Background
Autism spectrum disorder (ASD) is primarily characterized by impairments in social communication skills and the presence of repetitive and stereotypical behaviors [
1]. Recent research shows that many genes implicated in ASD are involved in formation and regulation of synaptic pathways and neural connections [
2,
3]. Given that the pruning and modification of such synapses and neural pathways are partly shaped by experience [
4,
5], studies of brain connectivity in very young children with ASD, or in infants ‘at risk’ for developing ASD, are critical.
Structural and functional MRI studies in adults with ASD have predominantly reported weaker long-range cortico-cortical connections [
6,
7]. The studies that used electro- (EEG) and magneto-encephalography (MEG) also primarily found reduced functional connectivity, although the results were more variable (see Additional file
1). In contrast, more recent fMRI studies in younger (pre-pubertal) children with ASD have reported
increased functional connectivity in brain networks [
8‐
10]. Further, functional hyper-connectivity associates with greater severity of autism symptoms in children [
8]. In line with the increased functional connectivity findings, recent diffusion tensor imaging studies revealed that young children and toddlers with ASD have atypically early maturation of white matter [
11‐
14].
The presence of different patterns of connectivity abnormalities in children and adults with ASD may reflect an atypical trajectory of brain development. The early overgrowth of white and gray matter during infancy and toddlerhood in individuals with ASD is followed by normal or decreased growth during later childhood [
15‐
18]. This pattern of atypical development seems to be a particular feature of the temporal, frontal, and cingulate cortices that play crucial roles in attention, emotions, and social cognition [
15,
18‐
21]. It is possible that the early increases in fMRI connectivity reported in younger children with ASD reflect their early brain overgrowth. On the other hand, the under-connectivity reported in adolescences and adults with ASD may be linked to the reduced brain growth during later childhood and/or to some compensatory processes.
Siblings of children with ASD have an increased risk of developing the disorder [
22]. Therefore, studies in infant siblings of children with ASD allow one to study the early signs of the disorder before reliable diagnosis is obtained in later childhood. A recent longitudinal study found atypical developmental trajectories of cortical fiber tracts in at-risk infants who displayed ASD features at 2 years of life [
14]. Specifically, development of most fiber tracts studied in the infants with later ASD was characterized by higher fractional anisotropy (FA) values at 6 months, followed by slower change over time relative to high-risk infants with few autistic traits assessed at 24 months. Thus, by 24 months of age, those with emerging ASD had lower FA values.
Structural and functional connectivity are not unequivocally associated [
23]. It is therefore unclear if, and how, early differences in structural connectivity might affect emerging functional connections to result in later diagnosed ASD or, indeed, how early differences in functional connectivity might sculpt later appearing differences in structural connectivity. While in adults functional connectivity has been mainly investigated with fMRI, this method has practical limitations in infants and young children. Even in the youngest children, however, functional connectivity can be measured noninvasively with EEG. While motion inside the MRI scanner may create artifacts presenting a serious problem for analysis of connectivity [
24], EEG connectivity is unlikely to be affected by motion in a similar adverse way because the EEG electrodes are in fixed locations on the scalp. In addition, the high time resolution of the method allows one to measure phase coupling of the oscillatory EEG signals indicating synchronization of neuronal populations on a millisecond timescale [
25].
Although EEG has been widely used to investigate possible neural markers of ASD in adults and children, only two studies have investigated ‘ongoing’ EEG in infants at risk for later autism [
26,
27], and neither of these measured connectivity. Tierney and colleagues [
27] and Bosl and colleagues [
26] describe differences in quantitative EEG features between infants at high and low familial risk for ASD (i.e., infants who had an older sibling with ASD), but the actual outcome of infants ‘at risk’ was not known at the time of investigation.
In the present study, we used high-density EEG to examine functional connectivity in a cohort of 14-month-old infants at low or high familial risk for ASD and then assessed symptoms of ASD at age 3 years. Different frequencies of EEG and MEG oscillations are associated with different cognitive, emotional, and motor processes [
25,
28,
29] and may reflect activity in different brain networks. Further, atypical EEG connectivity in neuropsychiatric disorders may be specific to particular frequency bands [
30,
31]. In this study, we recorded EEG while infants attended to a series of videos and then analyzed connectivity within the infant alpha band. There were several reasons for focusing on the alpha frequency band. First, the alpha rhythm is intimately related to attention processes in adults [
28,
32] and infants [
33] and alpha-range connectivity increases during states of attentiveness [
32,
34]. Therefore, we expected the presence of reliable alpha-range connectivity during sustained attention to the videos in our study. Second, alpha activity is to a lesser extent than theta activity modulated by inter-individual differences in emotional and cognitive engagement [
35,
36]. Therefore, we expected that the inter-individual variations in these uncontrolled factors during passive viewing of the videos would contribute less to the alpha then into the theta frequency band. Third, the presence of alpha peak in the infant EEG power spectra suggests the presence of distinct rhythmic processes at this frequency and, correspondingly, good signal-to-noise ratio. Fourth, alpha activity is less prone to contamination by muscle artifacts than the EEG signal of higher frequencies [
37,
38] and is less prone to contamination by movement artifacts than the delta activity. Considering all these factors, we hypothesized that in infants any altered coupling between cortical regions during attention to videos would be most reliably detected in this frequency band.
The magnitude of alpha-range connectivity can be modulated by a participant’s functional state and behavior [
34,
39,
40]. In order to assess the effect of ongoing behavioral state on EEG connectivity, and to control for possible outcome group differences in behavior, we complemented the EEG analysis with an analysis of infants’ behavior during the EEG recording session.
Siegel et al. [
25] point to several important challenges for M/EEG studies of functional brain connectivity. First, due to volume conduction, the electrical potentials generated by neuronal activity are not only measured in the direct vicinity of neuronal sources but can also be measured at distant sites, substantially limiting the interpretation of between-channel synchronization measured by, e.g., coherence of phase locking. The authors suggested that one effective strategy for addressing this confounding factor is to confine the analysis to noninstantaneous correlations, e.g., by looking at the phase-lagged part of coherence. In the present study, we analyzed phase-lagged connectivity using a recent method characterized by improved sensitivity [
41]. Second, Siegel et al. note that differences in measured correlations can be driven by differences in the signal-to-noise ratio (SNR). Even if the true correlation between two brain sources does not differ between experimental groups, differences in the signal amplitude alone lead to differences in the SNR and thus to differences in the connectivity measures. To exclude this confounding factor, we controlled for differences in signal amplitude between the groups. Third, a large number of interactions between regions/electrodes imposes the problem of multiple comparisons. To circumvent this problem, we applied the method for statistical analysis on large networks—network-based statistics (NBS) [
42].
Discussion
We observed increased phase-lagged alpha EEG connectivity in a group of 14-month-old infants who were later diagnosed with ASD at 3 years of age. The elevated connectivity was present despite the lack of measurable differences in behavior during the EEG data collection or differences in EEG spectral power. Further, in participants with ASD, this measure at 14 months correlated with the severity of restricted and repetitive behaviors at 3 years.
Studies in adults show a correlation between alpha EEG coherence and the structural integrity of white matter [
50]. The high alpha-range connectivity in HR-ASD infants in our study may reflect the early maturation of white matter tracts previously reported in toddlers and young children with ASD [
11‐
14]. The increases in white matter in ASD appear to reduce in toddlerhood [
14] or later childhood [
51,
52], finally resulting in predominantly hypo-connectivity between cortical areas in adults with ASD [
7]. These age-related differences in white matter integrity may potentially explain the fact that the alpha-range hyper-connectivity detected in infants who later go on to have ASD in our study is later followed by alpha-range hypo-connectivity in adolescence and adulthood [
31,
39,
53‐
56].
It is also possible that the atypical connectivity observed in ASD is a sequela of an altered excitation/inhibition (E/I) ratio [
57]. The number of parvalbumin-expressing inhibitory interneurons [
58], as well as GABA receptors and the receptors’ benzodiazepine binding sites [
59], is reduced in autism, suggesting reduced cortical inhibition. In typical individuals, the glutamate/GABA ratio correlates positively with functional connectivity in the default mode network [
60]. It is therefore possible that the elevated connectivity that we observed in HR-ASD infants results from their elevated E/I ratio. The later decrease in connectivity in adolescents and adults with ASD may then reflect a compensatory reaction to reduce cortical excitability. The future studies may help to investigate the putative link between abnormal neural connectivity and elevated E/I ratio in ASD by looking at connectivity measures in relation to ongoing and event-related gamma oscillations [
61,
62].
The increased alpha-range connectivity in HR-ASD infants was widespread across the scalp (Figures
2A and
3), a finding that is in accord with DTI evidence on the presence of atypical development trajectories for white matter in all of the major tracts studied in toddlers with high autism traits [
14] and with recent fMRI findings in children with ASD [
8]. Nevertheless, clustering of significant dbWPLI differences over the anterior, and particularly left frontal (Figure
2B, C, see also Additional file
2: Figure S8), regions implies some degree of regional specificity in the atypical functional connectivity and suggests that the left frontal cortex may be particularly strongly over-connected in some infants who later go on to an ASD diagnosis. Prominent developmental atypicalities have previously been reported in the frontal lobes of children and adults with ASD [
15,
16,
63]. Interestingly, Bashat and colleagues [
11,
13] found abnormally mature white matter predominantly in the left frontal lobe in preschool children with ASD.
Alpha-range connectivity in infants who went on to ASD correlated with their later severity of RRB (Figure
4,
lower panel). This finding suggests that alpha hyper-connectivity may reflect an important aspect of atypical brain function in many cases of emerging ASD. The link between hyper-connectivity and severity of RRB has previously been reported in resting-state fMRI studies. Uddin et al. [
9] observed that hyper-connectivity in the ‘salience network’ predicted greater incidence of RRB in 10-year-old children with ASD. Delmonte et al. [
64] observed a link between the severity of RRB and elevated fronto-striatal connectivity in adolescents with ASD. Interestingly, correlations between structural changes in the striatum and repetitive behaviors have been previously reported [
65,
66] and may be secondary to changes in frontal areas or fronto-striatal connectivity [
65]. In adults with ASD, the higher functional connectivity between left dorsal ACC and the frontal eye field was associated with more severe RRB, even though connectivity between these structures was reduced in the ASD participants relative to neurotypical controls [
67]. A recent morphological study of short-range intrinsic cortico-cortical connections also reports an association between frontal connectivity and the ADI-R RRB score in adults with ASD [
68]. Specifically, the decreased ‘wiring cost’ (and thus potentially increased cortico-cortical connectivity) in several frontal regions in participants with ASD correlated with their tendency to engage in repetitive behaviors. Although the sources of the elevated alpha connectivity in our study cannot be localized precisely based exclusively on surface EEG, our results accord well with the previous studies implicating frontal hyper-connectivity in repetitive behaviors in ASD.
It should be noted that
reduced connectivity in young children with ASD has also been previously reported [
69]. Using fMRI, Dinstein et al. [
69] found that toddlers with ASD had reduced synchronization between corresponding ‘language areas’ of the two hemispheres during sleep. Moreover, the reduced functional connectivity correlated with poorer language abilities and with greater social and communication deficits. Further studies are therefore needed to investigate in ASD possible differences in functional connectivity measured at faster (EEG phase lag) and slower (fMRI) timescales, as well as modulations of connectivity by participant’s age and functional state.
It has been suggested that in adults with ASD, local hyper-connectivity in the frontal cortex occurs together with long-range hypo-connectivity [
7,
16]. We have found elevated connectivity in infants with later ASD irrespective of between-electrode distance (Figure
3). Notably, the phase-lagged connectivity assessed in the present study may underestimate short-range connections as they are more likely than the long-range connections to occur without measurable phase lag. Other methods, such as, for example, alpha to gamma phase-amplitude coupling, may provide better estimates of short-range connectivity in ASD [
54].
Although no EEG connectivity studies have been done so far in infants with later diagnosis of ASD, such studies have been previously performed in diagnosed children and adults. The studies in adults and adolescents with ASD mainly reported decreased MEG and EEG connectivity in the alpha frequency range [
31,
39,
53‐
56], although negative findings have also been reported [
70,
71]. Studies in children produced more variable results, with findings of decreased [
72,
73], not changed [
49,
74], both decreased and increased [
75], or increased [
76] alpha-range connectivity in ASD. Apart from differences in experimental samples, there are other factors that might potentially contribute to the findings. Among these are the choice of frequency band and reference electrode for EEG recording (e.g., Common Mode Sense-Driven Right Leg reference [
74], linked earlobes [
73,
76], Laplacian [
75], or average reference [
72]), presence of group differences in alpha power (and SNR) [
72,
76] or lack of information on such differences [
49,
74,
75], and differences in experimental paradigms, such as undefined state [
49], rest with eyes closed [
73], photo-driving [
76], or visual stimulation with long (seconds) intervals [
72,
74]. Some of these factors could influence group differences in alpha-range connectivity in our study. Moreover, the majority of the previous studies in children assessed functional connectivity using coherence—the measure that is sensitive to volume conduction. The application of the phase-lagged connectivity in our study should favor detection of the ‘true’ functional connectivity differences that could otherwise have been altered or canceled out by volume conduction effects that might be stronger in individuals possessing greater alpha power.
Although EEG and MEG have been previously used to study functional connectivity in ASD (see Additional file
1 for review), only one study, similarly to our present study, analyzed phase-lagged connectivity [
74]. Boersma and colleagues recorded EEG in children (mean age 3 years) during presentation of visual stimuli but did not find group differences in low-frequency (4–10 Hz) connectivity. One possibility is that the elevated alpha range connectivity in ASD is more prominent during infancy and then decreases to the third year of life. Another possibility is that connectivity differences can still be detected in 3-year-olds with ASD but in narrower age-appropriate frequency bands and/or with a greater temporal quantity of data.
Figure
2C suggests that not all participants who were later diagnosed with ASD in our study had increased alpha-range connectivity as infants. Alterations in functional brain connectivity in ASD may depend on the presence of a risk allele of a particular ‘connectivity-crucial’ gene [
77,
78] or on a combination of genes unique to an individual. Possibly, the atypical connectivity patterns characterize only a subset of the HR-ASD infants with such genetic predisposition. Our findings indicate that hyper-connectivity during infancy may be a feature of participants who later go on to show higher levels of restrictive and repetitive behavior (Figure
4). Considering the small group size of the HR-ASD infants in our study, further studies with larger samples are needed to elucidate behavioral features that may differentiate children with and without early alpha-range hyper-connectivity.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
ME, GD, TC, and MJ designed and conceived the experiment. ME and the BASIS Team collected the data. EO and the BASIS Team cleaned and analyzed the data. EO, EJ, GD, TC, and MJ contributed to the interpretation and writing. TC and the BASIS Team contributed to the clinical assessment and cohort characterization. All authors read and approved the final manuscript.