Predictors of Cognitive Test Patterns in Autism Families

Transcription

1 J. Child Psychol. Psychiat. Vol. 40, No. 7, pp , 1999 Cambridge University Press 1999 Association for Child Psychology and Psychiatry Printed in Great Britain. All rights reserved $ Predictors of Cognitive Test Patterns in Autism Families S. E. Folstein S. L. Santangelo New England Medical Center Tufts University School of Medicine, Boston and Eunice Kennedy Shriver Center for Developmental Disorders, Waltham, U.S.A. New England Medical Center Tufts University School of Medicine and Harvard University School of Public Health, Boston, U.S.A. S. E. Gilman J. Piven New England Medical Center Tufts University University of Iowa, Iowa City, U.S.A. School of Medicine, Boston, U.S.A. R. Landa J. Lainhart Johns Hopkins University, Baltimore, U.S.A. University of Utah School of Medicine, Salt Lake City, U.S.A. J. Hein M. Wzorek New England Medical Center Tufts University School of Medicine, Boston, U.S.A. Children s National Medical Center, Washington DC, U.S.A. In a case-control study of cognitive performance, tests of intelligence, reading, spelling, and pragmatic language were administered to the parents and siblings of 90 communityascertained probands with autism (AU group) and to the parents and siblings of 40 similarly ascertained probands with trisomy 21 Down syndrome (DS group). The two samples were comparable for age and parents education; both groups were well-educated and had aboveaverage intelligence. AU parents scored slightly but significantly lower on the WAIS-R Full Scale and Performance IQ, on two subtests (Picture Arrangement and Picture Completion), and on the Word Attack Test (reading nonsense words) from the Woodcock-Johnson battery. There were no differences between AU and DS siblings. As in earlier studies, AU parents, more often than DS parents, reported a history of early language-related cognitive difficulties; we were not able to replicate this in siblings. AU parents who reported such difficulties scored significantly lower on Verbal IQ, spelling, and the nonsense reading test. AU parents without a history of early language-related cognitive difficulties often had a Verbal IQ that exceeded Performance IQ by more than one standard deviation. AU siblings with early language-related difficulties had similar findings: lower Verbal IQ, poorer spelling, and poorer reading scores, compared to AU siblings without such a history. Parents with a positive history also scored worse on a measure of pragmatic language,the Pragmatic Rating Scale, but not on measures of social-related components of the broader autism phenotype. We propose that cognitive differences in a subset of autism family members are manifestations of the language-related component of the broader autism phenotype, and separate from the social-related component. This is consistent with the hypothesis that there are several genes that may interact to cause autism which segregate independently and have distinguishable manifestations in family members. The hypothesis would be further supported by finding different patterns of genetic loci linked to autism in families where one or both parents has language difficulties. Keywords: Autistic disorder, cognition, genetics, intelligence, language, reading disorder, spelling disorder. Abbreviations: ADI: Autism Diagnostic Interview; AU: autistic group; DS: Down syndrome group; GORT: Gray Oral Reading Test; MSAAC: Maryland Society for Autistic Adults and Children; PAS: Personality Assessment Schedule; PDD: pervasive developmental disorder; PRS: Pragmatic Rating Scale; WRAT: Wide Range Achievement Test. Requests for reprints to: Susan Folstein, MD, Department of Psychiatry, New England Medical Center, 750 Washington Street, 1007, Boston, MA 02111, U.S.A. ( susan.folstein es.nemc.org). 1117

2 1118 S. E. FOLSTEIN et al. Introduction The meaning of cognitive abnormalities in the phenomenology and cause of autism has not been resolved. While mental retardation is present in 75% of cases and is associated with a worse prognosis, classical autism is also seen in the presence of normal or superior intelligence, and IQ has never been one of the diagnostic criteria for autism. However, even those individuals with autism who are normally intelligent have great difficulties with certain aspects of language and reading. One way to address the question of the role of genes for cognition and language in the etiology of autism is to study these features in the parents and siblings of autistic individuals, compared with controls. Patterns of Wechsler IQ subtest scores are remarkably uneven in autistic individuals, reflecting the strengths and weaknesses seen clinically. Tasks requiring abstract language or sequencing are poorly performed, whereas those testing either rote memory or visual-spacial capacity are usually performed well (reviewed by Siegel, Minshew, & Goldstein, 1996). Autistic individuals show this same pattern of strengths and weakness regardless of Full Scale IQ, but the relationship between Verbal and Performance scores changes as Full Scale IQ increases. In autistic persons with lower Full Scale IQs, Performance IQ usually exceeds Verbal IQ; in autistic subjects with higher Full Scale IQs, this discrepancy is not marked, and studies of normally intelligent autistic persons (Ruhl, Werner, & Poustka, 1995) or persons with Asperger syndrome (Klin, Volkmar, Sparrow, Cicchetti, & Rourke, 1995) have often found higher Verbal than Performance IQs. As is true for intelligence tests, some aspects of reading can be normal (e.g. decoding), whereas comprehension is profoundly impaired in autistic individuals (Minshew, Goldstein, Taylor, & Siegel, 1994; Whitehouse & Harris, 1984). Do these distinctive patterns of cognitive abnormality reflect the genetic liability to autism? It is now well established that the etiology of autism is largely genetic (reviewed in Santangelo & Folstein, 1999), with heritability estimated at over 90% (Bailey et al., 1995). Based on the very large differences between the concordance rates in monozygotic and dizygotic twins, combined with the rarity of cases in extended families, it seems most likely that autism is caused by the convergence of at least three or four interacting genes (Pickles et al., 1995; Van Eerdewegh, unpublished analysis). Several traits have been detected in the parents of autistic children that are conceptually parallel to the diagnostic criteria for autism. As a group, these traits have been called the broader autism phenotype, and are hypothesized to be manifestations of the genetic liability to autism. The characteristics so far identified include social reticence (Piven et al., 1994; Santangelo & Folstein, 1995), pragmatic language deficits (Landa, Folstein, & Isaacs, 1991; Landa et al., 1992; Wolff, Narayan, & Moyes, 1988), and rigid and obsessive traits (Bolton et al., 1994; Eisenberg, 1957; Piven, Palmer, Jacobi, Childress, & Arndt, 1997). Under the model of several independently inherited, interacting genes causing autism, the expectation would be that one or more genes is contributed by each parent unless one parent was autistic, in which case he or she could contribute all the requisite hypothesized genes. Thus, we have hypothesized (Santangelo & Folstein, 1999) that each component of the broader autism phenotype is inherited separately and manifested independently in nonautistic family members, who may have only impaired language, social reticence, or rigid obsessive features. Do cognitive difficulties exist in family members of autistic individuals, and if so do they constitute a fourth aspect of the broader autism phenotype, or are they another reflection of the language-related component? In three family history studies, parents have reported cognitive deficits in themselves and their nonautistic children, more often than controls (Bartak, Rutter, & Cox, 1975; Bolton et al., 1994; Wzorek et al., 1989). Both parents and siblings of autistic probands were more likely than controls to report delay in onset of speech, articulation defects, trouble learning to read, or spelling problems. Only Szatmari et al. failed to confirm this finding (1993) when comparing their families (ascertained through a proband with either autism or pervasive developmental disorder PDD) to families with a mentally handicapped child. However, they did find significant differences in the frequency of language-based cognitive difficulties as well as social traits when the autism PDD families were compared with the families who had adopted autistic or PDD children (Szatmari et al., 1996). Despite the consistent findings on family history, however, studies that have directly tested reading, spelling, or general intelligence in the parents or siblings of autistic children have not reported abnormalities. Studies of Parents Two research groups (Dor-Shav & Horowitz, 1984; Freeman et al., 1989) reported normal intelligence in the parents of autistic children (AU) but did not test a comparison group. Three controlled studies have been published recently. Szatmari et al. (1993) compared 97 parents of a child with either autism or PDD and 54 parents of children with Down syndrome (DS) and low birthweight on five WAIS subtests (Vocabulary, Digit Span, Comprehension, Block Design, and Digit Symbol). He found no differences between cases and controls on these individual subtests; Full Scale IQ scores were not estimated. In the London family study (Fombonne, Bolton, Prior, Jordan, & Rutter, 1997), parallel in design to the one reported here, WAIS-R IQ and reading tests were carried out on 160 AU parents and 42 DS parents. Performance IQ was significantly lower than Verbal IQ for AU parents, while the opposite was found in the DS control group. Since this is an atypical pattern for a normal control group, the significantly better verbal subtest scores of the AU parents were difficult to interpret. This pattern of slightly higher Verbal than Performance IQ in parents was also found in multiplex autism families (Piven & Palmer, 1997). Those AU parents also scored lower than controls on the Picture Completion and Object Assembly subtests. Results of reading tests in parents were reported by both the Fombonne and Piven studies. In the Fombonne study, there were few differences between cases and

3 PREDICTORS OF COGNITIVE TEST PATTERNS IN AUTISM FAMILIES 1119 controls. In the Piven study of the parents of two or more autistic children, parents scored lower than controls only on the Passage Comprehension test of the Woodcock- Johnson Battery; this study did not test spelling. Studies of Siblings The paucity of cognitive difficulties observed in parents may be partly due to selection bias insofar as individuals with the most severe cognitive difficulties would be unlikely to become parents. Consequently, if genes related to cognition are important to the etiology of autism, cognitive difficulties in nonautistic siblings who are not selected in this way would be expected. Cognitive difficulties, by parental report, have been found in several twin (Bailey et al., 1995; Folstein & Rutter, 1977) and controlled family history studies (Bartak, Rutter, & Cox, 1997; Bolton et al., 1994; Piven et al., 1997). These include both language-related difficulties and mental retardation. However, the early suggestions of more general cognitive deficits in siblings (August, Stewart, & Tsai, 1981; Minton, Campbell, Green, Jennings, & Samit, 1982) have not been confirmed by more recent studies of IQ (Freeman et al., 1989; Szatmari et al., 1993). In the London family study, siblings of autistic children scored slightly but significantly higher on WAIS-R WISC-R Full Scale and Verbal IQ than did siblings of DS controls. The AU siblings had slightly higher Verbal than Performance IQ, as did their parents; control siblings, like their parents, had somewhat higher Performance than Verbal IQ. Thus, like the parents, severe cognitive difficulties in siblings have not been found in any recent study. August et al. (1981) tested reading and spelling in the siblings of probands with autism and Down syndrome. A number of the siblings were poor readers, but their reading scores were consistent with what would be expected from their IQs, suggesting that cognitive difficulties in family members were related to mental retardation rather than to autism. Siblings in the large sample from the Utah epidemiological study of autism (Freeman et al., 1989) scored similarly to published norms on the Wide Range Achievement Test (WRAT), and Szatmari et al. (1993) found no differences on the WRAT between siblings of autism PDD probands and siblings of controls. Reading tests that particularly tax phonetics or comprehension were not administered. The family study of cognitive performance reported here was part of a multifaceted study comparing parents and siblings of community-ascertained probands with autism (AU group) with control parents and siblings of probands with Down syndrome (DS group); cases and controls had comparable educational levels. The study was designed to test the hypothesis that the first-degree relatives of probands with autism would exhibit differences on measures of social, language, and cognitive traits compared to controls with Down syndrome. Differences on these measures, if observed, would support the theory that the genetic liability to autism is manifested in these traits. Our results largely confirm other recent studies: AU parents had slightly but significantly higher Verbal than Performance IQ and also performed worse on nonsense word reading. No case control differences were found in siblings. AU parents (although not siblings) were more likely than controls to have had a history of early language-related cognitive difficulties. Both AU parents and siblings who had such a history had patterns of IQ, reading, and spelling scores that are consistent with deficits in reading and language, relative to their IQ and education. Methods Sample Ninety families with a child with autism (AU) and 40 families with a child with trisomy 21 Down syndrome (DS) were recruited. DS families were selected as the comparison group in order to control for the presence of a severely handicapped child in the family, and at the same time study family members who were not at increased genetic risk for mental retardation or other cognitive or social difficulties. AU and DS families were first recruited in the community through the Maryland Society for Autistic Adults and Children (MSAAC) and the Maryland Down Syndrome Congress, respectively. When those sources were exhausted, additional families were ascertained through all special education classes in the Baltimore County Public Schools, a district that is mostly middle- to upper-class. The autism sample was enriched for female probands who were recruited from several other sources, mainly through the files of an autism clinic. In the letter asking for volunteer families, we stated that we were studying biological causes of autism. We did not disclose the genetic hypothesis, but did say that we would be testing family members. For the MSAAC families, we obtained permission to telephone those families on the mailing list who did not respond to our invitation to participate, in order to check for biases in the sample. We were particularly concerned that the sample not be biased toward families with more than one autistic child or families with additional handicapped children. Those who declined to participate had older children, but there was no suggestion of bias toward genetic cases among the volunteers. Families were excluded who did not speak English as their first language because many of the tests and examinations measured aspects of language. Because this study was so multifaceted and required so many visits to the families homes, we did not attempt to maintain blindness. Autism was diagnosed using the Autism Diagnostic Interview (ADI; Le Couteur et al., 1989), and probands met both ADI criteria and DSM-III-R (American Psychiatric Association, 1987) criteria for autism at age 5. Some of the older probands no longer met the full DSM-III-R criteria at the time of the study, but continued to meet ADI criteria. Autism probands with fragile-x (one undiagnosed case was found), tuberous sclerosis, neurofibromatosis, and severe cerebral palsy were excluded. So that family characteristics could be studied according to the level of the proband s IQ (Table 1), the autism sample was stratified to include approximately equal numbers of probands in four IQ categories, based on nonverbal IQ: under 30 (N 19), (N 22), (N 15), and above 70 (N 34). When possible, proband IQ was based on tests from the medical record. If there were several acceptable measurements of nonverbal IQ, we used the test carried out nearest to age 12. We chose tests according to the following hierarchy: Wechsler Intelligence Scale for Children (Wechsler, 1974) was used in preference to the Merrill-Palmer Scales (Stutsman, 1931), which were preferred over the Leiter International Performance Scales (Arthur, 1952). IQ testing was carried out for those probands for whom an acceptable test was not found in the record.

4 1120 S. E. FOLSTEIN et al. Procedures Family history interview. For this study we developed, in collaboration with our colleagues in London, a structured family history interview for autism families (Bolton et al., 1994; Pickles et al., 1995; Piven et al., 1997) that included items related to cognition, social interactions, personality attributes, and psychiatric disorders. For this report, we analyzed four items: onset of phrase speech, severe abnormalities of articulation, difficulty with learning to read, and difficulty with spelling. Early language-related cognitive difficulties were scored as present when any of these four items was rated as definite (a direct report from the subject or a subject s parent of delays or specific remedial interventions) or probable (highly likely to be present, but no specific remediation was reported). Direct tests and interviews. Parents and siblings were interviewed and tested using measures of social interaction (Piven et al., 1994; Santangelo & Folstein, 1995; Santangelo et al., 1999), social and narrative language (Landa et al., 1991, 1992), and psychiatric disorders (Piven et al., 1991), as well as the cognitive measures reported here. The present report includes those parents and siblings who had IQ testing (Table 2). Parents and siblings aged 16 and over were tested using the WAIS-R; siblings aged 6 15 were given the WISC-R. Throughout the study, the verbal subtests Arithmetic, Vocabulary, Comprehension, and Digit Span were administered and the performance subtests Picture Arrangement, Block Design, and Object Assembly were used. After the first year, the Similarities and Picture Completion subtests were added. Reading and spelling tests necessarily varied somewhat with age. Subjects of all ages were administered the Gray Oral Reading Test (Wiederholt & Bryant, 1986). In children, reading comprehension was tested using the comprehension test from the Kaufman Tests of Educational Achievement (Kaufman & Kaufman, 1985). In a preliminary analysis made after a year of data collection, there were no group differences on any of the reading tests in either adults or children. Therefore, at that time, we added the word attack test from the Woodcock-Johnson battery, which requires the subject to read nonsense words aloud (Woodcock, Johnson, & Bonner, 1991). This test is also standardized for a wide age range and was suitable for administration to all subjects. To test spelling, adults were administered the Schonell spelling test (Schonell & Schonell, 1960), while subjects under 18 years of age were given the spelling test from the Kaufman battery. We also report here the relationship between another language-related test, the Pragmatic Rating Scale (PRS) and the cognitive findings. The PRS is an interview-administered rating to document errors in pragmatic language and speech that are observed during an interview. Instructions for rating can be found in Landa et al. (1992). The total score significantly distinguished AU from DS parents. (A subset of these interviews was scored blind to family membership from audio tapes; the relative scores of AU and DS family members was not different from the unblind ratings.) In addition, we report the relationship between the cognitive findings and two measures of the social aspects of the broader autism phenotype: the score on the Friendship Interview (Santangelo & Folstein, 1995; Santangelo et al., 1999) and the Aloof item from the Personality Assessment Schedule (PAS) (Piven et al., 1994). The Friendship Interview was developed for this study. The global friendship rating reflects both the number of friends and the nature of the friendships including intimacy and reciprocity. The PAS is based on an interview developed by Tyrer to assess personality traits and disorders (Tyrer, 1988). Traits are rated as normal (0, 1) or causing difficulty (2) in one or more situations (work, family, peers). Aloofness is defined by a lack of interest in or enjoyment from being with people, and is often accompanied by a preference to be alone. Significant case control differences were found on both the Friendship and Aloof measures. Analysis All analyses were run in SAS (1990). Odds ratios and chisquare tests of significance were used for univariate analyses of discrete variables, and t-tests were used to compare case and control differences on continuous variables. Multivariate analyses were carried out using logistic regression. Results Sample Although the AU probands were somewhat older than the DS probands (Table 1), the parents and siblings were very similar in age and education (Table 2). Thus, we compared levels and patterns of test performance without adjusting for age or education in either parents or siblings. The parents who had IQ testing included 94% of the eligible AU parents ( ) and 93% of the eligible DS parents (75 80). Of the eligible siblings (aged 6 and over), 82% of the AU siblings (87 106) and 88% of the DS siblings (64 73) were tested. Only one sibling of an autistic individual probably had severe enough traits to be called PDD or Asperger syndrome. He was included in the analyses because he generally functioned in the community, holding a job and living independently. He had never been taken for an evaluation. Wechsler Intelligence Scales Parents. Whereas both groups of parents had above average Full Scale IQs (Table 3), as expected from our ascertainment sources and the parents educational status, the AU parents scored slightly but significantly lower than the DS controls on Performance IQ, and thus on Full Scale IQ as well. AU parents scored significantly lower than DS parents on the Picture Completion and Picture Arrangement subtests. Verbal IQs were not different. Although the magnitude of these IQ differences was small, it was highly significant, and consistent with Table 1 Demographic Characteristics of Probands Autism (90 families) Down syndrome (40 families) Probands (N) Agea:Mean (SD) (6 9) 9 56 (5 9) Age rangea Sex ratio (M:F) 69:21 23:17 IQ groups (N) a In years.

5 PREDICTORS OF COGNITIVE TEST PATTERNS IN AUTISM FAMILIES 1121 Table 2 Demographic Characteristics of Parents and Siblings with IQ Tests Autism (90 families) Down syndrome (40 families) Parents (N) Agea: Mean (SD) 45 6 (9 8) 44 8 (9 3) Age rangea Educationa: Mean (SD) 15 7 (2 7) 15 4 (2 7) Education rangea Occupation Professional 54% 56% Skilled, nonmanual 15% 15% Skilled, manual 8% 4% Homemaker 17% 23% Other 6% 2% Siblings (N) Male: Female 42: 45 28: 36 Agea: Mean (SD) 18 9 (10 3) 18 5 (11 1) Age rangea Race (% of families) White Black, Hispanic, Asian 6 0 a In years. the two other controlled studies that used the complete Wechsler scales (Fombonne et al., 1997; Piven & Palmer, 1997). No mental retardation was seen; no AU or DS parent had either a Verbal or Performance IQ below 75. However, the variation in scores on all scales was significantly greater for AU than for DS parents. There was a significant Verbal-Performance difference for AU but not DS parents, with the AU parents having higher average Verbal than Performance IQ. When we stratified the analysis by sex, a significant Verbal-Performance difference was only observed among mothers, although just as many fathers as mothers had Verbal IQ scores which exceeded their Performance IQs by at least 15 points or 1 SD. To test whether these findings were in some way related to our inclusion of 19 families in which the autistic probands had IQs under 30, a population in which autism can be difficult to diagnose, data were analyzed excluding the 28 parents and 18 siblings of these families. The findings were essentially the same (data not shown). Analyses omitting the parents whose IQ had been estimated from fewer WAIS-R subtests also gave very similar results. Siblings. There were no significant case-control differences in any of the IQ variables for siblings except that, as in the parents, there was more variation in the scores of the AU siblings (Table 3). The findings were the same when older siblings tested on the WAIS-R and younger siblings tested on the WISC-R were considered separately and when siblings of autistic probands with IQs less than 30 were omitted (data not shown). No AU siblings had a Full Scale IQ less than 74; the lowest DS sibling score was 84. Tests of Reading and Spelling Parents. Despite giving histories of having had trouble with reading and spelling as children, the AU parents as a group scored like the DS parents on the WRAT wordreading test and on the Schonell Spelling test. However, they scored lower on the Woodcock-Johnson word attack Table 3 Wechsler Intelligence Scale Scores in AU and DS Parents and Siblings AU DS WAIS-R WISC-R scorea Mean SD Mean SD t df p Parentsb Full Scale IQ Verbal IQ n.s. Performance IQ Picture Arrangement Picture Completion Siblingsc Full Scale IQ n.s. Verbal IQ n.s. Performance IQ n.s. Verbal-Performance differenced Mean SD t p Parents AU DS n.s. Siblings AU n.s. DS n.s. a All parents and siblings aged 16 and over were tested using the WAIS-R. b Fifty-four of the 166 AU parents and 7 of the 75 DS parents had IQs estimated from 4 verbal and 3 nonverbal subtests. c Seventeen of the 87 AU siblings and 8 of the 64 DS siblings had IQs estimated from 4 verbal and 3 nonverbal subtests. d Paired t-test.

6 1122 S. E. FOLSTEIN et al. Table 4 Direct Tests of Reading and Spelling Autism DS Mean SD Mean SD t df p Parents Spelling Schonell Test Total words correct n.s. Spelling age n.s. Reading Gray Oral Test Comprehension (standard score) n.s. Passage (standard score) n.s. Woodcock-Johnson Nonsense Word Readinga Reading age Reading grade Siblings Spelling Schonell Test, 18 years Mean age Total words correct n.s. Spelling age n.s. Kaufman, 18 years Mean age Spelling (standard score) n.s. Reading Kaufman, 18 years Mean age Comprehension (standard score) n.s. Gray Oral Test, all ages Mean age Comprehension (standard score) n.s. Passage (standard score) n.s. Woodcock-Johnson Nonsense Word Reading (all ages)b Mean age of subjects Reading age n.s. Reading grade n.s. a Woodcock-Johnson scores are calculated using the 54 AU parents and 36 DS parents tested. b Woodcock-Johnson scores are calculated using the 33 Autism siblings and 29 Down Syndrome siblings tested. test (Table 4), which requires the subjects to read aloud nonsense words that are phonetically plausible. Siblings. As was true for IQ and reading and language history, AU and DS siblings scored almost identically on all the tests of reading and spelling (Table 4). History of Early Difficulties with Language and Reading AU parents, significantly more often than DS parents, were given a rating of either definite or probable for one of several language-related problems they reported they had during childhood: late onset of phrase speech, significant articulation defects, trouble learning to read, or trouble with spelling (Table 5). AU parents were nearly three times as likely as DS parents to have reported such a history (p 003; OR 2 84, CI: ). In both the cases and controls, fathers were more likely to report early language-related cognitive difficulties than mothers. When only those with a definite rating on one of these questions were included, the numbers were smaller, but the differences remained significant using a one-tailed Fisher s exact test. Justification for use of a one-tailed test was based on the findings of the prior studies. For siblings, using both definite and probable ratings of early language-related difficulties, cases and controls were not significantly different. When only definite ratings were used, 11% of AU and 7% of DS siblings were affected, but the absolute number affected was low, and the difference was not statistically significant. Again, the findings were almost identical when the parents and siblings of the low IQ cases were omitted from the analyses. A History of Early Language-related Cognitive Difficulties and Direct Test Results We reasoned that, like adults with a history of language and reading difficulties, those family members with a positive history of early reading and language difficulties would be likely to demonstrate persisting difficulties on language-related intelligence and achievement tests. We

7 PREDICTORS OF COGNITIVE TEST PATTERNS IN AUTISM FAMILIES 1123 Table 5 Early Language-related Cognitive Difficulties in AU and DS Parents and Siblings AU DS χ p Odds ratio (CI) pa Parents (N) Definite & probable 26% 11% ( ) 006 Definite only 14% 6% ( ) 039 Siblings (N) Definite & probable 22% 21% 0 06 n.s ( ) n.s. Definite only 11% 7% 0 71 n.s ( ) n.s. a Fisher s exact (one-sided). Subjects with any one of the following were coded as having early language-related cognitive difficulties: late onset of phrase speech, articulation defect severe enough to cause difficulty in understanding the child at age 5, trouble learning to read, or spelling problems. Use of one-sided test was based on the consistent finding in three other studies that autism family members were more likely than controls to have early language-related cognitive difficulties. Table 6 Early Language-related Cognitive Difficulties (ELRCD) by Direct Tests in AU Parents ELRCD absent ELRCD present Mean SD Mean SD t df p Education (years) n.s. WAIS-R Full Scale IQ n.s. Verbal IQ Performance IQ n.s. Spelling (Schonell) Total Words Correct Reading Gray Oral Reading Test Comprehension n.s. Passage n.s. Woodcock-Johnson Nonsense Word Reading Reading age Reading grade PRSa WAIS-R Verbal-Performance Differenceb Mean SD t p ELRCD present ELRCD absent n.s. PAS ELRCD absent ELRCD present χ p Aloof (% positive) n.s. Friendship (% above cutoff) n.s. a On the PRS, higher scores indicate a worse performance. The mean (SD) score for the DS control parents (N 76) was 1 63 (2 8). The score was not significantly different from the score of the AU parents who did not have a history of early cognitive difficulties. b Paired t-test. therefore carried out a within-group comparison of the autism family members who were history positive and history negative. Autism parents. For the AU parents, education and Full Scale IQ were unrelated to their having reported early difficulties (Table 6), but language-related aspects of IQ were related: Verbal IQ was lower for these parents, as were the Vocabulary and Comprehension subtests. In contrast to the group of AU parents as a whole, those with early difficulties had no overall verbal performance difference. On tests of reading and spelling, parents who reported early language-related difficulties scored significantly lower on the Woodcock-Johnson nonsense reading test and on the Schonell Spelling test. Parents with early language difficulties had an average Woodcock-Johnson grade (years education) score of 9 4 years; their actual average years of education was 15 2 years. The AU parents without any complaint of early difficulties had a more striking advantage on Verbal IQ than did the AU parents as a whole. Of the 27 parents whose Verbal IQ exceeded their Performance IQ by more than 15 points, 26 reported no early cognitive difficulties. Their scores on reading and spelling tests were like those of the DS control parents. We also compared history positive and history negative AU parents on three other measures that had showed case-control differences in our earlier analyses of these same families. Many items of the PRS (Landa et al., 1992) are closely related to language competency. The

8 1124 S. E. FOLSTEIN et al. Table 7 Early Language-related Cognitive Difficulties (ELRCD) and Direct Tests in AU Siblings ELRCD absent ELRCD present Mean SD Mean SD t df p Age n.s. WAIS-R WISC-Ra Full Scale IQ Verbal IQ Performance IQ n.s. Spelling Schonell ( 18 yrs) Total words correct Kaufman ( 18 yrs) Standard score Reading Kaufman ( 18 yrs) Standard score Gray Oral Reading Test Comprehension n.s. Passage n.s. WAIS-R Verbal-Performance differenceb Mean SD t p ELRCD present n.s. ELRCD absent n.s. a Wechsler Subtest scores were not analyzed because of difficulties in combining those from the WAIS and WISC. b Paired t-test. other two measures, the overall score on the Friendship Interview (Santangelo et al., 1999) and the Aloof item from the PAS (Piven et al., 1994; Tyrer, 1988), were hypothesized to be more related to the social than the language component of the broader autism phenotype. Those parents who reported early cognitive difficulties scored significantly worse on the PRS while the scores for parents without early language difficulties were similar to the DS control parents. For the friendship score and the Aloof item from the PAS, there were no differences between the parents with and without a history of language-related cognitive difficulties; both groups scored significantly higher (worse) than controls. Multivariate logistic regression was employed to assess the marginal effects of the following measures on reporting early language-related cognitive difficulties in AU parents: Verbal IQ, Performance IQ, Schonell Spelling Test (total words correct), and two Gray Oral Reading Test (GORT; Wiederholt & Bryant, 1986) items. The nonsense reading test was not included because of the consequent reduction in sample size due to the fact that the test was administered to less than half of the sample. The only significant predictor from this model was the Schonell spelling score. Autism siblings. There were no significant differences between cases and controls on the variable early language-related cognitive difficulties. However, when AU siblings who did have such a history were compared with those who did not, the differences were similar to those found in parents (Table 7). Those AU siblings with positive histories (N 19) had significantly lower Full Scale and Verbal IQ and lower spelling scores than those with negative histories (N 66). The school-aged siblings also scored lower on the Kaufman Reading Comprehension test; adult siblings were not given this test. We were unable to translate the Kaufman scores into a grade estimate (years of education). Too few siblings were given the nonsense reading test for meaningful comparisons to be made. DS parents and siblings. DS parents with and without early language-related difficulties scored similarly on all the tests. However, these estimates are likely to be imprecise since only 9 of the 75 DS parents who completed IQ tests reported early language-related difficulties. (Data not shown.) In contrast to the AU siblings, the DS siblings with positive (N 12) and negative (N 46) histories of language-related cognitive difficulties performed almost exactly the same on all IQ reading and spelling tests. One possible explanation for this failure to find a difference is that the parents of the DS siblings appeared to have a low threshold for reporting articulation difficulties; 9% (N 8) had probable ratings. When the analysis was restricted to DS siblings who reported a definite history of language-related cognitive difficulties, the DS siblings with a positive history (N 4) had a significantly lower Full Scale IQ, but had reading and spelling scores that were similar to those with a negative history (data not shown). Results in Relation to Proband Characteristics There was no relationship between the autistic probands nonverbal IQ category and any of the test results of either their parents or their siblings. Nor did parents

9 PREDICTORS OF COGNITIVE TEST PATTERNS IN AUTISM FAMILIES 1125 history of early language-related cognitive difficulties relate to the IQ of their autistic children. (Data not shown.) Results in Relation to Sex of Parents Compared with AU fathers, DS fathers had a higher mean Full Scale IQ [ vs , t(87) 2 18, p 032] and Performance IQ [ vs , t(111) 2 10, p 038]; there were no differences on any IQ tests between AU and DS mothers. When AU fathers and mothers were compared, fathers scored better on Performance IQ [ vs ; t(164) 2 04, p 043] and nonsignificantly better on Full Scale IQ [ vs ; t(164) 1 93, p 055], while mothers scored slightly higher on the Schonell Spelling Test [92 48 vs ; t(162) 2 01, p 053]. In the DS families, fathers scored significantly better than mothers on all three IQ scales [Full Scale: vs , t(73) 3 04, p 003; Performance: vs , t(56) 2 62, p 011; Verbal: vs , t(73) 2 28, p 026]. Discussion Summary of Findings The parents and siblings of community-ascertained AU and DS families, comparable for parents education and social class, scored similarly on tests of IQ, reading, and spelling. As expected for their education level (over 15 years on average), both case and control family members had above-average intelligence. There were some differences between parents of AU and DS probands: AU parents had slightly but significantly lower Full Scale and Performance IQ, higher Verbal than Performance IQs on average, and scored lower on a nonsense word reading test. Further, significantly more AU parents than DS parents (24% vs. 12%) reported having had language-related difficulties during childhood: either late onset of speech, articulation difficulties, or (most commonly) difficulty with reading and spelling. There were no significant case-control differences among siblings in IQ, reading, spelling, or a history of early language problems. The autism family members (both parents and siblings) who reported such a history had different patterns of cognitive test scores than those without early languagerelated problems. Parents and siblings with a history of complaints had lower Verbal IQs, no Verbal-Performance difference, and were worse spellers. Parents had grade level scores on the nonsense reading task more than 5 years lower than their actual years of education; siblings scored lower on reading comprehension. The cognitive difficulties in parents were related to their scores on the PRS but not to two measures of sociability. Strengths and Limitations of the Design The strengths of this study include the community ascertainment of the families, which makes the participants largely unselected for registration with a university clinic or prior participation in research projects, and the almost identical mean and range of education of both parent groups, which allowed us to compare patterns of cognitive function in subjects who had comparable overall attainments. Every effort was made to recruit all families who made inquiries, not only those families who found it easy to cooperate with our multifaceted and time-consuming research protocol. The research staff, all professional clinicians, exerted considerable effort to retain all families who entered the study. The sample was not, however, drawn randomly from a geographically defined population and probably includes more well-educated families than the general population of families with an autistic child. Although this is a limitation, it does mean that we cannot attribute any reading and spelling difficulties to limited intelligence or education. Other limitations include the small size of our control group and the fact that some tests were added after the project started, both of which limited the value of some within-group analyses due to small sample sizes. An additional disadvantage is the wide age range of the siblings, making it necessary to use more than one test of spelling and reading in this group, which limited our power since the sample sizes on a given test were not always large. We decided not to combine adult and school-aged siblings in the analyses because the reading tests used with adults and children were not comparable. Comparison with Other Studies Our findings in the sample as a whole agree with the majority of studies, which have found that although parents and siblings of autistic children are more likely than controls to report a history of early language-related cognitive difficulties, they score similarly to controls with similar education (or population-based norms) on most tests of intelligence, reading, and spelling. All studies agree that the family members of autistic individuals do not, as a group, have deficits on standard tests of intelligence, reading, or spelling. Ours is the third recent study to find slightly but significantly higher Wechsler Verbal than Performance IQ scores in parents of AU individuals (Fombonne et al., 1997; Piven & Palmer, 1997). Siblings in the London study had a similar Verbal-Performance difference, but we were not able to confirm this in our sibling sample. This pattern is similar to that found in studies of autistic individuals with normal IQs and in individuals diagnosed with Asperger syndrome. However, the individual subtest scores found in these three recent studies do not conform to the pattern seen in AU individuals. In both this and the Piven study, AU parents scored more poorly than controls on Picture Completion. The subtest differences in the London study are difficult to interpret because their control group had an unexpected IQ pattern, with the Performance greater than Verbal IQ. The Verbal-Performance differences, while consistent across these three studies, are quite small. Rather different patterns emerged when we divided the family members into those with and without early language delays. It may be, therefore, that the small Verbal-Performance differences seen in the sample as a whole may reflect a composite of two (or more) different patterns on intelligence testing.

10 1126 S. E. FOLSTEIN et al. The biggest difference between our results and those of Fombonne et al. (1997) is that we were unable to confirm the higher rate of early language-related cognitive difficulties in siblings. There is no clear explanation for this, except to suggest random sampling differences. Also, when we related test performance to aspects of the broader phenotype (albeit differently formulated, as described below), the London group found stronger differences in siblings than in parents; our differences were about the same in both siblings and parents. They interpret their data as consistent with a genetic model where family members who had severe manifestations of the broad phenotype would be unlikely to become parents, a bias that is not applicable to siblings. It is not at all clear, however, that there would be any bias against mating due to the small magnitude of cognitive differences observed in siblings in the Fombonne study. Patterns of Test Performance According to History of Early Language-related Cognitive Difficulties We found robust and consistent differences on many of the tests between AU family members who did and did not report early language-related difficulties. The finding that both subgroups of AU parents had IQ patterns that were different from controls suggests the possibility that most AU parents may be atypical in their cognitive patterns, and in more than one way. The AU parents without early language-related cognitive difficulties had a larger Verbal IQ-Performance IQ difference than the group as a whole, but scored similarly to controls on the tests of spelling and nonsense word reading and on a measure of pragmatic language. The pattern of reading and spelling scores found in the parents with early language-related difficulties is reminiscent of patterns seen in adults with a history of reading difficulty. Their spelling was poor, and although they performed normally on the GORT, their scores were well below those expected for their education on the more challenging task of reading nonsense words. This test often remains deficient in compensated adults with a history of dyslexia (Felton, Naylor, & Wood, 1990). Piven et al. (1997) reported similar findings: poor scores on reading comprehension in parents ascertained through two autistic children. Fombonne et al. (1997) compared family members with and without the broad autism phenotype, as elicited by the family history interview used in both studies. Using questions from this interview, they produced a composite score for social interaction, communication (comparable to our early language-related cognitive history ), and rigid or stereotyped behavior. They then analyzed the relationship between this composite broad phenotype score and results of cognitive tests. Siblings with the broad phenotype had similar findings to our family members with a positive history of language difficulties: lower Verbal IQs, equal Verbal and Performance IQs, and a higher frequency of poorer readers and spellers. However, the parents did not have such a clear pattern of scores. In this study, we analyzed two components of the broader phenotype separately for their relationship to a history of reading and language difficulties: an additional measure of pragmatic language, the PRS, and two measures unrelated to language, the Aloof measure from the PAS and the Friendship score. This was based on the following logic. First, the most plausible genetic model for autism suggests that several individual genes, some contributed by each parent, interact to cause autism. Second, we reasoned that each individual component of the broader phenotype, conceptually parallel to each of the three criteria for autism (Bolton et al., 1994; Piven et al., 1997) social reticence, pragmatic language difficulties, and difficulty with change may be the phenotypic manifestation of one or more of these individual genes. Third, most of the twin and family history evidence suggests that the cognitive difficulties seen in family members are mainly language related, and this was supported by our finding a relationship between Verbal IQ, reading, and spelling and an early history of language-related difficulties. This hypothesis would be further supported if a different language-related measure, the PRS, showed a similar relationship to cognitive test scores as did the history of language-related cognitive difficulties, and if measures unrelated to language, but to a separate component of the broader phenotype social reticence were unrelated to cognitive test scores. Our findings are consistent with that logic, and we therefore propose that some aspects of reading and intelligence tests can detect the language component of the broader phenotype, and that only some family members (perhaps a quarter of our sample) of autistic children have this type of language-related aspect of the broader phenotype. This conclusion can be most directly refuted by finding that the theorized component genes for autism do not segregate with individual traits in family members. Further indirect support for the hypothesis that separate genes contribute to the autism phenotype code for particular aspects of the disorder comes from recently published genetic linkage studies. The genome screen published by the International Molecular Genetic Study of Autism Consortium (IMGSAC; 1998) gave strong support of linkage to autism for a region on chromosome 7q, the same chromosomal region implicated in a linkage study of a family with developmental language disorder (Fisher, Vargha-Khadem, Watkins, Monaco, & Pembrey, 1998). The International Autism Consortium data were further supported by their more recent abstract (Maestrini & IMGSAC, 1998). Tomblin, Nishimura, Zhang, and Murray (1998) recently reported an association study in an epidemiologically ascertained sample of children with developmental language disorder, which also suggests a locus for this same area of chromosome 7. Acknowledgements The authors would like to thank Erica Bisson for her help in the statistical review and editing of this manuscript. This research was supported in part by National Institutes of Health Grants MH01338 (SLS), MH55135 (SEF), and MH39936 (SEF). References American Psychiatric Association. (1987). Diagnostic and statistical manual of mental disorders (3rd ed., revised; DSM- III-R). Washington, DC: Author.