Genetic and Environmental Etiologies of Reading Disability: A Twin Study Michele C. LaBuda J. C. DeFries University of Colorado Boulder, Colorado
Previous twin studies of reading disability employed a comparison of concordance rates in identical and fraternal twin pairs as a test for genetic etiology. Recently, a statistically more powerful multiple regression analysis of twin data has been formulated to assess the importance of genetic factors in the development of reading difficulties. Application of this analysis to twin data from the Colorado Reading Project yields definitive evidence for a genetic etiology. Results from this study suggest that approximately 40 percent of the deficit observed in the disabled readers is due to genetic factors, 35 percent is due to environmental influences shared by members of twin pairs, and about 25 percent is the result of environmental factors unique to the individual and~or error variance. Introduction
The familial nature of reading disability has long been recognized. For example, in 1905, Thomas n o t e d that congenital word-blindness "frequently assumes a family t y p e " (Thomas 1905, p. 381). Of the seven cases described by Thomas, two were brothers and a third was a m e m b e r of a family in which the m o t h e r and six siblings had also experienced reading difficulties. Such familial transmission for reading disability has subsequently been d o c u m e n t e d in considerable detail. For recent reviews, see Finucci (1978) and DeFries, Vogler, and LaBuda (1986). This work was supported in part by a program project grant from the NICHD (HD-11681). We wish to acknowledge the invaluable contributions of staff members of the many Colorado school districts and of the families who participated in the study. Correspondence and requests for reprints should be addressed to Michele C. LaBuda, Institute for BehavioralGenetics, Campus Box 447, University of Colorado, Boulder, CO, 80309. Annals of Dyslexia, Vol. 38, 1988. Copyright ©I988 by The Orton Dyslexia Society ISSN 0474-7534
131
132
THEORETICAL SEEDS
The fact that reading disability tends to cluster within families is consistent with a genetic etiology; however, members of intact families share a variety of environmental factors which could also account for familial transmission. Twin studies, in contrast, provide more definitive evidence for the heritable nature of reading disability. Identical or monozygotic (MZ) twins are genetically identical, whereas fraternal or dizygotic (DZ) twins share 50 percent of their genes on average; thus, a greater similarity between members of MZ as opposed to DZ twin pairs may be ascribed to their greater genetic similarity, assuming, of course, that the extent to which environmental influences shared by members of MZ and DZ twin pairs is comparable. The validity of this "equal environments assumption" for cognitive measures has recently been discussed by Vandenberg (1984). Previous twin studies of reading disability have typically employed a comparison of MZ and DZ concordance rates as a test for genetic etiology. A pair is concordant if both members of the pair are affected, but discordant if only one twin is disabled. Therefore, to the extent that reading disability is heritable, a higher concordance rate would be expected for MZ twin pairs. In a recent review of previous twin studies, we summarized data from over 85 MZ and 120 DZ twin pairs in which at least one member of each pair had been diagnosed as being reading disabled (LaBuda and DeFries 1988). The sample sizes are generally small for any individual study; thus there is considerable variation in the concordance rates among studies. For example, MZ concordance rates vary from 21 percent to 100 percent whereas DZ concordance rates vary from 17 percent to 45 percent. Although this variation among studies diminishes the confidence to be placed in any individual report, in each study the MZ concordance rate exceeded that for the DZ twin pairs. Part of the variability among concordance rates in previous twin studies is almost certainly due to methodological differences, including differing diagnostic criteria. Obviously, the choice of different cut-off points on a continuous measure, such as reading performance, to diagnose reading disability will yield different concordance rates. Because of this difficulty with concordance rate comparisons, we recently formulated a multiple regression analysis of continuous twin data that provides a highly flexible and statistically powerful test of genetic etiology (DeFries and Fulker 1985; LaBuda, DeFries, and Fulker 1986). This methodology has recently been employed to analyze data collected from a sample of twins participating in the Colorado Reading Project (CRP) and has yielded evidence of a significant genetic etiology for reading disability (DeFries, Fulker, and LaBuda 1987; LaBuda and DeFries 1988). Results of preliminary analyses suggest that 30 percent to 40 percent of the reading deficit exhibited by disabled readers in the CRP is due to heritable influences. Results obtained from subsequent analyses of CRP twin data are described in the present report, the objectives of which are threefold: (1) to describe the multiple regression analysis of twin data as a test of
GENETIC AND ENVIRONMENTAL ETIOLOGIES
133
genetic etiology; (2) to apply this analysis to data obtained from a larger and more stringently selected sample of twins tested within the CRP; and (3) to assess the environmental etiologies of reading disability.
Method
The logic underlying the multiple regression analysis of selected twin data as a test for genetic etiology is illustrated in Figure 1. The top portion of Figure I depicts a hypothetical distribution of twin data for a
Unselected ins
I
I
I I I
/f--xL DZ Cotwins
:
.
Figure 1. Hypothetical distributions for reading performance of an unselected sample of twins, monozygotic (MZ) cotwins of reading-disabled probands and dizygotic (DZ) cotwins of reading-disabled probands. When scores are standardized by subtracting the mean of the unselected population (ix) and dividing by the proband mean (P), differences among the proband and cotwin means are functions of the relative contributions of genetic (hl), shared environmental (c~), and unique environmental (e~) factors to the etiology of reading disability. (From DeFries and Fulker 1988).
134
THEORETICAL SEEDS
measure such as reading performance. When affected individuals (probands) have been identified due to extreme scores on such a dimension, the distributions of scores for both MZ and DZ cotwins of probands would be expected to regress back toward the mean of the unselected population. However, to the extent that heritable influences contribute to the reading deficit manifested by the probands, the distributions of MZ and DZ cotwin scores should regress differentially to the unselected population mean. Because MZ twins are genetically identical, whereas DZ twins share approximately half their genes, the distribution of DZ cotwin scores should regress further toward the mean of the unselected population than would that of the MZ cotwin scores. Therefore, when MZ and DZ proband means are equal, a t-test of the difference between MZ and DZ cotwin scores would provide a test of genetic etiology. The multiple regression analysis of twin data, however, provides a more general test which is also more flexible and statistically more powerful. The multiple regression model for the analysis of selected twin data to detect genetic etiology is as follows: C = blP + b2R + A,
where C, the cotwin's score, is predicted from P, the proband's score, and R, the coefficient of relationship (1.0 for MZ twins and 0.5 for DZ twins). A is the regression constant. The partial regression of cotwin's score on proband's score, bx, is a measure of average twin resemblance whereas b2, the partial regression of cotwin's score on the coefficient of relationship, is equivalent to twice the difference between MZ and DZ cotwin means after covariance adjustment for the difference between MZ and DZ probands. The significance of b2, therefore, provides a direct statistical test for genetic etiology. Furthermore, when scores are standardized by subtracting the mean of the unselected population and dividing by the proband mean, b2 provides a measure of the contribution of heritable influences (h2) to the performance deficit of the probands. The reading deficit exhibited by probands, of course, may be due, at least in part, to environmental influences. Environmental effects may be partitioned into those shared by members of a twin pair (e.g., gestational age) and those which are unique to the individual (e.g., brain injury to one member of the twin pair during the birth process). Just as the difference between the standardized MZ and DZ cotwin means may be used to estimate the extent to which the reading deficit of probands is due to heritable influences, differences between means may be used to estimate the relative contributions of environmental factors to the etiology of reading disability. As illustrated in Figure 1, the difference between the average proband score (P) and that of MZ cotwins (CMz) is due to environmental factors unique to the individual (e2) since members of MZ twins pairs share all hereditary influences and those environmental influences common to members of twin pairs. In contrast, the difference between the DZ proband and cotwin means is due to both heritable (I/2 h 2) and unique
GENETIC AND ENVIRONMENTAL ETIOLOGIES
135
environmental influences (e~). DZ cotwins deviate from the mean of the unselected population (p,) to the extent that both genetic (1/2 h~) and shared environmental influences (c2) are important (see DeFries and Fulker [1988] for derivation of these expected deviations). Thus, these 2 yield further insight into the etiology of three parameters, h 2, c~, and eg, reading disability. Estimates of these parameters, along with their associated standard errors, were obtained in the present study of minimization of a weighted least squares function using the generalized numerical optimization package MINUIT (CERN 1977).
Subjects Twin pairs were identified through cooperating school districts within a 150 mile radius of Denver, Colorado. Parents of twins were contacted in order to obtain permission to access the child's school records. Records were screened for any indication of reading difficulties, such as low reading performance, referral to resource rooms for special instruction, etc. When at least one member of a twin pair exhibited a positive school history for reading problems, both children were administered an extensive psychometric test battery which includes measures of general reading and cognitive ability as well as more specific measures of reading and language processes. In an independent sample of 140 non-twin reading-disabled subjects and 140 controls, approximately 95 percent of the subjects were correctly classified by a discriminant function analysis based upon data from six tests also administered in the twin study: the Reading Recognition, Reading Comprehension, and Spelling subtests of the Peabody Individual Achievement Tests (Dunn and Markwardt 1970); the Coding and Digit Span subtests of either the Wechsler Intelligence Scale for Children--Revised (Wechsler 1974) or the Wechsler Adult Intelligence Scale--Revised (Wechsler 1981); and the Colorado Perceptual Speed Test (DeFries et al. 1981) correctly classified approximately 95 percent of the subjects. These discriminant function weights were also applied to the twin data to compute composite scores. In the present study, diagnostic criteria included evidence of reading problems in school, a discriminant score in the disabled range, Verbal or Performance IQ score of at least 90, no known history of emotional, behavioral, or neurological problems, and no uncorrected auditory or visual acuity deficits. If only one member of a twin pair was diagnosed as reading disabled, the subject was obviously the proband. If, on the other hand, both members of the twin pair met the criteria for reading disability, the twin with the lower discriminant score was designated the proband. Results in the present report are based upon data from 75 MZ and 59 DZ twin pairs in which at least one member of each pair was diagnosed as reading disabled. Control children were also tested and were matched to the disabled subjects, whenever possible, on the basis of age, gender, and zygosity in order to provide a comparable group of normal-reading peers.
136
THEORETICAL SEEDS
Zygosity determination was conducted t h r o u g h the use of selected items from the Nichols and Bilbro (1966) questionnaire. The questionnaire has a reported accuracy of 95 percent; however, in doubtful cases, blood samples were drawn and analyzed by the Minneapolis War Memorial Blood Bank. The average age of the twins in the sample was 12.7 and all twins came from English-speaking, middle-class homes.
Results
The average scores of MZ and DZ probands and cotwins for the discriminant function composite, as well as the individual tests which comprise the composite, are presented in Table I. The scores have been standardized based u p o n the control sample means and standard deviations in order to facilitate comparisons a m o n g tests. For each measure, the reading-disabled probands score significantly lower than the controls. Significant twin resemblance occurs for all measures, as indicated in Table I by the highly significant bl coefficients estimated by the multiple regression model. Multiple regression analysis as a test for genetic etiology is most appropriate for the variable u p o n which selection occurred (i.e., the discriminant score). As may be seen in Table I, the MZ and DZ proband means are quite comparable, but there is a substantial difference between the means of the MZ and DZ cotwins. The difference between these means, after covariance adjustment for the slight difference between MZ and DZ probands, is highly significant as indicated by the b2 coefficient and provides compelling evidence of a genetic etiology for reading disability. With regard to the individual tests, the difference between the means of DZ probands and cotwins is consistently greater than Table I Mean Standardized Test Scores and Partial Regression Coefficients Descriptive Statistics Monozygotic Measure
Dizygotic
Proband Cotwin Proband Cotwin
Discriminant Reading Recognition Reading Comprehension Spelling Coding Colorado Perceptual Speed Digit Span *p < .05 (one-tailed).
Partial Regression Coefficients
bl
b2 -1.11 ± .26*
- 2.89
- 2.20
- 2.96
- 1.70
.78 ± .08*
- 2.43
- 1.83
- 2.54
- 1.42
.68 +
-2.01 -2.01 -0.65
- 1.48 - 1.92 - 1.47 -2.01 -0.71 -0.99
- 1.18 -0.98
- 1.13 -0.79
- 1.24 -1.22
-.96
-+ .27*
-1.18 .51 -+ .09* .55 +- .09* -0.75 .60 + .08*
-.51
+ .29*
-.92
-+ .29*
-.34
--- .30
-0.98 .52 ± .08* -0.71 .49 +- .09*
-.38
± .26
-.40
± .31
.08*
- 1.01
GENETIC AND ENVIRONMENTAL ETIOLOGIES
137
that between MZ probands and cotwins. The corresponding b2 coefficients are significant for Reading Recognition, Reading Comprehension, and Spelling, indicating significant genetic covariance with the discriminant score. Subsequent to the multiple regression analysis, a weighted least squares procedure was employed in order to estimate the contributions of genetic, shared environmental, and unique environmental factors to the etiology of reading disability. As may be discerned from Figure 1, the three parameters may be estimated by equating the standardized proband and cotwin means to their expected values: (1) the deviation of the proband mean from that of the unselected population includes h 2 + c2 + e~; (2) the deviation of the average MZ cotwin score from the population mean contains h~ + c2; and (3) the deviation of the DZ cotwin mean from the population mean equals .5 h~ + c2. For this analysis, an estimate of the population mean was obtained by using a weighted average of the control and proband means (95 percent and 5 percent, respectively), and the inverse of the variance of the mean was employed as a weight in each case. The estimates of h i, c~, and e 2 obtained from an analysis of the discriminant function score data are .394 -+ .125, .354 + .105, and .252 --.046, respectively; therefore, approximately 40 percent of the deficit of the reading-disabled probands may be attributed to genetic influences, 35 percent to those environmental factors shared by members of a twin pair, and 25 percent to a combination of environmental influences unique to the individual and error variance.
Discussion
The study of genetic and environmental etiologies of reading disability is important for a variety of reasons. As recently stated in a position paper issued by the National Joint Committee on Learning Disabilities (1981), "An understanding of etiological mechanisms (a) facilitates a determination of prognosis, (b) provides information to individuals and their families that helps to clarify their understanding of the manifest disorder(s), and (c) provides direction for research studies that will influence educational practice." Furthermore, results of such studies could yield improved risk estimates, earlier diagnosis, and more effective remediation. The multiple regression analysis of selected twin data is simple, flexible, and yields a statistically powerful test for genetic etiology. Moreover, the extent to which the observed reading deficit of probands is due to environmental effects common to twin pairs, as well as those unique to the individual, may also be assessed. In addition to the basic test for genetic etiology, the multiple regression model outlined in the present report may be extended in a variety of
138
THEORETICAL SEEDS
ways to test more searching hypotheses. For example, if data from individuals of two ostensibly different subtypes were available for analysis, the addition of two independent variables (a dumm~¢ variate, indicative of subtype classification, and the product of this variable and the coefficient of relationship) would provide a test of differential genetic etiology. Results from such an analysis could provide important evidence for the validity of alternative typologies. In a similar manner, a model could be formulated to test for differential genetic etiology as a function of gender, socioeconomic status, age, or any other variable of interest. Such analyses will be undertaken w h e n the sample of twins tested in the CRP becomes larger.
References CERN. 1977. MINUIT: A system for function minimization and analysis of the parameter errors and correlations. Geneva: CERN. DeFries, J. C. and Fulker, D. W. 1985. Multiple regression analysis of twin data. Behavior Genetics 15:467-473. DeFries, J. C. and Fulker, D. W. 1988. Multiple regression analysis of twin data: Etiology of deviant scores versus individual differences. Acta Geneticae Medicae et Gemellologiae. In press. DeFries, J. C., Fulker, D. W., and LaBuda, M. C. 1987. Evidence for a genetic aetiology in reading disability of twins. Nature 329:537-539. DeFries, J. C., Plomin, R., Vandenberg, S. G., and Kuse, A. R. 1981. Parent-offspring resemblance for cognitive abilities in the Colorado Adoption Project: Biological, adoptive, and control parents and one-year-old children. Intelligence 5:245-277. DeFries, J. C., Vogler, G. P., and LaBuda, M. C. 1986. Colorado Family Reading Study: An overview. In J. L. Fuller and E. C. Simmel (eds.). Behavior Genetics: Principles and applications IL Hillsdale NJ: Lawrence Erlbaum Associates. Dunn, L. M. and Markwardt, F. C. 1970. Examiner's Manual: Peabody Individual Achievement Test. Circle Pines, MN: American Guidance Service. Finucci, J. M. 1978. Genetic considerations in dyslexia. In H. R. Myklebust (ed.). Progress in Learning Disabilities, Vol. IV. New York: Grune and Stratton. LaBuda, M. C. and DeFries, J. C. 1988. Genetic etiology of reading disability: Evidence from a twin study. In G. Th. Pavlidis (ed.). Dyslexia: A neuropsychological and learning perspective. New York: John Wiley and Sons. In press. LaBuda, M. C., DeFries, J. C., and Fulker, D. W. 1986. Multiple regression analysis of twin data obtained from selected samples. Genetic Epidemiology 3:425-433. National Joint Committee on Learning Disabilities. 1981. Learning Disabilities: Issues on definition (A position paper of the National Joint Committee on Learning Disabilities, January 30, 1981). {Reprinted in Journal of Learning Disabilities 1987 20:107-108.] Nichols, R. C., and Bilbro, W. C. 1966. The diagnosis of twin zygosity. Acta Genetica 16:265275. Thomas, C. J. 1905. Congenital "word-blindness" and its treatment. The Ophthalmoscope 3: 380-385. Wechsler, D. I. 1981. Examiner's Manual: Wechsler Adult Intelligence Scale--Revised. New York: Psychological Corporation. Wechsler, D. I. 1974. Examiner's Manual: Wechsler Intelligence Scale for Children--Revised. New York: Psychological Corporation. Vandenberg, S. G. 1984. Does a special twin situation contribute to similarity for abilities in MZ and DZ twins? Acta Geneticae Medicae et Gemellologiae 33:219-222.
Previous twin studies of reading disability employed a comparison of concordance rates in identical and fraternal twin pairs as a test for genetic etiology. Recently, a statistically more powerful multiple regression analysis of twin data has been formulated to assess the importance of genetic factors in the development of reading difficulties. Application of this analysis to twin data from the Colorado Reading Project yields definitive evidence for a genetic etiology. Results from this study suggest that approximately 40 percent of the deficit observed in the disabled readers is due to genetic factors, 35 percent is due to environmental influences shared by members of twin pairs, and about 25 percent is the result of environmental factors unique to the individual and~or error variance. Introduction
The familial nature of reading disability has long been recognized. For example, in 1905, Thomas n o t e d that congenital word-blindness "frequently assumes a family t y p e " (Thomas 1905, p. 381). Of the seven cases described by Thomas, two were brothers and a third was a m e m b e r of a family in which the m o t h e r and six siblings had also experienced reading difficulties. Such familial transmission for reading disability has subsequently been d o c u m e n t e d in considerable detail. For recent reviews, see Finucci (1978) and DeFries, Vogler, and LaBuda (1986). This work was supported in part by a program project grant from the NICHD (HD-11681). We wish to acknowledge the invaluable contributions of staff members of the many Colorado school districts and of the families who participated in the study. Correspondence and requests for reprints should be addressed to Michele C. LaBuda, Institute for BehavioralGenetics, Campus Box 447, University of Colorado, Boulder, CO, 80309. Annals of Dyslexia, Vol. 38, 1988. Copyright ©I988 by The Orton Dyslexia Society ISSN 0474-7534
131
132
THEORETICAL SEEDS
The fact that reading disability tends to cluster within families is consistent with a genetic etiology; however, members of intact families share a variety of environmental factors which could also account for familial transmission. Twin studies, in contrast, provide more definitive evidence for the heritable nature of reading disability. Identical or monozygotic (MZ) twins are genetically identical, whereas fraternal or dizygotic (DZ) twins share 50 percent of their genes on average; thus, a greater similarity between members of MZ as opposed to DZ twin pairs may be ascribed to their greater genetic similarity, assuming, of course, that the extent to which environmental influences shared by members of MZ and DZ twin pairs is comparable. The validity of this "equal environments assumption" for cognitive measures has recently been discussed by Vandenberg (1984). Previous twin studies of reading disability have typically employed a comparison of MZ and DZ concordance rates as a test for genetic etiology. A pair is concordant if both members of the pair are affected, but discordant if only one twin is disabled. Therefore, to the extent that reading disability is heritable, a higher concordance rate would be expected for MZ twin pairs. In a recent review of previous twin studies, we summarized data from over 85 MZ and 120 DZ twin pairs in which at least one member of each pair had been diagnosed as being reading disabled (LaBuda and DeFries 1988). The sample sizes are generally small for any individual study; thus there is considerable variation in the concordance rates among studies. For example, MZ concordance rates vary from 21 percent to 100 percent whereas DZ concordance rates vary from 17 percent to 45 percent. Although this variation among studies diminishes the confidence to be placed in any individual report, in each study the MZ concordance rate exceeded that for the DZ twin pairs. Part of the variability among concordance rates in previous twin studies is almost certainly due to methodological differences, including differing diagnostic criteria. Obviously, the choice of different cut-off points on a continuous measure, such as reading performance, to diagnose reading disability will yield different concordance rates. Because of this difficulty with concordance rate comparisons, we recently formulated a multiple regression analysis of continuous twin data that provides a highly flexible and statistically powerful test of genetic etiology (DeFries and Fulker 1985; LaBuda, DeFries, and Fulker 1986). This methodology has recently been employed to analyze data collected from a sample of twins participating in the Colorado Reading Project (CRP) and has yielded evidence of a significant genetic etiology for reading disability (DeFries, Fulker, and LaBuda 1987; LaBuda and DeFries 1988). Results of preliminary analyses suggest that 30 percent to 40 percent of the reading deficit exhibited by disabled readers in the CRP is due to heritable influences. Results obtained from subsequent analyses of CRP twin data are described in the present report, the objectives of which are threefold: (1) to describe the multiple regression analysis of twin data as a test of
GENETIC AND ENVIRONMENTAL ETIOLOGIES
133
genetic etiology; (2) to apply this analysis to data obtained from a larger and more stringently selected sample of twins tested within the CRP; and (3) to assess the environmental etiologies of reading disability.
Method
The logic underlying the multiple regression analysis of selected twin data as a test for genetic etiology is illustrated in Figure 1. The top portion of Figure I depicts a hypothetical distribution of twin data for a
Unselected ins
I
I
I I I
/f--xL DZ Cotwins
:
.
Figure 1. Hypothetical distributions for reading performance of an unselected sample of twins, monozygotic (MZ) cotwins of reading-disabled probands and dizygotic (DZ) cotwins of reading-disabled probands. When scores are standardized by subtracting the mean of the unselected population (ix) and dividing by the proband mean (P), differences among the proband and cotwin means are functions of the relative contributions of genetic (hl), shared environmental (c~), and unique environmental (e~) factors to the etiology of reading disability. (From DeFries and Fulker 1988).
134
THEORETICAL SEEDS
measure such as reading performance. When affected individuals (probands) have been identified due to extreme scores on such a dimension, the distributions of scores for both MZ and DZ cotwins of probands would be expected to regress back toward the mean of the unselected population. However, to the extent that heritable influences contribute to the reading deficit manifested by the probands, the distributions of MZ and DZ cotwin scores should regress differentially to the unselected population mean. Because MZ twins are genetically identical, whereas DZ twins share approximately half their genes, the distribution of DZ cotwin scores should regress further toward the mean of the unselected population than would that of the MZ cotwin scores. Therefore, when MZ and DZ proband means are equal, a t-test of the difference between MZ and DZ cotwin scores would provide a test of genetic etiology. The multiple regression analysis of twin data, however, provides a more general test which is also more flexible and statistically more powerful. The multiple regression model for the analysis of selected twin data to detect genetic etiology is as follows: C = blP + b2R + A,
where C, the cotwin's score, is predicted from P, the proband's score, and R, the coefficient of relationship (1.0 for MZ twins and 0.5 for DZ twins). A is the regression constant. The partial regression of cotwin's score on proband's score, bx, is a measure of average twin resemblance whereas b2, the partial regression of cotwin's score on the coefficient of relationship, is equivalent to twice the difference between MZ and DZ cotwin means after covariance adjustment for the difference between MZ and DZ probands. The significance of b2, therefore, provides a direct statistical test for genetic etiology. Furthermore, when scores are standardized by subtracting the mean of the unselected population and dividing by the proband mean, b2 provides a measure of the contribution of heritable influences (h2) to the performance deficit of the probands. The reading deficit exhibited by probands, of course, may be due, at least in part, to environmental influences. Environmental effects may be partitioned into those shared by members of a twin pair (e.g., gestational age) and those which are unique to the individual (e.g., brain injury to one member of the twin pair during the birth process). Just as the difference between the standardized MZ and DZ cotwin means may be used to estimate the extent to which the reading deficit of probands is due to heritable influences, differences between means may be used to estimate the relative contributions of environmental factors to the etiology of reading disability. As illustrated in Figure 1, the difference between the average proband score (P) and that of MZ cotwins (CMz) is due to environmental factors unique to the individual (e2) since members of MZ twins pairs share all hereditary influences and those environmental influences common to members of twin pairs. In contrast, the difference between the DZ proband and cotwin means is due to both heritable (I/2 h 2) and unique
GENETIC AND ENVIRONMENTAL ETIOLOGIES
135
environmental influences (e~). DZ cotwins deviate from the mean of the unselected population (p,) to the extent that both genetic (1/2 h~) and shared environmental influences (c2) are important (see DeFries and Fulker [1988] for derivation of these expected deviations). Thus, these 2 yield further insight into the etiology of three parameters, h 2, c~, and eg, reading disability. Estimates of these parameters, along with their associated standard errors, were obtained in the present study of minimization of a weighted least squares function using the generalized numerical optimization package MINUIT (CERN 1977).
Subjects Twin pairs were identified through cooperating school districts within a 150 mile radius of Denver, Colorado. Parents of twins were contacted in order to obtain permission to access the child's school records. Records were screened for any indication of reading difficulties, such as low reading performance, referral to resource rooms for special instruction, etc. When at least one member of a twin pair exhibited a positive school history for reading problems, both children were administered an extensive psychometric test battery which includes measures of general reading and cognitive ability as well as more specific measures of reading and language processes. In an independent sample of 140 non-twin reading-disabled subjects and 140 controls, approximately 95 percent of the subjects were correctly classified by a discriminant function analysis based upon data from six tests also administered in the twin study: the Reading Recognition, Reading Comprehension, and Spelling subtests of the Peabody Individual Achievement Tests (Dunn and Markwardt 1970); the Coding and Digit Span subtests of either the Wechsler Intelligence Scale for Children--Revised (Wechsler 1974) or the Wechsler Adult Intelligence Scale--Revised (Wechsler 1981); and the Colorado Perceptual Speed Test (DeFries et al. 1981) correctly classified approximately 95 percent of the subjects. These discriminant function weights were also applied to the twin data to compute composite scores. In the present study, diagnostic criteria included evidence of reading problems in school, a discriminant score in the disabled range, Verbal or Performance IQ score of at least 90, no known history of emotional, behavioral, or neurological problems, and no uncorrected auditory or visual acuity deficits. If only one member of a twin pair was diagnosed as reading disabled, the subject was obviously the proband. If, on the other hand, both members of the twin pair met the criteria for reading disability, the twin with the lower discriminant score was designated the proband. Results in the present report are based upon data from 75 MZ and 59 DZ twin pairs in which at least one member of each pair was diagnosed as reading disabled. Control children were also tested and were matched to the disabled subjects, whenever possible, on the basis of age, gender, and zygosity in order to provide a comparable group of normal-reading peers.
136
THEORETICAL SEEDS
Zygosity determination was conducted t h r o u g h the use of selected items from the Nichols and Bilbro (1966) questionnaire. The questionnaire has a reported accuracy of 95 percent; however, in doubtful cases, blood samples were drawn and analyzed by the Minneapolis War Memorial Blood Bank. The average age of the twins in the sample was 12.7 and all twins came from English-speaking, middle-class homes.
Results
The average scores of MZ and DZ probands and cotwins for the discriminant function composite, as well as the individual tests which comprise the composite, are presented in Table I. The scores have been standardized based u p o n the control sample means and standard deviations in order to facilitate comparisons a m o n g tests. For each measure, the reading-disabled probands score significantly lower than the controls. Significant twin resemblance occurs for all measures, as indicated in Table I by the highly significant bl coefficients estimated by the multiple regression model. Multiple regression analysis as a test for genetic etiology is most appropriate for the variable u p o n which selection occurred (i.e., the discriminant score). As may be seen in Table I, the MZ and DZ proband means are quite comparable, but there is a substantial difference between the means of the MZ and DZ cotwins. The difference between these means, after covariance adjustment for the slight difference between MZ and DZ probands, is highly significant as indicated by the b2 coefficient and provides compelling evidence of a genetic etiology for reading disability. With regard to the individual tests, the difference between the means of DZ probands and cotwins is consistently greater than Table I Mean Standardized Test Scores and Partial Regression Coefficients Descriptive Statistics Monozygotic Measure
Dizygotic
Proband Cotwin Proband Cotwin
Discriminant Reading Recognition Reading Comprehension Spelling Coding Colorado Perceptual Speed Digit Span *p < .05 (one-tailed).
Partial Regression Coefficients
bl
b2 -1.11 ± .26*
- 2.89
- 2.20
- 2.96
- 1.70
.78 ± .08*
- 2.43
- 1.83
- 2.54
- 1.42
.68 +
-2.01 -2.01 -0.65
- 1.48 - 1.92 - 1.47 -2.01 -0.71 -0.99
- 1.18 -0.98
- 1.13 -0.79
- 1.24 -1.22
-.96
-+ .27*
-1.18 .51 -+ .09* .55 +- .09* -0.75 .60 + .08*
-.51
+ .29*
-.92
-+ .29*
-.34
--- .30
-0.98 .52 ± .08* -0.71 .49 +- .09*
-.38
± .26
-.40
± .31
.08*
- 1.01
GENETIC AND ENVIRONMENTAL ETIOLOGIES
137
that between MZ probands and cotwins. The corresponding b2 coefficients are significant for Reading Recognition, Reading Comprehension, and Spelling, indicating significant genetic covariance with the discriminant score. Subsequent to the multiple regression analysis, a weighted least squares procedure was employed in order to estimate the contributions of genetic, shared environmental, and unique environmental factors to the etiology of reading disability. As may be discerned from Figure 1, the three parameters may be estimated by equating the standardized proband and cotwin means to their expected values: (1) the deviation of the proband mean from that of the unselected population includes h 2 + c2 + e~; (2) the deviation of the average MZ cotwin score from the population mean contains h~ + c2; and (3) the deviation of the DZ cotwin mean from the population mean equals .5 h~ + c2. For this analysis, an estimate of the population mean was obtained by using a weighted average of the control and proband means (95 percent and 5 percent, respectively), and the inverse of the variance of the mean was employed as a weight in each case. The estimates of h i, c~, and e 2 obtained from an analysis of the discriminant function score data are .394 -+ .125, .354 + .105, and .252 --.046, respectively; therefore, approximately 40 percent of the deficit of the reading-disabled probands may be attributed to genetic influences, 35 percent to those environmental factors shared by members of a twin pair, and 25 percent to a combination of environmental influences unique to the individual and error variance.
Discussion
The study of genetic and environmental etiologies of reading disability is important for a variety of reasons. As recently stated in a position paper issued by the National Joint Committee on Learning Disabilities (1981), "An understanding of etiological mechanisms (a) facilitates a determination of prognosis, (b) provides information to individuals and their families that helps to clarify their understanding of the manifest disorder(s), and (c) provides direction for research studies that will influence educational practice." Furthermore, results of such studies could yield improved risk estimates, earlier diagnosis, and more effective remediation. The multiple regression analysis of selected twin data is simple, flexible, and yields a statistically powerful test for genetic etiology. Moreover, the extent to which the observed reading deficit of probands is due to environmental effects common to twin pairs, as well as those unique to the individual, may also be assessed. In addition to the basic test for genetic etiology, the multiple regression model outlined in the present report may be extended in a variety of
138
THEORETICAL SEEDS
ways to test more searching hypotheses. For example, if data from individuals of two ostensibly different subtypes were available for analysis, the addition of two independent variables (a dumm~¢ variate, indicative of subtype classification, and the product of this variable and the coefficient of relationship) would provide a test of differential genetic etiology. Results from such an analysis could provide important evidence for the validity of alternative typologies. In a similar manner, a model could be formulated to test for differential genetic etiology as a function of gender, socioeconomic status, age, or any other variable of interest. Such analyses will be undertaken w h e n the sample of twins tested in the CRP becomes larger.
References CERN. 1977. MINUIT: A system for function minimization and analysis of the parameter errors and correlations. Geneva: CERN. DeFries, J. C. and Fulker, D. W. 1985. Multiple regression analysis of twin data. Behavior Genetics 15:467-473. DeFries, J. C. and Fulker, D. W. 1988. Multiple regression analysis of twin data: Etiology of deviant scores versus individual differences. Acta Geneticae Medicae et Gemellologiae. In press. DeFries, J. C., Fulker, D. W., and LaBuda, M. C. 1987. Evidence for a genetic aetiology in reading disability of twins. Nature 329:537-539. DeFries, J. C., Plomin, R., Vandenberg, S. G., and Kuse, A. R. 1981. Parent-offspring resemblance for cognitive abilities in the Colorado Adoption Project: Biological, adoptive, and control parents and one-year-old children. Intelligence 5:245-277. DeFries, J. C., Vogler, G. P., and LaBuda, M. C. 1986. Colorado Family Reading Study: An overview. In J. L. Fuller and E. C. Simmel (eds.). Behavior Genetics: Principles and applications IL Hillsdale NJ: Lawrence Erlbaum Associates. Dunn, L. M. and Markwardt, F. C. 1970. Examiner's Manual: Peabody Individual Achievement Test. Circle Pines, MN: American Guidance Service. Finucci, J. M. 1978. Genetic considerations in dyslexia. In H. R. Myklebust (ed.). Progress in Learning Disabilities, Vol. IV. New York: Grune and Stratton. LaBuda, M. C. and DeFries, J. C. 1988. Genetic etiology of reading disability: Evidence from a twin study. In G. Th. Pavlidis (ed.). Dyslexia: A neuropsychological and learning perspective. New York: John Wiley and Sons. In press. LaBuda, M. C., DeFries, J. C., and Fulker, D. W. 1986. Multiple regression analysis of twin data obtained from selected samples. Genetic Epidemiology 3:425-433. National Joint Committee on Learning Disabilities. 1981. Learning Disabilities: Issues on definition (A position paper of the National Joint Committee on Learning Disabilities, January 30, 1981). {Reprinted in Journal of Learning Disabilities 1987 20:107-108.] Nichols, R. C., and Bilbro, W. C. 1966. The diagnosis of twin zygosity. Acta Genetica 16:265275. Thomas, C. J. 1905. Congenital "word-blindness" and its treatment. The Ophthalmoscope 3: 380-385. Wechsler, D. I. 1981. Examiner's Manual: Wechsler Adult Intelligence Scale--Revised. New York: Psychological Corporation. Wechsler, D. I. 1974. Examiner's Manual: Wechsler Intelligence Scale for Children--Revised. New York: Psychological Corporation. Vandenberg, S. G. 1984. Does a special twin situation contribute to similarity for abilities in MZ and DZ twins? Acta Geneticae Medicae et Gemellologiae 33:219-222.
