Here's Looking at You, Kid:  Neural Systems Underlying Face and Gaze Processing in Fragile XSyndrome FREE

Fragile X syndrome (fraX) is an inherited neurodevelopmental disordercaused by disrupted expression of the fragile X mental retardation (FMR1) gene. In this relatively common syndrome, FMR1 gene expression is reduced, and the resulting lack of FMR1 protein(FMRP) protein leads to altered synaptic function and abnormal dendritic spinemorphology.¹ Fragile X syndrome is an X-linkedcondition; therefore, female subjects have 1 affected and 1 unaffected X chromosome,resulting in FMRP levels that are reduced by approximately 50%. In addition,the level of FMRP can vary among affected male and female individuals, thuscontributing to variation in the level of impairment. As a neuropsychiatricphenotype, fraX is associated with increased risk for impairment in severaldomains, including sustained attention, working memory, visuospatial analysis,visuomotor coordination, executive function, and social function.²

Social difficulties, often associated with a subset of autistic behaviors,are one of the earliest and most maladaptive symptoms of fraX. These difficultiesmay worsen during childhood, thus having a potential long-term effect on thechild's adaptive behavior and mental health. One hallmark of fraX is the propensityto avoid eye contact and turn away during a social greeting, even while offeringa handshake or socially acceptable remark.³ Maleindividuals also show abnormalities in social play with peers and during verbaland nonverbal social communication.⁴ Younggirls with fraX are usually less affected than boys but also show social deficits,including withdrawal and avoidant behavior.⁵ Evenfemale subjects with fraX who have IQs in the normal range exhibit socialdysfunction, which may be related to reports of increased frequencies of socialanxiety and mood disorders⁶ in these subjects.The social avoidance seen in individuals with fraX has been attributed tohyperarousal during social situations.^7- 8 Together,these studies suggest that brain systems that process socially relevant stimulimay function differently in fraX than in typically developing subjects.

In typically developing subjects, the neurofunctional correlates ofsocial cognition have been investigated by observing responses to faces andthe direction of eye gaze. Neuroimaging studies have found greater activationof the fusiform gyrus (FG) in response to faces compared with letter strings,scrambled faces, houses, or human hands.^9- 11 Also,patients with lesions in this area have difficulty recognizing faces,¹² and intraoperative recordings from the FG show increasedresponses to faces.^13- 14 However,the specificity of FG activation remains in question. Recently, Gauthier andcolleagues^15- 16 suggested thatvisual expertise in general, rather than the analysis of faces in particular,recruits the capabilities of the FG. An understanding of the variables thatmodulate FG activation in response to faces is beginning to emerge. Rossionand colleagues¹⁷ showed that the left FG wasparticularly responsive to parts of faces rather than whole faces, whereasthe opposite was true for the right FG. Forward and angled faces have beenshown to evoke a greater FG response than profiles of faces.¹⁴ Attentionto faces increases FG activation,¹⁸ and reducedaccuracy at matching faces may decrease FG activation.¹⁹

The perception of gaze direction is less well studied, but neuroimagingdata suggest the involvement of the superior temporal sulci (STS)^20- 21 and the middle and inferior temporalgyri.²² Some studies suggested that avertedgaze activates the STS more than does direct gaze,²⁰ whereasother studies found no differences in STS activation to averted vs directgaze.^22- 23 A recent parametricanalysis showed that increasing proportions of direct gaze are associatedwith increasing blood flow in the STS.²⁴

We hypothesized that if the FG and the STS are typically activated inresponse to face and gaze stimuli, then studying activation of these regionsin individuals with fraX may help us begin to understand the nature of socialdifficulties in this condition. Several studies have shown that individualswith fraX, unlike individuals with autism, recognize and recall faces normally.^2,25 Therefore, we predicted that the perceptionof faces, which is attributed to the FG, may function normally in fraX. However,because of the symptom of gaze aversion, the STS may show altered functioning,and brain regions associated with anxiety may show increased activation infraX. We addressed our hypothesis by presenting 4 types of stimuli: Forwardand angled faces with direct and averted gaze. Direct gaze and forward facessignify socially relevant stimuli. Averted gaze and angled faces are appropriatecontrol stimuli because they possess similar visual characteristics. Combinedwith the inclusion of a homogeneous patient group with an identifiable etiology,this design allowed us to provide a foundational study of social informationprocessing in fraX. To our knowledge, this study is the first to use event-relatedfunctional magnetic resonance imaging (fMRI) to examine functional brain abnormalitiesunderlying the perception of face and gaze stimuli in fraX.

Fifteen female subjects with a diagnosis of fraX (fraX subjects) wererecruited through advertisements in national and regional fraX newslettersand referrals from physicians. Fifteen typically developing female controlsubjects were recruited through advertisement within the local community.We included only female subjects because they have milder symptoms than malefraX subjects and therefore are more likely to perform the task accuratelyand tolerate the scan. Including only female subjects also removes intersubjectvariance attributable to sex.

All subjects reported that they were right-handed. All fraX subjectshad the FMR1 full mutation, as confirmed by DNA (Southernblot) analysis. Written informed consent was obtained from all participants,and the human subjects review committee at Stanford University School of Medicine,Stanford, Calif, approved all protocols.

All of the control subjects were medication free. Of the final 11 fraXsubjects included in the study, 8 were medication free. The remaining 3 fraXsubjects were taking the following medications: (1) guanfacine hydrochloride(Tenex) and venlafaxine hydrochloride (Effexor); (2) paroxetine (Paxil) andmethylphenidate hydrochloride (Ritalin Hydrochloride); and (3) paroxetine,methylphenidate, and levothyroxine sodium (Synthroid). Subjects stopped takingmethylphenidate and guanfacine for 24 hours before the scan, and continuedtaking all other medications. The fraX group consisted of 9 white, 1 Hispanic,and 1 Pacific Islander subject. The final control group consisted of 9 white,1 Hispanic, and 1 Asian American subject.

The final subject groups did not differ significantly in age (fraX group,mean age, 16.4 years [SD, 4.09 years; range, 10-22 years]; control group,mean age, 15.5 years [SD, 3.41 years; range, 10-22 years]; F_1,20<1).The IQs were measured using the Wechsler Intelligence Scale for Children IIIfor subjects younger than 17 years and the Wechsler Adult Intelligence ScaleIII for subjects 17 years and older. One fraX subject was removed for havingan IQ significantly below the group mean (>2.5 SDs). To reduce the IQ disparitybetween groups, 3 control subjects were removed for having an IQ significantlygreater than the group mean (>120, or 1.33 SDs). The final Full-Scale IQ scoresfor the fraX group were in the average range of intelligence (80-111), witha mean score of 93.7 (SD, 10.4). Full-Scale IQ scores for the control groupranged from 85 to 120, with a mean score of 107.0 (SD, 11.2). A between-groups t test verified that the control group had significantlyhigher Full-Scale IQ scores than the fraX group (F_1,20 = 8.24 [P<.01]).

Color photographs of faces of 120 college-aged models with neutral facialexpressions were taken against a common, solid-color background at a distanceof about 2 m. Thirty photographs from each of the following 4 categories wereused: (1) face forward with direct gaze, (2) face forward with averted gaze,(3) face angled with direct gaze, and (4) face angled with averted gaze. Angledfaces and averted gaze were turned approximately 45° away from the camera.The photographs included 66 men and 54 women. Of these, 93 were white and27 were African American, Hispanic, Asian American, or Indian. Sex and racewere distributed similarly across stimulus categories. Examples of the stimuliare shown in Figure 1.

A research investigator (K.M.) practiced a training version of the taskwith each subject until confident that she understood the instructions andcould respond accurately. Subjects performed 2 tasks. The event-related taskused a jittered stimulus presentation, with a mean intertrial interval of1572 milliseconds (SD, 1805 milliseconds)²⁶ anda range of 0.25 to 4.25 seconds. Stimuli were presented using PsyScope software,²⁷ which also triggered the initiation of the fMRI scanby sending a transistor-transistor logic pulse to the scanning processor.Stimuli were projected onto a screen attached to the head coil. Subjects lookeddirectly upward at a mirror to view the stimuli. Each stimulus was presentedfor 1750 milliseconds, followed by a 250-millisecond duration fixation cross.Subjects were instructed to use the right index finger to press a button ifthe person in the photograph was looking at them, and to use the right seconddigit to press another adjacent button if the person was looking away fromthem. Correct and incorrect responses and reaction times were recorded ifthey occurred between 150 and 2000 milliseconds after the stimulus. Each subjectperformed 2 runs of the event-related task, with each run lasting 4 minutes32 seconds. The runs were separated by 2 to 3 minutes to prevent subject fatigue.In each run, 15 stimuli of each condition were presented, so 2 runs contained30 stimuli per condition.

The same stimuli described above were then presented in a block designto derive functional regions of interest (ROIs) defining the FG and STS. Inparticular, a block design was used to measure responses to all types of stimulicombined. This method of defining ROIs by the location of significant activationin response to related stimuli has been used in previous studies²⁸ andis a methodologically attractive way to create subject-specific ROIs basedon functional anatomy. The task consisted of 8 alternating epochs, each lasting30 seconds and presenting 15 stimuli for 1750 milliseconds each, with a 250-millisecondinterstimulus interval. Half of the epochs contained a mix of all 4 typesof face stimuli, whereas the alternating epochs contained scrambled pictures.Subjects were asked to indicate the direction of gaze for the faces and toalternate pressing the first and second button in response to the scrambledpictures.

Images were acquired on a 1.5-T GE scanner (General Electric Company,Milwaukee, Wis) using a custom-built whole-head coil that provides a 50% advantagein signal-to-noise ratio over that of the standard GE head coil.²⁹ Acustom-built head holder was used to prevent head movement. Eighteen axialslices (6-mm thickness; 1-mm gap) parallel to the anterior and posterior commissuresand covering the whole brain were imaged using a T2-weighted gradient echospiral pulse sequence (repetition time [TR], 2000 milliseconds; echo time[TE], 40 milliseconds; flip angle, 89° and 1 interleave³⁰;field of view [FOV], 240 mm; in-plane resolution, 3.75 mm). To help localizeactivation, a high-resolution T1-weighted, spoiled GRASS (gradient recalledacquisition in a steady state) image (SPGR) 3-dimensional MRI sequence (TR,24 milliseconds; echo time, 5 milliseconds; flip angle, 40°; FOV, 240mm; 124 sagittal slices; 256 × 192 matrix; resolution, 1.5 × 0.9× 1.2 mm) was also collected.

A 3-way analysis of variance (ANOVA) was used to examine task accuracy(percentage correct) and response time. The factors included face orientation(forward and angled), gaze orientation (direct and averted), and group (controland fraX).

Functional images were reconstructed by means of the inverse Fouriertransform for each of the 186 time points into 64 × 64 × 18 imagematrices (voxel size, 3.75 × 3.75 × 7 mm). Functional MRI datawere analyzed using SPM99 software (Statistical Parametric Mapping 99; WellcomeDepartment of Cognitive Neurology, Institute of Neurology, University College,London, England). Images were corrected for movement using least squares minimizationwithout higher-order corrections for spin history, and normalized to stereotaxicMontreal Neurologic Institute coordinates.³¹ Imageswere then resampled every 2 mm using sinc interpolation and smoothed witha 4-mm gaussian kernel to decrease spatial noise.

The general linear model and the theory of gaussian random fields implementedin SPM99 were used to complete statistical analyses of fMRI data.³⁰ For each subject, activation was calculated at eachvoxel and corrected for temporal autocorrelation. Confounding effects of fluctuationsin the global mean were removed by proportional scaling. Low-frequency noisewas removed by applying a high-pass filter (0.5 cycle/min) to the fMRI timeseries at each voxel. A temporal smoothing function (gaussian kernel correspondingto dispersion of 8 seconds) was applied to the fMRI time series to enhancethe temporal signal-to-noise ratio.

For each subject, a t score image was generatedfor each contrast of interest, including (1) forward compared with angledfaces, collapsed over gaze orientation, and (2) direct compared with avertedgaze, collapsed over face orientation. Individual contrast images were combinedinto a group image using a random-effects model, which provides a strongergeneralization to the population.³² All comparisonsbetween the fraX and control groups were controlled for differences in IQby including IQ as a covariate. A mask was used to remove group differencesarising from deactivation. Voxelwise t statisticswere normalized to z scores to provide a statisticalmeasure independent of sample size. Significant clusters of activation weredetermined using the joint expected probability of height (z>1.96 [P<.05]) and extent of z scores (P<.05),³³ yieldinga clusterwise significance level of P = .05, correctedfor multiple comparisons. The MNI coordinates were converted to Talairachcoordinates using procedures described by Brett.³⁴ Activationfoci were superimposed on high-resolution T1-weighted images and localizedwith reference to the stereotaxic atlas of Talairach and Tournoux.³¹ Because the contrasts examined in this study werechosen a priori, activations from other contrasts are not reported.

An ROI analysis was used to explore activation of the FG and STS withingroups and in response to each type of stimulus. The boundaries of anatomicalregions were drawn on each subject's spatially normalized structural MRI.These regions were further constrained by a mask constructed from each subject'sactivation during the block design task, which combined all face and gazestimuli (minus scrambled stimuli), thresholded at P<.05.When conjoined with the anatomical outline, each subject's FG and STS regionswere defined structurally and functionally. The ROI activation was calculatedas the percentage of significant voxels (z>1.67)for each of the 4 stimulus conditions. A repeated measures ANOVA was conductedfor each ROI with the factors of face orientation (forward and angled), gazeorientation (direct and averted), hemisphere (right and left), and group (fraXand control).

The anatomical boundaries of all regions were drawn on coronal slicesusing BrainImage software.³⁵ The FG beginsat the coronal slice containing the largest cross section of the anteriorcommisure and proceeds back to the posterior transverse collateral sulcus.The medial border is the collateral sulcus, and the lateral border is theoccipitotemporal sulcus. The superior temporal region includes the STS andthe superior and middle temporal gyri. It begins at the onset of the anteriortemporal pole and terminates posteriorly at the disappearance of the inferiortemporal sulcus.

Three fraX subjects were removed from the sample for scoring below 50%on the gaze discrimination task, leaving only those subjects whose task accuracyindicated understanding and compliance with task demands. One control subjectwas removed for failure to respond to the task. Thus, the following resultsinclude 11 subjects in the fraX group and 11 subjects in the control group.

Group means and SDs for task accuracy are summarized in Table 1. Main effects were found for group, face orientation, andthe interaction between face and gaze orientation. Control subjects had significantlygreater accuracy compared with the fraX group (F_1,19 = 13.39 [P = .002]). For all subjects, accuracy was greater forforward than for angled faces (F_1,19 = 10.88 [P = .004]). The significant interaction between face and gaze orientation(F_1,19 = 10.60 [P = .004]) showed thatfor forward faces, accuracy was greater for direct gaze than for averted gaze,but for angled faces, accuracy was not different for direct compared withaverted gaze conditions. This effect was driven primarily by the fact thatall subjects had greater accuracy responding to forward faces with directgaze than to any other stimulus.

Full-Scale IQ scores were significantly correlated with task accuracyfor both groups combined (r = 0.56 [P = .003, 1-tailed]). However, this appeared to reflect a categorical(ie, group) effect only, since IQ and task accuracy were not significantlycorrelated within each of the groups (control group, r =0.36 [P = .14]; fraX group, r =0.35 [P = .15]). Therefore, all subsequent between-groupanalyses were covaried for IQ, which also controlled for between-group differencesin task accuracy.

Table 1 shows the groupmeans and SDs for response time. No group differences were found in responsetime. Main effects were found for face orientation and the interaction betweenface and gaze orientation. For all subjects, reaction time was quicker forforward than for angled faces (F_1,19 = 5.07 [P<.04]). The interaction between face and gaze orientation (F_1,19 = 35.03 [P<.001]) showed that for forwardfaces, response time was similar for direct and averted gaze, but for angledfaces, response time was longer for direct than for averted gaze (F_1,19 = 21.2 [P<.001]).

There were no significant between-group differences in the comparisonof forward faces with angled faces (collapsed over gaze orientation). TheROI analysis was used to investigate activation of specific brain regionswithin each group and for each type of stimulus.

Three clusters of significantly greater activation were found in thecontrol group. Activation maxima included the STS, lingual gyrus, and cerebellum.Other significantly activated regions are listed in Table 2 and shown in Figure 2.

Two clusters of significantly greater activation were found in the fraXgroup. Activation maxima included the right insula and cerebellum. Other significantlyactivated regions are listed in Table 2 and shown in Figure 2.

A significant interaction was detected between group and face orientation(F_1,20 = 9.20 [P = .007]). Although controlsubjects had significantly greater FG activation in response to forward comparedwith angled faces (post hoc t₁₁ = 6.42[P = .02]), fraX subjects had no difference for forwardcompared with angled faces (P = .09) (Figure 3A). Activation of the FG to angled faces was not significantlydifferent between the fraX and control groups (post hoc F_1,19 =0.309 [P = .59]). A significant interaction betweengroup and hemisphere (F_1,20 = 5.75 [P =.03]) showed that controls had significantly greater right than left FG activationto all stimuli, whereas fraX subjects had no hemispheric differences in FGresponse (Figure 3B). Activationof FG in the left hemisphere is not significantly different between groups(F_1,19 = .55 [P = .47]).

The analysis of the STS region showed significant main effects of groupand hemisphere. Control subjects had greater STS activation than fraX subjectsin response to all stimulus conditions combined (F_1,19 = 6.11 [P = .02]; Figure 4).In addition, all subjects had greater STS activation in the right hemispherecompared with the left (F_1,20 = 14.61 [P =.001]).

We examined abnormalities in neural responses to face and gaze stimulito investigate the basis of alterations in social behavior, such as gaze aversion,in fraX individuals. Although all subjects had IQ scores within the averagerange of intelligence, the fraX group had lower IQ scores and decreased accuracyin determining gaze direction compared with controls. The ROI analysis ofthe FG showed a significant interaction between group and face orientation.Control subjects had greater FG activation to forward than to angled faces,whereas subjects with fraX had no difference in activation to forward comparedwith angled faces. Controls had significantly greater STS activation to allstimuli compared with fraX individuals. Therefore, our results suggest thatgaze avoidance in fraX individuals may be related to reduced ability to perceivegaze and decreased specialization in the perception of face orientation.

The results of the whole brain analysis of the FG could be interpretedas contradictory to the results of the ROI analysis of the same region. TheROI analysis showed that controls had significantly greater activation toforward than to angled faces, but the fraX group had no difference for forwardvs angled faces. Therefore, for the whole brain analysis, we would expectcontrol subjects to have greater FG activation than the fraX subjects forthe group comparison of forward minus angled faces. However, the whole brainanalysis showed no group differences. We believe that intersubject variabilityin the location of peak FG activation was responsible for this disparity.Group differences in FG activation in the whole brain analysis were in thesame direction as the ROI-based results (control > fraX) but did not reachthe significance threshold (z = 3.69 [P = .001 uncorrected; P = .21 corrected]).This indicates how the ROI analysis was helpful in controlling for individualvariation in neuroanatomy, as intended.

Activation of FG is reliably found in response to faces, and is typicallygreater for forward than for angled faces,^9- 11 perhapsbecause forward faces are perceived as more socially relevant. Lack of fusiformspecialization may be associated with a relatively greater tendency of fraXindividuals to look at faces when the faces are looking away.^36- 37 Thus,they may develop a normal ability to process faces, but no preference forforward faces. On the other hand, fraX individuals avoid social gaze altogether,and therefore may not develop a normal ability to process gaze.

Another possibility is that the FG finding indicates difficulty processingangled faces, since subjects with fraX had significantly lower accuracy whenresponding to angled faces compared with forward faces. Decreased accuracyhas been associated with increased activation of other visual cortical regions.³⁸ However, decreased accuracy also has been associatedwith decreased FG activation in previous studies involving control subjects.^16,19 Our data do not suggest that FG activationis related to accuracy in the current study. A post hoc analysis showed thatthere was no correlation between FG activation and accuracy for both groupscombined or considered separately (combined, Pearson r =−0.10 for left FG and r = 0.09 for right FG;fraX subjects, r = 0.15 for left FG and r = 0.06 for right FG; control subjects, r =−0.05 for left FG and r = 0.03 for right FG).

Our results suggest that social abnormalities in fraX individuals arealso related to a reduced ability to perceive social gaze, as evidenced bydecreased STS activation to all categories of stimuli. Anatomical abnormalitiesin the STS have previously been reported in fraX.³⁹ Activationof this region has been associated with the perception of social gaze in controlsubjects.^20- 21 Since fraX individualstypically avoid social gaze, they may not develop a normal ability to processgaze. Of course, we cannot determine whether alterations in the STS causegaze aversion behavior, or whether gaze aversion behavior results in changesin these brain regions.

An alternative explanation for these findings is that fraX subjectslooked away from the photographs. Although we cannot rule this out, becausewe did not measure eye movements, we did not find different activity in thefrontal eye fields (Brodmann areas 6/8) of fraX subjects compared with controls.Also, fraX subjects did not have decreased accuracy in response to forwardgaze stimuli. Finally, the participants in this study were chosen to be mildlyaffected patients with less severe gaze aversion behavior.

Social problems in fraX have been attributed to hyperarousal and anxiety.^7- 8 Although our study did not directlyassess the role of anxiety in social gaze perception, we did not see increasedactivation of the amygdala to direct gaze in fraX subjects. However, we sawincreased activation of the right anterior insula, ventral prefrontal cortex,and midbrain. The ventrolateral prefrontal cortex has sensory and limbic inputs⁴⁰ and is activated during the experience of emotions,including sadness⁴¹ and anxiety.⁴² Similarly,the midbrain has been activated in neuroimaging studies of several emotions,including anxiety.⁴³ Insula activation is relatedto the experience of visceral and emotional symptoms, including chest constriction,fear, and uneasiness.^44- 45 Activationof this constellation of regions suggests an increased emotional responseto direct gaze in the fraX group. However, activation in these regions isnot specific to anxiety. In addition, greater left posterior insula activationwas seen in the control group. More research will be needed to determine whethersocial anxiety is related to altered gaze processing in fraX. It is possiblethat the photographs did not provoke anxiety in the fraX individuals becausethe scan did not involve an actual social situation, but only stimuli associatedwith social interactions. The knowledge that no real social interaction wouldtake place during the fMRI task could have helped to lessen the social anxietytypically observed in these subjects.

This study is significant in distinguishing brain responses to socialstimuli in fraX from those previously reported in autism. A subset of thesocial deficits seen in fraX is observed in autism, and hence the 2 developmentaldisorders have long been compared and contrasted.⁴ Forexample, both fraX and autistic individuals experience difficulties with verbaland nonverbal social communication.⁴⁶ However,unlike autistic children, children with fraX show a propensity to engage insocial behaviors with caregivers,⁴ and theycorrectly identify facial and auditory emotion.^25,47 Cohenand colleagues^36- 37 showed that,although both fraX and autistic children avoided social interactions, malefraX subjects avoided a stranger more than a parent, and autistic childrenavoided stranger and parent equally. Furthermore, analysis of dyadic socialgaze patterns suggest that male fraX subjects are sensitive to eye gaze butavoid it because they find it aversive, whereas autistic subjects are insensitiveto gaze and do not engage in social gaze, probably because of lack of interestor attention.^36- 37,48 Ourstudy furthers the distinction between fraX and autistic subjects by showingdifferences in neurofunctional responses to social stimuli. Previously, Schultzand colleagues⁴⁹ found reduced right FG activationto forward faces in autistic subjects. Similarly, Pierce and colleagues⁵⁰ found low or absent FG activation in response toface stimuli in autistic subjects, but not controls. In contrast, our studyfound activation of the FG in response to forward and angled faces in subjectswith fraX, although with less differentiation of FG response to face orientation.

Finally, this study did not test whether activation differences in theFG and STS in fraX are specific to face and eye gaze stimuli or can be generalizedto other visual stimuli. Also, further studies are needed to determine whetherincreased anxiety is related to STS dysfunction in fraX. Monitoring heartrate and eye gaze during scanning may help us to answer these questions infuture investigations.

Corresponding author: Amy S. Garrett, PhD, Department of Psychiatryand Behavioral Sciences, Stanford University School of Medicine, 401 QuarryRd, Stanford, CA 94305 (e-mail: agarrett@stanford.edu).

Submitted for publication March 11, 2003; final revision received July24, 2003; accepted August 5, 2003.

This study was supported by grants MH19908, MH64708, MH01142, MH50047,and HD31715 (Dr Reiss) and HD40761 (Dr Menon) from the National Institutesof Health, Bethesda, Md, and a gift from the Lynda and Scott Canel Fund forFragile X Research.

We thank Noah Merin and Chris White for data collection, David Hessl,PhD, for subject recruitment and testing, and Gary Glover, PhD, for technicalexpertise.