The study group comprised 43 6-year-old children of whom 22 were born EPT (gestational age <28 completed weeks) and 21 were born FT (gestational age ≥37 completed weeks). The children were originally recruited for a larger multi-methodological follow-up study (Rahkonen et al., 2013, 2014). The EPT children were born between 2006 and 2008 and treated after birth at the neonatal intensive care unit of the Helsinki University Hospital, Finland. The FT children, including 10 new controls recruited at 6 years of age, were born healthy in the Hospital District of Helsinki and Uusimaa between 2006 and 2009. For this study, we only included EPT children without major brain abnormalities, i.e., grade III–IV intraventricular hemorrhage in cranial ultrasound in the neonatal period or moderate-to-severe white matter abnormality (Woodward et al., 2006) in brain magnetic resonance imaging (MRI) at term age. The included FT children had no known brain abnormalities or neurological disorders.

Altogether, 25 EPT children and 25 FT children underwent the MEG recording. We subsequently excluded three EPT children and four FT children from the present analysis due to an incomplete measurement session (one EPT child and one FT child), extensive movement (two EPT children), technical problems (two FT children), or epileptiform abnormalities in MEG (one FT child). Table 1 presents details of the clinical characteristics of the 43 participating children included in the final analysis.

The Ethics Committee for gynaecology and obstetrics, pediatrics and psychiatry of the Hospital District of Helsinki and Uusimaa approved the study protocol. All participants and their parents signed an informed consent in accordance with the Declaration of Helsinki.

Obstetric and neonatal data were collected from hospital records and a parental questionnaire. Gestational age was determined from the first-trimester ultrasound when available. Birth weight z-scores for gestational age, sex, plurality, and parity were based on the Finnish growth reference data (Sankilampi et al., 2013). General cognitive ability (full scale intelligence quotient) at 6.5 ± 0.1 (mean ± SD) years of age was assessed using five subtests of the Finnish edition of the Wechsler Preschool and Primary Scale of Intelligence – Third Edition, WPPSI-III, (Wechsler et al., 2009). Handedness was determined according to the writing hand. Maternal education was obtained by a parental questionnaire. Educational attainment was classified from low (elementary school) to high (university).

Magnetoencephalography was recorded in the BioMag Laboratory in a magnetically shielded room (Euroshield, Finland) with a helmet-shaped sensor array consisting of 306 independent channels: 204 gradiometers and 102 magnetometers [Elekta Neuromag Vectorview (n = 34) or TRIUX (n = 9, all FT children), Elekta Oy, Helsinki, Finland]. Only gradiometer data were used in the analysis. The sampling rate was 1004 Hz (TRIUX: 1000 Hz) and the measuring band pass from 0.03 to 251 Hz (TRIUX: 0.03 – 330 Hz). An individual Cartesian coordinate system was constructed for each participant by digitizing the preauricular points and nasion before the recordings. Four (Vectorview) or five (TRIUX) position indicator coils were attached to the head and their positions relative to the anatomical landmarks were digitized. In the beginning of each recording block, the coils were fed with currents to determine the head position relative to the sensors. Continuous head position measurement was applied during the measurement. The activity of the thenar muscles and wrist/finger flexors of the right hand and forearm, respectively, were measured with two bipolar electromyography (EMG) traces.

During the measurement, the child lay supine with his/her head in the measurement helmet. Since the head of a 6-year-old child is small for the adult-size helmet, the head was supported from both sides with thin soft cushions to prevent head movements. During all recording blocks (except for the REST condition with eyes closed), the child was encouraged to look at a colorful picture on the ceiling of the shielded room above his/her head in order to prevent excessive eye-movements. One of the researchers was always in the measurement room with the child to give instructions and ensure optimal behavior. One parent was also allowed to join in.

Tactile stimulation was applied to the volar side of the distal phalanges of the index and little finger of the left hand. The stimulus was a gentle tap on the skin, delivered by a plastic membrane that expanded with delivered air puffs (Somatosensory Stimulus Generator, 4-D NeuroImaging Inc., San Diego, CA, USA). The stimulation was given in a semi-random predefined sequence (to the index finger with 60% probability and to the little finger with 40% probability) with a 2-s inter-stimulus interval between consecutive stimuli. One block lasted approximately 9 min with at least 250 stimuli.

The head movement compensation, i.e., the data conversion to the initial head position, as well as the suppression of external magnetic interference from outside the brain, were calculated from the raw MEG data with the spatiotemporal signal space separation method (Taulu and Simola, 2006) in the MaxFilter^® software (Elekta Neuromag^®; Elekta Oy, Helsinki, Finland) using a correlation limit of 0.98 and a 16-s time window. Then, individual spectral content of the data during eyes open and eyes closed was calculated for each subject from the resting data. Furthermore, the child’s head position in the MEG helmet was visualized with the BrainStorm software (Tadel et al., 2011) using an MRI template for 6.5-year-olds (Evans, 2006; Sanchez et al., 2012; Richards and Xie, 2015; Richards et al., 2016). In addition, we calculated the amount of head movements (average translational and rotational speed) in the NO-TASK condition. There were no significant differences in these parameters between the EPT and FT children.

For the TASK condition, EMG traces were inspected visually to identify epochs with correct responses, i.e., squeezing for the little finger stimulation and not squeezing for the index finger stimulation. Only the correct epochs were used in further analyses.

The reactivity of the sensorimotor oscillations in alpha (7–12 Hz) and beta (15–25 Hz) bands was analyzed using the temporal-spectral evolution method (Salmelin and Hari, 1994), where the data are first band-pass filtered, rectified, and finally averaged, time-locked to the tactile stimulation. This demonstrates event-related changes in the average amplitude level of oscillatory activity in the given frequency bands in the same units as the original data. Before calculating the amplitude envelopes, the somatosensory evoked fields (to finger stimulation) were subtracted from the raw data and artifacts from cardiac activity removed by the signal space projection method (Taulu and Simola, 2006). Then, the amplitude envelopes were calculated from all gradiometer channels for each subject in the two frequency ranges, independently for both index and little finger stimulation and for both TASK and NO-TASK conditions (all steps performed using the Graph software, Elekta Neuromag^®; Elekta Oy, Helsinki, Finland). In the TASK condition, only epochs with correct behavioral responses were included in the averages. Finally, two gradiometer channels in each hemisphere showing the clearest modulation of the oscillation level were averaged and these averages, one in each hemisphere, were used for amplitude comparisons. Averaging two channels was done to add objectivity to the choice, as e.g., suppression and rebound might be maximal in neighboring channels. The amplitude envelopes were then low-pass filtered at 10 Hz. After visual inspection (Figure 1), three time windows were chosen for the main analysis: from 100 to 600 ms post-stimulus including the main suppression, from 600 to 1200 ms, and from 1200 to 1700 ms. The mean oscillation amplitude during each of these three time-windows was calculated for each subject and each condition. Relative amplitudes for each time-window were then calculated as percentages of the 300-ms prestimulus baseline. Statistical analysis was performed on these relative amplitudes. In addition, the modulation of occipital alpha-band oscillations was calculated (average of eight occipital channels) from the same three time-windows from epochs of the index-finger stimulation in the TASK and NO-TASK conditions.

The equivalent current dipoles in the primary somatosensory cortex (SI) for the first prominent activity peak after the index finger stimulation in the NO-TASK condition were calculated using a spherical head model with the source modeling program (Elekta Neuromag^®; Elekta Oy, Helsinki, Finland). Source-level oscillations and their modulations for this SI dipole were then calculated using the Graph program (Elekta Neuromag^®; Elekta Oy, Helsinki, Finland).

Statistical analyses were performed using the IBM SPSS Statistics software v. 22 (SPSS Inc, Chicago, IL, USA). Baseline characteristics between the study groups were compared with Student’s T-test, Fisher’s Exact Test or Chi-square Test according to the variable type and expected frequency within cells. Behavioral results in the Go/NoGo task between and within groups were compared with Mann–Whitney U-test and Wilcoxon signed rank test. The correlation between these behavioral results and the intelligence quotient was performed with Spearman’s rho. MEG results were compared with a repeated measures ANOVA (refer to the results section for details of each comparison). Greenhouse-Geisser corrections were used when needed. P-values less than 0.05 were considered statistically significant. Post hoc comparisons (performed with Student’s T-test) were corrected for multiple comparisons with Bonferroni correction when several comparisons were carried out from the same data, e.g., when oscillation levels after index and little finger stimulation were compared in all time-windows separately. In these cases, the original p-value was multiplied by the number of comparisons. If this corrected p-value was less than 0.05, the difference was considered statistically significant. For the source-level analysis, Student’s T-test or, in case of non-normal distribution of the data, Mann–Whitney U-Test was used.

The EPT children performed worse than the FT children in the TASK condition requiring a squeeze for the little-finger (target) stimulation and no squeeze for the index finger (non-target) stimulation (Table 2). In addition, it was more difficult for the EPT children to refrain from squeezing for the non-target finger stimulation than to respond with a squeeze for the target finger stimulation. The FT group had a similar tendency, but the difference between correct response rates for the target and non-target finger was not significant (Table 2). Intelligence quotient had no significant correlation with the rate of correct behavioral responses (Spearman’s correlation, p = 0.06 for index finger and p = 0.14 for little finger).

Tactile stimulation of the fingers induced, expectedly, a prominent suppression of the sensorimotor oscillations on the contralateral-to-stimulation hemisphere within the first time-window. This suppression was, however, significantly weaker in the EPT than FT children, which may reflect reduced ability to engage the stimulus-relevant brain networks into active information processing. In a recent experiment, a group of autistic children showed reduced posterior alpha-band suppression in comparison with typically developing children while pantomiming tool use (Ewen et al., 2016). Decreased modulation of alpha oscillation was interpreted to indicate decreased task-related activation.

In adults, the post-stimulus suppression is followed by a rebound of the oscillations over the baseline levels. In the present data set, some individual traces showed such a rebound, but it did not reach significance in the group-averaged data either in FT or EPT children, alpha or beta band. A previous study (Gaetz et al., 2010) suggests that at least post-movement beta rebound, thought to reflect motor inhibition, gradually develops with age, and 4–6-year-olds still show significantly reduced post-movement beta rebound in comparison to older groups. It is thus possible that the stimulus-induced rebound also develops with age and was therefore not well seen in our 6-year-old subjects.

In the TASK condition, the EPT children responded less often to the target stimulation, as well as more often to the non-target stimulation than the FT children. In addition, their performance was even less precise for the non-target than target finger, indicating that it was more difficult for them to refrain from responding to the non-target index finger stimulation than to respond to the target little finger stimulation. These results are in line with those of others who have studied response inhibition in preterm children with a Go/NoGo task (Mulder et al., 2011; Orchinik et al., 2011; Baron et al., 2012; Loe et al., 2015).

The poorer performance of the EPT children in the NoGo TASK (requiring response inhibition) was reflected in the modulation of sensorimotor brain oscillation levels. In FT children, in the left hemisphere—contralateral to the hand squeezing the toy—the relative sensorimotor alpha oscillation level was higher in the NoGo TASK compared with the NO-TASK condition. In other words, the oscillation level was enhanced when the FT children were actively refraining from squeezing, in contrast to when they were not paying attention to the stimulation, even though the behavioral response—no squeeze—was the same in both conditions. The EPT children, interestingly, showed no such difference between the NoGo TASK and NO-TASK conditions. We suggest that the enhanced sensorimotor oscillations in the left hemisphere when refraining from squeezing (i.e., NoGo) reflect active motor response inhibition. The lack of this effect in EPT children seems to therefore reflect altered brain mechanisms employed to actively inhibit a motor response.

In the Go TASK condition, the toy-squeeze induced a remarkable suppression on the contralateral-to-squeeze hemisphere in both EPT and FT children with a peak during the middle time-window. The peak latency of this long-lasting suppression is evidently a consequence of the reaction time after the stimulation was felt, as the squeeze started around 325 ms after the stimulation. Although, by visual inspection, the squeeze-induced suppression in the left hemisphere seemed stronger in the FT children, the suppression was also significantly present in the EPT children. As we could not adequately quantify the motor activity of the squeeze, we did not compare the suppression in the Go TASK condition between FT and EPT children, since group differences in the squeeze itself could have as well explained differences in the oscillation levels.

In a previous somatosensory Go/NoGo study in adults, a NoGo condition induced a prominent increase in alpha (and beta) level in frontal and central areas (pooled data from several sensors over both hemispheres; Nakata et al., 2013). In our study, even in the FT children the oscillation level was elevated only slightly, and merely on the contralateral-to-squeeze hemisphere. This alpha-increase in the NoGo TASK condition was seen by visual evaluation in individual subjects mostly on channels contralateral to the omitted squeeze over the central or fronto-central areas. One explanation for the more local/moderate elevation levels in our young subjects might be that, as response inhibition in a Go/NoGo task is known to develop with age (Levin et al., 1991; Brocki and Bohlin, 2004), its cortical expression reflected in elevated oscillation levels may simply not yet be fully developed in 6-year-old children. As with the EPT children, they did not show this elevated alpha level at all. However, two cross-sectional research reports suggested that, in response inhibition task performance, very preterm children approaching adolescence may catch up with their FT peers (Aarnoudse-Moens et al., 2012; Ritter et al., 2013). Whether the lack of sensorimotor oscillation enhancement during response inhibition in our group of 6-year-old EPT children would be a delay rather than a deficit remains a prospect for future research.

Neural substrates of response inhibition have been widely studied with fMRI. These studies have revealed a wide cortical network underlying response inhibition (for reviews, see e.g., Simmonds et al., 2008; Swick et al., 2011; Criaud and Boulinguez, 2013). This wide activation may, however, be at least partly explained by non-specific paradigms also activating several other cognitive processes that are intrinsically related to response inhibition and difficult to disentangle, including, e.g., attention, working memory, and response selection (Criaud and Boulinguez, 2013). FMRI studies of response inhibition in preterm-born adolescents/adults have shown differences in BOLD responses in several brain areas (but not the sensorimotor areas) in the absence of behavioral differences (Nosarti et al., 2006; Lawrence et al., 2009). The brain activation differences were suggested to represent altered neuronal pathways for response inhibition in the preterm-born group. In another study, preterm children (6–7 years) with intrauterine growth restriction exhibited higher variability in executing or inhibiting a response than preterm-born children without growth restriction and also ‘decreased reliance on brain regions classically involved in successful response inhibition’ (Réveillon et al., 2013).

The lack of activation/deactivation of sensorimotor cortices in fMRI studies of response inhibition is striking. Nakata et al. (2008) reported activation of some sensorimotor areas in addition to several other brain areas in NoGo trials, requiring inhibition of movement after electric stimulation of hand nerves. However, the activation was not contrasted with the same stimulation in a no-task condition, so it remains unknown if it was enhanced during response inhibition. More time-sensitive methods, however, do show changes in sensorimotor-cortex activity. In an MEG study on response inhibition, the NoGo stimulus started to activate the motor cortex 100–120 ms after the stimulus onset, just like the Go-stimulus, with a consequent decay of activity before 200 ms, indicating initial preparation for and subsequent inhibition of the movement (Endo et al., 1999). In a transcranial magnetic stimulation (TMS) study, motor evoked potentials (MEPs) were suppressed during a NoGo condition, reflecting inhibition in the pyramidal tract (Hoshiyama et al., 1996). Furthermore, behavioral and electrophysiological (decreased MEP amplitudes) response inhibition was shown to be paralleled by enhanced oscillatory alpha activity over cortical sensorimotor areas (Hummel et al., 2002). It is thus plausible that the elevated oscillation levels in our study reflect inhibition of motor output.

Two separate spectral peaks in the alpha range were detected in spectral analyses in all conditions, one over the occipital and the other over sensorimotor areas. In both EPT and FT children, the peak frequency (in the REST condition) of the sensorimotor alpha (∼9.3 Hz) was slightly but significantly higher than the occipital alpha (∼8.4 Hz). This is in agreement with the slightly higher sensorimotor than occipital alpha frequencies reported in adults (Storm van Leeuwen et al., 1978). In our study, the alpha frequencies in the sensorimotor or occipital areas did not differ between FT and EPT children. In other brain areas, differences have been shown by Doesburg et al. (2011): compared with full-term children, preterm born children (≤32 week of gestation) had lower peak alpha frequencies in frontal areas and lower alpha power in bilateral frontal and temporal areas.

In the present study, the level of the occipital alpha was significantly enhanced during the first time-window in the NoGo TASK condition, while attending to the somatosensory stimuli, compared with the occipital alpha level with the same index-finger stimulation in the NO-TASK condition when the stimuli were not attended. Several studies suggest that both in visual and somatosensory modalities, cue-induced alpha modulation might serve as a gating mechanism for attention. Alpha band oscillations have been suggested to reflect frontally controlled disengagement of task-irrelevant regions (Haegens et al., 2010). For example, cued visual spatial attention leads to increased amplitude of occipital oscillatory alpha activity in the hemisphere contralateral to the unattended visual field (Worden et al., 2000). The same holds for the somatosensory area: attention increased the power of sensorimotor alpha and beta oscillations in the hand area of SI when attention was directed to the foot, and decreased the power when attention was cued to the hand (Jones et al., 2010). This reciprocal effect is also intermodal: visual processing increases the sensorimotor alpha and decreases the visual alpha, and movement or somatosensory attention decreases the sensorimotor alpha and increases the visual alpha (Pfurtscheller, 1992; Anderson and Ding, 2011). A similar intermodal effect was seen in the early time-window in our study – when the children were attending to the somatosensory stimuli, the visual-alpha levels were enhanced.

The relative sensorimotor oscillation level associated with the response inhibition in the FT group was higher in the TASK NoGo compared with the same finger stimulation in the NO-TASK condition. Also, the relative occipital alpha levels in the occipital channels were higher in the TASK NoGo compared with the NO-TASK condition. The natural doubt arises whether this response inhibition effect is a ‘leakage’ of the occipital alpha to the sensorimotor areas. The occipital alpha level was, however, significantly higher in the TASK than NO-TASK condition only during the first time-window, whereas the sensorimotor alpha level continued to be higher in the TASK than NO-TASK during the middle and last time-windows. Thus, the occipital alpha cannot explain the enhancement of the oscillation level in the sensorimotor channels, and the response inhibition effect seen as enhanced sensorimotor alpha levels during the TASK NoGo vs. NO-TASK condition is, therefore, genuine.

One of the methodological limitations of the study is that the main analysis was carried out on the sensor level. The sensor level analysis should not have affected most of our results, as they were gained from comparisons that were based on relative changes within the study groups. Results of between-group comparisons could have been affected by, e.g., different head shapes and consequently different distances of relevant cortex from the sensors, resulting in smaller amplitude of the measured oscillation. For between-groups comparisons, we therefore calculated the prestimulus oscillation level and relative suppression from the SI dipole source waveform. As the equivalent dipole source calculation takes into account the head position and distance of the measured activity from the sensors, we obtained oscillation values from the SI dipoles independent from the head size/form. The post-stimulus suppression, measured from the SI source waveform data, was indeed weaker in the EPT than the FT children in these comparisons.

In addition, the last nine FT children were recorded with an upgraded MEG-device (TRIUX), whereas all EPT children were recorded with the previous device (Vectorview). We do not believe the change in the device had significant impact on our results, however, as a preliminary analysis including only the FT children (n = 14) measured with the Vectorview-device yielded the same main results (Pihko et al., 2015).

In this study, we excluded children with major abnormalities in brain imaging because these lesions might have affected the MEG results. Subsequently, none of the EPT children in this study had cerebral palsy or severe cognitive impairment. Excluding the most severely affected patients from our study resulted presumably in better behavioral performance and possibly less altered MEG results among the EPT children.

The number of subjects—22 EPT children and 21 FT children—was limited, but sufficient to detect at least the most robust differences. Since the relatively small sample size of the subject groups, we used the most conservative correction for multiple comparisons (Bonferroni) to bring out merely the most clinically relevant results. Confidence intervals further support these results and our conclusions.