Human Psychobiology Lab, Experimental Psychology Deparment, University of Sevilla, Seville, Spain

Behavioral Methodology Lab, Experimental Psychology Deparment, University of Sevilla, Seville, Spain

Abstract

Background

The peri-adolescent period is a crucial developmental moment of transition from childhood to emergent adulthood. The present report analyses the differences in Power Spectrum (PS) of the Electroencephalogram (EEG) between late childhood (24 children between 8 and 13 years old) and young adulthood (24 young adults between 18 and 23 years old).

Results

The narrow band analysis of the Electroencephalogram was computed in the frequency range of 0–20 Hz. The analysis of mean and variance suggested that six frequency ranges presented a different rate of maturation at these ages, namely: low delta, delta-theta, low alpha, high alpha, low beta and high beta. For most of these bands the maturation seems to occur later in anterior sites than posterior sites. Correlational analysis showed a lower pattern of correlation between different frequencies in children than in young adults, suggesting a certain asynchrony in the maturation of different rhythms. The topographical analysis revealed similar topographies of the different rhythms in children and young adults. Principal Component Analysis (PCA) demonstrated the same internal structure for the Electroencephalogram of both age groups. Principal Component Analysis allowed to separate four subcomponents in the alpha range. All these subcomponents peaked at a lower frequency in children than in young adults.

Conclusions

The present approaches complement and solve some of the incertitudes when the classical brain broad rhythm analysis is applied. Children have a higher absolute power than young adults for frequency ranges between 0-20 Hz, the correlation of Power Spectrum (PS) with age and the variance age comparison showed that there are six ranges of frequencies that can distinguish the level of EEG maturation in children and adults. The establishment of maturational order of different frequencies and its possible maturational interdependence would require a complete series including all the different ages.

Background

The adolescent period is a crucial developmental moment of transition from childhood to emergent adulthood. It is also a period in which many different types of mental and behavioral problems can arise

The EEG is not stable over development, it is changing until arriving to the typical adult pattern. Several topics have been in the focus of the research investigating EEG development in control children: EEG power developmental trajectories

EEG power developmental trajectories, the establishment of developmental equations

The delta rhythm is the main activity in the first two years of life. In contrast, delta waves are not observed in normal adult EEG, in awake and relaxed states. However, the delta waves are characteristic of NREM sleeps stages III and IV also called Slow Wave Sleep in both adult and children

The study of EEG power maturation during development is particularly important because EEG rhythms can be affected or modulated by developmental disorders

With increasing age, lower frequencies decrease and higher frequencies increase

The continuous or discontinuous nature of the developmental trajectory of EEG power is under a certain controversy, from a continuous linear change, as shown by the developmental equations

Changes in frequency

Other important parameter changing during childhood is the frequency of brain rhythms. The frequency of the alpha rhythm is increasing during development. At the age of primary school, the alpha rhythm presents an average frequency of 10 Hz. This corresponds to the average frequency of the adult’s EEG. This value is achieved at the age of 10

Changes in topographies

The presence of theta rhythm in posterior regions is common in children between 7 and 10 years old

Shaw et al.

Relation to cognitive maturation

Some studies have shown that brain and cognitive maturation are intimately associated. For example, Hudspeth and Pribram

Two of the basic developmental phenomena are related to myelination of axons and to synaptic pruning

Furthermore, in studies of sleep EEG, Feinberg and Campbell

Principal component analysis of the EEG

Principal Component Analysis (PCA) allows the multivariate EEG data be explained by a small number of latent variables (principal components). In previous reports, using the principal component analysis

The main objective of this work is to study the EEG power pattern of maturation in human beings in the period from late childhood to peri-adolescent period, which are the periods surrounding the very important developmental period of adolescence.

Although adolescence is considered to be a period of major evolutionary changes at psychological and physiological levels, the EEG does not change significantly. At the age of 13, the teen shows an EEG similar to mature pattern

Methods

Experimental procedure

Subjects

The study included a sample comprising 48 subjects (27 women and 21 men), aged between 8 and 23. Of the total, 45 were right-handed and 3 left-handed. The total sample was divided into two subgroups: a group of children and a group of young adults.

The group of children consisted of 24 subjects aged between 8 and 13 (mean age ± SD age: 10.1 years ± 1.41). Of these, 13 were females and 11 males (22 right-handed and 2 left-handed). The young adult group consisted of 24 subjects aged between 18 and 23 (mean ± SD age: 20.5 years ± 1.3). Of these, 14 were females and 10 males (23 right-handed and 1 left-handed).

Children and young adults did not report any neurological or psychological disease or impairment. Both groups were extracted from middle class socioeconomic background. Children were normal in academic records, and young adults were college students. Experiments were conducted with the informed and written consent of each participant (parents/tutors in the case of the children) following the Helsinki protocol.

Electrophysiological recording

The EEGs were recorded during three minutes of spontaneous activity (which does not involve any explicit cognitive task) keeping the eyes open. The subjects were recorded at different times of the day, between 11 AM and 8 PM. No information about previous sleep was required. They were obtained from an average reference of 20 scalp sites of the International 10–20 system (Fp1, Fp2, F3, F4, F7, F8, Fz, FCz, T7, T8, C3, C4, Cz, P7, P8, P3, P4, Pz, O1, O2), using tin electrodes mounted in an electrode cap (EASYCAP, Herrsching-Breitbrunn, Germany) with two additional electrodes (M1, M2). Ocular movements (EOG) were recorded from two electrodes at the outer canthus of each eye for horizontal movements and one electrode under the left eye for vertical movements that were referred to FP1. All the scalp electrodes were re-referenced off line to the mastoid average (M1 + M2/2). Impedance was maintained below 10 Kilo-Ohms (KΩ). Data were recorded in Direct Current (DC) mode at 512 Hz, with a 20,000 amplification gain using a commercial Analog Digital (AD) acquisition and analysis board (ANT). Data were not filtered during registration. We asked the subjects to stay calm and looking at the screen for three minutes. Following to the three minutes of spontaneous EEG, the subjects were recorded in an ERP experiment which lasted for 20 minutes. The Event Related Potential (ERPs) results have already been reported

Data analysis

A 0.1 Hz high-pass filter and a 20 Hz low-pass filter (zero-phase, low cutoff of 6 db/octave, high cutoff of 48 db/octave, BESA software) were applied to the data. The artefacts in the resulting EEG recordings were corrected by an artefact correction protocol. The algorithm used for the artefact correction was based on PCA (BESA software). This method splits the EEG components associated with cerebral activity from artefacts (ocular movements, muscular or cardiac activity), on the basis of spatial distribution, after which the EEG can be reconstructed free of artefacts

After the correction of artefacts, an artefact rejection protocol was applied. All the epochs for which the EEG exceeded ±100 Microvolts (μV) in any channel were automatically discarded. The resulting records were reviewed manually and rejected those segments that appeared to be outside the parameters of brain activity.

The next step was to compute the Power Spectrum (PS) of the epochs by means of the Fast Fourier Transform (FFT). This tool consists of a mathematical function that transforms data that belongs to time domain into frequency domain

The analyzed time in the power spectral analysis were less than those initially recorded (90 epoch of 2 seconds) due to the elimination of epochs containing artifacts. The average time analyzed in the records of the group of children was 2'28'' (mean ± standard deviation: 2'28'' ± 0.41 minimum time 1'32'') while the average time recorded in the records of the group of young adults was 2'40'' (mean ± standard deviation: 2'40'' ± 0.45; minimum time: 2'20''). The frequency resolution was 0.51 Hz. Therefore, 39 frequencies (from the range 0–0.51 Hz to 19.38-19.89 Hz) were obtained, although the graphics have been rounded (20 Hz like extreme value). The EEG frequencies over 20 Hz were excluded from the analysis in order to reduce the impact of possible electromyography contamination of EEG recordings. Therefore, the original data matrix consisted of 960 rows (48 subjects x 20 electrodes) and 39 empirical variables in the matrix columns (39 frequencies).

For certain applications, the PS data were exported in selected pre-defined frequency bands for each subject: Delta (1 – 4 Hz), Theta (5 – 8 Hz), Alpha (9 – 12 Hz) and Beta (14 – 19 Hz) or selected based on the results. For certain applications, the PS averaged values for each band were collapsed by regions (anterior, central and posterior). For these particular exports, the electrodes that composed the anterior area were Fp1, Fp2, F7, F3, Fz, F4, F8; the central area was composed by FCz, T7, C3, Cz, C4, T8 and the posterior area by P7, P3, Pz, P4, P8, O1, O2. In these cases, the matrix for data analysis comprised 12 columns (4 bands x 3 regions) and 48 rows (subjects). For some applications, the same electrodes were collapsed but for different frequency ranges, which in the course of the analysis appeared as more suited for understanding the developmental issue.

Statistical analysis

Using Statistical Package for the Social Sciences (SPSS) 14.0, a mixed-model **AN**alysis **O**f **VA**riance (ANOVA) was applied to the logarithm of the PS of absolute power to compare the children and young adults groups. The inter-subject factor was the variable age group with two levels: children and young adults. The within-subjects factors were the bands (four levels: delta, theta, alpha, beta) and the regions (3 levels: anterior, central, posterior). P values were computed using the Greenhouse-Geisser correction. The same type of ANOVA was performed to the six frequency ranges that in the course of the analysis appeared as more suited to the developmental issue (low delta, delta-theta, low alpha, high alpha, low beta and high beta frequencies).

In addition to the general ANOVA, the ratio of mean PS of adults with the mean PS of children was computed. The ratio of variances was also computed. This ratio was computed independently for each electrode and frequency and provides a landscape of the different frequency ranges in which different maturational trends can be obtained. Afterwards, and complementing the broad band ANOVA analysis, a narrow band mean comparison in terms of frequencies and spatial resolution was computed for the differences between children and young adults. T-tests were computed between children and young adults in each single electrode and frequency. For the mean comparisons of absolute power, a more detailed graphic with the different levels of statistical significance was also computed. No multiple corrections were applied to the t-tests because of the high statistical interdependence between the data, as proved by (i) the landscape of t-test across electrodes and frequencies, (ii) the high correlation between the PS of different frequencies and the small number of components explaining the variance of the whole population of data. The high internal dependence between the different empirical variables would potentially produce a huge number of false negative results if a conservative correction as Bonferroni was used. Anyway p-values in the t-test were very small, indicating big differences in the mean PS between children and young adults.

In addition to the narrow band between-subjects means comparisons, the between-subjects variance equality and correlation with age has been computed using F-Levene for variance comparisons and Spearman correlation coefficient respectively. These computations were obtained by each frequency and electrode. In addition to the statistical comparisons, we were interested in testing if there were different levels of maturation in different frequency ranges. For this purpose, the correlation with the EEG power in all electrodes and frequencies considered was computed. The topographical representation of this correlation was obtained in order to establish the level of maturation of different frequencies and scalp locations.

The mean and variance comparisons and Spearman correlations were computed in absolute and relative power. The relative power was obtained using Matlab 7.0 and using the following formula (eq. 1)

PRfi: Relative Power for frequency i

PAfi: Absolute Power for frequency i

With all the previous analysis we expected to obtain an idea of the different maturational trends that would appear in different frequencies and electrodes.

Cross-frequencies correlations

All frequency variables (39 frequencies) were correlated against the other in order to observe if there are patterns of covariation between different frequencies. This was done independently for the sample of children, for young adults and for the total sample. The two-tailed statistical significance of the correlations between different frequencies was estimated taking into account the number (N) of subjects (N = 24 for the children and young adults group, and N = 48 for the total sample). The correlation matrix was expressed in a color code to better appreciate the different cross-frequencies correlation patterns, and if they were different in children and in young adults.

Principal component analysis

With this method, it is possible to identify the latent components that explain the variance of the experimental data

In order to establish the physiological meaning of each component, the loading factors of each empirical variable (the 39 frequencies considered) for the extracted components were represented in a color coded display. The loading factors are the correlation coefficient between a given empirical variable (PS for a given narrow band frequency) and the component scores of a given component. The two-tailed statistical significance of the loading factors was estimated taken into account the number of subjects (N = 24 for the children and young adults group, and N = 48 for the total sample). Component scores provide the value that a certain case (subject in a given electrode) has for a certain latent variable (the component). The correlation of the component scores with the subject age would give an indication on how a given component captures a maturational trend. The total number of component scores for each extracted component was 960 (48 subjects x 20 electrodes).

An important point was to find homologous components in the children, adults and total sample. To that end, two different methods were applied: the simultaneous representation of loading factors of the components versus different frequencies for both age groups, and the topographical representation of the component scores of the different components.

For the loading factors versus frequencies representations, these components which presented a similar profile were considered homologous across groups. In order to better observe if there was a similarity between components, the correlation between the loading factors of candidates to be homologous components was obtained. For the topographical maps of the component scores, those obtained by individual subjects were averaged independently for the children, young adults and total group, and represented by means of the EEGLAB topoplot function

Specific questions that would be addressed with the PCA method and which are relevant to the understanding of the maturation of spontaneous brain rhythms are:

1) The comparison of the profile of the loading factors and the topographies of the different variables in children and young adults will assess whether the structure of the EEG is similar in both age groups or different.

2) If component scores of a certain component are inversely correlated with different frequencies, it would suggest an opposite maturational trend of certain frequencies during development.

3) By representing the loading factors vs. frequency of the homologous components of children and young adults, it would be possible to observe if the frequencies that are related to a certain component are shifted in frequency in late childhood as compared with young adulthood.

Results

Power spectrum

Figure

Power spectrum (PS) (Figure 1A) and the corresponding standard error (Figure 1B) of children and young adults in electrodes Fz, FCz, Cz, Pz, O1 and O2

**Power spectrum (PS) (Figure ****A) and the corresponding standard error (Figure ****B) of children and young adults in electrodes Fz, FCz, Cz, Pz, O1 and O2.** Notice the higher PS in children than in adults. The insets in electrodes O1 and O2 show that the alpha peak is reached at lower frequencies in children than in young adults.

**
ELECTRODES
**

**O1**

**O2**

**Mean**

**Standard deviation**

**Mean**

**Standard deviation**

9.223

.926

9.350

.898

10.094

1.345

10.200

1.258

In Figure

Frequency histograms of the averaged PS values of frontal, central and posterior electrodes in the delta, theta, alpha and beta bands

**Frequency histograms of the averaged PS values of frontal, central and posterior electrodes in the delta, theta, alpha and beta bands.** In all bands the children presented a higher mean than the young adults. Each histogram is created with 24 children or adults. The ANOVA statistics are referred in the results section. Averaging was obtained across subjects and electrodes as described in the methods section.

However, children should not have exactly the same rate of brain maturation in all the considered frequencies. For this reason, we computed for each electrode and frequency the ratio between the mean PS of young adults and children (Figure

Ratio of PS mean (3A) and variance (3B) of young adults and children

**Ratio of PS mean (3A) and variance (3B) of young adults and children.** Figure **A**. The figure shows the ratio of mean PS in young adults with PS in children (above). Figure **B** The ratios of variance are shown. In the case of variance, the color code has been saturated to 1.3. Notice the presence of six bands in which these parameters are relatively steady.

The latter results suggest reorganizing the collapsing of frequencies for these new bands that seemed to reflect better the EEG bands maturational trends. Figure

Frequency histograms of the averaged PS values of frontal, central and posterior electrodes in the low-delta, delta-theta, low alpha, high alpha, low beta and beta bands, as suggested by Figure

**Frequency histograms of the averaged PS values of frontal, central and posterior electrodes in the low-delta, delta-theta, low alpha, high alpha, low beta and beta bands, as suggested by Figure ****.** In all bands the children presented a higher mean than the young adults. Each histogram is created with 24 children or adults. The ANOVA statistics are referred in the results section. Averaging was obtained across subjects and electrodes as described in the methods section.

However, the collapsing of electrodes in a few scalp regions and broad frequency bands could have missed certain patterns of maturational trends across frequencies and scalp locations. Therefore, we decided to make mean comparisons by means of t-tests (Figures

**T-test mean comparisons of PS of young adults and children.** Each pixel in the image represents the p-values for the

Mean (Figure 6A), Variance Comparisons (Figure 6B) and Spearman correlations (Figure 6C) for absolute and relative power

**Mean (Figure ****A), Variance Comparisons (Figure ****B) and Spearman correlations (Figure ****C) for absolute and relative power.** The p-value of the **A**), the p-value of the F-Levene testing homogeneity of variance of children and young adults (6**B**), and the Spearman correlation of PS with age (6**C**) are displayed. Notice the differential pattern of absolute (left) and relative (right) power. Rho: Spearman correlation coefficient.

Spearman correlation of PS of young adults and children with age

**Spearman correlation of PS of young adults and children with age.** Each pixel in the image represents the correlation coefficient in a given electrode (rows) and frequency (columns). The highest correlation coefficients correspond to the delta-theta range. The topographical maps of the mean of correlation coefficient for the frequency ranges indicated by lines are also displayed. Notice that correlation coefficient is more negative (better correlation with age) in anterior and central electrodes than in posterior electrodes for most frequencies. Rho: Spearman correlation coefficient.

The level of signification for mean and variance comparisons and the correlation coefficient of PS vs. age would be indicators of differential trends of maturation of the different frequencies in different regions of the scalp. The signification of the t-test maps (Figure

Figure

In Figure

An important point we would like to highlight in the analysis of PS in young adults and children is the comparison of results when absolute and relative power are compared, and when fractions of low frequency/high frequency PS values are considered.

With regard to the comparisons low/high frequency, the comparisons between the low delta (Figure

Fractions of low frequencies/high frequencies in three different scalp locations

**Fractions of low frequencies/high frequencies in three different scalp locations.** The fractions of Low delta (8**A**) and Delta-Theta (**B**) Spectral Power are represented. Notice that the fractions of low delta did not have a difference between age groups, while those of delta-theta present a higher value in children than in adults indicating the dramatic decrease of delta-theta power in the adolescent period.

Correlational and principal component analysis

The Spearman correlation matrix of 39 empirical variables (from 0 to 20 Hz) was computed. The correlation analysis performed with all frequencies tries to understand the internal structure of covariation of spontaneous EEG (Figure

A-C. Cross-frequency PS Spearman correlation (Rho) (left), loading factors of PCA analysis in the non-rotated PCA (middle) and rotated PCA (right)

**A-C. Cross-frequency PS Spearman correlation (Rho) (left), loading factors of PCA analysis in the non-rotated PCA (middle) and rotated PCA (right).** In the left side of Figures **A**-**C**, it appears the cross-frequency correlation matrix for the children, young adults and the total sample. The Loading factors obtained from the non-rotated (middle) and varimax rotated (right) PCA are shown in the middle and the right of Figure **9D.** The Spearman correlation with age of the component scores was also computed (Figure **D**) for both non-rotated and rotated PCA.

Non-rotated and Varimax rotated PCA were computed in the matrix for 39 empirical variables (from 0 to 20 Hz). Figure

Scree plot

**Scree plot.** The image shows the explained variance by each of the components extracted in the Varimax-rotated PCA of PS data. The Scree plot is displayed for the children (10**A**), adults (10**B**) and total sample (10**C**).

Loading factors of the components for non-rotated and rotated PCA are shown in the middle and right side of Figure

Loading factors of the first component of the non-rotated PCA incorporate the individual variability, as it is shown by the high correlation with all the frequencies, indicating an individual variability meaning for this factor. Interestingly, the second component shows a reverse pattern between delta and alpha frequencies, but only for the total sample, indicating that during development, a latent factor should be acting with opposite effects in these two frequency ranges. For Varimax rotated PCA, the most interesting aspect is the excellent segregation of different frequency bands (Figure

Correlation with age of component scores was also computed (Figure

**Levene test:**

**Mean comparison:**

The F values of Levene test (homogeneity of variances) and One-way ANOVA (two between-subjects levels: children and young adults) are displayed. The rhythms associated to each component are indicated from component 1 (C1) to component 9 (C9) . In the case of C1 the term Beta-EMG refers to the possible contamination of the Beta rhythm by electromyographic signals.

C1 (Beta - EMG)

(F[1, 958] = 28.936), p < .001

(F[1, 705.346] = 6.488), p < .001

C2 (Delta - Theta)

(F[1, 958] = 361.333), p < .001

(F[1, 561.066] = 32.359), p < .001

C3 (Alpha)

(F[1, 958] = 48.738), p < .001

(F[1, 695.931] = 2.455), p < .014

C4 (Mu)

(F[1, 958] = 6.373), p < .012

(F[1, 691.765] = −1.610), p < .108

C5 (Low Alpha)

(F[1, 958] = 117.238), p < .001

(F[1, 522.495] = 2.954), p < .003

C6 (High Alpha)

(F[1, 958] = 72.075), p < .001

(F[1, 741.052] = .975), p < .330

C8 (Low Delta)

(F[1, 958] = 76.309), p < .001

(F[1, 728.777] = 4.176), p < .001

C9 (Mu)

(F[1, 958] = 15.962), p < .001

(F[1, 917.456] = 1.842), p < .066

In order to identify the physiological meaning and the homology of Varimax components in the two age groups, two different strategies were followed: the representation of loading factor vs. frequency, and the topographical maps of component scores. Two components would be considered homologous if they share similar patterns in both. In the case that two components are homologous, it would also indicate that the general structure, in terms of topography and frequency of a given rhythm, is already mature in late childhood.

Varimax rotated loading factors vs. frequency appear in Figure

Loading factors versus frequency

**Loading factors versus frequency.** 11**A**. The values of the loading factors vs. EEG frequency of the components in which similar patterns in children and young adults were obtained are represented. The Pearson correlation coefficient (r) computed comparing the loading factors of children and young adults are also represented. Notice that in most cases the r value is relatively high. For the C2 of children, two components of young adults were considered homologous due to the similar pattern of the loading factors, but also because similarities in the topographies of the component scores (see Figure **B**. The values of the loading factors versus EEG frequency of the components obtained from the total sample. Components 4 and 9 seem to explain the same range of frequencies and present similar topographies (see Figure

In Figure

Topographical representation of PS and component scores of topographies of each component

**Topographical representation of PS and component scores of topographies of each component.** Topographies of component scores of children, young adults and the total sample are displayed. For the PS map, the frequency in which the loading factor of the component peaked (see Figure

Table

One last interesting suggestion from the analysis of loading factors (Figure

Discussion

Present results showed that the well described pattern of decrease in EEG power with age would be organized in six different frequency ranges. The correlation pattern of different frequencies suggests a certain asynchrony in the maturation, which would be a consequence of the six frequency ranges with different EEG power ratios between children and adults in the peri-adolescent period. The narrow-band frequency correlation with age analysis corroborated the previously described pattern of an earlier maturation of posterior regions with respect to anterior regions. But low frequencies present a lower rate of maturation in the peri-adolescent period than alpha and beta rhythms when criteria based in mean comparisons and correlations comparisons are used. However, the delta-theta/high frequencies ratios were decreasing with age. The principal component analysis allowed extracting the following basic brain rhythms in children and adults with very similar topographies: beta, theta (and anterior delta), high alpha (occipito-temporal), occipital alpha, low alpha (parieto-occipital), low delta and the mu rhythm. The scalp alpha rhythm and all the extracted alpha sub-components and the extracted mu rhythm peaked at a higher frequency in young adults with respect to children. Given the different pattern of age differences when absolute and relative power are compared, some caution must be taken when extrapolating conclusions from one to the other type of analysis.

EEG power differences around the adolescence period

Children showed greater absolute spectral power than young adults in the four standard frequency bands. Spectral power decrease during maturation is a general finding

The obtained differences in spectral power in present report were higher in the delta and theta rhythms, as it has been previously described

However, in the case that EEG power maturation is already almost completed in our 18–23 years old sample, the following order of maturation for EEG absolute power frequencies can be proposed: high alpha, followed by low alpha and high beta, the low beta and low delta and finally the delta-theta range. This possible order of maturation is different to the general view of low frequencies maturing earlier than high frequencies

Interestingly, when the cross-frequencies EEG power correlations are obtained, the young adults obtained a higher pattern of correlation than the children. The latter result and the previously described six frequency ranges differential maturation indicate that there is possibly a certain asynchrony in the maturation of the different rhythms. Somsen et al.

Mention apart requires the obtained result that the EEG power maturation in the range of very low delta (range 0 to 0.51 Hz) seems to be different than in delta-theta frequency range (.51-7.65 Hz). However, it must be kept in mind that this frequency is within the range of the high pass filter located at 0.1 Hz and 6db per octave frequency cutoff. Nevertheless, records in young adults and children received the same type of filtering, and therefore, it is difficult to assume that this difference is a consequence of signal processing. Tenke and Kayser

The topographical pattern of the correlation of PS with age shows that, in general, correlation is higher in anterior than in posterior sites for the frequencies considered in present report, suggesting that maturation is progressing from posterior to anterior sites. The correlation with age pattern is quite similar to the mean PS comparison between age groups. In this sense, the correlation with age pattern can be considered as a measure of the size effect of mean comparisons. The present results extend to a narrowband PS analysis, previous results on broad band showing that maturation in general progresses from posterior to anterior sites. Hudspeth and Pribram

In this section of differences in PS between children and young adults, the different pattern of maturation when absolute and normalized power is used must be remarked. The normalized PS measures have been used to minimize the impact of the strong individual component in PS amplitude values and to increase test-retest reliability of EEG measurements

The results obtained in present report show the characteristic higher normalized PS of low frequencies in children over young adults, and the higher PS in beta of young adults with respect to children. Correlation with age showed a similar pattern. This result has been classically obtained, showing the typical movement from slow to fast waves with age

Principal component analysis

The purpose of using a PCA analysis would be to find different sub-components in the brain rhythms which are not obvious in the scalp EEG, and demonstrate if they show some maturational trends. In this approach, we used Varimax rotated and non-rotated approaches. Both approaches are mathematically valid, the rotated Varimax approach tries to accommodate the data variance in a few components in order to simplify the physiological interpretation, and for this reason, we have chosen this approach for the interpretation of the results, except for the results obtained in the non-rotated PCA in the first component and in the third component.

The loading factors of the first component of non-rotated PCA were very high and the component scores were correlated with age. This first component of non-rotated PCA could be related to the "general alert" factor described by Lazarev

The pattern of loading factors in rotated components, in terms of loading factors vs. frequency and topographies of component scores, would allow to characterize the physiological meaning of different components and to prove if children and young adults present similar components. All components obtained in children presented an homologous component in young adults, although the order of components in children and young adults obtained by the amount of explained variance was different. These homologies between children and young adults, which are also present in similar PS topographies, implicate that the structure of the EEG is already present in children in the pre-adolescent period. Different components were extracted corresponding to beta-EMG, delta-theta, a temporo-occipital high alpha, occipital alpha, parieto-occipital low alpha, low delta and mu rhythm. Beta rhythm, given the extension to lateral sites would have an important contamination from Electromyography (EMG) in these lateral sites. These extracted rhythms broadly correspond to the frequency ranges which present a certain asynchrony in the maturational pattern.

It is remarkable that peaks of loading factors in children occur at lower frequencies than adults in most cases, and this is particularly clear in the three different sub-components of alpha and in the mu rhythm. The topographic representation of PS and component scores comparing homologous components between two age groups showed very similar topographies, despite the fact that the same frequencies were not represented for the PS of children and young adults.

When loading factors of the different alpha range sub-components of children and young adults are represented, a clear same displacement from lower to higher frequencies occurs from children to young adults. The scalp frequency shift to high frequencies during maturation has been proposed to be due to an increase in the contribution of high alpha to the alpha peak

A final point which deserves some comments is the previously obtained posterior extension of theta rhythm in children with respect to young adults. It has been proposed that theta band shows a maturational progression from posterior to anterior areas

The present results clearly show that the EEG structure is already present in late childhood, although with a higher PS and lower frequency than young adults. The maturation occurs in an asynchronic manner for different frequencies and locations.

Some important limitations of present report are (i) that present study is only able to capture differences between the different age groups and not inside a certain age group, in fact, a continuous recording for all possible age groups would provide a better description of EEG maturation, (ii) the cohort character of present study, which only measures differences between age groups, which is an indirect measure of maturation. However the normal population character of the subjects in this experiment suggests that the obtained results would have a strong relationship with normal EEG developmental maturation in the peri-adolescent period, and (iii), given the circadian and homeostatic pattern of changes in beta, alpha and theta rhythms

Conclusions

In this study we conclude the following:

→ Children have a higher absolute power than young adults for frequency ranges between 0–20 Hz. For relative PS children present a higher power than young adults in delta and theta range, while young adults have an increased spectral power in very low range (0 to 0.51 Hz) and beta frequencies.

→ The age comparisons of mean PS, the correlation of PS with age and the variance age comparison showed that there are six ranges of frequencies that can distinguish the level of EEG maturation in children and adults. Frequencies are: very low delta, delta- theta, low alpha, high alpha, low beta and high beta.

→ Both PS: absolute and relative showed good correlations with age, indicating that PS can be regarded as a maturation index of the human electroencephalogram. But they presented different topographies, indicating that some caution must be taken when comparing absolute and relative power.

→The age group mean comparisons of PS and the PS correlation with age suggest a postero-anterior maturation of the PS with the following order: High alpha, followed by low alpha and high beta, low beta and low delta and finally the delta-theta range.

→ Cross frequency correlation matrices suggest an asynchrony in the maturation of the different brain rhythms around peri-adolescent period.

→ The component 1 of non-rotated PCA would explain the individual variability.

→The representation of loading factors of different frequencies suggests that the increase in frequency with maturation occurs in most of the frequency ranges, and particularly in the alpha sub-components and mu rhythm.

→The component scores and PS topographies show very similar topographies in young adults and children, reinforcing the idea of a completed structure of the EEG rhythms in the pre-adolescent period. Interestingly, in the alpha range, a parietal low alpha, a posterior alpha, an occipito-temporal high alpha and a mu rhythm can be distinguished in the 7–13 Hz range.

Abbreviations

(PS): Power Spectrum; (EEG): Electroencephalogram; (PCA): Principal Component Analysis; (SD): Standard Deviation; (KÎÂ©): Kilo-Ohms; (DC): Direct Current; (AD): Analog Digital; (ERPs): Even Related Potential; (BESA): Brain Electrical Source Analysis; (μV): Microvolts; (FFT): Fast Fourier Transform; (Hz): Hertz; (SPSS): Statistical Package for the Social Sciences; (ANOVA): **AN**alysis **O**f **VA**riance; (PRfi): Relative Power for frequency i; (PAfi): Absolute Power for frequency i; (N): Number; (EMG): Electromyography.

Competing interests

'The authors declare that they have no competing interests'.

Authors’ contributions

EI: Data Analysis and rational of the study. CI: Data Analysis and rational of the study. MI: Data Analysis and rational of the study. C: Data Analysis and rational of the study. AM: Data Analysis. CM: Data recording, Analysis, and rational of the study. All authors read and approved the final manuscript.

Authors’ information

EI: Degree in Psychology. PhD student. CI: Degree in Psychology. PhD student. MI: Degree in Psychology. PhD student. C: Degree in Psychology. PhD student. AL: Professor of methodology of Behavior Sciences, Phd. CM: Professor of Psychobiology Phd.

Acknowledgements

We want to thanks to Carmen Gómez-Sos for her English language editorial assistance.