Neural variability: friend or foe?

1 Neural variability : friend or foe?Ilan Dinstein1, David J. Heeger2, and Marlene Behrmann31 Department of Psychology, Ben Gurion University, PO Box 653, Beer Sheva 84105, Israel2 Department of Psychology and Center for Neural Science, New York University, 6 Washington Place, New York, NY 10003, USA3 Department of Psychology, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, PA 15213, USAA lthough we may not realize it, our brain function variesmarkedly from moment to moment such that our brainresponses exhibit substantial variability across trialseven in response to a simple repeating stimulus. Shouldwe care about such within-subject variability ? Are theredevelopmental, cognitive, and clinical consequences tohaving a brain that is more or less variable/noisy?Although Neural variability seems to be beneficial forlearning, excessive levels of Neural variability are appar-ent in individuals with different clinical disorders. Wepropose that measuring distinct types of Neural variabil-ity in autism and other disorders is likely to reveal crucialinsights regarding their neuropathology.

2 We further dis-cuss the importance of studying Neural variability moregenerally across development and aging in of Neural variabilityMoment-to-moment Neural variability is generated by manyneurophysiological mechanisms. At the single cell level,these include the noisy response characteristics of periph-eral sensors [1], the stochastic nature of synaptic transmis-sion [2], and the dynamic changes caused by neuraladaptation [3] and synaptic plasticity [4]. At the neuralnetwork level, additional variability is generated by adjust-ments of the excitation/inhibition balance [5], changes inattention and arousal levels [6], continuous interaction andcompetition across large Neural populations [7], and distrib-uted neuromodulation effects [8]. Working together, thesemechanisms (and others) generate substantial variabilitysuch that Neural responses to even a simple, mundanestimulus differ markedly across trials of an experiment[9 13].Estimating the amount of Neural variability associatedwith each of the sources described above in humans isdifficult.

3 However, when measuring within-subject neuralvariability using neuroimaging and electrophysiology tech-niques, it is possible to decompose Neural variability intovariability that appears in early versus late parts of thestimulus/task evoked response, variability that is specificto a local brain area versus variability that is shared acrossthe entire brain, and ongoing Neural variability thatappears in resting-state recordings where the stimulusor task are absent (Box 1). Do these distinct measures ofneural variability tell us anything about the integrity ofthe individual s brain function? Are particular levels ofneural variability indicative of the individual s perceptualand cognitive abilities or their clinical state?Excessive Neural variability in autismAutism is a developmental disorder which is diagnosedbased on the presence of specific behavioral symptoms thatinclude social communication difficulties, abnormal senso-ry sensitivities, and repetitive behaviors [14]. Prominenthypotheses about autism posit that it may result fromexcitation inhibition imbalances [15 17], abnormalitiesin genes that govern Neural migration and proliferation[18,19], and synaptic maturation and transmission[20,21].

4 Such fundamental Neural alterations are hypothe-sized to create widespread Neural processing abnormalities[22,23], which may include excessive Neural variability /noise [16,24,25].In support of this hypothesis, several fMRI [26 28] andelectroencephalography (EEG) [29,30] studies havereported that brain responses of high-functioning individ-uals with autism exhibit excessive trial-to-trial variabilityin comparison to brain responses of matched controls. Intwo of these studies, we examined the trial-to-trial vari-ability of fMRI response amplitudes when participantswere presented with simple visual, auditory, or tactilestimuli in three independent experiments. Although themean fMRI response amplitudes across trials in eachsensory domain were indistinguishable across the twogroups, the standard deviation across trials was approxi-mately 20% larger in the autism group in all three sensorysystems (Figure 1). These findings suggest that excessiveneural variability is a widespread phenomenon apparentin the responses of multiple sensory systems (and poten-tially other brain systems as well) in excessive Neural variability was apparent inthese sensory-evoked responses, it was not present incomparisons of ongoing ( resting state ) Neural fluctuations[26,27], suggesting that increased Neural variability inautism was specifically associated with sensory evokedprocesses rather than with ongoing Neural that, by dissociating Neural variability in these twodistinct situations (resting state and stimulus-evoked), it ispossible to separate the contribution of different underly-ing physiological mechanisms that drive Neural variability (Box 1).

5 Additional analyses revealed that the level ofneural variability within each participant was consistentacross the different sensory experiments (Figure 2), sug-gesting that Neural variability was a stable characteristicof each participant s brain, and this was similarly evidentacross all examined sensory 2015 Elsevier Ltd. All rights reserved. author: Dinstein, I. Trends in Cognitive Sciences, June 2015, Vol. 19, No. 6 How might excessive Neural variability be associatedwith the core social, sensory, and repetitive behaviorsymptoms that define autism? We speculate that variable,unreliable Neural responses in multiple sensory and asso-ciative brain areas during early autism development maycreate an unstable and unpredictable perception of theenvironment. Increased Neural noise may be associatedwith more Neural plasticity, as demonstrated in someanimal models of autism [16] (also see section below aboutdevelopment). In such a situation, individuals with autismmay indeed find it difficult to learn the correct probabilitiesand statistics of external events and, therefore, exhibitdifficulties in predicting their environment [31 34].

6 Thisunpredictability may be particularly accentuated in socialsituations where humans (unlike objects) display a widevariety of variable social and emotional cues, which mustbe perceived using multiple sensory modalities [35]. Devel-oping under such conditions may motivate an infant toretract from social interactions and engage in repetitivebehaviors (often involving objects) that are likely to gener-ate more predictable Neural responses. In addition, exces-sive Neural variability in sensory and motor systems mayexplain why individuals with autism exhibit balance pro-blems, motor clumsiness [36], differences in visual percep-tion [24,25] (see section below about perception in autism),and abnormally large behavioral variability in trial-to-trialreaction times [37,38], eye saccade accuracy [39], reachingmovement accuracy [40], and pitch of voice during speech[41]. Previous studies have also suggested that unstable/noisy Neural networks are more likely to develop epilepticseizures [15], which are indeed more prevalent in autismthan in the general population [42].

7 Although excessive Neural variability has so far beenreported only in sensory and motor systems of individualswith autism, these findings may indicate a more fundamen-tal and widespread physiological alteration in autism thatmight perturb Neural processing across many brain sys-tems. Recent theoretical discussion on this topic has pro-posed that both reduced and increased endogenous neuralnoise at the level of the single neuron or small-scale, localneural circuits may generate the increase in large-scaletrial-to-trial variability demonstrated by the EEG and fMRIstudies mentioned above [43,44]. Systematic characteriza-tion of Neural variability at different levels of sensory,emotional, and social processing, and at different stagesof development (with a particular focus on early develop-ment), is highly warranted for assessing these influential hypothesis is that autism is causedby abnormalities in synchronization of Neural activityacross distant brain areas as assessed with functionalconnectivity techniques that measure the correlation inactivity across brain areas [45,46].

8 Numerous neuroimag-ing studies have indeed reported that toddlers [47], chil-dren [48], adolescents [49], and adults [50,51] with autismexhibit abnormal functional connectivity in contrast tomatched controls. In addition, studies using diffusion ten-sor imaging (DTI) techniques have revealed that children,Box 1. Measuring multiple components of Neural variabilityin humansNeural variability can be separated into several distinct andmeasurable components. When recording brain activity usingEEG, it is possible to separate variability at different latencies fromstimulus or task onset. For example, when performing an eventrelated potential (ERP) analysis, it is possible to identify commonlydescribed early and late peaks of the ERP, such as the N1 and P2 ofan auditory evoked response, which are thought to representdifferent underlying sensory and cognitive processes [91], andassess the trial-to-trial variability of each (see Figure 3A in maintext). When recording brain activity with fMRI, it is possible toseparate variability across space.

9 For example, it is possible toseparate local variability from global variability by measuringtrial-by-trial variability in a local region of interest before and afterregressing out the global mean grey matter time-course (seeFigure 3B in main text). Finally, variability associated with ongoing( , resting state) activity can be separated from variabilityassociated with stimulus-evoked or task-evoked activity by comput-ing trial-to-trial variability before and after stimulus/task onset, byassessing variability on trials where the stimulus was absent (seeFigure 3A in main text), or by examining longer resting-staterecordings during which no stimulus is presented and no task isperformed. Such comparisons have been carried out extensively inelectrophysiology studies with animals [13,92], but have rarely beenperformed in human EEG and fMRI estimating Neural variability in humans using neuroima-ging techniques, it is important to remember that each technique isprone to large sources of measurement noise.

10 For example, fMRIscans are susceptible to head-movement artifacts [93], and EEGmeasures are susceptible to saccade and eye-blink artifacts [94]. It is,therefore, particularly important to measure and control external(non- Neural ) sources of variability when attempting to characterizetrial-to-trial Neural variability .(A) amplitude(% signal change)Time (s)VisSomOrange: au smBlue: controlAudVisSomAud0(B) (C) amplitude(% signal change)Response st. dev.(% signal change) 0 6 912 TRENDS in Cognitive Sciences Figure 1. Excessive Neural variability in adults with autism demonstrated across the visual (Vis), auditory (Aud), and somatosensory (Som) systems. (A) Meanhemodynamic response time-courses from a single subject with autism and a single control subject in an auditory experiment. Error bars, standard error across trials.(B) Mean response amplitudes, averaged across trials and across subjects in each group. (C) Standard deviations of response amplitudes across trials. Orange, autism; blue,control.