THE ART OF GENE HUNTING

Once it has been established from twin and/or adoption studies that a disorder has a high heritability (i.e., that genetic factors are likely to make a significant contribution to a person's risk of developing impairment), there are a number of ways of tracking down which specific genes may be involved (Botstein & Risch, 2003). Broadly speaking, the available strategies fall into two categories, “candidate-gene” approaches and “positional” methodologies, although with the recent sequencing of the human genome (Wolfsberg, McEntyre, & Schuler, 2001), the distinctions between these are beginning to blur. Candidate-gene studies rely on the existence of a prior hypothesis regarding the biological pathways underlying a trait. Researchers then select genes that are already known to be involved in the relevant processes and analyze these in subjects affected with the trait, in the hope that they will discover gene variants (alterations in gene sequence) that are associated with increased risk. The vast majority of studies of attention-deficit/hyperactivity disorder (ADHD) have exploited this kind of approach, basing the choice of candidate gene on well-established findings relating to efficacy of drug-based therapies. In particular, genes encoding proteins related to dopaminergic systems (most famously, DRD4 and DAT1) have been the subject of intense study, given that stimulant medication, primarily methylphenidate, can be effective for treating symptoms of ADHD (DiMaio, Grizenko, & Joober, 2003). However, for most other neurodevelopmental disorders virtually nothing is known about the potential molecular processes that are involved, making the selection of appropriate candidate genes problematic. The human genome comprises around 24,000–30,000 genes and a large proportion of these, perhaps more than half, are expressed (switched on) in the brain at some point during an organism's life; how does one then choose which gene to analyze? This is where the so-called positional approaches come into their own, because they have proved successful in pinpointing a risk gene without requiring any prior understanding of etiological pathways (Botstein & Risch, 2003).

At their most simple, positional strategies involve the analysis of inheritance patterns of chromosomal sections as they are passed from generation to generation in a “multiplex” family, that is, a family containing multiple individuals affected with a particular disorder. This kind of endeavor allows a geneticist to locate the chromosomal interval that is most likely to contain the gene that is influencing the disorder under study. Typically the region might span tens or perhaps hundreds of genes (even when pooling findings from many multiplex families), but this nevertheless represents a substantial reduction in the number of potential genes to be studied (compare to the number encoded by the entire genome). Researchers can then focus their efforts on this handful of genes and examine each one in affected subjects, looking for the specific mutation(s) that are contributing to the disorder. As mentioned, there is increasing overlap between positional and candidate gene approaches; it is now common for geneticists to localize a chromosomal region of interest in multiplex families and then prioritize candidate genes from this region based on clues to their likely function. For example, if investigating a chromosomal interval potentially involved in a neurodevelopmental trait, one might prioritize for study a gene previously implicated in neuronal migration or axon growth over one that is known to have a more general “housekeeping” role in many cell types. It is worth bearing in mind, however, that nature will often surprise us; sometimes the most unlikely candidates may turn out to be the true culprits.

At a practical level, how do we localize the risk genes involved in a particular trait or traits, in this case developmental disorders of speech and language? We first recruit families in which multiple individuals display speech and language impairment and take DNA samples from the family members. Then we establish the inheritance patterns of different chromosomal regions as they are passed on through each family, by analyzing the “genotypes” of each sample at a series of naturally occurring DNA variants, known as polymorphic markers. Most of these markers occur in the gaps between the coding parts of genes and can be thought of as signposts marking out particular chromosomal intervals. During transmission of genetic material from one generation to the next, corresponding sections of maternal and paternal chromosome copies are sometimes exchanged (genetic recombination). As a result, the genomic makeup of any individual represents a patchwork of maternal and paternal gene variants from previous generations (Pääbo, 2003). In a typical multigenerational family, genotyping of several hundred DNA markers allows us to reconstruct the inheritance patterns of virtually all chromosomal sections (this process is known as a genome-wide scan or screen). We can then test whether the transmission of any particular chromosomal region is linked to the occurrence of speech and language problems in the family. This is called “linkage mapping,” because the existence of recombination allows us to “map” potential risk genes to specific chromosomal locations. Simple statistical tests can be used to indicate whether linkages are likely to represent real findings or have instead arisen due to chance.

The success of carrying out traditional linkage mapping in a multigenerational family depends partly on the number of affected and unaffected individuals, but also on a series of assumptions about the genetic basis of a trait. For example, it assumes the involvement of only one risk gene and requires specification of a particular inheritance model (dominant, recessive, linked to the sex chromosomes, etc.). In general, the majority of neurodevelopmental disorders, including most types of speech and language impairment, violate these crucial assumptions (Fisher & DeFries, 2002; Fisher et al., 2003; Folstein & Rosen–Sheidley, 2001). Multiple risk genes could be implicated, modes of inheritance are often obscure, and different genes (with different inheritance patterns) may be involved in different families. This can be a problem, since it is often necessary to pool data from multiple families in order to provide enough power to detect linkage. Moreover, it is difficult to reliably map linkages due to occurrences of incomplete penetrance (individuals with increased genetic risk who remain unaffected) and phenocopies (individuals with low genetic risk who still develop language problems; Lander & Schork, 1994). This underlying genetic complexity, coupled with difficulties regarding the precise specification of the phenotype (i.e., diagnosis of disorder), has represented a major impediment to progress in mapping genes involved in common neurodevelopmental disorders. New strategies are emerging to handle genetic and phenotypic complexity, but these will not be discussed here. For more detailed accounts of genotyping, genome-wide screens, linkage mapping, and handling complex traits, see recent reviews describing the state of the art in genetic dissection of neurodevelopmental disorders (Fisher & DeFries, 2002; Fisher et al., 2003; Folstein & Rosen–Sheidley, 2001).

THE KE FAMILY: BUCKING THE TREND OF GENETIC COMPLEXITY

In the early 1990s an intriguing case emerged in the literature that appeared to contravene the common observation of genetic complexity in neurodevelopmental disorders (Gopnik & Crago, 1991; Hurst, Baraitser, Auger, Graham, & Norell, 1990). This was a large three-generation family, usually referred to as KE, in which 15 individuals suffer from developmental impairments in speech and language acquisition. Remarkably, the inheritance of these problems seemed to suggest involvement of a single copy of one mutated gene, transmitted from generation to generation of the family (i.e., a monogenic dominant pattern of inheritance). Consistent with this, around half of the individuals in the family had escaped from the disorder, presumably because they had not inherited the mutated copy of the putative risk gene. The large number of affected subjects in this family, coupled with the fact that the affected children had grown up in extremely similar environments to their unaffected siblings, represented persuasive (albeit indirect) evidence of a strong and simple genetic effect.

Given the unusual nature of this family, the characterization of their disorder rapidly became both an issue for scientific discussion (Alcock, Passingham, Watkins, & Vargha–Khadem, 2000; Gopnik & Crago, 1991; Ullman & Gopnik, 1999; Vargha–Khadem et al. 1998; Vargha–Khadem, Watkins, Alcock, Fletcher, & Passingham, 1995; Watkins, Dronkers, & Vargha–Khadem, 2002) and subject to a certain degree of media speculation (reviewed in Pinker, 2004). The ensuing debate generated extensive comments in various publications, most recently by Fisher et al. (2003) and Marcus and Fisher (2003). Therefore, I will limit myself here to a brief summary of the current consensus (where it exists) regarding the deficits in affected KE members and highlight a few key questions that remain unresolved. The problems inherited in the KE family appear to be restricted to the central nervous system (CNS), an issue I will return to later in the article. One of the most well documented aspects of the disorder is a developmental dyspraxia of the orofacial system; all affected members have difficulty in controlling the complex coordinated mouth movements that underlie speech articulation (Alcock et al., 2000; Vargha–Khadem et al., 1995, 1998; Watkins et al., 2002). Although in some cases the intelligibility of speech can improve over time in response to intensive speech therapy, oromotor deficits remain clearly detectable even in adulthood. As such, tests of orofacial praxis and/or repetition of pronounceable nonsense words (nonwords) represent the most robust methods for diagnosing disorder in the children and adults of the KE family (Watkins et al., 2002). Affected individuals perform normally on basic tests of limb praxis and when making single simple oral movements (Alcock et al., 2000; Watkins et al., 2002). Thus, it is likely that the difficulties reflect deficits in complex motor sequencing. However, it remains unclear whether these deficits are confined to the oromotor system or instead represent a more generalized impairment in motor control that impacts primarily on speech due to the particularly intense demands that human spoken language places on the speech articulators.

The articulation difficulties of affected members are accompanied by deficits in language-related skills, extending to both the expressive and receptive domains and affecting written as well as spoken language (Gopnik & Crago, 1991; Ullman & Gopnik, 1999; Vargha–Khadem et al., 1995; Watkins et al., 2002). For example, affected individuals have poor written verbal fluency and have problems spelling nonwords (Watkins et al., 2002). Grammatical abilities are also affected; the disorder impacts on comprehension of complex syntactical structure (evaluated via picture selection), and generation of word inflections or derivations in accordance with grammatical rules (Gopnik & Crago, 1991; Ullman & Gopnik, 1999; Watkins et al., 2002). It is worth emphasizing the point that the affected members of the KE family are still able to produce and comprehend language and can communicate with their peers to a certain extent; the disorder appears to involve weakening of language-related abilities across the board, rather than a total absence of any particular linguistic or grammatical skill (Watkins et al., 2002). A remaining area of controversy is the issue of whether the profile of expressive and receptive difficulties observed in the affected members may simply be a secondary result of impaired speech articulation in early development, itself due to a basic motor control deficit (Alcock et al., 2000). Alternatively, it could be hypothesized that the mutation in the KE family has multiple parallel effects on distinct aspects of brain development (this is a common phenomenon in developmental biology, which is known as pleiotropy). In fact, at the neurological and molecular levels, a range of different potential scenarios might account for the relationship between oromotor praxis and linguistic development in the disorder of the KE family (and indeed, in normal speech and language). This question may eventually be resolved through integration of data from different disciplines.

There has been some discussion regarding the nonverbal cognitive abilities of affected subjects in the KE family, fueling a misconception that the disorder is best viewed as a class of general mental retardation. It is indeed accurate to say that the mean performance IQ of affected KE members is significantly lower than that of unaffected subjects in the family (Vargha–Khadem et al., 1995), primarily due to poor performance on only one of the nonverbal subtests used to assess IQ (Watkins et al., 2002), but this is probably a red herring. In the KE family there are unaffected individuals with low nonverbal IQ and affected individuals with normal nonverbal IQ. It is unlikely that the skewed profile of speech and language difficulties observed in affected individuals can be explained by an overall domain-general reduction in cognitive ability. Research into common developmental disorders involving language has shown that the presence of early linguistic problems impacts on other areas later in life, including nonverbal abilities (Rutter & Mawhood, 1991; Tallal, Townsend, Curtiss, & Wulfeck, 1991).

FROM DISORDER TO GENE AND BACK AGAIN

Despite the complexities of the cognitive profile of the disorder segregating in the KE family, the inheritance pattern strongly suggested a simple genetic basis. Using traditional linkage mapping with DNA markers of the kind I described earlier in the article, we were able to formally prove this, and localize the putative defective gene to a small interval on one section of chromosome 7 (Fisher, Vargha–Khadem, Watkins, Monaco, & Pembrey, 1998). Due to the availability of human genomic sequence data in online databases, we could determine that there were around 70 genes in this genomic region (Lai et al., 2000), and so began to screen a subset of these in search of a mutation. These efforts were cut short by the discovery of a child (unrelated to the KE family) who had a similar diagnosis of severe verbal dyspraxia accompanied by substantial impairment of expressive and receptive language (Lai et al., 2000). In this case the disorder was associated with a gross chromosomal rearrangement (of a type known as a translocation) in which a section of chromosome 5 had been exchanged with a section of chromosome 7. Intriguingly, the point of breakage (the “breakpoint”) on chromosome 7 was localized to the same interval that had been implicated in the KE family, and so we hypothesized that disruption of the same gene was responsible in both cases. Precise mapping of the breakpoint in the child with the translocation indicated that it was directly interrupting a novel gene that had not, at the time of discovery, been fully characterized (Lai et al., 2001). A mutation search of this same gene in the KE family then revealed a single tiny alteration to its coding sequence, found in every affected individual. In all the unaffected family members (plus a large panel of unrelated control subjects), the coding sequence of the gene was normal (Lai et al., 2001).

How could a minor change in one letter of DNA within this novel gene be responsible for such a major alteration in brain function in the affected KE individuals? In order to answer this question we took the DNA sequence of the gene and predicted the resulting amino acid sequence of the protein that it was likely to encode (this is a standard code-cracking procedure; invariant rules describe how consecutive triplets of nucleotide letters get translated into strings of amino acids during protein assembly in the cell). We then compared the amino acid sequence to databases of current known proteins that had already been studied. This comparison identified a small stretch of sequence (called a “forkhead-box” or FOX domain) in the predicted protein, found in very similar form in a group of over 40 other human proteins (Carlsson & Mahlapuu, 2002). Mammalian forkhead proteins are clustered into related subgroups, depending on the degree of similarity between their forkhead domains, and each subgroup is arbitrarily assigned with a different letter of the alphabet (Kaestner, Knöchel, & Martinez, 2000). The protein encoded by the novel gene we had identified was given the official symbol FOXP2, because it was the second member to be discovered in the P subgroup of FOX proteins.

The presence of the forkhead domain gave the first pointers to both the function of the FOXP2 protein and the significance of the gene mutation in the KE family (Lai et al., 2001). The forkhead box is an 80–100 amino acid domain whose three-dimensional structure allows it to bind to DNA. Intensive studies of forkhead proteins in a variety of species from fungi to mice have reliably established that they act as transcription factors, helping to regulate the abundance of other proteins in the cell (Carlsson & Mahlapuu, 2002). They do this by binding to stretches of DNA found upstream of the coding parts of genes, and then are able to influence the amount of transcription of the gene in question. (During transcription the information encoded in the chromosomal DNA of a particular gene is converted into an intermediate messenger molecule, made of RNA. The message is further processed and then passed on to the protein building machinery of the cell where it acts as a template for making the encoded protein. The quantity of messenger molecules is a major factor determining the abundance of resulting protein, although there are other points of regulation in the system. Thus, regulation of transcription is a powerful tool for modifying protein levels.) A key aspect of forkhead protein function is that they do not act indiscriminately on all genes. Instead each different forkhead protein is able to regulate transcription of only a subset of genes, known as downstream targets. The specificity of any particular forkhead protein is partly determined by the structure of its forkhead-box domain, leading it to bind more strongly to some DNA sequences than others (Carlsson & Mahlapuu, 2002). The choice of targets influenced by a forkhead protein may also vary depending on the cellular environment (due to interactions between forkhead proteins and other regulatory cofactors) and could involve activation (increased transcription) or repression (reduced transcription) of gene expression.

Given their function in controlling profiles of gene expression in the cell, forkhead proteins (like other transcription factors) can have diverse roles in multiple biological pathways. Many of the known forkhead proteins are critical regulators of embryological development, involved in the sequence of ontogenetic events that turn a small bundle of undifferentiated cells into a complex and highly organized fetus with manifold specialized tissues and organs (Carlsson & Mahlapuu, 2002). As such, mutation of different forkhead genes has been linked to a range of developmental disorders in both humans and mice; for example, damage to the FOXC1 gene leads to glaucoma and other eye-related disorders, whereas disruption of FOXP3 causes a severe form of immune deficiency (Lehmann, Sowden, Carlsson, Jordan, & Bhattacharya, 2003). For dominantly inherited forkhead-related disorders, where there is one mutant copy and one normal copy of the forkhead gene in question, it is likely that abnormalities result from reduced levels of functional forkhead protein (Lehmann et al., 2003). This idea that forkhead dosage can be critical for normal development is supported by studies of gene disruption in mice. For example, mice have their own version of the FOXC1 gene (involved in eye disorders in humans), called Foxc1. If both copies of mouse Foxc1 are deleted, so that the mouse cannot make any Foxc1 protein at all, then this is lethal; the mouse dies early on in development (Kume et al., 1998). However, if only one copy is deleted, so that the mouse is able to make Foxc1 protein but at reduced levels to normal, then it survives but its eyes do not develop properly (Smith et al., 2000).

This brings us back to FOXP2 and its connection with developmental speech and language disorder. In the case of the child with the translocation, his chromosomal rearrangement directly disrupts one of his copies of FOXP2 (most likely rendering it totally inactive) while leaving the other intact (Lai et al., 2001). Therefore, his disorder appears to be associated with reduced dosage of functional FOXP2. In the KE family, there is a different type of gene disruption, a mutation of a single nucleotide, but again this affects only one copy of FOXP2 in affected subjects. The KE mutation is predicted to change an amino acid that occurs at a critical point in the forkhead domain of the FOXP2 protein, thus interfering with function. In fact, recent biochemical and functional studies of other forkhead proteins has strongly confirmed this prediction, showing that mutation at this position can disrupt cellular localization, DNA-binding activity, and transcriptional regulation of mutant protein (Saleem, Banerjee–Basu, Berry, Baxevanis, & Walter, 2003). In addition, more and more subjects are now being identified who have speech and language abnormalities associated with hemizygous deletions (i.e., deletions affecting only one chromosomal copy) involving FOXP2 (e.g., Liegeois et al., 2001) although other neighboring genes are also deleted. Based on the clinical data and knowledge of mechanisms of action of forkhead proteins we have hypothesized that reduced amounts of functional FOXP2 protein at a key stage of neurogenesis, and/or in a particular region of the CNS, leads to abnormalities in neural structures that are important for development of speech and language abilities (Lai et al., 2001). Thus, starting simply from the observation of a curious inheritance pattern in a multigenerational pedigree, and using a purely positional approach for gene hunting, it was eventually possible to identify the primary genetic cause, and also propose a plausible molecular mechanism to account for the disorder. However, identification of a mutated gene is not an end point, but rather a beginning, as I will demonstrate in the remainder of the article.

MORE THAN A NAME: FOLLOWING UP FOXP2

Discovering that disruption of FOXP2 is one cause of developmental speech and language disorder represents hard evidence that alterations to genes are able to influence the acquisition of verbal communication skills. As such, cognitive scientists may be tempted to insert the phrase “the FOXP2 gene” into their favorite hypotheses about the role of a gene or genes in human linguistic prowess, without necessarily modifying their original theories. As I outline below there are compelling reasons to be cautious about doing so. Regardless, FOXP2 is much more than an abstract name to hang a theory on; the gene provides a range of means to answer interesting questions about disease etiology, neurological mechanisms and evolution (Fisher, in press; Marcus & Fisher, 2003). Some of these questions have already been answered but others are currently the subject of intensive research in my laboratory and elsewhere.