Postmortem brain abnormalities of theglutamate neurotransmitter system

in autismA.E. Purcell, BA; O.H. Jeon, PhD; A.W. Zimmerman, MD; M.E. Blue, PhD; and J. Pevsner, PhD

Article abstract—Background: Studies examining the brains of individuals with autism have identified anatomic andpathologic changes in regions such as the cerebellum and hippocampus. Little, if anything, is known, however, about themolecules that are involved in the pathogenesis of this disorder. Objective: To identify genes with abnormal expressionlevels in the cerebella of subjects with autism. Methods: Brain samples from a total of 10 individuals with autism and 23matched controls were collected, mainly from the cerebellum. Two cDNA microarray technologies were used to identifygenes that were significantly up- or downregulated in autism. The abnormal mRNA or protein levels of several genesidentified by microarray analysis were investigated using PCR with reverse transcription and Western blotting. �-Amino-3-hydroxy-5-methyl-4-isoxazoleproprionic acid (AMPA)- and NMDA-type glutamate receptor densities were examined withreceptor autoradiography in the cerebellum, caudate-putamen, and prefrontal cortex. Results: The mRNA levels of severalgenes were significantly increased in autism, including excitatory amino acid transporter 1 and glutamate receptor AMPA1, two members of the glutamate system. Abnormalities in the protein or mRNA levels of several additional molecules inthe glutamate system were identified on further analysis, including glutamate receptor binding proteins. AMPA-typeglutamate receptor density was decreased in the cerebellum of individuals with autism (p � 0.05). Conclusions: Subjectswith autism may have specific abnormalities in the AMPA-type glutamate receptors and glutamate transporters in thecerebellum. These abnormalities may be directly involved in the pathogenesis of the disorder.

NEUROLOGY 2001;57:1618–1628

First described more than 50 years ago,1 autism re-mains an enigmatic, pervasive developmental disor-der. The lifelong syndrome is characterized by threemajor symptoms: deficits in social interaction, lan-guage abnormalities, and stereotyped repetitive pat-terns of behavior.2 Additional characteristics of thedisorder are that three to four times more men areaffected than women3 and by adulthood about one-third will have had at least two unprovoked sei-zures.4 The prevalence of autism is generallyconsidered to be two to five in 10,000.3 A recentstudy, however, suggests autism may be consider-ably more common, with a prevalence greater thanone in 1000.3

Autism, a behaviorally defined disorder, is theconsequence of a variety of genetic and nongeneticcauses. Although it is therefore difficult to determinethe primary cause of autism, there has been a grow-ing interest in identifying common pathways thatare affected as a result of the disorder. Autism isthought to reflect a nonprogressive brain impair-ment; however, only a few studies have examined

alterations in postmortem brain from affected indi-viduals. Neuropathologic studies have detected ab-normalities in regions including the brainstem,hippocampus, and other diencephalic structures.5

Thus far, the most consistent findings are the abnor-malities found in the cerebella of persons with au-tism, which is the region we chose to examine in thisstudy. A 35 to 50% reduction in the number of Pur-kinje cells in autism cerebellum when compared withnormal cerebellum has been demonstrated.5-7 SeveralMRI studies suggest that the cerebellum may besmaller in individuals with autism.8,9 Apart fromneuroanatomic evidence of brain abnormalities, weknow little, if anything, about brain disturbances atthe molecular level, perhaps because of the limitedavailability of postmortem brain samples from indi-viduals with autism.

In seeking to understand the molecular patho-physiology of autism in brain, cDNA microarraytechnology represents an efficient way to screenthousands of genes at once for expression differenc-es.10,11 This approach may be useful in identifying

From the Departments of Neurology (Drs. Jeon, Zimmerman, and Pevsner and A.E. Purcell) and Neuroscience (Dr. Blue), Kennedy Krieger Institute, and theDepartments of Neuroscience (Drs. Jeon, Blue, and Pevsner and A.E. Purcell) and Neurology (Dr. Blue), Johns Hopkins University School of Medicine,Baltimore, MD.Supported by grants from Solving the Mystery of Autism, Target: Autism Genome, the Autism Research Foundation, and the Developmental DisabilitiesMental Retardation Research Center (HD24061).The Harvard Brain Tissue Resource Center is supported by PHS grant number MH/NS 31862. The University of Miami/University of Maryland Brain andTissue Banks operate under NICHD contract #N01-HD-8–3284.Received April 11, 2001. Accepted in final form July 4, 2001.Address correspondence and reprint requests to Dr. J. Pevsner, Department of Neurology, Kennedy Krieger Institute, 707 North Broadway, Baltimore, MD21205; e-mail: pevsner@kennedykrieger.org

1618 Copyright © 2001 by AAN Enterprises, Inc.

cellular pathways that have been perturbed as a re-sult of autism and in discovering genes that serve asdiagnostic markers. Most human studies usingcDNA microarrays have focused on cell lines or tu-mors. However, high-density microarrays also pro-vide a useful tool for the study of brain tissue,including diseases such as autism, in which neuronalcell lines or animal models are not available.12 Stud-ies have used microarrays to measure gene expres-sion in brain samples from patients with neurologicdisorders such as MS13 and AD.14 In this study,cDNA microarrays are used to examine globalchanges in gene expression in the postmortem cere-bella of individuals with autism, a region withknown pathology.

Materials and methods. Tissue samples. Frozen post-mortem brain samples from the cerebellar cortex of nineautistic (A1 to A5; A7 to A10) and 18 control subjects (C1to C9; C11 to C19) were obtained from the Harvard BrainTissue Resource Center (Belmont, MA), with the supportof the Autism Research Foundation (Boston, MA) and theUniversity of Maryland Brain and Tissue Bank (Baltimore,MD). Most of these samples were from the cerebellum;however, several samples from the prefrontal cortex andcaudate-putamen were also provided (table 1). The diagno-sis of autism was assigned by a neurologist based on Diag-nostic and Statistical Manual–IV criteria15 and, in five ofthe nine cases, the Autism Diagnostic Interview–Revised.16

Additional frozen samples from the cerebellar cortex of oneautistic (A6) and one control subject (C10) were obtainedfrom the University of Miami (Miami, FL) in conjunctionwith the Maryland Brain and Tissue Bank. The diagnosisof autism was assigned by a neurologist based on theChildhood Autism Rating Scale.17 An additional four con-trol postmortem brain samples (C20 to C23) from the pre-frontal cortex or caudate-putamen were provided by theMaryland Brain and Tissue Bank. Of the 10 subjects withautism examined in this study, four (A1, A2, A4, and A10)

had a history of seizures, whereas the other cases had nohistory of seizures (A3, A5, A7 to A9) or information wasunavailable (A6). Additional clinical data for the patientswith autism, including cause of death, IQ, and medicationstatus prior to death are provided (see table 1, supplemen-tary content). The mean age (� SD) of all 10 patients withautism was 19 years (� 14 years), and the mean age (�SD) of the 23 control subjects was 22 years (� 15 years).Autism and control groups were matched for age, gender,and postmortem interval (PMI) within each experiment.

Autism and control cerebellum samples (~250 mg) wereprepared for protein analysis by homogenizing in SHEEPbuffer (0.32 M sucrose, 4 mM HEPES [4-(2-hydroxyethyl)-1-piperazineethanesulphonic acid]), 0.1 mM egtazic acid,0.1 mM ethylenediaminetetra-acetic acid, and 0.3 mM phe-nylmethane sulfonylfluoride). This preparation was thencentrifuged at 2000 � g for 10 minutes at 4 °C. The super-natant was collected and protein concentrations were de-termined by a protein assay,18 using bovine serum albuminas a standard. All tissue homogenates had a pH �6.3,which has been shown to be optimal for mRNA preserva-tion in human postmortem brain.19 Total RNA from eachbrain sample was isolated using TRIzol (Gibco-BRL,Gaithersburg, MD). RNA yield was determined by measur-ing the absorbance. The A260/A280 ratio for each samplewas confirmed to be between 1.9 and 2.0, the ratio of pureRNA (data not shown). Integrity of RNA was also assessedby electrophoresing samples on an ethidium bromide–stained, denaturing agarose gel (data not shown).

cDNA array hybridization and analysis. Gene expres-sion was analyzed using the Atlas Human Neurobiologyarray (Clontech Laboratories, Palo Alto, CA), which con-sists of 588 human cDNA spotted in duplicate onto a nylonmembrane. A total of eight individual hybridizations wereperformed with autism (A1, A2, A4, A9) and control (C7 toC9, C11) cerebellum samples. Fifteen micrograms of totalRNA from each sample were used to isolate poly A� RNAand synthesize radiolabeled cDNA probes as directed bythe manufacturer. Hybridization intensity of labeled prod-

Table 1 Clinical characteristics of postmortem brain samples

Autism Control

CaseAge,

y SexPMI,

h Region CaseAge,

y SexPMI,

h Region CaseAge,

y SexPMI,

h Region

A1 27 M 15 C, CP, F C1 6.0 M 21 C C12 53 M 17 C

A2 19 M 9.5 C, CP, F C2 16 M 13 C C13 20 F 21 C

A3 5.0 M 4.9 C C3 43 M 23 C C14 22 M 12 C

A4 20 M 15 C, CP, F C4 63 M 23 C C15 4.0 M 19 C

A5 54 M 4.3 C C5 28 M 24 C C16 5.0 M 15 C

A6 10 M 23 C C6 24 M 5.0 C, F C17 30 M 20 CP

A7 21 F 21 C C7 26 M 20 C C18 21 M 9.0 CP

A8 6.0 M 23 C C8 19 M 21 C, F C19 30 M 22 F

A9 19 M 15 C, F C9 20 M 16 C, CP C20 2.4 F 22 F

A10 9.0 M 24 C C10 5.0 M 19 C C21 25 F 16 CP, F

C11 19 M 17 C C22 23 M 4.0 CP, F

C23 22 M 13 CP, F

PMI � postmortem interval; C � cerebellum; CP � caudate-putamen; F � prefrontal cortex.

November (1 of 2) 2001 NEUROLOGY 57 1619

ucts was detected by phosphorimaging (Fujifilm BAS-2500;Fuji Photo Film Co., Edison, NJ). Atlas Image v.1.0 soft-ware (Clontech) was used to measure the spot intensitiesin the scanned array images. The intensity values abovebackground (n � 357) for each array were globally normal-ized to a mean of 1.5. This value of the mean was chosenbecause it was the closest to the original intensity meansof the arrays. For each gene, fold regulation was calculatedby averaging the four autism intensity values as well asthe four control intensity values. A ratio was then madebased on whether the gene was up- (autism/control) ordownregulated (�[control]/autism) in autism samples.

Gene expression was also evaluated with the UniGEMV2 array (Incyte Genomics, Palo Alto, CA), which contains9374 sequence-verified cDNA immobilized on a glass slide.One hundred micrograms of total RNA from nine autism(A1 to A6, A8 to A10) and from a pool of four control (C7 toC9, C11) cerebellum samples were sent as coded samplesto Incyte Genomics, where poly A� was isolated and cDNAprobes synthesized with the fluorescent markers Cy3 (au-tism) and Cy5 (control pool). Equal amounts of poly A�

were used to generate probes. Nine hybridizations wereperformed, one for each autism sample against the controlpool. For each array, Incyte Genomics measured the inten-sity values for each cDNA signal and also globally normal-ized the Cy5 signal to the Cy3 signal. To be included in theanalysis of an array, signals were at least 2.5-fold abovebackground level and constituted at least 40% of the de-

fined signal area. If a gene did not meet these criteria inmore than three of the nine arrays, the gene was excludedfrom further analysis. Internal array controls were notincluded in the analysis of microarray data. To calculatefold regulation of a gene, a ratio of autism signal intensityto normalized control pool signal intensity was calculatedfor each array. These ratios were then averaged across thenine experiments. Finally, fold regulation was calculatedbased on whether the gene was up- (average ratio) ordownregulated (�[1/average ratio]) in autism samples.

Reverse transcriptase PCR. First strand cDNA wassynthesized from total RNA using oligo d(T) with Super-script II reverse transcriptase as directed by the manufac-turer (Life Technologies, Rockville, MD). Approximately20 ng of cDNA were added to a 100-L PCR mix contain-ing 1.5 mM of MgCl2, 200 M of each nucleotide, 2.5 unitsof Taq DNA polymerase (Qiagen, Valencia, CA), and0.5 M of each gene-specific primer. Glyceraldehyde-3-phosphate dehydrogenase was used as a control to normal-ize gene expression. The primer sequences that were usedfor PCR are provided (table 2, supplementary content). Thereaction mixtures were subjected to 24 cycles of two-stepPCR (Clontech). Each cycle consisted of 20 seconds at 94 °Cand 1 minute at 70 °C. The entire program was followed by a5-minute extension period at 68 °C. The reactions were sub-jected to an additional successive three cycles and viewed onan ethidium bromide–stained agarose gel to determinewhen the amplification of each product was within the

Table 2 Most differentially expressed genes as identified by cDNA microarray analysis

Array

Upregulated genes in multiple autistic cases Downregulated genes in multiple autistic cases

Gene name GenBank ID Locus I C Gene name GenBank ID Locus I C

UniGEM V2 array Phospholipase A2 M21054 12q23–q24.1 1.85 1.03 KIAA0321 AB002319 14 �1.57

GABAA receptor �5 L08485 15q11.2–q12 1.75 Ribosomal proteinS29

AA715449 14 �1.41

Chemokine receptor 1 D10925 3p21 1.72 Heat shock 70-kdprotein 1

M59828 6p21.3 �1.41

EST AA946611 15q 1.64 SPS2 U43286 — �1.41

Apolipoprotein E K00396 19q13.2 1.64 1.35 Cytochrome P4503A5

X90579 7 �1.40

HLA-G antigen X17273 6p21.3 1.59 LIM protein AF061258 4q22 �1.35

Clusterin X14723 8p21-p12 1.56 DNA fragment AI697803 4p16.3 �1.35

Hevin X86693 4 1.54 EST AA007282 — �1.34

TU3A AF089853 3 1.54 EST AI031686 15 �1.33

MITF N34462 2 1.52 KIAA0913 AI200349 10 �1.33

Neurobiology array EAAT 1 U03504 5p13 1.39 1.86 Somatostatinreceptor 2

M81830 17q24 1.10 �1.62

Protease nexin 1 A03911 2q33–q35 1.76 ALAS 1 X56351 3p21.1 1.11 �1.51

Glutamate receptor,AMPA 1

M64752 5q31.1 1.37 1.55 Histidinedecarboxylase

X54297 15q �1.09 �1.47

GFAP J04569 17q21 1.47 1.52 Cannabinoidreceptor 1

X81120 6q 1.05 �1.40

RAP-1A M22995 1p13.3 �1.11 1.38 Acetylcholinesterase M55040 7q22 �1.12 �1.38

I � Incyte Genomics UniGEM V2 array, mean of nine autistic/control comparisons; C � Clontech Atlas Human Neurobiology array, mean of four autistic/control comparisons, fold regulation; GABA � -aminobutyric acid; EST � expressed sequence tag; TU3A � downregulated in renal cell carcinoma;MITF � microphthalmia-associated transcription factor; EAAT 1 � excitatory amino acid transporter 1; GFAP � glial fibrillary acidic protein; RAP-1A �

ras-related protein 1A; SPS2 � selenophosphate synthetase 2; ALAS 1 � 5-aminolevulinate synthase 1.

1620 NEUROLOGY 57 November (1 of 2) 2001

linear phase. Bands in the linear phase were quantified bymeasuring mean density using NIH image analysis soft-ware (NIH, Bethesda, MD). The ratio of the gene banddensity to the density of its glyceraldehyde-3-phosphatedehydrogenase band was calculated for each sample. Apaired t-test was performed to evaluate differences in geneexpression between control and autism groups. To assesswhether the differences between autism and controlgroups varied as a function of age or PMI, multiple regres-sion analysis was performed.

Subtractive hybridization and southern blotting. TheMicro-FastTrack isolation kit (Invitrogen, Carlsbad, CA)was used to purify poly A� RNA from a pool of total RNA(100 g) from four autism (A1, A2, A4, A9) and threecontrol (C7 to C9) cerebellum samples. Poly A� RNA wasreversed transcribed to cDNA, digested with RsaI, andsubtracted using PCR-Select cDNA Subtraction kit (Clon-tech). Pools of differentially expressed cDNA were sub-cloned into a TA-cloning vector (Invitrogen) andsequenced. The identity of band 4.1N (KIAA0338; Gen-bank accession number XM_047295) was determined byBasic Local Alignment Search Tool searching with �99%nucleotide identity over a region of at least 100 base pairsused as a criterion. The sequence of band 4.1N identified insubtraction spanned base pairs 3726 to 4122 (GenBankidentifier: AB002336). The band 4.1N clone isolated fromsubtraction was random prime labeled with Rediprime II(Amersham Pharmacia Biotech, Arlington Heights, IL) us-ing [32P]deoxycytidine triphosphate. This probe was thenhybridized to a nylon membrane containing pools of up-and downregulated cDNAs generated by subtractivehybridization.

Western blotting. Twenty micrograms of each cerebel-lum sample were loaded onto a 4 to 15% sodium dodecylsulfate polyacrylamide gradient gel (Bio-Rad, Hercules, CA)and electrophoresed at 60 mA for 90 minutes. A referencesample was electrophoresed on every gel so that individualgels could be compared. Proteins were transferred to Protrannitrocellulose membranes (Schleicher & Schuell, Keene, NH)in tromethamine/glycine buffer (2.5 mM tromethamine, 192mM glycine, 20% methanol) overnight at 30 mA. Nonspecificbinding sites were blocked for 1 hour at room temperature in50 mM of tromethamine buffer containing 150 mM of NaCland 0.1% polysorbate 20 with 5% nonfat dry milk. Immuno-staining was performed by incubating the primary antibodyfor 1 hour at room temperature and was followed by horse-radish peroxidase-conjugated secondary for 1 hour at roomtemperature. Signal was detected with an enhanced chemilu-minescence detection system. Bands were quantified by mea-suring mean density using NIH image analysis software. Tostandardize protein levels, the Western blots were reprobedwith an actin antibody. The ratio of the sample’s proteinband intensity to its actin band intensity was calculated. Toevaluate differences between autism and control samples,t-tests were performed. To assess whether the differencesbetween autism and control groups varied as a function ofage or PMI, multiple regression analysis was performed.

mAb339, a mouse anti--aminobutyric acid (GABA)A re-ceptor � chain monoclonal, and AB1504, a rabbit anti-glutamate receptor 1 polyclonal, were used at a dilution of1:50 (Chemicon International, Temecula, CA). A2066, arabbit antiactin polyclonal (Sigma Chemical, St. Louis,MO), was used at a dilution of 1:500 and a mouse anti–

neuron-specific enolase (NSE) polyclonal was used at adilution of 1:100 (Calbiochem, San Diego, CA). Rabbit an-ti–excitatory amino acid transporter 1 (EAAT 1) and anti-EAAT 2 antibodies were kindly provided by Dr. JeffreyRothstein. Rabbit polyclonal antibodies against metabo-tropic glutamate receptor 2/3 (AB1553; 1:100), NMDA re-ceptor 1 (AB1516; 1:100), and glutamate decarboxylase 1/2(AB1511; 1:500) were purchased from Chemicon Interna-tional. The rabbit antiband 4.1N antibody was supplied byDr. Solomon Snyder. The rabbit anti-�-amino-3-hydroxy-5-methyl-4-isoxazoleproprionic acid (AMPA) glutamate re-ceptor 2/3, rabbit antiglutamate receptor interactingprotein (GRIP), and rabbit antibodies against postsynapticdensity protein 95 (PSD-95) were generously supplied byDr. Richard Huganir. Horseradish peroxidase-labeled goatantimouse secondary and goat antirabbit secondary (Jack-son Immunoresearch Laboratories, West Grove, PA) wereused at a dilution of 1:5000.

Receptor autoradiography. Glutamate receptor densitywas evaluated in the cerebellum (A1, A3 to A7; C1 to C3,C5 to C10), prefrontal cortex (A1, A2, A4, A9; C6, C8, C19to C23), and caudate-putamen (A1, A2, A4; C9, C17, C18,C21 to 23) of autism and control postmortem brain tissue.Frozen brain samples were cut serially (20 m thick) on acryostat and then thaw-mounted onto Superfrost Plus–coated glass slides (VWR, West Chester, PA). Sectionswere processed for binding to AMPA and NMDA receptorsas described previously.20

Receptor densities were measured using a video-basedimage analysis system (Imaging Research, St. Catherines,Ontario). For each section, the densities of 25 small re-gions within the area of interest were averaged. The meanbackground was subtracted from this value. The averagedensities from four to eight separate sections from thesame brain sample were then averaged. The statisticalpackage S-PLUS 2000 (Insightful Corp., Seattle, WA) wasused to perform t-tests and multiple regression analysis foreach receptor type to determine whether receptor densitydifferences between autism and control samples varied asa function of age or PMI.

Results. Expression profiling using cDNA microar-rays. Two commercially available cDNA microarray tech-nologies were used to identify differences in geneexpression between autism and control postmortem cere-bellum. Radiolabeled cDNA probes from four autism andfour matched control samples were hybridized onto indi-vidual Atlas Neurobiology arrays (n � 8 arrays). Of 588cDNA on the array, 357 gave hybridization signals thatwere above background on every array. Onto the UniGEMV2 array, fluorescently labeled probes from nine autismsamples were each hybridized simultaneously againstprobes generated from the same control pool (n � 9 ar-rays). Of the 9374 cDNA, 8977 gave hybridization signalsthat were above background and met area criteria (see“Methods”) in at least six of the nine arrays. For eacharray technology, the top up- and downregulated genes inautism are shown in table 2. Fewer genes are listed fromthe Atlas Neurobiology array as the array contained farfewer genes than the UniGEM V2 array. Gene expressionresults were generally consistent between the two arraytechnologies (see table 2). A few of the genes, such as

November (1 of 2) 2001 NEUROLOGY 57 1621

GABAA receptor �5, lie within autism susceptibility locithat genome screens have identified.21

Differences in gene expression between the two clinicalgroups can be visualized by constructing a scatterplot ofautism intensity versus control intensity (figure 1). Eachpoint in the scatterplot represents one of the genes on thearray. The majority of genes were expressed at the samelevel in both autism and control cerebellum and thereforelie on or near the line that bisects the data. Genes that areupregulated in autism lie above the line, whereas genesthat are downregulated in autism lie below the line. Sev-eral genes that are significantly up- or downregulated inautism (see table 2) are labeled in the scatterplots. Al-though mean autism and control intensities were used forthe Atlas Neurobiology array scatterplot, intensity valuesfrom a representative UniGEM V2 array were used, asintensity values cannot be averaged across arrays usingthis technology. The expressed sequence tag (EST), whichis labeled with an asterisk in the UniGEM V2 scatterplot(see figure 1B), is highly downregulated in this array ex-periment; however, on average it is not downregulatedacross all nine experiments. The mRNA levels of genes

that were the focus of this study were consistently in-creased in autism cerebellum. Figure 1 shows the fold reg-ulation in all four (Atlas Neurobiology array; see figure 1A,right) or all nine (UniGEM V2 array; see figure 1B, right)array experiments. Histograms of all the ratios from allarray experiments were constructed to confirm that thedata were normally distributed (data not shown). The av-erage ratio � SD, calculated using fold regulations from allnine UniGEM V2 arrays, was 1.04 � 0.09 and for all fourAtlas Neurobiology comparisons was 1.02 � 0.15.

Reverse transcriptase PCR confirmation. Reverse tran-scriptase PCR was used to confirm the differential expres-sion of genes identified by the cDNA microarray analysis.Ten samples from the cerebella of subjects with autismand 10 matched controls were analyzed. Genes were cho-sen for RT-PCR from 1) the most up- and downregulatedgenes in table 2, 2) genes that showed consistent up- ordownregulation on both the Atlas Neurobiology and Uni-GEM V2 arrays, and 3) genes that have a strong rationalefor being involved in the pathogenesis of autism. Increasesin the mRNA levels of several genes in autism were con-firmed using RT-PCR (figure 2A). EAAT 1 was the most

Figure 1. Scatterplots show controlintensity versus autism intensity forgenes on the Atlas Neurobiology arrayand the UniGEM V2 (Incyte Genom-ics) array. Genes above the line areupregulated in autism whereas thosebelow the line are downregulated inautism. Most genes are expressed atsimilar levels in both clinical groupsand therefore lie close to or on theline. (A) Scatterplot of 357 gene inten-sities from the Atlas Neurobiology ar-ray. Each dot represents the averagegene intensity from four different ex-periments. Three upregulated as wellas two downregulated genes are indi-cated (arrows). The three graphs tothe right of the scatterplot representgenes of interest from the Atlas Neuro-biology array. The fold regulation ofthe gene in each of the four array ex-periments is plotted to demonstratethe variance in fold regulation be-tween experiments. EAAT 1 � excita-tory amino acid transporter 1; GluR1� glutamate receptor, ionotropic,AMPA 1; GFAP � glial fibrillaryacidic protein; ALAS 1 � 5-amino-levulinate synthase 1. (B) Representa-tive scatterplot of approximately 8977gene intensities from the UniGEM V2array. The scatterplot represents oneof the nine experiments. Three upregu-lated as well as two downregulatedgenes are labeled (arrows). The three

graphs to the right of the scatterplot represent genes of interest from the UniGEM V2 array. The fold regulation of thegene in each of the nine array experiments is plotted to demonstrate the variance in fold regulation between experiments.Hsp 70 � heat shock 70-kd protein 1; PLA 2 � phospholipase A2. *Expressed sequence tag (EST) with GenBank identifierN50593. †EST with GenBank identifier AA94611. GenBank identifiers for other genes labeled in both scatterplots areprovided in table 2.

1622 NEUROLOGY 57 November (1 of 2) 2001

upregulated gene in autism on the Atlas Neurobiology ar-ray. RT-PCR also demonstrated that EAAT 1 mRNA wasincreased in autism (p � 0.05). Glial fibrillary acidic pro-tein (GFAP) and glutamate receptor AMPA 1 (GluR1) wereadditional genes that the Atlas Neurobiology array demon-strated were upregulated in the cerebella of subjects withautism. Levels of GFAP and GluR1 mRNA, assayed byRT-PCR, were increased in the cerebellum samples of theautism group (p � 0.002 and p � 0.05).

GABAA receptor �5 and clusterin were among the mostupregulated genes on the UniGEM V2 array and also hadgreater mRNA amounts in autism when assayed by RT-PCR (p � 0.01). The C8 control sample in the GABAA

receptor RT-PCR (see figure 2A, dagger) was eliminatedfrom the statistical analysis of this gene because it wasdetermined to be an outlier. An EST (GenBank identifier:AA94611) was one of the top 10 upregulated genes on theUniGEM V2 array. This gene is of interest because it lieswithin one of the autism susceptibility loci, 15q.21 In addi-tion, its transcript level, determined by RT-PCR, was in-creased in the cerebella of individuals with autism (p �

0.005). Although glycoprotein M6b was not among the topregulated genes, both array technologies consistentlyshowed that glycoprotein M6b mRNA was increased inautism. This increase was confirmed by RT-PCR (p �0.05). This gene may also warrant further study because itlies on the X chromosome, a chromosome that has beenhypothesized to contain a defect in autism due to thegreater number of men with the disorder.3 Multiple regres-sion analyses determined that none of the differences inmRNA levels between clinical groups were a function ofage or PMI.

Transcript levels of several other genes were examinedby RT-PCR, including chemokine receptor 1, apoE,KIAA0321, heat shock 70-kd protein 1, protease nexin 1,5-aminolevulinate synthase 1, and acetylcholinesterase.However, no significant differences in the mRNA levelsbetween autism and control samples were detected (datanot shown).

mRNA levels of glutamate-related genes. Significantlyhigher levels of GluR1 and EAAT 1 mRNA in postmortemautism cerebellum were identified by microarray analysis

Figure 2. Levels of mRNA in 10 autism(A � autism) and 10 control (C � con-trol) postmortem cerebellum sampleswere compared using reverse transcrip-tase PCR (RT-PCR). The mean autismversus mean control band intensity (�SEM) is summarized to the right ofeach RT-PCR. The Y axis of each graphrepresents average normalized intensityof cDNA product signals. (A) RT-PCRconfirmation of altered mRNA levels inautism. All gene transcripts are in-creased in autism postmortem cerebel-lum (*p � 0.05). (B) Investigation ofmRNA levels of additional glutamate-related genes. Excitatory amino acidtransporter 2 (EAAT 2) and the AMPA-type glutamate receptors 2 and 3 haveincreased mRNA levels in autism (*p �0.05). Glyceraldehyde-3-phosphate de-hydrogenase band intensity (right,lower panel) was used for normaliza-tion of cDNA product in each sample.†Expressed sequence tag (EST) withGenBank identifier AA94611. GABA ��-aminobutyric acid.

November (1 of 2) 2001 NEUROLOGY 57 1623

and confirmed with RT-PCR. Therefore, the mRNA levelsof several other glutamate receptors and glutamate trans-porters were examined in autism. The same samples fromthe cerebellum of 10 subjects with autism and 10 controlsubjects were compared. The mRNA levels of two AMPA-type glutamate receptors, GluR2 and GluR3, were in-creased in cerebellum samples from individuals withautism (p � 0.002 and p � 0.03; see figure 2B). However,other glutamate receptor types such as kainate 1, metabo-tropic 3, and NMDA 1 displayed no significant differencein their expression levels (data not shown). The mRNA of aglutamate transporter, EAAT 2, was increased (p � 0.02)in samples from persons with autism. Multiple regressionanalysis demonstrated that none of the significant mRNAdifferences between autism and control groups were afunction of age or PMI. Members of other neurotransmittersystems were also examined in the same cerebellum sam-ples using RT-PCR. However, no significant differences inthe mRNA levels of dopamine receptor 2, neuropeptide Y6receptor, cholinergic receptor beta 1, purinergic receptorP2X 4, purinergic receptor P2Y1, adenosine A2b receptor,GABAA receptors �2, �3, and 3, the GABA transporter, orthe GABA synthesizing enzyme glutamate decarboxylase 1were found between autism and control cerebellum (datanot shown).

Subtractive hybridization. In a complementary ap-proach to microarray analysis, we performed subtractivehybridization to identify mRNA transcripts that were dif-ferentially regulated in the cerebellum of individuals withautism. Band 4.1N (KIAA0338; GenBank identifier:AB002336) was one of the 40 cDNA clones that were iso-lated. A complete analysis of the additional sequences canbe found elsewhere (Jeon and Pevsner, in preparation).Band 4.1N is a neuronal protein that binds to the C termi-

nus of GluR1 and may link it to the actin cytoskeleton.22

Subtraction identified this sequence as one that was up-regulated in autism cerebellum compared with control cer-ebellum. Southern blotting was used to confirm thepresence of this transcript in the upregulated pool of cDNA(data not shown).

Analysis of protein levels. Although levels of mRNA donot necessarily correspond to protein levels, it is of interestto determine whether the mRNA differences found in au-tism are also apparent at the protein level. The proteinlevels of these gene products were then examined by West-ern blotting (figure 3). Nine autism and 11 age- andgender-matched control postmortem cerebellum sampleswere analyzed.

The protein of the glial glutamate transporter EAAT 1,which appears as a doublet at approximately 60 kd, wasincreased threefold in autism cerebellum samples com-pared with controls (p � 0.001; see figure 3A). Becauseseizures are present in about 33% of individuals with au-tism,4 it may be that this increase is a compensatory mech-anism that functions to decrease glutamate excitation.Reviewing the four subjects with a history of seizures (A1,A2, A4, A10), however, samples A2 and A1 exhibit rela-tively low EAAT 1 protein levels compared with other au-tism samples. Sample A4 does indeed show high levels ofEAAT 1 protein, whereas sample A10 was not examined.Glutamate receptor AMPA 1 protein, also part of the glu-tamate system, is increased on average in autism cerebel-lum samples 2.5-fold (p � 0.0001). This may reflect ageneral disturbance in the glutamate system as a conse-quence of autism. Multiple regression analyses demon-strated that the increases in GluR1 and EAAT 1 proteinwere not a function of age or PMI. Using an antibodyspecific to the � chain of the GABAA receptor, Western

Figure 3. Protein levels of several genesin control (n � 11) and autism (n � 9)postmortem cerebellum samples. Quan-tification of immunoblots to the right ofeach blot shows mean autism versusmean control band intensities (� SEM).(A) Protein levels of genes identified bymicroarray analysis. Protein levels ofexcitatory amino acid transporter 1(EAAT 1) and glutamate receptor, iono-tropic, AMPA 1 (GluR1) are increasedin the cerebellum samples from autisticsubjects (*p � 0.001). Levels of neuron-specific enolase (NSE), a neuron-specificmarker, are similar in both clinicalgroups. The right panel, bottom, showsthe actin immunoblot used to standard-ize sample band intensity. (B) Proteinlevels of glutamate-related genes. Therewas an upregulation of the NMDA re-ceptor 1 and EAAT 2 as well as the glu-tamate receptor binding proteins band4.1N and glutamate receptor interact-ing protein (GRIP) (*p � 0.05). C �control; A � autism; R � reference.

1624 NEUROLOGY 57 November (1 of 2) 2001

blotting demonstrated that levels of this protein were in-creased, although not significantly (data not shown). Thisparticular antibody broadly recognized all � subunits andwas not specific for the �5 subunit. A potential upregula-tion of the �5 subunit protein could be masked by theprotein expression of other subunits. Protein levels of NSE,a marker that specifically recognizes neurons, were as-sessed to determine whether the number of neurons wassimilar in autism and control samples. Western blottingdemonstrated that the amount of NSE protein was similarin both clinical groups (see figure 3A).

Because there was a significant increase in both GluR1and EAAT 1 proteins in autism, Western blotting was usedto determine whether there were differences in the proteinamounts of additional glutamate-related genes in the samecerebellum samples. There was an increase in the proteinof EAAT 2 and the NMDA receptor 1 in autism (p � 0.04and p � 0.05; see figure 3B). Conversely, no significantdifferences were detected in the protein levels of glutamatedecarboxylase 1/2, metabotropic 2/3 receptors, or AMPA2/3 glutamate receptors (data not shown).

Protein levels of band 4.1N, which was identified bysubtractive hybridization, and other glutamate receptorbinding proteins were also assessed in the same cerebel-lum samples. Band 4.1N protein was increased in autismcerebellum (p � 0.05; see figure 3B). The C termini ofAMPA receptors have also been shown to interact withGRIP.23 GRIP protein was increased in samples from sub-

jects with autism (p � 0.04) when compared with controlsamples. However, PSD-95, a protein that associates withthe C termini of the NMDA receptor subunits,24 was notsignificantly different in autism cerebellum (data notshown). Multiple regression analysis demonstrated thatnone of the significant protein differences between autismand control groups were a function of age or PMI. Severalother genes involved in synaptic transmission were exam-ined by Western blotting. There was no significant differ-ence in GABAA receptor � subunit, glutamatedecarboxylase 1/2, purinergic receptor P2X 7, or dopaminereceptor 2 protein amount between control and autisticcases (data not shown).

Glutamate receptor autoradiography. The density ofNMDA- and AMPA-type glutamate receptors was exam-ined by receptor autoradiography, due to the mRNA andprotein abnormalities detected by Western blotting, RT-PCR, and microarray analysis. Glutamate receptor densitywas evaluated in the cerebellum of six subjects with au-tism and nine control subjects. AMPA glutamate receptordensity was decreased in both the granule cell layer (p �0.05) and molecular cell layer (p � 0.01) of the autismcerebellum (figure 4A). Multiple regression analysis dem-onstrated that these receptor density differences betweenautism and control groups were not a function of age orPMI. Unlike AMPA receptor density, NMDA receptor den-sity in these regions did not differ significantly betweencontrol and autism groups. Figure 4B shows NMDA and

Figure 4. Glutamate receptor autora-diography. (A) Density of AMPA- andNMDA-type glutamate receptors (�SEM) in the cerebellum (granule andmolecular cell layers), prefrontal cortex,and caudate-putamen of subjects withautism (filled bar) and control subjects(open bar). There is a decrease in thedensity of AMPA receptors in the gran-ule cell layer (*p � 0.05) and the molec-ular cell layer (**p � 0.01) of patientswith autism compared with matchedcontrols. (B) Pseudocolor images showsignificantly decreased [3H]AMPA bind-ing in the cerebellum of a representativeautism brain section (top, left) com-pared with a representative controlbrain section (top, right). There was nosignificant difference in [3H]CGP39653binding to NMDA receptors in a repre-sentative autism cerebellum section(bottom, left) compared with a represen-tative control cerebellum section (bot-tom, right).

November (1 of 2) 2001 NEUROLOGY 57 1625

AMPA receptor density in representative autism and con-trol cerebellum sections. The prefrontal cortex (four autis-tic versus seven control cases) and caudate-putamen (threeautistic versus six control cases) were also assessed fordifferences in AMPA or NMDA receptor density; however,no significant differences were detected (see figure 4A).

Discussion. Several biochemical differences havebeen reported in biologic samples from subjects withautism, yet few, if any, of these studies examinedpostmortem brains. The study of the brains of indi-viduals with autism is important because autism isconsidered a neurodevelopmental disorder and pa-thology in several regions is known to exist. Thestudy of postmortem human brain, however, pre-sents unique experimental challenges.12 As we areexamining postmortem samples long after the onsetof the disorder, it is more likely that we are identify-ing secondary consequences of the disorder. How-ever, at this time no animal model exists that can beused to study the developmental biology of autism.In addition, the quality of RNA isolated from post-mortem brain may be variable and can depend onthe agonal state (i.e., the events preceding death,such as hypoxia or other trauma to the brain) andthe PMI.25 The quality of RNA may be predicted fromthe pH of the brain,19 and in the present study thepH of samples was consistent and within an accept-able range. We also confirmed the integrity of RNAby gel electrophoresis and absorbance measure-ments. Multiple regression analysis demonstratedthat no mRNA or protein differences between thetwo groups were attributed to PMI. Previous studiesindicate that glutamate receptors remain stable forpostmortem and storage intervals that were similarto ours.26

Autism is a heterogeneous disorder and it is there-fore important to take into consideration clinicalvariations such as IQ and the presence of seizures.Because glutamate plays a role in the initiation andpropagation of seizures, we sought to correlate sig-nificant glutamate differences detected in this studyto subjects with a history of seizures. Four of the 10subjects with autism were affected by seizures; how-ever, we detected no significant correlation betweenthese cases and the magnitude of mRNA or proteinchanges. A majority of the prefrontal cortex andcaudate-putamen samples from the autism group ex-amined by autoradiography, however, were fromsubjects who had seizures, making this correlationdifficult. It is likely that the heterogeneity of thesesamples obscured the identification of additionalgenes that are abnormally regulated in autism. Inaddition, several of the patients were taking medica-tions, some of which could potentially affect geneexpression in brain.

cDNA microarray analysis provided several direc-tions for autism research by identifying moleculesthat have never been associated with autism. GFAPand clusterin mRNA were significantly increased inthe cerebellum samples from subjects with autism in

both RT-PCR and microarray experiments. In-creased mRNA levels of clusterin are seen in a vari-ety of disorders, such as AD; however, its physiologicfunction is unknown.27 One study found that proteinlevels of GFAP in CSF were almost three timeshigher in a group of 47 individuals with autism com-pared with similarly aged normal individuals.28 Inaddition, a significant increase in the incidence ofGFAP autoantibodies was found in the plasma of 53subjects with autism compared with 58 controls.29

GFAP, a biochemical marker of astrocytes, increasesduring astroglial activation.30 If the increase inGFAP mRNA we detected in autism is indicative ofan increase in GFAP protein, it is possible that reac-tive gliosis may contribute to autism pathophysiol-ogy. The few studies that have examinedpostmortem brain from subjects with autism havenot identified gliosis or an increase in GFAP pro-tein5; however, it may be mild or restricted to certainareas of the brain. Additional pathologic studies ex-amining a greater number of cases and a variety ofregions are therefore needed. Several autism suscep-tibility loci have been identified by genome-widescreens21 and many chromosomal aberrations havebeen linked with autism.31 A few genes that werewithin these regions were found to be up- or down-regulated in autism in our analyses, including anEST (GenBank identifier AA946611), GABAA recep-tor �5 subunit, and glycoprotein M6b.

cDNA microarray analysis also led to the identifi-cation of glutamate abnormalities in the cerebellumof individuals with autism. Abnormalities in seroto-nin, dopamine, norepinephrine, opiates, and severalother neurotransmitters have been reported in au-tism.32 The potential pathophysiologic role of gluta-mate, the most abundant excitatory neurotrans-mitter in the brain, has not been explored to thesame extent. There are several reasons that a gluta-mate hypothesis of autism is attractive. Receptorsfor glutamate are especially concentrated in regionsthat have been repeatedly implicated in autism, in-cluding the cerebellum and hippocampus.33,5 In addi-tion, autism is believed to be a developmentaldisorder of the brain and during development gluta-mate plays a crucial role in the formation of braincytoarchitecture by regulating processes such asneuronal outgrowth and synaptogenesis.34 The distri-bution, electrophysiology, and molecular characteris-tics of glutamate receptors change markedly duringthis period, making the formation of the brain vul-nerable to aberrations in glutamate neurotransmis-sion.34 In the adult brain, the NMDA-type glutamatereceptor is required for long-term potentiation, thephysiologic process underlying learning and memo-ry.35 It has been speculated that at least some of thebehavioral abnormalities of autism can be attributedto a memory deficit.36 Further, a study demonstratedthat glutamate may be important in the acquisitionof emotional behavior.37 Disrupted glutamate trans-mission could therefore account for a constellation ofthe cognitive deficits in autism. Finally, symptoms

1626 NEUROLOGY 57 November (1 of 2) 2001

elicited by glutamate receptor antagonist treatment,such as extreme focus on trivial details and heightenedor distorted pain perception, are similar to symptomsof autism.38 Blockade of glutamate receptors may alsoimprove autistic symptoms.39,40

Receptor autoradiography demonstrated thatAMPA receptor density was significantly decreasedin the cerebellum of subjects with autism. Expres-sion of these receptors could be decreased for a vari-ety reasons, including inefficient aggregation andimmobilization in the synapse or a decrease in thenumber of cell types that express AMPA receptors.Interactions with several proteins, including GRIPand band 4.1N, are thought to be important for thelocalization and synaptic expression of AMPA recep-tors on the cell surface.23 For instance, disruption ofthe interaction between GluR1 and band 4.1N incultured cells has been shown to reduce the level ofsurface AMPA receptors.22 In this study, Westernblotting demonstrated that GRIP protein was signif-icantly increased in postmortem autism cerebellum.We also showed that the mRNA and protein levels ofband 4.1N were expressed at abnormally high levelsin autism. Thus, abnormalities in the proteins thatare responsible for glutamate receptor synaptic local-ization, clustering, or immobilization, such as band4.1 or GRIP, may be the reason for reduced expres-sion of AMPA-type glutamate receptors. Althoughautoradiography determined there was a decrease inAMPA receptors, Western blotting indicated an in-crease in the AMPA 1 glutamate receptor in totalbrain homogenate. The explanation for this discrep-ancy is unknown; however, the ligands used in thereceptor density experiments broadly recognized alltypes of AMPA receptors. Kinetic analyses of gluta-mate receptor binding may help to address thisissue.

In addition to the abnormalities detected in theAMPA receptors, both the mRNA and protein levelsof EAAT 1 and 2 were significantly increased in au-tism cerebellum. EAAT 1 and EAAT 2 are predomi-nantly located on astroglia and their main functionis to remove glutamate from the extrasynapticspace.41 Studies have shown that the protein leveland activity of glutamate transporters are controlledby the extracellular glutamate concentration or byactivity-dependent mechanisms.42 This suggests thatincreased EAAT 1 and 2 protein levels in the autismgroup may be due to above-normal extracellular glu-tamate concentrations. In fact, cultured astrocytesexposed to glutamate for prolonged periods acquirean increased capacity for glutamate uptake and in-creased expression of EAAT 1 protein.43 Becausemany neurons in the cerebellum use glutamate as aneurotransmitter, and the glutamate transportersare the only mechanism to reduce concentrations ofextracellular glutamate, aberrations in their func-tioning may have severe pathologic effects such asexcitotoxicity.44 Alternatively, the increase in EAAT1 and 2 protein could be explained by an increasedlevel of glutamate transmission or innervation in

neurons, requiring increased expression of gluta-mate transporters. Currently, there is no pathologicevidence in autism to support this theory.

Pathologic studies of postmortem cerebellum ofautistic persons have shown variable reductions inthe number of Purkinje cells.5-7 The increased repre-sentation of molecules expressed on the remainingcells such as glia could therefore have resulted inincreased EAAT 1 and EAAT 2 protein. Microarrayanalysis of gene expression in autism cerebellum,however, did not detect significantly decreased ex-pression of neuron-specific markers. In addition,Western blotting of postmortem cerebellum homoge-nates demonstrated that the neuronal marker NSEwas unchanged in autism. Because the glutamatesystem is involved in the development of brain cyto-architecture, aberrations in the system during devel-opment may have disrupted synaptic connections orthe formation of certain cell types.34 This, in turn,may have affected the numbers and distribution ofglutamate receptors and glutamate transporters inthe brain of persons with autism. Additional patho-logic studies of postmortem brain from individualswith autism confirming Purkinje cell loss and de-scribing more fully abnormalities in cytoarchitecturewould be extremely informative. In the future, itmay also be interesting to assess the expression ofAMPA receptors and glutamate transporters in addi-tional brain regions with known pathology and atyounger ages, if postmortem tissue were available.

Additional research into the glutamatergic systemin autism is warranted as it may provide a rationalefor treatment. Clinical observations suggest that ket-amine anesthesia, which specifically blocks NMDAreceptors, has beneficial effects (calming, focused at-tention) on children with autism.40 However, moremust be known about the nature of the glutamatesystem’s involvement in the pathogenesis of autismin order to justify clinical research in this area. Thisis especially true because both increased and de-creased levels of glutamate have been shown to bedetrimental in the developing brain. Because autismis currently diagnosed solely on the basis of clinicalmanifestations, additional neurochemical researchwill be useful in identifying markers for the disorderor in identifying more homogeneous subgroups. For astudy of this nature, it would be crucial to obtain alarge number of quality postmortem brain samples,allowing for glutamatergic disturbances in each pa-tient to be compared in multiple brain regions acrossexperiments.

AcknowledgmentThe authors thank Karim Hyder and Tae H. Chong for theirtechnical assistance. The authors also thank Drs. Jeffrey Roth-stein, Solomon Snyder, and Richard Huganir for providingantibodies.

References1. Kanner L. Autistic disturbances of affective content. Nervous

Child 1943;2:217–250.

November (1 of 2) 2001 NEUROLOGY 57 1627

2. Rapin I, Katzman R. Neurobiology of autism. Ann Neurol1998;43:7–14.

3. Fombonne E. The epidemiology of autism: a review. PsycholMed 1999;29:769–786.

4. Olsson I, Steffenburg S, Gillberg C. Epilepsy in autism andautisticlike conditions: a population-based study. Arch Neurol1988;45:666–668.

5. Kemper TL, Bauman M. Neuropathology of infantile autism.J Neuropathol Exp Neurol 1998;57:645–652.

6. Williams RS, Hauser SL, Purpura DP, et al. Autism and men-tal retardation: neuropathologic studies performed in four re-tarded persons with autistic behavior. Arch Neurol 1980;37:749–753.

7. Ritvo ER, Freeman BJ, Scheibel AB, et al. Lower Purkinje cellcount in the cerebella of four autistic subjects: initial findingsof the UCLA-NDAC autopsy research report. Am J Psychiatry1986;143:862–866.

8. Hashimoto T, Tayama M, Murakawa K, et al. Development ofthe brainstem and cerebellum in autistic patients. J AutismDev Disord 1995;25:1–18.

9. Courchesne E. Brainstem, cerebellar and limbic neuroana-tomical abnormalities in autism. Curr Opin Neurobiol 1997;7:269–278.

10. Brown PO, Botstein D. Exploring the new world of the ge-nome with DNA microarrays. Nat Genet 1999;21(suppl 1):33–37.

11. Duggan DJ, Bittner M, Chen Y, et al. Expression profilingusing cDNA microarrays. Nat Genet 1999;21(1 suppl):10–14.

12. Colantuoni C, Purcell AE, Bouton CML, et al. High through-put analysis of gene expression in the human brain. J Neuro-sci Res 2000;59:1–10.

13. Whitney LW, Becker KG, Tresser NJ, et al. Analysis of geneexpression in multiple sclerosis lesions using cDNA microar-rays. Ann Neurol 1999;46:425–428.

14. Ginsberg SD, Hemby SE, Lee VM, et al. Expression profile oftranscripts in Alzheimer’s disease tangle-bearing CA1 neu-rons. Ann Neurol 2000;48:77–87.

15. American Psychiatric Association. Diagnostic and statisticalmanual of mental disorders, 4th ed. Washington, DC: Ameri-can Psychiatric Association, 1994.

16. Lord C, Rutter M, Le Couteur A. Autism Diagnostic Inter-view-Revised: a revised version of a diagnostic interview forcaregivers of individuals with possible pervasive developmen-tal disorder. J Autism Dev Disord 1994;24:659–685.

17. Schopler E, Reichler RJ, DeVillis RF, et al. Toward objectiveclassification of childhood autism: childhood autism ratingscale (CARS). J Autism Dev Disord 1999;40:343–355.

18. Bradford MM. A rapid and sensitive method for the quantita-tion of microgram quantities of protein utilizing the principleof protein-dye binding. Anal Biochem 1976;72:248–254.

19. Kingsbury AE, Foster OJF, Nisbet AP, et al. Tissue pH as anindicator of mRNA preservation in human post-mortem brain.Mol Brain Res 1995;28:311–318.

20. Blue ME, Naidu S, Johnston MV. Altered development ofglutamate and GABA receptors in the basal ganglia of girlswith Rett syndrome. Exp Neurol 1999;156:345–352.

21. Lamb JA, Moore J, Bailey A, et al. Autism: recent moleculargenetic advances. Hum Mol Genet 2000;9:861–868.

22. Shen L, Liang G, Walensky LD, et al. Regulation of AMPAreceptor GluR1 subunit surface expression by a 4.1N-linkedactin cytoskeletal association. J Neurosci 2000;20:7932–7940.

23. Dong H, O’Brien RJ, Fung ET, et al. GRIP: a synaptic PDZdomain-containing protein that interacts with AMPA recep-tors. Nature 1997;386:279–284.

24. O’Brien RJ, Lau LF, Huganir RL. Molecular mechanisms ofglutamate receptor clustering at excitatory synapses. CurrOpin Neurobiol 1998;8:364–369.

25. Johnston NL, Cervenak J, Shore AD, et al. Multivariate anal-ysis of RNA levels from postmortem human brains as mea-sured by three different methods of RT-PCR. StanleyNeuropathology Consortium. J Neurosci Methods 1997;77:83–92.

26. Freund G, Anderson KJ. Glutamate receptors in the frontalcortex of alcoholics. Alcohol Clin Exp Res 1996;20:1165–1172.

27. Rosenberg ME, Silkensen J. Clusterin: physiologic and patho-physiologic considerations. Int J Biochem Cell Biol 1995;27:633–645.

28. Ahlsen G, Rosengren L, Belfrage M, et al. Glial fibrillaryacidic protein in the cerebrospinal fluid of children with au-tism and other neuropsychiatric disorders. Biol Psychiatry1993;33:734–743.

29. Singh VK, Warren R, Averett R, et al. Circulating autoanti-bodies to neuronal and glial filament proteins in autism. Pedi-atr Neurol 1997;17:88–90.

30. Eng LF, Ghirnikar RS. GFAP and astrogliosis. Brain Pathol1994;4:229–237.

31. Gillberg C. Chromosomal disorders and autism. J Autism DevDisord 1998;28:415–425.

32. Anderson GM. Studies on the neurochemistry of autism. In:Bauman ML, Kemper TL, eds. The neurobiology of autism.Baltimore: The Johns Hopkins University Press, 1994:227–242.

33. Ozawa S, Kamiya H, Tsuzuki K. Glutamate receptors in themammalian central nervous system. Prog Neurobiol 1998;54:581–618.

34. Johnston MV. Neurotransmitters and vulnerability of the de-veloping brain. Brain Dev 1995;17:301–306.

35. Malenka RC, Nicoll RA. Long-term potentiation: a decade ofprogress? Science 1999;285:1870–1874.

36. Killiany RJ, Moss MB. Memory function in autism. In: Bau-man ML, Kemper TL, eds. The neurobiology of autism. Balti-more: The Johns Hopkins University Press, 1994:170–194.

37. Hoehn-Saric R, McLeod DR, Glowa JR. The effects of NMDAreceptor blockade on the acquisition of a conditioned emo-tional response. Biol Psychiatry 1990;30:170–176.

38. Carlsson ML. Hypothesis: is infantile autism a hypoglutamater-gic disorder? Relevance of glutamate–serotonin interactions forpharmacotherapy. J Neural Transm 1998;105:525–535.

39. Zimmerman AW, Myers SM. Magnesium, excitotoxicity, andgrowth of neuropil in autism. J Dev Learning Disord 1999;3:147–159.

40. Zimmerman AW, Hubner KF, Frye VH, et al. Ketamine anes-thesia for MRI and PET in children with autism: behavioralresponses and reduced cerebellar glucose utilization. Neurol-ogy 1990;40(suppl):359. Abstract.

41. Danbolt NC. The high affinity uptake system for excitatoryamino acids in the brain. Prog Neurobiol 1994;44:377–396.

42. Levy LM, Lehre KP, Walaas SI, et al. Down-regulation of glialglutamate transporters after glutamatergic denervation in therat brain. Eur J Neurosci 1995;7:2036–2041.

43. Gegelashvili G, Civenni G, Racagni G, et al. Glutamate recep-tor agonists up-regulate glutamate transporter GLAST in as-trocytes. Neuroreport 1996;8:261–265.

44. Maragakis NJ, Rothstein JD. Glutamate transporters in neu-rologic disease. Arch Neurol 2001;58:365–370.

1628 NEUROLOGY 57 November (1 of 2) 2001