Spatiotemporal Dynamics of Feature-Based Attention Spread: Evidence from Combined Electroencephalographic and Magnetoencephalographic Recordings

Discussion

The present study used simultaneous recordings of ERPs and ERMFs in human observers to investigate the spatiotemporal correlates of feature-based attentional selection at the neural population level. During the task subjects attended to a moving transparent surface in the left visual field and performed a motion discrimination task, while a task-irrelevant second surface located in the opposite visual field moved into different directions. This experimental design permitted to quantify the magnitude of ERPs and ERMFs evoked by the unattended surface while systematically varying the similarity between its motion direction and that of the attended surface. Our current results demonstrate a parametric feature-based attentional modulation of ERP and ERMF amplitudes, as a function of the similarity between the motion directions of the spatially attended and unattended stimuli. The time courses of the ERP and ERMF waveforms indicate that this attentional enhancement starts not before 200 ms poststimulus onset and originates from left middle occipitotemporal cortex, most likely corresponding to area V5/hMT. These findings provide strong support for the feature-similarity gain model by demonstrating that feature-based attention parametrically modulates direction-selective population activity within V5/hMT in a global manner. The timing indicates that the global spread of attention toward spatially unattended locations does not occur immediately after the selection of the attended feature pointing out to the time-consuming nature of this process.

The observed feature-based modulations were localized to the lateral middle occipitotemporal cortex, which corresponds well to area V5/hMT. This region, considered to be the homolog of the well described MT region in non-human primates, is specialized for the processing of motion information (Zeki et al., 1991; Ahlfors et al., 1999). Beyond purely sensory-driven effects, activity within this region can be markedly affected by attention. For example, V5/MT neurons increase their firing rate to attended compared with unattended motion stimuli (Cook and Maunsell, 2002, 2004), which is consistent with findings in human subjects (Corbetta et al., 1990, 1991; O'Craven et al., 1997; Schoenfeld et al., 2007). In addition to this global response enhancement, recent observations indicated that V5/MT activity could also be affected in a direction-selective manner. Feature-based attention modifies the response profile of direction-selective neurons within V5/MT multiplicatively: neurons whose feature preference closely match an attended motion direction increase their firing rate, while the firing of neurons tuned to opposite directions is suppressed (Treue and Martínez Trujillo, 1999; Martinez-Trujillo and Treue, 2004). As a consequence of these findings the ‘feature-similarity gain model’ has been formulated. This model posits that an individual neuron's response depends on the feature similarity between a current behaviorally relevant target and the feature preference of that neuron, resulting in an improved selectivity for the attended feature at the population level. Recently this hypothesis has also been confirmed using fMRI experiments in human observers with stimuli that were presented within the focus of spatial attention (Kamitani and Tong, 2006; Stoppel et al., 2011). Another fMRI study indicated that feature-based attention might also operate in a spatially global manner, i.e., for stimuli occurring at spatially unattended locations (Saenz et al., 2002). However, this study only compared stimuli that moved into the attended versus opposed to the attended direction. More importantly, due to the nature of the fMRI technique no inferences could be made on the timing. The current results extend these findings (Saenz et al., 2002) by demonstrating that electromagnetic population activity (i.e., ERP/ERMF amplitudes) to moving stimuli presented at unattended locations scales parametrically in dependence of their feature-similarity with respect to the attended stimulus.

Although fMRI studies provided a detailed picture on the anatomical structures modulated by feature-based selection, their temporal resolution is too limited to reveal the timing of the underlying attentional modulations. Fine-grained information about the time course of feature-based selection has therefore been determined primarily based on data from noninvasive EEG/MEG recordings in humans. By this means, previous studies indicated that the selection of task-relevant features is initiated in the time range of the N1-component, i.e., between 100 and 180 ms poststimulus onset (Harter and Aine, 1984; Kenemans et al., 1993; Motter, 1994; Anllo-Vento and Hillyard, 1996; Smid et al., 1999; Torriente et al., 1999; Kenemans et al., 2000; Martínez et al., 2001; Beer and Röder, 2004, 2005; Schoenfeld et al., 2007), which in most cases is reflected by a broad negativity over centro-posterior electrodes in the ERP (the so-called selection-negativity; Harter and Aine, 1984; Hillyard and Anllo-Vento, 1998). A common feature of these studies was that the stimulus eliciting the neurophysiological response was located in the attended part of the space.

Thus, the feature-based selection occurred at the spatially attended location. In the present study we observed an enhanced negativity over centro-posterior electrodes in the EEG, whose magnitude parametrically depended on the similarity between the motion directions of the attended and the unattended surfaces (Fig. 3). The corresponding modulations were also observed in the simultaneously recorded ERMF over occipitotemporal sensors. Importantly, the onset latency of these modulations was later (≈200 ms) than previously reported (Hillyard and Münte, 1984; Anllo-Vento and Hillyard, 1996; Karayanidis and Michie, 1996; Lange et al., 1998; Schoenfeld et al., 2007). Given that in the present experiment the motion probes were located in the unattended visual field, this latency difference is likely to reflect the temporal costs underlying the spread of feature-selective modulations toward spatially unattended locations.

However, there are also other explanations that might account for the observed latency difference. One possibility would be that the delay results mainly from time costs related to involuntary attention capture processes triggered by the task-irrelevant stimuli (Egeth and Yantis, 1997). On the other hand the initiation of an involuntary attention shift towards the task-irrelevant stimulus could be regarded as an integral process of the feature-selective attention spread that might be used to determine the most useful direction of propagation.

A second possibility would be that the observed latency difference rather reflects a delay in the general processing of the task-irrelevant stimulus than the time costs of the feature-based attentional spread. However, the processing of task-relevant stimuli within the same feature value (e.g., one color from another) at an attended location is typically enhanced at ∼100–110 ms poststimulus (Anllo-Vento et al., 1998). The processing of a task-irrelevant feature at the attended location is typically delayed by ∼50 ms (Schoenfeld et al., 2003). A general delay due to the task irrelevance per se would therefore be expected to be reflected in the electrophysiological recordings ∼150–160 ms poststimulus, which is different from the timing observed in the present study (∼200 ms). Task relevance is certainly a key factor with regard to the timing differences, but is unlikely to cause the entire delay observed.

Together with other findings from the literature, the current results point to a more general framework of the temporal dynamics of attentional spreading. Within an attended spatial location feature selection can be very fast (∼100 ms poststimulus onset). This is especially the case when an entire feature-dimension can be selected from another one (e.g., attending a stimulus' motion vs its color; Schoenfeld et al., 2007). Additional time costs are observed when attentional selection operates within a single feature-dimension (e.g., selecting one particular motion direction from another; Hillyard and Münte, 1984; Anllo-Vento and Hillyard, 1996; Karayanidis and Michie, 1996; Lange et al., 1998). Depending on the stimuli and task used, attentional selection occurs ∼110–160 ms poststimulus. In the current experiment the probe was located in the spatially unattended visual field. In this case the attentional selection occurred ∼50 ms later (∼200 ms poststimulus onset) indicating that the spread of attention across spatial locations takes ∼50 ms of time. Importantly, not only the spread of attention over space takes time. Studies on object-based attention could show that the attentional spread from an attended to an unattended feature of that same object took ∼40–50 ms. In this case both features were present at the same spatial location that was attended (Schoenfeld et al., 2003). A recent study showed that attention not only spreads across the same objects but also to other objects at spatially unattended locations if they share a task-irrelevant feature of an attended object (Boehler et al., 2011). In this case the attentional boost occurs even later, at ∼270 ms poststimulus. This suggests that the temporal costs for the attention spread from the task-relevant to the task-irrelevant object feature sum up to the cost for spreading from attended to unattended locations. Around 150 ms are needed for the selection of the task-relevant feature of a spatially attended object (Hillyard and Münte, 1984; Anllo-Vento and Hillyard, 1996; Karayanidis and Michie, 1996; Lange et al., 1998). Another 50–60 ms are needed for spreading to the task-irrelevant feature of the same object (Schoenfeld et al., 2003). The following spread toward an unattended spatial location would take another 50–60 ms (result of the current study). The resulting total time would be in the range between 250 and 270 ms, which is well in line with the aforementioned findings (Boehler et al., 2011).

In conclusion, the current results show that feature-based attention is associated with a global processing enhancement of the attended feature (Treue and Martínez Trujillo, 1999; Saenz et al., 2002; Kamitani and Tong, 2006). Neural responses elicited by spatially unattended task-irrelevant probes were modulated as a function of the degree of similarity between their movement direction and the task-relevant movement direction in the attended visual field. This provides strong support for the “feature similarity gain model” at the level of integrated population responses (Treue and Martínez Trujillo, 1999; Stoppel et al., 2011). Importantly, the global spread of feature-based attention does not occur suddenly following feature selection but rather appears to be a dynamic time-consuming process.