کنترل توجه: روابط زمانی در داخل شبکه فرونتو-جداری

Abstract Selective attention to particular aspects of incoming sensory information is enabled by a network of neural areas that includes frontal cortex, posterior parietal cortex, and, in the visual domain, visual sensory regions. Although progress has been made in understanding the relative contribution of these different regions to the process of visual attentional selection, primarily through studies using neuroimaging, rather little is known about the temporal relationships between these disparate regions. To examine this, participants viewed two rapid serial visual presentation (RSVP) streams of letters positioned to the left and right of fixation point. Before each run, attention was directed to either the left or the right stream. Occasionally, a digit appeared within the attended stream indicating whether attention was to be maintained within the same stream (‘hold’ condition) or to be shifted to the previously ignored stream (‘shift’ condition). By titrating the temporal parameters of the time taken to shift attention for each participant using a fine-grained psychophysics paradigm, we measured event-related potentials time-locked to the initiation of spatial shifts of attention. The results revealed that shifts of attention were evident earlier in the response recorded over frontal than over parietal electrodes and, importantly, that the early activity over frontal electrodes was associated with a successful shift of attention. We conclude that frontal areas are engaged early for the purpose of executing an attentional shift, likely triggering a cascade through the fronto-parietal network ultimately, resulting in the attentional modulation of sensory events in posterior cortices.

1. Introduction The human visual system sorts through massive amounts of sensory input, which is sampled almost continuously, to arrive at a coherent perception of the scene. This process of searching through the environment for behaviorally relevant information is a ubiquitous component of sensory processing, and it reflects the remarkable ability of the perceptual system to select dynamically information that is compatible with the current goal of the organism. Such perceptual selectivity, referred to as attention, is considered central to cognition, with selected or attended information subsequently receiving preferential or enhanced processing. One of the key elements to understanding attentional selection is to determine what representations are engaged by this process such that they serve as potential candidates for selection. Several possible representations have been identified including those that are space- (Eriksen and Hoffman, 1972, Posner et al., 1980 and Yantis et al., 2002), feature- (Corbetta et al., 1991 and Liu et al., 2003), object- (Corbetta et al., 2005b, Duncan, 1984, Kanwisher and Driver, 1992 and Shomstein and Behrmann, 2006), and/or modality-based (Bushara et al., 1999 and Shomstein and Yantis, 2004), and much recent psychophysical and imaging work has explored the similarities and distinctions between these forms of attentional selection and underlying representations. Of all of these different potential candidate representations from which selection can occur, selection from space-based representations is perhaps the most pervasive and fundamental. Not only do space-based representations reflect topographical organization and layout of early visual cortex, but these representations describe the sensory environment with a unique set of 3D identifiers (i.e., each stimulus in the sensory environment occupies a unique set of spatial coordinates), thereby facilitating location-based selection in a direct and isomorphic manner. This space-based selection is reflected in multiple visual cortical areas as increased activity of neurons representing the attended location (Bisley and Goldberg, 2003, Moran and Desimone, 1985, Saalmann et al., 2007, Somers et al., 1999 and Treue and Maunsell, 1996). The behavioral benefit of this enhanced neural selectivity is that stimuli that appear in attended spatial locations are processed more efficiently and more accurately than stimuli that appear elsewhere (Chawla et al., 1999, Posner, 1980, Treue and Martinez Trujillo, 1999 and Yantis et al., 2002). Despite the growing understanding of attentional selection gleaned from numerous studies, we do not yet have a full understanding of the mechanism that serves as the source to initiate the attentional orienting signal, which, ultimately, results in the neural modulation and behavioral benefit for attended locations. Investigations of this issue have uncovered a network of regions spanning frontal and parietal cortices that triggers a control signal for shifting from one representation to another, be it one that is space-based (Corbetta and Shulman, 2002, Hopfinger et al., 2000 and Serences and Yantis, 2007), feature-based (Greenberg et al., 2010 and Liu et al., 2003), or object-based (Shomstein & Behrmann, 2006). Although there is general consensus concerning regions that are engaged in this attentional shifting process, the relative contributions of the identified frontal and parietal regions have been difficult to characterize. Moreover, some studies have yielded conflicting findings, with several investigations suggesting that the initial spatial re-orienting signal is elicited by the frontal cortex, while others suggest that it is the parietal cortex that initiates the re-orienting signal with frontal cortex following suit (Brignani et al., 2009, Buschman and Miller, 2007, Green and McDonald, 2008 and Simpson et al., 2011). It should be noted that while most investigations of bottom-up attentional capture have convincingly demonstrated that the shifting signal originates over the parietal cortex (Fu et al., 2005, Green et al., 2011, Hopfinger and Ries, 2005, Leblanc et al., 2008 and Ptak et al., 2011), most of the controversy regarding the temporal relationship between the source signals over frontal or parietal cortex has been exclusive to the investigations of top-down attentional control. Part of the difficulty in determining the relative contribution of frontal and parietal regions to the attentional control signal lies in the fact that the neural profiles of these areas observed in response to the initiation of a spatial shift are similar, and, consequently, it is difficult to untangle and disambiguate their independent contributions. For example, both frontal and parietal regions contain topographically mapped priority maps. Single-unit physiology experiments with awake behaving monkeys have found evidence that both the frontal eye fields (FEFs) and the lateral intraparietal area (LIP) contain representations compatible with priority maps (Balan and Gottlieb, 2006, Bisley and Goldberg, 2010, Thompson and Bichot, 2005 and Thompson et al., 2005), usually assumed to be the first step in triggering the shift signal. Concordantly, functional imaging studies in humans have found that corresponding frontal and parietal areas contain topographic representations related to saccade planning and attention (Chiu et al., 2011, Esterman et al., 2009, Greenberg et al., 2010, Greenberg et al., 2012 and Silver and Kastner, 2009), suggesting that these areas in humans may also contain priority maps utilized for the upcoming shift of attention. Moreover, the shift-related signal elicited over frontal and parietal regions is similar with the result that both regions are best described as initiating a transient signal, as measured by both fMRI and ERP. This identified transient signal is interpreted as being responsible for issuing, or initiating, an attention control signal to switch the current spatial focus of attention but a more detailed account of the dynamics of these disparate regions remains elusive (Corbetta et al., 2000, Hopfinger et al., 2000, Rushworth et al., 2001 and Yantis et al., 2002). One possible clue that might assist in uncovering the relative contribution of frontal and parietal areas to the control of spatial attention lies in the ability to identify the relative timing of the corresponding activations in the different regions. Measuring event-related potentials (ERP) provides an ideal opportunity to exploit high temporal-resolution data and to examine the temporal relationship between the initiation of the spatial attentional control signal observed over the frontal and the parietal cortex. The goal of the present investigation was, thus, to elucidate the relative functional roles of two major nodes of the human attentional network, the frontal and parietal cortices, by focusing on the temporal relationships between these important subregions. In order to assess the relative timing of the contribution of frontal and parietal cortices to spatial shifts of attention, we adopted a two-pronged approach. First, we conducted detailed psychophysical investigations to determine the timing thresholds required, on an individual-by-individual basis, to initiate a spatial shift of attention so as to delineate the particular switch signature for each participant. At the same time, we determined a threshold at which each participant was able to detect a target after the switch of attention so that the signal for trials in which the shift was successful could be separated from trials in which it was not. Second, in a separate session, each participant's neural activity was recorded by ERP, while the individual completed the behavioral attentional shifting task with the unique parameters for stimulus presentation adopted from the individual thresholding session. Critically, these attentional switch thresholds ensured that we were indexing the ERP components that occurred before the attentional shift initiation (i.e., source of the attentional shifting signal) as opposed to those components that occur after the execution of the shift. In this way, we can isolate the components that are related to the initiation of a spatial shift of attention, rather than a host of perceptual/post-perceptual processes that are involved in target detection, more generally. Elucidating the neural mechanism of top-down spatial shifts of attention can also prove useful for understanding the behavioral deficits following damage to the parietal lobe. Clinical symptoms of hemispatial neglect have been strongly associated with damage to the parietal lobe including the temporo-parietal junction (TPJ) and the inferior parietal lobule (IPL) as well as connections between frontal and parietal cortices, all regions associated with shifts of spatial attention (Bartolomeo et al., 2007, Corbetta et al., 2005a, Friedrich et al., 1998, Ptak and Schnider, 2010, Shomstein et al., 2010, Thiebaut de Schotten et al., 2005 and Vallar and Perani, 1986).

3. Results Our goal was to investigate the temporal relationships among the key areas of the fronto-parietal network during a task that required the shifting of attention from one spatial location to another (Fig. 1 and Section 2). 3.1. Psychophysics of attentional shifting To examine whether the derived thresholds for target detection (with fixed 800 ms target-to-cue interval) varied as a function of condition or of side, the thresholds obtained for each participant were submitted to an omnibus ANOVA with target type (successful target identification after a shift or after a hold cue) and side (ipsilateral and contralateral to the target) as within-subjects factors. There was no significant difference in target duration as a function of whether the condition was a shift or hold, nor was duration affected as a function of side of space on which the target appeared (F < 1). There was also no interaction between target type and side of space (F < 1). The average threshold across participants for target duration was 134 ms with a subsequent average RT for target letter detection of 295 ms. Following this, we performed a similar analysis with the derived cue-to-target thresholds (with target duration fixed at individual thresholds); these were submitted to an omnibus ANOVA with cue-type (“4”, “2”, i.e., shift, hold) and field (left, right) as within-subject factors ( Fig. 2). The ANOVA revealed a significant main effect of cue-type [F(1,11) = 9.86, p < 0.05], with significantly longer thresholds required for shifts of attention (M = 286 ms) than for maintenance of attention (M = 163 ms). There was also a main effect of the field in which the cue appears [F(1,11) = 4.94, p < 0.05], with thresholds to shift attention from left to right being slightly shorter (M = 215 ms) than from right to left (M = 245 ms). This left/right difference potentially indexes the general view that the right hemisphere plays a greater role in attentional selection than does the left hemisphere ( Corbetta and Shulman, 2002 and Mesulam, 1999). The interaction between cue-type (shift/hold) and side (left/right) did not reach significance (all F < 1). Given these results, it is clear that while there is a small effect of the field in which the cues were presented, the primary difference in thresholds is due to the distinction between shift and hold cues. These results establish that it takes approximately 286 ms, on average, for participants to complete a successful shift of attention between hemifields, and this threshold provides a firm limit on the portion of the ERP response that should be considered as critical to accomplishing that shift rather than engaged in post-shift processing. Behavioral thresholds. The average behavioral thresholds derived for the ... Fig. 2. Behavioral thresholds. The average behavioral thresholds derived for the interval between a cue and target. The two red bars show the thresholds for left and right shift cues while the two blue bars show the same for hold cues. Note the slightly longer thresholds for right than left cues for both shift and hold cues. Note also the larger thresholds for the shift cues. The error bars in this and all other plots indicate the between-subject standard error. Figure options 3.2. ERP differences between successful shifts of attention and random letters In order to assess the relative timing of the contribution of frontal and parietal cortex to successful shifts of attention, we recorded participant's brain activity in a separate session while they performed the behavioral attentional shifting task with ERP, with the parameters for stimulus presentation adopted from the thresholding session. Critically, the individually established attentional switch thresholds allowed us to ensure that any effects we observe occurred prior to the completion of an attentional shift. First, following standard preprocessing (see Section 2) we extracted the grand averaged waveforms for Shift Hits (targets following a shift cue) and Random Letters (a random letter chosen from a no-target trial, within the temporal window of where the target would appear if it were a target-present trial) (Fig. 3). The comparison of these two conditions provides information about where and when the processing of the shift cue begins. The comparison of shift versus the baseline affords a clean determination of the shift (rather than any processes engaged in inhibiting the shift as might be true in the hold condition; see below for further analyses of the hold condition). We calculated the earliest separation times (see Section 2) between the shift and random letter conditions in the frontal and parietal ROIs. We first consider the response to contralateral stimuli – these are trials in which the shift cue (i.e., digit ‘4’) appeared on one side of space and the target appeared on the opposite side in the subsequent display (see Fig. 1 for examples). The waveforms showing the responses to contralateral stimuli are depicted in Fig. 3 (left panel), and suggest that the first significant divergence occurred in the frontal electrodes at 146 ms, well before the average behavioral threshold of 286 ms. The parietal electrodes showed the divergence only at 227 ms, a full 81 ms after the frontal electrodes but still in advance of the behavioral threshold. Qualitatively, note that the difference between frontal and parietal latencies of separation is substantial and greater than 25% of the available time range pre-shift (286 ms). These results suggest that the frontal cortex likely initiates the processing of the shift cue and then triggers the response of the parietal cortex. There was also a divergence between the conditions in the ipsilateral field in frontal electrodes prior to the behavioral threshold (170 ms). However, the divergence in the ipsilateral parietal electrodes (426 ms) occurred well after the behavioral threshold, implicating its involvement in post-perceptual and post-attentional shift processes. Differences between Shift Hits and Random. Raw ERP timecourses for Shift Hits ... Fig. 3. Differences between Shift Hits and Random. Raw ERP timecourses for Shift Hits and Random Letter trials. The first row shows these timecourses derived from the frontal electrodes, while the second row shows the same derived from the parietal electrodes. The first column shows responses to contralateral stimuli, while the second column shows the responses to ipsilateral stimuli. The red dotted lines show the average behavioral threshold for shift cues (see Fig. 2). The black dotted line indicates the first timepoint at which the Shift Hits and Random responses reliably separated in each plot. Note the later separation in parietal than frontal electrodes. Note also that the parietal electrodes produce very weak responses to ipsilateral cues and no separation in the responses until well past the behavioral threshold. Figure options In order to qualitatively visualize the spatial and temporal distribution of the difference between Shift Hits and Random Letters, we created a topographic plot of the difference between these waveforms across electrodes and time (Fig. 4A). As is evident from the plot, the earliest difference is a positive deflection across the contralateral frontal electrodes that then spreads to the ipsilateral frontal electrodes, followed by a quite spatially punctate negative deflection in the contralateral parietal electrodes. Note that the hemisphere engaged by the cues is always contralateral, with a strong flipping of the lateralization observed between left and right cues (Fig. 4B). Topography and timecourse of differences between Shift Hits and Random trials. ... Fig. 4. Topography and timecourse of differences between Shift Hits and Random trials. (A) The pattern of results was similar for left and right cues but flipped across hemisphere (see Fig. 4B). Therefore, the data have been collapsed over left and right presentations of cues by flipping the identity of electrodes across hemispheres to increase power and reduce complexity. The Shift cue or Random Letter occurred in the left hemifield (top inset) at time 0 (first topographic plot). The first significant difference between the two conditions occurred in the contralateral frontal electrodes (black arrow) at 146 ms (Fig. 3), spreading shortly thereafter to the ipsilateral frontal electrodes. The next significant difference emerged in the contralateral parietal electrodes (black arrow) at 246 ms (Fig. 3), which was before the mean behavioral threshold of 286 ms (dashed red line). (B) Plot of the difference between left and right cues in the critical time bins. Figure options Taken together, these results suggest that whereas contralateral frontal cortex participates in successful shifts of attention, both contralaterally and ipsilaterally and roughly at the same temporal point, the parietal engagement in shifts of attention appears to only be evident for contralateral but not ipsilateral shifts within the time range established for it to be functionally relevant. 3.3. Differences between shifts and maintenance of attention In order to investigate the difference in the ERP signal between processes engaged in shifting attention to a particular location versus maintaining attention on the same location, we next examined the responses to contralateral cues resulting in shift hits and hold hits (Fig. 5). Note that we focus only on the contralateral trials as the ipsilateral trials (see above) have a very delayed divergence. Interestingly, in both the frontal and parietal electrodes, these two conditions were not statistically different, in that the waveforms had the same basic components and appear qualitatively equivalent. Interestingly, this equivalence is apparent even though, in the case of the hold cues, most of the response occurs well after the behavioral threshold (168 ms), suggesting some variability in the signal. Differences between Shift Hits and Hold Hits trials. Raw ERP timecourses for ... Fig. 5. Differences between Shift Hits and Hold Hits trials. Raw ERP timecourses for Shift Hits and Hold Hits trials. The top plot is derived from frontal electrodes and the bottom from parietal electrodes. The red and blue dotted lines indicate the behavioral thresholds for shift and hold cues, respectively. Note the larger and earlier second component for Hold Hits compared to Shift Hits. Figure options Closer scrutiny of the waveforms, however, reveals that the shift and hold waveforms are not formally equivalent: relative to the hold cues, the peak response to shift cues was delayed in both ROIs and this is especially evident in the second major component of the response, which occurred just before the behavioral threshold for successful shifts of attention (286 ms). Peak times for this component, which occurred in the standard range for P2 (180–270 ms) were extracted for each individual participant and entered into a two-way ANOVA with ROI (frontal, parietal) and cue-type (shift, hold) as factors. There was a main effect of cue-type [F(1,11) = 9.80, p = 0.01], with longer latencies to peak for shift (frontal = 248 ms, parietal = 251 ms) than hold (frontal = 231 ms, parietal = 240 ms) cues ( Fig. 6). No other effects reach significance (all p > 0.1), though there was a weak trend for a main effect of ROI [F(1,11) = 2.56, p = 0.14]. These results suggest that shift cues trigger additional processing in frontal and parietal relative to hold cues. Analysis of the second component of Shift and Hold Hits trials. Latency and peak ... Fig. 6. Analysis of the second component of Shift and Hold Hits trials. Latency and peak response of the second component for Shift and Hold Hits trials in frontal (top row) and parietal (bottom row) electrodes. The first column shows the latency from stimulus onset to the peak response. The second column depicts the magnitudes of the peak response. Note the longer latency in both frontal and parietal electrodes for Shift compared to Hold hits. Note also the stronger response in frontal electrodes for the Shift compared to Hold cues. Figure options In addition to the apparent temporal disparity between the peak of the shift and hold cues, there was also a difference in the strength of the ERP components. To compare the signal magnitude in the ROIs, the absolute peak value for the component for each individual was entered into a two-way ANOVA with ROI (frontal, parietal) and cue-type (shift, hold) as factors, revealing an ROIxCue-type interaction [F(1,11) = 4.81, p = 0.05]. Subsequent pairwise comparisons revealed that this interaction arose from the component being stronger in frontal electrodes for shift (2.28 μV) than hold (1.6 mV) cues [t(11) = 2.04, p < 0.05]. This effect was absent in parietal electrodes [t(11) = 0.55] but showed a trend in the opposite direction with a slightly weaker response to shift (−1.85 μV) than hold (1.99 μV) cues ( Fig. 6). These results indicate that shift cues cause additional and delayed processing in frontal cortex relative to hold cues but that the signal is of greater magnitude when it emerges. Parietal cortex evidences a delay in the onset of the component, perhaps reflecting the delay in the signal from frontal cortex.