Atypical EEG complexity in autism spectrum conditions: A multiscale entropy analysis

Abstract

Objective

Intrinsic complexity subserves adaptability in biological systems. One recently developed measure of intrinsic complexity of biological systems is multiscale entropy (MSE). Autism spectrum conditions (ASC) have been described in terms of reduced adaptability at a behavioural level and by patterns of atypical connectivity at a neural level. Based on these observations we aimed to test the hypothesis that adults with ASC would show atypical intrinsic complexity of brain activity as indexed by MSE analysis of electroencephalographic (EEG) activity.

Methods

We used MSE to assess the complexity of EEG data recorded from 15 participants with ASC and 15 typical controls, during a face and chair matching task.

Results

Results demonstrate a reduction of EEG signal complexity in the ASC group, compared to typical controls, over temporo-parietal and occipital regions. No significant differences in EEG power spectra were observed between groups, indicating that changes in complexity values are not a reflection of changes in EEG power spectra.

Conclusions

The results are consistent with a model of atypical neural integrative capacity in people with ASC.

Significance

Results suggest that EEG complexity, as indexed by MSE measures, may also be a marker for disturbances in task-specific processing of information in people with autism.

Introduction

Physiological complexity, comprising the presence of non-random fluctuations over multiple time scales in the seemingly irregular dynamics of physiological outputs (Freeman, 1992, Glass and Mackey, 1992, Manor et al., 2010) is increasingly being recognized as contributing a novel descriptive approach to the investigation of typical and pathological developmental or degenerative states (Costa et al., 2002, Costa et al., 2005, Fallani Fde et al., 2010, Ouyang et al., 2010). Whilst the interpretation of the meaning of changes in complexity varies according to the physiological parameters studied and the developmental or clinical condition being investigated, there is nevertheless increasing evidence that a variety of pathological processes are associated with atypical and often, but not always, reduced measures of physiological complexity (Escudero et al., 2006, McIntosh et al., 2008, Kang et al., 2009, Istenic et al., 2010, Manor et al., 2010, Mizuno et al., 2010, Takahashi et al., 2010, Bosl et al., 2011). Regarding brain activity specifically, electroencephalographic (EEG) activity provides fine temporal resolution, making it particularly suitable for investigating inherently complex biological signals arising from brain systems regulated by multiple sources interacting with each other over different time scales, mechanisms, couplings and feedback loops (Bhattacharya et al., 2005, Fallani Fde et al., 2010, Ouyang et al., 2010).

There are reasons to suspect that autism spectrum conditions (ASC) may be associated with atypical patterns of brain complexity. ASC are a set of pervasive neurodevelopmental conditions with onset in early childhood and a wide range of life-long signs and symptoms that suggest an association with atypical functioning at a relatively profound level of brain functioning. Core features of ASC include a restricted repetitive range of behaviours, interests and activities; impairments in reciprocal social interactions; and qualitative disturbances in communication (American Psychiatric Association, 2000). In addition to these characteristic social and cognitive features, atypical patterns of sensory and motor functioning and integration are also increasingly recognised as features of ASC, with evidence of atypical visual perception (Simmons et al., 2009), including perception of biological motion (Kaiser et al., 2010), auditory perception (Hitoglou et al., 2010), somatosensory integration (Russo et al., 2010), and motor functions (Gidley Larson, 2006) as well as impaired sensorimotor integration (Haswell et al., 2009), motor planning and control (Jansiewicz et al., 2006, Rinehart et al., 2006, Freitag et al., 2007) and reduced adaptability to environmental changes (Russo et al., 2007, Thakkar et al., 2008, Foley Nicpon et al., 2010). There is also evidence that motor deficits do not occur in isolation from the social and cognitive features of ASC (Dziuk et al., 2007). In attempting to explain this wide range of features of ASC several explanatory models of brain functioning that suggest disturbances of underlying brain complexity have been proposed, including atypical neural connectivity (Belmonte et al., 2004, Courchesne and Pierce, 2005, Just et al., 2007, Barttfeld et al., 2011, Wass, 2011) and disrupted temporal integration of information (Brock et al., 2002, Rippon et al., 2007). Supporting the possibility of atypical functional complexity in autism, it has been observed that, in those without ASC, improved adaptability to cognitive demands is associated with increasing physiological variability reflected by greater scalp EEG complexity (McIntosh et al., 2008, Sitges et al., 2010) and that altered neural connectivity may be associated with atypical signal complexity in schizophrenia (Friston, 1996) and Alzheimer’s disease (Jeong, 2004). In addition, a recent study by Bosl et al. (2011) has shown a decrease in resting state EEG complexity in infants at high risk of ASC, when compared to normal controls, with low risk of ASC.

In order to examine whether ASC is associated with an atypical pattern of complexity of brain function we examined multiscale entropy (MSE) as a measure of physiological complexity in scalp-recorded EEG data in a group of adults with ASC and a matched typically developing control group. Entropy is a physical quantity that measures the order of a system. Regular systems have lower values of entropy, whilst totally irregular systems have very high values of entropy. However, regularity is not necessarily correlated with complexity. Random phenomena like white noise have very low regularity and will therefore present high values of entropy, but they do not have the structural richness of information over multiple spatial and temporal scales that characterises complex systems (Costa et al., 2002, Costa et al., 2005). In order to overcome this problem and differentiate between white noise and true complexity, Costa et al., 2002, Costa et al., 2005 introduced the method of MSE, which quantifies the complexity of a physiological signal by measuring the entropy across multiple time-scales, using a coarse-graining procedure. This model proposes that optimally functioning biological systems are modulated by multiple mechanisms which interact over multiple temporal scales. These processes generate complex data composed of overlapping signals from all the interrelating mechanisms. In these circumstances, MSE analysis will reveal a high value of entropy sustained for increasingly coarser time-scales. For random noise signals however, the system will show a decrease in entropy values as the time-scales increase. This is because a random white-noise signal has information only on the shortest time-scale; as the time-scales increase, since no new structures are revealed, the standard deviation of the signal decreases, causing a progressive decrease in the values of entropy with time-scale (Costa et al., 2005).

Brain activity in typical development from childhood to adulthood has been associated with increasing MSE (McIntosh et al., 2008), and in a study of adults with schizophrenia increased MSE has been observed in fronto-central and parietal regions (Takahashi et al., 2010) whereas age-related response to photic stimulation in typical individuals (Takahashi et al., 2009), and treatment of schizophrenia with antipsychotics, have been associated with reduced MSE. Alzheimer’s disease has been associated with several patterns of EEG complexity, with earlier studies reporting lower MSE (Escudero et al., 2006, Park et al., 2007) whilst more recently Mizuno et al. (2010) reported relatively decreased complexity over smaller timescales but relatively increased complexity at coarser timescales, possibly reflecting different modulating effects by separate neuropathophysiological mechanisms. Additionally, Bosl et al. (2011) have recently shown a decrease in resting state EEG complexity, at several stages of development, for infants at high risk of ASC, when compared to infants at low risk of ASC.

Whilst some MSE studies have analysed data collected during resting states (Escudero et al., 2006, Hornero et al., 2009, Takahashi et al., 2010, Bosl et al., 2011), others have employed activation or stressor tasks to explore responses to stimuli of relevance to the physiological or clinical process of interest (Takahashi et al., 2009, Manor et al., 2010, Sitges et al., 2010). Previous event-related potential (ERP) studies have found that face processing in some circumstances is impaired in people with ASC (O’Connor et al., 2005, Jemel et al., 2006, Churches et al., 2010) and in this study we analysed EEG recorded whilst participants observed images of faces and other objects.

In this investigation the first aim is to determine whether typical controls and a group with ASC have similar or differing patterns of MSE. We predict that those with ASC will have an atypical pattern of complexity, as reflected by significantly different MSE values at coarser time scales compared to controls. Secondly, given the reduced behavioural adaptability observed in ASC (Russo et al., 2007, Thakkar et al., 2008, Foley Nicpon et al., 2010), and the observation that in the general population greater adaptability is associated with higher MSE values (McIntosh et al., 2008), along with findings of reduced complexity in infants at high risk of ASC by Bosl et al. (2011), we hypothesise that MSE will be reduced over coarser time scales in the ASC group, when compared to MSE in the control group, during performance of a visual matching task. To address the question of the extent to which differences in EEG complexity may relate to group differences in EEG power spectra, we also conducted a traditional power analysis using the same EEG data. Based on results from previous studies (Milne et al., 2009, Raymaekers et al., 2009) we did not expect to find any differences in EEG power spectra between the ASC and the control groups.