Context Binding in Visual Working Memory Is Reflected in Bilateral Event-Related Potentials, But Not in Contralateral Delay Activity Part 1

Visual Abstract

Successful retrieval of a specific item from visual working memory (VWM) depends on the binding of that item to its unique context. Recent functional magnetic resonance imaging studies of VWM manipulating memory set homogeneity have identified an important role for the intraparietal sulcus in context binding, independent of any role in representing stimulus identity. 

Working memory and memory are two different concepts, but they are closely related. Working memory refers to the information we can remember for a short period, usually related to the task currently in progress. Compared to working memory, memory refers more to our ability to remember long-term, that is, our ability to store and retrieve memories over time.

Although working memory and memory are different, they are uniquely linked. A strong working memory usually predicts a better memory. This is because working memory is the "hub" of our long-term memory, and we must store information in working memory before it can be transferred to long-term memory. So, if our working memory is strong, it is easier for us to store information in long-term memory.

In addition, working memory can also be seen as part of the "attention control system" in the brain. Attention is key to how we remember information, and if we can't pay attention to certain information, we won't be able to remember it. Therefore, a good working memory is closely related to a good command of attention, which also helps us remember better.

In conclusion, a good working memory is critical to long-term memory capacity. Fortunately, there are several ways we can enhance our working memory and memory. For example, practicing memory games, getting enough sleep, and exercising can all enhance our working memory and memory. So, let's work on improving our working memory and memory so we can face challenges better. It can be seen that we need to improve our memory. Cistanche can significantly improve memory, because Cistanche can also regulate the balance of neurotransmitters, such as increasing the level of acetylcholine and growth factors. These substances are very important for memory and learning. In addition, meat can also improve blood flow and promote oxygen delivery, which can ensure that the brain receives sufficient nutrition and energy, thereby improving the vitality and endurance of the brain.

Click know supplements to boost memory

The current study explored whether the contralateral delay activity (CDA), which is an event-related potential (ERP) component derived from posterior electrodes that track the amount of information held in VWM, might also be sensitive to context-binding demands. In experiment 1, human participants performed lateralized delayed recognition with memory sets containing one, three, or five items that were drawn from the same category (orientations: “homogeneous”) or from different categories (orientation, color, and luminance: “heterogeneous”). Because the location and identity of the memory probe indicated the item to be retrieved, homogeneous trials placed higher context-binding demands. 

VWM capacity was higher in heterogeneous trials. ERPs contralateral (contra) and ipsilateral (ipsi) to the remembered stimuli were higher for homogeneous trials, but these differences were removed in the contra – ipsi subtraction that produced the CDA. In experiment 2, human participants performed lateralized delayed recall with memory sets of one or three items (homogeneous or heterogeneous). 

Behavior was superior for three-item heterogeneous trials than for homogeneous trials, with modeling revealing context-binding errors in the latter. Bilateral ERPs and CDA results replicated experiment 1. These results support that the CDA tracks the number of object files engaged by VWM and establishes that it is not sensitive to context-binding demands.

Significance Statement

The contralateral delay activity (CDA) tracks the number of items held in visual working memory (VWM), but it remains unclear which cognitive processes can influence it. For example, although VWM entails the representation of the identity of each item, it also requires the representation of the unique context of each item (e.g., where or in what order it appeared). Here we varied stimulus set homogeneity as a way of manipulating demands on context binding. Although this manipulation was successful at influencing behavior, the CDA was insensitive to it. 

This supports previous models suggesting that, although the CDA acts as an abstract marker of the number of items being held in VWM, it does not provide a direct measure of all the neurocognitive processes whose operations support performance on VWM tasks.

Introduction

For the past quarter century, research on human visual working memory (VWM) has been massively influenced by variants of the “change detection” task popularized by Luck and Vogel (1997). In its canonical form, it is a test of delayed recognition in which an array of to-be-remembered items (often colored squares) is presented, followed by a brief unfilled delay period, followed by a recognition probe. 

In some variants, memory is tested via re-presentation of the sample array, with one item changed with p = 0.5; in others, a single probe stimulus assesses memory for the item that had occupied the probed location. Critically, the number of items in the sample array (the “memory load”) varies from trial to trial, and performance as a function of load can be transformed to yield an estimate of “VWM capacity” (i.e., “Cowan’s k”; Cowan, 2001), a measure that has impressive psychometric properties (Luck and Vogel, 2013; Fukuda et al., 2015b; Hakim and Vogel, 2018). 

Initial results with these canonical variants of the task led to theoretical models of VWM capacity limitations arising from a slot-like architecture (Zhang and Luck, 2008), and objections to this interpretation, in turn, prompted the introduction of a “delayed estimation” variant of the task in which subjects were cued to recall the color of the probed item by responding with a color wheel (Bays et al., 2009). 

This modification allowed for estimation of the precision of recall, as well as of misbinding (i.e., “swap”) errors, and lent itself to modeling VWM as dependent on a finite resource that is spread evermore thinly as load increases (Ma et al., 2014; for review, see Postle and Oberauer, 2022).

One highly influential development in this literature has been the characterization of an event-related potential [ERP; derived from the electroencephalogram (EEG)] component that closely tracks estimates of k. 

This contralateral delay activity (CDA) component is obtained in a variant of the canonical “change detection” task in which two arrays are presented, one in each hemifield, and subjects are to encode only the array that is produced before array onset. The EEG data from posterior electrodes are then averaged as a function of whether they were located over the cortex that is contralateral or ipsilateral to the cued array, and the subtraction of the ipsilateral from the contralateral signal yields the CDA. 

The CDA, which corresponds to the delay period of the trial, is tightly correlated with individual differences in k, becoming increasingly negative as memory load increases, and saturates the individual’s capacity. Thus, the CDA has been interpreted by many as a neural correlate of the storage of information in VWM (Vogel and Machizawa, 2004; Luria et al., 2016).

Because of its close linkages to the behavioral and psychometric findings summarized above, empirical results with the CDA are often used to advance theoretical claims. Consider tasks that make no overt demands on VWM, but in which sustained activity can nonetheless be seen to be greater at contralateral than at ipsilateral electrodes, and to track estimates of k. In these cases, “CDAlike” activity is often taken as evidence for a contribution of VWM to these nominally non mnemonic tasks. For example, for the multiple object-tracking task, although the target items are always visible, the CDA-like signal recorded during the task has been taken as evidence that successful tracking requires a working memory-like operation (McCollough and Vogel, 2010). 

For visual search, the “contralateral search activity” observed during lateralized visual search has been interpreted as “memory in search” (Kundu et al., 2013), whereby subjects may hold in VWM a record of the items in the array that have already been visited, to avoid revisiting them (Emrich et al., 2009). Also in visual search, but addressing a different stage of processing, the gradual diminution, across consecutive trials requiring a search for the same target, of the CDA-like signal that spans the delay between target offset and search array onset has been interpreted as a “handoff” of the representation of the search template from VWM to long-term memory (Carlisle et al., 2011).
Because considerable theoretical weight is often conferred on the CDA, it is important to fully understand the neurocognitive factors that underlie it. To this end, the experiments reported here draw inspiration from previous studies using functional magnetic resonance imaging (fMRI) that raise questions about the specificity of CDAlike activity for the active retention of stimulus information during VWM. As a point of departure, the same journal issue in which the CDA was first described by Vogel and Machizawa (2004) also contained a report of results of an fMRI study showing a broadly consistent pattern of results. 

In this article, Todd and Marois (2004) reported that, like the CDA, delay-period fMRI signal in the intraparietal sulcus (IPS) increased monotonically with VWM load before saturating at k. Although this finding has been replicated previously (Xu and Chun, 2006), subsequent research has indicated that the delay-period activity of the IPS is also sensitive to factors other than stimulus representation per se. 

For example, in an fMRI study of VWM for the direction of motion, although Emrich et al. (2013) observed a monotonic increase of delay-period activity in IPS for loads of one versus two versus three directions of motion, multivariate pattern classification (MVPA) of this signal failed to find evidence for stimulus information. (In the occipital cortex, in contrast, despite the absence of above-baseline levels of fMRI signal intensity, MVPA decoding of stimulus information from delay-period signal was successful, and decoder performance declined linearly with memory load.) In a follow-up study, Gosseries et al. (2018) measured BOLD activity while subjects held in working memory the direction of one motion patch (1 M), the directions of three motion patches (3 M), or the direction of one motion patch plus the chrominance values of two static color patches (1M2C). The MVPA decoding results were generally consistent with those from Emrich et al. (2013), but it is the pattern of delay-period signal intensity in IPS that is of particular interest for our present purposes: it was equivalent for 1 M and 1M2C trials, and markedly higher for 3 M trials. 

Because 1M2C and 3 M trials both required the retention of three items, the difference between the two indicated that some factor other than the number of items, per se, contributed importantly to delay-period activity in IPS. [Note that although the difference in delay-period signal between 3 M and 1M2C might be explained, in part, by a difference in stimulus energy between the two conditions (i.e., working memory for three motion patches might drive IPS harder than working memory for one motion patch and two color patches), this same logic cannot account for the absence of a delay-period load effect between 1 M and 1M2C trials.]

The results from the study by Gosseries et al. (2018) were replicated and extended by Cai et al. (2020), the study that leads directly to the experiments reported here. In this fMRI study, subjects viewed sample arrays of one oriented-bar stimulus patch (1O), three oriented bars (3O), or one orientation patch, one color patch, and one luminance patch (1O1C1L), and recalled the probed item on an orientation wheel, a color wheel, or a luminance wheel. 

The results were broadly consistent with those from the studies by Emrich et al. (2013) and Gosseries et al. (2018)—delay-period activity in IPS was comparably low on 1O and 1O1C1L trials relative to 3O—and they also generated evidence for a role for IPS in an operation that might account for the patterns of 1 M = 1M2C, 3 M (Gosseries et al., 2018) and 1O = 1O1C1L, 3O (Cai et al., 2020): the binding of trial unique context to each item that must be retained in working memory. 

In particular, multivariate inverted encoding modeling of the fMRI signal at recall indicated that, on 3O trials, subjects with a higher probability of responding to nontargets (i.e., those who committed more “swap errors”) represented the location of the probed item, as well as its orientation, less strongly, and with less differentiation from nonprobed items. Furthermore, the delay-period signal in IPS predicted behavioral and neural correlates of context binding at recall. The logic of the experiments presented here was to record the EEG during the performance of lateralized variants of the task from the study by Cai et al. (2020), to assess the extent to which the CDA might also be sensitive to memory-set homogeneity, a finding that would suggest that the CDA might reflect, in part, context-binding operations in VWM.1
Experiment 1

Materials and methods

Twenty-eight right-handed volunteers (16 females; age, 18–25 years; mean age, 22.87; SD = 3.22) participated in the study for remuneration ($20/h). The n was selected to be comparable to previous studies of the CDA (Vogel and Machizawa, 2004; Luria et al., 2016). All subjects provided written informed consent according to the procedures approved Health Sciences Institutional Review Board. Subjects had normal or corrected-to-normal vision, no contraindications for EEG, and no reported history of neurologic or psychiatric disease.

Stimuli

Delayed-recognition trials were blocked by condition: homogeneous versus heterogeneous memory sets. Homogeneous trials presented one, three, or five oriented-bar stimuli (1O, 3O, and 5O) rendered as the black diameter (length, 1.6°; width, 0.08°) of a white circular patch, and drawn from a pool of nine possible orientations ranging from 10° to 180°, in 20° increments. 

Heterogeneous trials presented one, three, or five items drawn from the categories orientation, color, and luminance. Orientation stimuli were in the homogeneous condition. Color stimuli were drawn from a pool of nine equidistant colors along a circle in CIE L*a*b* color space (L = 70, a = 20, b = 38, the radius of 60; and thus varying markedly in hue and slightly in saturation), and presented on 1.6° diameter circles. Luminance stimuli comprised a gray annulus (diameter, 1°) inside a white ring (diameter = 1.6°, RGB values ([0, 0, 0]). 

The annulus could take on one of nine grayscale values ranging equidistantly from light gray ([0.03, 0.03, 0.03] to darkest gray ([0.97, 0.97, 0.97]). All stimulus arrays were presented within two 4°
7.5° rectangular regions that were centered 3° to the left and right of a central fixation. In one-item arrays, one stimulus appeared in the center of each rectangular region. In three-item arrays, one stimulus appeared in the center of each rectangular region and one each at the top and bottom corners nearest fixation. In five-item arrays, one stimulus appeared in the center of each rectangular region and one at each of the four corners (Fig. 1A).

Procedure

Experimental sessions comprised the following two tasks: delayed recognition (i.e., “change detection”), followed by an orientation discrimination task. (Orientation discrimination was conducted for a different purpose and will not be described here.) All stimuli and procedures were generated and presented in MATLAB (MathWorks) and Psychtoolbox-3 extensions (http://psychtoolbox.org).

In the delayed-recognition task, each trial began with the simultaneous onset of the fixation cross (which remained present throughout the trial) and the cues (appearing above and below fixation) indicating which display would be tested on that trial (left or right, unpredictably; equal number in each block; 200 ms). The ensuing unfilled cue–stimulus interval varied unpredictably in length (400–600 ms, jittered in steps of 50 ms), followed by the bilateral presentation of the sample arrays (750 ms). After a 900 ms unfilled delay, a probe appeared in the same location as one of the sample stimuli on the cued side (unpredictable on three-item and five-item trials; Fig. 1A). 

In each block, an equal number of probes matched or did not match the identity of the item that had appeared at its location, in an unpredictable sequence. Nonmatching probes were always drawn from the same category as the probed item but had a value that differed from it by 90° in stimulus space (each domain scaled to span 180°). Subjects were instructed to indicate their match/nonmatch judgment by pressing the “F” or “J” key (with the left or right index finger, respectively, on a keyboard resting in their lap; counterbalanced across subjects) within the 2 s that the probe appeared on the screen. Trial homogeneity was blocked (120 trials/block), and the two types of blocks were interleaved in random order. 

Load varied unpredictably within each block.

The six homogeneous blocks contained a total of 180 1O trials, 360 3O trials, and 180 5O trials; the nine heterogenous blocks contained a total of 540 one-item trials [180 O trials; 180 color (C) trials; 180 luminance (L) trials; 360 three-item trials (all 1O1C1L trials), and 180 five-item trials (60 1O1C2L, 60 1O2C1L, and 60 2O1C1L trials); the reason for the larger number of three-item trials was to achieve better sensitivity for multivariate inverted encoding modeling, the results of which we do not report here, and so details specific to this task are not presented]. Each heterogeneous block contained an equal number of 1O, 1C, and 1L trials. For three-item and five-item trials, the category configuration of the arrays was the same in both hemispheres, and an orientation always occupied the center position of the array. The delayed recognition portion of the experiment took;100 min to complete.

Behavioral analysis
To assess delayed recognition performance, we calculated Cowan’s k value in each memory condition following the formula: k = set size
(hit rate – false alarm rate), where hit rate corresponded to successful responses on nonmatch trials and false alarm rate corresponded to incorrect responses on match trials (Cowan, 2001). Sensitivity to our experimental manipulations was assessed with a two-way repeated-measures ANOVA, with the factors of homogeneity and set size. 

In the event of significant interactions, post hoc tests were conducted to further clarify the differences between homogeneous and heterogeneous memory sets in each set size. We only included data from the trials from heterogeneous blocks in which memory for orientation was probed. In this way, any differences found between the two homogeneity conditions could only be attributed to differences at encoding or during the delay, because the precise appearance of the probes, and the operations they prompted, were identical.

EEG methods

Data acquisition and preprocessing. During the performance of the behavioral tasks, EEG was recorded with an Eximia 60-channel amplifier (Nextim) with a sampling rate of 1450 Hz. The 60 Ag/AgCl electrodes were positioned according to the extended 10–20 system, with a reference electrode on the forehead. During the recording, impedances of all the channels were kept at,15 kV. EEG data preprocessing and analysis were conducted using the EEGLab toolbox in MATLAB (Delorme and Makeig, 2004) and customized MATLAB scripts (MathWorks). Eye movements were monitored with EOG electrodes placed near the external canthus of, and below, the right eye.

For the delayed-recognition task, raw voltage data were downsampled to 250 Hz offline, bandpass filtered (0.1;30 Hz), and segmented into epochs from 1.5 to 12.5 s relative to sample array onset. After the segmentation, bad EEG channels were identified by visual inspection and were interpolated using the “spherical” method in EEGLab. Next, baseline removal was conducted by subtracting the averaged activity from the 200 ms prestimulus interval, and epochs with baseline-corrected activity exceeding 100 mV at any electrode were discarded. 

Additionally, because horizontal eye movement can contaminate lateralized measures, we used a split-half sliding window approach (Adam et al., 2018; window size, 200 ms; step size, 20 ms; threshold, 20 mV) on the horizontal EOG signal. If the change in voltage from the first half to the second half of the window was .20mV, it was labeled an eye movement, and that epoch was rejected. For the remaining epochs, eye blinks and muscle artifacts were identified with independent component analysis (ICA) and removed. 

Finally, the post-ICA data were carefully visually inspected to catch any potential remaining artifacts and were rejected. After epochs were sorted by trial type, any subject with,75 trials was excluded from the EEG analyses. This resulted in the exclusion of data from 4 subjects, leaving a sample of 24 subjects whose EEG data were included in the analyses.

For more information:1950477648nn@gmail.com