**3. Neural substrates for working memory**

Over many years, numerous researchers have attempted to localize and characterize the neural implementation of VWM and dissociate its functions. Lesion studies have reported that damage to the prefrontal cortex (PFC) in monkeys impairs performance on DMS tasks with a short delay, but not on visual discrimination tasks that do not require maintenance of information (Goldman-Rakic, 1987). Likewise, electrophysiological recording studies of nonhuman primates have revealed sustained neuronal firing in the PFC during the retention interval of DMS tasks, and interpreted the activity as maintaining the previously presented representations (Fuster & Alexander, 1971; Kubota & Niki, 1971). Therefore, the PFC was believed to be the neural substrate for VWM over a longer period. Since then, numerous physiological studies have shown neurons specifically active during the delay period in a vast network of brain regions including the PFC (e.g., Funahashi, Bruce, & Goldman-Rakic, 1989), the posterior parietal cortex (e.g., Chafee & Godman-Rakic, 1998), and visual processing cortices (Bisley & Pasternak, 2000; Miyashita & Chang, 1988).

Consistent with this interpretation, human neuroimaging studies have also revealed that the blood flow in these regions continually increased during the retention interval (Courtney et al., 1997, 1998; Postle & D'Esposito, 1999). Although considerable evidence supports the sustained delay-period activity, DMS tasks include many requirements (e.g., preparation of actions) in addition to maintenance. Therefore, recent fMRI studies have assumed that the blood oxygen level-dependent (BOLD) signal captures a population of neuronal activity that

The Cowan's *K* value is obtained from the set size of each sample display as each subject's

Participants' response

Fig. 1. Combination of participants' response and trial type (change or not) in change

Unlike VWM, the WM capacity for executive functions has been evaluated using dual WM tasks. Although no interference exists between independent storage components such as visual and verbal storage, simultaneously processing more than 2 information sources that require mental manipulation and reactions is thought to be difficult. Previous studies have revealed a psychological refractory period, in which a second task elicits a longer reaction time when the interval between the first and second tasks (i.e., stimulus onset asynchrony; SOA) is short (Marois & Ivanoff, 2005). That is to say, if the 2 tasks seem to be processed simultaneously, the performance is degraded. This phenomenon is known to be a bottleneck

Over many years, numerous researchers have attempted to localize and characterize the neural implementation of VWM and dissociate its functions. Lesion studies have reported that damage to the prefrontal cortex (PFC) in monkeys impairs performance on DMS tasks with a short delay, but not on visual discrimination tasks that do not require maintenance of information (Goldman-Rakic, 1987). Likewise, electrophysiological recording studies of nonhuman primates have revealed sustained neuronal firing in the PFC during the retention interval of DMS tasks, and interpreted the activity as maintaining the previously presented representations (Fuster & Alexander, 1971; Kubota & Niki, 1971). Therefore, the PFC was believed to be the neural substrate for VWM over a longer period. Since then, numerous physiological studies have shown neurons specifically active during the delay period in a vast network of brain regions including the PFC (e.g., Funahashi, Bruce, & Goldman-Rakic, 1989), the posterior parietal cortex (e.g., Chafee & Godman-Rakic, 1998), and visual

Consistent with this interpretation, human neuroimaging studies have also revealed that the blood flow in these regions continually increased during the retention interval (Courtney et al., 1997, 1998; Postle & D'Esposito, 1999). Although considerable evidence supports the sustained delay-period activity, DMS tasks include many requirements (e.g., preparation of actions) in addition to maintenance. Therefore, recent fMRI studies have assumed that the blood oxygen level-dependent (BOLD) signal captures a population of neuronal activity that

processing cortices (Bisley & Pasternak, 2000; Miyashita & Chang, 1988).

detection paradigm (left) and a model of Cowan's formula (right).

VWM capacity

VWM capacity for a given material.

of the central executive function.

**3. Neural substrates for working memory** 

may reflect the representation of multiple items to be maintained, and have indeed shown that a subset of the distributed network demonstrated delay-period activity sensitive to the number of items in the sample display (Diwadkar et al., 2000; Glahn et al., 2002; Jha & McCarthey 2000; Linden et al., 2003). The VWM load-sensitive network includes the frontal, parietal, and visual cortices. Notably, some studies have revealed that activity in the posterior parietal cortex is correlated with the number of items to be remembered (Cowan's *K* value) and indicated that this area actually stored the representations (Kawasaki et al., 2008; Todd & Marois, 2004, 2005; Vogel & Machizawa, 2004; Xu & Chun, 2006).

In contrast to the posterior parietal and visual cortices, anterior regions including the frontal cortex have also been associated with executive processes such as attentional selection and manipulation of information (Curtis & D'Esposito, 2003). For instance, in studies using a spatial WM task that requires participants to memorize the spatial locations of simultaneous or sequentially presented items and, after a delay, select one relevant location, the prefrontal cortex has been reported to show transient activity during the selection period and no sustained activity during the retention interval (Rowe et al., 2000). Furthermore, the frontal cortex is particularly sensitive to the number of listed items to be maintained in VWM in the n-back task, which requires participants to maintain a series of items and their order, select a relevant item from VWM, and compare it with the earlier item (Smith & Jonides, 1999). Moreover, the frontal cortex is proposed to serve in maintaining task-specific goals (Miller & Cohen, 2001; Passingham & Sakai, 2004) and assist in maintaining high loads and/or long retention intervals (Braver et al., 1997; Linden et al., 2003).

Although, thus far, many neuroimaging studies have identified the neural substrate for the storage systems and central executive of WM, they have not dealt with how these brain areas temporally communicate. To address this issue, some studies have investigated the dynamic relationships governing brain activity by focusing on electroencephalograph (EEG) oscillations, which are closely related to synchronization of a large number of neurons underlying a particular function (Varela et al., 2001). Previous human scalp-recorded EEG studies have revealed modulated theta (about 4–8 Hz) and alpha (about 9–12 Hz) rhythms in distributed brain regions and phase synchronization between them during various WM tasks (Jensen & Tesche, 2002; Kawasaki & Watanabe, 2007; Klimesch et al., 2008; Mizuhara et al., 2004; Sauseng et al., 2005). Frontal theta activity in particular has been associated with the mental manipulation of WM, because these oscillations were enhanced in tasks such as mental calculation and image transfomation (Kawasaki et al., 2010). In contrast, posterior alpha activities are thought to be involved in the WM storage systems, because these oscillations are mainly observed in the retention intervals of many WM tasks. However, whether these oscillatory activities are increased or decreased during each WM period remains controversial. Furthermore, little is known regarding how these oscillations represent WM limitations; therefore, their detailed mechanisms have not yet been identified. To clarify the functional role of the theta and alpha oscillations in WM, the study described in the following 2 sections used EEG data measured during DMS and dual WM tasks to demonstrate 2 types of EEG activity that were correlated with the WM capacities for visual storage and central executive systems.

#### **4. EEG oscillations for visual storage capacity**

This section describes the investigation of EEG oscillatory activity correlated with VWM capacity, which aimed to identify the roles of different oscillations in the VWM storage

Human Oscillatory EEG Activities Representing Working Memory Capacity 255

Institute for Neural Computation, University of California, San Diego, CA) running under Matlab (Mathworks, Natick, MA). ICA components that were significantly correlated with vertical or horizontal EOGs were regarded as components related to eye movement or other artifacts and were reduced or eliminated from the data. The ICA-corrected data were

To accurately evaluate cortical activity under the scalp EEG electrodes without error due to volume conduction, we used a current source density analysis at each electrode position. The spherical Laplace operator was applied to the voltage distribution on the surface of the scalp using the following parameters: the order of the spline, m = 4, and the maximum degree of the Legendre polynomial, n = 50, with a precision of 10–5 (Perrin et al., 1989). Time-frequency (TF) amplitudes and phases were calculated by wavelet transforms based on Morlet's wavelets, having a Gaussian shape in the time domain (SD t) and frequency domain (SD f) around a center frequency (*f*) (Tallon-Baudry et al., 1997). The TF amplitude *E*(*t*, *f*) for each time point of each trial was the squared norm of the result of the convolution

where f = 1/(2t). The wavelet used was characterized by a constant ratio (f/f = 7), with *f* ranging from 1 Hz to 40 Hz in 0.5-Hz steps. The TF amplitude was averaged across single trials for events and conditions. The event-related TF amplitude was calculated by subtracting the baseline data measure in the ITI for each frequency band. For all statistical analyses, a nonparametric Wilcoxon signed-rank test was used across the events or conditions because the

Accuracy rates (percent correct) for lower numbers of presented objects were higher than those for larger numbers of presented objects (3 objects: 90.2 ± 2.0%; 6 objects: 72.6 ± 2.8%). A one-factor analysis of variance (ANOVA) revealed a main effect of the number of objects (*F*1, 26 = 24.3, *P* < 0.01) and the accuracy rates demonstrated a significant difference

The VWM capacity was estimated by Cowan's *K* formula (see Section 2; 3 objects: *K* = 2.41 ± 0.12; 6 objects: *K* = 2.71 ± 0.33). A one-factor ANOVA revealed no main effect of the number of objects (*F*1, 26 = 0.64, *P* = 0.43), and no significant difference between K-values was detected between 3 and 6 objects (*Z* = 1.18, *P* = 0.24). These results suggested that the VWM

Brain activity was evaluated using the averaged time-frequency amplitudes of the EEG data obtained during the DMS task. The EEG results demonstrated that parietal alpha amplitudes (about 12 Hz) sustainably and significantly increased during the retention intervals (POz electrode: *Z* = 2.11, *P* < 0.04), whereas enhancement of the frontal theta delayperiod amplitudes (about 6 Hz) was not observed (Fz electrode: *Z* = 0.18, *P* = 0.85). Frontal theta activity during maintenance of 6 objects was significantly higher than that for maintenance of 3 obejcts (3 objects: -0.28 ± 0.21 V; 6 objects: 0.55 ± 0.40 V; *Z* = 2.12, *P* < 0.04). In contrast, parietal alpha activity demonstrated an opposing pattern (3 objects: 2.06 ±

1/2 2 2 ( , ) ( ) exp( / 2 )exp( 2 ) *w t t t <sup>f</sup> t i ft* (4)

<sup>2</sup> *E t*(, ) (, ) () *f w t f s t* (5)

of the original EEG signal *s*(*t*) with the complex Morlet's wavelet function *w*(*t*, *f*):

distributions of the TF amplitude populations were far from Gaussian.

capacity in our experiments was limited to approximately 2.7 objects.

(Wilcoxon signed-rank test; *Z* = 3.71, *P* < 0.01).

**4.3 Results** 

recalculated using regressions on the remaining components.

systems (e.g., maintenance of high or low VWM demands). EEG data was measured during the DMS task.

### **4.1 Delayed matching to sample task**

Fourteen healthy, right-handed volunteers (10 male and 4 female; mean age = 25.6 ± 4.2 years, range 21–38 years) with normal or corrected-to-normal visual acuity, normal hearing acuity, and normal motor performance took part in the delayed matching to sample tasks. All participants gave written informed consent, which was approved by the Ethical Committee of the RIKEN (in accordance with the Declaration of Helsinki), before the experiments were performed.

Participants faced a computer screen and were asked to memorize the colors of 3 or 6 colored disks (size, 1° × 1°; color, white, red, green, blue, yellow, magenta, cyan, or orange) that were distributed at random locations within an invisible 3 × 3 cell matrix in a black rectangle (size, 10° × 10°) for 0.2 s (Fig. 2, sample display). After a 2-s retention interval, one disk was presented at one location within the sample array (test display), and participants were asked to judge whether its color matched the disk at the same location in the sample display via a button press while the fixation point was red for 2 s. In one trial, the color of the probe disk matched the sample disk, and in a second trial, the color of the probe disk did not match. After the judgment, a feedback stimulus indicating whether the answer was correct (O) or incorrect (X) was presented. The duration of the inter-trial interval (ITI) was 2 s. Each participant completed 4 separate sessions which consisted of 48 trials. A behavioral training session before the EEG-measurement sessions was provided for all participants.

Fig. 2. Task procedure for 1 trial of the delayed-matching-to-sample task.

#### **4.2 EEG measurements and analyses**

An EEG was continuously recorded using 60 scalp electrodes embedded in an electrode cap in accordance with the extended version of the International 10/20 System of Electrode Placement. The sampling rate was 500 Hz. Reference electrodes were placed on the right and left earlobes. Artifacts due to eye blinks and movements were detected by electrooculogram (EOG) electrodes placed above and below the left eye to monitor eye blinks and vertical eye movements, and electrodes placed 1 cm from the right and left eyes to monitor horizontal eye movements. Trials in which the amplitude of any electrode of an EEG epoch exceeded plus or minus 100 V were rejected from the offline analysis. These EEG data were amplified using NeuroScan equipment (Compumedics NeuroScan Corp., Charlotte, NC) and filtered with a band-pass range from 0.1 Hz to 50 Hz.

We analyzed the EEG data for the correct trials. These epochs were subjected to infomax independent component analysis (ICA) with the use of EEGLAB (Delorme & Makeig, 2004; Institute for Neural Computation, University of California, San Diego, CA) running under Matlab (Mathworks, Natick, MA). ICA components that were significantly correlated with vertical or horizontal EOGs were regarded as components related to eye movement or other artifacts and were reduced or eliminated from the data. The ICA-corrected data were recalculated using regressions on the remaining components.

To accurately evaluate cortical activity under the scalp EEG electrodes without error due to volume conduction, we used a current source density analysis at each electrode position. The spherical Laplace operator was applied to the voltage distribution on the surface of the scalp using the following parameters: the order of the spline, m = 4, and the maximum degree of the Legendre polynomial, n = 50, with a precision of 10–5 (Perrin et al., 1989).

Time-frequency (TF) amplitudes and phases were calculated by wavelet transforms based on Morlet's wavelets, having a Gaussian shape in the time domain (SD t) and frequency domain (SD f) around a center frequency (*f*) (Tallon-Baudry et al., 1997). The TF amplitude *E*(*t*, *f*) for each time point of each trial was the squared norm of the result of the convolution of the original EEG signal *s*(*t*) with the complex Morlet's wavelet function *w*(*t*, *f*):

$$w(t,f) = (\sigma\_t \sqrt{\pi})^{-1/2} \exp(-t^2 \;/\ 2\sigma\_t^2) \exp(i2\pi ft) \tag{4}$$

$$E(t,f) = \left| w(t,f) \otimes s(t) \right|^2\tag{5}$$

where f = 1/(2t). The wavelet used was characterized by a constant ratio (f/f = 7), with *f* ranging from 1 Hz to 40 Hz in 0.5-Hz steps. The TF amplitude was averaged across single trials for events and conditions. The event-related TF amplitude was calculated by subtracting the baseline data measure in the ITI for each frequency band. For all statistical analyses, a nonparametric Wilcoxon signed-rank test was used across the events or conditions because the distributions of the TF amplitude populations were far from Gaussian.

#### **4.3 Results**

254 Neuroimaging – Cognitive and Clinical Neuroscience

systems (e.g., maintenance of high or low VWM demands). EEG data was measured during

Fourteen healthy, right-handed volunteers (10 male and 4 female; mean age = 25.6 ± 4.2 years, range 21–38 years) with normal or corrected-to-normal visual acuity, normal hearing acuity, and normal motor performance took part in the delayed matching to sample tasks. All participants gave written informed consent, which was approved by the Ethical Committee of the RIKEN (in accordance with the Declaration of Helsinki), before the

Participants faced a computer screen and were asked to memorize the colors of 3 or 6 colored disks (size, 1° × 1°; color, white, red, green, blue, yellow, magenta, cyan, or orange) that were distributed at random locations within an invisible 3 × 3 cell matrix in a black rectangle (size, 10° × 10°) for 0.2 s (Fig. 2, sample display). After a 2-s retention interval, one disk was presented at one location within the sample array (test display), and participants were asked to judge whether its color matched the disk at the same location in the sample display via a button press while the fixation point was red for 2 s. In one trial, the color of the probe disk matched the sample disk, and in a second trial, the color of the probe disk did not match. After the judgment, a feedback stimulus indicating whether the answer was correct (O) or incorrect (X) was presented. The duration of the inter-trial interval (ITI) was 2 s. Each participant completed 4 separate sessions which consisted of 48 trials. A behavioral training session before the EEG-measurement sessions was provided for all participants.

Fig. 2. Task procedure for 1 trial of the delayed-matching-to-sample task.

An EEG was continuously recorded using 60 scalp electrodes embedded in an electrode cap in accordance with the extended version of the International 10/20 System of Electrode Placement. The sampling rate was 500 Hz. Reference electrodes were placed on the right and left earlobes. Artifacts due to eye blinks and movements were detected by electrooculogram (EOG) electrodes placed above and below the left eye to monitor eye blinks and vertical eye movements, and electrodes placed 1 cm from the right and left eyes to monitor horizontal eye movements. Trials in which the amplitude of any electrode of an EEG epoch exceeded plus or minus 100 V were rejected from the offline analysis. These EEG data were amplified using NeuroScan equipment (Compumedics NeuroScan Corp., Charlotte, NC)

We analyzed the EEG data for the correct trials. These epochs were subjected to infomax independent component analysis (ICA) with the use of EEGLAB (Delorme & Makeig, 2004;

**4.2 EEG measurements and analyses** 

and filtered with a band-pass range from 0.1 Hz to 50 Hz.

the DMS task.

**4.1 Delayed matching to sample task** 

experiments were performed.

Accuracy rates (percent correct) for lower numbers of presented objects were higher than those for larger numbers of presented objects (3 objects: 90.2 ± 2.0%; 6 objects: 72.6 ± 2.8%). A one-factor analysis of variance (ANOVA) revealed a main effect of the number of objects (*F*1, 26 = 24.3, *P* < 0.01) and the accuracy rates demonstrated a significant difference (Wilcoxon signed-rank test; *Z* = 3.71, *P* < 0.01).

The VWM capacity was estimated by Cowan's *K* formula (see Section 2; 3 objects: *K* = 2.41 ± 0.12; 6 objects: *K* = 2.71 ± 0.33). A one-factor ANOVA revealed no main effect of the number of objects (*F*1, 26 = 0.64, *P* = 0.43), and no significant difference between K-values was detected between 3 and 6 objects (*Z* = 1.18, *P* = 0.24). These results suggested that the VWM capacity in our experiments was limited to approximately 2.7 objects.

Brain activity was evaluated using the averaged time-frequency amplitudes of the EEG data obtained during the DMS task. The EEG results demonstrated that parietal alpha amplitudes (about 12 Hz) sustainably and significantly increased during the retention intervals (POz electrode: *Z* = 2.11, *P* < 0.04), whereas enhancement of the frontal theta delayperiod amplitudes (about 6 Hz) was not observed (Fz electrode: *Z* = 0.18, *P* = 0.85). Frontal theta activity during maintenance of 6 objects was significantly higher than that for maintenance of 3 obejcts (3 objects: -0.28 ± 0.21 V; 6 objects: 0.55 ± 0.40 V; *Z* = 2.12, *P* < 0.04). In contrast, parietal alpha activity demonstrated an opposing pattern (3 objects: 2.06 ±

Human Oscillatory EEG Activities Representing Working Memory Capacity 257

memory condition, the participants were required to repeat the mental manipulation 4 times, and then determine whether the position of the red circle which they mentally moved matched a probe visual stimulus (test display). In half of the trials, the probe stimulus matched the mental representation. In the remaining trials, the wrong probe was presented by changing only the fourth direction of movement from the initial position. The participants were asked to indicate via button press whether the probe stimulus was correct or not while the fixation point was red for 2 s. The duration of the ITI was 2 s. The size of the red circle and gridded squares was 1º 1º and 5º 5º (1º 1º per square), respectively. For the dual WM task, the participants were asked to complete an auditory WM task simultaneously to the visual tast (Fig. 3B). When the visual stimuli described above were presented on the computer screen, a word indicating a one-digit number was simultaneously presented as the auditory stimulus through the headphones of both ears for 1 s (sample display). The auditory WM task requried the participants to memorize and maintain the presented number with rehearsal in their minds and, after a 2-s retention interval, to update the number by adding the another presented one-digit number for 2 s. After this a total of 4 incidences of auditory and visual manipulation, auditory and visual stimuli were simultaneously presented again, and participants were required to judge whether or not they were identical to the manipulated mental representation for both auditory and visual tasks (test display). In half of the trials, both the auditory and visual probe stimulus matched the mental representations. In the remaining trials, the incorrect probe for either the auditory or visual stimulus was presented, similar to the single VWM condition. The button press, duration of the inter-trial interval, and creation of the stimuli

Fig. 3. Task procedure for one trial of the single visual WM (A) and dual WM (B) tasks.

All participants performed all the WM tasks with high accuracy rates (mean accuracy rate (± s.d.), 97.3 ± 4.7% and 91.1 ± 7.1% for visual and dual WM conditions, respectively).

were identical to the single WM condition.

**5.2 EEG measurements and analyses** 

**5.3 Results** 

The same methods were used as described in Section 4.2.

0.66 V; 6 objects: 0.45 ± 0.45 V; *Z* = 1.97, *P* < 0.05). Interestingly, frontal theta activity was significantly and positively correlated with the VWM capacity of the individual (Fz electrode: *r*(14) = 0.39, *P* < 0.05), whereas the parietal alpha activity was negatively correlated with the VWM capacity (Poz electrode: *r*(14) = -0.44, *P* < 0.05).
