Intelligence and Brain Myelination: A Hypothesis

Personality and Individual Differences, Vol 17, (December 1994) No. 6, 803-833.

• INTRODUCTION 1
• BRAIN ENERGY USE 2
• EFFECTS OF THICKER MYELIN ON BRAIN SIZE 2
• MAGNETIC RESONANCE IMAGING STUDIES 4
• SPEED-RELATED EVIDENCE 6
• Measures of variability in brain processes 11
• Skewness 12
• Reaction time variability increases rapidly with complexity 15
• Other reaction time considerations 16
• AGING AND MYELIN 17
• EVIDENCE FROM INBRED MICE STRAINS 22
• OTHER EVIDENCE 25
• BRAIN LIPIDS AND EVOLUTION 26
• Other Lipid and Intelligence Correlations 28
• Evidence from Premature Infants 29
• Other lipid evidence 31
• IMPLICATIONS YET TO BE TESTED 31
• CONCLUSIONS 34
• REFERENCES 34

Summary-Many observations concerning intelligence could be explained if much variance in intelligence reflects myelination differences. More intelligent brains show faster nerve conduction, less glucose utilization in positron emission tomography, faster reaction times, faster inspection times, faster speeds in general, greater circumference and volume, smaller standard deviation in reaction times, greater variability in EEG measures, shorter white matter T2 relaxation times, and higher gray-white matter contrast with magnetic resonance imaging. Also explainable are peculiarities of the increased reaction times and standard deviations with number of choices and complexity, reaction time skewness, the shorter latencies in evoked potentials, shorter latencies to the P300 wave, the high glial to neuron ratio in Einstein's brain, less glucose utilization per unit volume in large brains, certain results related to lipids, essential fatty acids, and cholesterol in adults and premature babies, and the survival of genes for lower intelligence. Children's improved performance with maturation might result from myelination. The slowing of response times with age, the decline in intelligence, and increased T1 relaxation times could be explained. Differential myelination in the mouse brain might be able to explain the heterosis observed for myelination, brain size, caudal nerve conduction velocity, and maze performance observed.
Keywords: intelligence, myelin, conduction velocity, glucose utilization, magnetic resonance imaging, reaction times, inspection time, fatty acids.

INTRODUCTION

Various researchers have reported the following differences between the brains of the highly intelligent and the less intelligent.

1. Intelligence is subject to considerable genetic influence.
2. Large brains are more intelligent.
3. The brain size versus intelligence relationship is pleiotropic.
4. More intelligent brains use less energy.
5. Large brains use less energy per unit volume.
6. Brains with faster nerve conduction are more intelligent.
7. The more intelligent have shorter white matter T2 relaxation times.
8. White grey matter MRI contrast decreases with age.
9. Responsetimes decline with age.
10. Intellectual performance declines withage.
11. Faster reaction times correlate with higher intelligence.
12. More variable reaction times correlate with lower intelligence.
13. Reaction time distributions are positively skewed.
14. The variability of choice reaction times increases more rapidly with complexitythan the median reaction times.
15. In the elderly, variability increases more rapidly with complexity than in young adults.
16. The genes for low intelligence have survived, even though intelligence appears to improve performance in virtually all mental functions.
17. In the mouse, heterosis for brain myelination, brain size, caudal (tail) nerve conduction velocity, and various aspects of behaviorexists.

What mechanism could make all these statements true? The high heritability for intelligence (recent references are provided by Hewitt & Last, 1984; Plomin & Loehlin, 1989; McCartny, Harris, & Bernieri, 1990; Bouchard, Lykken, McGue, Sega, & Tellegen, 1990; Bouchard, 1993; and Plomin, Chipuer, & Neiderhiser, 1994) is an important clue, since we know that genes code for biological differences. The inverse relation between size and energy use implies that much of the added matter in large brains uses little energy. Thus, it is probably not merely additional neurons. This added matter probably has a purpose (from evolutionary theory). Since conduction is faster in smart brains, the effect of this substance could be to speed up nerve conduction. The obvious candidate is myelin, since neurons sheathed with myelin have faster nerve conduction rates than those without myelin, and neurons with more myelin are also faster. Various features of the reaction time experiments could be explained by a process in which signals from one neuron accidentally cause signals in adjacent neurons, and these errors in transmission are more common where myelin was thinner.

The remainder of the paper will document the assertions made and explain how they could be explained by differences in myelination

BRAIN ENERGY USE

The brains of the more intelligent actually used less energy than brains of the less intelligent (Haier et al. 1988, also see Haier et al, 1992, and Haier, 1993). The problem is to explain how brains that actually use less energy function better. One possibility is that they are faster and less error prone because of more myelination. Because the myelin is chemically inactive (serving roughly the same function as the insulator on a cable), it uses very little glucose. Most energy is used in the movement of ions in and out of axons. Thus, lower energy use in the more intelligent could be merely the result of more relatively inert myelin. Alternatively, more intelligent brains are more "efficient" and somehow activate fewer neurons for any given problem. A correlation of the glucose consumption of eight individuals while doing the Raven's matrices (a non-verbal test that is considered a good measure of intelligence) with their performance on the test showed a statistically significant inverse relationship (Haier et al., 1988, see also Haier, 1993). Also, 10 mentally retarded showed higher glucose use than 10 age and sex matched controls of normal intelligence (Haier, et al. 1993). While the sample size was extremely small (but large enough to give a statistically significant result), other studies, using very small numbers of subjects, are consistent with this result (see Haier et al., 1992, for citations). For instance, Parks et al. (1988) found that the brains of subjects with higher verbal fluency used less energy. These results are all for non-diseased individuals. Studies including diseased individuals often show a positive correlation between energy use and intelligence, which is usually attributed to cell death or damage reducing both energy use and intelligence.

Diamond (1988) found that rats raised in an enriched environment (which in other studies has been shown to improve maze learning abilities) had lower glucose utilization in certain regions. These were the same regions which had enlarged with enrichment in other experiments. A major problem raised by the Haier et al. results is explaining why any brain would use other than the least energy intensive system.

EFFECTS OF THICKER MYELIN ON BRAIN SIZE

If neurons in some people have thicker layers of myelin than those of other people, this should increase the brain size of those with thicker myelin (all other things equal). This should produce a positive correlation between brain size and intelligence. Such a positive correlation has been found in many studies using the modern technology of magnetic resonance imaging (Willerman, Schultz, Rutledge, and Bigler, 1991; Andreason, et al. 1993; Raz, Torres, Spencer, Millman, & Sarpel, 1993; Haier et. al. 1993, Wickett, Vernon, & Lee, in press, and the two unpublished studies of Schultz and associates cited in Schultz 1993).The report by Andreason et al (1993, table 2) does contain one observation that may oppose thicker myelin leading to higher intelligence. Intelligence had a statistically significant correlation with the volume of grey matter in the brain. In contrast, the correlation with the volume of white matter, while positive, was not statistically significant. Thicker layers of myelin might be expected to increase the volume of white matter (if the number and diameter of neurons were held constant). However, a higher proportion of myelinated fibers might increase performance and myelin content while leaving volume unaffected (a possibility suggested by Schultz in a personal communication). In additon to the MRI studies, there are numerous studies using numerous studies using external head dimensions. Wickett, Vernon, & Lee, (in press) lists 39 samples from 24 research teams working over eight decades, all of which show a positive correlation between head size and a measure of intelligence. Lynn (1990 a, b, 1991a), Rushton (1990), Johnson (1991, 1991a, 1991b), and Jensen and Sinha (1993) have provided recent reviews and additional evidence.A peculiarity of some of the brain size data (at least that from Wickett et al. in press, and Andreason, et al. 1993) is that the correlations of performance and verbal tests with hemispheric size does not vary with the hemisphere, even though there is strong neurological evidence that the functions are localized in one hemisphere. The most plausible explanation is that whatever is causing the part of the variations in hemispheric volume that correlates with intelligence affects both hemispheres equally. Myelination could explain this if the genes (or environmental factors) that affect myelination act on the neurons throughout the brain. Then, a left hemisphere size measurement would be a good measure of the myelination of the right hemisphere, where much of the spatial performance processing is believed to occur, and likewise for right hemisphere size indicating myelinization on the left where verbal processing is believed to occur.Of course, a brain wide factor that affected growth in number of neurons could also produce this effect, but it seems more likely that the growth of each individual structure is influenced by separate genes. There is the Willerman, Schultz, Rutledge, & Bigler (1992) report that hemispheric volume differences predicts relative performance on verbal and performance tests. This effect might arise if the number of neurons was causing the part of the volume difference that was being picked up by relative volume. Since both brain size and intelligence exhibit high heritability, a natural question is whether the same genes produce both effects. Jensen & Johnson (1994), using data from 14,000 children from the Collaborative Perinatal Project, have shown that there is a correlation between head circumference (here a surrogate for brain size) in both 4 year olds and 7 year olds, although the correlation is much larger for 7 year olds. At age 7 there is a highly significant within family effect, such that the sibling with the larger head circumference (after controlling for age, height, and weight) has the higher IQ. This effect was found in the white male, white female, black male, and black female samples. Such a correlation makes it very likely that the observed relationship between head size (used as a surrogate for brain size) is due to one gene affecting both. Interestingly, the effect was not found at age 4. While it is always possible that a variable such as nutrition or exposure to disease is affecting both brain size and intelligence, this seems unlikely. Siblings are likely to have similar nutrition and disease exposure. Also, these variables would likely have affected brain size through an effect on body size, and since the weight and height were controlled for, any effect of disease or nutrition has probably been controlled for. Finally, disease and nutrition affect children before age 4, and the within family correlation was not found in the age 4 data. Traditionally, brain size intelligence relationships have been interpreted with the very plausible idea that brain size was a surrogate for the number of neurons. However, it is known that the number of neurons is determined at an early age. Thus, if there are large genetically produced differences in numbers of neurons that affect intelligence, one would expect a within-family correlation at age 4. However, Jensen and Johnson failed to find such an effect. This makes it unlikely that the brain size IQ correlation is really due to genetically caused differences in the number of neurons. An alternative is that genes somehow affect the size of some elements in the brain, and that this effect first appears between age 4 and age 7. Dendrite or axon size or lengths could differ. Possibly the differences in dendrite lengths that Jacobs, Schall, & Scheibel (1993) recently reported first emerges during that period. However, variation in the thickness of the myelin layers is a possible mechanism. Myelination of the cerebral cortex is still continuing after age four (Gibson, 1991, Table 4). "The corpus callosum . . . is not dark until middle childhood in humans. Long association fibers such as the cingulum, and the superior and inferior longitudinal fascicui are extremely protracted in their myelination. They are not dark until 3.5 years of age in the monkey and until the second decade of life in the human" (Gibson, 1991, 39-40). Dark here refers to the color observed in tissue slides after staining of the myelin. It is plausible that the genes that control the myelination of these tissues in middle childhood differ between individuals. If they do, it could explain the within family correlation that is found at age 7, but which has not emerged by age 4. It could also explain why the between families correlations are much stronger at age 7 than at age 4, since families also differ in the genes for myelination between ages 4 and 7, and their increased role could explain the increase in correlation.There is another peculiarity in the data which is explained by the myelination theory, but not as easily by there being more neurons. This is the observation (Hatazawa, Brooks, Di Chiro, & Bacharach, 1987; Yoshii, et al. 1988) that energy use (per cubic centimeter as measured by positron emission tomography) declines with brain size. The decline with brain size in energy use per unit volume was so great that the authors regarded it as a defensible hypothesis that the brainís total energy use was independent of brain size. This is what would be expected if the largest brains were more myelinated. If larger brains typically contained more neurons, one would expect them to use more energy (assuming energy use per neuron remains constant as the brain enlarges).

MAgNETIC RESONANCE IMAGING STUDIES

The strongest evidence that myelination improves mental performance is provided by the Schultz thesis (Schultz, 1991; Willerman, Schultz, Rutledge, and Bigler, 1992; 1994) which showed that grey-white matter contrast is greatest in the most intelligent. Further work showed that the difference in contrast was in the white matter, and due to T2 relaxation times being shorter in the more intelligent. The T2 relaxation time measures the rate at which the phase coherence in the signal arising from traverse magnetization is degraded by interactions with adjacent nuclei, with a shorter period indicating stronger interactions. The differences in the relaxation times are usually interpreted as varying with the amount of bound water, with protons in bound water being less free to return immediately to their original orientations after disturbance by a fluctuating magnetic field. It is often further interpreted as being related to the number of biological membranes in the immediate vicinity of the affected protons. Myelin consists of a series of concentric membrane layers around the neuron. The bound water is probably between the membrane layers in the myelin.

Schultz interprets the T2 differences as reflecting myelination differences. One of his strongest pieces of evidence is that the contrast in infants is very low and increases with age, roughly parallel to when the different structures are known to become myelinated from autopsy data. It is conceivable that the T2 variation reflects the number of layers of myelin, or the amount of bound water, rather than the absolute amounts of myelin, but this is unlikely.

The work of Koenig, Brown, Spiller, and Lundbom (1990) and of Koenig (1991) support this interpretation. The title of the latter's (1991) work, "Cholesterol of myelin is the determinant of gray-white contrast in MRI of brain" summarizes his opinion. His key piece of evidence was his Figure 1, showing an experiment comparing the contrast with pure water of two samples. One sample was 50% water and 50% the lipid, egg lecithin. This showed little contrast with pure water. However, when half of the egg lecithin was replaced with cholesterol, the contrast with pure water was much greater. This was interpreted as showing that cholesterol alters the lipid-water interfacial interactions that influence the behavior of water protons in a fluctuating magnetic field. Theoretical reasons (relating to the presence of an irregular surface with "potholes") are also presented to explain why (p. 288) "the geometry is ideal for magnetization transfer between the protons of the hydrogen-bonded waters and the cholesterol OH protons, as well as to the cholesterol C2, C3, and C4 protons just below the lipid-water interface."

It is known that myelination, in the sense of the thickness of the myelin layer, is correlated with axon size (Matthews, 1968; Friede and Samorajski, 1967; Bishop and Smith, 1964). Moreover, it appears that the increase in the thickness of the myelin layer is proportional to the thickness of the axons (Friede and Samorajski, 1967, p. 228).

The very fact that thicker axons have thicker myelin suggests that thicker myelin has some advantages, otherwise the myelin would be no thicker than required to function (perhaps 6 layers, judging from the Friede and Samorajski, 1967, report that fewer layers were seldom observed).

Thicker myelin must increase reliability, or speed, or both. There is probably considerable variability in the thickness of myelin along a single nerve fiber, and a greater average thickness could increase the thickness at the thinnest spots or reduce the number of such thin spots, even if the average is adequate.

Myelination is known to occur sequentially in the brain, and the order and timing in which myelination occurs in the brain is consistent with it causing the improvement in intellectual functioning that occurs as infants and toddlers mature (Gibson, 1991; Konner, 1991).

Delay of myelination is associated with clinically delayed development of motor milestones (Dietrich & Hoffman, 1992, p. 1071). More speculatively, later myelination may be related to children's increased ability to think abstractly as they age.

SPEED RELATED EVIDENCE

There are several early reports of statistically significant correlations between the speed of nerve conduction in the patellar reflex arc or the Achilles tendon reflex arc and intelligence (Vernon, 1990). Vernon and Mori (1992) have reported a correlation of peripheral nerve conduction velocity (interpreted as a surrogate for direct measures of brain nerve conduction velocity) with intelligence, although Barrett, Daum, & Eysenck (1990) and Reed & Jensen (1991) failed to find such a correlation, and Wickett & Vernon (in press) failed to replicate. Differences in myelin thickness could produce these differences, since myelin-sheathed neurons are faster. For already myelinated fibers, the effect of thicker myelin is to decrease the membrane capacitance (Koester, 1991b, p. 1036) and hence speed up the propagation of the action potential by reducing the time required to create a potential difference across the axon membrane (Koester, 1991a, p. 101).Reed and Jensen (1989, 1992) have developed a measure which is a good surrogate for nerve conduction velocity over a well defined path. Their measure was the time to the first electrical potential (the P100 visual evoked potential latency at the occiput) on the scalp after a visual stimuli, divided by the subject's head length. They found this to be correlated with scores on an untimed intelligence test. Another speed-related variable known to be correlated with intelligence is visual inspection time, the minimum time required to determine the longer of two tachistoscopically presented lines. It measures the minimum time the brain needs for a particularly simple visual discrimination. A review and meta-analysis (Kranzler & Jensen, 1989) found the best estimate of the correlation between IQ and visual inspection time to be -.54. The negative sign indicated that brains with shorter inspection times were more intelligent. More recently, Deary (1993) showed a strong correlation with performance IQ measures, although not with verbal ones. Again a measure of speed correlates with intelligence. However, Reed and Jensen (1993) reported that in their 147 subjects that visual pathway nerve conduction velocity and choice reaction times lacked an appreciable correlation with each other. This goes against the theory that myelination differences cause correlation of both with intelligence. This leads them to suggest that the more intelligent have shorter total length of cortical pathways, a possibility that might explain the lower energy use in the brains of the more intelligent. In contract to the visual inspection time results, auditory inspection time appears to measure a specific ability and not to correlate with intelligence (Langsford, Mackenzie, & Maher, 1994).Jensen (1982, 1992, 1993) and others (Jensen, Schafer, & Crinella, 1981; Barrett, Eysenck & Lucking, 1986; Lynn, 1991b; Lynn & Holmshaw, 1990; Lynn & Shigehisa, 1991, Beauducel & Brocke, 1993, Schweizer, 1993, also see the papers in Vernon, 1987) have shown that reaction time in various simple tasks negatively correlates with intelligence (i.e., faster reacting brains tend to be more intelligent). They have even been found to correlate with school marks (Van de Vijver & Willemse, 1991).Reaction times and intelligence correlate within families, as well as between families (Jensen, Cohn, & Cohn, 1989). The smarter sibling has the faster reaction time. Since siblings share many aspects of their environment (i.e. nutrition) this suggests the effect is not merely due to a shared aspect of the environment affecting both intelligence and reaction times. More importantly, this within family correlation suggests that both are being influenced by common genes that affect both traits. Myelination is a very plausible candidate for this trait.McGary-Roberts, Stelmack, & Campbell (1992), using apparatus that measures both reaction time and event-related potentials for 6 simple tests of mental function, have confirmed that both short reaction times and a short latency of the P300 wave of an event related potential correlate with intelligence (although the effect is not statistically significant for some items). Reed and Jensen (1992) also review evidence that P300 latency is inversely related to IQ. More recently a study (Polich & Martin, 1992) found P300 latency to correlate (inversely) with university grade point average, but, surprisingly, not with Raven's matrices scores (a good measure of intelligence). That event related potentials are indeed measuring intelligence is suggested by the fact that correlations of a composite average evoked potential measure with the various subtests of the Wechsler Adult Intelligence Scale correlate highly (Spearman rho of .95) with the factor loadings of these subtests on the general factor, g (Eysenck & Barrett, 1985). Polich, Ladish, & Burns (1990) showed that the latency of the P300 wave decreased as children (4 years to 20 years) matured, and that children that did worse on the Digit Span subtest of the Wechsler Adult Intelligence Scale had longer latencies, with both effects appearing about equally important on their Figure 2, and both being reported to be highly significant (at least one P<.001 for each electrode). The memory measure was statistically significant when age was controlled for. Unfortunately, results were not presented in such a way as to tell how much of age effect was left after the P300 latency (a speed measure) was controlled for. However, increased myelination is one of the features of brain maturation, and the simplest hypothesis to explain these results is that increased myelination increases the speed of brain processes, including the time to the peak of the P300 wave, and that this also somehow contributes to memory and intelligence. The Reed and Jensen nerve conduction velocity measurements, although over a specified nerve conduction pathway, are basically a variation of the latency studies made with evoked potential techniques. These show that certain waves are picked up earlier in the brains of the intelligent. The simplest interpretations of these results is that intelligent brains are faster. Many other simple measures of time required to do very simple tasks, such as determining whether a single digit was in a string of previously seen digits, or whether two words were the same or different, have been found to be correlated with intelligence (Vernon, 1983). The various measures of mental speed correlate with each other as well as with intelligence. Speed at the very simple task of drawing lines connecting numbers in order has a high correlation with IQ (Vernon 1993). Eysenck (1987, pp. 43-50) summarizes other studies correlating processing speed and intelligence. Vernon & Weese (1993) report a recent set of such experiments. Ho, Baker, & Decker (1988) found that 120 twin children's speed of naming of color, numbers, letters, and pictures was correlated with intelligence (statistically significant with r=.419), as was speed at symbol processing (also statistically significant with r=.418). Both intelligence and the speed measures were found to exhibit substantial heritabilities. Most importantly, correlated genetic effects appeared to underlie both intelligence and speed of processing, "lending support to the notion that speed and IQ may share some common biological mechanism" (p. 258).Most intelligence tests are timed, and reward speed. Thus, it might be thought that the correlation between speed and intelligence was due only to performance on timed tests benefitting from speed. However, Vernon (1983) shows that once general intelligence is controlled for, there is no significant correlation between mental speed and performance on the timed subtests of the intelligence test used. More importantly, Vernon & Kantor (1986) show that the reaction time and speed variables actually explain less of the variance of a timed intelligence test than that of an untimed administration of the same test.

Faster reaction times, faster nerve conduction, and shorter latencies for evoked potentials all represent doing the same things faster. Speed of simple operations and intelligence are found to be correlated, using several different methodologies. This suggests that a widespread property of the cerebral neurons is being measured, rather than just the ability of a particular part of the brain or nervous system to process information faster. Differences in the thickness of myelin layers or in the percentage of fibers myelinated are one possibility. Of course, other possibilities exist. Indeed, several independent factors derived from reaction times and related experiments appear to correlate with intelligence (Kranzler & Jensen, 1991; Vernon & Weese, 1993), suggesting there is more than a single physical difference.

Myelination in the human cortex continues throughout childhood (Yakovlev & Lecours, 1967), with the associative cortical areas showing increased amounts of myelin staining only by the second decade of life (Benes, 1989). Indirect evidence from interhemispheric coordination suggests that incomplete myelination of the corpus callasum leads to poorer childhood functioning on certain tasks (Hatta & Moriya, 1988). Because myelination is going on during childhood, myelination is a candidate (admittedly not the only one) to explain the improved intellectual functioning as children mature. Choice reaction times and their standard deviations, which are inversely correlated with intelligence, also decrease with age in children (Jensen, 1982, pp. 149-151). Gray-white matter contrasts increase as children mature (Schultz, 1991). In the elderly, reaction times and their standard deviations increase (Jensen, 1982), fluid intelligence decreases (Seligman, 1992), and gray-white matter contrast decreases (Schultz, 1991). Myelination could explain these changes with age.

It was discovered early in the study of intelligence that the ratio of mental age to chronological age (called the intelligence quotient or IQ) was approximately constant for an individual (Jensen, 1980, p. 105). This is consistent with there being some biological variable, such that children perform at the intellectual level of older children if their brain resembles that of older children with regard to the variable. Myelination is a plausible candidate for such a variable.

Kail & Bisanz (1992) showed that the time children took to do simple tasks declined at a decreasing rate with age, approaching asymptotically the adult levels. More interestingly, the decline was described by a formula in which the term (an exponential) describing the decline in time with age had the same value for five different tasks (mental rotation, name retrieval, memory search, visual search, and mental addition). Furthermore, in another experiment, the decline for six tasks was well described by the formula described in the earlier experiment, even though only two of the tasks were common to both experiments.

Kail (1990) has examined 72 studies of speed of mental processing that yielded 1,826 comparisons of children's average response times with adults' average response times for a specified task. He concluded that children take longer than adults on most tasks, and more strikingly, that children's response times increase linearly with the response times of adults on the same tasks. It appears that in doing a task both children's and adults' brains go through the same steps, but children do each step proportionately slower. As he points out, these results are consistent with age differences in processing time reflecting a general (non-task specific) component that changes rapidly during childhood and more slowly during adolescence. Myelination is a plausible candidate for the variable that causes the increased processing speed with maturation. Hale (1990) found similar results for 10, 12, and 15 year olds in comparison with adults.

This may be a good place to point out the remarkable resemblance between the performance of gifted 7th graders and Berkeley undergraduates on tests of information processing speed. Cohn, Carlson, & Jensen (1985) studied the speed of information processing in a group of gifted children who at ages of 12 to 14 years were succeeding in college courses in mathematics and science.

Very striking is their Fig. 2 (reproduced here as Figure 1), which shows the mean latency of various processing tasks for eight different tests in university students, gifted, and non-gifted students. The three lines are close to parallel, showing that the three groups have the same rank order and pattern of processing time for the different elementary processing tasks. The parallelism between university students and non-gifted children is an example of the similarity in processing times that Kailís (1990) literature analysis found to be common. More interestingly, the lines for the gifted students (12-14 years old) and the university students almost coincide with one another at the left hand side of the figure, with the line for the non-gifted 12-14 year olds being much further to the left, indicating processing times almost twice as great. The intraclass correlation between the 7th grade gifted and the university students was .98, the Pearson correlation was .99, and the Spearman rank correlation was .98 (p. 627). (Although Furneaux (1961) has argued that the log of response times may be better than absolute values, the similarity of the absolute values for the various groups suggests that such a transformation would have little effect. However, Hale, Myerson, & Wagstaff (1987) have shown that when response times for old adults are plotted against reaction times for young adults, a linear relationship is obtained if the logs are used, but not otherwise.)

figure 1 suggests that there is a basic neurological characteristic that determines processing speed, and that this characteristic changes systematically so as to decrease processing time as maturation occurs. This characteristic must essentially determine performance on these tests, and the gifted must be blessed with more of this characteristic. Life experience variables (such as culture) are unlikely to meet this condition, since the life experiences of 7th graders and university students are quite different.

Likewise, theories that hypothesize that variability in a particular part of the brain play a key role in explaining variability in certain of the tasks have difficulty in explaining the close resemblance between university students and the gifted 7th grade-aged students. For instance, since some tasks involve virtually no memory, some short term memory, and some accessing long term memory, the differences between the gifted and non-gifted are unlikely to result from factors that are peculiar to the memory mechanisms, or to parts of the brain that may be dedicated to memory (such as the hippocampus) or to overall planning (such as the frontal lobes). The gray matter of the brain appears to be highly differentiated, with different parts of the brain handling different functions.

It is unlikely that the differences between the gifted and the non-gifted involves the size of different parts of the gray matter or the anatomy of its neural connections, since it is hard to think of a mechanism that would consistently affect the gifted relative to the normal, while simultaneously having the same effect on university students relative to 7th grade students. The brain neurotransmitters appear to be highly differentiated, with different transmitters playing different roles in different areas and for different functions. Thus, their levels are unlikely to explain the pattern across tests.

Therefore, the difference between the gifted and non-gifted must be in a variable that affects, in a similar manner, many parts of the brain, and which also changes with age. One possibility is some unknown characteristic of the synapses, which causes them to act both faster and more consistently as children age, and which also exhibits considerable individual differences. The author knows of no such characteristics which change systematically with age during childhood (although such could exist).

However, the extent of myelination in the cerebral cortex is known to increase with age, and to continue increasing up to adulthood. It is plausible that myelination is greater in the intelligent, and presumably in the gifted. If the difference in the T2 relaxation time measurements indeed are due to differences in myelin content, as they seem to be, this would provide the evidence that myelination is greater in the gifted. It is plausible that the factors that determine the amount of myelin in one part of the brain are the same as those that determine it in other areas.

If children can't solve certain problems until certain areas have achieved a required degree of myelination, and children agree in the order of myelination (as they appear to) but differ in the speed, it would be possible to explain why children improve in ability as they age, and why mental age can explain children's ability in a wide range of areas. Those who achieve a certain pattern of myelination earlier would be able to perform as well as those older. This might explain why gifted 7th graders display a pattern of speeds that so closely resembles those of the Berkeley undergraduates whose intellectual performance they equal. Once myelination stops, the children whose brains were more myelinated would be those who had higher intelligences as adults. This would explain the correlation of childhood IQís with adult IQís. The inability to predict adult intelligences from those at very young ages (three or earlier) would be because the relevant areas had not begun to myelinate in any children, and hence there were no myelin related performance differences to observe.

If thicker myelin contributes to intelligence, and if some children have thicker myelin as adults because they accumulated myelin more rapidly as children, those children who will become the most intelligent will develop sufficient myelinization to lay down and retain memories at an earlier age. Rabbit & McInnis (1988) have shown that the more intelligent among the old recall memories from an earlier age than do the less intelligent.

Measures of variability in brain processes

The hypothesis that greater myelination improves mental performance was first suggested by Case (1985, pp. 377-381). In essence, he argued that less myelination led to greater cross-talk along neurons. Such cross-talk caused by transmission of signals across thin spots in the myelin layer could very well be the source of the random errors that both lower intelligence and raise the standard deviations for various measures of nervous system functioning. While Caseís hypothesis is a developmental one, proposed to explain why older children can do more than younger ones, it is easily extended to explain why children of the same age differ in intelligence (although he did not do so).Errors in transmission may explain the finding of Larson & Alderton (1990) that the highest correlations of intelligence with reaction time were found for the subject's worst performances, rather than with their best ones. Also (see their Figure 3), the difference in reaction time between the high intelligence and low intelligence subjects was minor for the quickest reactions, but much larger for the slowest ones (which took at least twice as long). Kranzler (1992a) also found that correlations with intelligence were greatest for the slowest reaction times for a variety of other reaction time tests. Jensen had earlier (1982) reported similar results.

Skewness

A particular form of a myelin related error hypothesis could explain the above findings, along with the puzzling finding of skewness in the typical individualís reaction time distribution. The hypothesis is that a thin spot in the myelin permits the signal to jump to an adjacent axon (i.e., a cross-talk error such as Case hypothesized). The turning on of the light in reaction time experiments is probably detected by discovering a new contrast between the image of the light and the adjacent area receiving no light. The fact that the reaction times are shorter for stronger stimuli, stimuli that cover a large area, stimuli impacting a dark-adapted eye, and even for binocular (as contrasted with monocular) stimuli (Woodworth & Schlosberg, 1954, Ch. 2) is consistent with at least part of the delay being due to the time required for a difference in signals from two parts of the retina having to reach some minimal value. With stronger stimuli creating more signals, the minimum level is reached quicker. Now imagine that a signal from the light's image jumps to an adjacent neuron that normally processes the signal from the dark area on the retina. The eye would no longer perceive as strong a contrast. However, by waiting a bit longer, the difference in the signals received from the two areas would reach the minimum significance needed to register that a light has come on. The net result would be that a processing error delays the recognition of the intensity difference that indicates a light has come on. The same argument could be extended to recognizing that lines differ in length (as in standard inspection time experiments), or that two stimuli differ in more complex ways. The criteria for recognizing a difference (in terms of differences or ratios of signals) would be set high enough for acceptable accuracy. The higher the noise to signal ratio, the longer the difference in signals must persist before a discrimination can be made. Striking evidence is provided by inspection time experiments where the subject must decide which of two briefly presented lines is longer. The shorter the exposure to the stimulus, the longer the time to respond (Vickers, Nettelbeck, & Willson, 1972, p. 280, p. 283). This counterintuitive result can be best explained by the brain having to wait longer for sufficient information to accumulate to permit a response. A plausible mechanism is for each release of neurotransmitter by certain afferent neurons to raise the level of a chemical in the receiving neuron, and for signals from other neurons to lower the level of this chemical. The neuron would fire only when the accumulated difference reached some critical level. Short stimuli result in weak signals and slow accumulation of the relevant chemical within receiving neurons, and hence slow responses. Reacting to the appearance of a stimulus, and discriminating between two stimuli, require observing the stimulus for long enough to separate the neuronal impulses created from background noise. This helps explain why reaction times and inspection times correlate with each other, and both correlate with intelligence. Any brain property that affects signal to noise ratios is likely to affect both inspection and reaction times. Now consider what happens if a signal transfers from one axon to another. An erroneous signal is created. Such an erroneous signal would delay the firing by reducing the difference between the number of positive and negative signals. Also, the "misrouting" could very easily have the effect of eliminating an expected positive signal (which is not propagated) and creating a negative signal. The missing positive signal could create a delay at the next processing stage, since the next neuron must wait longer for the required surplus of positive signals. However, the new (incorrect) negative signal is going to propagate through its own channels and produce one or more delays at other steps. Such a propagating error could affect multiple steps in the processing of a signal. Errors late in processing may cause only a single delay, but if one occurs at an earlier stage, delays at several steps would be created. Such a process could create the observed skewed reaction time distribution.It may not be obvious why a (intraindividual) skewed reaction time distribution is a puzzle. If the signal were processed sequentially in a series of steps, each taking about the same amount of time, and for each of these there was some variability, the total time required for the processing of the signal and for implementation of the actual reaction would be close to normally distributed. This would happen even if the distribution at each stage was not normal. (The Central Limit Theorem states that the distribution of the sum of a series of random variables approaches a normal distribution in the limit.) However, an individual's reaction times are not normally distributed. Instead, they are skewed (Jensen, 1982, p. 101; Woodworth & Schlosberg, 1954, Fig. 2-12, Fig. 39). Furthermore, they are skewed in such a way that the right tail is even more conspicuous for the retarded than for the normal (Jensen 1982, Fig. 17).It is not adequate to argue (as Jensen does in his 1982, 1987, and 1992 discussions) that there must be a physiological minimum, while the delays caused by errors are not bounded. A constant (the physiological minimum) plus the sum of a larger number of random and independently distributed delays should still be close to normally distributed. Thus, the skewness in the typical distribution implies that the processing errors are not independently distributed, or that a few large sources of error play an important role. The myelin theory can explain the lack of independence, if an early stray signal that propagates produces errors at successive steps. Eysenck (1987, p. 52) has argued that the brain must contain comparators which compare messages transmitted through different parallel channels and initiate action only when messages agree. Otherwise, they wait for more information. Skewness is especially likely if, after a repeat, there is the same probability of a second repeat at the same point. For instance, if there is a 50% chance of a delay, 50% of the signals will have no delay, 25% two delays, 12.5% three delays, etc. This is a highly skewed distribution. Although many mechanisms could produce such large errors, current leakage across myelin that created new erroneous signals would seem especially likely to do so. However, such skewness at an individual stage would produce large skewness for the reaction time only if most of the delays occurred during a single step. The sum of a series of serial delays, each with a skewed distribution, would be close to normal. Admittedly, there could be some alternative explanation for interdependence among the delays at different steps. A possible alternative would be some type of "attentional" variable in which signals were slowed down or speeded up as a group. If the attentional variable varied over periods of a few seconds or minutes, its presence might be detected by serial correlation in the reaction time of a particular individual. Laming (1988, pp. 74-75) has reported positive autocorrelation coefficients for fixed intertrial delays, although work reported by Greene (cited by Laming) shows less autocorrelation for experiments with random intervals between the warning stimulus and the stimulus to which a reaction was required. Most research on intelligence and reaction times, such as that by Jensen, uses random intervals. The theory that transfer of signals across the myelin layers is more frequent in the less intelligent could explain another puzzling finding. From visual inspection, the skewness appears much greater in the retarded than in normal individuals (Jensen, 1982, Fig. 17). Such a differential skewness is predicted by a theory in which the retarded experience more stray signals which then delay later processing. This greater skewness is essentially the same phenomenon as the small difference between the quickest reaction times of the intelligent and the less intelligent, but a large difference for the slowest reactions (reported by Jensen, 1982; Larson & Alderton, 1990; and by Kranzler, 1992a). The above theory predicts that reaction time skewness would correlate with intelligence. Juhel (1993) has reported such a correlation, and that it was stronger than the correlation of intelligence with standard deviations. However, Kranzler (1992b) did not find such a relationship. If there are more steps involved in reaction time problems with more choices (as the tendency for reaction times to increase with the number of choices suggests), the variability (measured by the standard deviation) will increase with the number of choices. This tendency for the standard deviation to increase with the number of bits of information being processed has repeatedly been found (Jensen, 1982, p. 130). It is not too surprising, as it is predicted by any model where errors introduce variability at each step, and more complex tasks involve more steps.

Reaction time variability increases rapidly with complexity

However, the above model has a less obvious prediction. For background, consider a model where the more complex tasks involve more steps (such as in Jensen, 1982). Now suppose the delays at each step are independent (such as might be true if they occurred only at synapses, and the cause of delays at one synapse did not affect delays at another). It would frequently happen that unusually large delays at one stage would be offset by unusually small delays at another. With such offsetting errors, the standard deviation would increase less than proportionately with the number of steps. More precisely, it would increase the square root of the number of steps, if all steps were identical. In the pyramidal system lying behind Hick's law (see illustration in Jensen, 1982, p. 128) in which the brain at each step reduces the uncertainty in the response by a factor of two (making the delay time proportional to the number of steps, and hence the number of bits of information involved), the standard deviation would increase less than proportionately with the number of bits. The data falsified the prediction. Even more surprising, the standard deviation increases more rapidly than median delay times (Jensen, 1982, 1992; Cohn et al., 1985, Fig. 3 and 4). As Jensen recognizes, this is a puzzle. He offers no solution to it. Even if the delays are perfectly correlated at each stage of the process, the standard deviations would increase only in proportion to the number of steps. Figure 2 shows the pattern which regularly emerges, using data kindly supplied from the relevant theses by Professor Jensen for physically active and educated elderly (aged 51-85, mean 68) (Ananda, 1985), and for university students (Vernon, 1983). The standard deviations are plotted versus the reaction times. The resulting curve is concave upwards, while it would be concave downwards if the reaction times were proportional to the number of steps required to solve the problem, and the times at each step were independent.

Somehow, additional variability is introduced in more complex processing, and this additional variability is more than proportional to the additional processing required. If errors in the early steps of processing actually lower the accuracy at the later steps, the rapid increase in variability with complexity could be explained. Furthermore, if the more error prone brains also had a tendency to introduce more errors into later steps, the more rapid increase in standard deviation in the less intelligent could be explained. Jensen (1982) has found that this is one of the most conspicuous differences between the more and less intelligent. Leakage of signals across myelin layers is one of the few mechanisms by which errors in early steps of processing could increase the standard deviation of later steps. Notice that the above leads to a different theory of why speed at elementary cognitive tasks is correlated with intelligence, as measured by the ability to successfully perform complex tasks. Jensen (1982, p. 122) has argued that short term memory traces in the brain decay rapidly, and a successful solution to a complex problem may require keeping several elements simultaneously in working memory. People whose brains work slowly will forget part of the problem before a solution is found. Thus, successful performance of complex tasks requires a fast brain. The chief problem with this theory is that the differences in reaction time between groups differing in intelligence are relatively small. For instance, Jensen reports (1987, p. 115) that the average 3 bit choice reaction time was 412 msec for gifted 7th graders, versus 523 msecs for average 7th graders. The average take 27% longer than the gifted. This might imply that the gifted could handle problems 27% more complex. Yet the difference in performance between the gifted and the average appears much greater than this. While the calculation is only illustrative, it is hard to explain how small speed differences produce large differences in ability to solve complex problems. Using intraindividual variability improves the situation somewhat, since average students' standard deviation was 118 msecs, versus 78 for the gifted (Jensen, 1987,p. 136). The average students' performances are 51% more variable than those of the gifted. This appears to be a more powerful effect, and if one interprets variability in aggregate times as indicating variability in individual steps, it becomes easier to imagine how the gifted derive their advantage. However, the complexity the gifted can handle still seems more than 51% above that which the average student can handle. Indeed, if this were all that was involved, one would expect average students to be able to solve the hardest problems the gifted can solve, although slower. Yet there are tasks the gifted can do that the average fail to do, even after repeated trials.The proposal that errors in processing at early steps somehow produces errors at later steps could produce an exponential increase in aggregate errors with problem complexity (just as the standard deviations of reaction times increase exponentially with problem complexity). This could explain why only a few individuals can solve the most complex problems. The leakage across the myelin layers is one of the few mechanisms which might produce this exponential increase in errors with complexity. There is one other implication of a system in which errors take the form of creation of erroneous signals. This is that total energy use will rise in brains that are more error prone. If these are in turn the less intelligent (as they are believed to be), the testable implication is that the brains of the less intelligent will use more glucose. This, as has been discussed above, is precisely what has been found. This provides a second mechanism (besides reducing the packing density of neurons) by which thicker myelin reduces energy use per unit volume.

Other reaction time considerations

Other reaction time considerations Jensen (1982, p. 115) notes that a satisfactory theory must account for the close relationship between a subject's fastest reaction times and his mean or median (an effect which has also been reported by Larson & Alderton, 1990, and by Kranzler, 1992a). He notes that correlations have been reported as high as .96 between the mean RT over 100 trials, and the average time for the fastest ten. This appears to be the same phenomenon as the linear relationship between standard deviations and median reaction times found by Myerson (undated). At first glance, this creates a problem for any theory that treats the standard deviation as measuring error proneness or noise, while treating the mean as measuring speed, since speed and error proneness need not be that well correlated if they are caused by different mechanisms. Thus, it appears that one mechanism must account for both, or at least explain a large common variance. Myelination can also explain, on a physical basis, a correlation between the fastest reaction times and the standard deviations. It appears to be the most plausible factor that could vary among individuals, and affect both processing errors and speed. Thicker myelin could both speed up impulse transmission and reduce the number of cross-talk errors. However, there is another mechanism that could explain a correlation between speed (as measured in the fastest reactions) and variability. The optimal criteria for a neuron to fire depends on how noisy the signals are. In a generally low noise system, a small difference between two signals can be safely interpreted as evidence of a real difference. Thus, low noise systems are likely to have the critical criteria set low, so that they respond quickly. Given individual variation in the noisiness of circuits, a well designed brain would have some mechanism for adjusting the criteria to the level of noise. High noise systems will have the criteria set higher. Even on those rare occasions when no errors occur, the receiving neurons will wait for the preset difference to accumulate (since they have no way of knowing that this time there were no errors). Hence, the noisier brains will have slower reaction times, even on the occasions when none of the possible random errors that cause delays occur. Thus, the correlation between the fastest reaction times and the standard deviations can be explained.

AGING AND MYELIN

Once adulthood is reached, reaction times start increasing again. An analysis of a large number of published studies, (Hale, Myerson, & Wagstaff, 1987) showed that response times could be well described by a general slowing of brain processes with age. They show the effect with Brinley plots (Brinley, 1965). These show the time the old take for a specific task versus the time young adults take for the same task. The times for the old are then found to be well approximated by a near linear (actually slightly curved) function of the times for young adults (usually university students). The times taken by young adults is usually interpreted as a measure of task complexity. It is generally found that the old take longer for all tasks, and that the lengthening of reaction times with age increases with complexity.

What is striking is that the time for old adults on a wide variety of tasks can be predicted so well from young adultsí times, even when the tasks differ in their apparent demands for short term memory or access to long term memory, or to resources localized in different parts of the brain. For instance, the times needed for mental rotations (which correlate with skills believed to be typically centered in the right hemisphere) and letter and word matching (which correlate with verbal skills believed to be typically centered in the left hemisphere) can be predicted from the same function. That such a good fit can be obtained by such a simple model, suggests some type of generalized slowing at the neurological level, rather than slowing limited to particular structures.

The relationship between measures of simple processing speed and fluid intelligence that is observed in the young is also found in the old. Speed on simple tasks correlates with mental performance on more complex tests of fluid intelligence well enough that most of the deterioration with age disappears if speed is controlled for (Hertzog, 1989; Salthouse, 1992; Salthouse & Babcock, 1991; Schaie, 1989).

Salthouse (1993c) has provided a table summarizing six major studies, all of which show reductions in the impact of age on cognitive performance after controlling for speed. All but one study reports that for all the tests given that at least half of the age related variance is eliminated by controlling for perceptual speed. (The exception is Salthouse and Mitchell, 1990, who report on spatial tests where controlling perceptual speed reduced the effect of age by from 20.3% to 92.9%.) Salthouse summarizes these studies by pointing out that, on average, age explains 15.8% of the variance, but only 3.5% after statistically controlling for the variation in a composite measure of perceptual speed. Thus, almost 80% of the age related variance in certain measures of fluid cognition is associated with variations in perceptual speed.

Since that review, Linderberger, Mayr, & Kliegl (1993) have reported that virtually all of the age related decline in intelligence (g) was mediated by measures of simple processing speed, and that 98% of the age related variance in general ability was shared with speed. Also in 1993, Salthouse (1993a) reported that the same equation could be used for both the young and old to predict performance on simple verbal tasks from speed and knowledge. Age seemed to have little independent effect, not mediated by speed.

The latency to P300 event-related brain potential is another measure of mental speed that correlates with intelligence (inversely). Interestingly, the latency increases with age in adults (Polich & Luckritz, in press), while it decreases as children mature. Thus, the pattern over the lifespan for P300 latency parallels that for other measures of mental speed.

The explanatory power of speed suggests that changes in some neurological variable both slow the brain and produce the poorer performance. Myelin degradations that increase at an accelerated rate with age is a plausible explanatory variable. It is not necessary that the myelin thickness actually decreases. Aging could impair the functioning of the myelin layers so as to increase errors. Other possible explanations exist. In particular, Cerella (1990) points out the cognitive slowing data is well explained by any mechanism that destroys or weakens random links in a neural network. Many age related changes in the brain are known (Rogers & Styren, 1987; Welford, 1984), most of which could affect speed or intelligence.

Age related myelin failures would be accompanied by increased intra-individual standard deviation in reaction times, and this has indeed been found (Salthouse, 1993b; Fozard, Thomas, & Waugh, 1976). A sample of the elderly (university educated) that averaged 59.19 ms which was appreciably above a weighted average of 34.61 ms from 11 university student samples that which had been obtained using the same methodology (calculated from Table 14 in Jensen, 1987).

An even more interesting feature is that the reaction times increase more rapidly with complexity in the old than in the young. Figure 3 shows reaction times versus number of bits for two samples measured in the same laboratory (that of Jensen, who was kind enough to send the data). The elderly sample (Ananda, 1985) consisted of 26 males, 50 females aged 51 to 87 (mean 68), physically active with a mean number of years of education of 15.25. The young sample consisted of 100 university students (Vernon, 1983). The elderly and the young differ very little in simple reaction times. However, they differ much more in choice reaction times, especially choice reaction times involving 8 choices (3 bits of information). As was discussed above, both samples are upwardly concave with complexity, an increase which is more than proportional to the time required to solve the problems and the presumed number of steps involved. However, the increase with complexity is much larger for the elderly. This is consistent with the myelin layers of the old permitting the creation of more erroneous signals. Each of these signals propagates creating additional delays at later stages.

Raz, Millman, and Sarpel (1990) using MRI have reported age-related prolongation in the white matter T1 decay time (Wahlund, et al 1990 have also reported this) which Raz interprets as reflecting age related declines in brain myelination. T1 measures the rate of decay of the magnetic signal due to interaction with nearby protons, and is usually interpreted as being an indirect measure of the water content. Myelin is a dense material relatively low in water. In particular, Raz reported that the grey-white matter contrast declined with age, virtually disappearing in the oldest individuals. The white matter T1 times showed a significant negative correlation with scores on the Cattell Culture Fair Test, considered a measure of fluid intelligence. Of course, even if a change in average T1 can be attributed to changes in myelin, MRI measurements lack the resolution needed to separate the decline in myelin accompanying neuronal death from a decline due to thinning or alteration of the layers surrounding particular neurons.

Another white matter change seen on MRI in healthy elderly persons correlates with attention and speed of mental processing (Ylikoski, Ylikoski, Erikinjuntti, Sulkava, Raininko, & Tilvis, 1993). The patchy or diffuse white matter changes known as leukoaraiosis correlates .48 (p<.001) with one measure of speed and attention (sum of Trail Making time and reading color names on the Stroop test), as well as with the Block Design of the WAIS-R and a language comprehension test. Since leukoaraiosis increased with age, its correlation with speed may merely result from both being a product of aging. However, the fact that leukoaraiosis still had a statistically significant effect on the above speed measure (r=.31) with age controlled for, suggests there may be a causal relationship here. Demyelination is one of several conditions stated to be related to leukoaraiosis. At a minimum, the Ylikoski et al. results suggests that white matter deterioration with aging can produce cognitive slowing.

General slowing with age could be explained by slowing within the axon of the neurons, at the synapses, by age related increases in the lengths of paths (presumably caused by certain links having disappeared due to neuronal death or deterioration), by the next neuron in a chain requiring a longer time to accumulate sufficient information for it to fire, or by a combination of these. A generalized slowing from any of these causes could adequately explain the data. Thus, the various hypotheses can not be distinguished by tests of functioning.

There is one aspect of the data that is not predicted by a model involving only across the board slowing. If the time required for a signal to traverse a neuron, or to cross a synapse was simply increased by a constant amount, the aging caused increase in delays with complexity would be proportional to the number of steps done. The number of steps might then be measured by the time required by the young (or another reference group) to do the task. This would predict a linear Brinley plot, with the time required by the old being a linear function of the time required by the young.

However, the slope of the Brinley plot exhibits a clear curvature, which disappears when the logarithms of the times replace the actual times (see diagrams in Hale et al., 1987). This suggests something more is going on than a simple slowing of all cognitive processes. A model is needed in which the delay per step increases with the number of steps involved. The above described model in which leakage across the myelin sheaths causes delays at subsequent steps is such a model. Later neurons must wait longer for the cumulated difference between two signals to indicate a statistically significant difference, at which point the neuron fires. The leakage across myelin layers model given above can be viewed as a version of information-loss model of Myerson, Hale, Wagstaff, Poon, & Smith (1990). They put forth a mathematical model with information loss (such as could be produced by leakage across the myelin layers) that accurately fits the data. Thus, the proposed model where brains differ in their ability to prevent propagation of errors across myelin layers can explain the age-related changes about aging that Myerson, et al. (1990) explain with their information loss theory. As put forth by Myerson, et al. (1990), the information loss theory does not make predictions about the standard deviations of response times, although a later paper did discuss them (Myerson, undated). Conceivably, the loss in information could induce a constant increase in time at each step, but with the increases being the same for all trials. This would leave the standard deviations unchanged. However, there is another possibility. Neurons might fire only when the accumulated surplus of positive signals over negative signals exceeded some amount. Stray signals emitted at random intervals (possibly due to leakage across myelin layers) would then increase the average time required for the neurons to fire. However, since the stray signals were emitted randomly, the magnitude of the delay would vary. As discussed above, the errors at different steps will be correlated with each other, thus causing the standard deviations to increase more rapidly with the number of processing steps than would be predicted from a simple model with random and independent delays at each step. By the time the later neurons in a more complex task were reached, many stray signals would have been created. However, the variability in the number of such signals would be quite large, causing the standard deviation to increase rapidly with complexity. Deterioration with age would then result in the intraindividual standard deviation increasing more rapidly with complexity for the old than for the young. Unfortunately, there is not enough published information on intraindividual standard deviations in aged individuals to fully test this prediction against alternative models with uncorrelated delays at each step. Few published reports of elderly reaction times provide intraindividual standard deviations.However, Smith, Poon, Hale, & Myerson (1989) provide some indirect evidence on standard deviations in an aged sample. They show a linear relationship between old adultsí reaction times and those of young adults for the 10th, 25th, 50th, 75th, and 90th percentiles of each individualís trials. All of these different points lie on a single straight line. This very striking finding is inconsistent with the variances at each stage being independent. The argument is as follows. The xth percentile time for an individual is the sum of his mean time plus a fraction of his standard deviation (assuming a normal distribution for reaction times on each task). Smith et al. show that the mean reaction times of the old increase linearly with the times of the young on the same tasks. Since there is a linear relationship between the xth percentile of reaction times for the old and mean times of the young, the standard deviations must be a linear function of the reaction times of the young. As pointed out above this is independent with independent delays at each stage, or with aging raising the variability at each step in when the delays at different steps are uncorrelated. The problem is to find an aging process that causes the aging related variability in processing time at each step to increase with the required number of steps in the required manner. Cascading errors from leakage across the myelin layers is one of the few processes that could produce the observed effects. Deterioration in the myelin layers with age could produce the hypothesized increase in variability with age. While the reaction time increase with age is well documented, as is the fact that this increase statistically explains the observed decline in performance on many tests, the magnitude of the observed slowing with age (1.5 to 2) appears too little to produce the observed decline in the ability to solve more complex cognitive problems. However, if longer reaction times are merely a sign of increased errors which, once created, propagate, it is very plausible that the increased error rate in the older, noisier systems could cause the last stages of processing to suffer from such high error rates that the most complex problems cannot be solved. This could explain the dramatic deterioration in the aged's ability to solve complex problems, as well as the rapid increase in their latencies with complexity. It is known that the complexity as measured in young adults by time taken to solve a problem correlates well with the percentage of young children who can solve the problem. Presumably, the same effect would be observed in comparing young adults with very old adults. Admittedly, other processes could produce the observed standard deviation increases with complexity and with age. Possibly, a task can be done by several sequences of neural actions, with some being quicker than others. Whether a quicker or longer route is used for each trial may depend on some random factor, such as whether certain neurons are ready to fire, or are still in the recovery period from having last fired. If the routes through the network are equivalent in the young and the old, but there is a general slowing, the result would be a spreading out of the distribution proportional to the degree of slowing. Myerson's recent showing (undated) that measures of intraindividual dispersion in response times are a linear function of median latencies, with the function appearing the same for both young adults and old adults is important, and is consistent with such a process. It is clearly inconsistent with a process where the delays at different stages are independent of each other, or where the older take longer because their brains go through more steps.

While there are several theories that could explain slowing with age, not all of them imply changes observable by MRI methods. However, changes in the myelin sheath, or disappearance of neurons and their myelin sheaths, are consistent with the prolongation of the T1 times described above. Thus, it is possible that myelin deterioration may help explain the age related declines in speed and intelligence.

Autopsy data has revealed that the myelin content of the human brain declines with age (Berlet & Volk, 1980). Ansari & Loch (1975) show that myelin basic protein (which accounts for about 30% of the proteins in adult myelin) in 7 individuals aged 72-78 was lower than in 7 individuals aged 45-65. The effect was quite striking, since the highest level in the old individuals was lower than the lowest in the middle aged individuals. However, some part of the age related decline in myelin may reflect an age related neuronal death, causing the disappearance of their myelin sheaths.

Thus, it is very plausible that changes in myelin could explain much of the age related decline in cognitive functioning. This could explain the general slowing, the increase in reaction times, the longer visual evoked potential latencies, the long P300 latencies, and the fact that measures of speed explain so much of the decline in cognitive performance. Not only are the MRI effects with age consistent with a myelin based theory, but the MRI measurements correlate with a test of intelligence.

EVIDENCE FROM INBRED MICE STRAINS

It was mentioned above that the timing of the development of different abilities in human infants and children is consistent with the timing of myelination. Additional evidence is provided by Wahlsten's (1975) research with multiple litters of 6 mice strains and their crosses, examining 14 measures of maturation, and myelination staining intensity for 80 different tracts per mouse. The timing of both maturation and myelination was expressed in terms of the age at which a standardized series of B6D2F2 hybrid mice achieved the same level. The timing of both myelination and maturation exhibited significant differences between the strains, indicating that both had a substantial degree of genetic variance. The timing of both exhibited heterosis for three crosses, as shown by each cross showing earlier development of both myelination and behavior than either of their parent strains. Since heterosis is not predicted by standard environmental theories, this is strong evidence that both age of maturation and myelination are subject to genetic influence. More striking is the correlation between the myelination age and the behavioral age. Their correlation was .937 (the equivalent ages were read from his Fig. 1) which is significant at better than the 1 in a thousand level. It is especially striking since both effects indicate heterosis, with the crosses maturing faster than either of the parent strains. This makes a causal relationship more likely. The behavioral indices were primarily reflexes (i.e., not measures of cognitive function). It is of course conceivable that some third genetically controlled variable controls maturation, and there is no causal relationship between myelination and timing of maturation. One piece of evidence against this general maturation hypothesis is that the thickness of the external granular layer of the cerebellum was also measured, and expressed as B6D2F2 hybrid age equivalents. This variable appears to have a weaker relationship with maturation, since it correlated only .640 with behavioral age. Since this thickness also reflects an aspect of brain maturation, the better fit of myelination makes it more plausible that the role of myelination is causal. While Wahlsten (1975) showed heterosis for myelination by measuring the darkness after staining, Seyfried & Yu (1980) showed it with chemical markers. They crossed only the C57Bl and the DBA strains, and also found heterosis. They indicated that the DBA strain is much higher in myelin than the C57BL, and two other strains, the BALB and the C3H.Although Seyfried & Yu do not comment on it, the crosses had larger brains (see the dry weights in their Table 1) than did either parental strain (except when measured at 7 days). This suggests that the added myelin was actually adding to brain dry weight rather than merely displacing other constituents. Since increased body size often accompanies increased brain size in crosses of different mouse strains (Leamy, 1985), it is possible that the brain size increase is incidental to a more general body size increase. Because there is very little myelin in the mouse brain before 10 days of age, the absence of heterosis in weight after 7 days (the weight for the cross is well between the weights for the two parent strains) is predicted if the heterosis in brain weight reflects the heterosis for myelin. For 14 days and after, when there is appreciable myelination, the brain weight heterosis is apparent. This evidence from mice makes it very plausible that genetically caused increases in human myelination would also add to brain size. If myelin indeed played a role in mice learning, crosses that show heterosis for myelin should also show heterosis for performance. Fortunately, these strains of mice are widely studied. While a complete search of the literature has not been made, there are reports of such effects. Olivero, Castellano, & Messeri (1972) reported that the DBA strain made fewer errors (69.3) than the C57BL (84.0) strain on a Lashley maze. The cross of the two strains made appreciably fewer errors (44.9) than either parent strain, with a reduced level continuing in the next generations (F2, 54.1, and F3, 38.3). However, the ability of the mice to learn to avoid a shock by running to another compartment when a light came on was also studied. Heterosis was not observed. However, strains are not always consistent across tests in mice, and effects of other genes differing between the strains could easily explain the differences across tests.Upchurch & Wehner (1989) crossed the DBA and C57BL strains and examined their spatial ability, tested by ability to find a submerged platform. Heterosis was observed, although the two pure inbred strains did not appear to differ in ability. Their literature review makes it clear that heterosis for learning is frequently found in this cross, but not always. Paylor, Baskall, & Wehner (1993) have argued that the DBA strain has a defective hippocampus, which could explain their poor performance on some tasks. If the relevant genes were recessive, it might also explain the heterosis in performance found when the DBA strain was crossed with the C57 strain. Anisman (1975) experimented with several tasks of increasing complexity in a DBA and C57BL cross (among others). He found that as the task complexity increased, intermediate inheritance was replaced by complete dominance, and then by over-dominance. If the more complex tasks required more "intelligence," and thicker myelin is more important for the more complex tasks, his results could be explained. If one does not wish to use the concept of intelligence, it is conceivable that leakage of signals prevents good performance on only the most complex tasks. Evolution could be expected to have provided the typical mouse with thick enough myelin to insure adequate performance on most ordinary tasks. Yet, successful performance on the most complex tasks may not have been critical enough for survival to justify the cost of providing enough myelin to always ensure the best performance. That heterosis caused increased brain myelin might speed up nerve conduction is suggested by the fact that heterosis is found for nerve conduction velocity in the mouse tail (Hegmann, 1972) when two other strains are crossed, the DBA (high myelin, according to Seyfried and Yu) and C3H (lower myelin). (The Seyfried & Yu (1980) report of lower myelin in these strains cites Seyfried, Glaser, & Yu (1979), who actually report ganglioside levels, which are interpreted as markers for myelin.) Interestingly, in another experiment, these same two strains also show heterosis for learning a water maze with visual cues, for the improvement with experience, and for reversal of learning of a T maze (Henderson, 1972), although poor visual capacity in the C3H strain may affect their visual learning of the water maze. Of course, since mice frequently exhibit heterosis for body size, it is possible that the crosses were large, had thicker axons, and hence showed faster nerve conduction velocity. Unfortunately, axon thickness was not measured. However, Hegmann (1979) published electron scanning microscope pictures of tail cross sections from strains bred for high and low caudal nerve velocity for ten generations. In these pictures, the strain bred for high conduction velocity appears to have thicker axons and thicker myelin. Incidentally, the fact that selection had produced significant differences in nerve velocity, shows that velocity exhibits genetic variability. While the genes that affect the tail nerve conduction velocity also increase sciatic nerve conduction velocity (suggesting a general effect on peripheral nerves), it is not known how they might affect the brain. Hegmann (1979) argued that they did affect the brain since certain behaviors, open field activity and defecation, varied in parallel with tail conduction velocities across his strains bred for high and low conduction velocities. These behaviors are not markers for intelligent behavior or for learning. If the behavior differences are indeed due to the same genes as affect peripheral nerve conduction velocity, it may be through affecting the fibers that transmit messages from the parts of the brain that control or inhibit such behaviors. There is no record that Hegmann, after having bred his strains for high and low velocity, tested them for maze running or other abilities. The fact that Hegmann was able to breed mice for high and low peripheral nerve velocities (measured in the tail), suggests the possibility of breeding them for variations in similar characteristics in the brain, and then testing the behaviors of the animals. The fact that mice strains differ in brain myelin, in brain size, in tail conduction velocity, and in ability to learn simple tasks, and frequently exhibit heterosis for such traits, suggests they would be a good model to study. If heterosis is exhibited for a physical trait, but not for a behavioral ability, it suggests that physical trait is not causally related to the behavior. For instance, since the size of the intra- and infrapyramidal mossy fiber field in the hippocampus exhibit intermediate inheritance for F1 crosses of the C57Bl and DBA strains (Crusio & Schwegler, 1987, as cited by Upchurch & Wehner, 1989), it is unlikely that this determines performance in the Morris water maze task, since heterosis is observed for this task (Upchurch & Wehner, 1989).If myelin is critical to mouse intellectual performance, myelin deficient mice would show worse performance. This prediction has been tested (Inagawa, Watanabe, Tsukada, & Mikoshiba, 1988). No differences were found in radial maze learning (not searching for food in arms that had previously been visited), which is behavior that may be built into the mouse, since it resembles their natural behavior. However, in reversal learning (in which the arm of the T maze containing food was changed, forcing learning of new behavior), the shiverer mutant had the worse performance, followed by the mld, followed by normal mice. Since the ranking of strains corresponded to the ranking by extent of myelin deficiency, it was concluded that myelin formation was related to learning, but not inborn, natural behavior.

OTHER EVIDENCE

In multiple sclerosis (caused by partial demyelination), reaction times increase (Arena, Mazzoni, Moretti, & Lepori, et al., 1986; Rao, St. Aubin-Faubert, & Leo, 1989) and intelligence declines (Medaer, de Smedt, Swerts, & Geutjens, 1984; Matthews, Compston, Allen, Martyn, 1991, pp. 56-57). Also, P300 latencies increase (i.e., slower processing times as in healthy subjects with poor intelligence test performance) (Rao, 1990, p. 175; Polich, Romine, Sipe, Aung, & Dalessio, 1992). Also, Polich, et al, (1992) showed that the visual evoked potential latency (which appears somewhat similar to the measurements that underlay the Reed and Jensen [1992] nerve conduction velocity measurements, which correlated with intelligence in university students) was significantly slowed (P<.001) in the multiple sclerosis patients. As discussed above, these latencies are a measure of cognitive speed which correlates with intelligence. The slower processing in those with long latencies for the P300 wave could be due to slower nerve conduction, or due to delays at the synapses. The fact that the demyelinating disease, multiple sclerosis, lengthens the latencies makes it more plausible that the variations observed between the bright and the less bright, the increase in adults with age, and the decreases observed as children mature, also reflects differences in the performance of the myelin sheath. Patients with abnormalities of later event-related potentials often have significantly delayed reaction times, interpreted as suggesting that lack of white matter integrity can affect both event-related potentials and reaction times (Newton, Barrett, Callanan, & Towell, 1989). Although multiple sclerosis involves only demyelination at particular spots, the effects of this selective demyelination, and the resulting interference between adjacent axons, may be qualitatively similar to the effects resulting from occasional naturally occurring thin spots in the myelin.Volpe (1991, p. 277) has hypothesized that "myelination could be the crucial process impaired in very-low-birth-weight infants with subsequent cognitive deficits." The evidence for this includes delayed myelination in infants after neonatal injury to the white matter, and that microcephaly due to perinatal injury in full term infants appears at a mean age of approximately nine months. Obviously, this hypothesis requires that the extent of myelination affect later intellectual performance. If adverse effects on myelination from injury lower intelligence, it is plausible that poorer myelination from other sources would have similar effects. Also, mentally handicapped children over 14 years of age, classified as autistic have significantly slower central conduction times (McClelland, Eyre, Watson, Calvert, & Sherrard, 1992). Because of the age distribution, this has been interpreted as being due to a maturational defect in myelination.

Diamond et al. (1985) reported that one area of Albert Einsteinís brain had an exceptionally high ratio of glial cells to neurons. Since glial cells create and maintain myelin (Vom Muralt, 1972, p. 5; Kandel, 1991, p. 22), this would be consistent with thicker myelin contributing to greater intelligence. Also, suggesting that a real effect exists, Diamond (1988) found that rats reared in an enriched environment had an elevated ratio of glial cells to neurons.

BRAIN LIPIDS AND EVOLUTION

Myelin contains high percentages of fatty acids and ìelongated and desaturated derivatives of linoleic acid (C18:n-6) comprise substantial percentages of myelin fatty acid compositions.î (Dewille & Horrocks, 1992, p. 214). Linoleic acid is an essential fatty acid which cannot be made by humans (Dewille & Horrocks, 1992, p. 216). Evidence is accumulating from animal studies and human clinical studies that long chain polyunsaturated acids are critical for normal brain functioning (see Koletzko, 1992).

It is possible that genes causing heavy use of linoleic acid, and its long chain polyunsaturated derivatives in brain myelin, produce problems if inadequate amounts are available. Linoleic acid derivatives are used in the peripheral nerve myelin, and in other body membranes. If more is removed from the blood to build brain structures, there is less left for other uses. Alternatively, having less available than is needed may lead to adequate quantities of myelin, but myelin of an inferior composition, such that the animal is less fit.

The brain apparently gets first claim on myelin lipids at times of scarcity of circulating lipids or their precursors (Dobbing, 1993). It then follows that other parts of the body must experience shortages if total intake is inadequate. Presumably, the more the brain is programmed to build myelin, the worse this deficit will be for other organs during famines. Thus, the curve for fitness versus myelination could be very flat, even though an intelligent brain can do virtually all tasks better than an unintelligent brain. Thus, the retention of the genes for low intelligence could be explained.

Myelination is one of the few theories that can explain why humans show a high degree of genetic variability for intelligence (evidenced by a high heritability, or by genetic variance being a high proportional of total variance). Intelligence (technically psychometric g) appears to be a trait that contributes to fitness, since the more intelligent do better at virtually all mental tasks, including memory, building things, disembeding figures from backgrounds (relevant to identifying camouflaged animals while hunting), and even relatively simple tasks such as reacting to lights, quickly identifying differences in length, scanning memory, etc. (Jensen, 1980, chap. 8). Many of these skills would have been of value even for pre-humans. For instance, spatial ability (measured from a standard intelligence test) correlates with throwing accuracy, a skill that probably contributed to hunting success in prehumans (Kalakowski & Malina, 1974).

Theories of differential intelligence that involve differences in network connectivity, neurotransmitter levels, enzymes, design of various structures, dendrite branching, etc. suffer from the problem that inferior designs should have been consistently selected against. Indeed, with regard to many of these variables, designs that are bad for humans were probably bad for the ancient ancestors of humans, and should have been eliminated long ago. A viable theory of the biological basis for intelligence (or at least for the genetic part of it) must identify a disadvantage to the genes for high intelligence in order to explain the persistance of the genes for low intelligence.

Besides brain size, and MRI parameters, the only physical difference reported in brains of different intelligences relate to dendritic measurements. Jacobs, Schall, & Scheibel (1993) have reported dendritic differences in the brains of those of different educational levels. Since similar changes have been produced in rodent brains by environmental enrichment, these differences are probably non-genetic in origin. There appear to be no major disadvantages to the dendrite patterns of the more intelligent, while there are disadvantages to low intelligence. Thus, the observated variation is probably not genetic in origin, since natural selection should have eliminated any such genetic variation long ago.

The only non-myelin based theories the author knows of that identify a disadvantage are those that assert that more intelligent brains contain more neurons, cortical columns, or similar energy and space using elements. These impose a burden on the rest of the body through increasing the requirements for food, and enlarging the head (which creates birth difficulties). While such theories may explain part of the genetic variance in intelligence, the brain size intelligence correlation is too small for them to explain much of it. If the theoretically necessary fitness penalty is imposed through a more intelligent brain using more energy (say by having more impulses per second in each neuron) run up against the difficulty that more intelligent brains use less energy per unit volume, as measured by positron emission tomograpy. Theories with no disadvantage to genes for intelligence can not explain low intelligence.

Other Lipid and Intelligence Correlations

Myelin contains a relatively high proportion of cholesterol and other lipids, including some derived from the essential fatty acid, linoleic acid, which humans cannot produce, and hence must have in their diet. While most variation between people in the thickness of their myelin layers is likely to be genetic, environmental variation due to diet and prenatal effects probably contribute to some variation. If there is variation from this source, several otherwise hard to explain facts could be explained.

In a sample of 209 students, those with high cholesterol levels had, on a Jensen type apparatus, quicker reaction times than those with low cholesterol for four (p<.007) and eight choices (p<.047), with one and two choice reaction times being in the predicted direction (Benton, 1994a). As discussed earlier, reaction times on this apparatu correlate with intelligence.

In a study of 7076 randomly chosen British adults, faster choice reaction times were associated with a high fat diet in both white collar (p<.0001) and blue collar (p<.0001) individuals (Benton 1994b). The effect was large enough so that the reaction times for white collar workers eating little fat were actually slightly below that for blue collar workers eating much fat. Since myelin contains large amounts of cholesterol, it is plausible that the plasma cholesterol effects relate to myelin. However, since the high fat diets described were probably high in animal fats, they may also have been rich in essential fatty acids. Assuming the effects reported by Benton are real, it remains to be established whether the diet leads to quicker reaction times, or whether brains that require large amounts of certain lipids produce a desire to eat foods containing these, or both are a result of something else.

Evidence that any lipid intelligence correlation might be due to non- genetic causes is provided by evidence from monozygotic twins (whose genes are identical). The member of a monozygotic twin pair who did better in high school has a statistically significant tendency to show higher high density lipoprotein cholesterol and lower low density lipoprotein cholesterol (Fabsitz, Feinleib, and Garrison, 1978, p. 75 and Table V).

Weiss (1984) has drawn attention to a correlation observed in trisomy 21 syndrome patients between erythrocyte glutathione peroxidase activity and intelligence, and to a higher average activity in an university population than in the general population. This has convinced him that psychometric intelligence correlates with interindividual differences in the rate of lipid peroxidation. Given that myelin is a lipid-protein combination, his speculations may be related to those offered here.

Babies whose mothers had had biliary tract disease were significantly lower in intelligence than the controls according to Churchill, Ayers & Caldwell (1967). They speculate that such disease might have lowered the availability of linoleic acid for fetal brain development. If environmental factors could produce such effects, genetic variability might also.

Studies (such as Rodgers, 1978) have reported higher intelligence in term babies fed human milk (rather than a formula). However, the results of many of these studies might be due to those of high socioeconomic status (who tend to be the more intelligent) doing more breast feeding, as they are known to do. However, there was an 8.3 IQ point improvement in the intelligence of premature infants (in a study of 300) fed motherís milk (Lucas, Morley, Cole, Lister, & Leeson-Payne, 1992). While it is possible that this effect is due to the higher intelligence parents choosing to provide milk, it is unlikely. The effect remains after controlling for social class and mother's education. Lower intelligence is found in children of mothers who desired to provide milk, but did not produce enough. Motherís milk contains essential fatty acid derivatives that are not in standard infant formulas.

Evidence from Premature Infants

Carlson (1993) has studied premature infants whose diet was supplemented with fish oils (which contains several long chain fatty acids, including docosahexaenoic acid) in order to improve their docosahexaenoic acid status (which retinas and human milk contain, but which is lacking in standard infant formulas). There were several reasons for the research project. Docosahexaenoic acid is concentrated in retinas, and contained in human milk, but absent from standard infant formulas. Breast fed term babies had higher levels of docosahexaenoic acid in their brain than those given a standard infant formula (Farquharson, Cokburn, Patrick, Jamieson, & Logan, 1992). Bourre et al. (1989) showed that extreme lowering of dietary levels of an essential fatty acid, a-linolenic acid, lowered the levels of docosahexaenoic acid in the brains of rats (including the myelin), and also adversely affected their learning abilities. Since myelin's docosahexaenoic acid concentration is normally low, and this fatty acid is found elsewhere in the brain, any effects of its consumption may be through effects on other parts of the brain. As Carlson et al. hoped, visual acuity was improved in the infants given the fish oil supplement.

However, an unexpected effect was a decline (relative to the control group) of infant length, weight, weight to length ratio, and head circumference. There was also a decline in plasma phospholipid arachidonate levels (abbreviated AA) (which is believed to be a result of the eicosapentaenoic acid unavoidably included in the fish oil competing for access to the enzymes that produce AA from the essential fatty acid, linoleic acid). AA status had a statistically significant correlation (positive) with growth in length, and weight through 96 weeks post conception.

Arachidonic acid status and head circumference correlated (r=.37, p<.01 and r=.34, p<.01) at the equivalent (prematurity adjusted) ages of 2 and 4 months, with the correlation declining at 6.5 months (.24, p<.07), 9 months (r=19, non-significant) and 12 months (r=.08). While the correlation could be spurious or due to any of several causes, an interpretation relevant to the hypothesis here is that lower AA availability lowered myelin production, since AA is an important component of myelin. This could have produced the change in head circumference. Very low birth weight infants (Hack, Breslau, Weisman, Aram, Klein, & Borawski, 1991) with subnormal head circumference at an equivalent age of eight months (prematurity adjusted), had lower IQ and academic achievement at eight years.

The Carlson et al. study tested infant functioning with the Bayley tests and with the Fagan Infantest. At a postconception age of 96 weeks, there was a statistically significant decline in the Fagan novelty preference score, which is known to be correlated with later intelligence. However, there was an increase in the number of discrete looks, which the authors point out may be a positive sign of intellectual performance. A myelin hypothesis interpretation of this experiment (which does demonstrate effects on human vision and brain functioning from experimental manipulation of essential fatty acid intake) interprets the Fagan test in the usual way as evidence of poorer mental functioning, and explains it by a lowered availability of arachidonate adversely affecting myelin production.

This interpretation might be supported by the fish oil supplemented infants having significantly lower Bayley Pschomotor development (p<.03) scores, although the decline in the Bayley mental test scores were not significant. There was a significant correlation between AA status and normalized growth, and between normalized growth and the 12 month psychomotor scores. If attainment of the types of performances measured by the Bayley tests is controlled by when relevant tracts myelinate (which Wahlsten's 1975 mouse evidence suggests), an adverse effect on the essential fatty acid derivative AA would predict the observed effect on scores. This interpretation would be consistent with the known failure of infant Bayley scores to predict later intelligence if virtually all children eventually develop adequate myelinization of these tracts, and the timing of such early myelination was not correlated with the later thickness for the intelligence related tracts. An environmental intervention that slowed myelination (by constraining AA availability) could have an effect on the timing of the development of the tracts that controlled the activities observed in the Bayley tests.

Admittedly, showing that correcting a probable nutritional deficiency improves mental functioning does not show that that genetic variability in essential fatty acid products explains variability in intelligence, but it makes such a hypothesis more plausible.

The Carlson et al. study showed statistically significant effects of AA status on the length and weight of the total sample of infants (which included those whose status was adversely affected by the fish oil supplements) through 96 weeks postconception, and on the control infants (whose diet was not supplemented with fish oil) between 2 and 6.5 months. The most impressive correlation was .53 (p<.001) at 4 months between length and AA status. It is not known why AA status affects weight and length, other than that AA is used in building many body membranes. Other studies report an inverse correlation between AA status and infant size (Ongari, Ritter, Orchard, Waddell, Blair, & Lewis, 1984; Koletzko & Braun, 1989 as cited by Carlson 1993), although since the AA status is not the result of a manipulation, they are less persuasive of a causal relationship. Taken together, these studies suggest that inadequate quantities of AA in the blood can limit bodily growth.

In the Carlson et al. study the variations in AA status were not due to differences in the availability of the raw material for AA, the essential fatty acid linoleic acid (as shown by the lack of a correlation between linoleic acid status and growth). However, since the body can make AA from linoleic acid, but cannot make linoleic acid, it is plausible that lack of the precursors for AA, possibly during famine, would have an adverse effect on growth. During such an essential fatty acid shortage, having genes that removed AA from the blood to make myelin might be fitness lowering, since the rest of the body would be adversely affected. This could explain why the genes for low intelligence (low myelin) had been retained. This effect would be most important among peoples whose natural diet was low in animal fats and other sources of linoleic acid or other w-6 fatty acids (see Miller 1994 for a discussion of possible differences among peoples under prehistoric conditions in the availability of animal fats).

Thus, the Carlson data is consistent with the hypothesis that using essential fatty acids to build brain myelin imposes the cost of sometimes being short of such essential fatty acids for the rest of the body. The net result is that fitness is not very sensitive to the gene frequencies relevant to myelination, thus permitting the survival of the genes for low intelligence.

Other lipid evidence

On an even more speculative level, there has been a worldwide rise in average intelligence. Lynn has proposed that this is related to improved nutrition (1990c, 1993). One of the most important nutritional changes has been an increase in the amount of animal fats in the diet. This increases the amounts of certain long chain fatty acids that are derived from essential fatty acids.

The strong human taste for fats is usually described as having evolved because of their high caloric content per gram (i.e. they represent an unusually concentrated energy source), but it is conceivable that it evolved, at least partially, because fats contain an essential raw material for producing the large human brain. Even today modern food technologists find it hard to formulate good tasting foods that do not contain fats (Gibson, 1992). More primitive hunting peoples show a strong preference for game animals that have a high fat content.

Eskimos, for instance, focus their caribou hunting on the fall migrations rather than the spring ones, reportedly because the fat content is very low after the winter. Lee (1969) describes a strong !Kung Bushmen preference for fat. Chimpanzees start eating the monkeys they kill by biting into the brain first (Goodall, 1986). Possibly this tendency is related to brains providing an essential nutrient that chimpanzees have evolved to desire.

IMPLICATIONS YET TO BE TESTED

The hypothesis offered by this paper is readily testable. Inspiring testing is one reason for writing this paper. Possible tests include:-The white matter of the more intelligent brains should contain more myelin. At least one brain collection is available for such studies. It contains both brains and IQ measurements made before death of the subjects (Witelson & McCulloch, 1991). One method is to measure substances diagnostic for myelin levels, such as myelin basic protein, or gangliosides. Alternative hypotheses testable by chemical means might be examined at the same time. Specific lipids, especially those in the omega 3 and omega 6 series (derived from essential fatty acids), might be correlated with intelligence. It would also be possible to look for specific compositional changes that correlated with age and intelligence, since these might indicate specific forms of deterioration that cause intelligence to decline with age.-Alternatively, the traditional method of sectioning and staining for myelin might be used. Such a study might very well measure other traits (neuron densities, thickness of fibers, density of fibers, etc. which might differ between brains). Since available brains tend to be of a variety of ages, with most from old people, hypotheses relating to age can be most easily checked.-At third approach would use the electron microscope on specific tracts.-Also, if this higher myelin content is the result of more glial cells, analysis similar to that conducted by Diamond et al. (1985) on Einstein's brain should show a higher ratio of glial cells.-The pattern of intra-individual standard deviations and skewness found for reaction times should be displayed by other speed-related measurements (such as inspection times, P300 latencies, time to perform various simple tasks, etc.). Standard deviations and skewness should correlate with intelligence. As task complexity increases (measured by processing time), the intraindividual standard deviations should increase. Furthermore, they should increase more rapidly than the square root of the central processing time (a surrogate for the number of steps involved). -If deterioration in myelin with aging produces the hypothesized effects of increased leakage, the relationships between intraindividual standard deviations and intelligence that have been found in teenagers and young adults will also be found in the old. If the intra-individual standard deviations are controlled for, the effects of age should be substantially reduced, or even eliminated. Testing may require only a new analysis of already collected reaction time data.-Variances in the integrity of the myelin layer should correlate with intelligence. Multiple sclerosis may be a suitable model although the lesions occur at specific spots, rather than across the whole sheath. If it is a reasonable model, variations in mental speed produced by the disease and measured by visual evoked potential methods, reaction times, or inspection times, should correlate well with intelligence. If the model is a good one, the function relating cognitive functioning to measures of speed should be similar in multiple sclerosis patients and normals, such that the decline in cognitive functioning accompanying a specified change in mental speed would be the same regardless of whether the decline was produced by the disease, or by normal variation.-If myelination in children produces effects similar to (but opposite in direction to) myelin deterioration in older individuals, and these effects are similar to those that separate the more intelligent from the less intelligent, plots of reaction times on various tasks against a standard group (Brinley plots) should be parallel for those of similar abilities but different ages, over a range from childhood to old age. A similar pattern should be obtained for plots of standard deviations, or even skewness on various tasks.-Children's intellectual performance should correlate with MRI myelination measures, such that children whose brains appear older on MRI also exhibit performance typical of a higher age level. Conversely, controlling for the level of MRI measured myelination should reduce the effects of age on children's performance and reaction times, and possibly eliminate the effect of age.-Intellectual performance in children should correlate with indirect measures of myelination of the corpus callosum. Tests exist which are intended to measure the transfer of information from one hemisphere to the other. For instance, Hatta & Moriya (1988) compared the ability to identify the location and sequence of taps to the finger by using the thumb of the same hand (no transfer of information between hemispheres), with the performance when identification is made using the thumb of the opposite hand. Differences in performance were greater for those under 10 years of age, which was interpreted as due to difficulty in transferring information between the hemispheres due to incomplete myelination of the corpus callosum. If this technique indeed measures myelination differences, and if myelination indeed affects intelligence, a correlation should be observed between the difference in performance in the two conditions, and overall intellectual performance. -If myelination accounts for the decrease in reaction time means and standard deviations, as well as the increase in ability with maturation and the change in white matter parameters, the variations across age in children should be parallel, with rapid growth occurring at the same ages in all of these measures.-There should be common genetic variance between reaction times and standard deviations for the various speed related measures and intelligence. Twin data could be used.-Since the timing of such growth spurts and plateaus is partially genetic, the profiles for monozygotic twins should be more similar than for dizygotic twins. The basic methodology would be similar to that of Wilson (1983). Also there should be highly significant cross monozygotic twin correlations between the timing of these variables (i.e. the timing of a plateau in twin 1ís reaction time standard deviations should occur at the same time as a plateau in twin 2ís intelligence). Because pre-school children can do simple reaction time tests, and reaction times and their standard deviations can be measured on an absolute scale, they are well suited for such longitudinal studies.-If an appreciable part of the genetic variance in both speed and standard deviation is due to genetically controlled myelination, within family correlations for reaction times and its standard deviations should be positive, and similar to the correlations between families. If they are controlled by different genes, the two traits could be inherited differently. Data such as that collected by Jensen, Cohn, & Cohn, (1989) could be used to answer this question.-To confirm the theory that magnetic resonance imaging methods are indeed measuring myelin levels, MRI scans of brains should be correlated with chemical and microscopic examination of tissue samples from cadavers. To insure a short time between MRI measure and the brain examinations, primates or other large experimental animals might be used.-The performance of experimental animals on reaction time tasks should correlate with the myelin content of their brains. Reaction time experiments are simple enough to be done by experimental animals. Cats have been trained to do choice reaction time tests (Jensen, personal communication). If the results in experimental animals were similar to those in humans, their myelin contents could be measured, thus testing the above theory. Of course, other theories regarding brain structure and reaction times could also be tested this way, permitting a comparison of their predictive power with that of the myelin theory.-Energy use differences (measured with positron emission tomography) between the more and less intelligent should increase rapidly with the number of processing steps required by the experimental task (as measured by reaction times).-Experimental work with inbred mice strains should show positive correlations between tests of mental performance, tail conduction velocity, and brain myelin content (measured after sacrifice of the animals).-Offspring of crosses of inbred mice lines differing in myelin (such as C57 and DBA) should show heterosis on tests of mental performance and caudal nerve conduction velocity, and after sacrifice of the animals tested, heterosis for brain myelin content, and caudal nerve myelin thickness. This test would be especially persuasive if heterosis was found for performance, conduction velocity, and myelin content, but was not found for other brain characteristics which might plausibly have affected performance, such as properties of the hippocampus. Even if it was decided that tasks that mice can perform were not good measures of what is called intelligence in humans, mice might still be good models for studying the determinants of myelination, axon thickness, and nerve conduction velocity.-If cross-talk really does occur between axons, it should be possible to demonstrate its presence in the laboratory. Possibly this could be done by sending signals down one axon, and picking up action potentials from others.

CONCLUSIONS

A number of different observations concerning intelligence could be explained if individuals with greater myelination are also more intelligent. These include faster nerve conduction in the more intelligent, less glucose utilization in the brains of the more intelligent, less glucose utilization per unit volume in the large brains, the higher glial to neuron ratio in the brain of Einstein, the intraindividual skewness in reaction times, the differential increase of standard deviations and means of reaction times with the number of bits, and the shorter T2 relaxation times in the white matter of the more intelligent. Other facts about child development and aging could be explained by increased myelination over time up to adulthood, and then myelin deterioration afterwards.

REFERENCES