The Neuroscience of Intelligence
Page 9
Are the factors that influence intelligence consistent across the entire range? Here is an interesting example of progress in testing hypotheses about not just whether genes are involved in intelligence, but how genes are involved. An important question is whether the genetic basis of high intelligence and any influence of environmental factors on the salient genes are the same for the average and lower parts of the normal distribution of intelligence test scores. High intelligence might result from different genetic and environmental factors than those that influence average and lower intelligence. The latter view is called the discontinuity hypothesis. One discontinuity hypothesis is central to the view that expertise associated with high intelligence is more a reflection of practice and motivation effects stemming from experiences rather than from inherited ability. Another discontinuity hypothesis is that different genes are involved in high intelligence than the genes involved in average intelligence. By contrast, the continuity hypothesis tests the view that the same genetic and environmental factors are at work throughout the intelligence distribution. The effects of each factor are additive so high intelligence reflects having more of the relevant genes and experiences.
In lieu of having specific intelligence genes to compare between a high IQ group and an average group, twin studies can test these competing hypotheses by comparing the groups for the proportions of genetic, shared, and non-shared variance. Simply put, the Discontinuity Hypothesis predicts the three components of variance would differ between high and average intelligence groups. A strong test of this was based on 9,000 twin pairs and 360,000 siblings sampled from 3 million 18-year-old males conscripted into military service in Sweden (Shakeshaft et al., 2015). All had completed a battery of cognitive tests from which a g-factor was extracted and the top 5% comprised the high-intelligence group (IQ estimated at greater than 125).
Several analyses were reported that were consistent in showing strong support for the Continuity Hypothesis and virtually no support for either the environmental or genetic Discontinuity Hypothesis. The authors concluded, “Stated more provocatively, high intelligence as we defined it appears to be nothing more than the quantitative extreme of the same genetic factors responsible for normal variation” (p. 130). They also cautioned that they did not have sufficient statistical power to determine whether the groups might differ if the high-intelligence group was defined more extremely, say at the upper .025% suggested for defining genius by Galton (1869), rather than the upper 5%.
Another twin study tested the important issue of whether genes account mostly for the g-factor or for specific cognitive domains that comprise g (Panizzon et al., 2014). These investigators used a large sample of middle-aged veterans (average age 55) from the Vietnam era in a longitudinal study of aging and found 346 pairs of identical twins and 265 fraternal twin pairs. Everyone had completed a battery of 10 cognitive tests representing four basic and well-established cognitive domains (verbal ability, working memory, visual–spatial reasoning, processing speed). Several alternative factor-analysis models of the relationship among tests and domains that did not place g at the peak of a hierarchy (see Chapter 1) were compared on estimates of variance accounted for by genes, shared environment, and unique environment. The model that best fit the data indicated that g in the hierarchical model was more heritable (86%) and accounted for more of the genetic effects in the specific domains than any other model. Although the researchers acknowledge some limitations in the study design, by directly testing alternative models, the results extend and strongly support the earlier research on g as the key common heritable factor underlying different mental abilities.
Additional progress is illustrated in elaborate new twin studies that not only have very large sample sizes but also combine DNA assessments and neuroimaging. These studies blend quantitative and molecular genetics and we will review them in Chapter 4 after the next chapter introduces neuroimaging. Before that, however, we continue here with some early studies of molecular genetics and the hunt for specific genes.
Textbox 2.1: Social class and intelligence
A widely cited study suggested the heritability of intelligence is stronger in families with high social–economic status (SES) and weaker in families with low SES (Turkheimer et al., 2003), but not all studies agree (Asbury et al., 2005; van der Sluis et al., 2008). Generally, SES is confounded with intelligence. On average people with high intelligence get higher-paying jobs and have more money to provide resources for children. They attain a higher SES directly or indirectly due in part to intelligence along with other factors (including luck). To the extent that intelligence is passed along by genes, SES effects independent of genetic influence on intelligence are difficult to assess. This underscores one difficulty in assessing gene–environment interactions. A recent meta-analysis of SES and heritability studies of intelligence suggests a more complex interaction between SES and intelligence (Bates et al., 2013) and there is some evidence that SES, education, and general intelligence have genes in common (Marioni et al., 2014).
In this context, a fascinating study of social class in Poland during its socialist years addressed this issue in an unusual way. This is an older study but quite illustrative (Firkowska et al., 1978). Here is the summary quoted directly from the published report: “The city of Warsaw was razed at the end of World War II and rebuilt under a socialist government whose policy was to allocate dwellings, schools, and health facilities without regard to social class. Of the 14,238 children born in 1963 and living in Warsaw, 96 percent were given the Raven’s Progressive Matrices Test and an arithmetic and a vocabulary test in March to June of 1974. Information was collected on the families of the children, and on characteristics of schools and city districts. Parental occupation and education were used to form a family factor, and the district data were collapsed into two factors, one relating to social marginality, and the other to distance from city center. Analysis showed that the initial assumption of even distribution of family, school, and district attributes was reasonable. Mental performance was unrelated either to school or district factors. It was related to parental occupation and education in a strong and regular gradient. It is concluded that an egalitarian social policy executed over a generation failed to override the association of social and family factors with cognitive development that is characteristic of more traditional industrial societies.” In the context of this chapter, the confounding of genetic and SES factors leads to a possible alternative conclusion: Any influence of social policy on mental performance failed to override the influence of genetic factors. The same confounding is apparent in new studies which suggest that SES accounts for brain differences underlying cognitive/achievement gaps, and we will detail them in Chapter 6.
2.5 Molecular Genetics and the Hunt for Intelligence Genes
Technological advances in measurement drive scientific progress. Until DNA technology developed to a point where the double helix could be chopped into precise fragments using cost-effective methods and the millions of pieces (base pairs among the smallest units) could be characterized statistically, the hunt for human intelligence genes could not advance in earnest. For decades, breeding experiments in mice had produced tantalizing evidence that learning how to navigate a maze to find cheese had a genetic basis. Some mice learned faster than others, and when the “smart” mice were bred with other smart mice, the offspring learned the maze faster. In 1999, genetic engineering was used for the first time to create smart mice that could learn a maze more quickly (Tang et al., 1999). The researchers named the strain of these mice “Doogie,” after a TV character that was a precocious teenager in medical school. This achievement (the mice, not the TV show) was based on considerable previous animal work that showed a certain synaptic receptor, NMDA (N-methyl d-aspartate), was involved in learning and memory. A single gene (NR2B) was found to regulate a part of this receptor’s function. The researchers spliced this gene into the DNA of ordinary mice embryos. The resulting Doogie strain of mice learned a series of t
asks faster than controls.
All neurotransmitters and receptors work in complex balances. In the world of synapses, too much or too little of any component can have deleterious or fatal consequences, so applying animal research findings to humans requires considerable patience and caution. Whether genetic manipulation of the NMDA receptor in humans might produce similar learning and memory enhancements, without serious side effects, is not yet known. This example illustrates that finding a gene related to something like learning or memory or intelligence is just a first step, albeit a challenging one even in animals. Determining how and why the gene functions, within a cascade of neurobiological steps and interactions, is even more difficult. Manipulating genetic effects to produce a desired outcome is not for the faint of heart or for impulsive personalities or short-term investors. Nonetheless, Doogie mice are a tantalizing example of the powerful potential for changing the determinism assumed when something has a strong genetic basis.
Separately from animal learning and memory research, the search for human intelligence genes began with a simple research strategy. DNA samples were collected from groups defined by IQ scores. Each participant’s DNA was fragmented into small pieces where genes could be identified. Textbox 2.2 describes key terms and methods used in DNA studies. These fragments from high- and low- (or average) IQ groups were compared and differences were noted as candidates for genes related to intelligence. This is a needle-in-the-haystack strategy because the number of fragments and individual genes or base pairs was in the many millions, the cost per individual was quite high, and the IQ groups also differed on many difficult-to-control characteristics in addition to IQ. Nonetheless, researchers were optimistic that specific intelligence genes would be discovered, especially as new DNA assessment technologies were developed. In fact, many candidate genes were identified using various quantification techniques of increasing sophistication.
Despite the daunting challenges of this search, a number of research groups around the world are using variations of this strategy to identify specific genes for intelligence. One interesting approach in a Japanese study used multiple DNA assessment techniques in a sample of 33 identical twin pairs who are discordant for IQ scores (Yu et al., 2012). That is, the twins within a pair had at least a 15-point difference in scores (one standard deviation). Using discordant identical twins (reared together), even a small sample, minimized irrelevant genetic and environmental factors and maximized the chances of finding salient differences in gene expression related to intelligence even if differences might be the result of epigenetic influences. The use of multiple methods of DNA analysis allowed for independent replications within the same sample. The outcome identified several possible differences in gene expression that suggest brain mechanisms that might be related to intelligence. The findings illustrate the complexity of gene expression and regulation that are far beyond our discussion here, but they demonstrate that finding genes is only the first step toward understanding exactly what the genes do and how they influence or regulate neurobiology and brain function on the molecular level.
Even at the early stage of the search for intelligence genes, the emerging data were consistent in two important ways. First, none of the candidate genes accounted for much variance in intelligence test scores. This was a disappointment to those who had hypothesized that a few genes would account for considerable if not most variance given the high heritability estimates for intelligence from twin studies. Other researchers recognized this reflected the complexity of the cognitive processes involved in intelligence and was consistent with Plomin’s prediction about numerous generalist genes, each accounting for a tiny portion of the variance in intelligence. This view has growing empirical support (Trzaskowski et al., 2013b). The second consistency, and far more distressing, was that none of the candidate genes identified in this early phase could be replicated in independent samples. Independent replication is a cornerstone and absolute requirement of scientific progress. Independent ideally means both a different investigator and a different sample. In this early phase there was considerable competition to find “the” genes and journals did not require independent replication for publication, even by the same investigators but in a separate sample.
There is no reason to list all the early candidate genes here along with the subsequent failures to replicate them. This disappointing state of affairs did not change for about two decades despite advances in the precision and cost-effectiveness of DNA technology and genomic information statistical analysis. One key problem was that most sample sizes were small and lacked statistical power to replicate them any tiny effects a gene might have on intelligence. As late as 2012, Chabris and colleagues summarized the search for intelligence genes in a comprehensive research paper titled: “Most Reported Genetic Associations with General Intelligence are Probably False Positives” (i.e., they are wrong). This group attempted to replicate 12 candidate genes for intelligence from published studies. They had access to three independent samples totaling over 6,000 people who had completed DNA analyses and intelligence testing. Their analysis was decidedly negative. None of the 12 candidate genes was associated with intelligence in a robust statistical manner. This failure to replicate intelligence genes, in a large sample with sufficient statistical power to find small effects if effects existed, was offset by successful replication in the same sample for control candidate genes related to Alzheimer’s disease and to body mass. The authors did not express discouragement at their failure to replicate genes for intelligence and concluded that even larger samples might be needed to find and replicate multiple genes if each one accounted for even tinier amounts of intelligence variance. They encouraged intelligence gene hunters to take part in multicenter consortia that could generate sample sizes of thousands of people (Chabris et al., 2012).
Textbox 2.2: Basic genetic concepts (also see Glossary)
The technologies and methods of DNA analyses used in molecular genetics are diverse, complex, and evolving at a rapid pace (Mardis, 2008). The basic thrust of progress is that the costs decrease and the precision and scope of analysis increases. There are now many studies of intelligence using DNA analyses. Here are some key terms used in the representative studies summarized in this chapter. A gene is the unit of inheritance. There are an estimated 19,000–22,000 genes, most still to be identified, distributed on the two sets of 23 chromosomes (one from each parent) in the human genome. Genomics is the term used to describe mapping genomes using many different methods. Every person has a unique genome, although most of the gene sequence is the same for everyone. Chromosomes are made of two strands of DNA molecules, the so-called double helix. During reproduction, the offspring randomly inherits these strands from each parent. The strands all are constructed from a combination of only four base molecules, (A)denine, (G)uanine, (C)ytosine, (T)hymine. These four arrange themselves in pairs across the two strands of DNA in the double helix like rungs on a ladder. Each member of a pair is inherited from one parent. On each rung, A and T, and G and C form pairs.
There are an estimated 3 billion of these “base pairs” (also called nucleotides) in the human genome. The order of these base pairs on a strand of DNA is the genetic code. All humans have nearly identical genetic codes and all the differences among individuals result from a relatively small portion of genetic variations. Genes create amino acids that form thousands of different proteins, and proteins are the building blocks of life that determine how an organism develops and functions at the cellular level. The sequence from amino acid creation to protein formation is called gene expression. RNA is similar to DNA, but RNA essentially translates the DNA code into amino acids and proteins. Genes can be active or inactive. They turn on and off during development and over the life span. The expression of a gene is regulated in part by methylation, one of several neurobiological mechanisms that can be influenced by non-genetic factors, including diet, illness, and stress (Jaenisch & Bird, 2003). Methylation is a process involved in many aspects of normal
and abnormal cell development. Of particular interest is that it can change the molecular structure of A and C in base pairs. These changes alter the expression of some human genes (Wagner et al., 2014), and importantly, the modified genes potentially can be inherited. Epigenetics studies how gene expression is modified by non-genetic factors.
DNA sequencing identifies the exact physical order of all the base pairs. Genes are contiguous segments of base pairs, although it is not always clear where one gene ends and another begins. Genes can have different forms as a result of getting half of the base pair from each parent. The form of a gene is called an allele. For example, a hypothetical gene for wrinkles on the chin may be expressed as WW and no wrinkles as ww. Each parent contributes either a W or a w, so the allele can be WW, Ww, wW, or ww. The inherited pair will determine whether the offspring has wrinkles or not.