Show Summary Details
Page of

Autism Endophenotypes and Quantitative Trait Loci 

Autism Endophenotypes and Quantitative Trait Loci
Autism Endophenotypes and Quantitative Trait Loci

Rita M . Cantor

Page of

PRINTED FROM OXFORD MEDICINE ONLINE ( © Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Medicine Online for personal use (for details see Privacy Policy).

Subscriber: null; date: 10 December 2017

Points of Interest

  • Twin, linkage, and association studies indicate that ASD is genetically complex, and may be the result of different subsets of multiple risk genes interacting to produce this phenotype.

  • Analytic strategies to reduce the complexity of ASD are needed to reveal its genetic etiology.

  • Identification of ASD-related quantitative endophenotypes that are heritable and have sufficient variation is an important first step.

  • Ascertaining study samples with ASD, measuring related endophenotypes in family members, genotyping, and conducting linkage analyses to detect quantitative trait loci (QTL) is an effective approach to reduce complexity.

  • For ASD, QTL studies have been conducted and loci have been replicated. In one case quantitative association studies reveal the predisposing gene.

1.0 ASD: Genetic But Not Mendelian

During the 6 decades since ASD was defined there has been a growing recognition of the importance of genes in its etiology. This was first revealed by twin studies; the ASD concordance rate in monozygotic (MZ) twins who share all of their genes has been estimated at 90%, while that rate in dizygotic (DZ) twins, who share on average 50 percent of their genes, is much lower, at 10% (Bailey et al., 1995). Despite this substantive support for genetics, efforts to identify ASD genes exhibit inconsistent linkage results and weak association effects (Lamb, Parr, Bailey, & Monaco, 2002). These are likely to reflect its genetic complexity (Veenstra-Vanderweele, Christian, & Cook, 2004). In contrast, single-gene Mendelian disorders that exhibit clear family segregation ratios of 50 percent for dominant traits and 25 percent for recessive traits, yield linkage signals that are usually unambiguously significant at one locus, allowing the responsible genes and their predisposing alleles to be localized and then identified by positional cloning techniques. As an example, the single gene for cystic fibrosis, an autosomal recessive Mendelian disorder, was one of the first to be discovered using linkage studies followed by positional cloning. Even though follow-up studies in patients from around the world indicate that there are over 1000 causal mutations in this gene (Farrell et al., 2008), linkage analysis was successful in localizing it as well as those for other Mendelian disorders.

1.1 Linkage Analysis for Mendelian Disorders

Linkage studies are designed to find the chromosome regions harboring the genes that contribute to phenotypic differences among individuals. The statistical algorithms to test for linkage have evolved over time to accommodate changes in the nature and complexity of the phenotypes and the numbers of markers available for mapping. However, all linkage studies require the ascertainment of pedigrees with members that show variation in the phenotype and have genotypes at informative markers. The statistical algorithms and computer programs for linkage analysis identify where the cosegregation of trait values and marker alleles within pedigrees is greater than what one would expect by chance alone. Linkage results at each locus reflect the statistical information that has been combined across all pedigrees in the analysis (Cantor, 2006; Ott, 1999).

Initially the traits in linkage studies were binary, such as the presence or absence of disease, and the models of inheritance used in the analyses were Mendelian dominant or recessive. Using model based or classical linkage analysis, the statistical evidence for linkage is evaluated by LOD scores. The LOD score is the log of the odds of linkage compared to the null hypothesis of no linkage at a marker. The LOD is calculated by comparing the statistical likelihood of the observed cosegregation of trait values and marker alleles in the study pedigrees to the statistical likelihood of the cosegregation in the same pedigrees at the same marker that is expected by chance alone when there is no linkage. Traditionally, a LOD greater than 3 (odds of 1000 to 1) has been taken as evidence supporting linkage to the region containing the genetic marker. Significant linkage implies that the disease gene is in the chromosome region containing the marker. Following a successful linkage analysis, positional cloning, association analyses, or sequencing studies designed to detect predisposing genes and causal alleles are conducted in the linked regions.

1.2 ASD is Genetically Complex

In addition to inconsistent linkage and association results, the genetic complexity of ASD is revealed by the large discrepancy between the MZ and DZ twin concordance rates. That is, when there is a multigene model with no allelic or gene interactions, the DZ twin concordance rate is expected to be about one half that seen in MZ twins. The large difference estimated for ASD (90% vs. 10%) is explained by a model of inheritance that includes numerous interacting genes and alleles (Ijichi et al., 2008; Risch et al., 1999), with each making a small contribution to the ASD phenotype. The complex genetics of ASD is also likely to include significant genetic heterogeneity (Happe, Ronald, & Plomin, 2006; Ronald et al., 2006). Thus, a realistic model of inheritance might include 20 or more predisposing ASD genes, with each having numerous risk alleles. One individual with ASD might have a particular subset of those genes and alleles, while another with ASD could have a different but overlapping or completely different subset. In addition, the environment is likely to play an important role, and its effects may be unequal among the affected individuals.

In summary, there are a very large number of genetic models that could each be invoked to explain the genetic risk for ASD, with none being correct. Linkage analyses of ASD cannot employ simple models of inheritance such as the dominant and recessive ones used for Mendelian disorders, as the true model is very complex and remains unknown. Given this genetic complexity, designing an appropriate gene identification study remains a challenge. However investigators continue to feel a pressing need to use the current molecular and analytic tools to tackle gene identification for ASD and other common complex disorders that are clearly genetic, but not Mendelian. The trajectory of this endeavor has been accelerated by the molecular and bioinformatics tools developed and refined in conjunction with the Human Genome (Roberts, Davenport, Pennisi, & Marshall, 2001) and HapMap (Manolio, Brooks, & Collins, 2008) projects.

1.2.1 Model Free Linkage for Complex Traits

Linkage methods that do not require a genetic model have been developed to address the genetic complexity. They are referred to as model free methods, and some approaches used in QTL mapping are based on them. While no model of inheritance is put forth, the principle on which the methods are based is that if a chromosome region contains a disease gene, siblings or other relative pairs who both have the disease will share marker genotypes in the region containing a disease gene more often than one would expect by chance alone. Thus, chromosomal loci exhibiting excess marker allele sharing identical by descent (IBD) (or the same allele inherited from common ancestors) by affected sibling pairs are linked to the disorder. Marker allele sharing IBD is illustrated in Figure 39-1.

Figure 39–1. Illustration of allele-sharing identical by descent.

Figure 39–1.
Illustration of allele-sharing identical by descent.

In this pedigree with a mother, a father, and their two children, the parents have genotypes A1A2 and A3A4 respectively. The mother passes A1 and the father passes A3 to their daughter. If the son gets A2 from the mother and A4 from the father, these two siblings share 0 alleles IBD, which is expected to occur 25% of the time. If the son gets A1 and A4, the siblings share one allele IBD. Sharing one allele could also occur if the son gets A2 and A3, so sharing one allele IBD is expected to occur 50% of the time. If, like the daughter, the son also gets A1 and A3 from their parents, the siblings share 2 alleles IBD, which is expected to happen 25% of the time. The null hypothesis of no linkage or no excess allele sharing has a probability distribution of 25%, 50%, and 25% for sharing 0, 1, or 2 alleles IBD. If, for example, in a panel of sibling pairs ascertained for ASD, an observed distribution of 15%, 40%, and 45% for a marker provides evidence of ASD linkage to the chromosome region containing the marker indicating the presence of an ASD gene at that locus.

Model free linkage analysis tests for deviations of the observed IBD distributions from the expected IBD sharing distribution under the null hypothesis of 25%, 50%, and 25%. Various statistics have been used to assess the significance of the deviations, and some methods provide a LOD score (Cantor, 2006; Ott, 1999). A threshold for significance should be set to declare linkage. Setting this threshold is not straightforward, but researchers often use a LOD score of 3 (or a p-value equivalent to it), which is the same as the one used for parametric linkage analysis. Replication in an independent sample is a hallmark of success in mapping complex traits. ASD, like many other genetically complex disorders, shows evidence of linkage to numerous chromosome loci, but only a few replications have been observed (Cantor et al., 2005). Thus, even model free linkage methods may not be adequate for a phenotype with this degree of genetic complexity. This suggests that ASD is appropriate for well-designed QTL studies.

1.3 Genomewide Association Studies

Currently, studies are employing an alternative to linkage analysis as the first step in gene detection. That is, preliminary linkage analyses are not conducted to localize genes, but the whole genome is tested for association with ASD or other complex disorders using 500,000 to one million single nucleotide polymorphisms (SNPs) to identify the predisposing genes and alleles (Simon-Sanchez & Singleton, 2008). The rationale is that linkage studies have some drawbacks, and the tools to identify risk genes through association analysis are now available (Frazer et al., 2007). One drawback is that linkage studies require the collection of families, which can be difficult to find and expensive to ascertain. Also, linkage analysis is not sensitive enough to detect the alleles that confer small risks.

Genomewide association studies (GWAS) conducted on large samples of cases and well-matched controls can identify alleles with small effects. They capitalize on the likely linkage disequilibrium between genetic risk variants and the nearby SNPs that have traveled with them on the same chromosome throughout the history of the population. It is thought that there are 10 million SNPs in humans, and that a large fraction of the genetic variants that contribute to the risk of disease will be captured by the commercial panels of 500,000 to 1,000,000 SNPs available for these studies. The mechanism by which fewer genotyped SNPs capture the variation of others is referred to as “tagging.” GWAS has met with success in a number of complex disorders, but in many of those cases the associated variants explain very little of the genetic risk for the trait (Willer et al., 2009). ASD GWAS studies are currently being conducted. In preliminary findings, significant associations with common variants have not been observed, and the focus has been mainly on rare copy number variants. Although genetic models are not needed for GWAS studies, strategies to reduce genetic complexity will be important. As with linkage, methods are available to test quantitative endophenotypes for association with SNP panels.

2.0 ASD: Reducing Genetic Complexity

Identifying the genetic contributions to complex disorders such as ASD has not been straightforward. Study designs and analytic methods that have been suggested (Almasy & Blangero, 2001), indicate that a more successful path to gene identification would include the identification and analysis of traits with simpler genetic models. There are two straightforward strategies to reducing complexity. The first is to stratify the members of the study sample by a factor thought to contribute to genetic heterogeneity. For example, one could know or assume that those with ASD who are nonverbal have a different set of predisposing genes than those who are affected and verbal. Depending on the analysis conducted, the members of the ASD study sample are stratified by a feature of ASD and separate gene finding analyses are conducted in the two samples.

A second strategy is to analyze a quantitative trait or endophenotype that is associated with ASD and known or assumed to capture a single or reduced number of genetic dimensions of the disorder. It is anticipated that the endophenotype is likely to result from fewer genes and alleles acting with less complexity. This approach can be successful if there are common genetic variants serving as the basis of the endophenotype (Wijsman, 2007). Using this strategy, genes predisposing to features of ASD are mapped through QTL analyses, an approach that is addressed here. To clarify, the goal is to identify heritable traits that have substantial variance and are known to be associated with or can contribute to a diagnosis of ASD. A common example illustrates this point. The genes that contribute to triglyceride levels, a continuous and heritable trait, are also likely to contribute to the risk for coronary artery disease (CAD), as high triglycerides are seen in those with CAD. Therefore triglyceride levels provide a good CAD endophenotype for QTL mapping. Identifying appropriate quantitative traits associated with ASD poses a greater challenge, because our understanding of its biology is not as well developed as that for CAD.

2.1 Quantitative Endophenotypes

The goal is to distill the genetically complex phenotype into more specific heritable features. That is, for many complex disorders including ASD, the diagnosis of being affected results from positive responses on a checklist of observed features or behaviors. For example, a genetically complex autoimmune disease like systemic lupus erythematosus is diagnosed if any of four out of eleven criteria are met (Hochberg, 1997); metabolic syndrome, which is also complex, is associated with diabetes and coronary artery disease and diagnosed when several criteria involving disturbances in metabolism and lipids are met (Qiao et al., 2009). Most relevant to ASD, neurobiological disorders are diagnosed when an individual exhibits a pattern of behaviors consistent with criteria listed in the Diagnostic and Statistical Manual (American Psychiatric Association, 1994). Focusing on single traits from these checklists may reveal entities that derive from the major effects of fewer genes with a simpler genetic model. Also, since the analysis of quantitative traits is more statistically powerful than the analysis of discrete traits, identifying quantitative features of the ASD phenotype can provide a robust approach to gene identification.

2.1.1 A Quantitative Trait Model

The genetic contribution to a quantitative trait or endophenotype can be illustrated by a simple arbitrary model with 3 genes. While the genetics of any “real” quantitative trait will be more complex, the principles illustrated by this model can be applied. In this hypothetical model, there are 3 genes each with 2 alleles contributing to Trait1 in an additive fashion. The alleles are “A” and “a” for gene1, “B” and “b” for gene2, and “C” and “c” for gene 3, and each capital letter allele contributes 3 to the trait value, while each lower case letter allele contributes 1. Thus, a person with genotype AA, Bb, and cc would have a trait value of 12. The maximum and minimum trait values are 18 and 6, respectively. Complexities might include gene-gene or gene-environment interactions that would alter the trait values. If the alleles with capital letters exhibit dominance at each locus, someone with genotype AA, Bb, and cc would have a trait value of 14. Interactions among loci would produce a wider number of possible trait values. Ascertainment of an informative study sample is critical, and gene finding efforts are enhanced if the sample is drawn from families exhibiting the entire range of the trait. These related individuals are expected to be correlated in their trait values because they share common genes. Identifying the genes by QTL analyses capitalizes on the expected relationships of trait values and marker allele sharing by the relative pairs in the pedigrees.

2.1.2 Three Criteria for Good QTL Endophenotypes

QTL mapping, like linkage analysis, is directed toward finding the chromosomal loci that harbor genes contributing to a trait, however here the trait is a quantitative endophenotype. Specific criteria must be met for an endophenotype to be appropriate for QTL mapping. The first is that the endophenotype exhibit substantial variation in the study sample under analysis. Achieving this can sometimes be difficult. A powerful study design consists of genotyped individuals in pedigrees ascertained for those with the complex disorder, where both the affected and unaffected are measured. The contrast of these individuals provides good power to detect the correlation of trait value differences and allele sharing that is used to map the genes. Thus, pedigrees with a large number of affected individuals are more powerful.

Sometimes the trait can only be measured in those who are affected. An example is the age of onset of the disorder. A common design consists of multiplex nuclear families with at lease two affected members measured for the endophenotype. This design will provide less power than one including unaffected individuals from the families, but it may be necessary for the proposed endophenotype. For other disorders, the endophenotype can only be assessed effectively in unaffected individuals. This will reduce the statistical power to identify the predisposing genes, as it leads to an enrichment of alleles that have very small effects on the trait value, since the bigger effects are likely to occur in the affected who have the disturbed trait values. Trait variation can be continuous, such as IQ, or ordinal, such as the degree of dysfunction (mild, moderate, or severe). A statistical power calculation taking the nature of the study sample, level of significance, and distribution of the trait can be conducted. It will allow the investigator to estimate the detectable effect size of the predisposing alleles. Since there is no simple approach that can be used in every situation, it is important to evaluate the distribution of the quantitative trait in the choice of analytic method and interpretation of results.

The second criterion for effective QTL mapping is that the endophenotype should be associated with the binary definition of the disorder. That is, one should be able to subdivide the values of the trait so that it correlates with binary definition of the disorder. Figure 39-2 illustrates this point. Here, in Figure 39-2a, the horizontal axis reflects the observed values of the trait, and the vertical axis reflects the percent or probability of each of those values. Hypothetically, the vertical line that is drawn divides those who are affected from those who are not. This indicates there is a perfect relationship between the endophenotype and the binary disorder. In reality, the relationship will not be that strong. This is illustrated in Figure 39-2b, where, as in 39–2a, the white color reflects the trait values of those who are affected and the grey color reflects the trait values of those who are not affected, but a clean vertical line cannot be drawn. Since the quantitative trait is not used to diagnose the disorder directly, there are no strict rules regarding this expected relationship between quantitative values and diagnosis, and the judgment of the investigator as to the acceptability of this correlation is important.

Figure 39–2. Relationships of a quantitative endophenotype and a binary diagnosis.

Figure 39–2.
Relationships of a quantitative endophenotype and a binary diagnosis.

The third critical criterion is that the endophenotype have a significant heritability. The heritability of a quantitative trait is the fraction of the trait variance that can be attributed to genes. Narrow heritability is operating if the genes act in an additive fashion, as illustrated by the example in section 2.1.1. Broad heritability is the estimate that includes the interactive effects of genes, also discussed in that section. Because these interactions are difficult to identify or predict, investigators usually make the simplifying assumption that all gene effects are additive, and they estimate the narrow heritability. The closer the genetic model is to one that is additive, the easier it is to map the QTL. Estimating Heritability

The heritability of a trait in a given population can be estimated by calculating the observed trait correlations in classes of relative pairs, such as siblings, and adjusting them for their degrees of relationship. The most informative pairs are MZ twins, as they share all of their genes and their correlation without an adjustment is a direct estimate of trait heritability. DZ twins share on average half of their genes and doubling their correlation provides an estimate of heritability. Their contrast, twice the difference in these correlations, provides a better estimate of heritability because it accounts for the sampling error and environmental correlations in these estimates. There are more complex methods to estimate heritability that involve analyses of trait variance (Rice, 2008).

Sometimes only sibling pairs are available for a heritability estimate. While the trait correlation for these pairs can be doubled for a rough estimate of trait heritability, it includes the effects of common family environment. Thus, it provides an upper bound for the heritability of the trait, which can be much smaller. Using multiple types of relative pairs provides a more refined estimate, but the “gold standard” is derived from data on MZ and DZ twin pairs. For most study designs, the heritability estimate has a standard error that can be used to test the null hypothesis that it is zero.

The heritability of a quantitative trait provides an upper bound for the heritability contributed by any one locus, referred to as the locus specific heritability. For example, if three genes contribute to the trait, the heritability will be divided among the three loci, although not necessarily evenly, giving some of the loci with bigger effects a better chance of detection by QTL analysis. If many genes contribute to the trait value, the heritability can be high and the locus specific heritabilities can all be low. Thus, a high heritability for a trait that is polygenic does not guarantee that a QTL will be identified. A well-defined study for a highly heritable trait, a sufficient sample size, and luck will all be important.

3.0 ASD: Quantitative Endophenotypes

The identification of quantitative endophenotypes for ASD is at an early stage of development, and QTL studies may be premature for many quantitative traits. However, several authors have addressed the value of identifying endophenotypes for general psychiatric disorders (Gottesman & Gould, 2003; Szatmari et al., 2007; Walters & Owen, 2007), and others provided general guidance for their definition in ASD (Allison et al., 2008; Chiu et al., 2008; Pickles et al., 2000; Viding & Blakemore, 2007; Wijsman, 2007). Their general guidelines for ASD suggest quantifying the features of a particular symptom and including normally functioning adults who also exhibit aspects of the trait, when this is feasible. The guidelines acknowledge that often the traits can only be assessed in those who are affected, limiting the range of variability. Suggested classes of traits include dimensions of mental skills that can explain features of ASD (Baron-Cohen, Wheelwright, Skinner, Martin, & Clubley, 2001). For example, employing an instrument to assess and measure the degree of central coherence could allow investigators to capture and quantify the skill that permits savant behavior, because it is thought to be explained by weak central coherence (Viding & Blakemore, 2007).

3.1 Behavioral Questionnaire Endophenotypes

Appropriate endophenotypes can come from instruments devised to assess the features of (Hoekstra, Bartels, Verweij, & Boomsma, 2007; Skuse, Mandy, & Scourfield, 2005) and diagnose ASD (Piven, Palmer, Landa et al., 1997). The Quantitative Checklist for Autism in Toddlers (QCHAT), a 25-item checklist with a score of 0–4 on each item provides a number of possible traits for analysis. These items are correlated with ASD, as the mean scores for those with ASD are significantly higher than for those who are unaffected (Allison et al., 2008). Additional analyses of the heritability or familiality of the scores would be important for further work. The broad autism phenotype (BAP) is described by features of language and personality, ranging from mild to severe. It is correlated with, but does not lead to autism, and can be quantified as a potential endophenotype (Losh & Piven, 2007; Piven, Palmer, Jacobi, Childress, & Arndt, 1997). The Broader Phenotype Autism Symptom Scale (BPASS) is a good source of endophenotypes. A trained clinician measures social motivation, social expressiveness, conversational skills, and flexibility in ASD-affected individuals and their family members along a quantitative continuum (Dawson et al., 2007; Dawson et al., 2002). Developers of this instrument anticipated the possibilities of QTL studies, as they identified BPASS traits that are familial (Sung et al., 2005) and correlated with ASD.

The Social Responsiveness Scale (SRS) (Constantino et al., 2006; Constantino & Todd, 2003) provides a quantitative score measuring the deficit in this defining feature of ASD. The SRS score is highly correlated with ASD and provides a measure of a single dimension of social skills. It has been validated, and found to be heritable in normal individuals, as assessed by twin studies. It can be administered to pedigree members and provides an excellent trait for QTL mapping. A small QTL study was conducted in nuclear families with ASD, which is discussed more extensively in section 5.0.

3.2 Behavioral Traits: Autism Diagnostic Instrument Revised (ADIR)

The ADIR, which has been used extensively to diagnose ASD (Lord, Rutter, & Le Couteur, 1994), provides a wealth of potential endophenotypes for QTL analyses. It assesses deficits in the domains of language and social skills and identifies the presence of stereotyped and repetitive behaviors. Quantification of these deficits and their specific features provides candidate endophenotypes for QTL analyses. These capture the global degree of deficit in language or social skills or focus on the quantitative assessment of a single dimension of language or social skills, such as the degree of delay in speech.

The ADIR is an assessment of approximately 100 ASD features by trained clinicians interrogating caregivers. Many questions address the levels at which the child exhibits particular deficits. The value for each item varies from 0 to 3, with a higher score reflecting a greater deficit. Thus, scores are elevated in those with ASD. A single item provides an ordinal score and combinations of items provide a more continuous range of values. Scores for item combinations can be achieved by direct addition or derived from multivariate statistical methods like factor analysis (Afifi, Clark, & May, 2004), which is applied to find a linear combination of the scores on individual items reflecting language deficits that are most correlated with each other in those with ASD. More sophisticated analytic methods are also available to find multivariate quantitative traits appropriate for QTL analysis (Kutner, Nachtsheim, & Neter, 2004).

3.3 Autism Diagnostic Observation Schedule (ADOS)

The ADOS is a semistructured assessment with diagnostic modules that are customized for both nonverbal and verbal patients and tailored for individuals that range in age from toddlerhood to adulthood. The assessment is open-ended, and an examiner uses a series of situations and interview questions to elicit behavior and verbal responses from the individual tested. An overall assessment is used to discriminate among forms of ASD, such as autism itself, pervasive developmental delay, or Asperger’s syndrome. Since it involves direct observation of a person’s behavior by an examiner who is taking careful note of traits and behaviors central to the diagnosis of autism, it is difficult to convert this instrument into a tool with quantitative traits. Thus, the ADOS is much less adaptable to QTL analyses than the ADI-R and the other instruments discussed in this chapter. These differences indicate that one must recognize that it is not only the quality of the traits that must be considered for QTL studies, but the nature of their measurement as well. An interactive tool for diagnosis may not adapt well to providing a consistent and well-defined phenotype.

3.4 Biological Endophenotypes

Although knowledge of the biology of ASD is currently limited, investigations suggest a number of ASD-correlated, heritable, and variable biological endophenotypes appropriate for QTL analyses. For example, an increased number of those with ASD exhibit a large head circumference for their age and sex (Fombonne, Roge, Claverie, Courty, & Fremolle, 1999; Lainhart et al., 2006; Sacco et al., 2007). Significant familial correlations in head circumference have been observed in family members of those with ASD, indicating this is a good ASD endophenotype for QTL analysis (Spence, Black, Miyamoto, & Geschwind, 2005). The sizes of certain structures of the brain have been implicated as causing the large head (Courchesne et al., 2007), and if the sizes are familial they may also be appropriate endophenotypes for QTL analysis. This endophenotype has already been associated with a genetic polymorphism (Conciatori et al., 2004). Increased blood serotonin levels have been reported in those with ASD (Cross et al., 2008; Weiss et al., 2006). The trait is familial, making it a good candidate for QTL analysis in pedigrees ascertained for individuals with ASD.

4.0 QTL Analyses

QTL analyses are conducted to localize genes that contribute to quantitative traits or endophenotypes. They were originally designed to localize genes in model organisms (Mackay, 2004) and in particular inbred strains of mice (Flint, Valdar, Shifman, & Mott, 2005). Inbred mouse strains are derived by breeding siblings for 20 generations until they are homozygous at almost every locus. Mating mice from two different inbred strains that differ significantly in their trait mean provides a straightforward approach to localizing genes contributing to that trait. Using this design, a significant test statistic identifies marker loci where mice differing in their genotypes also differ in their mean trait values. The differences are assessed by t-tests or Analyses of Variance (ANOVA) (Ott & Longnecker, 2000), and the loci with markers exhibiting significantly different trait means reveal the QTL. In inbred mice, gene identification at a QTL is a significant challenge, because they exhibit reduced genetic variability, making their QTL very broad, and thus encompassing many genes. Very large numbers of mice are needed to narrow the QTL by breeding, and complex study designs have been proposed to address this problem (Rockman & Kruglyak, 2008). After a QTL or gene is identified in mice, the human orthologs can be found using mouse/human comparative maps available at the Jackson Laboratories Web site (

4.1 Methods of QTL Analyses in Humans

This method of QTL analysis for inbred mice has been extended to address gene localization in outbred humans. Multiple analytic approaches are available, and factors that influence the choice are the degree of distributional normality of the trait and the configuration of the study sample. QTL analyses differ when the trait is continuous with a normal distribution and studied in several large pedigrees compared to when it is ordinal with a few categories and studied in a large number of nuclear families. Table 39-1 outlines some of the options in QTL analysis.

Table 39–1. QTL mapping approaches

Study Sample

Genetic Markers

Analytic Method



Large Pedigrees

Multiallelic Markers or


Analysis of Variance Components




The trait variance is partitioned into genetic and environmental components

Nuclear Families

Multiallelic Markers or



Kruskal Wallis test

Sibpal in SAGE

Nonparametric in GENEHUNTER

Identity-by-descent allele sharing in siblings is correlated with their trait differences

Parent/Child Trios


Family Based Association Test


Preferential Transmission of alleles to those with higher trait values

Cases and/or Controls


Quantitative Association Measured Genotype

Any General Statistical Package

Analysis of Variance or


4.1.1 Variance Components Analyses

If large pedigrees with most members measured for a normally distributed endophenotype are available, a variance component analysis is the most appropriate and statistically powerful approach. Computer programs that implement this method include SOLAR (Blangero & Almasy, 1996) (Almasy et al., 1999), MENDEL (Lange et al., 2001), and MERLIN (Abecasis, Cherny, Cookson, & Cardon, 2002; Heath, 1997). In each case, the trait variance is partitioned into components attributable to genes, the environment, and when appropriate, their interactions. The partition is achieved by comparing the degrees of the genetic relationships among the pedigree members and the covariances in their trait values, resulting in an estimate of trait heritability. Genetic markers are incorporated into the analysis to estimate marker allele sharing IBD among the relative pairs in the pedigrees and these IBD estimates are analyzed in relation to the covariances among the trait values. The QTL analysis tests whether including the effects of a gene contributing to the trait at that marker locus better explains the data. This is accomplished by a likelihood ratio test, which is similar to what is used by the LOD score linkage analysis. The test statistic compares the likelihood of the data with a trait influencing gene in the chromosome region included compared to the likelihood of the data without this gene. The loci with LOD scores exceeding a predetermined threshold are considered to be QTL, and the analysis allows for an estimate of the locus specific heritability at each QTL.

If the pedigrees under analysis have a large number of affected individuals, and the distribution of their trait values is bimodal, thus violating the normality assumption of this statistical method, transformation of the trait values to fit a normal distribution would satisfy the assumptions of the method. Alternatively, a nonparametric mapping method, such as those in given Table 39-1 may be more appropriate. Algorithms to conduct variance component QTL analyses can use a great deal of computer time, particularly if the analysis is multipoint, which incorporates the information from many markers to estimate the allele sharing IBD, rather than single point, where each marker is analyzed individually. Multipoint analyses consider more information simultaneously, and are thus more powerful statistically.

4.1.2 QTL Mapping for ASD Traits

Since individuals with ASD rarely have offspring, and their parents may respond to the burden of caring for a child with this disorder with stoppage (Slager, Foroud, Haghighi, Spence, & Hodge, 2001), ceasing to have children, it is difficult to identify large pedigrees with substantial numbers of ASD-affected individuals for variance components QTL studies. In this case, sibpair analyses are appropriate and families with affected sibling pairs are most often available for QTL studies. Here, the analysis finds QTL contributing to trait variation in those with ASD (Spiker, Lotspeich, Dimiceli, Myers, & Risch, 2002). A more powerful QTL analysis can be derived from panels of ASD parent/affected child trio samples that are being collected for genomewide association studies. Including an unaffected sibling measured for the endophenotype in the study sample provides a powerful contrast to the affected sibling. QTL mapping in these nuclear families can also employ a variance components approach when the trait is normally distributed.

For traits that are not normally distributed, several algorithms are available for sibpair QTL analyses. All are based on IBD sharing in sibpairs. The algorithms compare trait differences and IBD allele sharing, combining the results over the sibling pairs. It is expected that IBD sharing and trait differences will be negatively correlated at a QTL, because for each pair, the greater their allele sharing, the closer their trait values should be. This approach was originally proposed by Haseman and Elston (1972), and it is still in use today. The Statistical Analysis of Genetic Epidemiodogy (SAGE) software conducts these analyses ( Variations on this method include a nonparametric analysis in the GENEHUNTER software (Kruglyak, Daly, Reeve-Daly, & Lander, 1996). Using that algorithm, the trait differences are ranked over the complete sample of sibpairs and the differences are categorized by the observed IBD values in the pairs. The Kruskal Wallis test, a nonparametric ANOVA, is conducted (Ott & Longnecker, 2000). Consistent with other nonparametric statistical methods, using ranks as is done here prevents the undue influence of trait value outliers on the analysis.

4.2 Interpreting QTL Results

QTL are identified by setting a threshold for the test statistic that reflects the strength of the evidence at that locus. The test statistic follows a normal or t-distribution statistic or is reported as a LOD score, depending on the algorithm. Figure 39-3 illustrates the genomewide results of a QTL analysis in graphic form. The genome is represented across the horizontal axis, with the chromosomes listed from one to 22. The vertical axis is a value from the normal distribution. The test is one-sided, so that all the probability is in a single tail. The best result occurs on chromosome 1, with a z-score being a little greater than 3.5. The probability of observing this score or one that is larger when there is no endophenotype gene is approximately .000001, making this a likely QTL. If a z-score threshold is set at 3.0, QTL are also found on chromosomes 4 and 16.

Figure 39-4 illustrates the results of this QTL analysis on chromosome 1 in greater detail. The horizontal axis reflects the distance from the telomere of the chromosome in centimorgans (cM) and the vertical axis reflects the z-score at each point along the chromosome. As with all statistical tests, setting a less stringent p-value will provide more power to detect true signals, but will also permit more false signals to be included. Here we illustrate that QTL are usually wide and encompass many genes. Additional markers interspersed among those that are already in the analysis can often narrow the QTL. While it is more likely that the predisposing alleles are in genes under the peak of the QTL, stochastic variation in genotypes and trait values in the study sample may cause some variation in location of this peak. As with most linkage findings, replication is important. This principle is illustrated by the inclusion of a curve representing the QTL results on this chromosome in an independent sample.

Figure 39–4. QTL in original and replication samples.

Figure 39–4.
QTL in original and replication samples.

4.3 Association Studies at QTL

Once QTL have been identified, SNPs in the QTL can be genotyped for association analyses that can lead to gene identification. Table 39-1 gives information on two types of study designs for association analysis of quantitative traits. The first is to ascertain parent/child trios where the child is assessed for the endophenotype. The genotypes of both parents and the child are included to provide information on the alleles that are transmitted to the child and those that are not transmitted. The untransmitted alleles act as “controls” that are matched within each trio for ethnicity and other important factors. The Family Based Association Test (FBAT) (Horvath, Xu, & Laird, 2001) tests if the transmitted SNP alleles associate with high or low values of the endophenotype more than one would expect by chance alone, while taking into account the alleles that could have been transmitted. If additional members of a sibship are included with the trio, their associations will not be independent, and the level of significance of the association can be a false positive result. Corrections for nonindependence of the siblings in a sibship can be achieved by using an empiric p-value or by adjusting the variance of the test, as is programmed in the FBAT software (Horvath et al., 2001). The last line of Table 39-1 refers to the simple design where parents and siblings of affected individuals are not studied. The association test for the quantitative trait is referred to as a measured genotype analysis, which is conducted using an one way ANOVA (Ott & Longnecker, 2000).

Figure 39-5 illustrates the results of association analyses at a QTL. The horizontal axis is the number of base pairs along the chromosome, from 80,000,000 to 220,000,000 and the vertical axis represents the p-values of the test statistics for each SNP. The plot uses the negative of the log of this p-value, where the log of a small p-value becomes a positive number that increases as the result get more significant. In this figure, four SNPs are associated with the endophenotype when the criterion is set at the Bonferroni corrected value of p < .0001, for 500 SNPs. These SNPs do not necessarily contribute to variation in the endophenotype directly, but are more likely to be in linkage disequilibrium with SNPs that do. Follow-up studies might include replication in an independent sample and sequencing of a sample of those who have the associated variant to potentially identify the causal variants.

Figure 39–5. Association analysis of Snps in the QTL.

Figure 39–5.
Association analysis of Snps in the QTL.

5.0 ASD QTL Analyses

Investigators are just beginning to develop sufficient study samples and identify appropriate endophenotypes for effective QTL analyses in ASD. As these evolve, substantial findings may begin to emerge. Currently, there are a number of published manuscripts and abstracts reporting QTL results, and they are summarized in Table 39-2.

Table 39–2. QTL and quantitative association results in ASD family panels





(Location and p-Value)


Age at First Word

Alarcon et al., 2002

Alarcon et al., 2005

152 ASD AGRE Families

QTL 7q34-36 (p<.001)


Age at First Word

Schellenberg et al., 2006

222 CPEA Families

QTL 9q33-34 (p<.0008)


Age at First Word

Alarcon et al., 2008

172 and 304 AGRE Parent/ Child Trios

CNTNAP2 gene 7q35

Association (p<.002)

Replication (p<.005)

Social Responsiveness Scale (SRS)

Duvall et al., 2007

99 AGRE Families

QTL 11p12-13 (p<.0007)

ADIR Nonverbal Communication (NVC)

Chen et al., 2005

228 AGRE Families

QTL 1p13-q12


ADIR Nonverbal Communication (NVC)

Yoon et al., 2008

219 AGRE Families

QTL 1p13-q12

Replication (p<.0001)

ADIR Social Interaction Domain (SOC)

Behavioral Domain


Liu et al., 2008

976 AGP Families

QTL 12q13 (p<.002)

QTL 14q22 (p<.001)

Most of the QTL analyses reported in Table 39-2 have been conducted on samples of nuclear families drawn from the Autism Genetics Research Exchange (AGRE, resource, from which trait and genotype data are freely available to interested investigators (Geschwind et al., 2001). The AGRE study sample currently has about 1400 multiplex families ascertained for two children affected with ASD. The children in this sample have been diagnosed with ASD by trained staff using the ADIR. The data have been developed over the last 10 years, and thus the QTL findings given in Table 39-2 report AGRE samples of differing sizes, depending on when they were conducted and when the quantitative measure were included in the protocol. More complete samples are in the process of being collected.

In addition to providing data on the individual items from the ADIR (Lord et al., 1994), the AGRE data set includes individual responses on the ADOS (Lord et al., 2000), parent and teacher scores on the SRS (Constantino & Todd, 2003), and measurements of head circumference, all discussed as potential endophenotypes in section 3. Multiallelic genotypes and SNP data are available for QTL analyses. Densely spaced SNPs genotyped on the Affymetrix and Illumina platforms are available for quantitative association analyses (

5.1 QTL Analyses: “Age at First Word”

QTL analyses on data from the AGRE sample began in 2002 with the analysis of the ADIR caregiver report of “age at first word,” which is provided for each child who has been verbal at some time in their childhood. Delayed speech is associated with ASD, and this endophenotype is measured in months, providing a correlated trait with sufficient variability. The sibling correlation is .33, which is significantly different from 0, making this endophenotype a good candidate for QTL analyses. The QTL analysis was conducted using the nonparametric option of the GENEHUNTER software, and the strongest evidence for a QTL was found on chromosome band 7q34-36 (Alarcon, Cantor, Liu, Gilliam, & Geschwind, 2002). This sample was expanded, and the same analysis was conducted in the larger sample in 2005. While there was no formal replication, evidence supporting this QTL was attenuated but remained significant (Alarcon, Yonan, Gilliam, Cantor, & Geschwind, 2005).

A QTL analysis of “age at first word” was subsequently conducted in a different collection of multiplex families, also reported in Table 39-2. Schellenberg and colleagues conducted their study on a sample of 222 ASD multiplex United States families from the NIH Collaborative Programs of Excellence in Autism (CPEA) collection (Schellenberg et al., 2006). Using the ADIR and the ADOS for diagnosis, the families each had two children meeting the criteria for autism, pervasive developmental disorder, or ASD, which makes the CPEA selection criteria somewhat different from the AGRE selection criteria. The QTL analyses were also conducted using a different analytic algorithm, as the CPEA sample was analyzed using a variance components approach, while the AGRE sample was analyzed using the nonparametric approach. This CPEA study did not replicate the QTL on 7q34-36, but found one at 9q33-34. The lack of replication for this particular trait while disappointing, can be attributed to a reduced sample size, differing clinical ascertainment criteria, and a different analytic method. Using the same diagnostic criteria and analytic methods would be important for a replication study.

5.2 QTL Analyses: The Social Responsiveness Scale

The SRS is an ASD endophenotype that covers a wide range of values and discriminates well among those who are socially adept and those who are not. Twin studies indicate that scores on the SRS are heritable in normal individuals (Constantino & Todd, 2003). A small study was conducted on 99 AGRE families, and a QTL was identified on 11p12-13. The locus was also found to also be linked to the binary trait of ASD in that sample, and the SRS provided greater power (Duvall et al., 2007). The AGRE sample is continuing to expand and include the SRS assessment in the protocol. Future studies in much larger samples may reveal important ASD loci through the analysis of this quantitative endophenotype. Quantitative association studies will be an important component.

5.3 QTL Analyses: “Nonverbal Communication”

The ADIR subscore assessing deficits in nonverbal communication (NVC) provides an appropriate endophenotype for ASD QTL analyses. The NVC score is composed of the individual scores on seven items from the ADIR, where the deficits are categorized in an ordinal fashion from 0 to 3. NVC skills would allow the child to engage with others by doing such things as pointing to share information. Deficits would reflect a lack in the ability to communicate with others that may be considered independent of language skills. This endophenotype is best measured in those who have ASD, and it exhibits the full range of values in that sample. The NVC correlation in ASD affected siblings is .21, indicating it is familial as well as correlated with ASD and variable in the sample under analysis. A nonparametric QTL analysis of 228 AGRE families identified a QTL on 1p13-q12 (Chen, Kono, Geschwind, & Cantor, 2006), which has been replicated in an independent sample of 213 AGRE families, using the same multiallelic genotyping, ascertainment scheme, and analytic approach (Yoon, Alarcon, Geschwind, & Cantor, 2008). The result of p < .0001 in the first sample and p < .0001 in the second provide consistent evidence of and NVC QTL in this region. This QTL is very broad, and targeted SNP association studies and follow-up sequencing studies are revealing the genes in the region that contribute to this endophenotype.

5.4 QTL Analyses: Autism Genome Project

Large-scale studies of multiple ASD samples are an important focus for the Autism Genome Project (AGP). The AGP conducted a QTL study on its combined sample of ASD multiplex families ascertained at 10 coordinating sites in the United States and Europe over a 30-year period. This sample includes the CPEA and AGRE samples. While genotyping was conducted on all AGP families, the QTL analyses were done on a subset of 976 for whom the quantitative data could be made compatible among the sites (Liu, Paterson, & Szatmari, 2008). In preliminary analyses, the data from the different sites with varying ascertainment and diagnostic criteria were combined to reflect the same phenotypes. Fortunately, the genotype platform and analytic method was the same for all of the families. Two ADIR derived endophenotypes were analyzed in this sample. They are the ADIR total scores on the reciprocal social interaction domain total (SOC), with a heritability estimate of .35 and the restricted, repetitive, and stereotyped patterns of behavior domain (BEH), with a heritability estimate of .52. Since the data were only collected in sibling pairs, these estimates reflect the effects of common family environment, which are likely to be nontrivial. The QTL analyses were conducted using the variance components approach as programmed in the Merlin software (Abecasis et al., 2002). The best QTL signals for these endophenotypes were on chromosomes 12q13.11 (p = .002) for SOC and 14q22.1 (p = .001) for BEH. These results, which exhibit marginal significance in the largest sample analyzed for an ASD QTL, can be explained in a few ways. First the heritability estimates derived from the sibling pairs in this sample include common family environment indicating the true trait heritabilities may be considerably lower than .33 and .52. In addition, there may be many genes contributing to these endophenotypes, each with small locus specific heritabilities that cannot be detected, even with a study sample of this size. It is anticipated that other endophenotype studies in this large sample will result in stronger QTL.

5.4.1 Association of “Age at First Word”

An association study in a sample of the AGRE families was conducted within the 7q34-36 QTL to find the gene influencing “age at first word” in children with ASD. The density of the SNPs was substantial, providing more information than what would be seen in microarray SNP panel. The SNPs were tested for association individually, and haplotypes in a moving window were also tested. While several associations were observed, replication studies in an independent sample of AGRE families implicated only one haplotype that was located in the CNTNAP2 gene (Alarcon et al., 2008). Remarkably, the same gene showed association with the binary ASD phenotype in the AGRE families (Arking et al., 2008).

These QTL studies followed by association analysis and replication for “age at first word” provides strong support for the QTL approach proposed herein. QTL studies with larger samples and more refined endophenotype are likely to reveal other ASD genes.

5.4.2 A Roadmap for QTL Discovery

From a statistical genetics perspective, the analytic tools for QTL studies are in place. These tools await the application of suitable ASD phenotypes that are variable within the population under analysis, correlated with ASD, and exhibit evidence of a reasonable nonzero heritability. The nature of the study sample and distribution of the trait will limit the feasible analytic approaches.

These same design principles have been successfully applied to the study of coronary artery disease (CAD) through QTL analyses of heritable quantitative traits such as serum cholesterol and triglycerides levels. An important advantage for CAD studies, however, is that long-term epidemiology studies revealed important quantitative risk factors before genomewide studies became feasible.

Thus, it becomes critical to identify the important ASD quantitative traits, although the epidemiology studies are only in their very early stages. Perhaps the best approach currently is to create an environment where experts in the clinical and research features of ASD can participate in collaborative studies with investigators having QTL expertise and epidemiologists focused on revealing important ASD risk factors. It is expected such endeavors are likely to create a synergy among investigators that will lead to success.


The genetic complexity of ASD can be addressed by studying correlated quantitative endophenotypes that have simpler modes of inheritance. Successful ASD endophenotypes will exhibit wide variation and substantial heritability. QTL studies of these endophenotypes involve selecting an appropriate study sample that captures sufficient variation in those who are affected and possibly those who are unaffected. The choice of statistical algorithm for QTL mapping will depend on the individuals who can be studied and the distribution of the endophenotype. Although QTL studies for ASD are at an early stage, its application to ASD led to initial successes. The QTL for deficits in nonverbal communication has been replicated in an independent sample and awaits effective association studies. The CNTNAP2 gene in a language related QTL shows association with the language-related quantitative trait and ASD. Application of this approach to more refined endophenotypes in larger samples can lead to a substantial increase in our understanding of the genetic etiology of ASD.

Challenges and Future Directions

The identification of the best quantitative endophenotypes for QTL mapping remains a challenge. New study samples of ASD affected children and their unaffected offspring provide opportunities to pursue this approach. Synergy among experts on the ASD phenotypes, ASD epidemiology, and QTL mapping could energize the field. Once QTL are identified, it will be an interesting challenge to integrate GWAS, copy number variation, and sequence data into studies of these endophenotypes.

Suggested Readings

Alarcon, M., Abrahams, B. S., Stone, J. L., Duvall, J. A., Perederiy, J. V., Bomar, J. M., et al. (2008). Linkage, association, and gene-expression analyses identify CNTNAP2 as an autism-susceptibility gene. American Journal of Human Genetics, 82(1), 150–159.

Abrahams, B. S., & Geschwind, D. H. (2008). Advances in autism genetics: On the threshold of a new neurobiology. Nature Reviews Genetics, 9, 341–355.

Lynch, M., & Walsh, B. (1998). Genetics and the analysis of quantitative traits. Sunderland, MA: Sinauer Associates.


Abecasis, G. R., Cherny, S. S., Cookson, W. O., & Cardon, L. R. (2002). Merlin: rapid analysis of dense genetic maps using sparse gene flow trees. Nature Genetics, 30(1), 97–101.Find this resource:

Afifi, A., Clark, V. A., & May, S. (2004). Computer-aided multivariate analysis (4th ed.). Boca Raton, FL: CRC.Find this resource:

    Alarcon, M., Abrahams, B. S., Stone, J. L., Duvall, J. A., Perederiy, J. V., Bomar, J. M., et al. (2008). Linkage, association, and gene-expression analyses identify CNTNAP2 as an autism-susceptibility gene. American Journal of Human Genetics, 82(1), 150–159.Find this resource:

    Alarcon, M., Cantor, R. M., Liu, J., Gilliam, T. C., & Geschwind, D. H. (2002). Evidence for a language quantitative trait locus on chromosome 7q in multiplex autism families. American Journal of Human Genetics, 70(1), 60–71.Find this resource:

    Alarcon, M., Yonan, A. L., Gilliam, T. C., Cantor, R. M., & Geschwind, D. H. (2005). Quantitative genome scan and ordered-subsets analysis of autism endophenotypes support language QTLs. Molecular Psychiatry, 10(8), 747–757.Find this resource:

    Allison, C., Baron-Cohen, S., Wheelwright, S., Charman, T., Richler, J., Pasco, G., et al. (2008). The Q-CHAT (Quantitative CHecklist for Autism in Toddlers): a normally distributed quantitative measure of autistic traits at 18–24 months of age: preliminary report. Journal of Autism and Developmental Disorders, 38(8), 1414–1425.Find this resource:

    Almasy, L., & Blangero, J. (2001). Endophenotypes as quantitative risk factors for psychiatric disease: rationale and study design. American Journal of Medical Genetics, 105(1), 42–44.Find this resource:

    Almasy, L., Hixson, J. E., Rainwater, D. L., Cole, S., Williams, J. T., Mahaney, M. C., et al. (1999). Human pedigree-based quantitative-trait-locus mapping: localization of two genes influencing HDL-cholesterol metabolism. American Journal of Human Genetics, 64(6), 1686–1693.Find this resource:

    American Psychiatric Association. (1994). Diagnostic and statistical manual of mental disorders (4th ed.; DSM-IV). Washington, DC: Author.Find this resource:

      Arking, D. E., Cutler, D. J., Brune, C. W., Teslovich, T. M., West, K., Ikeda, M., et al. (2008). A common genetic variant in the neurexin superfamily member CNTNAP2 increases familial risk of autism. American Journal of Human Genetics, 82(1), 160–164.Find this resource:

      Bailey, A., Le Couteur, A., Gottesman, I., Bolton, P., Simonoff, E., Yuzda, E., et al. (1995). Autism as a strongly genetic disorder: evidence from a British twin study. Psychological Medicine, 25(1), 63–77.Find this resource:

      Baron-Cohen, S., Wheelwright, S., Skinner, R., Martin, J., & Clubley, E. (2001). The autism-spectrum quotient (AQ): evidence from Asperger syndrome/high-functioning autism, males and females, scientists and mathematicians. Journal of Autism and Developmental Disorders, 31(1), 5–17.Find this resource:

      Blangero, J., & Almasy, L. (1996). SOLAR: Sequential oligogenic linkage analysis routines. Population Genetics Laboratory Technical Report No. 6. San Antonio, TX: Southwest Foundation for Biomedical Research.Find this resource:

        Breiman L. (2001). Random forests. Machine Learning, 45, 5–32.Find this resource:

        Cantor, R. M. (2006). Linkage analysis. In A. E. H. Emery & D. L. Rimoin (Eds.), Principles and practice of medical genetics (5th ed.). London: Churchill Livingstone.Find this resource:

          Cantor, R. M., Kono, N., Duvall, J. A., Alvarez-Retuerto, A., Stone, J. L., Alarcon, M., et al. (2005). Replication of autism linkage: fine-mapping peak at 17q21. American Journal of Human Genetics, 76(6), 1050–1056.Find this resource:

          Chen, G. K., Kono, N., Geschwind, D. H., & Cantor, R. M. (2006). Quantitative trait locus analysis of nonverbal communication in autism spectrum disorder. Molecular Psychiatry, 11(2), 214–220.Find this resource:

          Chiu, P. H., Kayali, M. A., Kishida, K. T., Tomlin, D., Klinger, L. G., Klinger, M. R., et al. (2008). Self responses along cingulate cortex reveal quantitative neural phenotype for high-functioning autism. Neuron, 57(3), 463–473.Find this resource:

          Conciatori, M., Stodgell, C. J., Hyman, S. L., O’Bara, M., Militerni, R., Bravaccio, C., et al. (2004). Association between the HOXA1 A218G polymorphism and increased head circumference in patients with autism. Biological Psychiatry, 55(4), 413–419.Find this resource:

          Constantino, J. N., Lajonchere, C., Lutz, M., Gray, T., Abbacchi, A., McKenna, K., et al. (2006). Autistic social impairment in the siblings of children with pervasive developmental disorders. American Journal of Psychiatry, 163(2), 294–296.Find this resource:

          Constantino, J. N., & Todd, R. D. (2003). Autistic traits in the general population: a twin study. Archives of General Psychiatry, 60(5), 524–530.Find this resource:

          Courchesne, E., Pierce, K., Schumann, C. M., Redcay, E., Buckwalter, J. A., Kennedy, D. P., et al. (2007). Mapping early brain development in autism. Neuron, 56(2), 399–413.Find this resource:

          Cross, S., Kim, S. J., Weiss, L. A., Delahanty, R. J., Sutcliffe, J. S., Leventhal, B. L., et al. (2008). Molecular genetics of the platelet serotonin system in first-degree relatives of patients with autism. Neuropsychopharmacology, 33(2), 353–360.Find this resource:

          Dawson, G., Estes, A., Munson, J., Schellenberg, G., Bernier, R., & Abbott, R. (2007). Quantitative assessment of autism symptom-related traits in probands and parents: Broader Phenotype Autism Symptom Scale. Journal of Autism and Developmental Disorders, 37(3), 523–536.Find this resource:

          Dawson, G., Webb, S., Schellenberg, G. D., Dager, S., Friedman, S., Aylward, E., et al. (2002). Defining the broader phenotype of autism: genetic, brain, and behavioral perspectives. Developmental Psychopathology, 14(3), 581–611.Find this resource:

          Duvall, J. A., Lu, A., Cantor, R. M., Todd, R. D., Constantino, J. N., & Geschwind, D. H. (2007). A quantitative trait locus analysis of social responsiveness in multiplex autism families. American Journal of Psychiatry, 164(4), 656–662.Find this resource:

          Farrell, P. M., Rosenstein, B. J., White, T. B., Accurso, F. J., Castellani, C., Cutting, G. R., et al. (2008). Guidelines for diagnosis of cystic fibrosis in newborns through older adults: Cystic Fibrosis Foundation consensus report. Journal of Pediatrics, 153(2), S4–S14.Find this resource:

          Flint, J., Valdar, W., Shifman, S., & Mott, R. (2005). Strategies for mapping and cloning quantitative trait genes in rodents. Nature Reviews Genetics, 6(4), 271–286.Find this resource:

          Fombonne, E., Roge, B., Claverie, J., Courty, S., & Fremolle, J. (1999). Microcephaly and macrocephaly in autism. Journal of Autism and Developmental Disorders, 29(2), 113–119.Find this resource:

          Frazer, K. A., Ballinger, D. G., Cox, D. R., Hinds, D. A., Stuve, L. L., Gibbs, R. A., et al. (2007). A second generation human haplotype map of over 3.1 million SNPs. Nature, 449(7164), 851–861.Find this resource:

          Geschwind, D. H., Sowinski, J., Lord, C., Iversen, P., Shestack, J., Jones, P., et al. (2001). The autism genetic resource exchange: a resource for the study of autism and related neuropsychiatric conditions. American Journal of Human Genetics, 69(2), 463–466.Find this resource:

          Gottesman, I. I., & Gould, T. D. (2003). The endophenotype concept in psychiatry: etymology and strategic intentions. American Journal of Psychiatry, 160(4), 636–645.Find this resource:

          Happe, F., Ronald, A., & Plomin, R. (2006). Time to give up on a single explanation for autism. Nature Neuroscience, 9(10), 1218–1220.Find this resource:

          Haseman, J. K., & Elston, R. C. (1972). The investigation of linkage between a quantitative trait and a marker locus. Behavior Genetics, 2(1), 3–19.Find this resource:

          Heath, S. C. (1997). LOKI, ver 2.3.5: Markov chain Monte Carlo segregation and linkage analysis for oligogenic models. American Journal of Human Genetics, 61(3), 748–760.Find this resource:

          Hochberg, M. C. (1997). Updating the American College of Rheumatology revised criteria for the classification of systemic lupus erythematosus. Arthritis and Rheumatism, 40(9), 1725.Find this resource:

          Hoekstra, R. A., Bartels, M., Verweij, C. J., & Boomsma, D. I. (2007). Heritability of autistic traits in the general population. Archives of Pediatrics and Adolescent Medicine, 161(4), 372–377.Find this resource:

          Horvath, S., Xu, X., & Laird, N. M. (2001). The family based association test method: strategies for studying general genotype-phenotype associations. European Journal of Human Genetics, 9(4), 301–306.Find this resource:

          Ijichi, S., Ijichi, N., Ijichi, Y., Kawamura, Y., Hashiguchi, T., & Morioka, H. (2008). For others: epistasis and the evolutionary survival of an extreme tail of the quantitative distribution of autistic assets. Medical Hypotheses, 70(3), 515–521.Find this resource:

          Kruglyak, L., Daly, M. J., Reeve-Daly, M. P., & Lander, E. S. (1996). Genehunter, ver 2.1: Parametric and nonparametric linkage analysis: a unified multipoint approach. American Journal of Human Genetics, 58(6), 1347–1363.Find this resource:

          Kutner, M. H., Nachtsheim, C. J., & Neter, J. (2004). Applied linear regression models (4th ed.). Boston: McGraw-Hill.Find this resource:

            Lainhart, J. E., Bigler, E. D., Bocian, M., Coon, H., Dinh, E., Dawson, G., et al. (2006). Head circumference and height in autism: a study by the Collaborative Program of Excellence in Autism. American Journal of Medical Genetics. Part A, 140(21), 2257–2274.Find this resource:

            Lamb, J. A., Parr, J. R., Bailey, A. J., & Monaco, A. P. (2002). Autism: in search of susceptibility genes. Neuromolecular Medicine, 2(1), 11–28.Find this resource:

            Lange, K., Cantor, R. M., Horvath, S., Perola, M., Sabatti, C., Sinsheimer, J. S., et al. (2001). Mendel version 4.0: A complete package for the exact genetic analysis of discrete traits in pedigree and population data sets. American Journal of Human Genetics 69(supplement), 504.Find this resource:

            Liu, X. Q., Paterson, A. D., & Szatmari, P. (2008). Genome-wide linkage analyses of quantitative and categorical autism subphenotypes. Biological Psychiatry, 64(7), 561–570.Find this resource:

            Lord, C., Risi, S., Lambrecht, L., Cook, E. H., Jr., Leventhal, B. L., DiLavore, P. C., et al. (2000). The Autism Diagnostic Observation Schedule-Generic: a standard measure of social and communication deficits associated with the spectrum of autism. Journal of Autism and Developmental Disorders, 30(3), 205–223.Find this resource:

            Lord, C., Rutter, M., & Le Couteur, A. (1994). Autism Diagnostic Interview-Revised: a revised version of a diagnostic interview for caregivers of individuals with possible pervasive developmental disorders. Journal of Autism and Developmental Disorders, 24(5), 659–685.Find this resource:

            Losh, M., & Piven, J. (2007). Social-cognition and the broad autism phenotype: identifying genetically meaningful phenotypes. Journal of Child Psychology and Psychiatry and Allied Disciplines, 48(1), 105–112.Find this resource:

            Mackay, T. F. (2004). The genetic architecture of quantitative traits: lessons from Drosophila. Current Opinion in Genetics and Development, 14(3), 253–257.Find this resource:

            Manolio, T. A., Brooks, L. D., & Collins, F. S. (2008). A HapMap harvest of insights into the genetics of common disease. Journal of Clinical Investigation, 118(5), 1590–1605.Find this resource:

            Ott, J. (1999). Analysis of human genetic linkage (3rd ed.). Baltimore: The Johns Hopkins University Press.Find this resource:

              Ott, L., & Longnecker, M. (2000). An introduction to statistical methods and data analysis (6th ed.). Belmont, CA: Duxbury.Find this resource:

                Pickles, A., Starr, E., Kazak, S., Bolton, P., Papanikolaou, K., Bailey, A., et al. (2000). Variable expression of the autism broader phenotype: findings from extended pedigrees. Journal of Child Psychology and Psychiatry and Allied Disciplines, 41(4), 491–502.Find this resource:

                Piven, J., Palmer, P., Jacobi, D., Childress, D., & Arndt, S. (1997). Broader autism phenotype: evidence from a family history study of multiple-incidence autism families. American Journal of Psychiatry, 154(2), 185–190.Find this resource:

                Piven, J., Palmer, P., Landa, R., Santangelo, S., Jacobi, D., & Childress, D. (1997). Personality and language characteristics in parents from multiple-incidence autism families. American Journal of Medical Genetics, 74(4), 398–411.Find this resource:

                Qiao, Q., Laatikainen, T., Zethelius, B., Stegmayr, B., Eliasson, M., Jousilahti, P., et al. (2009). Comparison of definitions of metabolic syndrome in relation to the risk of developing stroke and coronary heart disease in Finnish and Swedish cohorts. Stroke, 40(2), 337–343.Find this resource:

                Rice, T. K. (2008). Familial resemblance and heritability. Advances in Genetics, 60, 35–49.Find this resource:

                Risch, N., Spiker, D., Lotspeich, L., Nouri, N., Hinds, D., Hallmayer, J., et al. (1999). A genomic screen of autism: evidence for a multilocus etiology. American Journal of Human Genetics, 65(2), 493–507.Find this resource:

                Roberts, L., Davenport, R. J., Pennisi, E., & Marshall, E. (2001). A history of the Human Genome Project. Science, 291(5507), 1195.Find this resource:

                Rockman, M. V., & Kruglyak, L. (2008). Breeding designs for recombinant inbred advanced intercross lines. Genetics, 179(2), 1069–1078.Find this resource:

                Ronald, A., Happe, F., Bolton, P., Butcher, L. M., Price, T. S., Wheelwright, S., et al. (2006). Genetic heterogeneity between the three components of the autism spectrum: a twin study. Journal of the American Academy of Child and Adolescent Psychiatry, 45(6), 691–699.Find this resource:

                Sacco, R., Militerni, R., Frolli, A., Bravaccio, C., Gritti, A., Elia, M., et al. (2007). Clinical, morphological, and biochemical correlates of head circumference in autism. Biological Psychiatry, 62(9), 1038–1047.Find this resource:

                Schellenberg, G. D., Dawson, G., Sung, Y. J., Estes, A., Munson, J., Rosenthal, E., et al. (2006). Evidence for multiple loci from a genome scan of autism kindreds. Molecular Psychiatry, 11(11), 1049–1060, 1979.Find this resource:

                Simon-Sanchez, J., & Singleton, A. (2008). Genome-wide association studies in neurological disorders. Lancet Neurology, 7(11), 1067–1072.Find this resource:

                Skuse, D. H., Mandy, W. P., & Scourfield, J. (2005). Measuring autistic traits: heritability, reliability, and validity of the Social and Communication Disorders Checklist. British Journal of Psychiatry, 187, 568–572.Find this resource:

                Slager, S. L., Foroud, T., Haghighi, F., Spence, M. A., & Hodge, S. E. (2001). Stoppage: an issue for segregation analysis. Genetic Epidemiology, 20(3), 328–339.Find this resource:

                Spence, S., Black, D., Miyamoto, J., & Geschwind, D. H. (2005). The occurrence of macrocephaly in autistic and non-autistic individuals from a large familial idiopathic autism sample (AGRE). 5th Annual International Meeting For Autism Research (IMFAR), Boston, MA (pp. S7.7).Find this resource:

                  Spiker, D., Lotspeich, L. J., Dimiceli, S., Myers, R. M., & Risch, N. (2002). Behavioral phenotypic variation in autism multiplex families: evidence for a continuous severity gradient. American Journal of Medical Genetics, 114(2), 129–136.Find this resource:

                  Sung, Y. J., Dawson, G., Munson, J., Estes, A., Schellenberg, G. D., & Wijsman, E. M. (2005). Genetic investigation of quantitative traits related to autism: use of multivariate polygenic models with ascertainment adjustment. American Journal of Human Genetics, 76(1), 68–81.Find this resource:

                  Szatmari, P., Maziade, M., Zwaigenbaum, L., Merette, C., Roy, M. A., Joober, R., et al. (2007). Informative phenotypes for genetic studies of psychiatric disorders. American Journal of Medical Genetics. Part B, Neuropsychiatric Genetics, 144B(5), 581–588.Find this resource:

                  Veenstra-Vanderweele, J., Christian, S. L., & Cook, E. H., Jr. (2004). Autism as a paradigmatic complex genetic disorder. Annual Review of Genomics and Human Genetics, 5, 379–405.Find this resource:

                  Viding, E., & Blakemore, S. J. (2007). Endophenotype approach to developmental psychopathology: implications for autism research. Behavior Genetics, 37(1), 51–60.Find this resource:

                  Walters, J. T., & Owen, M. J. (2007). Endophenotypes in psychiatric genetics. Molecular Psychiatry, 12(10), 886–890.Find this resource:

                  Weiss, L. A., Kosova, G., Delahanty, R. J., Jiang, L., Cook, E. H., Ober, C., et al. (2006). Variation in ITGB3 is associated with whole-blood serotonin level and autism susceptibility. European Journal of Human Genetics, 14(8), 923–931.Find this resource:

                  Wijsman, E. M. (2007). Statistical genetic approaches for analysis of autism and autism endophenotypes. 6th Annual International Meeting For Autism Research (IMFAR), Seattle, WA (pp. 84).Find this resource:

                    Willer, C. J., Speliotes, E. K., Loos, R. J., Li, S., Lindgren, C. M., Heid, I. M., et al. (2009). Six new loci associated with body mass index highlight a neuronal influence on body weight regulation. Nature Genetics, 41(1), 25–34.Find this resource:

                    Yoon, J. L., Alarcon, M., Geschwind, D. H., & Cantor, R. M. (2008). Replication and association analysis of a 1p13-q12 locus for nonverbal communication deficits in autism spectrum disorder. 7th Annual International Meeting For Autism Research (IMFAR) London, United Kingdom (pp. 105–109).Find this resource: