Intracranial capacity and brain volumes are associated with cognition in healthy elderly men

Subjects.

Subjects were 100 healthy, unmedicated male volunteers aged 65 to 70 years, recruited from the community. Volunteers with significant illness, as determined from clinical history and comprehensive blood sampling, were excluded. The study protocol was approved by the Lothian Health Ethics Committee.

Cognitive testing.

We used Raven’s Standard Progressive Matrices,13 the Logical Memory (immediate and delayed) subtest of the Wechsler Memory Scale,14 Rey Auditory-Verbal Learning Test (AVLT),15 the Visual Reproduction (immediate and delayed) subtest of the Wechsler Memory Scale,14 the Benton Visual Retention Test (BVRT),16 the Controlled Word Association Test,15 the Digit-Symbol Substitution Test from the Wechsler Adult Intelligence Scale,17 and the National Adult Reading Test.18

MRI.

Brain imaging was performed in an Elscint Prestige MR scanner operating at 1.9 T. Structural image acquisition followed a three-view localizer and consisted of a coronal T1-weighted three-dimensional gradient-echo sequence covering the entire brain and skull (echo time = 9.254, repetition time = 28.5, tip angle = 25°, slice thickness = 1.5 mm [no interslice gap], 18-cm field of view, matrix = 180 × 180).

Image analysis.

Image analysis was carried out on Sun workstations using Analyze software (Mayo Clinic, Rochester, MN). An intensity threshold separating the brain from the meninges was imposed for semiautomated analysis. Hippocampal formation was defined as subiculum, hippocampus proper, and dentate gyrus with the alveus and fimbria. The hippocampus was measured bilaterally using manual tracing from the first slice in which it appeared until the full extent of the crus fornicis appeared. In the posterior head of the hippocampus, an arbitrary judgment of the boundary between the hippocampus and the amygdala was made from the position of the temporal horn and the alveus or by comparing a slice with subsequent slices in which the division was more obvious.19,20⇓ A line was drawn at the angle between the superior and medial surfaces of the parahippocampal gyrus21 to separate the subiculum from the parahippocampal gyrus. Intrarater error was 4 ± 0.03% for left and 4 ± 0.04% for right hippocampus volumes, within limits identified in previous studies.22,23⇓ The prefrontal lobe was measured from the slice in which the frontal pole could be distinguished from the meninges. Measurements were made using automated methods with manual tracing to separate the lobes through the interhemispheric fissure. The last slice was that before the appearance of the genu of the corpus callosum. The anterior part of the temporal lobe was measured semiautomatically. Subsequently, a line was drawn manually along the lateral fissure to the Sylvian point and then diagonally to the superior-most point of the amygdalohippocampal complex and along the superior boundary of this structure. The last slice of the temporal lobe was the same as for the hippocampus. The intracranial area was measured in the midline sagittal slice of the sagittal localizer by manually tracing round the inner table of the cranial vault, along the superior surface of the floor of the frontal fossa, and across the pituitary fossa to the dorsum sella. Tracing continued down the posterior surface of the clivus and completed by a line joining the anterior and posterior rims of the foramen magnum (figure 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Intracranial area.

Statistical analysis.

Cognitive test scores were correlated with each other and with MRI data (including brain volumes, unadjusted and adjusted for intracranial area, created by saving residuals from a linear regression using Pearson correlation coefficient). Data reduction was performed using principal components analysis. To test competing models of the association between brain volumes and cognitive functions, we used structural equation modeling (see Deary, 199324 for an example) with the EQS program.25,26⇓

Correlations.

Of the 100 participants scanned, three were excluded when unexpected pathology was discovered: two because of congenital arachnoid cysts, and one because of a pituitary adenoma. For the correlational analyses, n = 97 for all comparisons except Raven’s Matrices (n = 95), from which two subjects were excluded because of incorrect completion of the answer sheets, Logical Memory (n = 96), for which one subject’s score was invalid in this test because of interruption of the testing session, and Visual Reproduction (n = 96), for which a further subject also had a test session interrupted. All subjects were aged between 65 and 70 years, with a mean age of 67.8 years at the time of cognitive testing. Controlling for age did not affect any of the results presented here.

The means and standard deviations for the cognitive tests are shown in table 1. Cognitive tests showed significant, positive intercorrelations throughout (table 2). Correlations were higher between tests thought to test the same specific abilities (e.g., Visual Reproduction and the BVRT) than between those testing different abilities. However, correlations were significant and positive for all the tests (r = 0.26 to 0.65). The descriptive statistics for the regional brain volumes and intracranial area are presented in table 3. Regional brain volumes were correlated significantly and positively, with right and left sides of the same region more strongly correlated than different regions. Intracranial area was significantly associated with all the brain regions measured (table 4).

Table 5 shows the correlation matrix between the cognitive tests and regional brain volumes and intracranial area. This shows that standard intelligence tests, including the National Adult Reading Test, the Digit-Symbol Substitution Test, and Raven’s Matrices, and tests of visuospatial memory were significantly and positively correlated with several brain region volumes. By contrast, tests of verbal memory and verbal fluency did not correlate significantly with any brain volumes. When brain volumes were adjusted for intracranial area, only the correlation between right temporal lobe volume and visual reproduction remained significant.

Data reduction.

Because of the large amount of shared variance throughout the different brain regions and across different cognitive domains, we subjected these variables to data reduction. Principal components analysis of the right and left hippocampi, temporal lobes, and frontal lobes revealed that the first unrotated component accounted for 59% of the total variance (n = 97). Thus, individual differences in these specific brain areas reflect, to a large degree, differences in overall brain size. Principal components analysis was performed on the following cognitive tests: verbal fluency, Digit-Symbol Substitution Test, AVLT, BVRT, Logical Memory, Visual Reproduction, and Raven’s Matrices (n = 95). The first unrotated component accounted for 50% of the total variance, which represents the typical amount of test covariance accounted for by the general cognitive factor.

Structural equation modeling.

The model applied to the current data is illustrated in figure 2. The arrows represent the relationships between the measured variables and the derived latent traits and also between the two latent traits. Numbers beside arrows that connect variables are parameter weights estimated by the EQS program: squaring them indicates the variance shared by adjacent variables. Because of the high correlations between volume measurements from the right and left sides, these were reduced (via summed standardized scores) to a single value for each brain region. The three brain region volumes were hypothesized to load on a general brain volume latent trait. The eight cognitive test variables were hypothesized to load on a single general cognitive latent trait. The correlation between these two latent traits was estimated as a free parameter in the model. After this correlation was estimated, we tested the hypotheses that there were additional, significant associations between specific brain regions and cognitive functions. Ninety-three subjects had complete data for the cognitive tests and MRI variables shown in figure 2. The statistical “fit” of the model in figure 2 is described first, and then the meaning of the model is explained.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Structural equation model of the relationships among individual brain region volumes, the general brain size factor, and the individual cognitive tests and the general cognitive factor. NART = National Adult Reading Test; RSPM = Raven’s Standard Progressive Matrices; DSST = Digit-Symbol Substitution Test; BVRT = Benton Visual Retention Test; VR = Visual Reproduction; AVLT = Rey Auditory-Verbal Learning Test; LM = Logical Memory; VF = verbal fluency. All relationships are at p < 0.05.

There are several different indications of fit adequacy in EQS. The average of the off-diagonal standardized residuals was 0.047, indicating that most of the covariance was accounted for by the model. The largest two residuals were between temporal lobe volume and AVLT and Logical Memory scores and were negative in direction. The χ² for the model was 35.6 (df = 41, p = 0.71). Nonsignificant χ² values indicate good fit. EQS gives three fit indices that have values between 0 and 1, with values >0.9 indicating good fit. For the model in figure 2, these were as follows: Bentler-Bonett normal fit index = 0.91; Bentler-Bonett nonnormal fit index = 1; and comparative fit index = 1.0. All indicate a well-fitting model. All of the pathways in the model are significant and the Wald test indicated that no pathways in the model should be dropped. In summary, the model in figure 2 has comprehensively good fit indices.

The well-fitting model in figure 2 shows that individual variation in the three brain areas substantially derives from overall brain size, here derived as a latent trait.

The covariance among the cognitive tests is successfully captured by hypothesizing that this derives substantially but not wholly from a general cognitive latent trait. These two latent traits (“overall brain size” and “general cognitive function”) correlate 0.42. Note that two verbal tests (National Adult Reading Test and verbal fluency) and two verbal memory tests (AVLT and Logical Memory) have correlated residuals. This reflects the similarity of the two pairs of tests, which are indicators of “group factors” of ability. Group factors are domains of mental ability that are more specific than the general factor. The Lagrange Multiplier Test showed that there were no additional significant paths between specific cognitive tests and brain size or between specific brain region volumes and the general cognitive factor.