Introduction

Autism spectrum disorders (ASD) are characterized by social and communication impairments, and by restricted and stereotyped behaviors[1]. ASD affect approximately 1 in 88 children and 1 in 54 males[2]. These disorders are highly heritable, with estimates ranging from 70 to 90%[3], and are known to have a high recurrence rate in siblings (10 to 20%[4]), yet, progress in identifying pathophysiological and etiological mechanisms has been limited.

It is likely that there are many causes of ASD. Several single-gene disorders (for example, Fragile X, tuberous sclerosis) and rare copy number variants (CNVs) (for example, 16p11 deletions, 15q13 duplications) appear to be strongly associated with ASD, but genetic syndromes, mutations, and single-gene etiologies account for only 10 to 20% of ASD cases, and many individuals with these genetic syndromes do not have ASD[5]. The majority of affected individuals appear to have more complex underlying genetic and epigenetic abnormalities, involving highly penetrant yet undiscovered rare mutations, or combinations of less penetrant and more common variants. The highly variable clinical presentation of ASD reflects this heterogeneity. Affected individuals vary greatly in the course of the disorder (around a quarter show significant developmental regression), associated medical conditions, behavioral challenges (for example, sensory issues, hyperactivity), and degree of intellectual impairment[6, 7].

Many studies have attempted to characterize the neural system abnormalities associated with ASD[8, 9]. Post-mortem studies have most consistently noted abnormalities in the limbic system and cerebellum[10]. Neuroimaging studies have identified abnormalities in brain and head size, and in cerebellum and limbic structures[11–13], with some individuals showing a pattern of early brain overgrowth[11, 14–16]. Functional MRI (fMRI) studies have reported abnormalities in individuals with ASD when performing various tasks involving language comprehension[17], working memory[18], face recognition[19, 20], and eye movements[9]. Compared with typically developing subjects, individuals with ASD usually express a diffuse network pattern with diminished activity in task-related regions and increased activity in task-unrelated regions[9, 20, 21]. When there is no task involved, individuals with ASD show functional underconnectivity in anterior-posterior connections[22] and reduced connectivity involving the medial prefrontal cortex and the left angular gyrus[23]. Moreover, a lack of deactivation in task-related regions during rest has also been reported in individuals with ASD[24]. Although post-mortem and structural MRI studies of ASD have provided promising insights, abnormalities often fail to provide a direct link to the clinical symptomatology of the disorder, with few exceptions[13, 25], a challenge for which neurophysiological studies offer advantages.

Electroencephalography (EEG), which primarily measures neurophysiological changes related to postsynaptic activity in the neocortex[26], has proven to be a powerful tool for studying complex neuropsychiatric disorders[27–30]. EEG has been the primary measure used to capture and characterize epileptiform and abnormal paroxysmal activity through the detection of focal spikes, which occur with increased frequency in ASD[31, 32]. Resting EEG studies have shown that 20% of individuals with ASD show epileptiform discharges at rest, typically without the presence of clinical seizures[33, 34]. Higher rates of epileptiform activity have also been reported in sleep studies; for example, Chez, et al.[35] reported that 61% of individuals with ASD and no clinical history of seizures displayed epileptiform abnormalities.

The most common way to characterize resting EEG is by breaking down the oscillatory patterns into bands of frequencies that share physiological properties. The typical clinically relevant frequency bands of EEG range from 0.3 to 100 Hz. Within the scope of the current paper, we focus on five frequency bands ranging from 1 to 100 Hz: delta (1 to 3 Hz), theta (4 to 7 Hz), alpha (8 to 12 Hz), beta (13 to 35 Hz), and gamma (>35 Hz). These historically documented frequency bands have attracted rapidly growing interest in clinical and cognitive neuroscience fields, and are believed to govern different cognitive processes[36]. Delta dominates deep sleep, and is thought to underlie the event-related slow waves seen in tasks for detection of attention and salience[37]. Theta is most commonly studied in relation to memory processes[38]. Alpha waves are present in relaxed awake individuals, and are associated with precise timing of sensory and cognitive inhibition[39]. Beta waves are associated with alertness, active task engagement, and motor behavior[40]. Finally, gamma waves, present during working-memory matching[41] and a variety of early sensory responses, are believed to facilitate feature binding in sensory processing[42, 43].

Additionally, EEG recordings can be used to assess functional connectivity between different brain regions over time via EEG coherence, and quantitative measurement of the relationship of frequency spectra between two EEG signals[44]. This advantageous feature can further our understanding of the impaired interactions between brain regions of individuals with ASD that have been suggested by functional MRI studies[45–51].

Advantages of resting-state EEG for studying brain dysfunction in ASD

Resting-state EEG studies are used to monitor brain activity in the absence of overt task performance or sensory stimulation. These measurements can identify abnormalities for which evoked potential studies, the most widely used approach in EEG research with ASD, are not well suited[52]. Indeed, task-dependent changes in brain function are difficult to interpret without fundamental knowledge of functional differences in individuals with ASD at rest. In task-based evoked potential studies, only time-locked neural responses to events of interest are studied; all other spontaneous activity is typically considered background noise[53–55]. Multiple studies have suggested that the brain is a system that operates intrinsically, with intrinsic resting-state integration. External sensory information interacts with, rather than determines, the operation of brain systems[55–57]. For example, many studies have shown that pre-stimulus EEG activities predict the event-related potentials for visual stimuli[58] or motor responses[59].

There are also several practical advantages of using EEG to study brain function in developmental disorders such as ASD. Compared with MRI, EEG can be used across a wider range of age groups and developmental abilities to study brain physiology, has a higher relative tolerance for movement, has higher temporal resolution, is more clinically available, and can be used to collect repeated measurements because (compared with positron emission tomography) it is non-invasive. Resting-state approaches do not require subjects to make a response. This element is particularly promising for studying more severely impaired and/or younger patients who may not be able to perform tasks accurately because of cognitive, physical, or developmental challenges. This is crucial for studying the abnormal maturational trajectory in ASD through early childhood. The literature on resting-state EEG in healthy individuals shows increased alpha power and coherence in individuals with ASD[60], as well as reduced power in low-frequency bands (delta, theta)[61] in adults relative to children. These findings reflect maturation of long-range cortico-cortical connections into adulthood.

Quantitative EEG of resting-state data also has promise as an approach for monitoring treatment outcomes. Pineda et al.[62] reported that individuals with ASD who received neurofeedback training on controlling neural oscillatory activity in the alpha or mu band displayed decreased mu power and coherence, as well as improved performance on an attention test and decreased scores on the Autism Treatment Evaluation Checklist[63]. Neurofeedback training (aiming at reducing theta activity while increasing beta activity) has also been reported to improve executive test performance in individuals with ASD (including attentional control, cognitive flexibility, and goal-setting) for up to 12 months[64, 65].

Despite these unique advantages, relatively few studies have used EEG to study resting-state brain alterations in ASD. In the present article, we review the existing literature on EEG resting-state abnormalities in ASD, discuss potential causes of inconsistencies between studies, and offer suggestions for future studies utilizing resting-state EEG to understand the pathophysiological mechanisms involved in ASD.

Resting-state EEG findings in ASD

Abnormal power

EEG power can be measured as either relative power or absolute power. Relative power is the amount of EEG activity in an individual frequency band divided by the amount of activity in all frequency bands. Absolute power is the amount of EEG activity in one band independent of activity in other bands. Relative power thus reflects the relationship between frequency bands, but does not yield an indication of the degree to which abnormal electrophysiological activity is present in a specific frequency band. Studies of ASD vary widely in the extent to which they present absolute power findings (Table 1), so interpretation of atypical inter-relationships between different frequency bands (relative power) can be advantageous for comparing frequency bands, but also confounds measurement of activity in the target band within any alterations that may occur in other frequency bands. Absolute power is in many ways preferable for developing an understanding of electrophysiological alterations in ASD, and for the interpretation of differences in relative power in this population[71].

In addition to differences between studies in approaches for characterizing EEG activity, studies of ASD are also confounded by the extreme behavioral and putative neurophysiological heterogeneity that characterizes this disorder. Studies vary widely in the demographic characteristics of their samples, and factors such as the age of subjects studied, and whether or not subjects with intellectual disability (ID) were included, may significantly affect study findings. Despite these important concerns, a relatively consistent and unique profile of electrophysiological abnormalities has emerged from resting-state EEG studies, which appears to be present across diverse patient populations. Specifically, excessive power at low-frequency (delta, theta) and high-frequency (beta, gamma) bands, but reduced power in the middle-range frequency band (alpha) (Figure 1) has been found at all stages of development and in children with and children without comorbid ID[72–75]. The excess in delta power has been found in both relative[72, 73] and absolute[73–75] powers, and in multiple brain regions, including the dorsal midline, parietal, right temporal[73] and frontal cortical[74, 75] areas, suggesting a widely distributed pattern of abnormality[72]. Similarly, enhanced low frequency relative[76, 77] and absolute[75, 78] theta (4 to 7 Hz) activity has been seen in both adults[76, 78] and children[75, 77] with ASD, primarily in the frontal and right posterior cortex. Enhanced power has also been reported in high-frequency relative beta (13 to 35 Hz) and absolute gamma (>35 Hz) bands in both adults[76] and children[77, 79] with ASD. Within the higher-frequency bands, the most significant alterations have been found in gamma power over occipital, parietal[76] and midline[77, 79] regions.

In contrast to the excess power displayed in low-frequency (delta, theta) and high-frequency (beta, gamma) bands, individuals with ASD show reduced relative[72, 73, 76] and absolute[80] power in middle-range (alpha) frequencies across many brain regions[72, 73], including the frontal,[76, 80], occipital, parietal[76], and temporal[80] cortex. This pattern indicates a U-shaped profile of electrophysiological power alterations in ASD in which the extremities of the power spectrum are abnormally increased, while power in the middle frequencies is reduced. Available evidence for this model is mostly supportive, but more hypothesis-driven work is needed to confirm and validate it.

We speculate that the etiology for this U-shaped profile may be attributed in part to abnormal functioning of gamma-aminobutryic acid (GABA)ergic tone in inhibitory circuitry, which influences the functional and developmental plasticity of the brain and is thought to modulate power in high-frequency and low-frequency bands while increasing the power of middle-range frequencies (alpha band)[81]. Activity in the gamma band that is visible in EEG seizure recordings has been linked to impairment of dendritic GABAergic inhibition[82]. However, increasing GABA concentration by administrating the GABA antagonist vigibatrin has been shown to increase resting delta power in both rats[83] and humans[84], suggesting that a simple decrease in GABA does not fully explain the U-shaped spectral profile in ASD. For example, thalamocortical delta oscillations are produced by an interaction between GABAergic interneurons and N-methyl-D-aspartate receptors on glutamatergic neurons, which are in turn modulated by dopaminergic neurons in the thalamus[85]. ASD power abnormalities may result from a complex pattern of neurochemical alterations that affect the physiology of inhibitory GABAergic inter-neurons and their modulation of excitatory activity in pyramidal cells.

There is evidence that GABAergic interneuron development and connectivity is disrupted in the prefrontal and temporal cortices in ASD[86], and that this disruption is relevant to excitatory/inhibitory balance[87]. The GABA agonist lorazepam has been shown to increase long-range cortical functional connectivity in the alpha and low beta ranges[88], suggesting an association between GABA tone and large scale cortico-cortical connectivity. The middle alpha frequencies have commonly been associated with an ‘idling’ state, or more recently with active inhibition[39, 89], a state that has been associated with GABAergic circuitry[90]. GABAergic abnormalities can also have early developmental consequences, as GABA acts as an excitatory trophic factor prenatally, guiding growth and connectivity of dendrites[91]. Abnormal embryonic GABA concentrations could lead to development of abnormal excitatory/inhibitory circuitry, causing long-term alterations in the entire oscillatory activity at multiple frequencies. GABA abnormalities could bias neural networks away from the state of active inhibition (alpha) and towards greater excitation (higher frequencies). Intermittent theta-burst stimulation has been shown to increase cortical inhibition in rat neocortex by reducing parvalbumin expression in fast-spiking interneurons[92]. Stimulation in both the delta and theta frequency bands also increases expression of GABA precursors in inhibitory cortical systems[93]. In such cases, increased low-frequency activity could be a compensatory mechanism in ASD to halt the proliferation of high-frequency excitatory activation produced by GABAergic dysfunction. This hypothesis is also consistent with several studies that have documented GABAergic abnormalities in individuals with ASD[94]. Fatemi et al.[95, 96] showed reductions in GABA receptor density in cerebellum and Brodmann’s areas 9 and 40[95, 96]. At the genetic level, it has been suggested that an interaction between GABA receptor subunit genes (GABRA4 and GABRB1) could be directly involved in the etiology of ASD by contributing to increased neuronal excitability, particularly during development, when GABRA4 mRNA levels in brain tissues are at their highest[94]. Increased excitation/inhibition ratios reflecting glutamatergic/GABAergic balance[97], induced by endogenously suppressed GABAergic inhibition[98–101], has been shown in some individuals with ASD. Alpha band power is thought to play an important role in top-down control of sensorimotor responses, including successful voluntary inhibition of contextually inappropriate responses[39]. Many children with ASD show increased levels of inattention and impulsivity[102, 103], which may be linked to increased rates of inhibitory control errors in affected individuals[104]. Although associations between alpha power and inhibitory control deficits in ASD have not been directly examined, the potential role of this neurophysiological mechanism in this domain of behavioral impairment merits investigation.

There arre inconsistencies between studies regarding the U-shaped pattern of power in ASD. Reduced frontal low-frequency delta has been reported in children with ASD without ID[77] and reduced delta has been seen in the frontal and temporal regions in children with high-functioning and low-functioning ASD[80]. A few studies reported no differences in alpha frequency bands in children with ASD[77, 105] and enhanced alpha power over frontal/parietal/midline regions in high-functioning individuals with ASD[106]. Dawson and colleagues[80] found reduced theta band activity in the frontal, temporal, and parietal regions, with no effect in the beta band. Differences in participant characteristics such as IQ and age might account for the inconsistencies in these findings. Within the delta band, all enhanced delta findings (supporting the U-shaped curve hypothesis) have been most robust in relatively low-functioning children (mean IQ = 37.[72]), those with over 20% mental age delay[74], and those with 28% lower intelligence score in ASD[73]. However, reduced delta power was reported in high-functioning children with ASD (mean Full Scale Intelligence Quotient of 93) by Coben et al.[77]. Findings of reduced alpha power (again supporting the U-shaped curve hypothesis) have been most consistent in low-functioning children with ASD[72, 73, 80], but there are examples of this pattern in relatively high-functioning adults with ASD[76]. Studies reporting enhanced or unaffected alpha power were reported for high-functioning children with ASD[77, 105, 106]. The relation between resting EEG abnormalities and level of intellectual disability, impairments in various behavioral domains, history of regression, and other clinical features of ASD will be an important focus of future research in this area as larger, well-characterized cohorts are studied.

Abnormal hemispheric asymmetry

In addition to spectral power differences in individuals with ASD, changes have been reported in the hemispheric asymmetry of brain neurophysiology. The majority of the existing resting-state EEG literature reports enhanced power in the left compared with the right hemisphere in individuals with ASD, across all frequency bands[72, 74, 78, 106, 107]. This asymmetry is generally much larger than the mild individually variable asymmetries seen in typically developing humans[106].

Cantor and colleagues[72] reported that subjects with ASD had enhanced power in the delta band, in the posterior-temporal, midline, and occipital regions of the left hemisphere. Similarly, Stroganova et al.[74] found enhanced delta power in the left hemisphere of individuals with ASD in the frontal, temporal, and parietal regions. In the theta band, left-hemisphere dominance in ASD was seen in frontal[74, 78], parietal[74], temporal[72, 74], and occipital[72, 74] regions. In the alpha band, left-hemisphere dominance in ASD was reported in multiple studies in mid-frontal[106, 107], temporal, parietal[72, 74], midline[72, 106], and occipital[72] regions. Finally, Cantor et al.[72] replicated the left-hemisphere dominance pattern in the beta band in posterior-temporal, midline, and occipital regions.

Left-hemisphere asymmetry in ASD is of clinical interest, given the common language abnormalities seen in ASD[108–111]. Increased resting power in the left hemisphere may contribute to left-hemisphere performance deficits by decreasing the signal-to-noise ratio during active tasks, similar to reports of increased background noise and behavioral performance impairment in the literature on schizophrenia[112, 113]. Left-hemisphere dysfunction may also be dependent on the task that subjects are performing. When performing tasks of executive functioning (for example, Go/No-go and Stroop tests), high-functioning adults with ASD had significantly increased activation restricted to the left hemisphere[114]. Left-hemisphere dysfunction has also been identified in smooth pursuit eye movements in individuals with ASD[115], as have left-lateralized alterations during an oculomotor serial reaction time task[116].

Nevertheless, as in many neuropsychiatric disorders, evidence of lateralized abnormalities has been inconsistent[74]. Dawson et al.[80] reported reduced delta power in the left mid-temporal cortex, and Lazarev et al.[105] reported no left/right-hemisphere differences frequency bands in the occipital cortex. However, in the same and subsequent reports, the same authors noted hyperconnectivity within the left hemisphere[117] and reduced power in the right hemisphere[105] in children with ASD during presentation of photic driving stimulation. Dawson and colleagues[80] utilized a relatively short (1 second) window to measure delta power, which may affect reliability of measured low-frequency activity. Lazarev and colleagues[105, 117] measured activity only in 14 relatively heterogeneous subjects with ASD, so those studies may have lacked statistical power to detect effects. Photic driving is a robust response, and would potentially be more sensitive to small pathological alterations in studies with small patient cohorts.

Abnormal coherence

Resting-state EEG studies of ASD have also documented reduced long-range coherence patterns[76, 77, 118, 119]. Weaker coherence between frontal and occipital regions was reported for delta[77, 118] and theta[77] bands, whereas Murias and colleagues[76] reported significantly reduced alpha coherence between the frontal cortex and the temporal, parietal, and occipital cortices. Duffy and Als[119] reported weaker left frontal-temporal connectivity within the beta band. This finding is similar to results from multiple fMRI studies that have shown reduced left frontotemporal connectivity during resting state[18, 46, 120]. These findings parallel those from Horwitz and colleagues[121], who used positron emission tomography to show reduced correlations in glucose metabolism between frontal and other cortical areas in resting adults with ASD. Generalized decreases in frontoparietal and fronto-occipital connectivity have also been reported in ASD during resting fMRI[22, 50]. These findings converge to suggest weakened long-range connectivity between the frontal lobe and other cortical regions. The frontal lobe plays a crucial role in higher-order cognitive, language, social, and emotional functioning[122]. Thus, it is not surprising that clear deficits in frontal lobe connectivity have been reported in ASD, as frontal lobe abnormalities have been proposed to play a key role in regulating a wide range of cognitive, sensory, and motor processes[123–125].

Findings on short-range connection patterns in resting EEG studies are less consistent. Both intrahemispheric and interhemispheric local coherences in all brain regions have been reported to be reduced in delta and theta bands[77], while a reduced local coherence over mid-frontal regions has been found in both the delta[118] and alpha band[76]. By contrast, enhanced local coherence has been found over the lateral-frontal region in the delta band[118], and over left frontal and temporal regions in the theta band[76]. Furthermore, these functional frontal deficits have been linked to structural abnormalities in the frontal lobe in ASD[126]. Although several studies have suggested excessive short-range connectivity in ASD, due to increased density of cortical mini-columns[123] and disproportionately increased white matter[127, 128], some diffusion tensor imaging (DTI) studies failed to report this pattern[49]. Although short-range coherence studies will require more investigation with newer high-resolution DTI techniques, in parallel with resting-state measures of frontal connectivity using EEG, weaker long-range coherence between frontal and other brain regions found in resting EEG studies suggests that the frontal lobe is less well integrated with other local cortical areas, in ways likely to have considerable neurobehavioral significance[123]. Further examination of this model is necessary to understand how these functional abnormalities relate to clinical phenotypes.

Crucial considerations

Although previous studies of resting-state EEG in ASD have identified abnormalities in low-frequency and high-frequency band power, connectivity, and lateralization of brain functions, there are multiple crucial methodological and clinical issues that warrant attention in reviewing this literature and in planning future research.

Future directions