Distortions and disconnections: Disrupted brain connectivity in autism

Abstract

The past few years have seen considerable interest in findings of abnormal brain connectivity in the autism spectrum disorders (ASD). We review recent work from neuroimaging and other sources, and argue that there is considerable convergent evidence suggesting that connectivity is disrupted in ASD. We point to evidence both of local over-connectivity and of long-distance under-connectivity, and describe some non-uniformities in this picture, most notably that disruptions appear more severe in later-developing cortical regions. We conclude by discussing a number of extant questions. Firstly, we consider whether aberrant connectivity should be seen as part of the primary pathogenesis of autism, or whether disrupted connectivity in ASD emerges over time. Secondly, we consider how the patterns of disrupted connectivity found in ASD might relate to those being found in a range of other disorders.

Introduction

The autism spectrum disorders (ASD) are neurodevelopmental disorders characterized by deficits in social interaction, communication, and stereotyped or repetitive behaviors (APA, 1994). Although the first report of functional under-connectivity in ASD came from Horwitz, Rumsey, Grady, and Rapoport (1988) it is only in recent years that the study of connectivity in ASD has attracted widespread attention (although see also Brock et al., 2002, Carpenter et al., 2001, Castelli et al., 2002, Rubenstein and Merzenich, 2003). Just, Cherkassky, Keller, and Minshew (2004) provided an influential formulation of what they dubbed “under-connectivity theory”, arguing that “autism is a cognitive and neurobiological disorder marked and caused by under-functioning integrative circuitry that results in a deficit of integration of information at the neural and cognitive levels”. In recent years, under-connectivity theory has attracted considerable attention as a number of studies have reported lower than expected between-region functional correlations on a range of tasks (e.g. Koshino et al., 2008), and DTI has revealed markers of disordered structural connectivity in individuals with ASD (e.g. Sundaram et al., 2008).

Belmonte et al. (2004) provided an important addendum to the Just et al. paper. They noted that whereas some authors discuss ASD as a problem of under-connectivity (Brock et al., 2002, Just et al., 2004), others treat the problem as one of over-connectivity (e.g. Rubenstein & Merzenich, 2003). Belmonte et al. reconciled these two ideas, suggesting that “high local connectivity may develop in tandem with low long-range connectivity” and that “high physical connectivity and low computational connectivity may reinforce each other by failing to differentiate signal from noise” (see also Johnson, 2005, Quartz and Sejnowski, 1997, Rubenstein and Merzenich, 2003).

Long-distance under-connectivity in ASD implies that “any facet of psychological function that is dependent on the coordination or integration of brain regions is susceptible to disruption, particularly when the computational demand of the coordination is large” (Just et al., 2004). “Core” autistic deficits in social interaction, language and repetitive and restrictive behaviors (APA, 1994) arise because these are the domains that place the largest demands on the time-sensitive integration of information from spatially discrete brain areas (Herbert, 2005, Just et al., 2004, Lewis and Elman, 2008; although see Mottron, Dawson, Soulieres, Hubert, & Burack, 2006). As such, the theory is in some ways a neural substantiation of central coherence theory (e.g. Happe, 1999, Happe and Frith, 2006) which posits “a cognitive style biased towards local rather than global information processing” (Happe, 1999), although Just et al. (2004) contest this claim, pointing out that central coherence theory postulates “a core deficit in central processing” (Happe & Frith, 2006) whereas “under-connectivity theory treats the coherence as an emergent property of the collaboration among brain centers” (Just et al., 2004).

Local over-connectivity has also been connected to findings of behavioral hyper-specificism and inferior generalization in ASD (Casanova et al., 2006a, Cohen, 2007; see also Gustafsson, 1997, McClelland, 2000), and to superior discrimination on certain tasks (Cohen, 2007; see also Mottron et al., 2006 for an excellent discussion of this), as for example in the embedded figures task (see also Plaisted et al., 1998, Shah and Frith, 1983).

Friston (1994) suggested that there are two guiding principles to neural function. The first is functional segregation and the second is functional integration, a process mediated by connectivity. Functional segregation describes the location of information processing; as Friston notes, the explosion of neuroimaging work over the last twenty years “has been extremely successful in establishing functional segregation as a principle of organization in the human brain” (Friston, 1994). The aspect that has received relatively less attention is the functional integration of information – how information is combined between different areas during the performance of particular tasks (Friston, 1994, Honey et al., 2009, Horwitz, 2003).

In this article we will distinguish two types of connectivity (following Friston, 1994, Horwitz, 2003, Sporns, 2007):

Structural connectivity refers to the physical or structural (synaptic) connections linking sets of neurons or neuronal elements, and to their associated structural biophysical attributes encapsulated in parameters such as synaptic strength or effectiveness (Friston, 1994, Sporns, 2007). Structural connectivity can describe how individual neurons are connected (at the micro scale) and also how different brain regions are connected (at the macro scale).

Functional connectivity refers to the degree to which activity in one area correlates with activity in another (David et al., 2004, Friston, 1994), or to the temporal synchronization of activation of two brain areas during task performance (Friston, 1994, Horwitz, 2003, Rippon et al., 2007). As a purely correlative measure, measurements of functional connectivity leave open questions of causation, and of whether control is symmetric, asymmetric, or coordinated by a third area (c.f. effective connectivity – Friston, 1994, Friston, 2009).

It is important to recognize that structural and functional connectivity are not necessarily co-referential. For example, tonic neuromodulatory changes can affect the functional activation patterns of a range of neurons without altering the structural connectivity (Friston, 1994; see also He et al., 2008, Honey et al., 2009, Sorg et al., 2007, Sporns, 2007, Sporns et al., 2000, Stephan et al., 2009).

The past few years have seen an explosion in interest in the tools that can be used to study brain connectivity (Friston, 2002, Honey et al., 2009, Horwitz and Glabus, 2005). Structural connectivity is being measured using MRI-based techniques such as diffusion tensor imaging (DTI) (Behrens et al., 2003, Conturo et al., 1999; see Karlsgodt et al., 2008 for an excellent brief summary) that examine the structural integrity of white matter tracts that are seen as the inter-areal macro structural correlates of brain connectivity (Fields, 2008, Oishi et al., 2008; although this may be a simplistic view – see e.g. Karadottir et al., 2008, Schummers et al., 2008, Ziskin et al., 2007). Tractography based on DTI (e.g. Hagmann et al., 2003, Hagmann et al., 2008, Lewis et al., 2009) can be used to trace fiber tracts, allowing association fiber pathways to be mapped in vivo. Problems remain with these techniques, though, particularly coping with multiple fiber orientations within a single voxel (Wedeen et al., 2008).

At the macro (whole brain) level, structural MRI has also been used to look at correlations between the size of different brain regions (e.g. Boucher et al., 2005), and a number of other techniques have also been used to aid the parcellation of structural MRI data, including voxel-based morphometry (VBM), a post hoc analysis that allows the relative volumes of different brain areas to be compared with greater precision than traditional morphometric techniques (e.g. Abell et al., 1999).

At a much higher spatial resolution, post mortem histological analyses (e.g. Casanova et al., 2006a) can trace neural connectivity at the micro- and meso-levels (such as mini-columnar structures), although sample sizes, particularly of immature subjects, remain (thankfully) very low.

Functional connectivity data on fMRI comes from studying correlations between activation patterns of particular voxels during task performance. EEG has been used to provide indices of connectivity in ASD in two ways – firstly, power in the higher frequency bands (particularly gamma, typically >25 Hz), that is thought (e.g. Csibra et al., 2000, Engel and Singer, 2001, Paik et al., 2009) to represent binding between spatially discrete areas of the brain (see e.g. Rippon et al., 2007, Thai et al., 2009). Secondly, coherence, which gives an index of the average degree of correlation between spatially discrete electrode groupings (see e.g. Lachaux et al., 1999, Murias et al., 2007, Thatcher et al., 2008).

For both fMRI and EEG, more advanced techniques have recently been developed that allow the calculation of effective connectivity – the influence that one neural system exerts over another (Friston, 1994, Friston, 2009, Sporns, 2007). These techniques, which include Granger causal modeling and dynamic causal modeling (see Friston, 2009 for a review) have not, to our knowledge, been applied to data from subjects with autism.

It is also interesting to note that some ingenious behavioral measures have been interpreted as evidence of abnormal functional connectivity. For example, Tommerdahl, Tannan, Holden, & Baranek, 2008 (see also Haswell, Izawa, Dowell, Mostofsky, & Shadmehr, 2009) investigated the absence of stimulus-driven synchronization effects on tactile perception in autism, a finding they suggest may relate to macro-columnar under-connectivity in the somatosensory cortex. Again, a full discussion of this work is beyond the scope of this review.

Finally, neural network modeling – which can model both structural and functional aspects – has played an important role in investigating how disrupted connectivity might affect subsequent autistic development at a variety of spatiotemporal scales (e.g. Noriega, 2008).

In addition to those areas we discuss in this paper, there has also been considerable discussion of the micro-structural correlates of connectivity, such as neurite morphology and synaptogenesis (see, for example, Persico & Bourgeron, 2006), signaling molecules such as HGF/MET, Reelin and neurotrophins (see e.g. Pardo & Eberhart, 2007), synaptic proteins (e.g. neuroligins – see Garber, 2007, Gutierrez et al., 2009) and neuro- and/or gliogensis (see e.g. Courchesne et al., 2007, McCaffery and Deutsch, 2005). A number of authors have also discussed the role that disruptions to various neurotransmitters (in particular GABA) might play during development (Delong, 2007, Kana et al., 2007, Van Kooten et al., 2005).

A full review of this material would render this paper indigestibly large; the interested reader is recommended to visit a number of excellent recent discussions on these subjects. Particular highlights include Courchesne et al. (2007, on the micro-structural correlates of early brain overgrowth), Herbert (2005, also on the mechanisms that might drive brain overgrowth), Pardo and Eberhart (2007, on how the micro correlates of structural connectivity might become disrupted as a result of aberrant development) and Persico and Bourgeron (2006, also on the micro correlates of disrupted structural connectivity).

In Sections 3 Hypothesis 2 – over-connectivity exists within local networks in ASD, 4 Hypothesis 3: connectivity disruptions are uniform across the whole brain in ASD of this article we review recent evidence regarding connectivity in ASD. For the sake of clarity, we have structured our discussion around three principal hypotheses. In each section, we review the evidence from a variety of methodologies that can be brought to bear in addressing each hypothesis.