Книги по МРТ КТ на английском языке / Functional Neuroimaging in Child Psychiatry Ernst 1 ed 2000

range, from childhood into adulthood, oVers promise for probing phenotypic variability (e.g., milder variants) in family genetic studies.

In Chapter 23, the editors address topics not covered in this book, such as promising new techniques, the combined uses of diVerent imaging modalities, and advances in research design and image analysis that allow for improved exploitation of exist-

ing imaging modalities. They highlight the need for integration with other aspects of clinical neuroscience and emphasize the focus within child psychiatry on executive and attentional dysfunctions, on emotion, and their interrelationships and neurochemistry. Emerging contributions of neuroimaging studies to child psychiatry are highlighted, and directions for future research proposed.

Introduction

The Welds of functional brain imaging and molecular genetics are at the forefront of biomedical research because these techniques are rapidly identifying new questions to be addressed and increasing the depth to which old questions can be addressed. The purpose of this chapter is to describe the techniques that will facilitate new understanding of neuroimaging results by adding a molecular genetic approach to examining neuroimaging phenotypes. It is hoped that these genetic techniques will enable researchers to identify the genes that play a role in the processes that underlie brain function.

What is meant by a neuroimaging phenotype, or trait? Examples include receptor density measured by ligands in positron emission tomography (PET) or the location and size of a brain region activated in response to a cognitive task measured by cerebral blood Xow techniques. A phenotype could also be a psychiatric diagnosis that has some relationship to neuroimaging measurements. DeWned broadly, a phenotype, or trait, is any observable characteristic that has a heritable component. In fact, it is the quantitative nature of neuroimaging that may provide an essential element for making signiWcant progress in Wnding the genes that underlie the biochemical and molecular pathways of brain function. The choice of the deWnition of the phenotype in a genetic study is of paramount importance for the success of the research enterprise, as is discussed in Chapter 19. The search for the genes responsible for a given phenotype needs to be guided by hypotheses, given that more than half of the 50±100000 human genes are expressed in the brain (Lewin, 1997).

A brief example of Mendelian genetics is provided initially to highlight the point that variation in a single gene can aVect many of the phenotypes measured in neuroimaging. This is followed by a description of complex genet-

ics ± the most likely to be encountered in neuroimaging studies ± in which several genes contribute to the phenotype of interest. The types of study design that are needed for complex genetics are then described, followed by the methods necessary to collect the genomic DNA, which is required for a genetic analysis of neuroimaging results. A glossary is provided at the end of the chapter for quick reference to the genetic terms encountered either within this chapter or within the material referred to during the chapter. Some readers may beneWt by referring to a general textbook of human genetics (e.g., Gelerter et al., 1998; Strachan and Read, 1996) to help to understand the following material.

Mendelian genetics: single gene mutations alter brain function

First, an important point of clariWcation: the terms altered genes or gene variants are used to avoid the common misnomer of a ªdisease geneº. The term disease gene suggests that an individual with the disease has a gene that nondiseased individuals do not. Rather, it is intended to refer to a mutated variant of the same gene that is inherited in a ªnormalº form by nondiseased individuals. This mutated variant is responsible for the inherited disease.

As of March 1998, about 46800 genes have been described in the human genome database (www.gdb.org), and 1325 genetic diseases have been mapped to chromosomal sites. Many human genes have been cloned and sequenced, including those that, when mutated, give rise to diseases relevant to neuroimaging, such as Huntington's disease (chromosome 4), Alzheimer's disease (chromosomes 1, 14, 21), neuroWbromatosis (chromosomes 17, 22), fragile X syndrome (chromosome X), and Lesch±Nyhan disease (chromosome X).

316 D. J. Vandenbergh

Lesch±Nyhan disease is a well-known example of a Mendelian disorder in which a mutated form of a single gene can cause a neuropsychiatric syndrome of motor and compulsive self-injurious symptoms (McKusick, 1990). At its simplest, a Mendelian disorder is one in which an individual can be deWned as either having, or not having, the disorder. The inheritance pattern is thus categorical ± one either inherits a mutated form (allele) of the gene or a nonmutated form. In the case of Lesch±Nyhan disease, the causative mutation is an inactive allele of the hypoxan- thine-guanine phosphoribosyl transferase (HPRT) gene, where, most frequently, one base pair of the gene is deleted (a point mutation). This single alteration causes the gene to encode a nonfunctional protein. The mutation is recessive, i.e., for the disease to be expressed, both inherited genes (one from each parent) need to be mutated; as a corollary, an individual that inherits one active allele from one parent and one inactive allele from the other parent is unaVected. In Lesch±Nyhan disease, the gene is located on the X chromosome. Therefore, females who carry only one aVected gene (carrier) do not express the disease but will pass the gene, and thus the disease, to 50% of their male oVspring. The mutation can also occur de novo during the gametogenesis. HPRT is a key enzyme in the nucleotide salvage pathway, and in the absence of active HPRT, uric acid accumulates to toxic levels. A result of this build up of uric acid is the development of multiple symptoms, including choreathetosis and self-destructive behavior. The relationship of the HPRT gene and brain function is not clear; however, it has been shown by neuroimaging that individuals with Lesch±Nyhan disease have reduced levels of the DOPA decarboxylase enzyme activity (Ernst et al., 1996) and the dopamine transporter (Wong et al., 1996).

The relationship of an inactive HPRT gene with decreased levels of DOPA decarboxylase and dopamine transporter, both detected by PET scan, raises an important question pertinent to genetics and neuroimaging. Is it possible that other partially active forms of HPRT, and perhaps other partially active forms of other genes, serve, in combination, to modify neurologic behaviors? Minor alteration of the levels of HPRT, DOPA decarboxylase, or the dopamine transporter may not lead to identiWable disorders, such as Lesch±Nyhan disease, but may still have some functional consequences in the brain. It is the detection and identiWcation of these types of gene, called ªsusceptibilityº or ªvulnerabilityº genes, that is the realm of complex genetics.

Before addressing complex genetics, the contribution of two other genetic phenomena, expansion of trinucleotide repeat DNA sequences and imprinting, to the characteris-

tics of the inheritance and course of a disorder will be mentioned brieXy. These events can complicate the tracking of inheritance of a trait and point to the dynamic nature of the genome, something not widely appreciated prior to molecular analysis.

Trinucleotide repeat expansion and anticipation

Trinucleotide repeat expansion mutations cause several inherited neurodegenerative diseases, including Huntington's disease and several forms of spinocerebellar ataxia, often with signiWcant neuropsychiatric symptoms. Expansion mutations have also been proposed to cause a number of other disorders, including bipolar aVective disorder, schizophrenia, and autism. For as yet unknown reasons, some repeated sequences of three nucleotides (e.g., CGG or CAG, where C is cytosine, G is guanosine, and A is adenosine) are unstable and can increase in length (expand) during DNA replication. Repeats of a given length increase in size when transmitted from parents to oVspring (intergenerational instability, ªmeiotic instabilityº) and often show size variation within the tissues of an aVected individual (somatic mosaicism, ªmitotic instabilityº). Repeat instability is clinically important, because longer repeats result in earlier age of onset and more severe disease phenotype. Therefore, a molecular explanation for anticipation (increasing disease severity in successive aVected generations) is given by the correlation of the length of a trinucleotide repeat to the severity of the phenotype.

Fragile X is an archetypal example of expansion of a trinucleotide repeat (de Vries et al., 1998). The mutation is caused by a repeat of the trinucleotide CGG in the promotor areas (DNA region that regulates the initiation and level of transcription of genes) of the FMR-1 gene located on the X chromosome. An allele with a relatively small number of the trinucleotides (50±200 triplets or 150±600 base pairs) is termed a premutation because individuals carrying this size repeat are normal. Individuals with the full mutation show an expansion of 200±1000 trinucleotide repeats. Alleles become expanded primarily through female gamete formation (see imprinting phenomenon below), but expansion can also occur postzygotically. In this latter process, the sex of the embryo also appears to inXuence expansion (greater in male than female embryos). The full mutation causes the range of physical and behavioral symptoms associated with the fragile X syndrome. Huntington's disease is another typical example of trinucleotide expansion. In this disorder, lower cognitive performance has been correlated with longer trinucleotide repeats in at-risk gene carriers (Jason et al., 1997).

Imprinting

Another important aspect of genetic transmission of disease is that the sex of the parent can diVerentially aVect the formation of repeats (imprinting phenomenon). Genomic imprinting is an epigenetic mechanism resulting in the preferential expression of the maternal or paternal alleles of a speciWc subset of genes. The basic mechanism of imprinting is methylation of DNA, in particular some cytosine residues when they are followed by a guanosine. Methylation is thought to regulate the ability of proteins in the nucleus of the cell to transcribe a gene in the general region of the methylated cytosine. Prader±Willi syndrome and Angelman syndrome are classical examples of the imprinting phenomenon (Feil and Kelsey, 1997). Patients with the former have neonatal hypotonia with failure to thrive, hyperphagia and severe obesity, hypogonadism, short stature, short hands and feet, mild mental retardation with learning disabilities, and obsessive-compulsive disorder. Symptomatology in Angelman syndrome includes ataxia, tremulousness, seizures, sleep disorder, hyperactivity, severe mental retardation with lack of speech, and a happy disposition with paroxysms of laughter. Both disorders are caused by the loss of function of imprinted genes (expressed from either maternal or paternal chromosome) in proximal 15q11-q13. However, deletions in Prader±Willi syndrome are of paternal origin whereas in Angelman syndrome they are of maternal origin. In approximately 2±4% of patients, a loss of function results from an imprinting defect. Molecular analysis of imprinting mutations that interfere with the appropriate establishment of the maternal and paternal epigenotypes has led to the identiWcation of imprinted transcripts that could be involved in determining which gene is imprinted in the germline.

Complex genetics: interaction of multiple genes alter brain function

Severe diseases that are attributable to a single gene, such as Lesch±Nyhan disease, are rare. More common are disorders that are heritable in part but do not display a one-to- one correspondence between the disorder's presence and a single gene. These disorders are referred to as complex. Complex disorders may not be as severe as Lesch±Nyhan disease but are present in a large enough proportion of the population that the aggregate eVect on society is large. In the case of complex disorders, several genes contribute to the disorder. Each of the altered genes alone is not suYcient or even necessary to cause the disorder, but the sum of

multiple small eVects from several altered genes leads to disease. Complex genetics are not limited to traits that we think of as diseases or disorders. Many traits demonstrate complex patterns of inheritance. Aspects of normal brain development are likely to be complex traits that depend on the action of multiple genes. For example, the volumes of speciWc brain nuclei can be inherited as complex traits. Complex traits can occur as a continuous (quantitative) characteristic, in which a normal distribution of values for the trait can be seen, or as a discontinuous (dichotomous) characteristic, in which a threshold value must be achieved to deWne the trait (for further description see Strachan and Read (1996) and Falconer (1989)).

A further level of complexity, in addition to the action of multiple genes, is the fact that a given constellation of altered genes (vulnerability or susceptibility genes) may only lead to a predisposition for the disease. This predisposition requires an interaction with environmental factors to produce the disease. Examples of disorders of this type that are actively investigated by neuroimaging methods are reading disorders (Grigorenko et al., 1997), attentiondeWcit hyperactivity disorder (ADHD) (Castellanos, 1997; Ernst et al., 1998), and substance abuse (Grant et al., 1996). Until recently these types of disorder were thought to be unapproachable by geneticists, because of, Wrst, the inability to Wt the disorders to traditional models of dominant or recessive inheritance, and, second, the recognition that a large number of genotypes would need to be examined.

Two major advances have occurred, or perhaps more accurately continue to occur, allowing the genetic analysis of complex disorders. The Wrst advance was the application of the polymerase chain reaction (PCR) (Saiki et al., 1988) to human genetics. The application of PCR to genetics allows a short region of DNA containing a polymorphism to be ampliWed. The detection of the form, or forms, of the polymorphism present in the ampliWed DNA leads to a genotype in a much more rapid manner than achieved with the older methods. Two types of polymorphism are commonly detected. First, sites in which one nucleotide has been replaced by another are known as restriction fragment length polymorphisms (RFLPs) (Fig. 18.1). RFLPs were originally detected by Southern blot procedures that were costly, especially in the amount of DNA used, and time consuming. The second common polymorphism is a variable number of repeated nucleotides, known as a microsatellite (Fig. 18.2). (The most common such repeat is the dinucleotide CA, but a repeat can consist of three or more nucleotides and it is not unusual for 50 or more nucleotides to be the repeat unit.) PCR methods have led to the rapid identiWcation of many regions of DNA containing polymorphisms scattered across the genome.

(a)

(b)

Fig. 18.2. Primers, shown as small arrows in (a) and (b), bind to two sites Xanking a polymorphism. The polymorphism is a diVerence in the number of dinucleotide repeats, CA in this case, such that the allele in (a) contains three copies of CA and that in (b) contains nine copies. Multiple copies of the DNA are synthesized by the polymerase chain reaction and the diVerence in size between these products can be determined by gel electrophoresis.

In addition to identifying many new polymorphisms, PCR methods also increased the rate of acquisition of an individual's genotype at each polymorphic site. These sites are called genetic markers and, with PCR, many markers can be tested in a short period of time. The identiWcation of candidate genes contributing to a complex trait requires the use of many genetic markers. These genetic markers may be anonymous pieces of DNA without a known function, but as long as their chromosomal location is known, it is possible to track the inheritance of the genetic marker and thus the DNA physically close to the marker on the chromosome. To date, about 19000 markers have been identiWed throughout the human genome (Web site www.gdb.org; sadly this database is no longer being maintained, and the latest data are from 9 March 1998). Additional modiWcations of PCR, such as multiple reactions in a single tube, are causing further increases in the rate of acquisition of genotypes to the point that questions that were unanswerable prior to 1987, when PCR was invented, are now feasible with the resources of one or a few laboratories.

The second advance was the recognition that the identiWcation of many genetic markers spaced closely along each chromosome would be more powerful for detection of the genes involved in complex diseases (Lander and Schork, 1994). The need for close spacing

between markers is akin to driving down a dark road trying to Wnd a barn by shining a Xashlight outside the side window. If the Xashlight is turned on once every mile, the chances of Wnding the barn are not great, but if the Xashlight is turned on every 20m the chances are much better. Additionally, it was recognized that the power to detect a gene's involvement in a disease would increase by analyzing two or more adjacent markers at a time, known as multipoint analysis. (If the genotypes are derived from a single chromosome they are called a haplotype.) Finally, new strategies were developed, such as the haplotype relative risk paradigm and the transmission disequilibrium test (TDT), which seek to maximize the information available by combining elements of traditional linkage (studying nuclear families) with elements of association tests (studying marker±phenotype relationship with, or without, requiring physical linkage). These methods are detailed below.

Study designs for complex traits in neuroimaging

Given that most traits measured in neuroimaging studies are genetically complex, how should one go about detecting the genes that underlie the trait? Psychiatric disorders are generally deWned by behavioral characteristics that may not directly reveal biological function. Neuroimaging data from brain disorders provide quantitative measurements (continuous variables), which is the type of phenotypic information that is amenable to analysis by modern genetic methods. Indeed, a qualitative phenotype for a disorder (either having or not having the disorder, dichotomous variable) may not adequately describe the variability found in the disorder. Results of genetic studies may eventually aid in deWning types and subtypes of disorder (nosology) (Leboyer et al., 1998), which in turn may aid in further reWning genetic searches in a bootstrap approach. A greater understanding of the genetic component of these disorders will enable diagnosis of disease and open new therapeutic approaches. Finally, understanding the genetic component will allow for more accurate deWnition of the environmental factors that interact with genes to cause disease.

The recognition of the diYculty in Wnding multiple gene variants that contribute to complex disorders by traditional methods (analysis of pedigrees) has led geneticists to use new approaches that are generally termed nonparametric. These approaches fall into two categories: analysis of small nuclear families, or parts of families (e.g., sibpairs), or analysis of cohorts of unrelated individuals (association studies). The approaches taken by these methods

are addressed brieXy to give a rudimentary understanding of how the methods can be applied to neuroimaging data. For further details, the reader is referred to reviews by Risch (Risch and Merikangas, 1996; Risch, 1997) and Rao (1998).

Sample size is a serious issue in genetic studies and often constitutes the limiting factor in the completion of genetic investigations. The number of individuals needed varies with the neuroimaging question being asked, and the genetic method that is appropriate to the question. Most importantly, the sample size is dependent on three factors related to the neuroimaging study. The Wrst critical factor is the error in measurement inherent in the neuroimaging technique. Because several genes may contribute to a small part of the data, the error in a neuroimaging measurement may be as large as, and mask, the eVect of any one gene. An example, explained below, is dyslexia. Imaging studies might identify neurologic aspects of dyslexia that are genetic, but the imprecision in neuroimaging results, and the diYculty of deWning a reading disability phenotype, will prevent easy detection of speciWc genes involved in the disorder. The second factor related to sample size is the fraction of the variance in measurement of the trait that is aVected by genes, or more simply its degree of heritability. Many traits may have signiWcant environmental components, making it diYcult to detect the genetic components. Most frequently, heritability in humans is measured in studies that compare the frequency of a trait in pairs of monozygous and dizygous twins. A pair of twins is said to be concordant if the trait is present in both. For traits that have a heritable component, the concordance rate is expected to be higher in monozygous twins, who share all of their genes, than in dizygous twins, who share only half of their genes. A second measurement of heritability is the risk to relatives, in which one compares the incidence of a trait in family members of selected individuals with the incidence in the general population. An increased incidence in relatives suggests (but does not prove) a heritable component. An example of a heritable trait relevant to neuroimaging is dyslexia, certain forms of which were proposed to be hereditary (Hallgren, 1950), with genetics accounting for approximately half of the variability (DeFries et al., 1987; LaBuda et al., 1993). Single word reading and phonological decoding and awareness are measures of dyslexia that are currently being examined in genetic studies (Grigorenko et al., 1997; see also Chapter 15) as more narrowly deWned traits, with the intent to minimize variability in phenotype classiWcation. This study by Grigorenko et al. required approximately 100 individuals to identify two chromosomal regions as important for aspects of dyslexia. Third, and Wnally, it is also necessary that any single gene's contribution to the trait be large

320D. J. Vandenbergh

enough to be detected in the presence of the eVect of other genes. With weak eVects from several genes, more individuals will be needed in the study. In general, DNA collected from a single study may not provide a large enough sample. Data from several sites will need to be combined in a metaanalysis to generate suYcient power to detect genetic inXuences on neuroimaging phenotypes (Rao, 1998). This strategy also demands that data are reliable and comparable across the various imaging centers.

The following three sections describe diVerent methods for detecting genes that contribute to a complex trait (Table 18.1). These methods do not rely on collecting the rare, large pedigrees typical of genetic studies of simple Mendelian genetic traits; instead, they focus on small nuclear families, siblings, or unrelated individuals.

Sib-pair designs

The genetic analysis of pairs of siblings examines the degree of similarity between aVected siblings for a quantitative phenotype and compares it with the number of alleles shared by the siblings at a speciWc marker (Haseman and Elston, 1972). It is not necessary to know the state of the parents with respect to the trait under study for this analysis. Computer programs are available for the analysis of such data. A recent computer program, the MAPMAKER/SIBS (Kruglyak and Lander, 1995) uses the genotype information for each sib-pair to estimate the ªidentical by descentº (IBD) status (see below) of each marker along the genome (parental information can be included to increase the power of the test). A major advantage of the sib-pair approach is that no prior assumption of the speciWc genetic model parameters is needed (nonparametric method). However, sib-pair analyses are not as powerful as family pedigree analyses and do not provide estimation of the strength of the linkage and the recombination fraction.

The analysis rests on the fact that, on average, siblings share one half of their genes (and genetic markers). At any single marker, the siblings may share 0, 1, or 2 alleles (for each of the pair of chromosomes that contains the marker), one being the average value. If each sibling inherited the same allele from one of their parents (the same ancestor), then the alleles are said to be IBD. Ideally, knowledge of the parents' genotype helps to assess the IBD status. For example, if each parent is heterozygous at a single genetic marker, and each of their alleles is diVerent (i.e., paternal alleles are A1 and A2 and maternal alleles are A3 and A4), and both of their oVspring have the A3 allele, then the oVspring must have inherited A3 from their mother. In the event that parental information is not avail-

able, the likelihood of IBD status can be estimated from known population frequencies of the alleles. The IBD status predicts the covariance between the sibs for the phenotype; consequently, the higher the IBD status the higher the covariance for a linked trait (Fulker et al., 1995). In other words, if the number of siblings sharing the alleles IBD is signiWcantly higher than expected by chance (50% of the time), then it is possible that a gene that accounts for variance in the trait is close to the marker being assessed.

Minimizing the number of sib-pairs may be possible by examining sib-pairs that are extremely dissimilar (discordant), and/or extremely similar (concordant) (Eaves and Meyer, 1994; Risch and Zhang, 1995). With these types of sampling from sib-pairs, traits for which heritability is in the range of 0.2 to 0.3 would require from 40 to 400 sibpairs, depending on allele frequencies in the entire population and other parameters (Risch and Zhang, 1996).

Overall, sib-pair analysis is the most powerful approach for detecting linkage of a region of a chromosome to a quantitative trait. Once one is conWdent that these results can be replicated, determining which gene in that region is responsible for the trait becomes the next big task.

Transmission disequilibrium test

The TDT (Spielman et al., 1993) is a recently developed test that was designed to take advantage of small nuclear families (two parents and an oVspring) as a way to avoid some of the pitfalls of detecting association (described below) and yet not require large pedigrees to detect linkage. The test requires that the oVspring be classiWed in a qualitative fashion (aVected/unaVected) but can be applied to quantitative traits by utilizing a cut-oV value of the measured traits. The best cut-oV point is one that is relevant to the biological underpinnings of the trait, but a relevant cut-oV may not be easily selected if little is known about the biology of the trait.

The TDT is based on the fact that a parent donates a single chromosome of each pair and, thus, one allele of any particular genetic marker. The donation of one of the two chromosomes in a pair is random; therefore, each chromosome, detected by genetic markers on the chromosome, should appear in oVspring 50% of the time (equilibrium). If transmission of a marker deviates from this expected frequency the marker is said to be in disequilibrium (hence the name of the test). The second chromosome of each pair is not transmitted, setting up a comparison of transmitted versus nontransmitted chromosomes by a chi-square test (see Table 18.2). For a marker to be informative, at least one parent must be heterozygous so that the two alleles can be distinguished. The advantage of this design is that both the

322D. J. Vandenbergh

Table 18.2. The transmission disequilibrium testa

Note:

a The 22 test is (b2 c)2/(b1 c) with one degree of freedom.

transmitted and the nontransmitted alleles are coming from a single group (the parents), avoiding the possibility that comparing unrelated individuals will Wnd allele frequency diVerences owing to ethnic background and not related to presence of disease. A further advantage is that the parents do not need to be phenotyped, which can result in signiWcant cost savings for expensive neuroimaging procedures.

A recent extension of the TDT has been proposed using sibships when parental data do not exist (Spielman and Ewens, 1998). This sib-TDT may be very useful in studies collected over long periods of time, and for those in which parents are deceased or unavailable. This method, rather than comparing the transmission of an allele from parent to aVected child, compares the genotypes of aVected and unaVected siblings. Spielman and Ewens (1998) show that these data may be combined with data from families analyzed by the original TDT to further increase the available sample.

TDT utilizes samples of multiple small families. The number of families, or sib-pairs in the case of sib-TDT, is dependent on the heritability of the trait, with increasing heritability needing smaller samples. Risch and Merikangas (1996) provide tables of predicted sample sizes using genetic risk ratios of a trait (deWned as the increased chance that an individual with a particular genotype has the trait) and the frequency of a disease allele in the population. For a genotypic risk ratio of 4, the number of families ranges from 48 to 235; a genotypic risk ratio of 2 requires 180±1970 families, and a genotypic risk ratio of 1.5 requires 484±7776 families. The range of families needed in the study is driven by the frequency of the disease allele (Risch and Merikangas, 1996). The TDT method works best with quantitative traits that have well-deWned cut-oV values that allow them to be analyzed as qualitative traits.

Population association design

The population association method is commonly used but has a serious caveat, which is the association test between

two groups of unrelated individuals. This method is also termed a case-control method, although this term should be reserved for studies that actually select a control individual for each case (disease) individual, based on minimizing any relevant diVerences between the two individuals. The association test's attractiveness is the ease with which the study can be conducted. The genotypes are identiWed for two groups of subjects, individuals with a disorder or trait and individuals without the disorder. The frequency of alleles is compared between the two groups by a chi-square test. It is possible to determine genotypes from each group at many sites along the chromosome and make repeated comparisons, although a statistical Bonferroni correction for multiple tests should be made. The primary caveat of this type of study is that a positive association does not imply any relationship between the gene and disorder. It is equally possible that individuals with the disorder, and those without, may actually be genetically diVerent subsets of a population; as a result, diVerences in allele frequency between the two groups would have nothing to do with the disorder but rather would reXect unanticipated ethnic diVerences (termed population stratiWcation).

Population-based studies of this type have lost favor because of the diYculty in replicating published Wndings, probably because of population stratiWcation. One example of an association that is frequently cited as being caused by stratiWcation is the relationship between novelty seeking and the dopamine receptor D4 (Benjamin et al., 1996; Ebstein et al., 1996; Malhotra et al., 1996; Vandenbergh et al., 1997b). True associations between a genetic marker and a trait can occur for only two reasons:

(i) the marker may actually have some functional consequence that directly aVects the phenotype being measured (functional allele), or (ii) if the marker does not alter function, then it must be extremely close to the polymorphic site of a gene that is the functional allele. These two sites on the chromosome, the marker and the functional polymorphism, must be so close that the two regions of DNA are always, or nearly always, co-inherited. This relationship of the co-inheritance of the two sites (loci) is known as linkage disequilibrium. There is very little information concerning the amount of linkage disequilibrium to expect between any two sites in large groups of people. It is clear that disequilibrium values will be high in regions of chromosomes that show high rates of recombination, or if two polymorphisms arose at evolutionarily ancient times allowing more time for recombination to occur between them. Recombination is the act of two sister chromosomes exchanging DNA (a normal part of meiosis) and results in segregation of markers that once were on a single chromo-

some onto separate chromosomes (Fig. 18.3). The linkage disequilibrium also depends on population history, with more mixing of populations decreasing its value (TishkoV, et al., 1996; Laan and Paabo, 1997; see also comments by Freimer et al., 1997). The lack of information concerning allele frequencies in ethnic groups makes the magnitude of the sample size diYcult to predict and, therefore, association studies are of little value to neuroimaging phenotypes.

One exception to this general limitation of association studies is the case of functional alleles of genes: alleles that alter protein function, or concentration, in some way, as opposed to alleles at anonymous sites that have no known eVect and are only markers. Functional alleles may allow for the inclusion of physiologic data to understand how a gene is related to the trait being studied. Several examples exist in the literature, such as alleles that alter the eVect of the enzyme catecholamine-O-methyltransferase (metabolizes dopamine) on measures of substance abuse (Vandenbergh et al., 1997a), or alleles of the cytochrome P450 gene (CYP2A6, which metabolizes nicotine) on measures of cigarette smoking (Pianezza et al., 1998). The example of measuring DOPA decarboxylase activity in individuals with Lesch±Nyhan disease was driven by the possible relationship of dopamine to self-injurious behavior (Ernst et al., 1996).

Whole genome scans to detect genes that alter quantitative traits

The study designs described involve testing many genetic markers in a ªwhole genome scanº. The results generated

deWne one, or more likely many, regions of the chromosomes in which diVerent genes reside that contribute to the measured trait. Each region is known as quantitative trait locus (QTL). Generally, the QTLs described in these studies are large enough that many genes exist in the QTL, and those that are not related to the trait must be winnowed out. At this point, several diVerent approaches may be necessary to narrow the selection process to Wnd the relevant gene. Further genetic studies that focus on reWning the position of the QTL may be necessary. Alternatively, searching for complementary DNA (cDNA) clones of the genes from the region, or genomic clones of the entire region, may be the next step before searching for allelic variants of the genes themselves. This process moves into a realm that combines molecular biology and genetics, and it is likely that this approach will grow and change rapidly in the next decade.

Genetic methods in neuroimaging studies: DNA collection

At its simplest, any genetic-neuroimaging study is the comparison of genotype status, determined from a research volunteer's DNA, with a particular phenotype, or trait, that may be derived from a neuroimaging study. In considering how to incorporate genetic analysis into a neuroimaging study, investigators need to plan very carefully the neuroimaging/genetic study design and not rely on the ease in collecting the DNA samples.

Indeed, collecting a sample of DNA from the research volunteers is a relatively easy addition to a study that