Although it is generally accepted that genetic factors play a major role in the etiology of autism spectrum disorders (ASDs), identification of specific genetic risk markers is complicated by the phenotypic complexity of clinical diagnoses. For example, the Diagnostic and Statistical Manual of Mental Disorders 4
ed. (DSM-IV) diagnostic criteria for autistic disorder (AD) require impairments in three domains: social interaction, communication and repetitive and stereotyped behavior. Each of these three domains has been shown to be heritable, but their covariation in the general population is modest, and genetic modeling suggests distinct genetic influences for each [3–5]. Thus, it has been argued that the ability to identify susceptibility loci for ASD would be increased if specific ASD/AD traits were used as phenotypes [3, 6].
Specific ASD/AD traits have been employed in genetic studies most often either to stratify pedigrees for linkage analysis or as the dependent variable in association tests for specific alleles. For example, the first approach has found stronger ASD linkage signals in pedigrees with more abnormal levels of phrased speech delay [7, 8], repetitive behavior [9–11] and savant skills , but there have been failures in replication . The second approach has resulted in significant genotype associations with repetitive behavior [14–16]. A third, less common approach has been to use the specific trait as a quantitative or qualitative phenotype in linkage analyses. For example, we used the Social Reciprocity Responsiveness Scale (SRS)  score as the phenotype in linkage analyses of multiplex ASD pedigrees (Coon et al., Genome-wide linkage using the Social Responsiveness Scale (SRS) in Utah autism pedigrees, submitted). Although each of these methods has merit, it should be noted that the first method attempts to reduce heterogeneity of the diagnostic phenotype by stratification on a specific trait, whereas the second and third approaches seek to identify risk markers for the trait itself.
Repetitive and stereotyped behavior is a promising candidate for further genetic study because it probably comprises at least two even more specific phenotypes that differ in their behavioral correlates, familiality, and relation to genetic linkage with ASD. The 'restricted and repetitive stereotyped behavior' (RRSB) domain of the Autism Diagnostic Interview--Revised (ADI-R) [18, 19] is a well-accepted measure of the repetitive behavior phenotype. To uncover the factor structure of RRSB, a variety of factor analytic techniques have been used with different subsets of RRSB items and with study populations that differ in ASD severity and ethnicity [11, 20–25]. Remarkably, in spite of their methodological differences, these analyses converge on a two-factor solution comprising 'repetitive sensory-motor actions' (RSMA) and 'insistence on sameness' (IS). RSMA items investigate repetitive physical mannerisms and unusual sensory interests, whereas IS items investigate compulsive behaviors. There are two exceptions to the common two-factor solution. First, an exploratory factor analysis of RRSB items  recovered essentially the same RSMA and IS factors but also found a third factor ('circumscribed interests'). This finding does not detract from the conclusion that RRSB comprises RSMA and IS, but rather suggests that RRSB may measure additional factors as well. Second, a principal components analysis of all ADI-R items identified six factors, including a 'compulsions' factor that contained some items from both the IS and RSMA factors, and a 'social intent' factor that combined social interaction items with the RSMA item of 'hand and finger mannerisms' . Despite this, the preponderance of statistical evidence indicates that RSMA and IS are distinct factors within the RRSB domain.
It is well established that IS and RSMA have different patterns of relationship with other ASD traits. Specifically, RSMA, but not IS, has been reported to be associated with lower IQ, less adaptive behavior, and later age of appearance of first words and phrases [6, 20, 21], which suggests that RSMA may be more correlated with ASD severity . These findings support the validity of treating IS and RSMA as different phenotypes.
There is more empirical support for a genetic effect on IS than on RSMA. Whereas modest evidence of familial concordance occurs for IS, no reported concordance occurs for RSMA [21, 25]. Thus, the IS factor may account for earlier findings that RRSB is familial [28, 29]. Indeed, Silverman et al.  reported that RRSB categories that include IS items were familial, whereas those that include RSMA items were not. Further, a linkage analysis across the 15q11-q13 region in a subset of families with the highest IS scores resulted in increased LOD scores for AD  over scores obtained without stratification. By contrast, stratification on RRSB or RSMA did not increase lod scores. Finally, obsessive compulsive disorder (OCD) features in parents were associated with IS, but not RSMA, in children with AD , which suggests that IS may be part of a broader autism phenotype of obsessive behavior.
We are not aware of previous genetic linkage studies with either IS or RSMA as the phenotype. The primary aim of the present study was to perform a genome-wide linkage analysis with both IS and RSMA as phenotypes using large extended ASD pedigrees. Thus, our goal was to identify genetic risk regions for IS and RSMA in ASD cases rather than to stratify on IS and RSMA to reduce ASD heterogeneity. Because IS and RSMA data were available only for ASD cases rather than for all pedigree members, we focused our analyses on these specific phenotypes in ASD cases and did not include clinically unaffected family members in this study. Signals obtained with these phenotypes were compared with those found in the same set of pedigrees using ASD diagnosis . Contrasting results obtained with IS and RSMA with those obtained by ASD categorical diagnosis addresses empirically the possibilities that more narrowly defined phenotypes improve linkage analysis signals, and that different narrowly defined phenotypes are associated with different loci. Secondary aims were to examine the correlates of IS and RSMA and to assess the heritability of both scales.