Response to Davis and Plaisted-Grant: Low or high endogenous neural noise in autism spectrum disorder?

As has been pointed out by Davis and Plaisted-Grant (2014), while there is a current trend towards fractionating the complex presentation of autism spectrum disorder (ASD) into components which may be more amenable to explanation in terms of genetic traits, there is still a strong temptation to look for what they term ‘heuristic theories’, which explain the entire phenotype. While often flawed in terms of their details, these theories can provide a framework in which to place targeted experiments which then further our knowledge of this enigmatic condition.

This is the context in which Davis and Plaisted-Grant (2014) place their ‘low-noise’ account of ASD. At first glance, this theory appears to speak directly against the suggestion, made by one of us (Simmons et al., 2009), that many of the behaviours and perceptual symptoms experienced by people with ASD may be explained by increased neural noise. However, a close look at the low-noise account suggests that the low-noise argument can be considered as complementary, rather than contradictory to the high-noise argument. As both the high- and the low-noise accounts of ASD demonstrate, compelling theories can be developed from relatively little direct evidence. What is required now is to put these theories to the test. In other words, we must be guided by data rather than by eloquence.

Both accounts implicate noise at the neural level, and conceive of noise as variability. In the high-noise account, noise is postulated to arise from increased neural cross-talk or mis-firing synapses, either, or both, of which may lead to increased neural variability across large-scale neural networks. Set within the context of stochastic resonance, such an increase in endogenous noise could, in some sense, provide a potentially neat explanation of sensory difficulties experienced in autism, by providing enhancement (and thus hyper-sensitivity) under some conditions and masking (and thus hypo-sensitivity) under others (Simmons et al., 2009). The low-noise account defines noise as variation in neural responses within small-scale neural networks. Within this framework, increased signal-to-noise ratio reduced switching between states and reduced overlap of stimulus representations are suggested consequences of low-noise.

There are two crucial assumptions within the low-noise account: (1) that the level at which low-noise occurs can be defined as variation in responses between neurons within a local network when responding to a single stimulus and (2) that this ‘local’ noise, leads to an increase in ‘global’ noise. Assumption 2 is particularly important to the account as it provides a way of reconciling low-noise in ASD with existing evidence of increased electroencephalogram (EEG) and blood-oxygen-level-dependent (BOLD) signal variability in ASD (Dinstein et al., 2012; Milne, 2011). An admirable quality of the low-noise account is that it centres the debate on the spatial scale at which increased or decreased noise may or may not occur in ASD. However, extending the assumptions outlined above, it becomes clear that the low-noise theory as articulated by Davis and Plaisted-Grant provides an alternative explanation for the origin of increased global noise, rather than offering a conflicting theory. Thus, a challenge to the field is to clarify the definition of neural noise. While the high-noise theory is predicated by increased neural variability, the low-noise theory neatly side-steps this issue by claiming that inter-trial variability in EEG or BOLD signals could be just as easily explained by decreased noise in individual neurons or local networks as by increased noise in individual neurons or local networks. A second challenge is to establish the spatial scale at which atypical noise exerts its effects on behaviour and perception in those with ASD. For example, if reduced local noise leads to increased global noise then do the perceptual and cognitive findings in ASD described by Davis and Plaisted-Grant arise from reduced local noise, increased global noise or both? How, and indeed can, these effects be disentangled empirically?

Davis and Plaisted-Grant have made a valuable contribution by highlighting the importance of considering spatial scale in theoretical accounts of noise. However, defending a high-noise position over a low-noise position at this stage seems to be somewhat premature given the lack of clarity regarding what constitutes noise. A case in point is the issue of whether the neural noise is additive or multiplicative, and what potential consequences this might have (Dinstein et al., 2012). We end this short response therefore on a cautionary note: while heuristic theories play an important part of refining scientific enquiry and stimulating new ideas, these theories are only useful when they are clearly defined and lead to clear and testable hypotheses. At present, it is unclear, at least to us, how reduced local and increased global noise can be separated experimentally within the current low-noise framework.