Investigation of Neuroanatomical Differences Between Autism and AspergerSyndrome FREE

Autism is a pervasive developmental disorder (PDD) defined by the followingtriad of behavioral characteristics: social and communication impairmentsin combination with restricted and repetitive behaviors.^1- 2 Manyautistic individuals have cognitive impairments; in the literature, subjectswith an IQ of less than 70 are typically designated as having low-functioningautism (LFA), and those with an IQ of at least 70 as having high-functioningautism (HFA).^3- 5 Aspergersyndrome (ASP), another PDD, is similar to autism, sharing features of socialimpairment and repetitive behaviors, in the absence of communication and cognitiveimpairments (ie, phrase language developed before 36 months of age and IQ≥70).^2,6- 7 Because persons withHFA also have IQs of at least 70, and the DSM-IV doesnot require a history of language delay for a diagnosis of autism, this createsa diagnostic overlap between HFA and ASP, resulting in many individuals withASP who meet DSM-IV criteria for autism.^4,8 As a result, many studies comparingHFA and ASP distinguish these 2 conditions according to history of phraselanguage development (HFA at 36 months or older, and ASP at younger than 36months).^9- 13

Owing to the diagnostic similarities between HFA and ASP, a debate isgrowing about the validity of ASP as a disorder distinctly different fromautism. Several studies have found differences between HFA and ASP on measuresof social skills,¹⁰ cognition/executive functioning,^14- 16 and motor ability^17- 18; others reported no differences onsimilar measures.^13,19- 23 Ithas been suggested that the spectrum of behavioral and cognitive patternsseen in individuals with PDD is driven by an underlying severity gradient.^23- 24 Subtypes of PDD align themselvesalong a severity continuum, beginning with LFA at one end, moving throughHFA, and ending with ASP.³ Others argue thatthese conditions may represent distinct neuropathological disorders with overlappingbehavioral and cognitive symptoms.⁵ In eithercase, the underlying neurodevelopmental mechanisms leading to these conditionsare unknown.

A number of structural magnetic resonance imaging (sMRI) studies ofthe brain in subjects with autistic disorder have revealed neuroanatomicalabnormalities of the corpus callosum,^25- 27 cerebellarvermal lobules VI and VII,^28- 29 andamygdala and hippocampus.^30- 31 Thesefindings were not always replicated,^32- 36 possiblyowing to differences between subject populations, scanning procedures, andimage processing methods between research sites.

The most consistent sMRI finding is increased brain volume in autisticsubjects.^37- 41 Thisfinding is consistent with reports of increased head circumference^42- 44 and brain weight⁴⁵ in autism. Increased brain volume was found regionallyin the parietal, temporal, and occipital but not the frontal lobes⁴⁶ and in cortical gray and cerebral white tissue.³⁸ There appear to be age effects on brain volume inautism; children tend to have larger brain volumes than older individualsrelative to age-matched controls.^38,41 Mostautism brain volume studies included subjects with LFA and HFA,^{37- 38,40,46} buta few studies were restricted to those with HFA.^39,41 Thesestudies did not specifically include subjects with ASP.

Of the few sMRI studies of subjects with ASP,^47- 50 onlyone investigated brain volume, and McAlonan and coworkers⁴⁹ reportedno difference in total hemispheric volume between ASP and control subjects.In a related study, Gillberg and de Souza⁵¹ usedhead circumference data to report macrocephaly in a subgroup of subjects withASP.

The first goal of the present study is to assess brain volumes in thefollowing 3 PDD groups: LFA, HFA, and ASP. In particular, we sought to determinewhether these PDD groups constitute distinct biological entities as evidencedby neuroanatomical measures. This is the first volumetric neuroimaging investigation,to our knowledge, that compares subjects with autism and those with ASP. Thisstudy is part of a larger project to investigate intersite differences thatmight explain the inconsistent replication seen in autism neuroimaging investigations.Thus, our second goal is to examine the differences and similarities in subjectpopulations and neuroimaging data between 2 sites when using the same subjectrecruitment strategies, scanning protocols, and data measurement procedures.

Subjects were boys ranging in age from 7.8 to 17.9 years who met eligibilityfor 1 of the following 4 subject groups: LFA (Full-Scale IQ [FSIQ], <70),HFA (FSIQ, ≥70), ASP, and age-matched controls. The HFA, ASP, and controlgroups were recruited jointly by the University of California–Davis(UC Davis) and Stanford University Medical School, Stanford, Calif (Stanford).The LFA group was recruited solely by UC Davis. Both sites recruited throughlocal parent networks and regional professionals who work with the PDD population.All subjects with PDD underwent screening and were excluded if they had anymajor medical (eg, fragile X syndrome) or psychiatric condition.

Control subjects were recruited at both sites through newspaper advertisementsand through friends of the subjects with PDD. The controls were matched bygroup age with the PDD subjects. All controls were in good physical healthand underwent screening to exclude neurological, developmental, or psychiatricdisorders. They also underwent screening for any psychiatric symptoms withthe Child Behavioral Checklist.⁵² The studywas approved by the institutional review boards of Stanford and UC Davis.Written consent was obtained from all subjects and their parents.

Subjects with PDD first underwent assessment and rating on the DSM-IV² criteria for a diagnosisof autism or ASP. They then underwent assessment using the Autism DiagnosticInterview–Revised (ADI-R)⁵³ and the AutismDiagnostic Observation Schedule–Generic (ADOS-G)⁵⁴ bytrained examiners (L.J.L. and B.L.G.-J.) who had each established reliabilitywith 1 of the developers of the instrument. Standardized cognitive testingusing the Wechsler Abbreviated Scale of Intelligence⁵⁵ wasadministered to all subjects with the exception of those with LFA, who wereadministered the Leiter International Performance Scale–Revised.⁵⁶

For inclusion in the LFA group, subjects had to have ADI-R and ADOS-Gthreshold scores for autism and an FSIQ of less than 70. Subjects with HFAhad to have ADI-R and ADOS-G threshold scores for autism, an FSIQ of at least70, and a history of phrase speech development at 36 months or older. TheASP group had to meet DSM-IV criteria for ASP orautism, an ADOS-G threshold score for autism or autism spectrum disorder,an FSIQ of at least 70, and a history of phrase speech development at youngerthan 36 months. Since many persons with ASP also meet ADI-R and DSM-IV criteria for autism,^8- 11 inthis study the primary distinguishing feature between individuals with HFAand ASP was a history of clinically significant language impairment; thisstrategy has been used in other studies.^9- 13 Insummary, subjects with LFA and HFA were differentiated by FSIQ scores, andsubjects with ASP and HFA were differentiated by age of phrase language development.

A total of 73 subjects underwent analysis in this study, including 13with LFA, 9 with HFA, 11 with ASP, and 11 controls from UC Davis and 9 withHFA, 10 with ASP, and 10 controls from Stanford.

Subjects at both sites participated in the sMRI protocol described inthis study. Subjects at Stanford also participated in a functional MRI protocoland thus had to remain alert throughout scanning. At Stanford, subjects firstunderwent screening with an MRI simulator. Those subjects with excessive headmovement were withdrawn from the study. In contrast, at UC Davis, subjectswith PDD who could not remain still underwent scanning under general anesthesia.Images from 29 subjects were acquired on a 3.0-T GE Signa whole-body echospeedMRI system (GE Medical Systems, Milwaukee, Wis) at the Richard M. Lucas Centerat Stanford, whereas images from 19 subjects were acquired on a 1.5-T GE SignaNeurovascular-optimized MRI system at UC Davis Imaging Research Center. Theremaining 25 subjects with PDD required general anesthesia; accordingly, theirdata were acquired on the 1.5-T GE Signa MRI system at the UC Davis MedicalCenter. A 3-dimensional volumetric radio-frequency spoiled-gradient echo pulsesequence was used to acquire all images in the coronal plane, with the followingparameters: repetition time of 35 milliseconds, echo time of 6 milliseconds,flip angle of 45°, number of signals is 1, matrix size of 256 ×192, field of view of 24 cm, full bandwidth of 32 kHz, and slice thicknessof 1.5 to 1.7 mm for 124 contiguous sections.

At Stanford, all 73 images were imported into the program BrainImage5.X⁵⁷ for masked semiautomated image-processinganalyses and brain volume measurements.⁵⁸ Theseprocedures were previously described and validated.^59- 61 Dataprocessing included removal of nonbrain tissue, correction of image inhomogeneity,data interpolation to cubic dimensions, and segmentation into gray tissue,white tissue, and cerebrospinal fluid (Figure1) for the following structures: cerebral lobes, subcortical nuclei,cerebellum, and lateral ventricles. Specific regions are parcelled and measuredusing a semiautomated stereotactic method.^58,62- 63

One of the purposes of using a multisite study design is to elucidatethe degree to which differences in methods between research sites might contributeto the variable structural MRI results reported in the autism literature.Although all MRI systems used in this study were GE Signa systems, they differedby magnetic field strength (3 T vs 1.5 T) and software (Stanford, GE HorizonLX Version 8.3; UC Davis Hospital, GE Horizon LX Version 8.3.5; and UC DavisResearch Imaging Center, GE Horizon LX Version 8.4M4).

An intersite MRI comparison was conducted using images from 3 normalvolunteer adults (1 man and 2 women) who underwent scanning during a 10-monthperiod with the 3 MRI systems described above. Images were acquired and analyzedwith the same pulse sequence and BrainImage 5.X⁵⁷ programas that used with the study subjects. Total brain and segmented tissue volumeswere compared between the 3 systems; a percentage difference between MRI systemsfor each subject was averaged across subjects to arrive at a mean percentagedifference for each volumetric measure. Only those volumetric values witha mean percentage difference of less than 5% between sites were used in thisstudy. Following are the observed mean percentage differences. For cerebrummeasures, these were 1.2% for cerebral total tissue, 1.8% for cerebral graymatter, and 2.3% for cerebral white matter; for cerebellum measures, 6.2%for cerebellar total tissue, 7.0% for cerebellar gray matter, and 17.4% forcerebellar white matter. Cerebral volumes for total, gray, and white tissueshad mean intersite differences of 3% or less and thus were used in the analysisof the 3 PDD and control groups.

We used analysis of variance (ANOVA) followed by Scheffé posthoc testing to assess the 4 subject groups for any differences in age andIQ (performance IQ [PIQ], verbal IQ [VIQ], and FSIQ). The ANOVA followed byScheffé post hoc testing also was used to examine the 2 sites for anydifferences in age and IQ for those subject groups who underwent scanningat both sites (HFA, ASP, and control). The subjects with LFA were excludedfrom the second analysis, because they were recruited only at UC Davis. Forthese 2 sets of analyses, we used parametric statistics, as the distributionsof the data did not violate assumptions of normality or homogeneity of variance.

We first applied a parametric method, ANOVA, for MRI volumetric analysis.Because the variance of MRI volumetric findings for the LFA group was largerthan for the other subject groups, analysis was repeated using nonparametricmethods (Kruskal-Wallis test with post hoc Mann-Whitney test). We assessedinteractions between volumetric measures with site, age, and IQ for the 3subject groups recruited from both sites (ASP, HFA, and control). Interactionterms were excluded from final ANOVA models if they did not approach or reachsignificance (P<.10). Finally, we used the Pearsoncorrelation coefficient to explore the potential effects of age and IQ onMRI volumetric values. We then compared these within-group correlations usingthe Fisher r-to-z transformation.For all analyses in this report, we used a P valueof .05 as a threshold for statistical significance.

Among the 73 boys recruited, PIQ ranged from 36 to 142. For the HFA,ASP, and control groups, VIQ scores ranged from 67 to 144 and FSIQ scoresranged from 70 to 140. The VIQ and FSIQ scores were not available for theLFA group, because they were administered the Leiter Scale, which only providesa PIQ.

Means and standard deviations for age and IQ of subject groups are displayedin Table 1. The subject groupsdid not differ significantly in age. There was a significant main effect ofsubject group on PIQ. Post hoc testing revealed that the LFA group had a lowerPIQ compared with the HFA (P<.001), ASP (P<.001), and control (P<.001)groups, whereas the HFA, ASP, and control groups were not significantly differentfrom each other. There also was a main effect of group (ie, HFA, ASP, andcontrol) on VIQ. Post hoc analyses revealed that the HFA group had a significantlylower VIQ compared with the ASP (P<.001) and control(P<.001) groups. There was no significant differencebetween the ASP and control groups in VIQ. Groups also differed on FSIQ, whichwas lower in the HFA group compared with the ASP (P =.03) and control (P = .001) groups, reflecting thepattern seen in VIQ.

Table 2 shows age and IQresults for both sites across the 3 subjects groups (ie, HFA, ASP, and control);the LFA group was excluded since they underwent imaging only at UC Davis.Stanford subjects were significantly older than UC Davis subjects. This differencein age between sites was seen across all 3 diagnostic groups, but was foundto be statistically significant only in the HFA group (F_1,16 =7.9; P = .01). There also was a significant between-sitedifference in IQ; UC Davis subjects had lower PIQ, VIQ, and FSIQ comparedwith Stanford subjects. The UC Davis HFA group had lower PIQ (P<.001) and lower FSIQ (P = .01) comparedwith the Stanford HFA group, but no differences in VIQ. The UC Davis ASP grouphad lower PIQ (P<.001), VIQ (P = .01), and FSIQ (P = .003) compared withthe Stanford ASP group. Analyses revealed no significant IQ differences betweencontrol groups from the 2 sites.

Results of MRI volumetric results for the 4 subject groups are shownin Table 3. There was a significantsubject group effect for cerebral gray tissue (F_3,69 = 4.37; P = .007; Figure 2)but no significant subject group effect for cerebral total or white tissue.Because of relatively larger variance within the LFA group compared with theother 3 groups, analyses were repeated using nonparametric methods (Table 3). These results were similar tothose obtained with ANOVA. Post hoc 2-group analyses using Mann-Whitney indicatedthat, compared with the control group, the LFA (P =.04) and HFA (P = .009) groups had enlarged cerebralgray matter volumes, whereas the LFA, HFA, and ASP groups were not significantlydifferent from each other.

Knowing that there were intersite differences for age and IQ, we thenlooked for effects of 2-way interactions between site and subject group usingage and PIQ as covariates and cerebral gray matter volume as the dependentvariable. We excluded LFA from this analysis. When all 2-way interactionswere included in an initial ANOVA model, the subject group × site 2-wayinteraction approached significance (F_2,57 = 2.6; P = .09) and was therefore included in the final analysis of covariancemodel. Consistent with the previous results, a significant main effect ofsubject group on cerebral gray matter volume was observed (F_2,57 =4.02; P = .02). Age also contributed significantlyto the final model (F_1,58 = 6.60; P =.01); decreasing cerebral gray matter volumes were correlated with increasingage across the 3 groups.

Within-group correlations between cerebral gray matter and the variablesof age and IQ are shown in Table 4.A negative correlation between cerebral gray matter volume and age was significantin the HFA group (r₁₆ = −0.53; P = .02). A significant negative correlation between cerebralgray matter volume and PIQ was observed for the HFA group (r₁₆ = −0.49; P = .04), anda positive correlation for the same variables approached significance forthe ASP group (r₁₉ = 0.42; P = .06). To rule out confounding relationships between age and PIQin the HFA and ASP groups, these correlations between cerebral gray mattervolume and PIQ were repeated using age as a covariate. These partial correlationswere of borderline significance for HFA and significant for ASP (Table 4).

Within-group correlations between cerebral white matter and the variablesof age and IQ are shown in Table 5.The correlation between cerebral white matter and age was significant forthe HFA, ASP, and control groups; white matter volume was observed to increasewith increasing age across the samples. Only the ASP group had a positivewithin-group correlation between cerebral white matter volume and PIQ andbetween cerebral white matter volume and VIQ. Significance was maintainedwhen analyses were repeated as partial correlations using age as a covariate(Table 5).

There were a few significant between-group differences for correlationsof volumetric measures with IQ (Figure 3 and Figure 4). The correlations for cerebralgray matter volume and PIQ for the HFA and ASP groups were in opposite directions;this between-group difference was significant (Figure 3). There also were between-group differences for cerebralwhite matter volume with PIQ correlations between the ASP and the other 2groups (HFA and control [Figure 4]).

Despite the use of similar recruitment strategies and scanning protocols,we found site differences in subjects by age and IQ. These differences werenot significant in the control group, but were significant in the HFA andASP groups. Site-specific differences in IQ may be related, in part, to differencesin subject enrollment; Stanford's PDD subject recruitment was limited to thosewith an FSIQ of 70 or above, whereas UC Davis' recruitment included IQs aboveand below 70. Another reason for site differences in subjects' age and IQis the subject retention and withdrawal practices dictated by differencesin MRI protocols between sites. University of California–Davis usedan sMRI protocol only and thus used general anesthesia for those subjectswith PDD who could not conform to the motion reduction requirements of MRI,regardless of IQ. In contrast, Stanford used sMRI and functional MRI protocolsand withdrew potentially eligible subjects owing to their inability to reducehead movement during the MRI simulation. As a result, Stanford withdrew 12subjects with PDD, who had an average age of 10.3 years (IQs not measured),and UC Davis anesthetized 12 subjects with HFA and with ASP, who had an averageage of 10.7 years and an average VIQ and PIQ of 84 each. As Stanford was withdrawingyounger subjects and possibly subjects with lower IQs, UC Davis was retainingsimilar subjects, resulting in age and IQ differences between sites. Poorreplication of autism MRI results may thus be partially explained by differencesin eligible subject pools stemming from differences in recruitment and MRIprotocols.

Other sources of variation in published PDD neuroimaging results aredifferences in MRI methods, including field strengths of the MRI system, variationsin image acquisition protocols, rater training, methods of image processing,and statistical methods. In this 2-site study, we used uniform imaging acquisitionand analysis protocols but different MRI systems with different field strengths(two 1.5T systems at UC Davis and a 3.0-T system at Stanford). Results ofthe reliability test of 3 volunteer subjects revealed differences of greaterthan 5% in cerebellar but not cerebral measures. To minimize the effects ofusing different MRI systems on brain volumetrics, we limited our analysesto cerebral volumes. If these MRI reliability results are replicated in alarge group of subjects, this may shed light on the lack of agreement betweenneuroimaging studies, particularly when a cerebellar tissue segmentation procedureis used. Other possible sources of poor replicability between MRI studiesare site differences in imaging acquisition and analysis protocols. Thesecould not be addressed in this study, because we used similar protocols. Onthe basis of these results, MRI reliability analysis should become a standardprocedure for multisite neuroimaging studies.

The HFA and LFA groups both had enlarged total cerebral gray mattervolumes compared with the control group. In a similar study, Courchesne etal³⁸ reported increased cortical gray mattervolume in young autistic children aged 2 to 3 years, but not in older autisticsubjects aged 6 to 16 years. Courchesne and colleagues³⁸ studiedonly a few subjects in the age range reported in the present study, and thusthey may not have had the power to detect a difference. In their study andthe present one, cerebral gray tissue volume was observed to be reduced withincreasing age in autistic subjects. The enlarged cerebral gray matter volumeseen in the present study is congruent with findings of enlarged brain volume^37,41 in autism and a neuropathologicalstudy⁴⁵ showing increased cortical volume inautistic adults.

There also appears to be a relationship between cerebral gray mattervolume and IQ in autism that is unrelated to mental retardation. That is,within the HFA group, there was a tendency for individuals with large cerebralgray matter volumes to have lower PIQs. This negative correlation stands incontrast to analyses of typically developing children and adolescents, inwhom larger brain volumes are associated with higher IQ.⁶⁴ Twoprevious investigations^40- 41 failedto find a significant correlation between total brain volume and IQ in autisticsubjects. This indicates that the brain-IQ relationships in autism may pertainonly to cerebral gray matter and not to total brain tissue.

Increased cerebral gray tissue in autism may be due to abnormalitiesin gray tissue development. Neuropathology studies of autism have revealedcerebral gray matter abnormalities that include an increased number of minicolumnsper unit area along with fewer neurons per minicolumn,⁶⁵ smallerand more densely packed neurons in the anterior cingulate gyrus and limbicsystem,⁶⁶ and an approximately 50% reductionin protein levels of the enzymes that synthesize γ-aminobutyric acidand glutamic acid decarboxylase in parietal and cerebellar cortices.⁶⁷ Overexpression of specific neuropeptides and neurotrophinswere reported in neonatal blood of infants who were later diagnosed as havingautism.⁶⁸ Overall, a growing body of literaturesupports the conclusion that abnormalities in gray matter development area defining feature of autism.

In the present study, we noted that the LFA group had an unusually largevariance in cerebral total tissue. This suggests that, neuroanatomically,the LFA sample represents a more heterogeneous population than HFA or ASPsamples. Increased heterogeneity implies a greater mixture of disparate etiologies,some of which may be unidentified single-gene disorders. The probability thatLFA is more heterogeneous than HFA has previously been discussed.⁶⁹

To our knowledge, this is the first neuroimaging study to investigatedifferences in brain volumetric measures between subjects with ASP and thosewith autism. When ASP and HFA are distinguished by timing of language development,as in this study, there are no differences in cerebral volumetric measures(total, gray, and white tissue) between these 2 PDD subgroups. Also, no differenceswere observed between ASP and control groups on these same measures, a findingconsistent with the report of McAlonan and colleagues,⁴⁹ whofound no differences in total cerebral volume in ASP adults compared withcontrols.

In the current study, the mean cerebral gray matter volume for the ASPgroup was intermediate between means for the HFA and control groups; thismay indicate a continuum in which cerebral gray matter volume increases withthe severity of the PDD condition. Using a different MRI technique, voxel-basedanalysis, 2 investigations^49- 50 reportedgray tissue differences in ASP subjects compared with controls. A neuropathologystudy⁷⁰ reported abnormal minicolumn architecturein ASP subjects similar to that described in autistic subjects, suggestinga common underlying neuropathology.

We also have preliminary evidence that HFA and ASP may differ from eachother in specific brain-behavior relationships. First, the HFA group had theatypical pattern of decreasing PIQ associated with increasing gray mattervolume, whereas the ASP group had the typical pattern of increasing PIQ associatedwith increasing gray matter volume.⁶⁴ Second,there was a strong correlation between PIQ and cerebral white tissue volumein the ASP group that differed significantly from the HFA and control groups.Previous studes^64,71 in typicallydeveloping children have suggested that IQ is not related to white tissuevolume. This functional white tissue difference between ASP subjects and controlsmay be congruent with another study,⁴⁹ whichused MRI voxel-based analysis and reported white tissue differences betweenASP subjects and controls. These suggested brain-behavior differences betweenHFA and ASP, based on exploratory analyses, are somewhat speculative and requireconfirmation.

Our attempt to determine whether HFA and ASP disorders are conditionson a continuum or are distinct biological entities was only partially successful.On the single measure of cerebral gray tissue volume, these conditions appearto represent a continuum of severity, with autism exhibiting the greatestaberrant neurodevelopment. However, on multiple measures (ie, brain-behaviorcorrelations of IQ with specific cerebral volumes) there is preliminary evidenceof fundamentally different patterns of neurodevelopment between HFA and ASPsubjects. These findings are based on differentiating HFA and ASP by historyof language development. These dissonant neuroimaging results reflect thepresent literature on behavioral and cognitive studies of HFA and ASP.^4- 5 Rinehart et al⁵ concludedthat results of behavioral and cognitive studies "suggest that it is prematureto rule out the possibility that autism and Asperger disorder may be clinically,and possibly neurobiologically, separate."^5(p768) Family studies⁵ indicate thatASP may be genetically different from autism. Our results suggest that whenHFA and ASP are differentiated by history of language development, as theyare herein, qualitative differences may surface when patterns of muliple measuresare examined.

The 2-site design of this study is both a limitation and a strength.Use of different MRI systems and subject groups (ie, differences on age andIQ) introduces confounding variables and is a limitation. However, the 2-sitedesign uncovers those variables that may explain the poor replicability ofprevious autism MRI investigations and thus is a strength.

We were able to address the known intersite differences in age and IQby statistically accounting for the effects of age and IQ on the brain volumecomparisons. Differences in MRI system field strength were addressed by limitingthe analyses to only those volumetric measures with good intersite reliability;this restricted the analyses to measurements focused on the cerebrum. Theseadjustments may not completely address all intersite differences. Thus, thisstudy needs to be replicated using a intersite design with greater attentionto common subject enrollment and withdrawal practices and MRI procedures (ie,sMRI vs functional MRI and the sedation protocol). In an ideal design, travelingsubjects should be incorporated for MRI reliability.

Increased sample size would have permitted more robust statistical comparisonof the 4 groups. Greater numbers would have given us more power to detectdifferences where they exist. Since there are age effects on brain development,a prospective study design in which the same subjects undergo scanning everyfew years into early adulthood should give us the best method to determinedifferences in gray and white tissue volumes in individuals with PDD.

Corresponding author: Linda J. Lotspeich, MD, Department of Psychiatryand Behavioral Sciences, Stanford University School of Medicine, 401 QuarryRd, Stanford, CA 94305 (e-mail: Linda.Lotspeich@stanford.edu).

Submitted for publication February 11, 2003; final revision receivedAugust 12, 2003; accepted October 7, 2003.

This study was supported by grants MH01142, MH50047, and HD31715 (DrReiss) and MH01832 (Dr Lotspeich) from the National Institutes of Health,Bethesda, Md, and a grant from The MIND Institute, Davis, Calif (Drs Amaraland Reiss).

This study was presented as a poster at the International Meeting forAutism Research; November 1, 2002; Orlando, Fla.

We thank Cindy Johnston, John Ryan, and Meridith Brandt for their contributionto the collection of data, and the study subjects and their families for theirparticipation.