Large-scale coupling in neuronal activity is essential in all cognitive functions, but its emergence and functional correlates are poorly known in the human newborn. This study aimed to characterize functional connectivity in the healthy human newborn, and to identify the changes in connectivity related to vigilance states and to maturation during the early postnatal weeks. We recorded active and quiet sleep of 38 sleeping newborn babies using multichannel electroencephalography (EEG) at 2 neonatal time points. Functional connectivity between brain areas was quantified with 3 different metrics: phase–phase correlations, amplitude–amplitude correlations (AACs), and phase–amplitude correlations. All functional connectivity measures changed significantly between vigilance states and matured rapidly after normal birth. The observed changes were frequency-specific, most salient in AAC coupling, and their development was compatible with the known development of structural cortico-cortical connectivity. The present findings support the view that emerging functional connectivity exhibits fundamental differences between sleep states months before the onset of genuine EEG signatures of sleep states. Moreover, our findings also support the idea that early cortical events entail different mechanisms of functional coupling needed to provide endogenous guidance for early activity-dependent development of brain networks.
Spontaneous brain activity exhibits large-scale spatio-temporal correlations that are observed both in neuronal electric activity and in fluctuations of cerebral blood flow (Biswal et al. 1995; Jerbi et al. 2010), and they are correlated with behavioral states (Raichle 2010; Palva and Palva 2011a; Hutchison et al. 2013). It is now well documented in the direct electrical measures of neuronal activity that a range of brain functions emerge in these networks that bind together into dynamically changing functional constellations (Bressler and Menon 2010; Uhlhaas et al. 2010; Palva and Palva 2011a).
Little is known about the early ontogenetic emergence of functional communication in the newly formed neuronal networks, though recent anatomical studies have revealed detailed insight into the microscopic development of brain-wide networks in the human fetus (Kostović and Jovanov-Milošević 2006; Kostović and Judaš 2010). The first step to understand the emerging dynamics in early brain function would be to identify the neuronal mechanisms that mediate interareal interactions in the newborn brain. Some features of large-scale spatial coordination in brain electric activity have been reported in sleeping human newborns (Vasung et al. 2011; Tokariev et al. 2012; Kostović et al. 2014; Omidvarnia et al. 2014). Prior work on preterm infants has disclosed 2 modes of brain activity (Vanhatalo and Kaila 2006) where level of cortical activity alternates in sub-second time scales from very low level of activity to erratically appearing bursts of very high activity, hence called low and high modes, respectively (Omidvarnia et al. 2014). The subsequent connectivity analysis showed that these early brain modes exhibit drastically different levels of spatial correlation where most of large-scale correlations are related to epochs of high mode, or high cortical activity (Omidvarnia et al. 2014). Such “functional bimodality” is fully compatible with the emerging view in developmental neurobiology that high activity mode associates with self-organizing spontaneous activity transients (SATs, also known as bursts in the electroencephalography (EEG) literature Vanhatalo and Kaila 2006; Palmu et al. 2010; Vanhatalo and Kaila 2010), while the low activity mode represents low-amplitude intervals between SATs (Vanhatalo et al. 2005; Tolonen et al. 2007). This bimodality also decreases towards the term age (Omidvarnia et al. 2014) as expected from the decline of SAT events (Vanhatalo et al. 2005; Myers et al. 2012).
Neuronal interactions in large-scale networks result in several kinds of correlation structures that reflect distinct facets of them. Representing neuronal activity as phase and amplitude dynamics in limited frequency bands, 3 salient forms of interareal correlations have been considered in many studies. Phase–phase correlations (PPCs), that is, phase synchrony, reflect consistent sub-second timing relationships in neuronal excitability windows, which regulate neuronal spiking and hence can both facilitate and suppress interareal communication (Singer 1999; Fries 2005; Womelsdorf et al. 2007; Palva and Palva 2011b). Amplitude–amplitude correlations (AACs), on the other hand, are characteristic in multisecond time scales and reflect co-modulations of overall neuronal activity levels and correlations in gross cortical excitability fluctuations (Hanslmayr et al. 2011; Palva and Palva 2011a; Hipp et al. 2012). Finally, phase–amplitude correlations (PACs), a.k.a. nested oscillations (Vanhatalo et al. 2005), refers to a cross-frequency interaction where the amplitude of the fast oscillation is correlated with the phase of the slower oscillation. Several studies have shown that PAC is widespread in brain dynamics and may be an essential systems-level mechanism for integrating and regulating large-scale neuronal processing distributed concurrently to several frequency bands (Vanhatalo et al. 2005; Canolty and Knight 2010; He et al. 2010; Engel et al. 2013). PAC may reflect either a modulation of the fast oscillations by the slower oscillation or the phase-locking of the slow oscillation to the amplitude peaks of the faster oscillation, or a bidirectional combination of these effects.
In the present study, we aimed to characterize neuronal interactions in the healthy human newborn brain. In particular, we hypothesized that mechanisms of neuronal interactions differ with respect to their underlying oscillatory frequencies and spatial extent. We also hypothesized that these network mechanisms change rapidly during the time of normal human birth (i.e., term age), and that these mechanisms would have functional correlates with behavioral states.
Materials and Methods
We studied sleeping newborn infants with 2 EEG recordings with an interval of a few weeks. This study design allowed us to assess the individual maturation around term age. Functional, or behavioral, correlates of the physiological phenomena observed in EEG were assessed by comparing 2 vigilance states: active sleep (AS) and quiet sleep (QS). During the awake state, newborn infants exhibit very frequent movements causing excessive artefacts, hence high quality multiminute EEG recordings needed for functional connectivity analyses can only be obtained during sleep.
Subjects and Recordings
The EEG was recorded during daytime sleep from 38 healthy fullterm born babies recruited as a control group for several research projects running in parallel in BABA center (eng.babacenter.fi) at Children's Hospital, Helsinki University Hospital. The gestational age was 40.55 ± 1.09 (mean ± SD) weeks. The EEG signals were collected at sampling frequency Fs = 500 Hz using NicOne EEG amplifier (Cardinal Healthcare/Natus, USA) and EEG caps (sintered Ag/AgCl electrodes; Waveguard, ANT-Neuro, Germany) with 28 channels positioned according to the International 10–20 standard (Waveguard, ANT-Neuro; for further details of the newborn EEG recording method, see Vanhatalo et al. 2008; Stjerna et al. 2012 and http://eng.babacenter.fi/videos/). For sleep-state assessment, we also included the following polygraphic channels: chin electromyogram, electrocardiogram, eye movements, and respiratory sensors. The study was approved by the Ethics Committee of the Children's Hospital, Helsinki University Central Hospital. Prior to recordings, parents or guardians had given an informed consent.
EEG recordings from these babies were done within the first few days after delivery (Rec1; mean ± SD: 41.26 ± 1.18 weeks) and again at a fixed conceptional age (CA) of 42 weeks (Rec2; mean ± SD: 42.60 ± 0.45 weeks). The postnatal age at Rec1 was 5.2 ± 1.5 (mean ± SD) days, hence the infants had passed the postnatal adaptation that in the EEG is commonly considered to take up to 3 days (Scher et al. 1995). The time difference between Rec2 and Rec1 was 13.9 ± 7.2 (mean ± SD) days, so our study focused on analysing connectivity maturation within the earliest neonatal weeks, the time before infants undergo their shift from neonatal to infant neurobehavior (Guzzetta et al. 2005). The recording session was continued until baby had gone through both vigilance states during sleep: AS and QS, as defined conventionally by observing criteria from both EEG and polygraphic channels (André et al. 2010).
Visual EEG review was done using NicoletOne Reader software. Good quality and relatively artefact-free EEG epochs of AS and QS were identified for further analyses after sleep-state annotation according to standard criteria with EEG and polygraphic channels (André et al. 2010). Visual rating was defined by consensus agreement (MV and SV) when needed.
Data epochs for further analysis were selected from both Rec1 and Rec2, and from both vigilance states. This yielded 4 groups of continuous 5-min EEG epochs for comparisons: AS-Rec1 (18 subjects), AS-Rec2 (33 subjects), QS-Rec1 (18 subjects), and QS-Rec2 (37 subjects). The epochs were converted to EDF format to be quantitatively analyzed using custom scripts in Matlab (Version R2012b, MathWorks, Natick, MA, USA) and Labview (Version 2012, National Instruments, Austin, Texas, USA) programming environments.
The EEG epochs were first rereferenced into 19 derivations (Fig. 1A) of current source density (CSD) transform (Perrin et al. 1989; Kayser and Tenke 2006a; Kayser and Tenke 2006b; Tenke and Kayser 2012; see also CSD Toolbox http://psychophysiology.cpmc.columbia.edu/software/CSDtoolbox). We used the following settings: spline flexibility m = 4 and smoothing factor λ = 0. This montage was chosen after detailed comparison of CSD and average montage (for details, seeTokariev et al. 2015).
The EEG signals were filtered with a set of finite impulse response filters into 13 frequency bands with central frequencies Fc = 0.5, 0.7, 1, 1.4, 2, 2.8, 4, 5.7, 8, 11.3, 16, 22.6, and 32 Hz. Pairs of low-pass and high-pass filters were used to filter out EEG signals within the frequency bands of interest around the Fc. Pass-band cut-off frequencies were taken as 0.85·Fc–1.15·Fc while stop-band cut-offs at 40 dB attenuation were 0.5·Fc and 1.5·Fc, respectively.
Finally, Hilbert transform was applied to filtered EEG time series to obtain the complex representation of signals and thereafter the amplitude and phase time series were used in subsequent connectivity analyses.
Functional connectivity was assessed between all pairs of EEG electrodes (edges; N = 171) for all frequency bands in each baby. To examine PPC, we used the phase locking value (Jervis et al. 1983; Lachaux et al. 1999) between the phase time series, whereas AAC was estimated with the Pearson correlation coefficient (CC) of the amplitude time series:
where φX and φY are the instantaneous phases of the analyzed complex time series X and Y. Their absolute values are Xabs and Yabs with respective standard deviations , and mean values , . While N denotes a time window length, k is the sample number, j is an imaginary unit, and |·| indicates the absolute value.
These connectivity measures are widely studied in adults (reviewed in Engel et al. 2013), as well as reported in neonates (Tokariev et al. 2012). Both PLV and CC are generally sensitive to volume conduction related to linear mixing. Unlike in adults, however, such mixing is small in the neonatal EEG (Odabaee et al. 2013) because the neonatal skull conductivity is much higher than that of the adults (Odabaee et al. 2014). Moreover, the CSD montage used in this study effectively minimizes volume conduction effects. Functional interaction measures PLV and CC were used to test whether EEG networks change their “spatial extent” and “strength” between vigilance states and with the maturation. Cross-frequency interactions via PACs were estimated using nestedness coefficient (NC), which refers to PLV between the phase of the low-frequency oscillations and the phase of the amplitude envelope of higher frequency oscillation after filtering with the same filter used for the lower frequency band (Vanhatalo et al. 2005). Notably, NC was always computed between 2 frequency bands within the same signal.
Spatial extent was quantified as the fraction of edges (connectivity density, K) that were significantly different between the tested groups, or showed correlation with the age in the given channel pair. Estimate K was computed for both PLV and CC at each frequency band; its sign (increase (K+) or decrease (K−)) indicates the fractions of statistically significant observations in the group comparisons. At the same time, the network strength assessment was done using absolute values of the edges. In this context, we formed spatial groups to compare connectivity estimates within (intra) and between (inter) hemispheres (Fig. 1A), as well as the ratio of intrahemispheric to interhemispheric synchrony at single-subject level.
Grand mean amplitude value over 19 channels in each subject was computed for every frequency band and for both vigilance states, which allowed the comparison of amplitudes between conditions, as well as their correlation with the age for each condition separately. We also studied the correlations between connectivity measures and interelectrode distance (IED) for every electrode pair to assess whether spatial smearing via volume conduction could potentially confound the findings (however, see Odabaee et al. 2013, 2014). Finally, we compared 2 possible montages, CSD and average montage, by analyzing the distribution bias in mean phase difference (Δθ) between signal pairs (see
Group comparisons were performed with Wilcoxon sign rank test, and maturational changes were assessed with Spearman CC (r). Significance level was set to P < 0.05. Multiple comparisons were controlled as described earlier (Palva et al. 2010) by removing 5% from the total number of tests (Nt), that is, the expected fraction of false positives from the all statistically significant observations so that the least significant observations were removed. In estimating the spatial extent, statistical tests were pooled together from all frequency bands (Nt = 171 × 13 = 2223). In comparing conditions (Rec1 vs. Rec2 or AS vs. QS) or correlation with the age, we combined hemispheric metrics from all frequency bands, both from PLV and CC in each case (Nt = 3 × 13 × 2 = 78). Likewise, post hoc control was used for all corresponding tests for grand NC comparison (Nt = 40) and average NC at every electrode comparison (Nt = 760) between vigilance states, amplitude group comparison (Nt = 39), and correlation with age (Nt = 26).
The present work characterized functional connectivity in the multichannel neonatal EEG to assess its changes during early weeks after birth and between vigilance states. Vigilance state effects were assessed by comparing AS and QS periods, while the maturational changes were examined by estimating the correlation between functional connectivity and the CA at the time of EEG recording. Functional connectivity was estimated by PPC, AAC, and PAC. Both PPC and AAC analyses were done at the level of each pair of EEG electrodes, and the findings were expressed as the fraction of statistically significant connections (edges of the connectivity graph), which mirrors the spatial extent of the network change. We also computed hemispheric metrics where sets of signals were grouped together to examine changes that may occur at spatially larger level, especially inter- versus intrahemispheric level. The analysis of PAC, in turn, was computed for each EEG signal separately as a measure of a cross-frequency interaction between local slow (nesting) and faster (nested) oscillations. The main findings are schematically summarized in Table 1.
|AS → QS||Maturation|
|Amp.||↑ (<11.3 Hz)|
↓ (>11.3 Hz)
|↑ (2–5.7 Hz in QS)|
n.s. in AS
|PPC||↑ (<1 Hz and >4 Hz)|
↓ (1.4–2.8 Hz)
|AAC||↓ (<1.4 Hz)|
↑ (4–16 Hz)
|↓ (All freq. in QS)|
n.s. in AS
|PAC||↑ (All freq.)||↓ (>10 Hz)|
|AS → QS||Maturation|
|Amp.||↑ (<11.3 Hz)|
↓ (>11.3 Hz)
|↑ (2–5.7 Hz in QS)|
n.s. in AS
|PPC||↑ (<1 Hz and >4 Hz)|
↓ (1.4–2.8 Hz)
|AAC||↓ (<1.4 Hz)|
↑ (4–16 Hz)
|↓ (All freq. in QS)|
n.s. in AS
|PAC||↑ (All freq.)||↓ (>10 Hz)|
Note: This table presents a collection of the most prominent and global changes that take place during the vigilance state shifts (from AS to QS), as well as during early postnatal maturation (from Rec1 to Rec2). The direction of change is shown with arrows for the depicted frequency range in amplitudes (Amp.), PPC, AAC, and PACs. Effects that were weak (asterisk, *) or nonsignificant (n.s.) are not depicted here.
PPC Relates to Vigilance State in a Wide Range of Frequencies
Comparison between vigilance states (Fig. 2A) showed that functional connectivity via PPC is widely changed when the infant shifts from one state to the other. In QS, brain had more significantly stronger connections at the lowest (<1 Hz) and higher (∼4–22.6 Hz) frequencies, whereas AS exhibited more significantly stronger connections at the middle range (∼1–4 Hz). Topological inspection of selected frequency bands suggested that these differences have no visually apparent spatial constellations. The global difference was confirmed by hemispheric metrics where corresponding frequency ranges showed significant difference in both intra- and interhemispheric PPC (Fig. 2B). Interhemispheric PPC was significantly higher in QS at the lowest (<1 Hz) and most of higher (>8 Hz) frequencies, while it was significantly higher in AS at the range between them (1.4–4 Hz). The same pattern with significantly higher PPC during QS was seen at the lowest (0.5 Hz) and highest (4–22.6 Hz) frequencies in the intrahemispheric PPC. Comparison of the intra- versus interhemispheric PPC showed that intrahemispheric PPC was more prevalent in QS than in AS, especially at midrange frequencies (1.4–11.3 Hz).
Early Postnatal Maturation Does Only Affect PPC at Lower Frequencies
Correlation of PPC to infant's age showed significant developmental changes within the few weeks time studied around term age. Interestingly, this maturation was differently expressed between the vigilance states as well as among the oscillatory frequencies (Fig. 2C), and it was generally smaller in magnitude compared with differences between vigilance states (Fig. 2A). There was also a notable spatial heterogeneity in the way how individual connections were dependent on maturation, and no apparent topological patterns could be observed. A closer comparison of maturational changes between vigilance states suggested that AS was associated with relatively more changes in PPC in slow (0.5–0.7 Hz) and middle frequencies (1.4–5.7 Hz), whereas maturational changes during QS were related to low delta (1.4–2 Hz) and highest frequencies. Inspection of spatial topology of significant signal pairs (Fig. 2C) revealed a visually apparent abundance of interhemispheric connections at 2 Hz during QS. This observation was confirmed by the hemispheric metrics (Fig. 2D) that showed a significant maturational increase in interhemispheric PPC at 1–2 Hz as well as a decrease in the ratio of intra-/interhemispheric PPC at the same frequencies.
AAC Shows Wide, Frequency-dependent Shifts Between Vigilance States
AS and QS were associated with wide, significant, and frequency-dependent differences in AAC (Fig. 3A). At frequencies ≤2 Hz, most of the connections (up to 87% at 0.5 Hz) had significantly higher AAC in AS compared with QS. QS, in turn, was associated with a significantly increased AAC in most signal pairs in the higher frequencies (up to 86% at 8 Hz). Hemispheric metrics showed significant differences in both intra- and interhemispheric connections at the same frequency bands (Fig. 3B) suggesting that these effects are global. Comparison of intra- versus interhemispheric AAC between vigilance states suggested that a shift from AS to QS was related to a shift from intrahemispheric to interhemispheric AAC at the lowest frequencies (<1 Hz).
Early Postnatal Maturation of AAC is Wide Scale and State Specific
Correlation analysis between AAC and CA at the time of EEG recording showed that postnatal maturation was associated with significant decreases in AAC in large number of connections throughout the whole frequency range, with the most notable changes in the low (0.7–1 Hz) and higher (8–22 Hz) frequencies (Fig. 3C). Visual inspection of topographic plots of the significant signal pairs (Fig. 3C) suggested wide unselective distributions. A significant maturational change was seen during AS in only few edges at the lowest frequency examined (0.5 Hz). Hemispheric metrics showed that significant decreases in AAC coupling were found in both intra- and interhemispheric connections at a wide range of higher frequencies (>2 Hz; Fig. 3D). Low-frequency (0.7–1 Hz) intrahemispheric AAC coupling was also reduced with maturation. While the maturational decrease of higher frequency AAC coupling was global (Fig. 3C), there was a stronger relative decrease in the intrahemispheric AAC coupling at all lower (<2 Hz) frequencies (Fig. 3D).
PAC is Frequency-dependent and Strongly Related to Vigilance State
Nestedness of all higher frequencies was examined with respect to 2 lower frequencies, 0.7 Hz and 2 Hz (named PAC0.7 and PAC2, respectively). The lower of these frequencies (0.7 Hz) was selected to represent the slow wave component, which is known to nest higher activities in the neonatal EEG (Vanhatalo et al. 2005). Notably, EEG has been reported to exhibit nestedness across a wide range of frequencies in the adults as well (He et al. 2010), and we wanted to at least partially distinguish between SAT-related versus SAT-unrelated nestedness. We therefore selected a higher frequency band (2 Hz) for comparison to the SAT-related slow oscillations (Tolonen et al. 2007; Tokariev et al. 2012). We found that PAC0.7 was higher than PAC2 throughout the whole range of higher frequencies, and the finding was similar in both vigilance states (left side on Fig. 4A,B). Moreover, a vigilance state shift from AS to QS was related to a very wide scale, statistically significant increase in PAC0.7 with all high frequencies. Comparison of individual EEG signals showed that the vigilance state effect was significant in 53–84% of electrodes (left panel on Fig. 4C). In contrast, the change in PAC2 between AS and QS was only seen between a limited (5.7–16 Hz) frequency range (right side on Fig. 4A,B), and this was only seen in a small proportion (5–37%) of EEG signals (right panel on Fig. 4C). Taken together, these findings suggested that nestedness was much stronger within lowest frequencies and it was strongly and globally modulated by the vigilance state. To complement the summary Table 1, we also compared the first and the second recordings and found that PAC decreases with maturation at PAC decreased at higher (>10 Hz) frequencies in both vigilance states.
Developmental Changes in Oscillation Amplitudes
It was theoretically possible that maturational changes in the oscillation amplitudes could affect the development of functional connectivity described above. This may be even a significant source of error in PPC studies in the adults (Palva et al. 2005; Palva and Palva 2012) and it cannot be a priori ignored in the neonatal EEG that is known to change its spectral properties rapidly. We found that the amplitudes in QS were higher than in AS up to 11.3 Hz, whereas the amplitudes at frequencies above this up to 32 Hz were higher in AS (Fig. 1B). With respect to maturational changes (Fig. 1C) in AS, only one frequency (Fc = 11.3 Hz; r ≈ −0.3) showed a significant relationship, however amplitudes in QS were significantly increased with age at the midrange (Fc = 2–4 Hz; r ≈ 0.4) and decreased at the higher frequency (Fc = 22.6 Hz; r ≈ −0.3). These frequency profiles were not directly comparable to our PPC and ACC findings (below), suggesting that our connectivity findings are not explained by changes in amplitudes.
Our study shows that key measures of functional connectivity can be readily assessed from the EEG of the human newborn, and we provide a comprehensive frequency-dependent account of these connectivity metrics. Moreover, we show that functional brain connectivity is highly dynamic in the newborn infants so that it changes between vigilance states and matures rapidly over the couple weeks after normal term birth. Our work provides the first systematic insight to the electric brain networks during early postnatal period, the most critical few weeks window in the whole human life. The ability to study robust functional and developmental correlates of network states in the newborn opens unprecedented possibilities to advance understanding in newborn neurocognitive functions, as well as to examine the pathological network mechanisms underlying neurological adversities that are unfortunately common in this age group (Bonifacio et al. 2011).
Spatial Patterns of Connectivity
Our findings together show significant maturational and state-related changes in the neuronal coupling, however the spatial network topologies did not follow stationary constellations as often reported in prior functional magnetic resonance imaging studies (fMRI; Fransson et al. 2007; Smyser et al. 2011; Ball et al. 2014; Collin et al. 2014). Instead, our present findings are fully compatible with the idea established in recent adult studies that electrophysiological brain networks have rapid spatial and temporal dynamics (Betzel et al. 2012; Brookes et al. 2014). In neonates, some of the observed spatial variability between patients may arise from the high spatial complexity of the newborn scalp EEG signals (Odabaee et al. 2013; Odabaee et al. 2014), which leads to technical challenge in accurate neuroanatomical comparison between individuals. Such technical rather than physiological variability will readily bias statistical comparisons toward a more conservative end. Hence, our hereby reported large numbers of individual connections with significant changes related to either maturation or vigilance state speak for genuinely robust underlying effects in functional connectivity.
Connectivity Measures Reflect Different Modes of Early Cortical Operation
It is well established that the visually observed modes of spontaneous (or “background”) EEG activity are distinct between AS and QS (André et al. 2010). EEG during AS is characterized by continuous fluctuations at mixed frequencies, whereas during QS, EEG activity exhibits marked intermittency, (“trace alternant”; André et al. 2010) where SATs, that is, bursts of activity (Vanhatalo et al. 2005) erupt from the low-amplitude background. Recent studies merging knowledge from animal experiments and advanced human recordings have led to a unified framework (Vanhatalo and Kaila 2006; An et al. 2014) that consolidates early EEG development into evolving combination of 2 cortical mechanisms: First, there is ongoing spontaneous activity that is predominant in AS while it is also seen with lower amplitudes during periods between SATs in the QS (Vanhatalo et al. 2005). Conceivably, this ongoing activity accounts for much of the PPC measured in the newborns, especially during AS. Second, there is an immature type of brain network activity that is only seen from early prematurity to few weeks after term birth, or neonatal period (Vanhatalo et al. 2005; Seelke and Blumberg 2010; Myers et al. 2012). This is characterized by intermittent occurrence of complex multi-frequency events (please see Vanhatalo and Kaila 2010 for further discussion on the nomenclature relative to clinical literature). In the fullterm newborn, these events are the dominant mode during QS, however they are also seen to less extent during AS. Due to their wide band and high-energy content, they likely account for much if not most of the AAC coupling described herein, and moreover, they are the cortical network mechanism underlying the strong nestedness in the early EEG activity (Vanhatalo et al. 2005).
PPC Coupling Relates to Anatomical Maturation
Our PPC findings together showed large-scale reorganization of functional connectivity as the infant shifts between vigilance states. This observation is nicely compatible with the known early histological development of cortical networks. During AS, PPC is relatively stronger at lower frequency and in interhemispheric connections, while it becomes relatively stronger at higher frequency and intrahemispheric connections during QS. Prior histological studies of fetal human brain have shown that intrahemispheric cortico-cortical connections precede the growth of interhemispheric connections, and that precise networks required for higher frequency PPC grow over long period after establishing the overall wiring (Kostović and Jovanov-Milošević 2006; Kostović and Judaš 2010). These would predict, that PPC is first established at lower frequencies and shorter distances that may operate with less anatomical precision, hence the intrahemispheric connections would mature first. Indeed, we found such frequency dependence where PLV–IED relationship declined more rapidly towards higher frequencies (2B). The difference between vigilance states is hence compatible with an idea that, during QS brain is occupied by training the precise intrahemispheric connectivity (cf. Brockmann et al. 2011; Minlebaev et al. 2011; Kirkby et al. 2013; An et al. 2014), while AS is related to synchronizing larger neuronal ensembles between hemispheres. Notably, comparable relations, low-frequency PPC coupling in AS as opposed to high frequency PPC in QS, were also observed in AAC (see below).
AAC Coupling Declines Rapidly with Maturation
Earlier studies have shown that the incidence of the main cortical mechanism underlying AAC coupling, the SAT events, decreases from prematurity toward term age (Vanhatalo et al. 2005; Myers et al. 2012). Also density of spatial amplitude correlations between high-frequency activity are reported to decline during the last trimester of pregnancy, or from prematurity to term age (Omidvarnia et al. 2014). Our present findings show that the developmental decline in AAC is particularly rapid during the couple of weeks after term birth. The best proxy to AAC in the clinical EEG review is visual assessment of interhemispheric (Lombroso 1979; Räsänen et al. 2013; Koolen et al. 2014), seen as an apparent co-occurrence of SAT/burst events in 2 hemispheres during QS. interhemispheric increases during the last trimester (Meyerson 1968; Stenberg 1973; Lombroso 1979), however it does not change significantly across the immediate perinatal period studied in our present work (Koolen et al. 2014). Our present results extend this by showing how AAC coupling develops differentially between and within hemispheres. We report that maturation of the as yet unexplored intrahemispheric coupling is relatively stronger than the maturation between hemispheres.
Functional Connectivity via PPC and AAC may be Combined via Nestedness
On top of the generally perceived dichotomy between PPC and AAC coupling (Engel et al. 2013), our present results suggest that, at least in the early developing networks, these 2 coupling mechanisms may not be independent. We show that the transition from AS to QS is related to a significant wide spread increase in AAC at higher frequencies, as well as with a frequency-dependent increase in PPC at both the lowest and higher frequencies. This apparently complex observation finds an intriguing resolution in the occurrence of SATs, the multifrequency events that dominate QS: our findings are compatible with the idea that the lowest frequency PPC would relate to large-scale spatial correlation of SAT events (Omidvarnia et al. 2014). Then, spatial synchrony of SATs implies synchronized bouts of higher frequency oscillations, that is, AAC coupling, which would directly give rise to intermittent increase in PPC at these higher frequencies.
Developmental significance of this coupling mechanism is shown in experimental work where cortical connections are thought to develop with guidance of these intermittent events (Yang et al. 2009; Marcano-Reik et al. 2010; Brockmann et al. 2011; Minlebaev et al. 2011; An et al. 2014). They suggest that SATs, or their equivalents in the experimental models, would function as spatially coordinated training mechanism for faster oscillations during the early developmental period when first neuronal connections have grown and brain is transitioning from experience-independent to the experience-dependent phase in development (Blankenship and Feller 2010; Hanganu-Opatz 2010; Colonnese and Khazipov 2012). A direct experimental testing of this mechanistic hypothesis may be not possible in the human newborn, however several experimental animal studies suggest that early network training is genuinely focused on cortical mechanisms that are comparable to SATs in the human EEG (Blankenship and Feller 2010; Brockmann et al. 2011; Colonnese and Khazipov 2012; Kirkby et al. 2013; An et al. 2014).
Neonatal Changes in Functional Coupling Correlate with the Shift in Brain–Environment Relations
Finding rapid developmental increase of both AAC and PPC in intrahemispheric connectivity raises an intriguing view. The shift in functional networks is developmentally timed to the phase when the brain is probably shifting from endogenously guided brain activity in utero to a more environmentally influenced brain function after neonatal period. Prior work on human brain function has found it broadly challenging to examine developmental trajectories that span over the neonatal time (Stjerna et al. 2015), also called neonatal gap, because of the qualitative shift after neonatal period at multiple levels of brain functions, ranging from molecular mechanisms (Vanhatalo et al. 2005; Hanganu-Opatz 2010) to neuronal network behaviors (Vanhatalo and Kaila 2006; Colonnese et al. 2010; Kilb et al. 2011; Omidvarnia et al. 2014), as well as in neurobehavioral performance (Guzzetta et al. 2005). Neurodevelopmentally, this period is characterized by the shift from precritical period with experience-independent maturation to critical period with increasingly stronger experience-dependence (Feller and Scanziani 2005; Espinosa and Stryker 2012). Our observations together suggest that functional brain networks are rapidly adapted for this shift and life ex utero. However, our present work from healthy fullterm babies cannot distinguish whether the developmental change is a reaction to the onset of rich sensory experience ex utero more than a part of ontogenetic scheduling. These 2 effects could be distinguished by studying the same developmental period around term age in very preterm born babies whose postnatal adaptation has passed several months earlier.
A.T., M.V., J.M.P., and S.V. designed the research and wrote the paper. A.T. and M.V. analyzed data.
This work was supported by Academy of Finland (276523, 288220), Sigrid Jusélius Foundation, Finnish Cultural Foundation (00140992), Päivikki ja Sakari Sohlberg Foundation as well as by the Helsinki University Hospital .
We want to greatly acknowledge Ms Susanna Stjerna for the highly skilled newborn EEG recordings, as well as the parents for allowing us to study their infants. Conflict of Interest: None declared.