Intraclass correlations (adjusted for age and education) for monozygotic (MZ) and dizygotic (DZ) twin pairs on 4 components of verbal learning and memory.
Estimates of percentage of variance due to genetic, shared environmental, and nonshared environmental influences as derived from the best-fitting maximum likelihood model of twin similarity on specific components of verbal learning and memory.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Swan GE, Reed T, Jack LM, et al. Differential Genetic Influence for Components of Memory in Aging Adult Twins. Arch Neurol. 1999;56(9):1127–1132. doi:10.1001/archneur.56.9.1127
To investigate the relative proportion of genetic and environmental contributions to verbal memory in community-dwelling World War II veteran twins.
The California Verbal Learning Test (CVLT) was administered to 94 monozygotic (MZ) and 89 dizygotic (DZ) elderly male twin pair participants in the fourth examination of the National Heart, Lung, and Blood Institute Twin Study.
Subjects voluntarily participated on an outpatient basis at a research or medical center facility in 1 of 4 sites in the United States.
Subjects had a mean age of 71.8 years (SD, 2.9 years), a mean educational level of 13.6 years (SD, 2.8 years), and no history of stroke and/or a Mini-Mental State Examination score of 23 or greater.
Main Outcome Measures
Twin pair similarity in performance on 4 factor analytically derived components of the CVLT measuring verbal learning and memory, response discrimination, learning strategy, and recognition memory.
The MZ intraclass correlation was significantly larger than the DZ correlation for verbal learning and memory (I<.001) but not for the other 3 components of memory. Using maximum likelihood methods, the best-fitting genetic model indicated that verbal learning and memory has a substantial genetic component (56% of total variance), whereas response discrimination has a much smaller, although still detectable, genetic component (24% of total variance). There is no evidence of genetic influence on learning strategy or recognition memory.
Differential contribution of genetic and environmental influences to specific components of memory suggest that, in this group of elderly male twin pairs, some components may be more amenable to intervention than others.
WITH THE enormous increase in the elderly population of the industrialized nations expected to occur during the next 20 to 30 years,1 the determination of genetic and environmental influences on functioning in late life will take on greater importance. The purpose of our investigation was to contribute to the further genetic refinement of one of the most important behavioral markers of cognitive health: verbal memory. In a recent study, McClearn et al2 reported evidence of genetic influence on globally assessed memory in twins aged 80 years or older. The extent to which specific components of memory, of which there are several, have detectable genetic variance requires further investigation.
The influence of genetic factors on general intelligence and its components is now well characterized in children and middle-aged adults.3-6 Previous work2,7-11 suggests that the influence of genetic factors on cognition extends to late life as well. Although environmental factors would be expected to have a greater influence on cognition in older than in younger adults, studies show that genetic factors explain about half of the variance in intelligence and neuropsychological test performance in both age groups.2,10,11
Relatively little is known, however, about the genetics of specific forms of memory in older adults.4,10 Continued genetic investigations of elderly individuals are important because they provide information on the extent and the limits of genetic influence on component cognitive processes. Knowledge of subcomponents of cognition in the elderly that are determined, in part, by genetic factors will aid in the development of a refined phenotype, a critical step toward the establishment of linkage of specific aspects of cognition with gene variants.12 Moreover, the identification of component processes that have minimal or no genetic variance is important, because these processes may well be more amenable to environmental intervention in the aging adult.
Our analysis focuses on twin pair similarity in performance on the California Verbal Learning Test (CVLT),13 a relatively new neuropsychological test based on methods and constructs of cognitive science.13,14 A recent meta-analysis of the literature found the CVLT to be among the most sensitive of all tests to cognitive deficits associated with abnormal aging.15 The CVLT not only assesses, through recall and recognition trials, the amount of verbal information learned, but also provides various indices of how information is learned by characterizing the strategies used and the types of errors made.
Our data were collected as part of the fourth examination follow-up of the National Heart, Lung, and Blood Institute Twin Study, a longitudinal investigation of the genetic and environmental determinants of cardiovascular and cerebrovascular disease risk factors.16-18 Twins who agreed to participate in the fourth examination were given detailed explanations of the procedures and were asked to sign informed consent forms before any testing.
To assess bias due to nonparticipation in the follow-up examination, we compared nonparticipants with participants in the 1995-1997 follow-up on demographics and blood pressure (BP) history category during examinations 1 through 3. Nonparticipants were significantly older at examination 3 (63.7 vs 62.3 years; P=.007) and had higher average BP during the first 3 examinations (systolic BP, 133.7 vs 129.1 mm Hg; P<.001; diastolic BP, 82.3 vs 80.9 mm Hg; P=.04). These 2 subgroups did not differ on education level or cognitive performance at the third examination.
The CVLT is a multiple-trial recall and recognition word-list learning test. The subject is read a list of 16 nouns (list A) at 1-second intervals in fixed order for 5 learning trials. After each trial, the subject is asked to recall as many words as possible in any order (ie, free recall). The words on list A are drawn from 4 semantic categories (tools, fruits, clothing, and spices and herbs), with no consecutive words from the same category. After the fifth trial, an interference list of 16 new nouns (list B) is presented, which shares 2 categories from list A and has 2 unshared categories. On cued recall trials, subjects must recall the list A words as the examiner specifies each category in turn. Potential encoding deficits are indicated by differences in the number of words recalled under free and cued recall conditions. Free and cued recall of list A are tested immediately (short delay) and again after 20 minutes (long delay). Retention is measured through differences in short and long delay. The CVLT ends with a recognition and response discrimination task. As each word on a 44-word list is read aloud, the subject must indicate whether it is a target word (from list A) or a distractor. Some distractors share semantic categories with the target words, and others sound alike. Both word lists on the CVLT are introduced as shopping lists because of their similarity to everyday activities.14 We used a total of 27 CVLT subscores that reflect specific types of verbal memory capabilities in our analyses.
A total of 492 individual twins were given the CVLT. Of this number, 366 were members of complete pairs (94 monozygotic [MZ] pairs, 89 dizygotic [DZ] pairs). Twin pairs were seen on the same day and given the CVLT by separate, experienced examiners. The CVLT was administered in accordance with published instructions, and scoring was completed using the CVLT scoring software.19
Fifty individuals of the sample of complete pairs were identified on the basis of medical history as having had a stroke or a score of less than 23 on the Mini-Mental State Examination.20 Of these, 18 were MZ twins (7 pairs, 4 singletons) and 32 were DZ twins (13 pairs, 6 singletons). The incidence of stroke was 5.9% in MZ and 6.3% in DZ twins (χ21=0.04; P=.84). The prevalence of low scores on the Mini-Mental State Examination was 4.7% in MZ and 7.5% in DZ twins (χ21=1.72; P=.19). Compared with individuals free of stroke and with scores of 23 or greater on the Mini-Mental State Examination at the time of assessment, those with stroke or impairment were older (73.1 vs 71.8 years; t491=−2.87; P=.004), had fewer years of education (12.3 vs 13.6 years; t491=3.10; P=.002), and scored significantly lower on 19 of the 27 CVLT measures of learning and memory (Table 1). Of the 8 subscales on which no differences were observed, 4 consisted of perseveration errors during various recall trials, 2 consisted of learning strategies (semantic or serial clustering), and 2 consisted of unrelated indices (intrusions during long-delay free recall and response bias during the recognition trial). In the remaining sample of twin pairs available for analysis (n=298; 84 MZ pairs and 65 DZ pairs), the prevalence of health conditions (defined on the basis of self-report that was verified by medical record review) was as follows: diabetes, 10.1%; hypertension, 47.6%; and coronary heart disease, 33.1%.
Because of the high intercorrelations among the CVLT measures, we used a principal components analysis to reduce redundancy.21 The application of principal components analysis to the 27×27 covariance matrix resulted in 8 orthogonal (ie, independent) factors with eigenvalues greater than 1.0. After varimax rotation, these 8 factors accounted for 76.0% of the total variance. Application of the scree test (a visual inspection of the plot of percentage of total variance accounted for by each factor) suggested the retention of the first 4 factors, accounting collectively for 57% of the variance in the CVLT (Table 2). This factor structure is consistent with previously published multivariate analyses of the CVLT.22-24
Standardized scoring coefficients from each of the 3 factors were then used to generate the following 4 composite standardized scores (mean, 0; SD, 1) for each individual included in this analysis: verbal learning and memory, response discrimination, learning strategy, and recognition.
The association between each of the 4 memory components with age and education tended to be small although, in several instances, significant. Verbal learning and memory were positively associated with education (r=0.19; P<.001). Response discrimination was positively associated with age (r=0.12; P=.03), whereas recognition was negatively associated with education (r=−0.18; P=.002). Although the magnitudes of these correlations were generally small, we believed it appropriate to residualize each of the component scores for age and education before genetic analysis.
Genetic model fitting was performed on the MZ and DZ variance-covariance matrices calculated for each of the 4 CVLT factors. A genetic model specifies the variation in phenotype to be due to genotype and environmental influences. Sources of variation considered in biometric genetic analyses include additive genetic variation due to the sum of effects of individual alleles at all loci, dominance genetic variation due to the interaction of alleles at a given locus and between loci, shared familial environmental effects, and random individual environmental variation that is not shared by family members. The relative contribution of genetic and environmental influences to individual differences in performance on the CVLT were estimated by maximum likelihood, using the computer program TWINAN92 (Version 1.0; software available from Christopher Williams, PhD, Department of Mathematics and Statistics, University of Idaho, Moscow, ID 83843).
Goodness of fit was assessed by likelihood-ratio χ2 tests that test the agreement between the observed and predicted variance-covariance matrices in MZ and DZ twins. A large χ2 (corresponding to a low probability) indicates a poor fit; a small χ2 (accompanied by a high P value) indicates that the data are consistent with the model. Submodels were compared using hierarchical χ2 tests, where the χ2 for a reduced model is subtracted from that of the full model. The degrees of freedom in such tests are equal to the difference between the degrees of freedom for the full and the degrees of freedom for the reduced model.25
Table 1 shows means and SD of subjects' scores on the 27 CVLT subscales included. Overall, the performance of this sample on the various CVLT measures is consistent with previously published norms for individuals of this age and education level.13,20
The first factor (30% of total CVLT variance; Table 2), verbal learning and memory, was marked by very high loadings on measures of verbal learning and memory, including list A total recall, short-delay free and cued recall, and long-delay free and cued recall (all loadings >0.85) and moderate loadings on list A in the first trial, list B, recall consistency, learning slope, and discriminability (all loadings >0.45). Individuals with low scores on this factor are likely to use inconsistent learning and recall strategies.13
The second factor (accounting for 12% of total variance), response discrimination, was marked by strong loadings on response discrimination, including the number of short- and long-delay free and cued intrusions (all loadings >0.76) and moderate loadings on 3 of the 4 measures of perseverations (long-delay cued and short-delay free and cued; all loadings >0.40). Individuals with high scores on this factor may have problems discriminating relevant from irrelevant responses.13
The third factor (accounting for 8% of total variance), learning strategy, was marked by a strong negative association with serial clustering (−0.71) and moderate negative associations with recall of the primacy and recency regions of list A (−0.49 and −0.37, respectively). This factor was positively associated with semantic clustering (0.53) and the recall of the middle region of the word list (0.76). High scorers on this factor tend to use semantic clustering, a strategy reflective of the extent to which the examinee actively imposes an organization on the list of items according to shared semantic features. These individuals also tend to recall words from the middle region of the word list, which suggests an active vs a passive learning and recall strategy.13
The fourth of the retained factors (accounting for 7% of total variance), recognition, was marked by strong positive associations with indices of recognition memory such as hits (0.76) and response bias (0.88). Individuals with high scores on this factor are more likely to have encoded the target words, although they may not have been able to retrieve them.13
No significant mean differences between MZ and DZ twins on the 4 retained factors were observed. On 3 of the 4 CVLT components, there were no differences in variance between the 2 zygosities. Dizygotic twins, however, did show significantly less variance on response discrimination than did MZ twins (DZ/MZ mean squares ratio, F128,153=0.66; P=.025).
Intraclass correlations within each zygosity for each of the 4 CVLT components adjusted for age and education are shown in Figure 1. The MZ intraclass correlations were significantly different from 0 for verbal learning and memory (r=0.56; P<.001), response discrimination (r=0.28; P=.006), and learning strategy (r=0.26; P=.008). The DZ intraclass correlation was significantly different from 0 for learning strategy only (r=0.28; P=.009). Neither the MZ nor the DZ intraclass correlation was significantly different from 0 for recognition performance. The intraclass correlation was significantly greater within MZ than DZ pairs for verbal learning and memory (t103=3.38; P<.001), but not for the other 3 CVLT components.
Figure 2 shows the residual proportion of additive genetic and environmental (shared and nonshared) variance estimated from the best-fitting maximum-likelihood model of twin variances and covariances on each of the factor scores residualized for age and education. For verbal learning and memory, the best-fitting model included additive genetic (56%) and nonshared environmental (44%) effects (χ24 goodness of fit, 6.65; P=.15). This model provided a significantly better fit to the observed data than did a purely environmental model (χ21=28.24; P<.001) as well as a model that included shared and nonshared environmental effects (χ21=13.91; P<.001). For response discrimination, the best-fitting model included a small additive genetic effect (24%) and a much larger nonshared environmental effect (76%) (χ24 goodness of fit, 13.03; P=.01). Although this model provided a significantly better fit to the data than did one that specified only nonshared environmental effects (χ21=7.12; P=.007), it did not provide a better fit than one that specified shared and nonshared environmental effects (χ21=0.70; P=.40). Thus, the interpretation of the observed source of variation as genetic is equivocal. The best-fitting model for learning strategy included only environmental effects, with 27% due to common environmental influences and 73% due to nonshared environmental factors (χ24 goodness of fit, 0.95; P=.92). This model provided a significantly better fit to the data than did one that specified only nonshared environmental factors (χ21=10.80; P<.001). In the case of recognition, no evidence of familiality was observed; the best-fitting model included only a component for unique, nonshared environmental influences (χ24 goodness of fit, 7.20; P=.12).
To our knowledge, this is the first study to evaluate the relative proportion of genetic and environmental contributions to important indices of memory in an aging population as measured by the CVLT. Whereas verbal learning, encoding, and retrieval appear to have substantial genetic underpinnings, response discrimination, learning strategy, and verbal recognition appear to be much more products of environmental influences, such as life-long familial, educational, and/or work experiences. The relative lack or complete absence of genetic influence on response discrimination, learning strategy, and verbal recognition suggests that these components of memory in old age may be subject to direct intervention without regard to genotype of the aging adult.
The conventional wisdom in developmental genetics research states that environmental influences become progressively more important and that genetic influences become correspondingly less important with the passage of time and the accumulation of disparate environmental experiences. With regard to measures of general intelligence, the opposite appears to be the case, with numerous studies showing no decrease or even an increase in the strength of the influence of genetics on cognitive abilities in older adults.2,4 Our finding for performance on general verbal learning and memory is consistent with those of previous studies showing that at least some aspects of cognition in aging adults maintain a substantial additive genetic component.11
Our results also reveal that additive genetic influences are not uniform across the various elements of verbal learning and memory assessed by the CVLT. Three factors—response discrimination, learning strategy, and recognition—appear to be influenced largely or entirely by environmental factors. In contrast to the models for response discrimination and learning strategy, the best-fitting model for recognition included only nonshared environmental influences. A similar pattern was observed in a study of primary cognitive abilities (learning, recall, reaction time, and stimulus discrimination) in children aged 6 to 12 years, leading the investigators to conclude that information-processing–based measurement of various components of ability as opposed to tests of global abilities provides clearer insight into the extent and limits of genetic influence.26 In our study, response discrimination and learning strategy were influenced to varying degrees by shared environmental factors that could include effects of common familial, school-based, or peer sources on the development of the cognitive tools necessary for effective verbal learning and memory. Recognition, on the other hand, appears to be influenced exclusively by nonshared environmental influences that, in the present case, could be represented by different marital, social, occupational, and health histories as adults in these twin participants.
A genetic factor underlying the observed results for verbal learning and memory may be the apolipoprotein E (APOE) gene and its variant, ϵ4, located on chromosome 19q13.2 and implicated in Alzheimer disease (AD) in numerous association studies.27 That the effect of genotype may extend to individual differences in verbal memory is suggested by a previous cross-sectional APOE ϵ4 carrier-noncarrier comparison of performance on the CVLT, in which carriers scored lower on 5 of the memory subscales consisting of verbal learning and memory found to have a genetic component in our analysis. No difference between carriers and noncarriers was observed on a composite recognition score similar to that found not to have a genetic component herein.28
The sex (entirely men) and ethnic composition (almost exclusively white) of this cohort places restrictions on the extent to which the reported results are generalizable to women and other ethnic groups. There is growing evidence of sex differences in the change in brain structure seen with aging.29 Similarly, ethnic variation in elevated BP such as that found in African American individuals may place them at heightened risk for cerebrovascular outcomes in late life.30 Individual differences in brain structure and/or cerebrovascular disease could influence performance on the CVLT.13
Because diagnostic procedures to define the presence of AD with a reasonable degree of certainty were not available to our study, we cannot state with unqualified confidence that our results are not due to undetected disease. We can say, however, that on the basis of medical record review, none of the individuals in this analysis had clinically defined AD. Moreover, individuals with severe cognitive impairment would have been unable to complete the neuropsychological test sequence, resulting in missing examination data. The logistical design of the study assumed a nonclinical, community-based cohort capable of, in some cases, extensive travel and overnight stay to participate in the examination. The required level of functioning would have further eliminated seriously ill individuals from participation. Finally, the exclusion of individuals on the basis of stroke history or cognitive impairment would have eliminated individuals with more subtle deficits not resulting in nonparticipation or missing data. For these reasons, we believe it to be less likely that our findings result from undetected AD. More likely, because of the bias away from seriously ill individuals inherent in community-based studies such as this, the results reported herein probably underestimate the true size of the genetic involvement in verbal learning and memory.
Although the approach of this analysis was to partition the observed variance in neuropsychological test performance into genetic and environmental compartments, recent evidence suggests that gene×environment interaction (eg, head injury in APOE ϵ4 carriers) can play a dramatic role in increasing the odds of dementing illness.31 With the further refinement of the verbal memory phenotype, it will become possible to conduct more robust linkage studies. With the identification of specific genotypes associated with specific impairments in memory, it will then be possible to identify individuals with increased vulnerability to memory impairment. The final step would then include the design of studies to identify the effect of genotype and the environment (eg, low education level, excessive alcohol use). Knowledge of susceptible individuals and risky environments should aid in the targeting of efforts to prevent or slow the progression of age-related memory impairment.
Accepted for publication December 9, 1998.
This study was supported by grant HL51429 from the National Heart, Lung, and Blood Institute, Bethesda, Md.
We thank Ruth Krasnow, Lois Abel, MA, Danene Clements, Sandy Kirkwood, Carol Miller, Ann Von Essen, Sharyn Moore, MA, and Kim McNulty for assistance with data collection and analysis.
Reprints: Gary E. Swan, PhD, Center for Health Sciences, SRI International, 333 Ravenswood Ave, Menlo Park, CA 94024 (e-mail: firstname.lastname@example.org).