Elsevier

Brain Research

Volume 1411, 9 September 2011, Pages 41-56
Brain Research

Research Report
Event-related potential (ERP) measures reveal the timing of memory selection processes and proactive interference resolution in working memory

https://doi.org/10.1016/j.brainres.2011.07.004Get rights and content

Abstract

Behavioral studies show that no-longer-relevant information, although presumably removed from working memory (WM), still engenders proactive interference (PI). However, the timing of selecting items within WM and resolving PI is relatively unknown. To assess this, we recorded ERPs during WM from 20 young adults. In all conditions, a 4-digit display was followed by a cue indicating which digits to remember. In the selection condition, 2 digits were cued. The reaction time difference between the intrusion probe, a match of a to-be-rejected digit, and the non-intrusion probe, which did not match any of the 4 digits, was reliable, indicating a robust effect of PI. In the neutral-2 (remember 2 digits) and − 4 (remember all 4) conditions, participants maintained the digits following the cue. Relative to neutral-4, selection elicited larger positivity at parietal sites (approximately 260 ms) and negativity at frontal sites (approximately 420 ms). Relative to the non-intrusion probe ERP, that to the intrusion probe was more negative over frontal scalp (approximately 500 ms). We conclude that initial selection occurs over parietal cortex and reflects top–down attention to task relevant items, whereas the subsequent negativity may reflect inhibition of no-longer-relevant items over frontal cortex. The probe-locked ERPs suggest that the frontal negativity (approximately 500 ms) reflects the final resolution of PI.

Highlights

► We studied the timing of selection and inhibitory processes in working memory. ► The selection of task-relevant information is reflected by a parietal positivity. ► Inhibition of no-longer-relevant information may be reflected by a frontal negativity.

Introduction

Working memory (WM) refers to the short-term storage and manipulation of information (Baddeley, 1986, Baddeley and Hitch, 1974). Baddeley (1986) proposed that WM was comprised of a central executive and two slave systems, the phonological loop and the visuospatial sketchpad. The function of the central executive is poorly understood in comparison with the two slave systems. It is considered an attentional-control system that has limited capacity (Baddeley, 2003). One of the important roles of the central executive is to select relevant information while disregarding or inhibiting irrelevant information (Baddeley, 1996). This type of control system also appears in Cowan's (1999) embedded-processes model. In his model, information is represented at three levels: information within the focus of attention, information not in the focus of attention but currently activated, and non-activated information in long-term memory. The focus of attention refers to the maintenance of a small set of information that is readily accessible and can be selected from WM and manipulated directly. The mechanism of selecting information from WM has been found to be similar to the endogenous shifting of attention to a cued spatial location (Griffin and Nobre, 2003, Nobre et al., 2004). In other words, selection of task-relevant information from a currently activated memory set requires an inward focus that serves to maintain task-relevant items in the face of competition from their task-irrelevant counterparts.

Several behavioral experiments have been directed at assessing the nature of the focus of attention and the effect of distracters on WM performance (Makovsik and Jiang, 2007, Makovski et al., 2008, Oberauer, 2001, Oberauer, 2005a, Oberauer, 2005b, Yi et al., 2009). For example, to study the fate of task-irrelevant items in WM (presumed to be outside the focus of attention), Oberauer (2001) had participants view a memory set of 2 to 6 words. After a delay, a selection cue indicated that participants should focus on a subset of the words (1 or 3 words) while ignoring the others (1 or 3 words). At the probe stage, participants had to decide whether the probe was one of the words inside (i.e., task-relevant) or outside (task-irrelevant) the focus of attention. Oberauer (2001) found that the ignored words, although presumably removed from the focus of attention, still engendered proactive interference (PI). That is, reaction time (RT) to a probe that was task-irrelevant (i.e., to be ignored) was longer than that to a probe that had not appeared in the initial memory set (Oberauer, 2001). In sum, the no-longer-relevant words still affected recognition performance.

Oberauer (2001) also varied the time interval between the selection cue and the probe to investigate the time at which no-longer-relevant information was removed from the focus of attention. He found that the set size (1 or 3) of the to-be-ignored information did not affect RT up to 1 s after the selection cue was presented. Consequently, Oberauer concluded that, while the relevant information was maintained within the focus of attention to enable matching with the probe, no-longer relevant information was removed from that focus 1 s following the cue and no longer occupied the central capacity of WM.

Although behavioral experiments have shed light on the mechanisms underlying the selection and removal of items from WM, RT and accuracy rates are the final endpoints of decision processes and cannot easily be queried to determine the processes that unfold between the selection cue and the subsequent probe. By contrast, although temporal information is limited, fMRI-derived hemodynamic activity time-locked to the selection cue can yield information about the brain regions and the putative processes that follow the cue (Lepsien et al., 2005, Lepsien and Nobre, 2007, Nee and Jonides, 2009, Oh and Leung, 2010). In Lepsien et al.'s study, for example, following the presentation of a memory set of colors, an informative cue indicated which items within the memory set were to be selected and brought within the focus of attention. By contrast, an uninformative cue did not provide any information on which items within the memory set were to be selected. Therefore, all items had to be maintained until the probe was presented. The behavioral results showed that RT was faster and accuracy was higher to a probe following the informative compared to the uninformative cue. This result suggested that the informative, selection cue facilitated recognition of the probe. The left inferior frontal gyrus (IFG), insula, supramarginal gyrus (SMG), the right precuneus, inferior parietal sulcus (IPS), and the pre-supplementary area (SMA) showed greater activation when selecting task-relevant colors following an informative compared to an uninformative cue. Because color selection could not be initiated following an uninformative cue, Lepsien et al. (2005) concluded that their findings suggested that a network of frontal–parietal regions, especially the lateral–prefrontal (PFC) and posterior–parietal cortices, were critically involved in selecting task-relevant information within WM.

By time-locking hemodynamic activity to the presentation of the probe, other researchers have focused on the neural underpinnings of PI resolution at the decision stage. These investigators used the recent-negative probe paradigm originally introduced by Monsell (1978). In this paradigm, participants are tested in a short-term recognition task with two types of probes. A recent-negative probe is an item that was presented on the previous trial but is an irrelevant item on the current trail. By contrast, a non-recent negative probe does not match either an item from the current or previous trial. The recent-negative probe creates PI, which is confirmed by comparing its RT and accuracy to those of the non-recent negative probe. These comparisons show that RTs are longer and accuracy lower to the recent-negative probe (Monsell, 1978). Data from several investigations have demonstrated that, in accord with the behavioral data, activation of the left IFG was greater to the recent negative relative to the non-recent negative probe (D'Esposito et al., 1999, Jonides et al., 1998, Nee et al., 2007, Postle et al., 2001, Zhang et al., 2003). On these bases, the left IFG is thought to be implicated in resolving PI engendered by the no-longer-relevant item (see Jonides and Nee, 2006 for a review).

Despite the importance of these findings, and the putative association of the lateral PFC and posterior parietal cortices with the processes involved in the selection of information in WM and the resolution of PI, few studies have demonstrated that selection-related brain activity and PI resolution are related. In an attempt to shed light on this association, Yi et al. (2009) had participants remember a list of four digits. There were two types of cues that followed the initial memory set. A selection cue (an arrow pointing to the left or right) indicated which two of the four digits were to be remembered for the current trial. A neutral cue (a double-headed arrow) indicated that participants had to remember all four digits. Following the selection cue, three types of probes could be presented: a positive probe (which matched one of the to-be-remembered digits), a non-intrusion probe (which did not match any of the previously presented digits) or an intrusion probe (which matched a previously presented, but to-be-ignored, digit). At the probe stage, participants had to decide whether the probe matched one of the to-be-remembered digits. A reliable PI effect was found, which was indicated by longer RTs to the intrusion relative to the non-intrusion probe. fMRI results showed that activation differences between the selection and neutral cues in the left middle frontal gyrus (MFG), inferior parietal lobe (IPL) and precuneus predicted the behavioral measure of PI, i.e., the larger the activation differences, the lower the PI. The results suggested that hemodynamic activity in these frontal and parietal regions might have reflected control processes that served to proactively reduce the interference induced by the intrusion probe.

Despite knowledge of the underlying brain regions, the timing of the processes implicated in the selection of information in WM is relatively unknown. In order to understand better the temporal patterning of processes contributing to selection and the resolution of PI, it is critical to use a technique that can track cognitive processes in a manner consistent with the speed at which they unfold. The event-related potential (ERP) provides high-resolution temporal information in the millisecond range. Accordingly, we collected ERP data during a modified version of the Yi et al. (2009) task design. The purpose of the current study was to examine the timing of processes comprising the selection of task-relevant information in WM and how these processes contribute to interference resolution at the probe stage.

Probe-related processes in WM and the resolution of PI have been assessed in a small set of ERP investigations that have offered fairly consistent results. For example, Tays et al., 2008, Tays et al., 2009 assessed PI resolution in a variant of the recent-negative probe paradigm described earlier. In their study, the recent negative probe elicited a greater frontal negativity (FN) between 400 and 500 ms than the non-recent negative probe. As the no-longer-relevant information had to be rejected in order to perform accurately at the probe stage, Tays and colleagues suggested that this component could have reflected the inhibition or suppression of the task-irrelevant information. The greater FN in the condition requiring inhibition or suppression of a no-longer relevant item is consistent with several reports concerned with the brain's electrical activity to stimuli that engender the inhibition of pre-potent response tendencies, such as incongruent Stroop (Hanslmayr et al., 2008, Liotti et al., 2000, West, 2003), incongruent Eriksen-flanker (Van Veen and Carter, 2002), and No-go trials in the Go/No-go paradigm (Eimer, 1993, Falkenstein et al., 1999, Jodo and Kayama, 1992, Kok, 1986). All of these events elicit larger negativities within the same general time frame as the FN. Taken as a whole, these data provide evidence in support of the hypothesis that the FN reflects inhibitory processing, although other alternatives, such as conflict detection, conflict monitoring, and mismatch detection are possible (see the review by Folstein and Van Petten, 2008 and immediately below). We postpone a treatment of other alternatives for the processes reflected by frontal N2 (i.e., FN) components to the Discussion section.

In a recent-probe experiment relatively close in design to the one reported here, Du et al. (2008) interpreted their negative-going (labeled N2) data differently than proposed immediately above. These investigators assessed PI in WM using a condition similar to the selection condition of Yi et al. (2009). Participants viewed a 4-letter memory set followed by a cue which always indicated the 2 letters that had to be ignored. Following the cue, a positive probe (one of the two, to-be-remembered letters), a negative probe (which did not appear in the 4-letter memory set) or a lure letter (which matched one of the to-be-ignored letters) could be presented. Lure items engendered a reliable PI effect. That is, RTs to lure items were prolonged relative to those to negative probes. One of the critical ERP findings was that, relative to the ERP to the positive probe, which comprised a large, parietally-maximal P3b, the two negative probes elicited anteriorly-distributed N2 components at about 300 ms that were reliably more negative-going than the ERP to the positive probe. Hence, they could have reflected, according to the nomenclature of Folstein and Van Petten (2008), a mismatch process that served to distinguish targets from non-targets. Furthermore, in accord with this hypothesis, the negative probe, which was relatively unfamiliar, was of larger magnitude than the lure item, which was highly familiar. Taking their results as a whole, Du et al. (2008) suggested that the N2 was sensitive to factors influencing the resolution of PI at the probe stage.

Nonetheless, as far as we can determine, investigators have not assessed the neural events that intervene between the cue and the probe. Hence, to enable an assessment of the relation between WM selection processes and PI resolution, we used a design similar to that of Yi et al. (2009). ERPs were recorded to obtain estimates of the timing of selection processes at the cue stage and the processes involved in PI resolution at the probe stage. As noted earlier, in Yi et al.'s (2009) study, there were two types of cues, selection and neutral. The activation difference between them was interpreted as reflecting memory selection. However, WM load could have been confounded with memory selection, because WM load was four items after the neutral cue and two items after the selection cue. Hence, the activity differences between the selection and neutral cues could have been impacted by selection processes as well as the unequal memory load between these conditions. Therefore, if the same brain region were sensitive to selection as well as memory load, activity to the selection cue would tend to increase while activity sensitive to memory load, which is lower in the selection condition, would tend to decrease.

Hence, to differentiate these two effects, a second, neutral-2 condition was added in the current experiment. In this condition, participants had to remember 2 digits initially and maintain them following the neutral-2 cue. Thus, there were three cue conditions. In the selection condition, a 4-digit display was followed by a selection cue indicating which two of the four digits to remember. In the neutral-2 and neutral-4 control conditions, participants had to remember, respectively, either 2 or 4 digits and maintain them following the corresponding cue, after which selection was not necessary. At the probe stage, the participant had to judge whether the probe matched one of the to-be-remembered digits. As the selection and neutral-4 conditions had the same initial memory load, our main focus was on a comparison of these two conditions. Compared to the neutral-4 condition, the selection of relevant information and the inhibition of no-longer-relevant information would be recruited by the selection cue. The advantage of adding the neutral-2 condition was that it would enable a comparison of memory load effects between the neutral-2 and the selection cue. If electrical activity is sensitive to memory load, it should be similar in the selection and neutral-2 conditions. If, by contrast, neural activity is sensitive to selection processes, activity engendered by the selection cue should be enhanced compared to the neutral-2 cue. Hence, we could determine whether the effect was due to memory load or selection.

As noted earlier, the electrical activity underlying PI resolution at the probe stage has more often been assessed than selection-related activity elicited by the cue. Hence, predictions were based mainly on those studies involving PI resolution during the probe stage. Based on the investigations that have observed the frontally-oriented FN component in situations where information must be suppressed or inhibited in order to perform adequately, an expectation was that we might observe a negative-going component at frontal sites when contrasting the selection with the neutral-4 cue, because the former should initiate processes that serve to inhibit the no-longer-relevant information indicated by the cue. However, it should be noted that, if indeed a negativity were to result from this contrast, it could reflect the inhibition or suppression of mental representations of the irrelevant digits. By contrast, the larger negativity we expected at the probe stage to the intrusion relative to the non-intrusion probe would most likely reflect processes instrumental in inhibiting inappropriate motor responses to a highly familiar, but no-longer-relevant digit. Nonetheless, whether this latter activity would be similar to or different from cue-related activity was difficult to predict. Then again, there was some basis for expecting them to be similar because they both could reflect inhibitory processes. On the other hand, they might also be expected to reflect different processes because of the type of events that require inhibition — mental representations in one case and inappropriate motor responses in the other.

Although ERP studies of cue-related activity in WM tasks are rare, there are some studies to which we can turn for aid in predicting the kind of neural activity we might expect to be elicited by the selection cue. For example, Kiss et al. (1998) studied WM using a task in which a series of digits (2 to 9 digits) was presented sequentially followed by a probe digit pair. Participants were required to judge whether the digit pair matched the last 2 digits in the series. As participants did not know when the digit series would end and the probe pair would occur, they were required to maintain the last 2 digits in WM, while removing the no-longer-relevant digits from the focus of attention. As serial position increased, (i.e., up to 9 digits) positive activity over centro-parietal scalp sites was found to increase. This could have been due to the necessity to maintain the last 2 digits within the focus of attention while suppressing or removing the previous digits from WM. Similarly, in the selection condition of the current study, participants had to shift attention to the task-relevant digits indicated by the direction of the arrow while suppressing the task-irrelevant digits. Thus, we predicted a larger centro-parietal positivity to the selection relative to the neutral-2 and neutral-4 cues. This positivity could reflect the internal shifting of attention to the task-relevant information.

Section snippets

Behavioral data

Fig. 1 depicts the behavioral results.

Discussion

One of the main goals of the current study was to understand better the timing of selection-related process in WM engendered by the cue and the subsequent resolution of PI at the probe stage. Indeed, using the precise temporal patterning of the ERP technique, selection-related positive activity was reliable between 330 and 370 ms, and may have had its onset as early as 260 ms over parietal cortex. This activity was then followed, at about 420 ms, by negative-going activity over frontal cortex.

Participants

Twenty-three young adults participated after responding to recruitment ads posted on the internet and in the vicinity of the Columbia University Medical Center. Three participants were removed from the final analyses. We deleted one participant's data because of poor behavioral performance (3 standard deviation outlier from the group mean). Another participant's data were removed because there were less than 20 trials in one or more of the critical averages. A third participant's data were

Acknowledgments

The present work was supported by Grant #AG005213 from the NIA and the New York State Department of Mental Hygiene. We thank Timothy Martin, Julianna Kulik and Beatrice Bleier for assistance with data collection. We thank Dr. Doreen Nessler and Dr. Hoi-Chung Leung for their helpful comments in the discussions of these data. We thank Dr. Edward Smith for his helpful comments on a previous version of this manuscript. We thank Mr. Charles Brown, III for his help in the programming of the

References (56)

  • I. Kiss et al.

    Control processes in verbal working memory: an event-related potential study

    Brain Res.

    (2007)
  • R.T. Knight et al.

    Prefrontal cortex regulates inhibition and excitation in distributed neural networks

    Acta Psychol. (Amst)

    (1999)
  • A. Kok

    Effects of degradation of visual stimulation on components of the event-related potential (ERP) in go/nogo reaction tasks

    Biol. Psychol.

    (1986)
  • J. Lepsien et al.

    Directing spatial attention in mental representations: interactions between attentional orienting and working-memory load

    NeuroImage

    (2005)
  • M. Liotti et al.

    An ERP study of the temporal course of the Stroop color-word interference effect

    Neuropsychologia

    (2000)
  • G. McCarthy et al.

    Scalp distributions of event-related potentials: an ambiguity associated with analysis of variance models

    Electroencephalogr. Clin. Neurophysiol.

    (1985)
  • S. Monsell

    Recency, immediate recognition memory, and reaction time

    Cogn. Psychol.

    (1978)
  • D.E. Nee et al.

    Common and distinct neural correlates of perceptual and memorial selection

    NeuroImage

    (2009)
  • D.E. Nee et al.

    Neural mechanisms of proactive interference-resolution

    NeuroImage

    (2007)
  • F. Perrin et al.

    Spherical splines for scalp potential and current density mapping

    Electroencephalogr. Clin. Neurophysiol.

    (1989)
  • M.D. Rugg et al.

    Event-related potentials and recognition memory

    Trends Cogn. Sci.

    (2007)
  • N.K. Squires et al.

    Two varieties of long-latency positive waves evoked by unpredictable auditory stimuli in man

    Electroencephalogr. Clin. Neurophysiol.

    (1975)
  • W.J. Tays et al.

    General proactive interference and the N450 response

    Neurosci. Lett.

    (2009)
  • R. West

    Neural correlates of cognitive control and conflict detection in the Stroop and digit-location tasks

    Neuropsychologia

    (2003)
  • J.X. Zhang et al.

    Frontal activations associated with accessing and evaluating information in working memory: an fMRI study

    NeuroImage

    (2003)
  • J.X. Zhang et al.

    Electrophysiological correlates of proactive interference in the ‘Recent Probes’ verbal working memory task

    Neuropsychologia

    (2010)
  • A.D. Baddeley

    Working memory

  • A.D. Baddeley

    Exploring the central executive

    Q. J. Exp. Psychol. A

    (1996)
  • Cited by (0)

    View full text