visual attention in autism families: ‘unaffected’ sibs...

Visual attention in autism families: ‘unaffected’sibs share atypical frontal activation

Matthew K. Belmonte,1,3 Marie Gomot2,3 and Simon Baron-Cohen3

1Department of Human Development, Cornell University, Ithaca, NY, USA; 2INSERM U619, Universite Francois-Rabelais, CHRU de Tours, France; 3Autism Research Centre, Department of Psychiatry, University of Cambridge, UK

Background: In addition to their more clinically evident abnormalities of social cognition, people withautism spectrum conditions (ASC) manifest perturbations of attention and sensory perception whichmay offer insights into the underlying neural abnormalities. Similar autistic traits in ASC relativeswithout a diagnosis suggest a continuity between clinically affected and unaffected familymembers. Methods: We applied fMRI in the context of a non-social task of visual attention in order todetermine whether this continuity persists at the level of brain physiology. Results: Both boys withASC and clinically unaffected brothers of people with ASC were impaired at a visual divided-attentiontask demanding conjunction of attributes from rapidly and simultaneously presented, spatially disjointstimuli and suppression of spatially intervening distractors. In addition, both groups in comparison tocontrols manifested atypical fronto-cerebellar activation as a function of distractor congruence, and thedegree of this frontal atypicality correlated with psychometric measures of autistic traits in ASC andsibs. Despite these resemblances between the ASC and sib groups, an exploratory, hypothesis-gener-ating analysis of correlations across brain regions revealed a decrease in overall functional correlationonly in the ASC group and not in the sibs. Conclusions: These results establish a neurophysiologicalcorrelate of familial susceptibility to ASC, and suggest that whilst abnormal time courses of frontalactivation may reflect processes permissive of autistic brain development, abnormal patterns of func-tional correlation across a wider array of brain regions may relate more closely to autism’s determi-nants. Keywords: Autism, broader autism phenotype, visual spatial attention, functional MRI, frontalcortex, cerebellum, intraparietal sulcus, cingulum, functional connectivity.

Autism is a behaviourally defined condition whosediagnosis depends on a ‘triad’ of deficits comprisingimpaired social interaction, impaired communica-tion, and restricted interests and repetitive behav-iours. Symptoms in thesedomains vary greatly acrossindividuals,makingautism the extremeof a spectrumor continuum of abnormalities whose milder end isnot necessarily distinct from variation in the generalpopulation (Constantino & Todd, 2003). This contin-uous variation often manifests in autism families asthe Broader Autism Phenotype, a subclinical patternof cognitive and behavioural traits involving mildsocial and communicative deficits (Piven, Palmer,Jacobi, Childress, & Arndt, 1997; Baron-Cohen &Hammer, 1997) and associated with a cognitive stylebiased towards local details (Happe, Briskman, &Frith, 2001; Baron-Cohen &Hammer, 1997). Despitemany behavioural and psychometric results relatingASC and the Broader Autism Phenotype (reviewed inBelmonte et al., 2004b), comparatively little attentionhas been given to physiological investigations of thesefamilial patterns.

Three studies have uncovered tantalising abnor-malities in the neural bases of social perception infirst-degree relatives: The brains of parents ofautistic children seem less specialised for face per-

ception, lacking the normal right-lateralised aug-mentation of the N170 event-related potential tofaces, and lacking the normal correlation betweenN170 amplitude and face expertise (Dawson et al.,2005). Parents also hyperactivate frontal cortexduring the Reading the Mind in the Eyes Test (Baron-Cohen et al., 2006b) – perhaps as a compensatorystrategy for this perceptual difference. Likewise,clinically unaffected sibs of children with autismhave abnormally small amygdalae, gaze at eyesabnormally infrequently, and activate the rightfusiform face area less than controls do (Dalton,Nacewicz, Alexander, & Davidson, 2007). Dawson etal. (2005, p. 693) have suggested that such socialperceptual impairments may develop from a con-vergence of abnormalities of social motivation andreward on the one hand, and fundamental abnor-malities of perception on the other. Indeed, early-developing, non-social perceptual abnormalities,such as sensory-seeking behaviours and longlatency to disengage attention, combine with earlysocial abnormalities in predicting later diagnosis ofautism (Zwaigenbaum et al., 2005). It thus seemswarranted to examine the neural bases of non-socialas well as social traits as potential markers offamilial autism susceptibility.

One possible unifying theme between social andnon-social domains is the notion of selective deficitsConflict of interest statement: No conflicts declared.

Journal of Child Psychology and Psychiatry 51:3 (2010), pp 259–276 doi:10.1111/j.1469-7610.2009.02153.x

� 2009 The AuthorsJournal compilation � 2009 Association for Child and Adolescent Mental Health.Published by Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA

in ‘complex’ processing. Within the domain of visualperception, autistic impairments at detectingcoherent motion (Milne, Swettenham, & Campbell,2005) and second-order contrast (Bertone, Mottron,Jelenic, & Faubert, 2005) have led to the hypothesisthat first-order perceptual computations may takeprecedence over more complex perceptual process-ing (Bertone, Mottron, Jelenic, & Faubert, 2003),analogously to the autistic profile of precedence oflocal over complex information processing in higher-order cognitive domains (Minshew, Goldstein, &Siegel, 1997). In both of these cases, the cognitiveand the perceptual, ‘complexity’ is somewhatdomain-specific in its particulars (Minshew et al.,1997, p. 312) but can be understood in general as arequirement for rapid and integrative processing ofinformation from multiple inputs (Minshew et al.,1997, p. 313).

One method of quantifying this integrativedemand draws on the cognitive theory of relationalcomplexity. Simply put, the relational complexity of acognitive or perceptual operation is the number ofinteracting variables (the integrative aspect of Min-shew’s sense of complexity) that must be representedsimultaneously (the rapid aspect) in the process ofperforming the operation (Halford, Wilson, & Phil-lips, 1998). For instance, evaluating second-order(textural or multiplicative) contrast requires theintegration of local luminance contrasts over aninterval of spatial positions and thus depends on theluminance · position interaction, whereas first-order(luminance or additive) contrast can be assayed froma single local contrast signal. Although the theory ofrelational complexity has found most of its applica-tion at more abstract, cognitive levels of processing(e.g., in characterising the complexity of false-belieftasks (Andrews, Halford, Bunch, Bowden, & Jones,2003)), as a computational description it can beapplied equally well in the domain of perceptualintegration. This notion of complexity-specific defi-cits across domains relates well to ideas of atypicalneural connectivity in autism (Brock, Brown, Bou-cher, & Rippon, 2002; Belmonte et al., 2004a; Just,Cherkassky, Keller, & Minshew, 2004; Courchesne& Pierce, 2005).

A recent elaboration of the Enhanced PerceptualFunction theory of autism (Mottron, Dawson, Sou-lieres, Hubert, & Burack, 2006) cites abnormallyhigh salience at high levels of specialisation for thecentral, high-magnification portion of the visual field(consistent with behavioural and physiological find-ings of a narrow, ‘spotlight’ distribution of spatialattention in autism (Townsend & Courchesne,1994)), and at low levels of the hierarchy of visualprocessing abstraction (consistent with behaviouralfindings of a precedence of a bias for enhanced localprocessing over intact global processing (Plaisted,Swettenham, & Rees, 1999; Iarocci, Burack, Shore,Mottron, & Enns, 2006)). This combination of biasestowards central-field locations and towards low

levels of hierarchical processing predicts impairedresponse to peripherally located (lower salience)targets in conflict with centrally located (high sal-ience) distractors, especially when such conflictoccurs in low-level (high salience) properties such asorientation and colour.

Distracting stimuli can exert an unusually largeeffect on reaction time in people with autism (Bu-rack, 1994), and fMRI suggests that distractor sup-pression in autism may evoke atypically strongactivation in posterior intraparietal sulcus (Belmonte& Yurgelun-Todd, 2003b), a region also associatedwith suppression of visual distractors in typicalindividuals (Wojciulik & Kanwisher, 1999; Belmonte& Yurgelun-Todd, 2003a; Serences et al., 2005)). Weapplied behavioural and fMRI measures to cha-racterise the event-related dynamics of distractorsuppression in ASC, to discern whether the patternof subclinical features in family members mightextend to this non-social domain of visual percep-tion, to ascertain whether any such behaviouraldistinction might be associated with differences inneurophysiology, and to determine whether thedegree of correlation of functional activationamongst brain regions might differentiate familymembers with and without ASC. Boys with ASC,clinically unaffected brothers of people with ASC,and unrelated, typical controls performed a taskdemanding selective attention to colour and orien-tation in briefly displayed sine-wave gratings in dis-joint, peripheral locations, in the presence ofspatially intervening, centrally located distractorsthat were either congruent or incongruent to theseattended stimuli (Figure 1).

Methods

Participants

The study was approved by the Cambridge LocalResearch Ethics Committee. Informed assent wasobtained from each participant, and informed consentfrom each participant’s parent. Members of the ASCgroup, and the ASC sibs of members of the sib group,were identified from the Autism Research Centre vol-unteer pool, a database of individuals with pre-existingdiagnoses of autism or Asperger syndrome made

Figure 1 Examples of incongruent (left) and congruent(right) stimuli. The middle position in each 3 · 3hemifield was attended; all other positions weredistractors. Each stimulus array was presented for 167ms. Participants were asked to report a conjunction oforientation (with ignored colour) in one target andcolour (with ignored orientation) in the other

260 Matthew K. Belmonte, Marie Gomot, and Simon Baron-Cohen

� 2009 The AuthorsJournal compilation � 2009 Association for Child and Adolescent Mental Health.

according to DSM-IV-TR (American Psychiatric Associ-ation, 2000) criteria. The volunteer pool is restricted todiagnoses made by qualified clinicians at recognisedcentres. Controls were recruited from local schools.General developmental delay and other neurological orneuropsychiatric conditions were excluded, as was col-our-blindness. All participants had normal vision. IQwas assessed by the Wechsler Abbreviated Scale ofIntelligence (WASI), except for two participants in the sibgroup who refused IQ testing. For members of the ASCgroup, the Autism Diagnostic Interview – Revised (ADI-R; Lord, Rutter, & Le Couteur, 1994) was conductedwith one or both parents by a qualified rater (M.K.B.)1 Inaddition, all participants completed the adolescentversion of the Autism Spectrum Quotient (AQ; Baron-Cohen et al., 2001; Baron-Cohen, Hoekstra, Knick-meyer, & Wheelwright, 2006a), a validated screeningtool (Woodbury-Smith, Robinson, Wheelwright, &Baron-Cohen, 2005) which ruled out autism spectrumconditions in the sib group and the control group andconfirmed clinical and ADI-R diagnoses in the ASCgroup. Although there was no formal clinical evaluation,in no case did anymember of the sib group or the controlgroup present a clinical impression of ASC.

Participants were excluded if they did not meet ADI-Rdiagnostic threshold or if they scored less than 32 onthe AQ (in the ASC group), or if they were unable toperform the behavioural task or to tolerate the scannerenvironment, or if they moved excessively during thescan. One participant – the monozygotic twin of anincluded participant (Belmonte & Carper, 2006) – didnot meet ADI-R social and communicative criteria, andone participant met ADI-R criteria by one point but didnot meet the AQ criterion. One participant with ASCand two sibs were unable to perform the behaviouraltask. Three children with ASC, one sib, and two typicalcontrols were unable to tolerate the scanner. Two con-trols moved excessively during the scan. After theseexclusions, group sizes were 8 ASC, 7 sib, and 9 con-trols (Table 1). One sib was the brother of a member ofthe ASC group; the other sibs were unrelated to the ASCchildren enrolled in this study.

Task

The task required combining attention to location, col-our, and orientation, and was purposefully chosen to bedifficult since the interest was in maximising perceptualcomplexity and cognitive demand rather than perfor-mance. In order to avoid potential difficulties imple-menting complex verbal instructions, verbalexplanations were supplemented by visual illustrationsof target and non-target stimuli along with physical

rehearsals of target and non-target responses. Beforeentering the scanner, participants practised the task ona laptop computer until behavioural performance waslevel from one block of trials to the next, and until theyfelt comfortable with the task.

Each stimulus was a 3 · 6 array of oriented, colouredsine-wave gratings in Gabor patches, presented for167 ms. Gratings had a spatial frequency of 4.4/�, werecoloured either yellow-green (hue 50 in colour space) oryellow-orange (hue 95), and were oriented either 22.5�or 112.5� away from the horizontal (‘diagonal-up’ or‘diagonal-down’). Arrays were 5.5� wide and 2.75� tall,centred on a fixation cross 0.3� in width and height.Participants were instructed to attend simultaneouslyto the central grating in the 3 · 3 segment of the arrayin the left hemifield and the central grating in the 3 · 3segment in the right hemifield.

Imaging took place in two runs, separated by a briefrest break during which the participant remainedimmobile in the scanner. In the colour-orientation run,participants were instructed to press a button with theright index finger if the attended grating on the left wasorange and the attended grating on the right was diag-onal-up, and with the right middle finger otherwise.In the orientation-colour run, these criteria werereversed: participants pressed with the index finger ifthe grating on the left was diagonal-up and the gratingon the right was orange, and with the middle fingerotherwise. Each of the two runs comprised 166 stimu-lus arrays presented at 3 s intervals. The elements ofeach array were randomly chosen conjunctions of oneof the two colours and one of the two orientations; thesefour possible combinations yielded a 25% proportion oftargets. The order of the two runs was counter-balancedacross participants.

In the congruent condition, the two attended stimuliwere identical to each other and to the distractors in thespatially intervening, innermost column of each hemi-field, whilst the 10 outer distractors remained randomlychosen. In the incongruent condition, the attendedstimuli were uncorrelated with each other and with thedistractors. Thus the congruent condition could beperceptually represented by collapsing colour and ori-entation across spatial position, whereas the incon-gruent condition had greater relational complexitybecause it required the simultaneous maintenance ofthese three interacting variables (colour, orientation,position). Proportions of targets and non-targets werethe same in the congruent as in the incongruent con-dition. Participants were instructed to emphasiseaccuracy rather than speed, to slow themselves down soas to avoid impulsive responding, and to feel free to takeas much of the 3 s response interval as they needed.

Autism is characterised by an abnormally strongdependence of behavioural performance on attentionaland executive set: results differ markedly depending onwhether instruction and expectation bias towards astrategy of locking attention onto localised features ordistributing attention globally across stimuli (Plaistedet al., 1999; Iarocci et al., 2006), and trial-to-trialchanges from narrow to broad attentional scope exact aheavy price in terms of behavioural performance (Mann& Walker, 2003). Because of this consideration ofattentional set, our design goal was to establish abehavioural context in which the majority of stimuli

1 In order to respect the copyright on the ADI-R, the following

procedure was implemented: During the interview, questions

were read from one legally obtained, printed copy of the

research version of the ADI-R, which was used for all admin-

istrations, and notes and codings were typed into a laptop

computer. After the interview the resulting computer file,

which contained only the rater’s notes and codings and not the

text of the ADI-R, was scored automatically by software

developed by the rater. This software implements the ADI-R

scoring algorithm (which is not patented) but incorporates

none of the text of the ADI-R (which is copyrighted).

Visual attention in autism families 261


demanded distribution of attention across locationsand features. Each run therefore contained 34 con-gruent events embedded in a background of 132incongruent events, with the constraint that no twocongruent events were separated by fewer than twoincongruent events. The incongruent events thusestablished an attention-demanding baseline response,against which the congruent events were negativelycontrasted. Because participants had difficulty tolerat-ing long sessions in the scanner, and because ‘rest’ is inany case a poorly controlled task condition when itcomes to comparisons between autistic and non-autis-tic individuals (Kennedy, Redcay, & Courchesne, 2006),no rest condition was included.

Functional imaging

Echo-planar images were acquired on a 3 T whole-bodysystem consisting of a Bruker Medspec 30/100 spec-trometer (Ettlingen, Germany) and a 910 mm borewhole-body actively shielded magnet (Oxford MagnetTechnology, Oxford, UK), in 21 axial slices parallel tothe AC-PC plane, with TR 1.1 s, TE 27.5 ms, flip angle65.5�. Slices were 4 mm thick with a 1 mm gap.

A 20 cm field of view in a 64 · 64 matrix yielded an in-plane resolution of 3.125 mm. Imaging began 28.65 sbefore the onset of the first stimulus, and continued23.35 s after the onset of the last stimulus.

Functional images were corrected for head motion byregistering the entire image series from both runs to thelast image of the first run, using decoupled automatedrotational and translational motion correction (Maas,Frederick, & Renshaw, 1997), a method that uses ak-space representation of the images to separate rota-tional and translational (k-space phase) components.fMRI time series were high-pass filtered by detrendingin a moving window of 26 points (28.6 s), and tempo-rally interpolated to correct for slice acquisition timing.

Statistical analysis

d¢ scores from the behavioural data were subjected to a3 · 2 · 2 (group · run · congruence) analysis of vari-ance, with post hoc t-tests.

Voxel time series were correlated with an idealhæmodynamic response to the incongruent events,computed by a balloon hæmodynamic model (Buxton,Wong, & Frank, 1998) with the following parameters:

Table 1 Psychometric and behavioural data. The table lists age, IQ, AQ, ADI-R scores (for the ASC group), and d¢ for incongruentand congruent conditions, with means and standard deviations for each group

Age (Years)

WASI

AQ

ADI-R d¢

VIQ PIQ Social Communication Repetitive Behaviours Incongruent Congruent

Normal13.09 118 112 20 1.91 2.2115.45 129 120 16 2.16 2.5512.75 119 118 4 1.86 1.6512.67 134 99 2 1.12 1.3614.41 133 121 8 0.58 0.9114.89 123 126 15 1.60 2.1014.56 107 119 2 1.97 2.0614.63 136 111 24 2.57 1.9112.70 125 131 26 1.75 2.08

13.91 125 117 13.0 1.72 1.871.09 9 9 9.4 0.62 0.49

Sib11.93 121 129 7 1.34 1.9115.81 121 103 7 0.91 0.2914.05 123 119 6 1.75 2.2311.24 120 99 4 0.67 1.1611.40 3 1.04 1.5311.46 136 131 23 1.58 2.4613.57 4 1.12 1.64

12.78 124 116 7.7 1.20 1.601.74 7 15 6.9 0.38 0.72

ASC13.45 106 88 41 15 15 6 0.40 0.5215.55 126 93 43 26 17 8 1.00 1.4612.47 105 100 46 26 24 11 1.23 1.3311.44 109 105 39 29 24 10 1.01 1.6314.13 113 120 46 24 26 10 1.25 1.3514.68 88 93 32 20 14 4 1.50 1.6815.11 143 119 37 26 21 6 1.53 1.4310.89 147 141 36 19 11 6 1.81 1.58

13.47 117 107 40.0 23.1 19.0 7.6 1.22 1.371.72 20 18 5.0 4.7 5.5 2.5 0.43 0.37



mean venous transit time at rest s0 = 2s, resting oxygenextraction fraction E0 = 0.4, resting blood volumeV0 = 0.01, maximum normalised inflow rate fmax = 1.7,outflow exponent fa = 0.5, outflow linear coefficient fb= 1.866. Inputs to this model were set with rise time aninvariant 4 s, sustain time equal to the behaviouralresponse latency on each trial, and instantaneous falltime. These input parameters to the balloon modelyielded a smooth output peaking near 5.5 s anddecaying near zero within 9 s (a curve similar to astandard c variate, but based on an explicit physicalmodel and accounting for trial-to-trial variation inresponse latency). Overlapping responses were sum-med. Only the events associated with correct responseswere included in this analysis; intervals containinghæmodynamic responses associated with incorrectbehavioural responses were specifically excluded fromthe correlation computation.2

The resulting correlation images were transformed toTalairach space (Talairach & Tournoux, 1988) withinterpolation to 1 mm2 voxels, and convolved with a6.25 mmFWHMGaussian kernel. To identify structuresactivated within each of the three participant groupsconsidered separately, a voxelwise one-sample t-test wasthresholded at a two-tailed probability of .01 and acluster size of 470 ll.3 Activated structures were labelledby hand with reference to automated labellings gener-ated by the TalairachDæmon (Lancaster et al., 2000). Toidentify structures differentially activated in the threegroups, between-groups t testswereappliedonly to thosevoxels significantly activated or deactivated in any of thethree within-group tests. These t images were threshol-ded at a two-tailed probability of .05.

To explore activations across time as well as acrossspace, average fMRI responses were computed for eachparticipant. At each voxel in the Talairach-transformedimage, and for incongruent and congruent events sep-arately, the 17 scans (18.7 s) following an event wereaveraged temporally across events and spatially over aradius of 3.125 mm, and normalised as percent signalchange from time zero.4 The congruent average wassubtracted from the incongruent average to form a dif-ference average consisting only of the BOLD effectassociated with decreased perceptual congruence. (Thissubtraction method is analogous to the computation ofa ‘difference wave’ in the context of event-relatedpotentials.) To reduce noise, each of these single-participant averages underwent two iterations of three-point smoothing. Pointwise t-tests compared thesedifference averages across groups, and grand averageswere computed for each of the three participant groups.

In any clinical study, confounds of diagnostic groupwith behavioural performance can be difficult to avoid,and the directionality of the relationship betweenbehaviour and physiology can be unclear. Even ourexclusion of trials with incorrect behavioural responsesstill allows some possibility that observed group differ-ences in physiology might have been induced by dif-ferences in behavioural performance. To evaluate thispossibility, an index of the ‘abnormality’ of BOLDresponse was computed, and regressed against d¢score: First, individual fMRI time series within eachregion of interest were correlated against the groupgrand average within that region, for each individual ineach of the ASC and sib groups. That is to say, eachASC time series was correlated against the ASC grandaverage, and each sib was correlated against the sibgrand average. In addition, these same individualaverages in the ASC and sib groups were correlatedagainst the grand average from the normal group.

After Fisher’s z transformation, the individual’s cor-relation against the normal time series was subtractedfrom the individual’s correlation against their owngroup’s time series, yielding a normally distributedmeasure of where each individual lay on a continuumbetween identity with the normal time series and iden-tity with their own group’s abnormal (ASC or sib) timeseries. These abnormality indices were then regressedagainst d¢ scores from the incongruent condition, usingeither linear or quadratic regression, whichever madethe best fit in each case. Within the sib and ASC groups,a positive effect of d¢ on the abnormality index wouldindicate that those individuals who were most drivingthe observed abnormal group time series were thosewith the most normal behaviour, whereas a negativeeffect would indicate that the abnormal physiologicaleffect were associated with abnormal behaviour, andthe absence of any measurable effect would suggestthat degrees of physiological and behavioural abnor-mality were not strongly associated.

For an exploratory, hypothesis-generating evaluationof functional correlation between brain regions, a set of38 bilateral (76 total) regions of interest (Table 2) wasidentified by selecting all regions activated in any of thegroup t images, as well as several regions activated justbelow the cluster-size threshold. Correlation matricesfor averaged time series in all these regions were com-puted within each participant separately, and con-verted to z¢ scores by Fisher’s transformation. z¢ scoreswere compared using t tests within each region pair inthe correlation matrix, both within and between groups.In addition, z¢ scores pooled across region pairs werecompared between groups using an analysis of varianceand post hoc t-tests, with degrees of freedom adjusted toreflect the number of regions (that is, the number ofindependent observations) rather than the number ofregion pairs.

To relate levels of autistic traits to levels of functionalactivation within brain regions, the normality index(z-transformed correlation with the normal groupaverage fMRI time series, as above) was modelled as afunction of AQ score in each of four bilateral, distinctregions of interest (see Table 2) along the length of eachhemisphere’s middle frontal gyrus, from Brodmannarea 10 near area 46, to area 9 near area 8 (SAS PROCMIXED with fixed effects of AQ, group, hemisphere and

2 In every participant sufficient trials for averaging remained

after exclusion of incorrect responses – see the behavioural

accuracy figures.3 According to an empirical null distribution of maximum

cluster sizes computed by t-test over 7 independent null data

sets with the identical 6.25 mm FWHM blurring, these voxel-

probability and cluster-size constraints implemented a cluster-

level tail probability of .05.4 Though 18.7 s is a much longer averaging epoch than is

typical in event related fMRI studies, this interval was neces-

sary in order to capture the longer-lasting, secondary modu-

lations induced in the two experimental groups.



region, and random effect of participant). Within theASC group, ADI-R social, communicative and repetitivebehaviours subscores each were similarly modelled(SAS PROC MIXED with fixed effects of ADI-R, hemi-sphere and region, and random effect of participant).

To relate levels of autistic traits to levels of functionalcorrelation between brain regions, z¢ for functional cor-relations (computed as above), averaged within eachparticipant across all pairs of regions, wasmodelled as afunction of d¢ from the behavioural task and as a func-tion of AQ score (SAS PROC MIXED with fixed effects ofd¢ or AQ and group, and random effect of participant). Inthe ASC group, z¢ was also modelled as a function ofADI-R social, communicative, and repetitive behavioursubscores (SAS PROCMIXED with fixed effects of ADI-Rsubscore and random effect of participant).

Results

Behaviour

Despite an instruction to focus on accuracy ratherthan speed, mean response time for the ASC groupwas slightly but significantly shorter than those forthe typical and sib groups: 753 ± 34 ms versus889 ± 35 ms and 918 ± 52 ms (mean ± S.E.M.),respectively (F(2,84) = 4.40, p = .0152). Accuracy,the behavioural variable of interest, showed a pat-tern of differences distinct from that of the reactiontimes (Figure 2): d¢ scores differed significantly bydiagnosis (F(2,84) = 5.99, p = .0037), being lowerthan normal in the sib group (t(62) = 2.36, p =

.0217), and lowest in the ASC group (t(66) = 3.42,p = .0011). (In terms of the proportion of correctlyclassified stimuli, accuracy for incongruent andcongruent conditions, respectively, was 80% and84% in the normal group, 71% and 76% in the sibgroup, and 68% and 74% in the ASC group.) Thus, interms of this behavioural measure, the sibs’response partially mimicked that of ASC.

Table 2 Regions of interest for the exploratory analysis ofcorrelation (Talairach coordinates). Abbreviations: L left, Rright, r rostral, d dorsal, a anterior, p posterior, l lateral, mmedial, OFC orbitofrontal cortex, ACC anterior cingulate cor-tex, Cb cerebellum, Thal thalamus, Ins insula, IFG inferiorfrontal gyrus, MFG middle frontal gyrus, MedFG medial frontalgyrus, PreCG precentral gyrus, PhippG parahippocampalgyrus, Hipp hippocampal formation, STG superior temporalgyrus, Cu cuneus, PCu precuneus, FusG fusiform gyrus, IPsintraparietal sulcus, SMG supramarginal gyrus, AG angulargyrus, PCC posterior cingulate cortex, SPL superior parietallobule, IPL inferior parietal lobule, LingG lingual gyrus, IPTOposterior intraparietal sulcus near transverse occipital sulcus.

Region x y z

LOFC )8 58 0ROFC 8 58 0LrACC )5 14 19RrACC 5 14 19LalCb )31 )53 )41RalCb 31 )53 )41LamCb )14 )37 )31RamCb 14 )37 )31LPyramis )16 )70 )20RPyramis 16 )70 )20LDeclive )4 )63 )18RDeclive 4 )63 )18LmdThal )7 )19 0RmdThal 7 )19 0LaIns )42 14 2RaIns 42 14 2LIFG )46 23 7RIFG 46 23 7LMFG1 )25 43 35RMFG1 25 43 35LMFG2 )33 31 21RMFG2 33 31 21LMFG3 )46 19 31RMFG3 46 19 31LMFG4 )39 22 44RMFG4 39 22 44LMedFG )7 14 44RMedFG 7 14 44LdACC )4 )2 50RdACC 4 )2 50LPreCG )35 )11 34RPreCG 35 )11 34LPhippG )32 )30 )3RPhippG 32 )30 )3LHipp )33 )12 )14RHipp 33 )12 )14LpSTG )51 )43 19RpSTG 51 )43 19LmCu )2 )80 34RmCu 2 )80 34LPCu7 )7 )46 37RPCu7 7 )46 37LPCu19 )29 )65 23RPCu19 29 )65 23LlFusG )40 )68 )9RlFusG 40 )68 )9LaIPs )29 )53 40RaIPs 29 )53 40LSMG )40 )43 28RSMG 40 )43 28LAG )37 )66 30RAG 37 )66 30LaSTG )51 4 1RaSTG 51 4 1LPCC )7 )24 20

Table 2 (Continued)

Region x y z

RPCC 7 )24 20LpostCu )6 )82 12RpostCu 6 )82 12LlSPL )25 )60 45RlSPL 25 )60 45LpSPL )13 )66 50RpSPL 13 )66 50LLingG )4 )73 0RLingG 4 )73 0LmFusG )25 )71 )13RmFusG 25 )71 )13LSTG42 )56 )38 2RSTG42 56 )38 2LmSPL )8 )66 32RmSPL 8 )66 32LIPTO )36 )75 31RIPTO 36 )75 31LIPL )30 )30 31RIPL 30 )30 31LlCu )7 )65 33RlCu 7 )65 33



In order to ensure that these effects were not beingdriven by possible outliers, the analyses were re-runexcluding the two participants in the sib group withthe largest VIQ–PIQ discrepancies and the partici-pant in the ASC group with the lowest PIQ and thelowest d¢. Response time differences were again sig-nificant but slight (799 ± 29 ms versus 889 ± 35 msand 1021 ± 50 ms, F(2,72) = 6.76, p = .002). d¢ againdiffered significantly by diagnosis (F(2,72) = 3.86,p = .0255), but with this exclusion of outliers fromthe sib group the d¢ difference was driven only by theASC group (t(62) = 2.65, p = .0102). In addition, withthis exclusion of outliers the expected main effect ofcongruence appeared, F(1,72) = 4.53, p = .0367,with d¢ greater in the congruent condition than in theincongruent condition.

Functional mapping: typical controls

Details of event-related activation foci within eachgroup and in contrasts between groups are pre-sented in Table 3, and representative group imagesin Figure 3. In controls, low congruence activated adistributed network of regions including an atten-tion-related region in left anterior lateral cerebellum(Allen, Buxton, Wong, & Courchesne, 1997),5 bilat-eral middle frontal gyri, dorsal anterior cingulum,right anterior intraparietal sulcus, right supramar-ginal gyrus, and left superior parietal lobule. Alsoactivated were left mediodorsal thalamus, left para-hippocampal gyrus, right lateral cuneus, rostralanterior cingulum, left anterior insula, and a loca-tion in left medial fusiform gyrus consistent with thelocation of the V8 colour region (Hadjikhani, Liu,Dale, Cavanagh, & Tootell, 1998).

Functional mapping: ASC

Comparing regions of statistically significant activa-tion (see Table 3 for t values and cluster extents), theASC group was notable for atypical frontal andsuperior parietal activations including a lack ofactivation in dorsal anterior cingulum and lack ofsignificant activation in left anterior insula, leftmiddle frontal gyrus, and superior parietal lobe. Inaddition, activations arising within the autism groupbut not significant in the between-groups compari-son included extensive cerebellar activation centredin the left pyramis, activation of orbitofrontal cortex,right posterior superior temporal gyrus, and (aspredicted) the posterior extent of right intraparietalsulcus near its junction with the transverse occipitalsulcus, a localisation consistent with the represen-tation of parafoveal space along the medial wall of

the intraparietal sulcus (between foveal space at thefundus and far peripheral space on the adjacentgyrus) within area V7 (Swisher, Halko, Merabet,McMains, & Somers, 2007).

Functional mapping: sibs

The sib group was remarkable for what seemed anoverall lack of significant activations – at least incorrelation with our activation model validated in anormal pilot group. The only regions significantlydifferentially active in sibs in this contrast betweenhigh and low congruence were rostral anterior cin-gulum and also the retrosplenial portion of posteriorcingulum, a region associated with establishingspatial expectancy (Small et al., 2003).

Time courses: primary phase

Examination of time courses for all regions activatedin any of the groups revealed several frontal, cere-bellar, and occipitoparietal regions (Figure 4) wherecongruence in the current event affected BOLDresponses during the events immediately sub-sequent, in the ASC and sib groups. The resultingabnormally prolonged BOLD responses thereforewere not detected by our model based on normal pilot

Inco

ngru

ent

Con

grue

nt

0

0.5

1

1.5

2

2.5

3

autismsibnormal

d’

Figure 2 d¢ scores as a function of diagnostic groupand stimulus congruence. For each participant sepa-rately, the d¢ scores for incongruent stimuli and forcongruent stimuli are joined by a line. Despite a largeamount of variance within groups, both the sib groupand the ASC group differed significantly from controls

5 In 2 controls, 3 participants with ASC, and 3 sibs the cere-

bellar attention area lay just beyond the imaged volume and

could not be evaluated. The result remains statistically sig-

nificant, though, even given this reduced number of observa-

tions.



Table

3Talairach

coord

inatesofactivationsin

allregions,within

and

betw

een

participantgroups.Within-group

activationsare

p<0.01

ateach

voxel,

inclusters

ofatleast470

ll(equivalentto

12voxels

atth

escanresolution).Effects

ofbetw

een-groupscomparisonare

p<0.05ateachvoxel,withcomparisonsrestrictedto

regionsactivatedwithin

groups.Eacht

valueis

anavera

geoverth

eentire

cluster

Structu

reBA

Norm

al

Sib

Autism

Autism

–Norm

al

Sib

–Norm

al

Sib

–Autism

xy

zVol.

� tx

yz

Vol.

� tx

yz

Vol.

� tx

yz

Vol.

� tx

yz

Vol.

� tx

yz

Vol.

� t

OFC

10

16

58

)2

901

+5.05

14

60

)2

567

)2.84

rACC

24,32

317

24

500

+3.96

414

28

482

+4.93

612

29

130

+3.01

LalC

b)28

)51

)36

1342

+3.79

RamCb

19

)41

)30

483

+4.06

LPyra

mis

)16

)72

)25

621

+4.10

)10

)50

)36

117

)2.87

)18

)74

)26

160

)2.79

LmdThal

)9

)19

3970

+3.96

)10

)17

4508

)2.79

LaIns

13

)37

13

2548

+4.09

)39

13

389

)2.46

)38

11

026

)2.04

LMFG

9)39

28

27

2078

+3.95

)42

23

32

661

)2.73

)37

29

31

10

)2.55

RMFG

927

39

36

1986

+3.86

36

34

33

2217

+4.30

28

16

24

122

)2.56

26

38

37

610

)2.22

44

124

297

)2.76

MedFG

6)1

951

525

+4.51

)8

14

46

85

)2.15

dACC

24,32

)2

148

1267

+4.02

)3

)1

50

508

)2.58

)4

248

44

)2.00

LPhippG

30

)32

)34

)2

945

+4.17

)34

)33

)5

770

)3.03

RpSTG

22

50

)43

21

625

+4.34

53

)44

20

95

)2.19

LlFusG

19

)41

)63

)23

1175

+4.45

)44

)66

)25

912

+3.23

)42

)66

)25

322

)2.65

RaIPs

727

)53

43

1024

+3.93

28

)56

42

245

)2.46

19

)54

38

10

)2.79

31

)41

39

125

)2.84

RSMG

40

44

)49

29

1188

+4.14

40

)43

28

57

)2.51

30

)50

47

353

)2.64

PCC

29,30

3)26

19

582

+4.80

3)26

19

440

+2.81

2)26

21

220

+2.58

LlSPL

7)20

)64

42

1923

+3.97

)9

)66

35

134

)2.01

)12

)65

50

30

)2.07

LmFusG

19

)26

)72

)14

471

+4.36

)25

)75

)12

32

)2.38

RIPTO

19

30

)66

28

2181

+4.22

30

)65

12

479

)3.43

RlC

u19

9)66

35

929

+4.22

7)66

35

529

)2.72

21

)65

26

76

)2.59



data. In general, these responses occurred in twophases, a primary phase from 0 to 7 s and a sec-ondary and overlapping phase from 7 s to about 18 s.

In the primary phase, the typical group tended toactivate dorsal anterior cingulum more than the ASCgroup (t(15) = 2.04, p = .059), and amongst the sibstherewas considerable variationwith no clear pattern(no significant contrasts between sibs and ASC norbetween sibs and normal). In contrast, all threegroups activated bilateral middle frontal gyri. Inter-estingly, in the cerebellar attention area sibs activatedmore strongly (t(9) = 2.50, p = .034), whilst ASC didnot differ significantly from controls. Although thesecingulate and cerebellar activations distinguished theASC and sib groups from controls, occipitoparietalregionsweremore variable across participants andnosignificant group differences were found.

After exclusion of the aforementioned two sib andone ASC behavioural outliers, re-running this anal-ysis maintained or augmented all the results: fordorsal anterior cingulum, controls > ASC,t(14) = 2.65, p = .019 and also sibs > controls,t(12) = 2.26, p = .044 and sibs > ASC, t(10) = 4.10,p = .0022; for the cerebellar attention area, sibs >

controls, t(8) = 3.16, p = .0133 and also controls >

ASC, t(9) = 2.31, p = .047.

Time courses: secondary phase

The secondary phase extended well past the usualdecay period of hæmodynamic response to a singletrial – that is, this phase consisted of alteredresponses not to the event of interest, but rather toevents immediately succeeding the event of interest:the manipulation of congruence within the initialevent induced a longer-lasting change in the brainresponse to subsequent events. This unusualresponse arose in ASC in the cerebellar attentionarea (t(10) = 3.98, p = .0026). The sibs manifested asimilar pattern of prolonged activations, with astronger effect that was significant in a larger set ofregions. Sibs activated more strongly than controlsin the cerebellar attention area (t(9) = 6.25,p = .00015), left (t(14) = 3.18, p = .0067) and right(t(14) = 2.25, p = .041) middle frontal gyri, and rightposterior intraparietal sulcus (t(14) = 3.75,p = .0022). In all these regions, the response in sibswas even greater than that in ASC.

Figure 3 Activation images in within-group averaged echo-planar slices, illustrating group differences in patterns ofactivation. See Table 3 for contrasts. Left and right are true left and right



Again after exclusion of the behavioural outliers,these results were maintained. The ASC group againactivated the cerebellar attention area more thannormal, t(9) = 3.13, p = .012. In the sibs, contrastswere significant for greater activations in the cere-bellar attention area (t(8) = 7.51, p = .000069), leftmiddle frontal gyrus (t(12) = 3.73, p = .0029) with alike trend in right middle frontal gyrus (t(12) = 1.99,p = .070), and right posterior intraparietal sulcus(t(12) = 3.39, p = .0054).

Relation of behavioural and physiologicaldifferences

Sibs showed a strong quadratic relationship betweend¢ and atypicality of the fMRI time series in bilteralmiddle frontal gyri (t(4) = +3.43 and p = .0264, samevalues in both gyri separately), as well as a trendtowards a similar quadratic relationship in rightposterior intraparietal sulcus (t(4) = +2.41, p =

.0733). The ASC group showed a quadratic rela-tionship in dorsal anterior cingulum (t(5) = +3.95,p = .0108) and a borderline negative linear relation-

ship in right posterior intraparietal sulcus (t(6) =)2.42, p = .0515) (Figure 5).

Functional correlations between regions

Though no between-groups comparisons of func-tional correlations within individual region pairssurvived Bonferroni correction for multiple compari-sons, functional correlation pooled across all possiblepairs of regions of interest was significantly lower inASC than in the other two groups (omnibus F(2,1821) = 266.89, post hoc t(1291) = 20.71 betweenASC and controls, t(1139) = 19.84 between ASC andsibs, but t(1215) = ).72 between sibs and controls).

Exclusion of the behavioural outliers strengthenedthe large difference between ASC and the other twogroups (omnibus F(2, 1593) = 308.19, post hoc

t(1215) = 21.04 between ASC and controls,t(913) = 22.55 between ASC and sibs), and alsorevealed a comparatively slight yet statistically sig-nificant difference between sibs and controls(t(1065) = )4.53), in a direction opposite to that ofthe contrast between ASC and controls: that is, in

R MFGL MFGdACC

−0.2

−0.1

0

0.1

0.2

0.3

0.4

0.5 0.5

−0.3

−0.3

−0.2

−0.1

0.1

0.2

0.3

0.4

0.5 R IPTOR PCCL ant. lat. cerebellum

0

16 180 2 4 6 8 10 12 140 2 4 6 8 10 12 14 16 4 6 8 10 12 14 16 18 18 0 2seconds

sig

nal c

hang

e, in

cong

ruen

t − c

ongr

uent

Figure 4 fMRI time series for regions in which time series examinations revealed activation characteristics notcaptured by the hæmodynamic model. Dorsal anterior cingulate activation in sibs was highly variable across indi-viduals. Delayed activations in ASC and sibs arose in middle frontal gyrus, the cerebellar attention area, and pos-terior intraparietal sulcus. ASC activation of posterior cingulum, though subthreshold, was closer to that of sibs thancontrols. Time courses are plotted for controls (red), ASC (blue), and sibs (magenta). Bars represent standard error ofthe mean



this omnibus test in which co-activation data werecollapsed across all possible pairings of brainregions, the five sibs remaining after exclusion of twobehavioural outlier sibs manifested functionalco-activation slightly greater than controls’, unlikethe ASC probands in whom functional co-activationwas markedly less than controls’. (Despite thisincrease in overall co-activation across all pairs, theexclusion of outliers actually produced fewer indi-vidual pairs of regions that reached the uncorrectedsignificance threshold a = .05, owing to the decreasein the number of observations in the sib sample.)

Figure 6 presents graphically the results of com-parisons within region pairs, uncorrected for multi-ple comparisons and (for conservatism) withoutexclusion of outliers. Most evident in the controlgroup was a strong coupling between dorsal anteriorcingulum, cerebellum, superior parietal lobule, andlingual gyrus and other occipital regions. This groupalso manifested strong correlations within a visualprocessing network comprising fusiform gyrus,cuneus, precuneus, and posterior superior temporalgyrus, a stimulus evaluation network comprisingsuperior temporal gyrus, insula, and inferior frontalgyrus, and a response network comprising middlefrontal gyrus, dorsal anterior cingulum, and leftmedial and precentral gyri. The ASC group, in con-

trast, was typified by correlations between cuneus,precuneus, and middle frontal gyrus, and quies-cence of dorsal anterior cingulum. Significant cor-relations in ASC also existed between middle frontalgyrus and rostral anterior cingulum and orbitofron-tal cortex, regions involved in affective control. Thecomplete sib group activated a dorsal anteriorcingulum network similar to that of the controlgroup, albeit one with lesser correlation with middlefrontal gyrus and superior parietal lobule.

Relation to psychometric measures of autistic traits

In the model of normality of brain activation as afunction of AQ score, t scores for all middle frontalgyrus (MFG) regions in all participant groups –including the control group – were negative. Thisrelationship was significant in posterior MFG in ASC(left MFG region #4 Kenward-Roger t(45.7) = )2.09,p = .0421; right MFG region #3 t(45.7) = )2.22,p = .0315), and in anterior and posterior MFG in sibs(region #1 left and right t(56.6) = )3.56, p = .0008and t(56.6) = )2.00, p = .0502, respectively; region#3 left and right t(56.6) = )3.21, p = .0022 andt(56.6) = )2.11, p = .0392; region #4 left and rightt(56.6) = )2.48, p = .0160 and t(56.6) = )4.29,p < .0001). Similarly, in the ASC group t scores for

Figure 5 Relationship of behavioural accuracy to differences in fMRI time course. Each participant has one datumon the abscissa (d¢) and two on the ordinate (Fisher’s z-transforms of correlations with the participant’s own group’saverage fMRI time series (circles) and with the control group’s (triangles)). Regression curves are illustrated wheresignificant relationships between d¢ and zgroup ) znormal have been detected



all three ADI-R subscores in all MFG regions werenegative; ADI-R social and communicative sub-scores were strongly negatively associated with thenormality index in most regions of posterior MFG(left MFG region #3 social t(55) = )2.12, p = .0388,communicative t(55) = )2.07, p = .0432; rightMFG region #3 social and communicative botht(55) = )2.93, p = .0049; left MFG region #4 socialt(55) = )2.73, p = .0085, communicative t(55) =)2.92, p = .0051), and trended more weakly inanterior MFG (left MFG region #1 social t(55) =)2.18, p = .0335, communicative t(55) = )2.10,p = .0404; right MFG region #1 social t(55) = )1.99,p = .0511, communicative t(55) = )1.89, p = .0636).(See Table 2 for Talairach coordinates.) Therewere no significant associations with the ADI-Rrepetitive behaviours subscore, although in thiscase again all eight t scores (independent testsin four bilateral, distinct regions) were negative.

In the models of within-participant mean func-tional correlation as a function of behavioural d¢,AQ score and ADI-R subscores, the only significant -effect – and only marginally so – was that of AQ in theASC group (Kenward-Roger t(15.5) = )2.06,p = .0567). Functional correlation was unrelated

to any of these variables in the control or sibgroups, andunrelated tod¢ orADI-R in theASCgroup.

Discussion

Behaviourally, although the ASC group respondedmore quickly than either the controls or the sibs, theASC group was less accurate than the sibs, who inturn were less accurate than the controls. Thisfinding of autistic impairment in a spatial divided-attention task echoes psychophysical and electro-physiological observations of an atypical spatialdistribution of visual attention in autism (Townsend& Courchesne, 1994).

Physiologically, incongruence-related activationsin the controls occupied a widespread network offrontal, cerebellar, and parietal attention regions,whereas the ASC group activated a cerebellar regionoutside the attention area, did not phasically acti-vate frontal and parietal attention regions, but didactivate posterior visual regions and also orbito-frontal cortex. These findings appear to confirm alarge number of previous results from various cog-nitive tasks, suggesting hypoactivation of frontalcortices in autism and hyperactivation of posterior

Figure 6 Correlations of event-related response amongst 38 bilateral pairs of brain regions. Within-group and be-tween-groups comparisons significant at two-tailed a = 0.05 are denoted in colour (see the scale bar at top left). Thecomparison between ASC and the other two groups was statistically significant when data were pooled acrossregions, as reflected in the overall weakness of correlations in the ASC group (top right panel) versus strength in thecontrol and sib groups (top left two panels). Group comparisons within regions, however, have not been corrected formultiple comparisons and therefore are considered exploratory



cortices subserving lower levels of processing (Haist,Adamo, Westerfield, Courchesne, & Townsend,2005; Silk et al., 2006; Belmonte et al., 2004b), aswell as results on orbitofrontal activation possiblyrelated to arousal (Belmonte & Yurgelun-Todd,2003b), anterior cingulate hypoactivation (Mundy,2003; Gomot et al., 2006), and hypoactivation in thecerebellar attention region with hyperactivation ofother cerebellar regions (Allen & Courchesne, 2003;Allen, Muller, & Courchesne, 2004). In the sibs, incontrast, differential activations between the con-gruence conditions were limited to only two regions:rostral anterior cingulum and retrosplenial cortex.

These modelled activations, however, do not tellthe whole story. Atypical activation maps may arisebecause the activation really isn’t there, or becausethe activation is atypically timed and fails to onsetand/or to resolve within the modelled time interval.Examination of time courses revealed ASC andsib activations whose atypically delayed and pro-longed timing had prevented detection in the whole-brain analysis.6 That is, the ASC and sib groups didactivate fronto-cerebellar attention systems, butthese activations arose too late to be useful dur-ing behavioural response to the trial of interest,instead manifesting during the trials immediatelysubsequent.7

It is notable that sibs do not attain normative lev-els of performance: the difference in BOLD responsein a direction aligned with that of behavioural per-formance between the sib and ASC groups but stillopposite that of behavioural performance betweenthe sib and control groups is what suggests theoperation of a compensatory process. Also of interestis that these delayed activations are even greater inmagnitude in sibs than in ASC, suggesting that thesibs may more efficiently or completely implementthis putative compensatory process. This differencemay reflect either the absence in the sibs of some ofthe genetic or environmental factors that confer lia-bility to autism, or the presence of genetic or envi-ronmental factors that protect against autism.

These findings of differential timing form animportant counterpoint to the oft-repeated conten-tion that people with autism simply do not activatemany of the same brain regions activated in controls.This delayed response is consistent with delayedactivation in an evoked-potentials study of responseto peripheral visual stimuli (Townsend et al., 2001),and with studies of attention shifting showing that

impaired performance normalises when two to threeseconds are allowed in which to accomplish the shift(Townsend, Singer-Harris, & Courchesne, 1996). Itis also consistent with many clinical behaviouralobservations and self-reports highlighting theimportance of having explicit instruction and time toprepare responses to unpredictable stimuli.

It is important to note that the ASC group’s patternof prolonged response could not have been directlyassociated with motor activation, since the ASCgroup’s motor responses were faster than normal.Group differences in accuracy also are unlikely to bea causal factor in these physiological differences,since only the trials associated with correctbehavioural responses were included in the aver-ages. It is, of course, possible that group differencesin impulse control, distractibility and concentrationmay be seen as a confound in the current results, ifsuch differences are not viewed as part of the syn-drome of ASC. In any case, although the group dif-ferences in brain response that we have presentedmight permit stronger inferences in the context ofequal behavioural responses between groups, suchbehavioural equality is often not the case in studiesof autism and other clinical populations, and braindifferences may cause behavioural differences justas well as the converse.

The quadratic effect of d¢ on the abnormality offMRI time series in key prefrontal and posterior brainregions suggests that at least in the sibs, the group’satypical physiology was not driven solely by abnor-mally low behavioural performers – indeed, Figure 5illustrates that the participants with the greatest d¢scores, those that fell within the normal range (seeTable 1), were amongst those with the most delayedand prolonged physiological responses – a relation-ship verified by inspection of each individualparticipant’s time series in comparison to the groupand normal control averages. The quadratic natureof this relationship further suggests the intriguingpossibility of a heterogeneous effect of atypicalphysiology on behaviour, one in which heightenedprefrontal activation can either succeed or fail incompensating for atypical neural and cognitive pro-cesses. With regard to right intraparietal sulcus inparticular, it is interesting to speculate that thenegative linear effect in ASC might be akin to the leftside of the sibs’ quadratic effect.

The negative t scores in all four bilateral (eighttotal), distinct middle frontal gyrus (MFG) regions inall three groups suggest a general relationshipbetween AQ and frontal activation: the greater theAQ, the more delayed and prolonged the fMRI timecourse. This relationship manifested in posteriorMFG (Brodmann area 9) for the ASC and sib groups,and also in anterior MFG for the sib group. Withinthe ASC group, ADI-R subscores for social andcommunicative but not repetitive behaviours weresimilarly related to MFG atypicality. These relation-ships to social and communicative impairments

6 Our chosen contrast cannot, of course, distinguish between

hyperactivation following incongruent trials and hypoactiva-

tion following congruent trials. In either case, though, the

resemblance between ASC and sibs is of interest.7 Note that our constraint on the separation of congruent and

incongruent events allows us to meaningfully subtract

responses up to two trials out. The resulting 9 s interval

(current trial plus two subsequent trials) includes the onset of

the second phase of the response.



arise in the context of a non-social task of visualattention, suggesting a common substrate for socialand non-social applications of cognitive control.Such a commonality between social and non-socialdomains makes sense in the context of Minshewet al.’s notion of a domain-general impairment incomplex processing, and in the context of abnormalneural connectivity which has been suggested as abasis for that impairment. It also fits with Dawson etal.’s notion of perceptual abnormality as a contrib-uting cause of autistic social deficits, and with thepossibility of remediating those deficits by slowingthe physical presentation of social stimuli (Tardif,Laine, Rodriguez, & Gepner, 2007) to match theslowed pace of frontal activation.

A limit to this study is the lack of gaze trackingduring performance of the task. Gaze monitoringwithin the scanner environment was a difficulttechnical obstacle at the time these data were col-lected, and without gaze data one cannot absolutelyrule out the possibility that group behavioural andneurophysiological differences may have been drivenby group differences in eye direction. However, acombination of features of the experimental para-digm, the analytical procedure, and the brain acti-vation data seem to render this possibility unlikely:since stimuli were presented only for 167 ms, par-ticipants would not have had time to implementsaccades away from the stimulus location; further-more, since the analysis included only the trialsassociated with correct responses, during theseanalysed trials participants were likely to have beenlooking at the stimulus location when the stimuluswas presented; and in addition, the ASC group’sstrong activations in the visual regions left fusiformgyrus and right posterior intraparietal sulcus (V7)suggest that visual stimuli were indeed being pro-cessed during these trials.

The qualitative resemblances between ASC andsibs that we have presented, though interesting,raise a troubling question: if clinically unaffectedsibs of people with ASC are ‘slightly autistic’ on thesebehavioural and physiological measures of visualattention (as has already been shown in the case ofhigher-level social cognitive measures (Constantinoet al., 2006)), then what is it that makes some peoplein these families autistic and some not? As notedabove, we and many others have suggested thatfunctional connectivity may be an important metricin autism. We found anatomically distributed pat-terns of functional correlation in the control and sibgroups, including a visual attention network com-prising many frontal, cerebellar, parietal, andoccipital regions. The ASC group, in contrast, man-ifested overall much lower functional correlation (inan analysis pooled across all region pairs), and whatcorrelation did exist seemed concentrated (in anexploratory analysis uncorrected for multiple com-parisons) between middle frontal gyrus, cuneus andprecuneus – a pattern that seemingly bypasses most

of the frontal processing resources brought to bear inthe control and sib groups, and is consistent with arecent result on fronto-posterior connectivity in ASCduring a task of facial emotion recognition (Wickeret al., 2008).

This result of reduced inter-regional correlation isconsistent with a previous finding of high variabilityand low correlation of resting-state metabolismbetween brain regions (Horwitz, Rumsey, Grady, &Rapoport, 1988), and with recent results ofdecreased functional connectivity in autism during avariety of cognitive tasks (Just et al., 2004; Koshinoet al., 2005; Kana, Keller, Cherkassky, Minshew, &Just, 2006; Just, Cherkassky, Keller, Kana, & Min-shew, 2007). Importantly, although the sibs didresemble the ASC group in terms of activations, theydid not measurably resemble the ASC group in termsof overall inter-correlation. If there is any bifurcationin the developmental landscape, beyond which onepath converges to autism and another escapes it, thecurrent study suggests that this distinction may bemost expressed in terms of long-range functionalcorrelations, reflecting co-activation between brainregions. Activity within regions, on the other hand,may form more of a continuum in keeping with thebehavioural measures of this and other studies, andmay be permissive of ASC but not determinative. Asattention is known to act at least in part by modu-lating functional connectivity between brain regions(Friston & Buchel, 2000;Wang et al., in press), per-haps the particularly strong (albeit abnormallytimed) activations in non-ASC sibs in brain regionsassociated with attentional/executive functions maybe what preserves functional connectivity in thesesibs, rescuing them from the ASC phenotype.

A perennial question in fMRI studies of ASC is theextent to which the high-functioning subpopulationthat is capable of tolerating the scanner environ-ment may generalise to the wider ASC population.This is an especial concern in the context of thecurrent study, where six of the eight participantswith ASC had VIQ > PIQ, with two in the verysuperior range of VIQ. Overall, IQ measures in theASC group had four times the variance of those inthe other two groups, and even by this larger vari-ance measure mean VIQ and PIQ in the ASC groupwere half a standard deviation lower than in theother two groups. Despite this peculiarity of oursample, the results seem consistent with previousfindings of altered attention and altered patterns ofbrain activation and brain connectivity in autism,and are at least useful in establishing a hypothesisof familial patterns of atypical fronto-cerebellaractivation and more ASC-specific patterns ofdecreased functional correlation.

It is also important to note that the alterations infunctional co-activation that we have described donot necessarily reflect alterations in functionalconnectivity. Whereas functional connectivity asmeasured by fMRI was originally described as



low-frequency signal correlation during rest, ourmeasure of functional co-activation is driven by taskevents, localised in higher frequency bands, anddependent on the identification of activated brainregions. Our result of decreased co-activation in theASC group therefore may reflect a significantly lowernumber of activated regions, a significantly greatervariability in the anatomical loci of activation acrossindividuals, or atypically low functional connectivity– all of which findings have been reported in autism(Muller, 2007). Although sorting out these potentialmechanisms of apparently decreased co-activationremains a topic for further study, the specificity ofthis effect to ASC and its absence in the sib groupmay prove to be a significant clue as to the process ofautistic brain development.

The current study establishes several imperativesfor future neurophysiological investigations of aut-ism. First, familial endophenotypes are significantand span task domains and levels of analysis. Muchcan be gained from combining behavioural andphysiological methods, correlating across social andnon-social task domains, and comparing ASC andnon-ASC family members. At the time of this writing,only two other functional imaging studies haveexamined clinically unaffected family members:Baron-Cohen et al. (2006b) studied parents ofchildren with ASC and reported preliminary findingsof hypoactivation of extrastriate cortex during theEmbedded Figures Test and hyperactivation of leftinferior frontal gyrus during the Reading the Mind inthe Eyes Test; however, as this study included neitheran ASC comparison group nor an age-matched con-trol group, it is difficult to make inferences regardinggroup similarities. Dalton et al. (2007) studied sibs inan age range slightly broader than this study’s, andfound decreased amygdala volume, along with de-creased gaze fixation on eyes and decreased activa-tion of right fusiform gyrus during a face recognitiontask. Future studies of brain activation and func-tional connectivity might very usefully address thepossible relationships between such social percep-tual characteristics and non-social traits, in familymembers with andwithout ASC. Second, in studies of

autistic functional anatomy – as in studies of devel-opmental disorders in general (Johnson, Halit, Grice,& Karmiloff-Smith, 2002) – differences in the timingof activations may be even more important than dif-ferences in their anatomical loci, and models basedon normal time courses may miss out the mostinteresting results. In future, model-free analysesusing multivariate methods such as partial leastsquares or independent components analysis will becrucial to such investigations. Third, both abnor-malities in regional time courses and abnormalities ininter-regional functional connectivity may be moreprecisely characterised by complementing fMRI withmore temporally specific methods such as quantita-tive EEG (Brock et al., 2002) and MEG. These tech-niques benefit from recent advances in high-densityrecording hardware and multivariate analysis. Wehope that autism research as a field will take up theseimperatives, as we intend to do.

Acknowledgements

M.K.B. was funded by a Young Investigator Awardfrom Cure Autism Now, and M.G. by grants from theFondation pour la Recherche Medicale, the RegionCentre, and the Fondation d’Entreprise FranceTelecom. MRI scanning took place at the WolfsonBrain ImagingCentre, Cambridge, andwas fundedbyan MRC grant awarded to Simon Baron-Cohen. Wethank the MIT Student Information Processing Boardfor providing computing facilities, and in particularAlex Rolfe for helping with disaster recovery. Most ofall, thanks are due to our participants and their fam-ilies for their time and enthusiasm.

Correspondence to

Matthew K. Belmonte, Department of HumanDevelopment, Martha Van Rensselaer Hall, CornellUniversity, Ithaca, NY 14853-4401, USA; Tel:+1.617.715.2049; Fax: +1.607.255.9856; Email:[email protected]

Key points

• In addition to abnormalities of social cognition, people with autism spectrum conditions manifest ab-normalities of non-social attention and sensory perception.

• Similar traits in non-autism-spectrum relatives suggest a continuity between clinically affected and un-affected family members.

• In autism-spectrum individuals, in the context of a non-social task of visual attention, activation withinprefrontal cortex is abnormally delayed and prolonged, and the functional connectivity between brainregions is abnormally low.

• Unaffected sibs show similarly abnormal prefrontal activation, but intact functional connectivity. Greatermagnitude of prefrontal activation in unaffected sibs suggests successful compensatory processing.

• In individuals both on and off the autism spectrum, delayed prefrontal activation in this non-social taskcorrelates with psychometric measures of autistic social and communicative traits.



References

Allen, G., Buxton, R.B., Wong, E.C., & Courchesne, E.(1997). Attentional activation of the cerebellum inde-pendent of motor involvement. Science, 275, 1940–1943.

Allen, G., & Courchesne, E. (2003). Differential effectsof developmental cerebellar abnormality on cognitiveand motor functions in the cerebellum: An fMRI studyof autism. American Journal of Psychiatry, 160, 262–273.

Allen, G., Muller, R.-A., & Courchesne, E. (2004).Cerebellar function in autism: Functional magneticresonance image activation during a simple motortask. Biological Psychiatry, 56, 269–278.

American Psychiatric Association. (2000). Diagnosticand statistical manual of mental disorders (4th edn,text rev.). Washington, DC: American PsychiatricAssociation.

Andrews, G., Halford, G.S., Bunch, K.M., Bowden, D., &Jones, T. (2003). Theory of mind and relationalcomplexity. Child Development, 74, 1476–1499.

Baron-Cohen, S., & Hammer, J. (1997). Parents ofchildren with Asperger syndrome: What is the cogni-tive phenotype? Journal of Cognitive Neuroscience, 9,548–554.

Baron-Cohen, S., Hoekstra, R.A., Knickmeyer, R.C., &Wheelwright, S. (2006a). The Autism-Spectrum Quo-tient (AQ) – adolescent version. Journal of Autism andDevelopmental Disorders, 36, 343–350.

Baron-Cohen, S., Ring, H.A., Chitnis, X., Wheelwright,S., Gregory, L.J., Williams, S.C.R., Brammer, M.J., &Bullmore, E.T. (2006b). fMRI of parents of childrenwith Asperger syndrome: A pilot study. Brain andCognition, 61, 122–130.

Baron-Cohen, S., Wheelwright, S., Skinner, R., Martin,J., & Clubley, E. (2001). The Autism-SpectrumQuotient (AQ): Evidence from Asperger syndrome/high-functioning autism, males and females, scien-tists and mathematicians. Journal of Autism andDevelopmental Disorders, 31, 5–17.

Belmonte, M.K., Allen, G., Beckel-Mitchener, A., Boul-anger, L.M., Carper, R.A., & Webb, S.J. (2004a).Autism and abnormal development of brain connec-tivity. Journal of Neuroscience, 24, 9228–9231.

Belmonte, M.K., & Carper, R.A. (2006). Monozygotictwins with Asperger syndrome: Differences inbehaviour reflect variations in brain structure andfunction. Brain and Cognition, 61, 110–121.

Belmonte, M.K., Cook, E.H. Jr, Anderson, G.M.,Rubenstein, J.L.R., Greenough, W.T., Beckel-Mitch-ener, A., Courchesne, E., Boulanger, L.M., Powell,S.B., Levitt, P.R., Perry, E.K., Jiang, Y., DeLorey, T.M.,& Tierney, E. (2004b). Autism as a disorder of neuralinformation processing: Directions for research andtargets for therapy. Molecular Psychiatry, 9, 646–663.Unabridged edition at http://www.cureautism-now.org/conferences/summitmeetings/

Belmonte, M.K., & Yurgelun-Todd, D.A. (2003a).Anatomic dissociation of selective and suppressiveprocesses in visual attention. NeuroImage, 19,180–189.

Belmonte, M.K., & Yurgelun-Todd, D.A. (2003b). Func-tional anatomy of impaired selective attention and

compensatory processing in autism. Cognitive BrainResearch, 17, 651–664.

Bertone, A.,Mottron, L., Jelenic, P., &Faubert, J. (2003).Motion perception in autism: A ‘complex’ issue. Jour-nal of Cognitive Neuroscience, 15, 218–225.

Bertone, A., Mottron, L., Jelenic, P., & Faubert, J.(2005). Enhanced and diminished visuo-spatial infor-mation processing in autism depends on stimuluscomplexity. Brain, 128, 2430–2441.

Brock, J., Brown,C.C., Boucher, J., &Rippon,G. (2002).The temporal binding deficit hypothesis of autism.Development and Psychopathology, 14, 209–224.

Burack, J.A. (1994). Selective attention deficits inpersons with autism: Preliminary evidence of aninefficient attentional lens. Journal of Abnormal Psy-chology, 103, 535–543.

Buxton, R.B., Wong, E.C., & Frank, L.R. (1998).Dynamics of blood flow and oxygenation changesduring brain activation: The balloon model. MagneticResonance in Medicine, 39, 855–864.

Constantino, J.N., Lajonchere, C., Lutz, M., Gray, T.,Abbacchi, A., McKenna, K., Singh, D., & Todd, R.D.(2006). Autistic social impairment in the siblings ofchildren with pervasive developmental disorders.American Journal of Psychiatry, 163, 294–296.

Constantino, J.N., & Todd, R.D. (2003). Autistic traitsin the general population: A twin study. Archives ofGeneral Psychiatry, 60, 524–530.

Courchesne, E., & Pierce, K. (2005). Why the frontalcortex in autism might be talking only to itself: Localover-connectivity but long-distance disconnection.Current Opinion in Neurobiology, 15, 225–230.

Dalton, K.M., Nacewicz, B.M., Alexander, A.L., &Davidson, R.J. (2007). Gaze-fixation, brain activa-tion, and amygdala volume in unaffected siblings ofindividuals with autism. Biological Psychiatry, 61,512–520.

Dawson, G., Webb, S.J., Wijsman, E., Schellenberg, G.,Estes, A., Munson, J., & Faja, S. (2005). Neurocog-nitive and electrophysiological evidence of alteredface processing in parents of children with autism:Implications for a model of abnormal development ofsocial brain circuitry in autism. Development andPsychopathology, 17, 679–697.

Friston,K.J.,&Buchel,C. (2000).Attentionalmodulationof effective connectivity from V2 to V5/MT in humans.Proceedings of the National Academy of Sciences of theUnited States of America, 97, 7591–7596.

Gomot, M., Bernard, F.A., Davis, M.H., Belmonte, M.K.,Ashwin, C., Bullmore, E.T., & Baron-Cohen, S.(2006). Change detection in children with autism:An auditory event-related fMRI study. NeuroImage,29, 475–484.

Hadjikhani, N., Liu, A.K., Dale, A.M., Cavanagh, P., &Tootell, R.B.H. (1998). Retinotopy and color sensitiv-ity in human visual cortical area V8. Nature Neuro-science, 1, 235–241.

Haist, F., Adamo, M., Westerfield, M., Courchesne, E.,& Townsend, J. (2005). The functional neuroanatomyof spatial attention in autism spectrum disorder.Developmental Neuropsychology, 27, 425–458.

Halford, G., Wilson, W.H., & Phillips, S. (1998). Pro-cessing capacity defined by relational complexity:Implications for comparative, developmental, and



cognitive psychology. Behavioral and Brain Sciences,21, 803–831.

Happe, F., Briskman, J., & Frith, U. (2001). Exploringthe cognitive phenotype of autism: Weak ‘centralcoherence’ in parents and siblings of children withautism: I. Experimental tests. Journal of Child Psy-chology and Psychiatry, 42, 299–307.

Horwitz, B., Rumsey, J.M., Grady, C.L., & Rapoport,S.I. (1988). The cerebral metabolic landscape inautism. Intercorrelations of regional glucose utiliza-tion. Archives of Neurology, 45, 749–755.

Iarocci, G., Burack, J.A., Shore, D.I., Mottron, L., &Enns, J.T. (2006). Global–local visual processing inhigh functioning children with autism: Structural vs.implicit task biases. Journal of Autism and Develop-mental Disorders, 36, 117–129.

Johnson, M.H., Halit, H., Grice, S.J., & Karmiloff-Smith,A. (2002). Neuroimaging of typical and atypical devel-opment: A perspective frommultiple levels of analysis.Development and Psychopathology, 14, 521–536.

Just, M.A., Cherkassky, V.L., Keller, T.A., Kana, R.K., &Minshew, N.J. (2007). Functional and anatomicalcortical underconnectivity in autism: Evidence froman fMRI study of an executive function taskand corpuscallosummorphometry.Cerebral Cortex,17, 951–961.

Just, M.A., Cherkassky, V.L., Keller, T.A., & Minshew,N.J. (2004). Cortical activation and synchronizationduring sentence comprehension in high-functioningautism: Evidence of underconnectivity. Brain, 127,1811–1821.

Kana, R.K., Keller, T.A., Cherkassky, V.L., Minshew,N.J., & Just, M.A. (2006). Sentence comprehension inautism: Thinking in pictures with decreased func-tional connectivity. Brain, 129, 2484–2493.

Kennedy, D.P., Redcay, E., & Courchesne, E. (2006).Failing to deactivate: Resting functional abnormali-ties in autism. Proceedings of the National Academy ofSciences of the United States of America, 103, 8275–8280.

Koshino, H., Carpenter, P.A., Minshew, N.J., Cherkas-sky, V.L., Keller, T.A., & Just, M.A. (2005). Functionalconnectivity in an fMRI working memory task in high-functioning autism. NeuroImage, 24, 810–821.

Lancaster, J.L., Woldorff, M.G., Parsons, L.M., Liotti,M., Freitas, C.S., Rainey, L., Kochunov, P.V., Nicker-son, D., Mikiten, S.A., & Fox, P.T. (2000). AutomatedTalairach atlas labels for functional brain mapping.Human Brain Mapping, 10, 120–131.

Lord, C., Rutter, M., & Le Couteur, A. (1994). AutismDiagnostic Interview – Revised: A revised version of adiagnostic interview for caregivers of individuals withpossible pervasive developmental disorders. Journalof Autism and Developmental Disorders, 24, 659–685.

Maas, L.C., Frederick, B.D., & Renshaw, P.F. (1997).Decoupled automated rotational and translationalregistration for functional MRI time series data: TheDART registration algorithm. Magnetic Resonance inMedicine, 37, 131–139.

Mann, T.A., & Walker, P. (2003). Autism and a deficit inbroadening the spread of visual attention. Journal ofChild Psychology and Psychiatry, 44, 274–284.

Milne, E., Swettenham, J., & Campbell, R. (2005).Motion perception and autistic spectrum disorder: Areview. Cahiers de Psychologie Cognitive, 23, 3–34.

Minshew, N.J., Goldstein, G., & Siegel, D.J. (1997).Neuropsychologic functioning in autism: Profile of acomplex information processing disorder. Journal ofthe International Neuropsycholical Society, 3, 303–316.

Mottron, L., Dawson, M., Soulieres, I., Hubert, B., &Burack, J. (2006). Enhanced perceptual functioningin autism: An update, and eight principles of autisticperception. Journal of Autism and DevelopmentalDisorders, 36, 27–43.

Muller, R. (2007). The study of autism as a distributeddisorder. Mental Retardation and Developmental Dis-abilities Research Reviews, 13, 85–95.

Mundy, P. (2003). Annotation: The neural basis of socialimpairments in autism: The role of the dorsal medial-frontal cortex and anterior cingulated system. Journalof Child Psychology and Psychiatry, 44, 793–809.

Piven, J., Palmer, P., Jacobi, D., Childress, D., & Arndt,S. (1997). Broader autism phenotype: Evidence froma family history study of multiple-incidence autismfamilies. American Journal of Psychiatry, 154, 185–190.

Plaisted, K., Swettenham, J., & Rees, L. (1999). Chil-dren with autism show local precedence in a dividedattention task and global precedence in a selectiveattention task. Journal of Child Psychology andPsychiatry, 40, 733–742.

Serences, J.T., Shomstein, S., Leber, A.B., Golay, X.,Egeth, H.E., & Yantis, S. (2005). Coordination ofvoluntary and stimulus-driven attentional control inhuman cortex. Psychological Science, 16, 114–122.

Silk, T.J., Rinehart, N., Bradshaw, J.L., Tonge, B.,Egan, G., O’Boyle, M.W., & Cunnington, R. (2006).Visuospatial processing and the function of prefron-tal-parietal networks in autism spectrum disorders: Afunctional MRI study. American Journal of Psychia-try, 163, 1440–1443.

Small, D.M., Gitelman, D.R., Gregory, M.D., Nobre,A.C., Parrish, T.B., & Mesulam, M.M. (2003). Theposterior cingulate and medial prefrontal cortexmediate the anticipatory allocation of spatial atten-tion. NeuroImage, 18, 633–641.

Swisher, J.D., Halko, M.A., Merabet, L.B., McMains,S.A., & Somers, D.C. (2007). Visual topography ofhuman intraparietal sulcus. Journal of Neuroscience,27, 5326–5337.

Talairach, J., & Tournoux, P. (1988). Coplanar Stereo-tactic Atlas of the Human Brain. Stuttgart: Thieme.

Tardif, C., Laine, F., Rodriguez, M., & Gepner, B.(2007). Slowing down presentation of facial move-ments and vocal sounds enhances facial expressionrecognition and induces facial-vocal imitation inchildren with autism. Journal of Autism and Devel-opmental Disorders, 37, 1469–1484.

Townsend, J., & Courchesne, E. (1994). Parietal dam-age and narrow ‘spotlight’ spatial attention. Journal ofCognitive Neuroscience, 6, 220–232.

Townsend, J., Singer-Harris, N., & Courchesne, E.(1996). Visual attention abnormalities in autism:Delayed orienting to location. Journal of theInternational Neuropsychological Society, 2, 541–550.

Townsend, J., Westerfield, M., Leaver, E., Makeig, S.,Jung, T., Pierce, K., & Courchesne, E. (2001).Event-related brain response abnormalities inautism: Evidence for impaired cerebello-frontal



spatial attention networks. Cognitive Brain Research,11, 127–145.

Wang, L., Liu, X., Guise, K.G., Knight, R.T., Ghajar, J.,& Fan, J. (in press). Effective connectivity of thefronto-parietal network during attentional control.Journal of Cognitive Neuroscience.

Wicker, B., Fonlupt, P., Hubert, B., Tardif, C., Gepner,B., & Deruelle, C. (2008). Abnormal cerebral effectiveconnectivity during explicit emotional processing inadults with autism spectrum disorder. Social Cogni-tive and Affective Neuroscience, 3, 135–143.

Wojciulik, E., & Kanwisher, N. (1999). The generality ofparietal involvement in visual attention. Neuron, 23,747–764.

Woodbury-Smith, M.R., Robinson, J., Wheelwright, S.,& Baron-Cohen, S. (2005). Screening adults forAsperger Syndrome using the AQ: A preliminarystudy of its diagnostic validity in clinical practice.Journal of Autism and Developmental Disorders, 35,331–335.

Zwaigenbaum, L., Bryson, S., Rogers, T., Roberts, W.,Brian, J., & Szatmari, P. (2005). Behavioral manifes-tations of autism in the first year of life. InternationalJournal of Developmental Neuroscience, 23, 143–152.

Manuscript accepted 20 July 2009



visual attention in autism families: ‘unaffected’ sibs...

Documents