Development of the spontaneous activity transients and ongoing cortical activity in human preterm babies.

Abstract

Recent experimental studies have shown that developing cortex in several animals species, including humans, exhibits spontaneous intermittent activity that is believed to be crucial for the proper wiring of early brain networks. The present study examined the developmental changes in these spontaneous activity transients (SAT) and in other ongoing cortical activities in human preterm babies. Full-band electroencephalography (FbEEG) recordings were obtained from 16 babies at conceptional ages between 32.8 and 40 wk. We examined the SATs and the intervening ongoing cortical activities (inter-SAT; iSAT) with average waveforms, their variance and power, as well as with wavelet-based time-frequency analyses. Our results show, that the low frequency power and the variance of the average waveform of SAT decrease during development. There was a simultaneous increase in the activity at higher frequencies, with most pronounced increase at theta-alpha range (4-9 Hz). In addition to the overall increase, the activity at higher frequencies showed an increased grouping into bursts that are nested in the low frequency (0.5-1 Hz) waves. Analysis of the iSAT epochs showed a developmental increase in power at lower frequencies in quiet sleep. There was an increase in a wide range of higher frequencies (4-16 Hz), whereas the ratio of beta (16-30 Hz) and theta-alpha (4-9 Hz) range activity declined, indicating a preferential increase at theta-alpha range activity. Notably, SAT and iSAT activities remained distinct throughout the development in all measures used in our study. The present results are consistent with the idea that SAT and the other ongoing cortical activities are distinct functional entities. Recognition of these two basic mechanisms in the cortical activity in preterm human babies opens new rational approaches for an evaluation and monitoring of early human brain function.

Full text

Page 1

DEVELOPMENT OF THE SPONTANEOUS ACTIVITY TRANSIENTS
AND ONGOING CORTICAL ACTIVITY IN HUMAN PRETERM BABIES
M. TOLONEN,aJ. M. PALVA,bS. ANDERSSONa
AND S. VANHATALOc*
aDepartment of Pediatrics, Hospital for Children and Adolescents,
University Hospital of Helsinki, Helsinki, Finland
bNeuroscience Center, University of Helsinki, Helsinki, Finland
cDepartment of Clinical Neurophysiology, University Hospital of Hel-
sinki, PO Box 280, FIN-00029 HUS, Helsinki, Finland
Abstract—Recent experimental studies have shown that de-
veloping cortex in several animals species, including hu-
mans, exhibits spontaneous intermittent activity that is be-
lieved to be crucial for the proper wiring of early brain net-
works. The present study examined the developmental
changes in these spontaneous activity transients (SAT) and
in other ongoing cortical activities in human preterm babies.
Full-band electroencephalography (FbEEG) recordings were
obtained from 16 babies at conceptional ages between 32.8
and 40 wk. We examined the SATs and the intervening ongo-
ing cortical activities (inter-SAT; iSAT) with average wave-
forms, their variance and power, as well as with wavelet-
based time-frequency analyses.
Our results show, that the low frequency power and the
variance of the average waveform of SAT decrease during
development. There was a simultaneous increase in the ac-
tivity at higher frequencies, with most pronounced increase
at theta-alpha range (4–9 Hz). In addition to the overall in-
crease, the activity at higher frequencies showed an in-
creased grouping into bursts that are nested in the low fre-
quency (0.5–1 Hz) waves. Analysis of the iSAT epochs
showed a developmental increase in power at lower frequen-
cies in quiet sleep. There was an increase in a wide range of
higher frequencies (4–16 Hz), whereas the ratio of beta (16–
30 Hz) and theta-alpha (4–9 Hz) range activity declined, indi-
cating a preferential increase at theta-alpha range activity.
Notably, SAT and iSAT activities remained distinct through-
out the development in all measures used in our study.
The present results are consistent with the idea that SAT
and the other ongoing cortical activities are distinct func-
tional entities. Recognition of these two basic mechanisms in
the cortical activity in preterm human babies opens new
rational approaches for an evaluation and monitoring of early
human brain function. © 2007 IBRO. Published by Elsevier
Ltd. All rights reserved.
Key words: FbEEG, DC-EEG, EEG, infraslow, neonatal, pre-
mature.
The traditional view about spontaneous neonatal brain
activity (i.e. background electroencephalography (EEG)) is
by and large focused on the development of the visually
apparent “continuity” of the EEG tracing (Torres and
Anderson, 1985; reviewed in e.g. Lamblin et al., 1999).
This thinking is built upon the observations from a high
pass-filtered EEG signal (cf. Vanhatalo et al., 2002, 2005a)
where intermittent delta-appearing waves are separated
by relative flattening, hence giving the appearance of a
discontinuous activity (called trace discontinu or trace al-
ternant, see e.g. Lamblin et al., 1999). The clinical view
about EEG maturation is further complemented by reason-
ing that, over the course of development, the relatively
flatter periods in the EEG trace become progressively
shorter so that the initially distinct activity bursts extend
and merge together to become visually “continuous”
shortly after full term (FT) age. Whereas this traditional and
empirically built framework has been successful in clinical
use, it has not been able to provide a neurophysiological
framework for the interpretation of the neonatal brain ac-
tivity.
It has been established in the basic developmental
neuroscience, that intermittent slow transients of sponta-
neous activity are a characteristic feature of various imma-
ture brain structures in all as yet studied species (reviewed
in e.g. O’Donovan 1999; Ben-Ari, 2001; Khazipov and
Luhmann, 2006; Vanhatalo and Kaila, 2006; see also Ad-
elsberger et al., 2005). Such activity is now widely believed
to be crucial for the establishment of the first neuronal
connections during the time period before external input
via sensory systems begin to drive the activity-dependent
development and plasticity (cf. Pallas, 2001). We have
recently demonstrated in human preterm babies that spon-
taneous activity transients (SAT) of this kind are present
and readily recorded during the developmental time win-
dow that equals to the last trimester of normal human
pregnancy (Vanhatalo et al., 2002, 2005a; see also Thord-
stein et al., 2006). In agreement with a number of prior
animal studies, our recordings showed that majority of
brain activity is initially confined to these SAT events, and
that disappearance of SATs takes place near term age,
paralleling the maturation of functional cortical inhibitory
(GABAergic) neurotransmission (Vanhatalo et al., 2005a;
Vanhatalo and Kaila, 2006, Dzhala et al., 2005).
Our recent observations, as well as the rapidly expand-
ing literature on animal studies, have made it plausible to
propose that the EEG activity in preterm humans consists
of two distinct types of activity (reviewed in Vanhatalo and
Kaila, 2006): 1) SATs that have a specific developmental
function unique to the last trimester of pregnancy, and 2)
*Corresponding author. Tel: ?358-50-528-6119; fax: ?358-45-1990-485.
E-mail address: sampsa.vanhatalo@helsinki.fi (S. Vanhatalo).
Abbreviations: ANOVA, analysis of variance; CA, conceptional age;
EEG, electroencephalography; ePT, early preterm; FbEEG, full-band
electroencephalography; FFT, fast Fourier transformation; FT, full-
term; iSAT, inter-spontaneous activity transient; RMS, root mean
square; SAT, spontaneous activity transient; TF, time-frequency.
Neuroscience 145 (2007) 997–1006
0306-4522/07$30.00?0.00 © 2007 IBRO. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.neuroscience.2006.12.070
997

Page 2

the other cortical activity (hereafter called inter-spontane-
ous activity transient, iSAT) that represents the progres-
sively increasing ongoing brain activity, which continues
throughout the rest of the life.
In the present study, we set out to test the idea that
these two distinct components of brain activity do also
show differential development during preterm period. Be-
cause EEG activity in a sleeping preterm baby co-fluctu-
ates with sleep states, showing either more continuous or
more discontinuous EEG states (which roughly correspond
to active or quiet sleep states, respectively), we also de-
cided to study brain activity differentially in these EEG
states. The present results are consistent with the idea that
SAT and iSAT are two distinct brain activities with different
developmental trajectories.
EXPERIMENTAL PROCEDURES
Subjects and EEG recording
We studied 16 neurologically healthy infants (conceptional age
(CA) from 32 to 46 wk) during postprandial sleep. Scalp EEG was
recorded with a direct current (DC) -coupled amplifier (i.e. full-
band electroencephalography, FbEEG; reviewed in Vanhatalo et
al., 2005b,c) and four to eight Ag AgCl electrodes (type LP220, In
Vivo Metric, Healdsburg, CA, USA) with acquisition reference
placed on the ipsilateral mastoid (cf. Vanhatalo et al., 2002,
2005a). Most recordings were unilateral because of the height of
the electrode holders (8 mm). Polygraphic channels (electrocar-
diography, electrooculogram, electromyogram) and continuous vi-
sual observation were used to aid vigilance state classification and
artifact recognition. Baseline drifts and skin-borne potentials were
eliminated by short-circuiting them at the epithelium (Picton and
Hillyard, 1972; see also, Vanhatalo et al., 2005c). Signals were
acquired at 500 Hz by a 12-bit data acquisition board, and ana-
lyzed offline. The experiments were conducted with the under-
standing and the written (informed) consent obtained from the
parents. This study was approved by the Ethics Committee of
Hospital for Children and Adolescents, Helsinki University Central
Hospital.
EEG analysis
Only sleep periods were included in the analysis, and the EEG
was initially categorized into discontinuous (trace discontinu and
trace alternant) and continuous EEG states using conventionally
filtered signal (i.e. high pass filter at 0.5 Hz), which closely corre-
sponded to the quiet and active sleep stages, respectively. Fully
continuous epochs with slow wave sleep in full-term infants were
excluded from the analysis. Further time series analysis was
performed on the EEG signal without high pass filtering from the
occipital derivation re-referenced to vertex (Cz).
Identification and collection of iSAT/SAT epochs
Quantitative analysis of epochs containing iSAT and SAT were
made from visually collected, 10-s-long segments of EEG (n?797;
515 epochs from discontinuous EEG in 14 babies and 282 epochs
from continuous EEG in 13 babies). For generation of the data-
base containing these epochs, the EEG signal was down-sampled
to a sampling rate of 125 Hz and low pass filtered at 30 Hz (stop
band 45 Hz), and the epochs were centered to the visually iden-
tified onset of SAT, in order to have an array of 5-s signals from
iSAT and SAT in the same EEG epoch. We excluded epochs with
visually detectable artifacts in any part of the 10-s epoch, or if
there was another (i.e. very closely preceding) SAT within the
5-s-long iSAT period.
SATs are characterized by an increase in the activity at mul-
tiple higher frequency bands, ranging from the infraslow (Van-
hatalo et al., 2004) to higher frequency activities, so that the higher
activities are “nested” in the lower frequency waves (see also
Vanhatalo et al., 2005). Thus, SATs were identified by searching
for a simultaneous increase in infraslow and higher frequencies. In
practice, this was readily achieved by generating two copies of the
same trace: One trace was used for looking at the low frequency
deflections (taken from a non-filtered; i.e. FbEEG trace) and the
other trace was high pass filtered (passband 10 Hz, stop band
6 Hz; amplified 15 times more) for aiding visual identification of the
increases in the faster activity. We chose to study iSAT activity
from the period immediately preceding SAT because we felt that
any other criteria would be inherently more ambiguous. Namely, it
has been already reported that the length of SATs may vary by
several seconds (Vanhatalo et al., 2002, 2005a), while the mean
frequency of SAT occurrence is about 10 s (evidence reviewed in
Vanhatalo and Kaila, 2006). These facts together make it very
difficult if not impossible to define other unambiguous onset times
for iSAT epochs.
Signal analysis
The EEG activity was analyzed by four complementary methods:
First, superimposing and averaging (see Fig. 2A) was performed
to estimate how constant the iSAT/SAT epochs were within a
given age group, and how did they change during development.
Variance analysis (see Fig. 2B) was performed to see possible
developmental changes in the variability in the amplitude of iSAT/
SAT epochs. Second, root mean square (RMS) (see Fig. 3) was
calculated from the iSAT and SAT epochs (after low pass filtering
at 5 Hz; stop band at 10 Hz) to quantify the change in the total
power of the slow signal component. Third, amplitude spectra (see
Fig. 4) were computed from both iSAT and SAT epochs to dem-
onstrate whether persistent (vs. transient, cf. wavelets below)
changes occurred at specific frequencies. Amplitude spectra were
computed for detrended 5-s-long epochs, one for iSAT and one for
SAT, without windowing. We used amplitude spectra (i.e. square
root of power spectra) in order not to add unnecessary non-
linearity into the data visualization. In addition to the naïve ampli-
tude spectrum, we also calculated normalized amplitude spectra
by using the iSAT period of each EEG epoch as reference. Fourth,
we used Morlet wavelets (m?6.5; frequency separation 1 stan-
dard deviation) for visualizing the time-frequency characteristics of
iSAT/SAT epochs (see Fig. 6). The amount of activity at specific
high frequency bands was further quantified by taking the cumu-
lative amount (i.e. time integral) of the signal envelope filtered with
wavelets within the given frequencies (iSAT from 1 to 4 s before
and SAT from 0 to 3 s after visually identified onset of SAT). The
wavelet-filtered bands were: 4–9 Hz band was obtained from five
wavelets with the center frequencies from 4.6–8.6 Hz, 9–16 Hz
band was obtained from four wavelets with the center frequencies
from 9.9–15.9 Hz; 16–30 Hz band was obtained from four wave-
lets with the center frequencies from 18.4–29.7 Hz.
Data of all these analyses were calculated first for each
individual, and then grouped into three age groups: early preterm
(ePT, CA ?36 wk; n? 7), preterm (PT, CA 36–39 wk; n?5) and
FT (CA ?39 wk; n?4).
For statistical analysis, we calculated variance, Spearman
(one way) correlation coefficients, as well as an analysis of vari-
ance (ANOVA). We considered P-value 0.05 or lower to be sta-
tistically significant.
Methodological considerations
While the earlier literature have almost exclusively used fast Fou-
rier transformation (FFT) for frequency domain analysis of neona-
tal EEG, we felt it was necessary to supplement FFT with the other
three quantitative methods (superimposing/averaging, RMS and
M. Tolonen et al. / Neuroscience 145 (2007) 997–1006 998

Page 3

wavelets) for the following reasons: 1) FFT is based on an as-
sumption that the signal is stationary, whereas the recurrent SAT–
iSAT epochs are not stationary. 2) The slowest signals have so
low frequencies (0.05–0.2 Hz; see Vanhatalo et al., 2005a), that
FFT can not reliably quantify them because of the unavoidably
short analysis window. 3) RMS is a superior method in estimating
the “magnitude,” or power, of irregular signals, such as the in-
fraslow deflections in SATs, and 4) wavelets have a superior
temporal resolution for time frequency (TF) analysis. The temporal
resolution of wavelets is inversely correlated to the frequency of
interest, which becomes important in the present application
where SAT epochs are a priori known to comprise a wide range of
frequencies with only short bursts of higher frequency activity.
The present study was based on a visual identification of
SATs, which is conventionally considered the method of choice for
identifying prominent EEG events. It is possible, though, that
some bias might be introduced to the visual selection of qualified
(i.e. collected) SAT epochs, especially so at times when the
ongoing EEG activity is relatively higher in amplitude (e.g. during
continuous EEG state). We feel, however, that this potential bias
would at most lead to an underestimation of the developmental
changes reported in the present study.
RESULTS
Visual inspection of the preterm EEG revealed conspicu-
ous SATs during both continuous and discontinuous EEG
states, and at all CAs (Fig. 1; see also Vanhatalo et al.,
2005a/Fig. 1). The SATs were visually characterized by
the co-incidence of the infraslow (?0.5 Hz) activity, which
typically contained one to five repeats of nested low fre-
quency waves (0.7–1.5 Hz). These waves, in turn, con-
tained another set of nested (i.e. overriding) waves at
several higher frequency bands (Fig. 1C, see also Van-
hatalo et al., 2005a). Such co-incidence in multiband ac-
tivity was also seen as a significant (P?0.01) intra-individ-
ual correlation in the amount of activity in 16–30 Hz band
to the amount of activity in 9–16 Hz band (r?0.84) and
4–9 Hz band (r?0.75). High pass filtering of the signal at
0.5 Hz (i.e. the typical lower limit for routine clinical EEG)
shows that such manipulation results in a severe distortion
of SATs, making them appear as delta frequency waves
(Fig. 1A; see also Vanhatalo et al., 2002), which has been
traditionally considered the hallmark of premature EEG.
Quantitative analysis of SAT
Visual comparison of individual SATs at different ages
shows that the slow “carrier wave” of SATs was often
higher in amplitude, and mono- or biphasic in the younger
preterms, whereas it more often was lower in amplitude
and consisted of multiple phases in the babies closer to
term age (Figs. 1 and 2A). This difference suggested that
the development of the total power of SAT and its peak
amplitude should be studied differentially, and that the
variability in the signal may be higher in the younger neo-
nates. The individual grand average of the total amount of
low frequency activity in continuous EEG states, estimated
by FFT and RMS, decreased significantly during preterm
development (r??0.53, ?0.05; see Figs. 3–5). The de-
crease in lowest frequencies was also significant in the
FFT analysis from the raw EEG epochs (Fig. 4). In addi-
tion, the peak variability of the instant SAT amplitudes
decreased during development in both EEG states (Fig.
2B), suggesting that the developmental changes in SATs
involve more than a mere change in the peak amplitude or
power of the lowest frequencies. Comparison of the low
frequency (?2 Hz) spectra of SATs in continuous and
discontinuous EEG states showed that, while they are
quite different early in development, they become strikingly
similar closer to term age (see the lowest graphs in Fig. 4).
A close inspection of the FFT graphs (Fig. 4) sug-
gested possible small differences between the youngest
and the oldest infants at higher (?2 Hz) frequencies, which
appeared statistically significant after normalization of the
amplitude spectra from discontinuous EEG states (Fig. 5,
upper right graph). Due to the poor temporal resolution of
FFT method, and the episodic nature of the higher fre-
quency oscillations (see Fig. 1C), we reasoned that a TF
analysis of 2–35 Hz activity with wavelets could better
reveal possible developmental differences at higher fre-
quencies. Subtraction of the values of ePT from those of
FT (see the lowest TF graphs in Fig. 6) shows that the
increase in about 8–20 Hz activity at FT is increasingly
organized into bursts, especially so during discontinuous
EEG state, but also to some degree in the continuous EEG
state. Quantification of the absolute cumulative power by
wavelets of the first 3 s of SAT activity during discontinu-
ous EEG states demonstrated a statistically significant
correlation with age (r?0.62, P?0.05 for 4–9 Hz, and
r?0.55, P?0.05 for 4–30 Hz). The increase in lower (the-
ta-alpha) frequencies (4–9 Hz) was more pronounced (Fig.
6B), and clearly seen also in the FFT analysis (see Fig. 5,
hump in the spectra from both EEG states). Moreover, this
was also reflected in the age-dependent decline in the
activity ratio of higher frequencies to lower frequencies
(16–30 Hz vs. 4–9 Hz; see Fig. 6B) during both discontin-
uous (r??0.57, P?0.05) and continuous (r??0.419,
P?0.151; a trend only) EEG states.
Quantitative analysis of iSAT
Visual comparison of the iSAT epochs at different CAs
shows expectedly (cf. Lamblin et al., 1999; Vanhatalo and
Kaila, 2006) that the overall iSAT activity during discontin-
uous EEG state increases during development, while the
change is visually not so obvious during continuous EEG
states (Fig. 2A). This is also seen quantitatively in the
grand average of the total low frequency power (RMS and
FFT) of iSAT, which shows a maturational increase in
EEG activity during discontinuous EEG states (RMS:
r?0.78; P?0.01). Interestingly, comparison of the low
frequency (?2 Hz) spectra of iSATs in continuous and
discontinuous EEG states showed that, despite their
differences early in development (see Fig. 4; two upper
left graphs), they become quite similar closer to term
age (see the lowest graphs in Fig. 4).
A further inspection of the FFT analysis of iSATs raises
an idea that there may be an increase in the amplitude of
higher frequency (?2 Hz) activity in both continuous and
discontinuous EEG states. While the increase in higher
frequencies was statistically significant in the normalized
amplitude spectra from discontinuous EEG state (Fig. 5
M. Tolonen et al. / Neuroscience 145 (2007) 997–1006999

Page 4

upper left graph), the developmental change in the low
frequency activity (below 2 Hz) was opposite between
EEG states. Next we examined the idea that the little
age-dependent difference at higher frequencies could be
better seen in the TF analysis with wavelets. Gross inspec-
tion of the data at group level showed remarkably similar
TF characteristics in both ePT and FT babies (Fig. 6).
Subtraction of ePT from FT, however, discloses a marked
increase in power between 2 and 30 Hz during discontin-
uous EEG, and between 5 and 30 Hz during continuous
EEG states. This increase was also seen at the individual
level where the cumulative power at 4–9 Hz and 9–16 Hz
bands showed a statistically significant correlation with
age (r?0.83, P?0.01 and r?0.77, P?0.01, respectively;
see also Fig. 6B) during discontinuous EEG. The accen-
tuated increase in power at 4–9 Hz frequency band was
also seen in the FFT analysis (see Fig. 5; a hump in the
spectra from iSAT during discontinuous EEG), and re-
flected in the age-dependent decline of the power ratio
(16–30 Hz vs. 4–9 Hz) during both discontinuous and
continuous EEG (r??0.83, P?0.01, and ?0.73, P?
0.05, respectively; Fig. 6B).
Fig. 1. SATs in the premature EEG. (A) Four-minute-long epoch of a FbEEG recordings of a premature neonate (32.8 wk after conception)
demonstrates conspicuous cortical activity which consists of slow, very high amplitude transients with nested higher frequency activity. These SATs
are present during both continuous and discontinuous EEG activity. SATs are also observed in the full-term neonate (39 wk). (B) Representative SATs
are indicated and shown on an expanded scale in the three insets (B; a and b, 32.8 wk; c, 39 wk). For a comparison between FbEEG tracing and the
tracing obtained by filter settings of the conventional clinical EEG, the long EEG segments have been high-pass filtered at 0.5 Hz. It is notable that
this leads to a severe distortion of the SAT events. (C) TF analysis with wavelets of a long SAT event (35 wk CA) demonstrates that SATs are
characterized by intermittent bursts of higher frequency activity that are grouped into bursts nested in the low frequency (about 1 Hz) events, which,
in turn are nested in the infraslow transient (carrier wave). Derivation of all signals in Fig. 1 is O1-Cz; a.e.?amplitude envelope; very slow drifting of
the baseline is removed by high pass filtering at 0.015 Hz.
M. Tolonen et al. / Neuroscience 145 (2007) 997–10061000

Page 5

DISCUSSION
While EEG recordings of human preterm babies have
been available for routine clinical practice for several de-
cades (history reviewed in Lamblin et al., 1999), the anal-
ysis of neonatal EEG has been driven by empiric correla-
tions between EEG and subject’s clinical state rather than
by a physiological hypothesis. The present study extends
and departs from the available literature on EEG spectral
features in two aspects. First, our data segmentation, the
basis of all spectral analysis, was driven by our a priori
hypothesis, and second, we utilized several complemen-
tary analysis paradigms to reduce the ambiguity that is
inherent to all mathematical analyses (see the section
Methodological considerations). The conflicts between the
nature of neonatal EEG signal and the a priori assumptions
of the common spectral analysis tools pose limitations to
the interpretation of most existing literature. Despite these
concerns, our present observations are fully in line with the
general findings from the earlier reports that there is an
initial abundance of slow activity, which is gradually super-
seded by the increasing amount of higher frequency activ-
ity (see e.g. Scher et al., 1994; Lamblin et al., 1999;
Vanhatalo et al., 2005a; Okumura et al., 2006). The mat-
urational decrease in amplitude described by some earlier
work (see e.g. Lamblin et al., 1999, page 28) is not in
conflict with our results, because the earlier studies have
reported jointly on all EEG signal components, including
the slowest events and their filter-distorted rebound com-
ponents (cf. Vanhatalo et al., 2002, 2005a,b,c). An analy-
Fig. 2. Development of the low frequency waveform characteristics of iSAT and SAT. (A) Superimposition of all collected 10-s-long epochs with
iSAT/SAT are categorized into three different age groups, as well as into discontinuous and continuous EEG periods. The grand average of each group
(thick black line) demonstrates a striking overall similarity of the average SAT waveform throughout the development, whereas the variance (B) of the
instant amplitude demonstrates that the amplitudes of SATs are more variable in the younger babies. The developmental decrease in the variability
of SAT activity is clearly present in both EEG states during the first 1–2 s of SATs. Epochs of iSAT activity during discontinuous EEG period undergo
a progressive increase in amplitude (and in the variance), whereas amplitude variability is decreased in iSATs of continuous EEG. Superimposed
traces are taken from signals after low -pass filtering at 5 Hz. The vertical black bar above the variability graphs indicates time periods with significant
differences (ANOVA) between all age groups, whereas the gray bar indicates portions of the epochs that were significant between youngest and oldest
age groups. Graphs in A include all traces (n?797) used for the analysis in this study, and the average waves represent a grand average of all the
traces shown in gray in the same plot.
100
50
0
35
40 wks
200
100
0
35
RMS
RMS
40 wks
Discontinuous EEG
Continuous EEG
iSAT
SAT
Fig. 3. Development of the total power of iSAT and SAT. Estimation of
the total power of the iSAT and SAT episodes by RMS shows that
there is a clear increasing trend in power in iSAT during discontinuous
EEG, as well as a clear decrease in SAT power during continuous
EEG.
M. Tolonen et al. / Neuroscience 145 (2007) 997–10061001

Page 6

sis by visual perception from such traces does obviously
result in a notion that gross amplitude decreases with
aging (see also Vanhatalo et al., 2005a), whereas our
present study presents quantification of the higher fre-
quencies that are characteristic of more mature EEG
activity.
The present work was set out to examine the recently
emerged idea (discussed in detail in Vanhatalo and Kaila,
2006; see also Khazipov and Luhmann, 2006), that brain
activity in human preterm babies is composed of two dis-
tinct components: i) SATs which are present until FT age
only, and likely have a unique role in the network devel-
opment, and ii) the other activity (called iSAT in the present
paper) that represents the gradually emerging ongoing
cortical activity. According to this idea, it was plausible to
expect that the developmental changes in SAT activity are
Fig. 4. Development of the amplitude spectra of iSAT and SAT. Amplitude spectra calculated with FFT of SAT and iSAT epochs during both EEG
states demonstrates that the most pronounced developmental changes in all groups are in the amount of low frequency (below 2 Hz) activity. Black
bars in each graph depict the parts of spectra that was significantly (ANOVA) different between groups. In the row Continuous vs. Discontinuous, the
horizontal bars depict the portions of the above amplitude spectra that show a statistical difference (ANOVA) between discontinuous and continuous
EEG states in each age group (gray for ?36 wk and black for ?38 wk). Toward FT age, the low frequency power spectra becomes strikingly similar
between discontinuous and continuous EEG in both iSAT and SAT epochs, as seen from the ratio of FFTs from continuous/discontinuous EEG state
approaching value 1 at term age (see the lowest FFT plots). The thick lines represent grand averages of the younger (red) and older (blue) subjects,
whereas individual averages are shown with thin stippled lines (?36 wk: n?214 epochs during discontinuous EEG and n?99 epochs during
continuous EEG; for the group ?38 wk, n?217 and 54, respectively; total number of iSAT/SAT epochs analyzed for generation of these FFT graphs
is 584).
M. Tolonen et al. / Neuroscience 145 (2007) 997–10061002

Page 7

different from those of iSAT activity, whereas the develop-
mental changes within SAT and iSAT activities may be
rather similar regardless of the EEG states. In line with this
idea, the developmental change during SATs of lowest
frequencies (0.2–2 Hz) was similar in both EEG states, i.e.
a decrease of the activity in the FFT analysis, as well as a
decrease in the variance of instant amplitude (which
mostly reflects low frequencies). Moreover, in spite of the
marked differences in the low frequency (0.2–2 Hz) spec-
tra of iSAT and SAT epochs (see Fig. 2C), the ratio of the
FFT spectra between discontinuous and continuous EEG
states approached value 1 at term age in both iSAT and
SAT, respectively. This is compatible with the idea that in
spite of the markedly different overall appearance of EEG
signal during different vigilance states, among themselves
SATs and iSATs represent similar types of brain activity. In
addition, there was also a significant correlation between
the age and the amount of higher frequencies, most nota-
bly at theta/alpha range (4–9 Hz). This increase was seen
during both SAT and iSAT, but it was more pronounced
and more continuous during iSAT (i.e. spread throughout
the iSAT period; see Fig. 6A, lowest TF plots) than during
SAT. This is compatible with the idea that iSAT activity
develops gradually into the ongoing cortical activity (cf.
Vanhatalo and Kaila, 2006) which shows no such visually
observable temporal grouping. In contrast, SAT activity
preserves its overall temporal structure with activity group-
ing throughout the preterm development until they disap-
pear altogether soon after term age (see Vanhatalo et al.,
2005a).
We extend the prior literature by showing that the
developmental increase in higher frequency cortical activ-
ity is partly frequency dependent so that the increase in
4–9 Hz range is most notable, whereas the highest fre-
quencies (16–30 Hz) stay around the same or slightly
decrease. It is well known that the development of ongo-
ing, background activity in the posterior areas at theta/
alpha range goes through a successive increase in fre-
quency during the first year in life (Eeg-Olofsson, 1980;
Stroganova et al., 1999; Marshall et al., 2002). Our obser-
vation of the increasing amount of continuous 4–9 Hz
activity suggests that the development of the mechanisms
responsible for a continuous activity at these frequencies
may begin already during the last trimester of pregnancy.
The paucity of changes in the higher (beta and gamma)
frequencies is compatible with the idea that generation of
Fig. 5. Normalized amplitude spectra of iSAT and SAT epochs in preterm and fullterm babies. This figure shows the amplitude spectra of SAT and
iSAT epochs during both EEG states, after normalization by using iSAT period of each EEG epoch as the reference. Note the different amplitude
scaling in the FFT of SATs during discontinuous EEG state (upper right corner). Black bars in each graph depict the parts of spectra that were
significantly (ANOVA) different between groups. In the row Continuous vs. Discontinuous, the horizontal bars depict the portions of the above
amplitude spectra that show statistical difference between discontinuous and continuous EEG states in each age group (gray for ?36 wk and black
for ?38 wk). For details of epochs used to generate these figures, see the legend of Fig. 4.
M. Tolonen et al. / Neuroscience 145 (2007) 997–10061003

Page 8

these oscillations typically requires well-formed structural
intracortical inhibitory circuitry and its underlying molecular
machinery (such as carboanhydrase VII and KCC2 ex-
pression, reviewed in Rivera et al., 2005; Payne et al.,
2003), which are not yet mature at this developmental
phase (reviewed in Kostovic and Judas, 2002; Kostovic
and Jovanov-Miloševic ´, 2006; for KCC2 in humans, see
also Vanhatalo et al., 2005a).
Experimental animal studies (e.g. Dupont et al., 2006;
Adelsberger et al., 2005) have shown, and the clinical
observations (reviewed in Lamblin et al., 1999) have indi-
rectly supported the idea that SAT activity may be focal
Fig. 6. TF analysis with wavelets of iSAT and SAT. (A) TF analysis with wavelets demonstrates roughly similar characteristics of the whole iSAT/SAT
epochs between ePT and FT babies. However, comparison of the FT data to the ePT data by subtraction (FT minus ePT) discloses distinct
developmental changes: The most conspicuous difference is the increase of the ?2 Hz activity throughout the iSAT period in both EEG states,
whereas the increase in the higher frequency activity during SAT is more grouped into distinct temporal bursts (see also Fig. 1B). The apparent
decrease in first-second activity of SATs in the FT babies during continuous EEG (Fig. 6, the lowest TF plot in the right column) is likely due to the
fact that visual SAT identification may favor collection of SATs with short duration and high amplitude in the ePT (see also Fig. 2A). TF graphs
representing each group (two upper rows) are generated as averages of 50 iSAT/SAT epochs in each group, hence altogether 200 iSAT/SAT epochs
were used for Fig. 6. Amplitude scaling (see the pseudocolor bar) in the upper two rows is from 0 to 25 ?V, in the subtraction images from ?5–5 ?V
(discontinuous EEG) and from ?3–3 ?V (continuous EEG). (B) Relation between cumulative power vs. CA is shown both for the theta/alpha (4–9 Hz)
frequency band, as well as for the ratio of theta/beta (4–9 Hz vs. 16–30 Hz) frequency bands. There was a conspicuous increase in the 4–9 Hz band,
and a decrease in beta/theta ratio in all EEG epochs. Scaling in the vertical (y) axis is arbitrary, and represents the area under the envelope curve
(i.e. cumulative amount of power (in uV) within the 3-s time epoch). See text for further details.
M. Tolonen et al. / Neuroscience 145 (2007) 997–10061004

Page 9

and purely cortical during early development, and it be-
comes globally more synchronized during the last trimester
of pregnancy (reviewed in Vanhatalo and Kaila, 2006).
Direct cortical and brain stem recordings from animals
(Stenberg, 1973), and indirect evidence of brain stem ac-
tivity by analysis of heart rate variability in humans
(Pfurtscheller et al., 2005) have shown that, toward the FT
age, the SAT activity ultimately develops into a partially
synchronous (i.e. coincident) activity that involves both
hemispheres, as well as deeper brain structures. Thus, the
extent of global spread of SAT events may reflect the
extent of functional network at the given developmental
phase (for details of the anatomical development, see
Kostovic and Jovanov-Miloševic ´, 2006). This evolving spa-
tial extent of SAT events offers a natural explanation also
for the developmental changes that we found in the low
frequencies: SATs in the ePTs are focal and hence tend to
show not only high amplitudes but also a high variability in
the amplitude due to the fact that a focal SAT activity may
have highly varying spatial relation to the electrode posi-
tions. In contrast, SATs in the FT babies are much more
widely spread, and hence their amplitude by default cannot
show such high amplitudes in the differential EEG record-
ings. It is also obvious that an increasing gyration together
with the only partial global synchrony of SATs in the FT
babies make them more likely to appear as longer events
with “alternating” activity between electrodes/brain areas
(see also Figs. 1A and 2/last row).
CONCLUSION
In conclusion, the present study underlines the importance
of exploiting the experimental literature for recognition of
the particular properties of developing cortical activity,
such as SAT and iSAT in preterm humans. At this devel-
opmental phase, cortex is still poorly connected to the
sensory input (see Vanhatalo and Lauronen, 2006), the
corticocortical connections are sparse, and the inhibitory
(GABAergic) transmission is not yet present (see Judas et
al., 2005; Kostovic and Jovanov-Miloševic ´, 2006; Van-
hatalo et al., 2005a). Such fundamental differences in the
preterm brain compared with adults make it obvious that
EEG activity, including its spectral properties, cannot be
interpreted within the same neurophysiological framework
as EEG activity at older ages (cf. Buzsaki and Draguhn,
2004; Steriade, 2006). In addition, our present and earlier
(Vanhatalo et al., 2005a) observations clearly emphasize
the need to recognize SAT activities in the neonatal EEG
classification, as they occur in both sleep states and
throughout the preterm life. This idea is fully in line with the
recent animal studies, and underlines the developmental
importance of SATs as a core element of developing brain
activity (Adelsberger et al., 2005; Khazipov and Luhmann,
2006; Vanhatalo and Kaila, 2006). Recognition of such a
close correlation between human and animal data will also
open new possibilities to the design of experimental and
interventional studies whose results may be easily inter-
polated to human recordings. Furthermore, recognition of
SATs as independent and developmentally crucial events
will open relatively straightforward, neurophysiolocally de-
signed strategies for long term monitoring of brain function
in the neonatal intensive care unit.
REFERENCES
Adelsberger H, Garaschuk O, Konnerth A (2005) Cortical calcium
waves in resting newborn mice. Nat Neurosci 8:988–990.
Ben-Ari Y (2001) Developing networks play a similar melody. Trends
Neurosci 24:353–360.
Buzsaki G, Draguhn A (2004) Neuronal oscillations in cortical net-
works. Science 304:1926–1929.
Dupont E, Hanganu IL, Kilb W, Hirsch S, Luhmann HJ (2006) Rapid
developmental switch in the mechanisms driving early cortical
columnar networks. Nature 439:79–83.
Dzhala VI, Talos DM, Sdrulla DA, Brumback AC, Mathews GC, Benke
TA, Delpire E, Jensen FE, Staley KJ (2005) NKCC1 transporter
facilitates seizures in the developing brain. Nat Med 11:1205–
1213.
Eeg-Olofsson O (1980) Longitudinal developmental course of electri-
cal activity of brain. Brain Dev 2:33–44.
Judas M, Rados M, Jovanov-Milosevic N, Hrabac P, Stern-Padovan R,
Kostovic I (2005) Structural, immunocytochemical, and MR imag-
ing properties of periventricular crossroads of growing cortical
pathways in preterm infants. AJNR Am J Neuroradiol 26:
2671–2684.
Khazipov R, Luhmann HJ (2006) Early patterns of electrical activity in
the developing cerebral cortex of humans and rodents. Trends
Neurosci 29:414–418.
Kostovic I, Judas M (2002) Correlation between the sequential in-
growth of afferents and transient patterns of cortical lamination in
preterm infants. Anat Rec 267:1–6.
Kostovic I, Jovanov-Miloševic ´ N (2006) The development of cerebral
connections during the first 20–45 weeks’ gestation. Semin Neo-
natol 11:415–422.
Lamblin MD et al. (1999) Electroencephalography of the premature
and term newborn. Maturational aspects and glossary. Neuro-
physiol Clin 29:123–219.
Marshall PJ, Bar-Haim Y, Fox NA (2002) Development of the EEG
from 5 months to 4 years of age. Clin Neurophysiol 113:1199–
1208.
O’Donovan MJ (1999) The origin of spontaneous activity in developing
networks of the vertebrate nervous system. Curr Opin Neurobiol
9:94–104.
Okumura A, Kubota T, Tsuji T, Kato T, Hayakawa F, Watanabe K
(2006) Amplitude spectral analysis of theta/alpha/beta waves in
preterm infants. Pediatr Neurol 34:30–34.
Pallas SL (2001) Intrinsic and extrinsic factors that shape neocortical
specification. Trends Neurosci 24:417–423.
Payne JA, Rivera C, Voipio J, Kaila K (2003) Cation-chloride co-
transporters in neuronal communication, development and trauma.
Trends Neurosci 26:199–206.
Pfurtscheller K, Muller-Putz GR, Urlesberger B, Dax J, Muller W,
Pfurtscheller G (2005) Synchronous occurrence of EEG bursts and
heart rate acceleration in preterm infants. Brain Dev 27:558–563.
Picton TW, Hillyard SA (1972) Cephalic skin potentials in electroen-
cephalography. Electroencephalogr Clin Neurophysiol 33:419–
424.
Rivera C, Voipio J, Kaila K (2005) Two developmental switches in
GABAergic signalling: the K?-Cl- cotransporter KCC2 and car-
bonic anhydrase CAVII. J Physiol 562:27–36.
Scher MS, Sun M, Steppe DA, Guthrie RD, Sclabassi RJ (1994)
Comparisons of EEG spectral and correlation measures between
healthy term and preterm infants. Pediatr Neurol 10:104–108.
Stenberg D (1973) Ontogenetic development of gross electrical
activity in the lower brain stem and neocortex of the guinea pig,
pp 1–66. PhD Thesis. University of Helsinki.
M. Tolonen et al. / Neuroscience 145 (2007) 997–1006 1005

Page 10

Steriade M (2006) Grouping of brain rhythms in corticothalamic sys-
tems. Neuroscience 137:1087–1106.
Stroganova TA, Orekhova EV, Posikera IN (1999) EEG alpha rhythm
in infants. Clin Neurophysiol 110:997–1012.
Thordstein M, Lofgren N, Flisberg A, Lindecrantz K, Kjellmer I (2006)
Sex differences in electrocortical activity in human neonates. Neu-
roreport 17:1165–1168.
Torres F, Anderson C (1985) The normal EEG of the human newborn.
J Clin Neurophysiol 2:89–103.
Vanhatalo S, et al. (2002) DC-EEG discloses prominent, very slow
activity patterns during sleep in preterm infants. Clin Neurophysiol
113:1822–1825.
Vanhatalo S, Voipio J, Kaila K (2004) Infraslow EEG activity. In:
Electroencephalography. Basic principles, clinical applications and
related fields, 5th edition (Niedermeyer E, Lopes da Silva F, eds),
pp 489–493. Baltimore: Williams & Wilkins.
Vanhatalo S, Palva JM, Andersson S, Rivera C, Voipio J, Kaila K
(2005a) Slow endogenous activity transients and developmental
expression of K?-Cl- cotransporter 2 in the immature human cor-
tex. Eur J Neurosci 22:2799–2804.
Vanhatalo S, Voipio J, Kaila K (2005b) Full-band EEG (FbEEG): an
emerging standard in electroencephalography. Clin Neurophysiol
116:1–8.
Vanhatalo S, Voipio J, Kaila K (2005c) Full-band EEG (fbEEG): a new
standard for clinical electroencephalography. Clin EEG Neurosci
36:311–317.
Vanhatalo S, Kaila K (2006) Development of neonatal EEG activity:
from phenomenology to physiology. Semin Fetal Neonatal Med 11:
471–478.
Vanhatalo S, Lauronen L (2006) Neonatal SEP: back to bed-
side with basic science. Semin Fetal Neonatal Med 11:464–
470.
(Accepted 23 December 2006)
(Available online 20 February 2007)
M. Tolonen et al. / Neuroscience 145 (2007) 997–10061006