Bates’s Emergentist Theory and Genotype/Phenotype Relations 237
Bates’s Emergentist Theory and its Relevance to Understanding Genotype/Phenotype Relations
Annette Karmiloff-Smith
Neurocognitive Development Unit, Institute of Child Health, London
Introduction
Elizabeth Bates (Liz, to us all) is a very dear friend and colleague, so I am sure that she will forgive me if I start the chapter that I have written in her honor with five quotations that will give her goose pimples and raise her blood pressure momentarily! My purpose of course is rhetorical, and I will use Bates’s Emergentist Theory to challenge the common theoretical thread that underlies each of these statements.
We argue that human reasoning is guided by a collection of innate domain-specific systems of knowledge. (Carey & Spelke, 1994)
The mind is likely to contain blueprints for grammatical rules... and a special set of genes that help wire it in place. (Pinker, 1994)
The genes of onegroup of children [Specific Language Impairment] impair their grammar while sparing their intelligence;the genes of another group of children [Williams syndrome] impair their intelligence whilesparing their grammar. (Pinker, 1999)
It is uncontroversial that the development [of Universal Grammar] is essentially guided by a biological, genetically determined program…Experience-dependent variation in biological structures or processes…is an exception…and is called ‘plasticity’. (Wexler, 1996)
The discovery of the [sic] gene implicated in speech and language is amongst the first fruits of the Human Genome Project for the cognitive sciences. (Pinker, 2001)
A naïve student reading the above literature would be forgiven if s/he were immediately to conclude that language and other domains of higher-level cognition are genetically determined, each ready to operate as soon as appropriate stimuli present themselves, with each functioning in isolation of the rest of the system, and that plasticity is not involved in normal development but merely acts as a response to injury. Our hypothetical student might well go on to infer that the main task for the scientist is to search for single genes that somehow map directly onto specific higher-level cognitive outcomes like grammar, face processing, number and the like. How has this position become so attractive to many linguists, philosophers and psychologists? Why does the popular press welcome this approach and jump at concepts like “a gene for language” or “a gene for spatial cognition”? What is the attraction of the notion that single genes code specifically for higher-level cognitive outcomes? Why are mouse models of human cognition, which often involve single gene knockouts and their relationship to phenotypic outcomes, not examined more critically? In the rest of this chapter, I will attempt to address these questions and show how a very different and more interactive theoretical approach—the Emergentist Theory embraced by Bates—offers a far deeper insight into the mind and brain of the developing child.
Mouse models of human cognition
Obviously, there can be no mouse models directly of human language. Apart from in cartoons, mice don’t communicate through a linguistically structured system like spoken or sign language. But there are many of other aspects of human cognition that have been modelled using knockout mice. Since later in the chapter I will discuss genetic syndromes and focus in particular on one neurodevelopmental disorder, Williams syndrome (WS), I will take as my examples in this section two mouse models of WS.
First, a word about the syndrome itself. This will be very brief since WS has now been extensively described in the philosophical, linguistic and psychological literature (see Donnai & Karmiloff-Smith, 2000, for full details). Suffice it to say that WS is a genetic disorder in which some 25 genes are deleted on one copy of chromosome 7, causing serious deficits in spatial cognition, number, planning and problem solving. Of particular interest to cognitive neuroscientists is the fact that two domains – language and face processing – show particular behavioral proficiency compared to the general levels of intelligence reached by this clinical group. Indeed, while IQ scores are in the 50s to 60s range, WS behavior on some language and face processing tasks gives rise to scores that fall in the normal range, whereas spatial cognition is always seriously impaired.
Twenty-five genes within the typical WS critical region have been identified, although the function of many of them is not yet known (Donnai & Karmiloff-Smith, 2000). In the middle of the deletion are the elastin (ELN) and Limkinase1 (LIMK1) genes that have been the focus of considerable research. ELN was initially hailed as the gene that caused the facial dysmorphology in WS (Frangaskakis et al., 1996), lose only one copy of ELN and yet have nothing of the WS elfin-like facial dysmorphology (Tassabehji et al., 1999). The only clear-cut genotype/phenotype correlation confirmed so far is between ELN and supravalvular aortic stenosis (a narrowing of the aorta in its trajectory to the heart), due to the role that ELN plays in connective tissue development. If ELN contributes in any way to the WS facial dysmorphology, it must be in interaction with other genes on chromosome 7 which are deleted in WS. More interesting, however, are genes that are expressed in the brain. Could having only one copy of any of these genes be insufficient for the normal development of certain aspects of cognition? Note that we all can have genetic mutations of various kinds without that seeming to affect outcome at all. Only some genes are dosage sensitive, necessitating two copies to function normally. Two of the WS deleted genes that are expressed in the brain have recently been singled out as being of potential interest to the cognitive outcome: LIMK1 which is contiguous with ELN, and CYLN2 which is at the telomeric end of the deletion. Let us look at each of these in turn.
LIMK1 is expressed in early embryonic brain development as well as postnatally. Its product is thought to contribute to synaptic regulation and the modification of dendritic spines. In an initial study (Frankaskakis et al., 1996), two families were identified in which some members had two genes deleted within the WS critical region on chromosome 7 – ELN and LIMK1 – whereas other members had no such deletion. The family members with the deletions were shown to have spatial impairments on tasks involving pattern construction and block building, whereas the others were unimpaired on these tasks. The authors concluded that they had “discovered that hemizygosity of LIMK1, a protein kinase gene expressed in the brain…leads to impaired visuospatial constructive cognition in WS” (Frangiskakis et al. 1996). However, a subsequent study by Tassabehji and collaborators (Tassabehji et al., 1999) examined four patients from three unrelated families who had deletions within the WS critical region, including ELN and LIMK1. They displayed no such spatial impairments. Indeed, unlike studies of patients with WS that consistently reveal a marked disparity between verbal and spatial scores, the patients tested in the Tassabehji et al. study displayed similar levels for both domains, with scores within the normal range. In fact, one of these patients, CS, was well above normal levels in all domains, despite in her case a very large deletion of some 20 of the genes in the WS critical region.
A mouse model created subsequently seems at first blush to challenge these findings. Meng and colleagues produced a LIMK1-knockout and showed that in the water maze the mouse had poor spatial learning, especially with respect to the flexibility required for reversal learning (Meng et al., 2002). This was hailed once again as showing that the deletion of LIMK1 in WS is the gene responsible for spatial learning. Note, though, that the knockout mouse was not only impaired spatially but also developed spinal cord deficits and a number of other problems. I will return to this point later.
A second mouse model of WS was created by Hoogenraad and collaborators, knocking out one copy of CYLN2 at the telomeric end of the WS critical region (Hoogenraad et al., 2002). CYLN2 is expressed in dendrites and cell bodies of the brain and its product is thought to contribute to the development of several areas such as the cerebellum, cerebrum, hippocampus and amygdala. The knockout mouse developed memory deficits, impairments of contextual fear conditioning and spatial deficits.
Our team subsequently identified a 3-year-old girl who has the largest deletion to date in the WS critical region without the fullblown WS phenotype. As well as deletions of one copy of ELN, LIMK1 and some 19 other genes, this patient also has CYLN2 deleted (one more gene than CS above) but not the final two telomeric genes in the WS critical region. The 3-year old has the typical short stature of children with WS, mild hypercalcaemia, mild stenoses, and a relatively mild dysmorphic face. However, her cognitive phenotype turns out to be different from the typical WS pattern, although this could of course change over developmental time. In this 3 year old, language scores were significantly worse than spatial scores. Indeed, unlike most children and adults with WS, she was in the normal range (lower end) on the pattern construction and block design sub-tests of the British Ability Scales-II. However, when we subsequently tested her in a human equivalent of the mouse maze (a pool filled with balls through which the child had walk in order to locate a gift box hidden in two different positions and searching from different starting points), our preliminary results pointed to significant impairments in this child, compared to a group of 3-year-old healthy controls as well as to individual controls matched on each of the other spatial tasks. (We have yet to test WS toddlers with the full WS deletion on this new task, but are about to embark on the comparison.)
What can we derive from these two mouse models compared the human tests done with various patients? First, there are particular advantages of mouse models of WS. All the genes involved in the WS critical deletion on chromosome 7 are present in the same albeit inverted order on mouse chromosome 5G. However, in my view, there remain a number of serious problems with mouse models in general, and with these two WS models in particular. First, identical genes in two species do not necessarily entail identical functions. This has to be demonstrated rather than simply assumed. Second, two species may not have identical timing of gene expression, even when particular genes are the same. Third, haploinsufficiency (one copy only) of genes can turn out to be lethal in one species, and give rise to no observable effects in the other, or one copy deleted in the human case may require both genes deleted in the mouse model to generate similar effects. Fourth, the models discussed above involve the deletion of a single gene in each case. Yet, WS is a multiple gene deletion. It is highly likely that the deleted genes interact with one another and, of course, with genes on the rest of the genome in very complex ways. One-to-one mapping between each single gene and each phenotypic outcome is most unlikely. Fifth, as hinted above, single gene knockouts do not only give rise to a single impairment, like spatial learning. As is to be expected given the complexities of gene expression, single knockouts frequently result in multiple impairments to various aspects of the resulting phenotype. And, whereas reports comparing mouse and human stress spatial impairments, little is made of the fact that other impairments, such as spinal cord deficits in the LIKK1 mouse knockout, do not occur in the human WS case. Sixth, in general single gene knockouts of a variety of different genes on the mouse genome frequently give rise to spatial impairments. This is probably due in part to the fact that the mouse repertoire tested is fairly limited, with spatial memory in mazes being one of the preferred measures. This leads to my final, and seventh point, i.e., that we often encounter direct generalizations from mouse to human that involve behaviors at very different levels. The demonstration of the existence of a mouse deficit in the water maze and a human deficit in block design is not, I submit, the same thing. The former involves navigational skills, i.e., the animal’s own body position and orientation needs to be represented in space. By contrast, in table-top tasks like block design, the position of the individual’s own body in space is not relevant to the task; it is a given. Rather, the featural and configural relationships between objects themselves need to be represented. Large-scale navigational and small-scale spatial representations may well call on different abilities. Relevant to this point is the fact that the 3–year old patient discussed above was significantly more impaired in our ball-pool version of the water maze in which she had to represent the position of her own body in space than on the tasks measuring table-top spatial construction.
In sum, mouse models of human genes, although often fascinating, must always be considered with caution when generalising to the human phenotype. Although mouse models of human language are not directly possible, there are many other ways to investigate the genetic contribution to speech and language. Clearly genes do play some role in language development. The question is what that role might be.
Genes and language
In 1998, and more specifically in 2001, the scientific world became very excited about the discovery of what came to be known as “the gene for speech and language” (Pinker, 2001). A British family, the now well-known KE family, had been identified, in whom an allelic variation in the FOXP2 gene in some family members gave rise to serious impairments in speech and language, whereas family members without this allele developed language normally (Vargha-Khadem et Al., 1998; Pinker, 2001). Is the FOXP2 allele novel to the human genome, and could it explain the onset of language in the human species? Does such a gene have a unique and specific effect only on speech and language in humans? Some researchers have claimed that this might indeed be the case (Pinker, 2001; but see more cautious discussion from evolutionary biologists in Enard et al., 2002). However, a closer look at the phenotypic outcome in the affected KE family members highlights the need to consider a far more complex story. First, the deficits are neither specific to language, nor even to speech output. The dysfunctions in the affected family members not only involve serious problems with oral-facial movements which impact on the development of language, but also affect particular aspects of the perception of rhythm as well as the production of rhythmic movements of the hands (Alcock, 1995; Varga-Khadem et al., 1998). Moreover, this gene cannot alone explain human language for several reasons. First, we must not lose sight of the general point that when a genetic mutation is found to cause dysfunction of a particular behavior, this does not mean that intactness of that same gene causes the proper functioning of the behavior. The effects of a single gene may represent a very minuscule contribution to the total functioning and yet be sufficient to disrupt it. Numerous analogies make this point obvious. For example, if the carburettor of a car is not functioning properly, the car will not run. But it is not the carburettor that alone explains how the car runs in normal circumstances; it is just a small part of an extremely complex machinery of interactions. But a second point is as important. It is highly unlikely that a single gene or even specific set of genes will explain the development of human language. In the vast majority of cases, genes involve many-to-many mappings, not one-to-one mappings. Furthermore, effects are not only in the direction of genes to phenotypic outcomes; the outcomes themselves can affect subsequent gene expression in return. Even in the case of single gene disorders (e.g., Fragile X syndrome), the phenotypic outcome displays multiple impairments, because the gene in question is deeply involved in synaptogenesis across the developing system (Scerif et al., 2003).