Child and adolescent development of the brain oscillatory activity during a working memory task

Brain and Cognition 167 (2023) 105969 Available online 21 March 2023 0278-2626/© 2023 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/bync-nd/4.0/). Child and adolescent development of the brain oscillatory activity during a working memory task Carlos M. G´ omez a , * , Vanesa Mu˜ noz a , Elena I. Rodríguez-Martínez a , Antonio Arjona a , Catarina I. Barriga-Paulino b , Santiago Pelegrina c a Human Psychobiology Laboratory, Experimental Psychology Department, University of Sevilla, C/ Camilo Jos´ e Cela S/N, 41018 Sevilla, Spain b Center for Biomedical Research, Universidade do Algarve, Portugal c Department of Psychology, University of Jaen, Jaen, Spain ARTICLE INFO Keywords: Working memory Brain oscillations Delayed match-to-sample test Ontogenetic development Developmental trajectory ABSTRACT The developmental trajectories of brain oscillations during the encoding and maintenance phases of a Working Memory (WM) task were calculated. The Delayed-Match-to-Sample Test (DMTS) was applied to 239 subjects of 6–29 years, while EEG was recorded. The Event-Related Spectral Perturbation (ERSP) was obtained in the range between 1 and 25 Hz during the encoding and maintenance phases. Behavioral parameters of reaction times (RTs) and response accuracy were simultaneously recorded. The results indicate a myriad of transient and sustained bursts of oscillatory activity from low frequencies (1 Hz) to the beta range (up to 19 Hz). Beta and Lowfrequency ERSP increases were prominent in the encoding phase in all age groups, while low-frequency ERSP indexed the maintenance phase only in children and adolescents, but not in late adolescents and young adults, suggesting an age-dependent neural mechanism of stimulus trace maintenance. While the latter group showed Beta and Alpha indices of anticipatory attention for the retrieval phase. Mediation analysis showed an important role of early Delta-Theta and late Alpha oscillations for mediation between age and behavioral responses performance. In conclusion, the results show a complex pattern of oscillatory bursts during the encoding and maintenance phases with a consistent pattern of developmental changes. 1. Introduction Working memory (WM) allows encoding, maintaining, recovering, and manipulating the information for the operations necessary to solve a task (Baddeley, 2012). When the manipulative aspect is not involved and the visual modality is used, the term visual short-term memory is also employed. In the present report, we will use the term WM because, in the Delayed Match-to-Sample Test (DMTS), like the Sternberg task, selecting among possible targets or deciding whether the current stimulus was previously presented can be considered an operation; although certainly, more complex operations as in N-back tasks could add the WM updating process. WM decays in a few seconds and has a limited storage capacity. These characteristics limit the possibilities of WM to process information. WM can hold information from several sources, such as long-term memory (LTM), newly acquired sensory information (external or interoceptive), and the products of cognitive processing, including the results of operations performed in WM. WM enables the performance of more complex processes such as learning, language, reading, reasoning, etc. Therefore, it is necessary for daily and curricular activities. The following operational definition summarizes most of the aspects described above. “Working memory refers to the retention of information in conscious awareness when this information is not present in the environment, to its manipulation, and to its use in guiding behavior” (Postle, 2006). Improvement in WM performance with age is associated with maturational brain changes. Age-related increases in structural connectivity within the frontal cortex and frontoparietal networks have been associated with age-related improvements in WM (Ostby et al., 2011; Zhou et al., 2014). Frontoparietal cortical thinning, a neural marker of brain maturation with age, is also associated with age-related WM improvement (Ostby et al., 2011). Regarding developmental changes in brain activation during WM, a meta-analysis comparing adolescents and young adults (Andre et al., 2015) showed increased activation of the WM network, including the left precuneus, inferior parietal * Corresponding author. E-mail addresses: [email protected] (C.M. G´ omez), [email protected] (V. Mu˜ noz), [email protected] (E.I. Rodríguez-Martínez), [email protected] (A. Arjona), cipaulino@ ualg.pt (C.I. Barriga-Paulino), [email protected] (S. Pelegrina). Contents lists available at ScienceDirect Brain and Cognition journal homepage: www.elsevier.com/locate/b&c https://doi.org/10.1016/j.bandc.2023.105969 Received 11 October 2022; Received in revised form 13 February 2023; Accepted 9 March 2023 Brain and Cognition 167 (2023) 105969 2 gyri, and the middle frontal gyrus. Furthermore, decreased activation in the right superior frontal gyrus, the left junction of the inferior parietal and postcentral gyri, and the left posterior cingulate gyrus was associated with age-related increases in WM performance. A meta-analysis of young children has shown that activations are not only limited to frontoparietal networks but also include the insula and the cerebellum (Yaple & Arsalidou, 2018). The role of the cerebellum in WM during development has also been demonstrated in cerebellar lesions in children (Vaquero et al., 2008). Electroencephalography (EEG) and magnetoencephalography (MEG) allow to make inferences about the timing and anatomical location of WM subprocesses, although with low resolution due to the phenomenon of volume conduction, and can be used in children and adolescents during WM experimental paradigms such as the DMTS, Sternberg task, and N-back. In the present report, we will focus on reviewing the results of the DMTS (and the Sternberg task), since the main goal is to analyze time–frequency changes in children, adolescents, and young adults during the DMTS task. Event-related potentials (ERPs) have been used to analyze the neurophysiology of WM development in the DMTS task. A reduction in amplitude with age has been found in visual ERPs during the encoding and retrieval phases (Barriga-Paulino et al., 2014, 2017). During the maintenance phase, a slow component appears, that has been associated with the maturation of the previously described negative slow wave (NSW) recorded during the retention phase in adults (Ruchkin et al., 1990). Neural source analysis of the NSW showed a more intense frontal Fig. 1. Mediation model. Graphic display of the mediation model for the relationship between age and behavioral measures (RTs or correct responses) mediated by ERSP parameters. Fig. 2. Regression of behavioral parameters with respect to age. The figure displays: 2A. Reaction Times (RTs), 2B. Percentage of incorrect responses, 2C. Percentage of omissions, 2D. Percentage of anticipations, and 2E. Percentage of correct responses linearly and inversely regressed with respect to age in days. Only the inverse regression is displayed, as it has a better fit. The subjects belonging to the different considered groups are indicated. Behavioral parameters of males and females were regressed together but displayed with a different color. C.M. G´ omez et al. Brain and Cognition 167 (2023) 105969 3 neural recruitment to retain items in WM with increasing age (Barriga- Paulino et al., 2014, 2015). The reduction in ERPs with age in the DMTS WM task was also observed in N-back (Pelegrina et al., 2020) suggesting the generality of this age effect, probably due to the more general process of synaptic pruning (Huttenlocher, 1990). Indeed, an inverse relationship of age with frontal-parietal gray matter volume (a proxy for synaptic pruning) and amplitude of low-frequency EEG bands has been found (Whitford et al., 2007), suggesting that synaptic pruning should be related to a reduction in scalp EEG amplitude due to a reduction in intracerebral EEG signal generators with age. Moreover, the reduction of scalp EEG with age has been confirmed with MEG, indicating that the reduction of scalp EEG would not be related to changes in skull thickness and/or mineralization with age (G´ omez et al., 2017). Rhythmic EEG signals generated by interconnected neurons produce brain oscillations at rest or linked to stimuli or responses. The theta band has been one of the most conspicuous brain oscillations related to WM. For instance, using intracranial recordings Raghavachari et al. (2001) showed an increase in theta band Event-Related Spectral Perturbation (ERSP) during a verbal Sternberg task, including the maintenance phase. The Sternberg task is similar in some aspects to the DMTS task. One of the important mechanisms of how theta would operate in WM tasks is through theta phase coupling with gamma amplitude during the retention period of the DMTS task (Sauseng et al., 2010), whereas alpha amplitude was associated with suppression of distracting information (Sauseng et al., 2009). Indeed, brain stimulation entrainment of contralateral frontal theta and ipsilateral alpha to the target to be remembered enhances WM performance in a retro-cue visuospatial memory task (Riddle et al., 2020). The role of theta in WM is not only related to item retention, but also to the subsequent operation to be performed: Theta increased during the expectancy period of a cued task, where subjects have to prepare for a cued no-go, prosaccade, or antisaccade task (Cordones et al., 2013). Theta is also related to the recovery of previously presented items, with the number of theta oscillation periods during the retrieval phase being related to the number of items retained in WM (Tesche & Karhu, 2000). One of the possible roles of the theta band in WM would be the nesting of higher frequency oscillations such as the gamma band (Buzsaki, 2006). Concerning the underlying mechanisms of WM for encoding the item trace, not only the direct effects of electrophysiological activities such as the ERPs and brain oscillations have been proposed, but also the facilitation of the synaptic pathways related to the WM processing, at least for the items outside of the attentional focus. The latter would provide a more efficient mechanism for encoding WM-related information (LaRocque et al., 2013). Using the DMTS and Sternberg tasks, several authors have obtained evidence of neural dynamics associated with WM operation, mostly in adults and in some cases comparing controls with subjects diagnosed with several neural impairments and diseases. Koshy et al. (2020) showed a decrease in alpha and beta (8–18 Hz) during the encoding and maintenance phases of a numerical task. A similar result was obtained during encoding of an item for retention in WM, suggesting a decrease in alpha and beta Event-Related Synchronization (Zammit et al., 2018) for both, young adults and adolescents, but there were no differences between both age groups. Lenartowicz et al. (2014; 2019) using EEG in children and adolescents showed a decrease in midoccipital alpha during encoding and an increase in midfrontal theta during the maintenance phase, accompanied by an increase in alpha during the maintenance period. A result also observed by Arjona et al. (2023) in children and adolescents. The increase of alpha synchronization during the maintenance period has also been described during a DMTS task in children (Sato et al., 2018). More recently, a reduction of gamma power and gamma power variability across trials with age have also been described during the DMTS task (McKeon et al., 2023). In an object recognition task, similar to DMTS, an increase in theta and delta power and a decrease in alpha and beta power was recorded in the encoding phase (Kang et al., 2018). During the maintenance phase, a theta increase and a beta decrease were observed. Alpha/beta desynchronization was also evident in a Sternberg task, in which all items to be remembered were presented simultaneously. Interestingly, while frontal and temporal alpha/beta desynchronization was maintained up to one hundred milliseconds before retrieval, occipital alpha/beta encoding Fig. 3. ERSP absolute values during the baseline period. 3A. Display the ERSP of all subjects, ordered by age, in a color code for the electrode Pz. 3B. Regression of the collapse of the ERSP for all frequencies and electrodes with respect to age using linear and inverse models, only the inverse correlation is shown, as it has a better fit. 3C. Spearman correlation values (Rho) of the ERSP values with respect to the inverse of age for all the electrodes. 3D. Log10 (p) of the Spearman correlation showed in 3C. The arrow indicates the limit for significance level after applying the FDR correction. C.M. G´ omez et al. Brain and Cognition 167 (2023) 105969 4 desynchronization became synchronized during the maintenance phase (Heinrichs-Graham & Wilson, 2015). Although an increase of theta and gamma during the delay period has generally been obtained, an increase of gamma and reduced theta in areas close to sequentially coregistered increased fMRI activity has also been reported (Khursheed et al., 2011). Although the improvement of performance with age in the WM DMTS task is well-established, as its relationship with the NSW, the developmental trajectory of brain oscillations, and its possible relationship with WM behavioral parameters are scarcely represented in the scientific literature (reviewed in G´ omez et al., 2018). Therefore, the objectives of the present report were to obtain the ERSP increases and decreases (synchronization and desynchronization, respectively), during the encoding and maintenance periods of the DMTS task, and to analyze its relationship with age and behavioral parameters. This approach would allow us to trace the developmental trajectory of transient oscillatory bursts related to the WM operation. It would be particularly important to define the electrophysiological indices of item retention in WM during the maintenance period, and its possible relationship with age, complementing previous results obtained with ERPs. The ERSP results previously reviewed suggest a synchronization of theta accompanied by a desynchronization of alpha and beta during encoding and a return to baseline at the end of the maintenance period for alpha and beta. There is no particular hypothesis about the developmental trajectory of brain oscillations during WM operation; however, one possibility is that brain oscillations follow the well-known decrease in absolute power with age that has been described for EEG oscillatory behavior during resting state (Segalowitz et al., 2010). A relationship between brain oscillation power and behavioral responses is also expected. 2. Methods 2.1. Sample Two hundred and fifty-one children, adolescents, and young adults participated in the study, recruited through advertisements placed in Fig. 4. Average of Absolute Event-Related Spectral Perturbation (ERSP) panels and topography for Low-Delta1, Delta-Theta1, Alpha1, and Beta1 during the encoding phase. 4A. Time-frequency display for 2 selected electrodes and the average of all electrodes. 4B. Absolute ERSP topographies. The selected values for the time–frequency windows appear in the figure. The arrows in 4A and 4B indicate the time–frequency window and location of the selected time–frequency windows. 4C. Absolute ERSP across time of the selected frequency windows for the three groups of age: children; adolescents; late adolescents and young adults. The overlying lines in 4C correspond to the ERSP time windows selected for topographies in 4B and subsequent analyses. L-D1: Low-Delta1, D-T1: Delta-Theta1, A1: Apha1, B1: Beta1, LA-YA: Late adolescents and young adults. C.M. G´ omez et al. Brain and Cognition 167 (2023) 105969 5 different schools and at the University of Sevilla. All participants reported no neurological or psychological diseases or impairments. Subjects who did not reach a minimum of 10 artifact-free trials were eliminated from the experimental sample, resulting in a final sample of 239 (121 males/118 females, from 6 to 29 years old; mean =14.06 years old, SD = ±6.06, with a higher number of subjects per year in children and adolescents, see Supplementary Fig. 1 for the histogram of subject’s age). For some data analysis and visualization, the total sample was divided into three age groups: children (from 2114 to 4007 days of age equivalent to 6–10 years, n =73), adolescents (4051–5787 days equivalent to 11–15 years, n =68) and late adolescents-young adults (LA-YA group) (5911–10804 days equivalent to 16–29 years, n =98). Experiments were conducted with informed and written parental consent, in accordance with the World Medical Association’s Code of Ethics (Declaration of Helsinki) for experiments involving humans https://www.wma.net/en/30publications/10policies/b3/index.htm. The experimental protocol was approved by the biomedical ethical committee of the Andalucía autonomous region. Fig. 5. Average of Absolute Event-Related Spectral Perturbation (ERSP) panels and topography for Low-Delta2, Delta2, Theta2, Alpha2, and Beta2. during the maintenance phase. 5A, B, and C as in Fig. 4. L-D2: Low-Delta2, D2: Delta2, T2: Theta2, A2: Alpha2, B2: Beta2, LA-YA: Late adolescents and young adults. C.M. G´ omez et al. Brain and Cognition 167 (2023) 105969 6 2.2. Experimental session 2.2.1. Delayed match-to-sample task The experimental session consisted of 128 trials (see supplementary Fig. 2), organized into 4 experimental blocks, each with 32 trials. The trial started with the presentation of the first stimulus (S1), located in the center of the screen, with a visual angle of 4.56 degrees on the horizontal meridian. S1 was presented for 1000 ms and had to be memorized by the subject (encoding phase). Next, a screen with a black background and a white fixation cross in the center appeared for 1500 ms; a period during which the subject had to keep the S1 in memory (maintenance phase). Then, two stimuli (S2) appeared (one of them the same image as the S1, and the other a new stimulus), one placed on the left side and the other on the right side of the screen. S2 was presented for 2000 ms (selection, matching, response, or retrieval phase). Retrieval will be the term used in the present report. The subject responded by pressing the left button (with his left hand) or the right button (with his right hand) of the answer box, depending on whether S2 equal to S1 appeared on the left or right side of the screen, respectively. The number of trials on each side were randomly counterbalanced. The presentation of the essays was also randomized so each subject played a different sequence. After the response, and at the end of the essay, the subject heard auditory feedback: a sound with a positive connotation if he/she hits the trial and one with a negative connotation if the response failed. Then a new trial started. Before starting the task, a practice block, consisting of 10 trials was provided to familiarize the subject with the test and ensure that the instructions were well understood. If the subject did not understand the task correctly, the practice block was repeated. The duration of the task was approximately 17 min. The S1 presentation would permit the study of encoding, the delay period for the maintenance phase and the S2 presentation would permit the study of the retrieval. The latter phase would not be studied in the present report. 2.2.2. EEG recording EEG was recorded during the DMTS task. Subjects were instructed to look at the screen, blink as little as possible, focus on the cross presented at the center of the screen, and follow the response instructions. Recordings were obtained from 32 scalp sites using the 10–20 international system and an average reference (FP1, FPz, FP2, F7, F3, Fz, F4, F8, FC5, FC1, FC2, FC6, M1, T7, C3, Cz, C4, T8, M2, CP5, CP1, CP2, CP6, P7, P3, Pz, P4, P8, POz, O1, Oz and O2). Tin electrodes mounted on an electrode cap (ELECTROCAP) were employed. Eye movements were recorded by four electrodes: two electrodes placed at the outer canthus of each eye for horizontal movements, and two electrodes placed above and below the left eye for vertical movements. All the scalp electrodes were rereferenced off-line to the average mastoid ((M1 +M2)/2), and impedance was maintained below 10 kΩ. Data were recorded using a commercial analog–digital acquisition and analysis board (ANT) in direct current mode at 512 Hz with a 20,000 amplification gain. No filtering was applied during the recording. 2.3. Data preprocessing EEG recordings were analyzed with EEGLAB2021.0 (Delorme & Makeig, 2004), MATLAB R2021b (MathWorks Inc., MA, USA), and FieldTrip-20210507 (Oostenveld et al., 2011). Data were filtered with a lowpass filter of 150 Hz and a high-pass filter of 0.2 Hz. The fifty Hz power line artifact was removed by employing the pop_cleanline function. Eye movements, blinking, and muscle artifacts were rejected with an independent component analysis (ICA). Independent components were identified by visual inspection (Delorme et al., 2007). After EEG reconstruction, all epochs in which the EEG exceeded ±150 ˆ AµV in any Table 1 P-values for the comparison with respect to baseline. The sign +or – indicates if it was higher or lower than the baseline. FDR corrected. p Low-Delta1 <0.001, + Low-Delta2 0.052, + Delta2 0.003, + Delta-Theta1 <0.001, + Theta2 0.059, + Alpha1 <0.001, - Alpha2 0.004, + Beta1 <0.001, + Beta2 O2 0.059, + Beta2 Fz 0.167, + Table 2 P-values for the comparison with respect to baseline divided by age group (Children; Adolescents; Late adolescents and young adults). The sign +or – indicates if it was higher or lower than the baseline. P Values Children Adolescents Late adolescents and Young Adults Low-Delta1 <0.001, +0.003, +0.045, + Low-Delta2 0.013, +0.819, +0.720, - Delta2 0.287, +<0.001, +0.090, + Delta- Theta1 <0.001, +<0.001, +<0.001, + Theta2 0.010, +0.418, +0.005, - Alpha1 <0.001, - <0.001, - <0.001, - Alpha2 <0.001, +0.819, - 0.005, - Beta1 <0.001, +0.003, +0.946, - Beta2 O2 0.912, - 0.819, +<0.001, + Beta2 Fz 0.593, - 0.551, +<0.001, + Table 3 One-way ANOVA of the ERSP in the ten selected time–frequency windows, with the age group as the between-subjects factor (df =236), and post-hoc analysis FDR corrected. Ch: Children; Ad: Adolescents; LA-YA: Late adolescents and young adults. One-way ANOVA Post-hoc Children/Adolescent Children/LA-YA Adolescent/ LA-YA F p < or > p < or > p < or > p Low-Delta1 8.992 <0.001 Ch > Ad 0.002 Ch > LA-YA <0.001 Ad > LA-YA <0.001 Low-Delta2 5.100 0.011 Ch > Ad 0.029 Ch > LA-YA <0.001 Ad > LA-YA <0.001 Delta2 1.978 0.141 Ch < Ad <0.001 Ch > LA-YA <0.001 Ad > LA-YA 0.019 Delta-Theta1 42.461 <0.001 Ch > Ad <0.001 Ch > LA-YA <0.001 Ad > LA-YA <0.001 Theta2 9.995 <0.001 Ch > Ad 0.029 Ch > LA-YA 0.007 Ad > LA-YA 0.948 Alpha1 2.195 0.126 Ch < Ad 0.035 Ch < LA-YA 0.122 Ad > LA-YA 0.421 Alpha2 31.046 <0.001 Ch > Ad <0.001 Ch > LA-YA <0.001 Ad > LA-YA 0.980 Beta1 13.937 <0.001 Ch > Ad 0.070 Ch > LA-YA 0.184 Ad > LA-YA 0.929 Beta2 O2 2.499 0.105 Ch < Ad 0.597 Ch < LA-YA 0.002 Ad < LA-YA 0.018 Beta2 Fz 2.550 0.105 Ch < Ad 0.029 Ch < LA-YA <0.001 Ad < LA-YA 0.088 C.M. G´ omez et al. Brain and Cognition 167 (2023) 105969 7 channel were discarded. As indicated previously, only the 239 subjects who had a minimum of 10 correct trials were analyzed. 3. Data analysis 3.1. Event related Spectral Perturbation (ERSP) Time-frequency analysis allows observation of dynamic changes in spectral power, measured as ERSP relative to baseline, and phase. In the present report, we will focus on ERSP after subtracting the ERSP value from the baseline power. Note that although other studies have used different types of relative measures (Cohen, 2014), in the present report we preferred to use absolute values to consider the possible high dependence of ERSP with age. To study the time–frequency behavior of the electrical activity elicited by the stimuli presented in the trial, single trials (from 1 s before S1 to 7 s after S1) were convolved with a complex Morlet wavelet, using a 7 cycles wavelet, through the ft_freqanalysis function of FieldTrip (Oostenveld et al., 2011). ERSP values were maintained as absolute values, as the main interest was to observe changes in ERSP across ages. The baseline was applied in the period −800 ms to −400 ms before S1, to minimize the overlap of slow frequencies between the baseline and the perturbations induced by the stimulus. The frequency range analyzed was 1–25 Hz. High-frequency analysis (>25 Hz) was avoided due to the described contamination of these frequency ranges by muscle artifacts (Goncharova et al., 2003). ERSP values were averaged across trials. To observe which frequency oscillations are the most important for each of the WM subprocesses, the average of all the subjects was computed. This procedure allows to eliminate any bias due to age in the selection of frequencies and time windows for further analysis. 3.2. Behavioral statistical analysis The developmental trajectories of Reaction Times (RTs) and the percentage of incorrect responses, anticipations, and omissions were obtained. Non-responses to S2 were considered omissions; responses on the side of the novel stimulus were considered incorrect responses, and the responses to S1 during the maintenance period or responses faster than 200 ms after S2 were considered anticipations. These behavioral parameters were linearly regressed vs age expressed in days. The inverse model, as an alternative model, was computed by regressing the behavioral parameters vs. the inverse of age. To correlate the behavioral Fig. 6. ERSP amplitude relationship with age during the encoding phase. 6A. Spearman correlation (Rho) of ERSP vs. age panels for selected electrodes and collapsing all electrodes (below). The arrows in the color bar indicate the limits for Rho significance after applying the False Discovery Rate 6B. Topographies of the ERSP vs. age Spearman correlation averaging the time–frequency windows indicated. The arrows from 6A to 6B indicate the corresponding region in the time- –frequency plots. 6C. Linear and inverse regressions of the ERSP with age, collapsing the correlation values in the electrodes and time–frequency windows indicated in Fig. 6B. Only the correlation with the better fit is displayed. L-D1: Low-Delta1, D-T1: Delta-Theta1, A1: Apha1, B1: Beta1, LA-YA: Late adolescents and young adults. C.M. G´ omez et al. Brain and Cognition 167 (2023) 105969 8 parameters with the ERSP, and to reduce dimensionality, the percentage of correct responses was also computed by excluding all error types. The effect of gender on the behavioral parameters was analyzed by t-tests comparing both groups. No significant differences were found in any of the gender comparisons; therefore, gender was excluded from further behavioral analyses. 3.3. ERSP analysis ERSP were analyzed in absolute values in two different analysis periods: baseline and post-S1 (encoding and maintenance periods). ERSP during baseline (-.8s to -.4s) were used to obtain the EEG spectral composition during baseline (time–frequency panels). These values were regressed against the inverse of age. Additionally, mean ERSP values were collapsed across frequencies and electrodes during the Fig. 7. ERSP amplitude relationship with age during the maintenance phase. 7A, B, and C as in Fig. 6. L-D2: Low-Delta2, D2: Delta2, T2: Theta2, A2: Alpha2, B2: Beta2, LA-YA: Late adolescents and young adults. C.M. G´ omez et al. Brain and Cognition 167 (2023) 105969 9 baseline period and correlated with the inverse of age. The latter approach allowed us to observe the age dependence of brain rhythms not yet engaged in the task. The use of time–frequency windows, in a non-biased manner, rather than continuous statistical testing such as the cluster-mass approach (Maris & Oostenveld, 2007; Oostenveld et al., 2011) across the whole time–frequency and electrode space, was preferred for ERSP analysis to reduce the dimensionality of the data, which would make the statistical power of the analysis difficult given the high data dimensionality in terms of frequencies, periods, and electrodes. Averaging across the 239 subjects of the time–frequency panels of the absolute ERSP values produced a landscape of the oscillatory response to S1 presentation across electrodes during the encoding and maintenance periods. Some time–frequency intervals showed increments or decrements of ERSP amplitude, with respect to baseline, after the S1 presentation, including the encoding and maintenance phases. By averaging all subjects across ages, it was possible to obtain ERSP time–frequency panels that were not biased by age. To capture the electrodes where these dynamic responses occurred, the ERSP in the selected time–frequency windows were plotted topographically using the EEGLAB topoplot function. These latter procedures allowed the extraction of the electrodes, time windows, and frequencies in which the responses to S1 were more evident during the encoding and maintenance periods. The following criteria were used to select time–frequency windows and electrodes for further analysis: (i) the p-value should be significant compared to the baseline in at least one of the age groups, (ii) the time window should be centered around the peak amplitude of the selected frequency, (iii) the selected electrodes should present the maximum amplitude when topographically represented. Note also that the fact of selecting the time–frequency windows by the around peak activity criterion is completely unbiased with respect to the age group comparison since these features are selected in the collapsed across ages. The selected time–frequency windows were named as LowDelta1 (electrodes: FC1, FC2, Cz; time window: 0–0.55 s, frequency window: 1–1.5 Hz), LowDelta2 (Pz; 1–2.3 s; 1–1.5 Hz), Delta2 (FC1, FC2, Cz; 0.95–2.3 s; 3–4.5 Hz), Delta-Theta1 (O1, O2,Oz; 0.15–0.35 s; 3.5–6 Hz), Theta2 (Cz, Cp1, Cp2; 0.85–2.4 s; 7–8 Hz), Alpha1 (O1, O2, Oz; 0.35–0.80 s; 9–11.5 Hz), Alpha2 (POz; 2.15–2.4 s; 8.5–10 Hz), Beta1 (Oz, 0.1–0.15 s; 15–17.5 Hz); and Beta2, extracted independently in two different electrodes (O2 and Fz; 1.65–1.85 s; 17.5–19 Hz). Note that the strategy used to extract the time and frequency windows defined for further analysis is not biased because it is based on the average of all subjects. Similarly, the selection of electrodes for statistical analysis was based on the topography of the whole sample. Therefore, any agerelated results cannot be attributed to a posteriori selection of electrodes or time and frequency windows. The effect of gender on the ten ERSP time and frequency windows described in the preceding paragraphs was analyzed by t-tests comparing males and females. No significant differences were found in any of the gender comparisons; therefore, gender was excluded from further behavioral analyses. To test if the visually selected time and frequency windows were significantly different from the baseline, a ttest was applied to the mean ERSP values across the selected electrodes of each selected time–frequency window. Baseline comparisons were also computed for each of the selected time–frequency windows for the three age groups (children, adolescents, and LA-YA). A one-way ANOVA with age as the between-group factor was computed for the ERSP in each time–frequency period. To have a more continuous definition of the relationship between ERSP and age, a regression approach was used to characterize the developmental trajectories, instead of the more classical post-hoc comparisons. Mean values of each of these time–frequency periods across the selected electrodes were obtained and regressed linearly and inversely to age. The regression was complementary to ANOVAs to assess the effect of age on ERSP and was used in addition to Table 4 P-values of the FDR corrected Spearman correlations between ERSP amplitudes with Rts and the percentage of correct responses. ERSP Rts %Correct Responses Rho p Rho p Low-Delta1 0.186 0.010 -0.164 0.020 Low-Delta2 0.078 0.235 -0.170 0.019 Delta2 0.202 0.005 -0.125 0.072 Delta-Theta1 0.478 <0.001 -0.330 <0.001 Theta2 0.338 <0.001 -0.163 0.020 Alpha1 -0.111 0.103 0.136 0.051 Alpha2 0.460 <0.001 -0.307 <0.001 Beta1 0.273 <0.001 -0.156 0.025 Beta2 O2 -0.105 0.118 0.208 0.004 Beta2 Fz -0.118 0.086 0.164 0.020 Table 5 Mediational analysis. Standardized direct and indirect effects of age through ERSP amplitudes on behavioral measures. Outcome ERSP Mediator a b a ×b 95% CI c’ c Response Time -0.641** Low-Delta1 -0.269** 0.023 -0.006 [-0.034, 0.022] -0.635** Low-Delta2 -0.171* -0.044 0.007 [-0.011, 0.026] -0.649** Delta2 -0.074 -0.010 0.001 [-0.007, 0.008] -0.642** Delta-Theta1 -0.499** 0.162* -0.081* [-0.140, -0.022] -0.561** Theta2 -0.337** 0.035 -0.012 [-0.048, 0.024] -0.629** Alpha1 0.125 0.058 0.007 [-0.007, 0.022] -0.649** Alpha2 -0.382** 0.177* -0.067* [-0.113, -0.022] -0.574** Beta1 -0.282** 0.058 -0.016 [-0.046, 0.014] -0.625** Beta2 O2 0.135 0.021 0.003 [-0.011, 0.017] -0.644** Beta2 Fz 0.135 -0.064 -0.009 [-0.024, 0.007] -0.633** Correct Responses 0.379** Low-Delta1 -0.269** -0.033 0.009 [-0.025, 0.042] 0.370** Low-Delta2 -0.171* -0.046 0.008 [-0.014, 0.029] 0.371** Delta2 -0.074 -0.111 0.008 [-0.009, 0.025] 0.371** Delta-Theta1 -0.499** -0.080 0.040 [-0.029, 0.109] 0.339** Theta2 -0.337** -0.060 0.020 [-0.023, 0.064] 0.359** Alpha1 0.125 0.075 0.009 [-0.008, 0.027] 0.370** Alpha2 -0.382** -0.376** 0.144** [0.080, 0.208] 0.235** Beta1 -0.282** -0.010 0.003 [-0.032, 0.038] 0.376** Beta2 O2 0.135 0.032 0.004 [-0.012, 0.021] 0.375** Beta2 Fz 0.135 0.115 0.016 [-0.006, 0.038] 0.363** Note: ** p <.001, * p <.05; p-values were adjusted using FDR; a: coefficients for the path from age to ERSP measures; b: coefficients for the path from ERSP amplitudes to behavioral measures (Rts and correct responses); a ×b: indirect effects of age on behavioral measures through ERSP amplitudes; c’: direct effects of age on behavioral measures; c: total effects of age on behavioral measures; CI: confidence interval for the coefficients a ×b. C.M. G´ omez et al.