Phenotype definition and verification of genetic influence

To make the sample population as homogeneous as possible, it is important to minimize the presence of known environmental causes of developmental disorders, such as fetal alcohol syndrome, early childhood lead exposure, significant family stresses, or hearing loss. This can be done through a structured review of the medical history, including prenatal history, social history, psychosocial assessment, and formal audiologic evaluation, including auditory brainstem responses if necessary. Attention to subtle dysmorphic features may also produce clues to an underlying etiology (see Raymond, 2002, for an overview). When dysmorphic features are present, a clinical dysmorphology examination is indicated to determine a syndrome diagnosis. With some phenotypes, specific medical/genetic tests should also be done to rule out known etiologies. For example, individuals with dysmorphic features and cognitive impairment without a known syndrome diagnosis should have high resolution chromosome studies, subtelomere FISH study, and DNA testing for fragile X syndrome. Evaluation for autism could include high resolution chromosomal analysis with attention to chromosome 15q deletions or duplications, chromosome 17 deletions in the region of Smith Magenis syndrome, DNA testing for fragile X syndrome and Rett syndrome, physical examination to rule out tuberous sclerosis, and brain magnetic resonance imaging (MRI) evaluation for cerebral dysgenesis or other malformations (G. B. Schaefer, personal communication).

Depending upon the phenotype, identification of comorbid developmental conditions may also be important, such as assessment for attention-deficit/hyperactivity disorder (ADHD) in children with reading disability or motor problems in children with language disorders. These can alert the investigator to possible phenocopies that are secondary to another disorder; alternatively, the presence of two traits might indicate a subtype in which both disorders result from the same underlying genotype. Similarly, level of intelligence may be relevant to the phenotypic presentation and identification of subtypes. An accurate family history is especially important in regard to the investigation of comorbid conditions, in order to determine if the phenotypes segregate independently.

Definition of the phenotype is critical and should follow accepted guidelines for specificity and sensitivity, as determined by comparisons to appropriate controls. Phenotypic features should be objectively quantified, and for linkage studies should be classified as “affected,” “unaffected,” or “uncertain.” In addition, phenotypes that will be relevant to genetic etiologies should meet additional criteria that demonstrate that they are consistent over time and within families. Phenotypes that provide useful clinical descriptions to guide therapy, for example, may not reflect underlying genetic causes. An example of this may be the subtypes of ADHD as defined by the Diagnostic and Statistical Manual of Mental Disorders—Fourth Edition (DSM-IV; American Psychiatric Association, 1994): hyperactive/impulsive, inattentive, and combined. Through family studies, Faraone, Biederman, and Friedman (2000) and Smalley et al. (2000) found that subtype classification appeared to be influenced more by environmental than familial influences. The ability to follow a phenotype in a family is also dependent upon clear knowledge of the expression of that phenotype across different ages. Ideally, this would utilize longitudinal data to identify the changing phenotype with development. The trajectory of change may also be salient for some phenotypes.

Two commonly used methods to assess the genetic contribution to a phenotype, and thus its usefulness in gene identification studies, are to determine its heritability and to confirm its segregation in families. A third measure that is often cited is the sibling relative risk, which is the ratio of the probability of recurrence in siblings of an affected individual compared to overall prevalence. Sibling relative risk is an important indicator of “familiality,” but does not distinguish effects of family environment from genetic similarity. Thus, higher relative risk for a phenotype suggests but does not guarantee a genetic influence, but a low sibling relative risk for a phenotype would discourage its use in further gene localization studies.

Heritability

Family or twin studies can establish the heritability of a phenotype, with preference given to those phenotypes which show higher heritabilities, generally at least 0.50. This has been used in the selection of phenotypes for reading disability by Gayán and Olson (2001), in which measures of phonemic awareness and orthographic coding showed heritabilities between 0.60 and 0.70. These “component” phenotypes were chosen based on theoretical conceptualization of the reading process and verified by research indicating that the skills differentiated dyslexics from typical readers, even when they were matched for reading age, and the subsequent tests of heritability demonstrated that they were genetically influenced. In contrast, a more global phenotype was defined statistically through the use of discriminant analysis which identified patterns of test scores that differentiated good and poor readers. This was also found to have a heritability around 0.54 (DeFries & Alarcón, 1996). Todd et al. (2001) have used a more complex statistical method, latent class analysis, to define subtypes of ADHD that were then found to have higher heritabilities than the DSM-IV subtypes, which agreed with the previous analyses of segregation of the phenotypes in families.

Segregation analysis

Formal segregation analysis can test whether the transmission of a phenotype is random or consistent with a model of inheritance (dominant, recessive, polygenic, or a combination of modes of inheritance) and can also yield important information about the penetrance of the phenotype. Highly penetrant phenotypes will be the most appropriate for subsequent gene localization studies. Segregation in families can also demonstrate whether different phenotypes can actually be manifestations of the same disorder, such as the finding that ADHD was inherited independently from Tourette syndrome (Pauls et al., 1986) but that obsessive–compulsive disorder could be part of the Tourette phenotype (Pauls et al., 1995). Finally, family studies and subsequent segregation analysis have been used to directly identify phenotype clusters that were genetically transmitted, as demonstrated for dyslexia by Raskind, Hsu, Berninger, Thomson, and Wijsman (2000). Building on this analysis, subsequent studies by this group showed that the transmission of a phonemic decoding phenotype was more consistent with a dominant trait influenced by only a few genes, whereas a word attack phenotype was more polygenically influenced, making it less attractive for gene localization and identification studies (Chapman, Raskind, Thomson, Berninger, & Wijsman, 2003). In the study of autism, cluster analysis of symptoms and verification by sibling similarity identified five phenotypes with genetic influence. One of these, a savant skills cluster, showed linkage to chromosome 15q in the Prader Willi/Angelman deletion region that has been implicated in autism in other studies (Nurmi, Amin, 2003). A “reverse” approach, in which autism subtypes were assessed for their contribution to linkage to the 15q region, identified an additional phenotype, insistence on sameness (Shao et al., 2003). These examples illustrate the value of careful phenotype definition in the detection of linkage, as well as the chance that several phenotypes may end up attributed to the same gene.

Syndromes caused by single genes

Careful phenotypic definition can be critical in identifying genes that influence a cognitive disorder and thus identify pathogenic mechanisms. Conversely, by starting with individuals with a genetic syndrome known to be caused by mutations in a single gene and then characterizing the cognitive phenotype, the phenotypic spectrum resulting from mutations in that gene can be appreciated and the environmental contributions to that variation can be studied. The phenotypic spectrum of such genes may even extend to include nonsyndromic conditions, thus identifying causes for isolated cognitive disabilities through a candidate gene approach. By characterizing and comparing the phenotypes related to different genetic causes, the influences of specific genes can be isolated; at the same time, genetic disorders which have different causes but overlapping phenotypes may reflect deficits in common neural pathways. Examples of this are the studies of phenotypic similarities and differences between syndromic and nonsyndromic causes of autism such as tuberous sclerosis (TS), fragile X syndrome, Rett syndrome, and idiopathic autism.

Studies of TS have focused on the neuroanatomical characteristics of individuals with TS and autism compared to those with TS alone. Because children with TS have characteristic brain growths (tubers), it was hypothesized that the location of tubers would differentiate the autistic and nonautistic subgroups, thus localizing a region of the brain important in autism. It was first thought that tubers in the temporal lobe were associated with autism (Bolton, Park, Higgins, Griffiths, & Pickles, 2002), but subsequent studies did not confirm that finding. In a subsequent study of a pair of identical twins with TS but discordant for autism, members of this same group found that, while the number of tubers was the same in the twins, the autistic twin had larger tubers and more severe epilepsy (Humphrey, Higgins, Yates, & Bolton, 2004). In a larger study that found no difference in tuber location between autistic and nonautistic individuals with TS, the authors pointed out two possible confounds; first of all, there are two genes that cause TS, so that the diagnosis of TS does not, in fact, define a genetically homogeneous subgroup, and secondly, use of a global definition of autism may obscure phenotypic subtypes within that diagnostic category that may show neuroanatomic differences (Walz, Byars, Egelhoff, & Franz, 2002). Thus, while the studies of TS demonstrate the utility of investigating a syndrome to uncover causes of autism, they point again to the need for thorough phenotype and genotype definition.

Comparisons have also been made of the autistic characteristics among boys with fragile X syndrome, nonautistic boys with fragile X, and children with idiopathic autism. Study of boys with fragile X autism versus nonautistic fragile X boys could determine the factors that produce autism in only some of them, and comparison with idiopathic autism could aid in defining the salient features of fragile X autism. Initial studies in children under 4 years of age indicated that there were no significant differences in the autism characteristics between the children with fragile X autism and idiopathic autism, based on standardized instruments used to diagnose autism (Rogers, Wehner, & Hagerman, 2001); however, subsequent studies of the autism phenotype in fragile X versus idiopathic autism did suggest differences in behavioral characteristics. For example, boys with fragile X autism were less impaired in reciprocal social interaction when compared to children with idiopathic autism, but their communication deficits and stereotypic behavior were more prominent, leading to a characteristic social behavioral profile (Kau et al., 2004). Similarly, examination of language showed that boys with fragile X autism had greater difficulties with nonverbal cognitive skills and expressive language when compared to nonautistic boys with fragile X and children with autism. Deficits in receptive language particularly discriminated the boys with fragile X autism from the nonautistic fragile X boys, which could lead to identification of the factors that cause autism in this subset (Philofsky, Hepburn, Hayes, Hagerman, & Rogers, 2004). Thus, dissection of the phenotype may demonstrate differences between genetically distinct forms of autism. These differences may also be reflected at the neuroanatomical level; analysis of MRI scans of the cerebellar vermis has shown that, when comparisons are made between children with Down syndrome, fragile X with autism, fragile X without autism, and children with idiopathic autism, children with idiopathic autism have smaller vermian lobules VI and VII, whereas these regions are larger in boys with fragile X autism (Kaufmann et al., 2003). Further studies of fragile X could identify phenotypic, neuroanatomic, molecular, and environmental variables that determine which boys develop autism and how the causes of their autism relate to causes of other forms of autism.

Contiguous gene syndromes

Syndromes that are actually attributable to deletion of several contiguous genes offer a means to identify specific genes that cause separate features of the syndrome (Cassidy & Morris, 2002). This has been most effectively demonstrated with Williams syndrome, which includes a distinctive behavioral and language phenotype, that is, “a relative strength in auditory rote memory and language and extreme weakness in visuospatial constructive cognition, and a typical personality that includes overfriendliness, anxiety, and attention problems” (Morris & Mervis, 2000). This syndrome is due to a 1.6-MB deletion on one chromosome 7 in the region q11.23. (The genotype was originally discovered by quantifying the degree of cardiovascular disease in individuals with familial supravalvar aortic stenosis, performing linkage studies to localize the causative elastin gene, and then finding that elastin was deleted in Williams syndrome; see Ewart et al., 1993.) By comparison of the phenotypes of individuals with different sized deletions encompassing different genes, the genes contributing to specific aspects of the syndrome can be identified. For example, the connective tissue abnormalities of Williams syndrome such as supravalvar aortic stenosis and hernias can be attributed to deletion of the elastin (ELN) gene, while the deficits in visuospatial constructive cognition can be attributed to the LIM-kinase 1 (LIMK1) gene deletion (Frangiskakis et al., 1996) and the cognitive impairment is most likely due to the deletion of one copy of the GTF2I gene (Morris et al., 2003). Again, the ability to distinguish the genotype–phenotype relationships depended upon careful phenotype definition, including the development of a Williams Syndrome Cognitive Profile that reliably defined and quantified the features that were specific to the syndrome (Mervis et al., 2000). A similar analysis of the behavioral profile over time in Williams syndrome will be required to discover candidate genes for anxiety and attention-deficit disorder in the disorder.

Chromosomal abnormalities are also being used to identify genes for reading disability and for autism. Linkage of reading disability to chromosome 15q has been reported by several groups using positional cloning methods, and ascertainment of a Finnish family with a history of reading disability cosegregating with a chromosomal translocation involving the critical section of chromosome 15 led Taipale et al. (2003) to the discovery of the gene that was disrupted by the translocation. This gene, EKN1, has been designated DYX1C1, indicating that it is a candidate gene for the first dyslexia locus described. Markers in the gene were also found to be associated with reading disability in a sample of disabled readers from Finland. The function of this gene is unknown, and replication of this finding in other populations has not been reported yet.

Association of autism with abnormalities of chromosome 15q11–q13 (usually maternally derived duplications) was first noted in 1994 by Bundey, Hardy, Vickers, Kilpatrick, and Corbett (1994), and positive linkage findings with the region were first reported in 1997 by Pericak–Vance et al. (1997). These chromosomal loci were confirmed in several studies (e.g., Cook et al., 1997; Repetto, White, Bader, Johnson, & Knoll, 1998; Schroer et al., 1998), although Bolton et al. (2001) did not find that individuals with the chromosome abnormality fit classic autism criteria. Precise identification of the genes contained in the duplication is important, not only because these would be candidates for the influence on autism or autistic-like symptoms, but because there is a normal polymorphic duplication involving adjacent genes (Fantes et al., 2002), and inclusion of that duplication would lead to erroneous results. Linkage and association studies have narrowed the candidate genes within the duplicated region to GABRB3, UBE3A, and ATP10C (Buxbaum et al., 2002; Cook et al., 1998; Nurmi et al., 2001; Nurmi, Dowd et al., 2003; Shao et al., 2003). Deletion of this region is also involved in the pathogenesis of some cases of Prader Willi and Angelman syndromes, and mutations of the UBE3A gene alone cause some additional cases of Angelman syndrome (Kishino, Lalande, & Wagstaff, 1997; Matsuura et al., 1997). Further, children with Angelman syndrome can have features that overlap with autism (Steffenburg, Gillberg, Steffenburg, & Kyllerman, 1996).

Another syndrome that has features of autism and can also show phenotypic overlap with Angelman syndrome is Rett syndrome. This X-linked dominant disorder is almost always caused by mutation in the MECP2 gene (Amir et al., 1999). In a few cases, MECP2 mutations have been found in boys with phenotypic Angelman syndrome but normal molecular studies of chromosome 15q (Watson et al., 2001).Two children with Rett syndrome, MECP2 mutations, and rearrangement of 15q11–q13 have also been described (Longo et al., 2004). In addition, a small study of expression of the MeCP2 protein found dysregulation of the protein in a few individuals with Rett syndrome without MECP2 mutations as well as in autism and Angelman syndrome (Samaco, Nagarajan, Braunschweig, & LaSalle, 2004). Mutations in other X-linked genes, neuroligin-3 (NLGN-3) and neuroligin-4 (NLGN-4), may also cause autistic spectrum disorders (Jamain et al., 2003).

The genetic and phenotypic relationships between the genes influencing autism, Angelman syndrome, and Rett syndrome have prompted Dr. Huda Zoghbi to postulate a molecular mechanism involving genes influencing these conditions, particularly accounting for their apparent postnatal onset. MECP2 regulates other genes through methylation, although its targets are not known. Its temporal expression pattern in the brain and the consequences of mutation in animal models suggest that it is involved in maintenance or regulation of neuronal plasticity underlying the acquisition and retention of learned skills. Similarly, the NLGN genes may both be involved in synaptic differentiation through interactions with other gene products. UBE3A is a ubiquitin ligase, that participates in the degradation of proteins and is regulated by methylation. Because mutations of MECP2 have been noted to produce features very similar to Angelman syndrome in a few cases, there may be a functional relationship between UBE3A and MECP2. This leads to the hypothesis that methylation directed by MECP2 may regulate UBE3A, which in turn may be involved in the degradation of a neuroligin or one of the proteins it interacts with in the synapse; thus, all of these genes may be involved in a pathway that regulates synaptic modulation or maintenance, affecting the plasticity of synapses in the brain (Zoghbi, 2003). As Dr. Zoghbi points out, these are testable hypotheses and this is an example of how knowledge of individual genes and phenotypes can help explain the pathogenesis of a complex disorder.