent the sum of synchronous currents caused by neuronal activity in a few local columns underneath an electrode (Elul, 1972; Nunez, 1995). The signals in the γ-frequency range reflect the strength of precise synchronization of neurons within the investigated local population (Eckhorn et al., 1993; Singer and Gray, 1995; Livingstone, 1996; Herculano-Houzel et al., 1999; Logothetis et al., 2001; Rols et al., 2001; Siegel and König, 2003). Recording field potentials with an implanted epidural array provides (i) a measure of synchrony for an entire population of neurons representing the stimulus and (ii) stable and comparable data over multiple recording sessions that allows for compilation of rare, attention-specific error trials.

Our investigations revealed a remarkably strong increase of oscillatory activity in the γ-frequency range, indicating a corresponding increase of local synchrony for attended as compared with non-attended stimuli. In particular, the highly specific modulation preceding the occurrence of attention-related errors indicated a functional role of neuronal synchronization as a mechanism of attention.

Materials and Methods

Behavioral Training and Visual Stimulation

Two male rhesus monkeys (Macaca mulatta) were trained to a novel shape-tracking task. For training and recording sessions the monkeys sat in a primate chair with the head restrained. Visual stimuli were presented with a frame rate of 100 Hz on a 21″ CRT screen 81 cm in front of the monkeys' eyes.

Each monkey started a trial (see Fig. 1 and movie in supplemental material) by pressing a lever after the appearance of a central fixation point on the screen. Some 650 ms later, two different stimuli with complex shape appeared at fixed positions in the left and right visual hemifield (S1). To direct attention to the location of the behaviorally relevant stimulus sequence (i.e. the target), it was cued within the first 200 ms of stimulus presentation by green coloring of the stimulus that faded to white within the subsequent 400 ms. After static presentation for 1300 ms, target and distracter underwent a morphing process. During this process each stimulus morphed through a sequence of different shapes, referred to as S2–S6 according to their sequential position. For each trial the required number of different shapes was randomly selected from a set of 10. The respective initial shapes of the target or distracter sequence reappeared at different, randomly selected points in time (S3–S6). Reoccurrence of the initial shape in the morphing sequence of the initially cued stimulus required the monkey to release the lever to be rewarded with a drop of fruit juice. A reappearance of the initial shape in the morphing sequence of the distracter had to be ignored. If the monkeys broke fixation (rectangular fixation window of ±0.75°), or responded too early or too late, the trial was aborted without reward.

Each stimulus shape was defined by 12 non-visible points interconnected by a smooth Bézier curve, which was 0.3° wide. During a morphing cycle (see Fig. 1), each of the shape-defining points was moved along a straight trajectory from its former position (PS) to its position in the following shape (PE). The position (P) on this trajectory was updated with each frame in 10 ms intervals: 

\[P{=}(P_{\mathrm{E}}{-}P_{\mathrm{S}})R_{\mathrm{Pos}}{+}P_{\mathrm{S}}\]
RPos is the relative position along the trajectory ranging between 0 and 1: 
\[R_{\mathrm{Pos}}{=}(((2t{-}\mathrm{{\phi}})/\mathrm{{\phi}})^{0.5}{+}1)/2{\ }\mathrm{for}{\,}t{\geq}\mathrm{{\phi}}/2\]
 
\[R_{\mathrm{Pos}}{=}({-}((\mathrm{{\phi}}{-}2t)/\mathrm{{\phi}})^{0.5}{+}1)/2{\ }\mathrm{for}{\,}t{<}\mathrm{{\phi}}/2\]
with time t and the morphing cycle's duration ϕ.

Figure 1.

Schematic illustration of the shape-tracking task. A trial started with the monkey pressing a lever after the appearance of the fixation point. After 650 ms, two complex shapes appeared to the left and right side of the fixation point. Transient coloring cued the behaviorally relevant shape. Following an initial static period, both shapes underwent a morphing process, during which the stimuli passed through a sequence of different shapes. As soon as the initial shape was repeated in the cued sequence, the monkey had to release the lever. The upper part of the figure illustrates the continuous morphing from one shape to the next during one morphing cycle while the lower part of the figure depicts the sequence of events over time for the attended stimulus. The solid line beneath the sequence of shapes symbolizes the current shape's increasing distance in shape space from Sn during the first half of a morphing cycle and its decreasing distance to Sn+1 during the second half of the morphing cycle. The shaded areas within the morphing period depict periods during which the monkey had to respond if the target's shape was identical with the one shown initially.

Figure 1.

Schematic illustration of the shape-tracking task. A trial started with the monkey pressing a lever after the appearance of the fixation point. After 650 ms, two complex shapes appeared to the left and right side of the fixation point. Transient coloring cued the behaviorally relevant shape. Following an initial static period, both shapes underwent a morphing process, during which the stimuli passed through a sequence of different shapes. As soon as the initial shape was repeated in the cued sequence, the monkey had to release the lever. The upper part of the figure illustrates the continuous morphing from one shape to the next during one morphing cycle while the lower part of the figure depicts the sequence of events over time for the attended stimulus. The solid line beneath the sequence of shapes symbolizes the current shape's increasing distance in shape space from Sn during the first half of a morphing cycle and its decreasing distance to Sn+1 during the second half of the morphing cycle. The shaded areas within the morphing period depict periods during which the monkey had to respond if the target's shape was identical with the one shown initially.

This results in a continuous change between slower morphing in the beginning and towards the end of each cycle, and faster morphing inbetween. The response window ranged from 400 ms before the end of a morphing cycle to 750 ms after the beginning of the next cycle. Stimuli were centered 0.9° below the horizontal meridian and 2.9° aside the vertical meridian, and covered a region of ∼4° × 4°, with the eccentricity ranging from 0.9 to 5.7° and the center at 3°. For an example of the morphing stimulus, see supplemental material.

For a control task (Fig. 8A) used to isolate possible memory-related activity, both morphing stimuli were substituted by sequences of 500 ms static stimuli separated by 900 ms delay intervals. The initial shape was presented for 1550 ms and the target was cued as described above. The monkeys had to respond to the reappearance of the initial stimulus within 1000 ms. Otherwise the task was identical with the main task described above.

In the main task, as well as in the memory-control task, we presented blocks of 60–90 trials with the target stimulus on the same side to support target cueing. Monkey M was tested in addition with the target position chosen randomly for each trial. Analysis of behavioral performance revealed no difference between the two conditions.

For partial retinotopic mapping of the lower visual field representation of V4 on the gyrus prelunatus, small white squares (0.4° × 0.4°) were flashed at different positions in the visual field while monkeys engaged in a fixation task and kept their gaze within ±0.75° around the fixation point.

Surgical Preparation

The monkeys were implanted with a headpost and a thin gold ring placed between the conjunctiva and the sclera of one eye for measurement of gaze direction using the indirect search coil method (Bour et al., 1984). After completion of the subsequent behavioral training, the monkeys were implanted with an epidural array of platinum-iridium electrodes placed over area V4. Based on maps of the monkey brain (Gattass et al., 1981, 1988; Paxinos et al., 2000), the intended position of the array was determined relative to anatomical landmarks. Stereotactic coordinates of these landmarks were derived from structural magnetic resonance images obtained for each animal from a 4.7 T Bruker Biospec scanner (Ettlingen, Germany). The precise location of the implanted electrode array was estimated postoperatively by the stereotactic coordinates determined during implantation, their comparison with structural magnetic resonance images obtained after implantation and morphological confirmation in one of the monkeys. The localization of the array was further improved and confirmed by the construction of a partial retinotopic map of area V4, based on recordings of γ-band responses to the small test stimuli described above with the implanted electrode array.

Recording

Field potentials were recorded from a chronically implanted array of epidural electrodes (36 in monkey M and 37 in monkey F) covering area V4. The electrode array consisted of a 0.1-mm-thick sheet of silicone (Goodfellow, Bad Nauheim, Germany), in which Teflon-coated platinum–iridium (90Pt/10Ir) wires (diameter 50 μm, Science Products, Hofheim, Germany) were inserted with a regular spacing of 3 mm. The electrode contact was an uninsulated loop (diameter 210–220 μm) positioned parallel to the dura. Two reference electrodes (platinum–iridium wire, 150 μm diameter) were placed frontally. In monkey F a third reference was attached to the rear side of the electrode array (platinum–iridium foil, diameter 4.5 mm, thickness 0.1 mm). Recordings were referenced to the latter electrode in monkey F and a frontal electrode for monkey M. Signals were amplified (×40 000 in monkey F, ×30 000 in monkey M, 1–150 Hz bandwidth) and continuously recorded at a sampling rate of 1 kHz.

All surgical and experimental procedures were performed in accordance with the European Communities Council Directive of November 24,1986 (86/609/EEC) and with the regulations for the welfare of experimental animals issued by the Federal Government of Germany and had been approved by the local authorities.

Data Analysis

After recording, field potentials were high-pass filtered with a digital filter (Butterworth IIR filter, cut-off frequency 0.65 Hz at 3 dB, forward and backward filtering to avoid phase shifts) to eliminate DC offset. Trials containing artifacts were rejected. To suppress the effect of the common reference and to minimize spatial smearing (Nunez et al., 1997), the current source density (CSD) (Gevins, 1984) with unit V/m2 was computed. For each time bin the second spatial derivative of the field potentials was computed with the Laplacian operator (Perrin et al., 1987), using Gaussian radial basis functions (RBF) for interpolation (Moody and Darken, 1989).

For analysis in the time–frequency (TF) domain, the current source density signal of each trial and electrode was convoluted (using a core routine provided by Torrence and Compo, 1998) with complex Morlet's wavelets

\(w(t,f_{0}){\,}{=}{\,}A{\,}\mathrm{exp}({-}t^{2}/2\mathrm{{\sigma}}_{t}^{2}){\,}\mathrm{exp}(2i\mathrm{{\pi}}f_{0}t),\)
with σf = 1/πσt (Kronlandt-Martinet et al., 1987). The wavelets have Gaussian shape both in time (standard deviation σt) and frequency (standard deviation σf around the center frequency f0). Wavelets were normalized so that their total energy 
\[E(f_{0}){=}{{\int}_{{-}{\infty}}^{{\infty}}}w(t,f_{0})\mathrm{d}t\]
was 1 for all values of f0. (Tallon-Baudry et al., 1997). This condition requires the normalization constant A to be (σt√π)−0.5. This results in a flat spectrum estimate of a white noise signal. The central frequency f0 ranged from 5 to 160 Hz according to the scheme of Torrence and Compo (1998), with a ratio f0f = 6. At 10, 60 and 160 Hz, this leads to wavelet durations (2σt) of 194, 32 and 12 ms with a spectral bandwidth (2σf) of 3.3, 19.7 and 52.6 Hz, respectively. The power spectral density (PSD) with unit (V/m2)2/Hz was computed by taking the square of the absolute value of the convolution's result and dividing it by the Nyquist frequency (500 Hz).

To test for evoked, i.e. stimulus-locked, components within the oscillatory activity, the evoked PSD was computed based on the CSD-signal averaged over trials. In the γ-band, the evoked PSD was found to be two orders of magnitude smaller than the induced PSD (the average over the PSD of each trial) and thus had no significant impact on the quantitative results presented. The time course of the maximum evoked PSD between 45 and 103 Hz is shown in Figure 6A.

The spatial distribution of averaged PSD in a TF window spanning two morphing cycles during shape tracking (3.35–6.15 s after trial start, 45–76 Hz) was calculated by interpolating the resulting values in space by means of Gaussian RBFs. To obtain the spatial distribution of the effect of attention, a map of this effect was constructed by subtracting a map of the non-attended condition from a map of the attended condition. All TF plots and mean PSD values were computed for data taken from the site defined by the peak in this difference map.

TF plots of the stimulus response were constructed from the results of the wavelet transform by subtracting the PSD distribution of baseline activity before stimulus onset (always taken between 50 and 600 ms after trial start) and normalizing to this distribution. The time course of the response observed in the γ-frequency range was extracted from these TF plots by taking the peak PSD value for each time bin in the frequency range between 45 and 103 Hz. As a measure of neuronal activity, the evoked CSD was calculated for each electrode and its spatial distribution was estimated by interpolating the values of the difference between the first positive and the subsequent negative peak between 50 and 110 ms after stimulus onset with Gaussian RBFs.

Analysis of Error Trials

Aside from correct responses, different types of errors could be distinguished for the shape-tracking task. A specific attention-related error occurred, when a monkey responded too early with respect to the sequence of the target, but at a point in time where the initial shape of the distracter sequence reappeared (distracter-related false alarm, see Fig. 7A). In contrast to unspecific false alarms, which do not fulfill this condition, and misses, the distracter-related false alarm indicates that the monkey has misdirected attention to the distracter instead of the target before responding. Unspecific errors are thought to be related to a variety of reasons that are not necessarily related to attention. Oscillatory activity associated with different types of errors and correct responses was compared within a TF window spanning frequencies in the γ-band (45–103 Hz) and a 650 ms interval preceding the first response, which is 400–420 ms before completion of a final morphing cycle. This time interval was chosen since the attentional state can be deduced most reliably for intervals directly preceding the behavioral response and because it framed differences between correct and error trials. The PSD for errors observed under conditions indicating distracter-related false alarms was corrected for the fraction expected to be due to unspecific false alarms. The observed probability for the occurrence of distracter-related false alarms (pDisObs) is therefore composed of the true probability for distracter-related false alarms (pDisTrue) and the true probability for unspecific false alarms (pUnspTrue): 

\[p_{\mathrm{DisObs}}{=}p_{\mathrm{DisTrue}}{+}p_{\mathrm{UnspTrue}}{-}(p_{\mathrm{DisTrue}}{\times}p_{\mathrm{UnspTrue}})\]
The true probability of distracter-related false alarms can be estimated since pDisObs is observed and pUnspTrue is approximated by pUnspObs, the observed probability of unspecific false alarms.

Because of the mixed reasons for errors, the average normalized γPSD observed for distracter-related false alarms is a composite of the γPSD associated with true distracter-related and with unspecific false alarms. The γPSD that must have occurred for trials in which the false alarm was truly distracter-related (PSDDisTrue) can be estimated based on pDisObs,pUnspObs and the γPSD observed for the two types of errors (PSDDisObs = γPSD observed for distracter-related false alarms; PSDUnspObs = γPSD observed for unspecific false alarms) by using the following relation: 

\begin{eqnarray*}&&(p_{\mathrm{DisObs}}{\times}\mathrm{PSD_{DisObs}}){=}(p_{\mathrm{DisTrue}}{\times}\mathrm{PSD_{DisTrue}})\\&&{+}(p_{\mathrm{UnspTrue}}{-}(p_{\mathrm{DisTrue}}{\times}p_{\mathrm{UnspTrue}}){\times}\mathrm{PSD_{UnspObs}})\end{eqnarray*}

Statistical Analysis

All statistical analysis was done with data taken from the electrode closest to the site of maximum difference between the attended and non-attended condition (see above). The significance of the difference between the γPSD observed for the attended and the non-attended condition was tested for the average PSD in a TF window spanning two morphing cycles (3.35–6.15 s after trial start) and a frequency range from 45 to 103 Hz using the Mann–Whitney test. To test the effect of behavioral errors on the attentional modulation of the signal, a two-way analysis of variance (ANOVA) with the factors behavioral relevance (target or distracter) and trial result (hit or error) was performed on γPSD values of correct and error trials. These values were first normalized to baseline activity (0.05–0.6 s after trial start) and then subjected to a logarithmic transformation. Normality of distributions and homogeneity of variances were confirmed. The error analysis was confined to responses around the positions S3 and S4 of the shape-tracking task, because only for these positions a sufficient number of error trials were available.

Results

To investigate the effect of selective attention on neuronal synchronization in a demanding perceptual situation, two macaque monkeys were trained to a novel shape-tracking task (Fig. 1, see movie in supplemental material) which requires sustained attention to one of two morphing shapes to identify the reoccurrence of its initial shape. Average performance during recording sessions was estimated over all but the last possible response window in which a response would have always been correct. Disregarding fixation errors, the monkeys performed correct for 76.9% of the trials (monkey M 81.9%, monkey F 71.8%). The response occurred 127 and 118 ms (median values for monkey M and F, respectively) before the end of the morphing cycle leading again to the shape shown initially (S1). Due to the changing speed of morphing during a morphing cycle, the shape at this time was already very similar to the final shape.

Oscillatory Responses in V4

To estimate a measure of overall synchronization within the population of responding V4 neurons, we recorded field potentials with a rectangular grid of epidural electrodes over area V4 and computed the corresponding CSD. Field potentials and corresponding CSD reflect the time course of the spatially weighted sum of synaptic currents caused by action potentials in the local population of neurons (Llinas and Nicholson, 1974) and are essentially measures of total synchronous activity (Elul, 1972; Nunez, 1995). Such signals can occur either time-locked to the stimulus (evoked activity) or without a fixed phase relation to the stimulus (induced activity). The evoked activity in the field potential (EP) and the CSD (EC) is shown in Figure 2. As expected for the continuously morphing stimulus there was a consistent evoked response with stimulus onset in both monkeys for both measures while evoked components during later parts of the trial were small. Analysis of the oscillatory components of the evoked response revealed only negligible power within the γ-band (see below, Fig. 6). Thus, the oscillations in the γ-band described subsequently are of the induced type, i.e. not phase-locked to the stimulus.

Figure 2.

Time course of the averaged evoked potential (A) and the averaged evoked CSD (an analog of the evoked potential) (B) time-locked to stimulus onset. Black lines correspond to signals with the target, gray lines to signals associated with the distracter stimulus. Both measures result in comparable plots. Comparison of the attentional conditions showed no consistent differences.

Figure 2.

Time course of the averaged evoked potential (A) and the averaged evoked CSD (an analog of the evoked potential) (B) time-locked to stimulus onset. Black lines correspond to signals with the target, gray lines to signals associated with the distracter stimulus. Both measures result in comparable plots. Comparison of the attentional conditions showed no consistent differences.

Analysis of the spatial distribution of oscillatory γ-band activity within the para-foveal lower field representation of area V4 revealed a strong and localized response to the stimulus. This is shown by the map of oscillatory activity in Figure 3A. It is based on the power spectral density of the CSD in the γ-band (γPSD) between 45 and 76 Hz computed for a time interval containing two consecutive morphing cycles (3.35–6.15 s after trial start). A distinct spot of enhanced γPSD is confined to a region of ∼7.2 × 4.3 mm (10.6 × 4.6 mm in monkey M) and exceeds baseline levels by a factor of 8.8 (3.1 for monkey M). The spot is absent, if no stimulus is present (Fig. 3D) as expected for a signal representing stimulus-related activity. The circumscribed spatial distribution of increased γPSD induced by the morphing stimulus suggests that it is generated by a local group of responsive neurons in area V4. To compare extent and position of the stimulus in the visual field with the location of enhanced γPSD in area V4, a retinotopic map was constructed for the region of interest. For this purpose, small bright rectangles (0.4° × 0.4°) were flashed at different visual field positions during a fixation task. Based on the spatial location of the maxima of the γPSD, the isoeccentricity lines of the visual field representation were estimated. The area of increased γPSD induced by the morphing stimulus extends from <1° to ∼6° of eccentricity in area V4 (Fig. 3A). This matches well with the eccentricity of the stimulus in the visual field, ranging from 0.8° to 5.8°, with the center at 3°. The elongated rather than circular form of the responding part of V4 reflects the distortion expected for a roughly circular stimulus area in a retinotopic map with an overrepresentation of the central visual field. Furthermore, the spatial position of the peak of the γPSD is close to the peak of the evoked current, the CSD-analog of the evoked potential caused by stimulus onset (Fig. 3E). Thus, the enhancement of γPSD, indicating enhanced oscillatory synchrony, is essentially confined to the region of area V4 that is expected to contain neurons responsive to the stimulus.

Figure 3.

Spatial distribution of PSD in the gamma band during shape tracking. Maps A–D are based on the PSD of the CSD signals in the gamma band (45–76Hz) computed in a time interval containing two consecutive morphing cycles (3.35–6.15 s after trial start) of the shape-tracking task. Dotted white lines indicate isoeccentricity values within the location of expected stimulus representation in area V4. For the attended condition (A) and the non-attended condition (B) a distinct peak appears at the expected position of the retinotopic map in area V4 that is modulated by attention. (C) The spatial distribution of the difference between the attended and the non-attended condition shows the spatial restriction of the modulatory effect of attention to the region of area V4 engaged in the processing of the stimulus. (D) The map of baseline activity before stimulus onset contains no peak in the expected position of area V4. Map E represents the difference between the first positive and the subsequent negative peak of the evoked response of the CSD signals between 50 and 110 ms after the onset of the initial stimulus. All data were taken from monkey F.

Figure 3.

Spatial distribution of PSD in the gamma band during shape tracking. Maps A–D are based on the PSD of the CSD signals in the gamma band (45–76Hz) computed in a time interval containing two consecutive morphing cycles (3.35–6.15 s after trial start) of the shape-tracking task. Dotted white lines indicate isoeccentricity values within the location of expected stimulus representation in area V4. For the attended condition (A) and the non-attended condition (B) a distinct peak appears at the expected position of the retinotopic map in area V4 that is modulated by attention. (C) The spatial distribution of the difference between the attended and the non-attended condition shows the spatial restriction of the modulatory effect of attention to the region of area V4 engaged in the processing of the stimulus. (D) The map of baseline activity before stimulus onset contains no peak in the expected position of area V4. Map E represents the difference between the first positive and the subsequent negative peak of the evoked response of the CSD signals between 50 and 110 ms after the onset of the initial stimulus. All data were taken from monkey F.

The Effect of Changing Selective Attention

If neuronal mechanisms of selective attention employ synchronized oscillatory patterns of activity, the oscillating current induced in a local group of neurons and thus the corresponding PSD should be larger for processing attended stimuli as compared with processing identical, but non-attended stimuli. The comparison of spatial distributions of γPSD showed indeed a much stronger signal for the stimulus serving as a target of attention (Fig. 3A) as compared with the same stimulus serving as a distracter (Fig. 3B). The spatial extent of this difference (Fig. 3C) was confined to the patch in area V4 where the response occurred.

Spectral composition and time course of oscillatory activity are shown in TF plots (Fig. 4) computed for responses at the location of maximum difference between attentional conditions. This site was virtually identical with the location of the maxima. In both conditions, there was an increase of PSD throughout stimulus presentation which was restricted to the γ-band and which changed in magnitude in different segments of a trial. The strength of PSD in the γ-band, observed in response to a target, exceeds the strength obtained for a distracter during the whole trial (Fig. 4C). Averaged over two morphing periods, the γPSD of the response in the range between 45 and 103 Hz increased from the non-attended (distracter) to the attended (target) condition by 43% for monkey F and by 73% for the better performing monkey M. This difference was highly significant in both monkeys (Mann–Whitney U-test, 542/521 and 561/662 trials for the attended/non-attended condition in monkey M and F, respectively, P < 0.0001 for both monkeys). At the same time, the normalized PSD in lower frequency bands either decreased or showed only a comparatively small increase (Figs 4C and 5).

Figure 4.

Time course of oscillatory activity during the shape-tracking task. Time–frequency plots of PSD normalized to baseline activity associated with the target stimulus (A), the distracter stimulus (B) and the difference between both conditions (C). Underneath each plot a schematic time course of the task is shown (for details see Fig. 1). The rectangle depicts the time–frequency window inside which the mean effect of attention was calculated. The γPSD is generally higher for the attended than for the non-attended condition. The attention-dependent increment is not constant over time but modulated with the change from one shape to the next. Note the restriction of the normalized response and also the attention-dependent increase of the response to the γ-band.

Figure 4.

Time course of oscillatory activity during the shape-tracking task. Time–frequency plots of PSD normalized to baseline activity associated with the target stimulus (A), the distracter stimulus (B) and the difference between both conditions (C). Underneath each plot a schematic time course of the task is shown (for details see Fig. 1). The rectangle depicts the time–frequency window inside which the mean effect of attention was calculated. The γPSD is generally higher for the attended than for the non-attended condition. The attention-dependent increment is not constant over time but modulated with the change from one shape to the next. Note the restriction of the normalized response and also the attention-dependent increase of the response to the γ-band.

Figure 5.

Attention-dependent change of normalized PSD in different frequency bands. Each panel shows the change of normalized PSD averaged over the frequency range specified below and within a time interval containing two consecutive morphing cycles (same time interval as the TF window in Fig. 3). Note the strong increment of normalized PSD in the γ-range, whereas in lower frequency bands normalized PSD either decreases or shows only a small increase.

Figure 5.

Attention-dependent change of normalized PSD in different frequency bands. Each panel shows the change of normalized PSD averaged over the frequency range specified below and within a time interval containing two consecutive morphing cycles (same time interval as the TF window in Fig. 3). Note the strong increment of normalized PSD in the γ-range, whereas in lower frequency bands normalized PSD either decreases or shows only a small increase.

The strength of the attention-dependent increase of γPSD varied over time in a manner consistent with changing behavioral demands during the trial. Figure 6 illustrates for both monkeys the time course of the maximum values of normalized γPSD (Fig. 6A) for target and distracter. In addition, the phase-locked component of the γPSD, which represents γ-band components of the target's evoked response, is plotted (dashed-dotted lines). The very small size of this component confirms the induced, i.e. non-phase-locked, character of the observed signals. The attention-specific relative increase of the maximum values of normalized γPSD is shown in Figure 6B. A first peak reflecting an increase of γPSD by more than 100% was observed in both monkeys between 150 and 650 ms (peaking at 330 ms) after stimulus onset. While it is likely that this effect essentially reflects a difference in attention during examination of the target's initial shape, it cannot be excluded that the difference of color between target and distracter during the cueing period may also have contributed to the first peak. After this initial maximum, the attention-dependent increase of γPSD was comparatively low (∼45% in monkey M and ∼25% in monkey F) within the remaining time of the static period. Similarly, there was no strong increase of the effect until the subsequent first morphing cycle turned to its end. During these periods the task did not require much attention. The initial shape is expected to be perceived and memorized within the first few hundred milliseconds of the 1300 ms static presentation interval and shape S2 never required a response since it was unavoidably different from the initial shape. The attentional demand increased again with morphing from shape S2 to S3 and to the subsequent shapes, which were possible repetitions of the initial shape and therefore required proper representation and comparison with the memorized initial shape. Within this interval, starting around the presentation of S3 and lasting to the end of the trial, a much larger effect of attention was observed, leading to a peak increase of the γPSD by more than 100% as compared with the non-attended case in both monkeys. In addition to this overall modulation along the major periods of the whole trial, the time course of γPSD and its difference between target and distracter was characterized by a modulation within each morphing cycle (Fig. 6). Factors that contributed to these systematic variations may include changing attentional demands due to the fixed temporal pattern of morphing cycles, the time course of formation and cessation of neuronal representations of individual shapes, as well as the rhythm of stimulus deformation. In summary, the results show that during the whole trial attention was associated with a strong increase of γPSD, indicating a strong increase of oscillatory synchrony that is restricted to the γ-range in a retinotopically defined region of area V4 that corresponds to the extent of the stimulus in the visual field.

Figure 6.

Time course of γPSD during the shape-tracking task. (A) Solid and dotted lines show the time course of the peak normalized PSD in the γ-range for the attended and non-attended condition, respectively. Dashed–dotted lines depict the time course of γPSD for the evoked current, i.e. the stimulus-locked component of the signal. The plot shows that during the response essentially no stimulus-locked γ-band activity could be observed. (B) Time course showing the increment of the γPSD from the non-attended to the attended condition. Underneath each plot a schematic time course of the task is shown (for details see Fig. 1). The panels show that γPSD is generally higher for the attended than for the non-attended condition. This attention-dependent increase is not constant over time but modulated with changing attentional demands in different periods of a trial. In addition, the γPSD is modulated with the transition from one shape to the next.

Figure 6.

Time course of γPSD during the shape-tracking task. (A) Solid and dotted lines show the time course of the peak normalized PSD in the γ-range for the attended and non-attended condition, respectively. Dashed–dotted lines depict the time course of γPSD for the evoked current, i.e. the stimulus-locked component of the signal. The plot shows that during the response essentially no stimulus-locked γ-band activity could be observed. (B) Time course showing the increment of the γPSD from the non-attended to the attended condition. Underneath each plot a schematic time course of the task is shown (for details see Fig. 1). The panels show that γPSD is generally higher for the attended than for the non-attended condition. This attention-dependent increase is not constant over time but modulated with changing attentional demands in different periods of a trial. In addition, the γPSD is modulated with the transition from one shape to the next.

Neural Correlates of Behavioral Errors

The hypothesis that oscillatory synchrony serves as a mechanism of attention predicts that a misdirection of attention should be associated with a corresponding relocation of the focus of synchronous oscillations. Behavioral errors, which indicate such a misdirection of attention, are therefore expected to be preceded by periods of decreased oscillatory γ-band activity for neurons representing the erroneously ignored stimulus. Conversely, oscillatory γ-band activity for the distracter that was mistakenly attended should have been enhanced. For other behavioral errors, which are not related to attention, no such changes are expected. In the following, we show that our results confirm these predictions.

In the present paradigm, behavioral errors appear either as misses when responses would have been required or as false alarms when the monkey responded even though it should not have done so. Within the false alarms two groups are distinguished: a distracter-related false alarm (Fig. 7A) is present if the monkey responded to the repetition of the distracter's initial shape within its sequence of morphing shapes. This error suggests a misdirection of selective attention to the distracter. Other false alarms, which do not coincide with the reoccurrence of the distracter's initial shape, are called unspecific false alarms (Fig. 7B). They are thought to be related to reasons that are not necessarily related to attention, like a failure in encoding the initial shape or errors in the organization of the motor response. Distracter-related false alarms have been found to occur with a probability 2.15 times higher than unspecific false alarms, suggesting that more than half of these errors are due to a misdirection of selective attention while the rest are not expected to differ from unspecific false alarms. For morphing cycles leading to S3 and S4, a sufficient number of error trials were available to allow for statistical analysis of the relation between error type and strength of γPSD.

Figure 7.

Strength of γ-band oscillations for different behavioral conditions. (A) Schematic of a distracter-related false alarm. For this type of error the monkeys respond within the shaded time interval while a repetition of the initial shape (RIS) occurs in the distracter sequence. This error suggests a specific redirection of selective attention from target to distracter. (B) Schematic of an unspecific false alarm. For this type of error neither in the target nor in the distracter sequence was the initial shape repeating when the error happened. (C) Strength of γ-band oscillations in the neuronal responses associated with correct trials, distracter-related false alarms, unspecific false alarms, and misses for target stimuli (shaded bars) and distracter stimuli (white bars). All values represent the γPSD of the response within a time window including 650 ms before the end of trials, normalized to the level of baseline activity. Values for distracter-related false alarms were corrected for the probability of unspecific false alarms (for details see Material and Methods). Note the specific reduction of γPSD for targets and the corresponding increase for distracters, if a distracter-related false alarm occurred.

Figure 7.

Strength of γ-band oscillations for different behavioral conditions. (A) Schematic of a distracter-related false alarm. For this type of error the monkeys respond within the shaded time interval while a repetition of the initial shape (RIS) occurs in the distracter sequence. This error suggests a specific redirection of selective attention from target to distracter. (B) Schematic of an unspecific false alarm. For this type of error neither in the target nor in the distracter sequence was the initial shape repeating when the error happened. (C) Strength of γ-band oscillations in the neuronal responses associated with correct trials, distracter-related false alarms, unspecific false alarms, and misses for target stimuli (shaded bars) and distracter stimuli (white bars). All values represent the γPSD of the response within a time window including 650 ms before the end of trials, normalized to the level of baseline activity. Values for distracter-related false alarms were corrected for the probability of unspecific false alarms (for details see Material and Methods). Note the specific reduction of γPSD for targets and the corresponding increase for distracters, if a distracter-related false alarm occurred.

Significant differences in strength of the stimulus-induced increments of γPSD occurred between distracter-related false alarms and correct responses. The differences were observed predominantly in the last morphing cycle of a trial during which the initial shape of the distracter or the target reoccurred. They were analyzed in a 650 ms time interval preceding the earliest responses, which occurred 400–420 ms before the completion of this final morphing cycle. As expected for correct trials, within this time interval, the γPSD was higher for attended stimuli than for distracter stimuli (Fig. 7C). In contrast, for distracter-related errors the normalized γPSD decreased for the target that should have been attended. Conversely, it increased for the distracter, which should have been ignored, but has been attended as indicated by the monkey's behavior. Two-way ANOVAs confirmed a highly significant dependence between the normalized γPSD and the interaction of the factors behavioral relevance (target or distracter) and trial result (hit or distracter-related false alarm) in both monkeys for both possible positions of matching shapes S3 and S4 [monkey F, S3: F(3,1595) = 146.68, P < 0.0001; S4: F(3,1548) = 12.85, P < 0.001; monkey M, S3: F(3,1127) = 27.55, P < 0.0001; S4: F(3,1133) = 6.13, P = 0.013]. This result is in line with the expectation that a supposed mechanism of attention should be less active for the representation of stimuli which the animal fails to attend and more for those which it has erroneously attended.

Given this highly significant effect for erroneous redirections of attention, we wanted to test the more advanced prediction of a quantitative reversal of γPSD that would indicate a reversal of oscillatory synchrony between target and distracter for distracter-related false alarms. The idealized expectation would be that γPSD rises for the distracter to the level reached by a target in correct trials. Conversely, it should drop for the target down to what is expected for a distracter in a correct trial. For unspecific errors that do not reflect failures of attention, γPSD associated with the target and the distracter should stay similar to those in correct trials. How does this idealized expectation of a complete reversal of γPSD between target and distracter compare with the observed data? Errors during the stimulus condition allowing for distracter-related false alarms do not always reflect such specific misdirections of attention since unspecific errors are also expected to happen at this time (see above). Therefore, the normalized γPSD observed in this condition was corrected for contributions from unspecific false alarms (see Materials and Methods). Figure 7C shows this corrected, normalized γPSD computed for truly distracter-related false alarms and the normalized γPSD observed for correct trials, trials with misses and trials with unspecific false alarms. For correct responses, the strength of the synchronous oscillatory activity was on average 81.6% higher for the target (gray bars) than for the distracter condition (white bars). In contrast, errors based on a misdirection of attention were preceded by an almost reverse pattern of the amount of oscillatory synchronous activity. For these truly distracter-related false alarms, the erroneous inattention to the target was associated with a reduction of 75.1% of the attention-dependent increase observed for targets as compared with distracters in correct trials. Conversely, the γPSD increased by 66.2% of the same span for responses to an erroneously attended distracter. No such effect was observed for errors that are not primarily based on a failure of attention. The normalized γPSD of targets as well as distracters are much less changed for unspecific false alarms and misses as compared with correct trials. Thus, there is a clear correlation between changes of the strength of oscillatory γ-band activity and changes of the direction of attention indicated by characteristic behavioral errors.

Control for Memory-related Activity

The behavioral paradigm used in this study requires attention to select and enhance the representation of a changing stimulus with complex shape, which needs to be compared with an initial shape presented at the beginning of a trial. Thus, in addition to selective attention, working memory is also expected to occur preferentially for the attended stimulus sequence. We tested whether at least part of the observed task dependent change of the PSD in the γ-band in area V4 was related to short-term memory. For this purpose, the continuous morphing from one shape to the next was substituted by a sequence of static stimuli flashed for 500 ms and separated by blank delay periods of 900 ms (Fig. 8A), thereby isolating memory-related activity during delay periods. Again, the monkeys were required to respond to the repetition of the target's initial shape.

Figure 8.

Time course of γPSD during the control task. (A) Schematic illustration of the control task. In this modification of the task the continuous morphing from one shape to the next was substituted by a sequence of static shapes presented for 500 ms and separated by delay periods of 900 ms. This control task served to test for memory-related changes of γPSD, which should appear in the delay periods, isolated from stimulus-related activity. (B, C) Conventions as in Figure 6A. A difference between the attended and the non-attended condition during presentation of the stimulus similar to that for the shape-tracking task (see Fig. 6) was seen. In the delay periods, the γPSD decays almost to baseline levels without discernible differences between the two behavioral conditions, suggesting that the effect during stimulus presentation cannot be explained by memory-related activity but reflects attentional modulations. Again the stimulus-locked component of the γPSD is negligible. Plots in (B) and (C) correspond to monkeys M and F, respectively.

Figure 8.

Time course of γPSD during the control task. (A) Schematic illustration of the control task. In this modification of the task the continuous morphing from one shape to the next was substituted by a sequence of static shapes presented for 500 ms and separated by delay periods of 900 ms. This control task served to test for memory-related changes of γPSD, which should appear in the delay periods, isolated from stimulus-related activity. (B, C) Conventions as in Figure 6A. A difference between the attended and the non-attended condition during presentation of the stimulus similar to that for the shape-tracking task (see Fig. 6) was seen. In the delay periods, the γPSD decays almost to baseline levels without discernible differences between the two behavioral conditions, suggesting that the effect during stimulus presentation cannot be explained by memory-related activity but reflects attentional modulations. Again the stimulus-locked component of the γPSD is negligible. Plots in (B) and (C) correspond to monkeys M and F, respectively.

During the presentation of the stimuli γPSD was somewhat smaller but comparable to the morphing task and larger for the attended than the non-attended condition (Fig. 8B,C). Within the tonic part of the response, which can be compared best with the observations for the morphing stimuli, there was an increase from the non-attended to the attended condition of 46.6 and 30.2% for monkeys M and F, respectively (300 ms window beginning 250 ms after stimulus onset). During the delay periods, γPSD decays nearly to baseline levels, with almost no difference between the attended and non-attended conditions. Therefore, γPSD that may depend on memory processes is negligible and cannot account for the large difference observed between the two attentional conditions during continuous stimulus presentation. This suggests that the large changes of γPSD observed in both tasks depend essentially on the direction of selective attention and cannot be explained by processes, which have to keep information throughout the trial.

To address the question of whether there are indications for a change in total neuronal activity due to different states of attention, we computed the evoked potential and its analog, the evoked CSD caused by a stimulus onset (averaged over data for S3 and S4) during the trial. The results obtained for both measures are presented in Figure 9 for both animals. For the major part of the time course of both measures there were no consistent differences between attentional conditions. To the end, a moderate difference in favor of the non-attended stimulus appeared that was restricted to a short period of ∼100 ms, peaking ∼250 ms after stimulus onset. Another, smaller difference also in favor of the distracter stimulus appears ∼350 ms after stimulus onset, lasting for ∼100 ms. Attention-dependent changes in firing rate would be expected to occur earlier, to be more persistent and in favor of the attended stimulus. Therefore, this late and temporally restricted difference provides no evidence for a substantial attention-dependent enhancement of neuronal activity throughout the response.

Figure 9.

Time course of the evoked potential (A) and the evoked CSD (an analog of the evoked potential based on the CSD) (B). Black lines correspond to signals caused by the target, gray lines to signals caused by the distracter stimulus at stimulus positions S3 and S4. Arrows correspond to on- and offset of the stimulus. Both measures result in comparable plots. For most parts of the time course there was no consistent difference between attentional conditions in both measures. To the end, a moderate difference in favor of the distracter stimulus appears that is restricted to a short period of ∼100 ms, peaking ∼250 ms after stimulus onset. Another, smaller difference also in favor of the distracter stimulus appears 350 ms after stimulus onset, lasting for ∼100 ms.

Figure 9.

Time course of the evoked potential (A) and the evoked CSD (an analog of the evoked potential based on the CSD) (B). Black lines correspond to signals caused by the target, gray lines to signals caused by the distracter stimulus at stimulus positions S3 and S4. Arrows correspond to on- and offset of the stimulus. Both measures result in comparable plots. For most parts of the time course there was no consistent difference between attentional conditions in both measures. To the end, a moderate difference in favor of the distracter stimulus appears that is restricted to a short period of ∼100 ms, peaking ∼250 ms after stimulus onset. Another, smaller difference also in favor of the distracter stimulus appears 350 ms after stimulus onset, lasting for ∼100 ms.

Discussion

Our novel shape-tracking task revealed a strong correlation between attention drawn to a stimulus and the strength of oscillatory activity in the local population of neurons responding in area V4. This attention-dependent increase reached more than 100%, was present throughout the trial and was confined to the γ-band. Behavioral errors indicating selection and further processing of the distracter instead of the target were associated with a considerable redistribution of oscillatory activity patterns in favor of the distracters' representation. Much less change was seen for errors not necessarily associated with a failure of attention. Further evidence for a close functional relationship between oscillatory γ-band activity and selective attention is the enhanced γPSD during trial periods requiring more attention as compared with periods with less attentional demand.

Modulation of Oscillatory Activity by Selective Attention

The strong dependence of oscillatory activity on the behavioral relevance of the stimulus indicates that the shape-tracking task effectively challenged cognitive mechanisms that employ oscillatory activity. Since the task required keeping the target's initial shape in memory, the observed effect may have been related to memory as well as to attention. However, if the considerable difference of γPSD between both behavioral conditions during stimulus presentation would have reflected a change of activity patterns of neurons serving working memory for the behaviorally relevant stimulus, a difference of comparable size should have occurred during the delay period, too. Since the increase of γPSD during delay periods is orders of magnitudes smaller than the difference to be explained during stimulus presentation, this main effect cannot reflect activity patterns of neurons keeping information in memory throughout the trial. This conclusion is further supported by previous investigations which did not find much evidence for substantial delay activity in area V4 (Chelazzi et al., 2001). Furthermore, field potential recordings in monkey temporal cortex revealed memory-related synchronous oscillatory activity for the β- instead of the γ-band (Tallon-Baudry et al., 2004). We therefore conclude, that the modulation of oscillatory activity in the γ-band is caused by attention selecting the sensory representation of the stimulus in area V4. This does not exclude the possibility that attentional selection of stimulus representations is required to enable interactions with memory processes in addition to its crucial role in form perception.

Behavioral Evidence

The hypothesis of a synchronization mechanism of attention predicts that a misdirection of attention should be associated with a failure of the respective neuronal population to synchronize its activity and hence a reduction of γ-band oscillations. In the shape-tracking task the distracter-related false alarm indicated the execution of a correct matching operation, but this occurred for the distracter instead of the target. For this failure to attend and process the instructed stimulus, γPSD stayed rather low for the erroneously ignored target, while it rose for the attended distracter close to the strength expected for a correctly attended target. The pattern of results shows that this reversing effect was not only statistically significant but astonishingly complete, reaching more than two-thirds of the idealized expectation for a complete reversal of synchronization strength between target and distracter. Given that reliability and strength of attention in error trials are not expected to be as good as in correct trials, the observed reversal was remarkably strong and indicates a close relation between selective attention and coherent neuronal oscillations in the γ-band. This conclusion is further supported by the specificity of the diametrical effect for errors of attention. Other errors than distracter-related false alarms may reflect e.g. difficulties to recall or compare the initial shape, or to organize the motor response. Confirming the prediction for errors not reflecting failures of attention, the strength of γPSD for these errors remained close to what was found in correct trials. Slight deviations are most likely due to second-order effects, like reduced attention after difficulties to perceive or remember an initial shape, a general decrease of the level of focused attention or a short distraction by external events.

In addition to the specific modulation of γ-band oscillations during attentional errors, the modulation of its time course can be related to variations of attentional demands. Periods during trials that systematically lack behaviorally relevant events are expected to go along with reduced attention (Ghose and Maunsell, 2002). Such a period roughly comprises the second half of initial shape presentation and the first morphing cycle in which no repetition of the initial stimulus can occur. Well in line with a functional role for coherent oscillatory activity in attention, the oscillatory signal was less enhanced for the target during this part of the task that requires less attention. Taken together, the strength of γ-band oscillations is closely related to different states of attention, and it fulfills the predictions for a mechanism of attention in surprising detail and is highly consistent for both animals.

Changing γPSD Implies Changing Neuronal Synchrony

The observed increase of oscillatory activity in the γ-band could in principle be due to two different mechanisms: first, it could result from enhancing the total synaptic activity within the population of neurons without changing the probability of synaptic events to lock to the temporal pattern of synchronized oscillatory population activity; or second, it could result from enhancing their synchronization to the common, synchronous oscillatory activity pattern. For the following reasons, the observed strong attention-dependent increase of the γPSD can be explained best by an enhancement of precise synchronization within the local population of neurons: (i) the summed extracellular current of a population of neurons determines the amplitude of the field potential and grows almost linearly with the number of neurons contributing synchronous activity, whereas the summed currents of non-synchronized activity changes only in proportion with the square root of the number of contributing neurons. Given the large number of neurons involved, non-synchronized activity therefore provides only a negligible contribution to the field potential as compared with synchronized activity (Elul, 1972; Nunez, 1995). Strong changes in the amplitude of the summed current and the associated field potential are therefore unlikely to occur without changes in synchrony of synaptic events. This conclusion is supported by experimental investigations that found enhanced oscillatory field potentials to be associated with increased synchronization between the contributing neurons (Mitzdorf, 1987; Eckhorn et al., 1989; Gray and Singer, 1989; Murthy and Fetz, 1992; König et al., 1995b; Siegel and König, 2003). A direct dependence between neuronal synchronization and the strength of γ-band activity recorded at the surface of cat visual cortex was demonstrated by Herculano-Houzel et al. (1999). In addition, Fries et al. (2001) observed for a color-change detection paradigm an attention-dependent increase of the local field potential together with an enhanced spike-field coherence. (ii) An increase of the average firing rate without changing the probability of spikes to lock to the pattern of synchronized population activity would at best result in an increase of oscillatory activity proportional to the rate change. However, attention-dependent rate changes in area V4 are small or absent if stimuli do not occupy the same receptive fields (Moran and Desimone, 1985; Luck et al., 1997; Desimone, 1998; Reynolds et al., 1999) and share the same feature domain (McAdams and Maunsell, 1999, 2000) as in the present study. (iii) A general increase of firing rate would raise oscillatory signals in all frequency bands (Shah et al., 2004), whereas the strong increment we observed in the γ-band was instead associated with small changes and even decrements of oscillatory signals in lower frequency ranges. These observed decrements are likely to reflect the redistribution of comparatively large amounts of activity from non- or weakly synchronized patterns to well synchronized activity patterns in the γ-frequency range. Such synchronization requires no change of firing rates since individual neurons do not need to contribute to each cycle of synchronous population activity. (iv) The slightly reduced evoked current in the attended as compared with the non-attended condition does not provide evidence for an enhanced firing rate, which could explain the observed increase of γ-band activity in response to an attended stimulus. In summary, the results suggest that attention specifically increases the amount of synchronous oscillatory activity in the neuronal population processing the attended stimulus, predominantly by increasing neuronal synchrony.

Functional Significance

Psychophysical investigations suggest that the critical contribution of attention to explicit shape perception is the selection of the target's features and the construction of a structural description (Wolfe and Bennett, 1997). In area V4, the features of an object's shape are represented by the activity of a subset of neurons responding to a rich variety of stimulus properties (Pasupathy and Connor, 1999, 2001). Attention needs first to select this population for further conjoint processing. The modulation of synchrony in cortical networks is a very efficient way to selectively increase the common impact of activity from a specific subpopulation of neurons (Segev and Rall, 1998; Usrey et al., 2000; Azouz and Gray, 2003). Synchronized output from the selected population (e.g. a population of V4 cells) will summate particularly well in target cells (e.g. a population of IT cells) and will therefore be more effective than other, non-synchronized input. Theoretical work (Niebur and Koch, 1994) suggests that attention-dependent synchronization in V4 would thereby contribute to the selective responses of inferotemporal neurons to attended stimuli and to the suppression of responses to non-attended stimuli being present simultaneously within their large receptive fields.

After stimulus selection the construction of a structural description, defining the overall shape of the target, requires specification of the relations between its features (Wolfe and Bennett, 1997). It has been demonstrated that relations between feature selective neurons can be expressed by the synchrony of their responses (Engel et al., 1991a,b; Freiwald et al., 1995; König et al., 1995a; Kreiter and Singer, 1996b). Thus, the putative neuronal mechanisms for attentional selection could also serve to represent the structural relations defining overall shape.

In summary, the results show that oscillatory synchronization exhibits important properties expected for a mechanism of attention: it increases strongly within the neuronal representation of the attended object whereas it is weaker elsewhere, and it changes in close correspondence with patterns of attention-specific behavioral errors. This indicates a functional role of oscillatory synchronization as a mechanism of attention, which in shape perception may serve both required functions: stimulus selection and construction of structural descriptions for objects with complex shape.

This work was supported by SFB 517 of the Deutsche Forschungsgemeinschaft, Center for Advanced Imaging, the Hanse Institute for Advanced Study (W.A.F.) and by grants from Volkswagenstiftung (S.M. and W.A.F.). The authors thank A. Besuch, K. Thoß and S. Wicker for technical assistance and animal care, and U. Ernst, K. Pawelzik and M. Borisov for fruitful discussions and support to establish data analysis methods. We are grateful to D. Leibfritz, V. Nebe and K. Bockhorst (Institute for Organic Chemistry, University of Bremen) for MRI scans.

References

Azouz R, Gray CM (
2003
) Adaptive coincidence detection and dynamic gain control in visual cortical neurons in vivo.
Neuron
 
37
:
513
–523.
Bour LJ, Van Gisbergen JAM, Bruijns J, Ottes FP (
1984
) The double magnetic induction method for measuring eye movement–results in monkey and man.
IEEE Trans Biomed Eng
 
31
:
419
–427.
Chelazzi L, Miller EK, Duncan J, Desimone R (
2001
) Responses of neurons in macaque area V4 during memory-guided visual search.
Cereb Cortex
 
11
:
761
–772.
Desimone R (
1998
) Visual attention mediated by biased competition in extrastriate visual cortex.
Philos Trans R Soc Lond B Biol Sci
 
353
:
1245
–1255.
Desimone R, Schein SJ (
1987
) Visual properties of neurons in area V4 of the macaque: sensitivity to stimulus form.
J Neurophysiol
 
57
:
835
–868.
Eckhorn R, Reitboeck HJ, Arndt M, Dicke P (
1989
) A neural network for feature linking via synchronous activity: results from cat visual cortex and from simulations. In: Models of brain function (Cotterill RMJ, ed.), pp. 255–272. Cambridge: Cambridge University Press.
Eckhorn R, Frien A, Bauer R, Woelbern T, Kehr H (
1993
) High frequency (60–90 Hz) oscillations in primary visual cortex of awake monkey.
Neuroreport
 
4
:
243
–246.
Elul R (
1972
) The genesis of the EEG.
Int Rev Neurobiol
 
15
:
227
–272.
Engel AK, König P, Singer W (
1991
a) Direct physiological evidence for scene segmentation by temporal coding.
Proc Natl Acad Sci USA
 
88
:
9136
–9140.
Engel AK, Kreiter AK, König P, Singer W (
1991
b) Synchronization of oscillatory neuronal responses between striate and extrastriate visual cortical areas of the cat.
Proc Natl Acad Sci USA
 
88
:
6048
–6052.
Freiwald WA, Kreiter AK, Singer W (
1995
) Stimulus dependent intercolumnar synchronization of single unit responses in cat area 17.
Neuroreport
 
6
:
2348
–2352.
Freiwald WA, Kreiter AK, Singer W (
2001
) Synchronization and assembly formation in the visual cortex.
Progr Brain Res
 
130
:
111
–140.
Fries P, Reynolds JH, Rorie AE, Desimone R (
2001
) Modulation of oscillatory neuronal synchronization by selective visual attention.
Science
 
291
:
1560
–1563.
Gallant JL, Braun J, Van Essen DC (
1993
) Selectivity for polar, hyperbolic, and cartesian gratings in macaque visual cortex.
Science
 
259
:
100
–103.
Gattass R, Gross CG, Sandell JH (
1981
) Visual topography of v2 in the macaque.
J Comp Neurol
 
201
:
519
–539.
Gattass R, Sousa APB, Gross CG (
1988
) Visuotopic organization and extent of v3 and v4 of the macaque.
J Neurosci
 
8
:
1831
–1845.
Gevins AS (
1984
) Analysis of the electromagnetic signals of the human brain: milestones, obstacles, and goals.
IEEE Trans Biomed Eng
 
31
:
833
–850.
Ghose GM, Maunsell JH (
2002
) Attentional modulation in visual cortex depends on task timing.
Nature
 
419
:
616
–620.
Gray CM, Singer W (
1989
) Stimulus-specific neuronal oscillations in orientation columns of cat visual cortex.
Proc Natl Acad Sci USA
 
86
:
1698
–1702.
Hahnloser R, Douglas RJ, Mahowald M, Hepp K (
1999
) Feedback interactions between neuronal pointers and maps for attentional processing.
Nat Neurosci
 
2
:
746
–753.
Herculano-Houzel S, Munk MH, Neuenschwander S, Singer W (
1999
) Precisely synchronized oscillatory firing patterns require electroencephalographic activation.
J Neurosci
 
19
:
3992
–4010.
Keil A, Gruber T, Müller MM (
2001
) Functional correlates of macroscopic high-frequency brain activity in the human visual system.
Neurosci Biobehav Rev
 
25
:
527
–534.
König P, Engel AK, Roelfsema PR, Singer W (
1995
a) How precise is neuronal synchronization?
Neural Comput
 
7
:
469
–485.
König P, Engel AK, Singer W (
1995
b) Relation between oscillatory activity and long-range synchronization in cat visual cortex.
Proc Natl Acad Sci USA
 
92
:
290
–294.
Kreiter AK, Singer W (
1996
a) On the role of neural synchrony in the primate visual cortex. In: Brain theory (Aertsen A, Braitenberg V, eds), pp. 201–227. Amsterdam: Elsevier.
Kreiter AK, Singer W (
1996
b) Stimulus dependent synchronization of neuronal responses in the visual cortex of the awake macaque monkey.
J Neurosci
 
16
:
2381
–2396.
Kronlandt-Martinet R, Morlet J, Grossmann A (
1987
) Analysis of sound patterns through wavelet transforms.
Int J Pattern Recognit Artif Intell
 
1
:
273
–302.
Livingstone MS (
1996
) Oscillatory firing and interneuronal correlations in squirrel monkey striate cortex.
J Neurophysiol
 
75
:
2467
–2485.
Llinas R, Nicholson C (
1974
) Analysis of field potentials in the central nervous system. In: Handbook of EEG and clinical neurophysiology (Stevens CF, ed.), pp. 62–92. Amsterdam: Elsevier.
Logothetis NK, Pauls J, Augath M, Trinath T, Oeltermann A (
2001
) Neurophysiological investigation of the basis of the fMRI signal.
Nature
 
412
:
150
–157.
Luck SJ, Chelazzi L, Hillyard SA, Desimone R (
1997
) Neural mechanisms of spatial selective attention in areas V1, V2, and V4 of macaque visual cortex.
J Neurophysiol
 
77
:
24
–42.
McAdams CJ, Maunsell JH (
1999
) Effects of attention on orientation-tuning functions of single neurons in macaque cortical area V4.
J Neurosci
 
19
:
431
–441.
McAdams CJ, Maunsell JH (
2000
) Attention to both space and feature modulates neuronal responses in macaque area V4.
J Neurophysiol
 
83
:
1751
–1755.
Mitzdorf U (
1987
) Properties of the evoked potential generators: current source–density analysis of visually evoked potentials in the cat cortex.
Int J Neurosci
 
33
:
33
–59.
Moody JE, Darken C (
1989
) Fast learning in networks of locally-tuned processing units.
Neural Comput
 
1
:
281
–294.
Moran J, Desimone R (
1985
) Selective attention gates visual processing in extrasriate cortex.
Science
 
229
:
782
–784.
Motter BC (
1993
) Focal attention produces spatially selective processing in visual cortical areas V1, V2, and V4 in the presence of competing stimuli.
J Neurophysiol
 
70
:
909
–919.
Murthy VN, Fetz EE (
1992
) Coherent 25- to 35-Hz oscillations in the sensorimotor cortex of awake behaving monkeys.
Proc Natl Acad Sci USA
 
89
:
5670
–5674.
Nakahara H, Wu S, Amari S (
2001
) Attention modulation of neural tuning through peak and base rate.
Neural Comput
 
13
:
2031
–2047.
Niebur E, Koch C (
1994
) A model for the neuronal implementation of selective visual attention based on temporal correlation among neurons.
J Comput Neurosci
 
1–2
:
141
–158.
Nunez PL (
1995
) Quantitative states of neocortex. In: Neocortical dynamics and human EEG rhythms (Nunez PL, ed.), pp. 3–67. New York: Oxford University Press.
Nunez PL, Srinivasan R, Westdorp AF, Wijesinghe RS, Tucker DM, Silberstein RB, Cadusch PJ (
1997
) EEG coherency. I. Statistics, reference electrode, volume conduction, Laplacians, cortical imaging, and interpretation at multiple scales.
Electroencephalogr Clin Neurophysiol
 
103
:
499
–515.
Olshausen BA, Anderson CH, Van Essen DC (
1993
) A neurobiological model of visual attention and invariant pattern recognition based on dynamic routing of information.
J Neurosci
 
13
:
4700
–4719.
Pasupathy A, Connor CE (
1999
) Responses to contour features in macaque area V4.
J Neurophysiol
 
82
:
2490
–2502.
Pasupathy A, Connor CE (
2001
) Shape representation in area V4: position-specific tuning for boundary conformation.
J Neurophysiol
 
86
:
2505
–2519.
Pasupathy A, Connor CE (
2002
) Population coding of shape in area V4.
Nat Neurosci
 
5
:
1332
–1338.
Paxinos G, Huang X-F, Toga AW (
2000
) The rhsesus monkey brain in stereotaxic coordinates. London: Academic Press.
Perrin F, Bertrand O, Pernier J (
1987
) Scalp current density mapping: value and estimation from potential data.
IEEE Trans Biomed Eng
 
34
:
283
–288.
Reynolds JH, Chelazzi L, Desimone R (
1999
) Competitive mechanisms subserve attention in macaque areas V2 and V4.
J Neurosci
 
19
:
1736
–1753.
Rock I, Gutman D (
1981
) The effect of inattention on form perception.
J Exp Psychol Hum Percept Perform
 
7
:
275
–285.
Rock I, Linnett CM, Grant P, Mack A (
1992
) Perception without attention: results of a new method.
Cognit Psychol
 
24
:
502
–534.
Rols G, Tallon-Baudry C, Girard P, Bertrand O, Bullier J (
2001
) Cortical mapping of gamma oscillations in areas V1 and V4 of the macaque monkey.
Vis Neurosci
 
18
:
527
–540.
Segev I, Rall W (
1998
) Excitable dendrites and spines: earlier theoretical insights elucidate recent direct observations.
Trends Neurosci
 
21
:
453
–460.
Shah AS, Bressler SL, Knuth KH, Ding M, Mehta AD, Ulbert I, Schroeder CE (
2004
) Neural dynamics and the fundamental mechanisms of event-related brain potentials.
Cereb Cortex
 
14
:
476
–483.
Siegel M, König P (
2003
) A functional gamma-band defined by stimulus-dependent synchronization in area 18 of awake behaving cats.
J Neurosci
 
23
:
4251
–4260.
Singer W (
1999
) Neuronal synchrony: a versatile code for the definition of relations?
Neuron
 
24
:
49
–65.
Singer W, Gray CM (
1995
) Visual feature integration and the temporal correlation hypothesis.
Annu Rev Neurosci
 
18
:
555
–586.
Steinmetz PN, Roy A, Fitzgerald PJ, Hsiao SS, Johnson KO, Niebur E (
2000
) Attention modulates synchronized neuronal firing in primate somatosensory cortex.
Nature
 
404
:
187
–190.
Tallon-Baudry C, Bertrand O, Delpuech C, Pernier J (
1997
) Oscillatory gamma-band (30–70 Hz) activity induced by a visual search task in humans.
J Neurosci
 
17
:
722
–734.
Tallon-Baudry C, Mandon S, Freiwald WA, Kreiter AK (
2004
) Oscillatory synchrony in the monkey temporal lobe correlates with performance in a visual short-term memory task.
Cereb Cortex
 
14
:
713
–720.
Tanaka K (
1993
) Neuronal mechanisms of object recognition.
Science
 
262
:
685
–688.
Torrence C, Compo GP (
1998
) A practical guide to wavelet analysis.
Bull Am Meteor Soc
 
79
:
61
–78.
Treisman A (
1995
) The perception of features and objects. In: Attention: selection, awareness, and control; a tribute to Donald Broadbent (Baddeley A, Weiskrantz L, eds), pp. 1–35. Oxford: Clarendon Press.
Treisman A (
1998
) Feature binding, attention and object perception.
Philos Trans R Soc Lond B Biol Sci
 
353
:
1295
–1306.
Treue S, Maunsell JH (
1996
) Attentional modulation of visual motion processing in cortical areas MT and MST.
Nature
 
382
:
539
–541.
Usrey WM, Alonso J-M, Reid RC (
2000
) Synaptic interactions between thalamic inputs to simple cells in cat visual cortex.
J Neurosci
 
20
:
561
–567.
Wolfe JM, Bennett SC (
1997
) Preattentive object files: shapeless bundles of basic features.
Vision Res
 
37
:
25
–43.

Author notes

1Brain Research Institute, Center for Emotional and Cognitive Sciences, University of Bremen, D-28334 Bremen, Germany and 2Department of Neurobiology, Harvard Medical School, 220 Longwood Ave, Boston, MA 02115, USA