Introduction

The number of representations one can simultaneously hold in visual working memory (VWM) is highly limited; behavioral studies have suggested that it is up to about four items in humans [1], [2]. This limitation has been thought to make the brain prioritize the processing of relevant over irrelevant information [3], [4], [5], [6], [7], [8]. Maintaining a limited number of representations in an active state by sustained attention plays a primary role in VWM, which is considered to be the interface through which attentional control mechanisms filter and select information from cluttered environments [9], [10]. Visual attention and VWM are intimately linked [9], [10], [11], although the exact sameness between the two is questioned [12], [13].

Recently, electrophysiological and neuroimaging studies have demonstrated that this capacity-limited memory system resides in the posterior parietal and occipital cortex [14], [15], [16]. The activity of the IPS has been assumed to reflect the number of representations held in VWM because the activity shows memory load-dependent responses [14], [16], [17], while some parts of occipital cortices have also been known to show similar memory load-dependent responses [14], [15], [16], [18]. It remains, however, to be demonstrated how differently each memory load-dependent area contributes to VWM, especially when task-irrelevant items are to be excluded.

Although working memory is often, because of its severely limited capacity, considered to store only necessary information [3], [6], [7], recent studies suggest that the frontoparietal network encodes not only necessary objects but also unnecessary objects so as to control occipital activity [19], [20]. In fact, Tsushima et al [21] found that representations of distractors in visual areas are not subject to effective inhibitory control when they are subthreshold and not represented in the prefrontal cortex. Furthermore, posterior parietal lesions have been shown to impair the filtering of distractors [22]. These results, together with findings showing that the frontoparietal network biases activity in the earlier visual pathway to enable effective processing of targets and distractors [19], [20], [23], suggest the need for representing distractors in the frontoparietal network to exert inhibitory control over visual areas.

Load-dependent responses in parietal and occipital cortices may thus reflect distinct aspects of attentional control in VWM; i.e., activity in the parietal cortex may reflect the “source” of stimulus filtering and selection, while that in the occipital cortex may reflect the target or output of that control process. Here, we investigated this issue by elucidating which regions are susceptible to task-relevant stimuli, using a change detection task in which participants are required to remember pre-specified targets and ignore distractors [5], [24]. We hypothesized that activity would decrease when items were task-irrelevant in the occipital cortex but not in the IPS, because the source (parietal) region has to handle task-irrelevant stimuli so as to modulate (e.g., suppress) activity in the target (occipital) region.

Materials and Methods

Design and Procedure

The experimental design and procedures have been described elsewhere in detail [24] and are summarized here. An example trial is depicted in Fig. 1A. Each 6-s trial started with a sample display containing one, two, four, or six red rectangles, or two red rectangles with two blue rectangles (resulting in five experimental conditions), presented for 150 ms. Each rectangle (approximately 1.8°×0.8°) had one of four orientations (vertical, horizontal, left 45°, right 45°) and was located 3.1° away from the fixation point. Following the sample display, a 1200-ms blank interval, and then a 2000-ms test display were presented. One of the red rectangles changed its orientation for half of the trials and did not change for the other half. Participants were required to indicate whether a red rectangle in a test display changed its orientation or not from a sample display during a test display phase while ignoring blue rectangles as distractors. Each functional run consisted of five experimental conditions and a non-event condition (only the fixation point was presented), with the order of conditions pseudo-randomized within runs. Participants completed four functional runs, each including 12 trials per condition.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Experimental protocol and behavioral results.

A. An example trial. A trial type with two targets and two distractors (distractor condition) is shown. Participants had to indicate whether a red rectangle changed its orientation, while ignoring blue rectangles as distractors. Stimuli are not drawn to scale. B. Behavioral results. The behavioral VWM estimate K (open circles) and accuracy (filled circles) are shown as a function of trial types. Error bars reflect ±1 SEM.

https://doi.org/10.1371/journal.pone.0038623.g001

We used a standard formula [1] to estimate the number of objects held in VWM for each set size. K = (hit rate + correct rejection rate−1)×S, where K is the VWM capacity estimate, and S is the display set size of the array.

Results

Imaging data

The analysis revealed regions whose activities significantly correlated with the number of objects held in VWM in the right IPS (x = 22, y = −55, z = 46; Fig. 2A left), the bilateral posterior lateral occipital cortex (LO; x = −32, y = −78, z = −4 for left; x = 38, y = −78, z = 1; x = 30, y = −80, z = 10 for right; Fig. 2A middle), and the right dorsal occipital cortex (x = 23, y = −94, z = 11; Fig. 2A right). The left IPS and the left dorsal occipital cortex showed significant activation when the threshold was relaxed twenty-fold and ten-fold, respectively. These regions exhibited qualitatively identical responses with each contralateral counterpart (F values <1 for both main effect of hemisphere and interaction between hemisphere and trial type).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Results of the ROI analysis.

A. Coronal views of the IPS (left, Bonferroni p<.05; y = −55), lateral occipital regions (middle, Bonferroni p<.05; y = −78), and the dorsal occipital cortex (right, the threshold was ten-fold relaxed from Bonferroni p<.05 for the purpose of displaying activation of the left; y = −91). B. BOLD signal magnitude as a function of trial types in each ROI. Error bars reflect ±1 SEM.

https://doi.org/10.1371/journal.pone.0038623.g002

BOLD signal magnitude as a function of trial type is shown for each activated area in Fig. 2B. The results for the three activated areas of the LO were pooled for further analysis because there was no significant main effect of area and interaction between area and trial type (F values <1). Activities in the LO, and dorsal occipital cortex increased as a function of VWM load (P values <001, between set size 1 and 6), while the IPS activity increased up to display set size 4 (P values <001 between set size 1 and 4) and thereafter became asymptotic (t<1, between set size 4 and 6). The IPS activity was better described by a K function than linear function (t₁₅ = 4.17, P<0.001) but this was not true of the LO and dorsal occipital activities (t values <1). This suggests the IPS, rather than occipital regions, to be a key neural locus of capacity-limited VWM, but does not necessarily mean the absence of a contribution from the occipital regions to VWM.

Importantly, whereas comparisons between the distractor condition and set size 4 in the IPS and LO areas did not show significance (t values <1), the dorsal occipital cortex exhibited a significantly lower response to the distractor condition relative to set size 4 (t₁₅ = 3.41, P<0.01). Differences between the distractor condition and set size 2 were significant in all ROIs (t₁₅ = 6.39, P<0.001 in the LO; t₁₅ = 5.18, P<0.001 in the IPS; t₁₅ = 2.75, P<0.02 in the dorsal occipital cortex).

Discussion

The present study investigated whether load-dependent responses in parietal and occipital cortices can be differentiated by behavioral relevance. We found that the dorsal occipital cortex showed less of a response to distractors than to targets, while IPS activity did not differ. These results are consistent with the idea that the parietal cortex subserves attentional control and the modulation of occipital activity may reflect output for that control process [19], [22], [27], [28], [29], [30], [31], [32]. The IPS might respond to task-irrelevant items because of the need to manage task-irrelevant information to avoid deleterious effects on behavior. These control processes might cause the modulation of activity in the dorsal visual cortex (see also [5], [20]), although the moderate decline of activity indicates distractor filtering is imperfect.

It is unlikely that the comparable BOLD response between targets and distractors in the IPS is due to the placing of equivalent processing weights on both targets and distractors. Because change detection performance for an array of two targets with two distractors was indistinguishable from that for two targets only, it would seem that participants could effectively prioritize targets and prevent distractors from affecting behavioral performance. Note that this does not necessarily mean perfect distractor filtering: previous behavioral studies, which used a similar change detection task assessed with distractor-change [24] or lure trials [26], have showed that distractors are not filtered perfectly in VWM, while leaving accuracy or memory capacity (K) unaffected [24], [26]. This imperfectness of distractor filtering may also be reflected in the dorsal occipital activity. Furthermore, one might argue that items in a test display increased the IPS activity under the distractor condition, because in a test display, more items were present under the distractor condition than two-target condition. However, the IPS showed the same response pattern even when a single item from a sample display was used as a probe instead of all items (see supplemental experiment in [24]). The results thus suggest that the IPS activity is not contaminated by the test display.

The coordinates of the IPS correspond to the superior IPS, which specifically processes featural information, in Xu's superior/inferior distinction of IPS [16]. This may indicate that task-irrelevant featural information is encoded in VWM (but see [33]). Note, however, that it remains unclear whether the IPS activity reflects task-irrelevant representations held in VWM, or if it instead reflects the requirement to focus on targets and/or ignore distractors. The IPS may act as a limited capacity ‘pointer’ system [34] in VWM that can help individuate task-irrelevant objects and help filter them out (see also [22], [35], [36], [37]), or may simply use more attentional resources to concentrate on task-relevant objects or to suppress task-irrelevant objects [38]; both processes are likely to result in IPS activation under the distractor condition. The present results do not distinguish between the two, but nevertheless suggest that, at least to some extent, the IPS processes task-irrelevant information.

The VWM load sensitive activity in the IPS has been considered analogous to the contra-lateral delay activity (CDA) [5], [15], [35], [39], [40]. The CDA is demonstrated to reflect individual differences in allocating VWM capacity [5]: the higher the memory capacity that one has, the more efficiently one can prevent irrelevant items from increasing CDA, i.e., consuming capacity. On the face of it, our results seem to be inconsistent with Vogel et al. [5] in the sense that the memory load sensitive region (i.e., the IPS) showed comparable response to both relevant and irrelevant stimuli. However, as their own scalp topography analysis [39] has shown, CDA is distributed over the parietal and occipital cortex and is computed using relatively lateral/posterior electrodes such as PO3/4. It is therefore possible that their neural evidence for efficient attentional filtering reflected in the CDA originates not only from the IPS but also from the occipital regions as shown by our results¹. The CDA and the IPS might reflect “somewhat distinct but overlapping” neural mechanisms supporting VWM [41]. In fact, Robitaille et al. [41] has shown that the two neural activities are not necessarily identical. Moreover, in the first place, sulcal activity is difficult to measure in electroencephalogram (EEG) [42], while this is not the case in fMRI. Note that activities in the LO and the dorsal occipital cortex did not reach asymptote at four objects, but this may be due in part to the objects and because the task used in the present study was relatively easy to memorize (see [16], [33]). The LO and the dorsal occipital cortex might have greater processing capacity than the IPS and/or process visual objects in different ways from the IPS.

Although we did not conduct retinotopic mapping and thus can not define the precise retinotopic location, the coordinates of our dorsal occipital activation closely match those of area V3a in a previous study [43]. Recent neuroimaging studies have begun to demonstrate that V3a is involved in figural processing [44], [45], [46]. V3a has a representation of the whole contralateral visual field and a relatively large receptive field, and thus is a likely candidate for early figural integration [44]. In particular, the finding by Scholte et al. [47] that the conscious detection of a segregated figure results in higher V3a responses is suggestive of the susceptibility of V3a to perceptual awareness or attentional manipulations. It therefore seems that the decline in the dorsal occipital activity under the distractor condition reflects attentional modulation of figural representations in the V3a. Although the role of V3a in figural processing is not well understood, the present results may support the notion that V3a is subject to attentional factors.

Finally, the differential responses to distractors we found among VWM load-sensitive regions indicate that each area contributes differently to the processing of task-irrelevant information. Consistent with previous studies [16], [48], we also found VWM load-sensitive activity in the LO. Given that the LO and IPS responded comparably to both targets and distractors, the LO may also contribute to processing of task-irrelevant objects [18], [49], [50], [51]. Note however that the activity of the LO, unlike the IPS whose activity reached asymptote at four objects, tracked the total number of objects in the display. This might reflect general object processing [52], [53], rather than capacity-limited VWM, in the LO. Occipital activations thus would not reflect the VWM capacity-limit itself, but might nevertheless support VWM by processing mid-level aspects of visual objects. Further study will be necessary to understand the particular role and/or the cooperation of intraparietal and occipital regions in the processing of task-irrelevant information (e.g., [54]). The decline of dorsal occipital activity under the distractor condition seems to be consistent with Vogel et al. We could not find, however, a significant correlation between individual capacity estimates (K_max) and encoding task-irrelevant objects into VWM (distractor filtering efficiency: α) in the dorsal occipital cortex (r = 124). This discrepancy might be because the number of targets and distractors was not necessarily equal between hemifields; e.g., targets were presented in one hemifield and distractors in the other hemifield at times, while targets and distractors were evenly distributed in each hemifield at other times. Alternatively, the CDA and the dorsal occipital activity may reflect different neural mechanisms supporting VWM. Note: Individual capacity estimates were derived from the maximum value of Cowan's K across all set sizes of that subject (K_max). Distractor. filtering efficiency was derived from the next formula: α  =  (F-D)/(F-T), where α is the filtering efficiency, F is the signal magnitude for the four targets condition, D is the signal magnitude for the two targets with two distractors condition, and T is the signal magnitude for the two targets alone condition [5].

Acknowledgments

We would like to thank Hiroyuki Tsubomi and anonymous reviewers for comments on earlier drafts of the manuscript.