Data were obtained from the Wisconsin Longitudinal Study (WLS), a random sample of 10,317 women and men who graduated from Wisconsin high schools in 1957, representing one-third of that age cohort (Hauser et al., 1998). In 1957, 75% of Wisconsin 18 year olds were high school graduates. Follow-up studies were conducted in 1975 with 9,138 (90.1%) surviving members of the original sample when they were, on average, 36 years old; in 1992 with 8,493 (87.2%) of the surviving respondents when they were in their early 50 s; and in 2004 with 7,265 (80.0%) of the surviving respondents when they were in their mid-60 s; and in 2011 with 5,969 (68.4%) of the surviving respondents when they were 71 years of age. In addition, parallel data collection procedures were conducted with one randomly selected sibling of a subset of the respondents in 1977, 1994, 2005, and 2011, with 5,823 siblings participating in one or more of these data collection points. The original respondents were members of a single cohort (high school graduates in the year 1957), but their siblings ranged from age 46 to 92 when the data reported here were collected (62.4% of the entire sample was age 71 or 72 at that time).

Although all of the original WLS participants were high school graduates, as were 93% of their siblings, WLS participants ranged in IQ score (as measured during high school) from a low of 61 (the floor of the test, described below) to a high of 145. Fifteen percent had IQ scores of 85 (1 SD below the mean) or below. This percentage is approximately the expected proportion of the population on the low end of the IQ distribution (16% of the population is expected to be 1 SD below the mean or lower). Reflecting Wisconsin’s population at the mid-20th century, the WLS sample is racially and ethnically homogeneous; 99.2% are White (84.2% of northern or central European heritage).

In 2006 and 2007, WLS collected saliva samples from participants using Oragene kits (DNA Genotek, Inc.) and a mailback protocol patterned closely on a previous Swedish study (see Rylander-Rudqvist et al., 2006). Oragene kits were selected because of their ability to be used in a mailback protocol (e.g., no need for immediate freezing) and their high average DNA yield (in our sample, median = 319 μg/mL, mean = 400 μg/mL, SD = 284 μg/mL). All participants provided informed consent under a protocol approved by the Institutional Review Board of the University of Wisconsin-Madison. 56% of WLS participants alive in 2006 provided saliva samples (n = 7044). Those who sent saliva had one-half year more schooling (13.9 years vs. 13.4 years, p < 0.001), three points higher IQ scores (103.2 vs. 99.5, p < 0.001), and higher high school rank (54.6 vs. 46.7, p < 0.001) than those who did not return saliva samples. Otherwise, they were representative of the WLS sample as a whole.

Of the 7044 saliva samples, 15 were not used for the present analysis because of ambiguous sex determination. Another 297 were not used because there was insufficient DNA for the CGG repeat assay. Therefore, the present study is based on 6732 cases whose saliva sample yielded sufficient DNA for the CGG repeat assay, of whom 3263 (48.5%) were males and 3469 (51.5%) were females. These cases included 1311 sibling pairs.

The number of FMR1 CGG repeats was determined for all biological samples using a PCR-based protocol that incorporated reagents developed and manufactured by Celera Corporation. The specific procedures we used were described previously (Seltzer et al., 2012). The protocol combined gene-specific primers that flank the CGG repeat region of the FMR1 gene with gender-specific primers, a polymerase mixture, and a reaction buffer that is optimized for amplification of GC-rich DNA. Besides CGG repeat data, this assay also detects the presence of X and Y chromosomes within a sample, enabling sex confirmation and identifying female samples with a single detectable CGG repeat (apparent homozygosity). Data were analyzed using GeneMapper^®; v. 4.0 (Applied Biosystems). CGG triplet repeats were calculated using the following formula: number of CGG repeats = (peak size–193)/3.

We defined the upper boundary of low-normal CGG repeats based on the distribution of CGG repeats in the WLS population. The mean was 30.6 CGGs, with a standard deviation of 3.8. Two standard deviations below the mean was 23 CGG repeats or fewer. For males, their FMR1 gene on the X chromosome had to have 23 or fewer CGG repeats to be classified into the low-normal group. Otherwise, they were classified as having normal CGG repeats (i.e., 24–40 CGG repeats). Males with 41 or more CGG repeats were omitted from the present analysis (n = 164).

For females, both of their FMR1 alleles had to have 23 or fewer CGG repeats to be included in the low-normal group. Females with 24–40 CGG repeats on both alleles were classified as having normal-length CGG repeats (n = 2452). Omitted from the present analysis were females with any gray zone or premutation-length CGG repeats (n = 295), as well as those who had one low-normal and one normal CGG repeats (n = 676). For females, we dropped those with one normal and one low-normal allele because we did not have access to X-inactivation data, and thus it was not possible to determine the proportion of normal alleles vs. the proportion of low-normal alleles that were active for any given sample member.

Based on these definitions, 341 males (11%) and 46 females (1.8%) were classified as having low-normal CGG repeats and 2758 males and 2452 females were classified as having normal-length CGG repeats. For males, the lowest number of CGG repeats in this sample was 9, whereas for females, the lowest was 11 CGG repeats.

Three rounds of data from the WLS were used to measure the variables for the present study: demographic data collected in 1957 and 2011, identification of number of children and children with disabilities in the 2004/06 and 2011 rounds of data collection, and outcome data collected in 2011, when respondents averaged 71 years of age.

For the present study we included three variables from the 1957 data: the years of education completed by respondent’s father and mother, and the respondent’s IQ score as measured in high school. Scores on the Henmon-Nelson Test of Mental Ability (Henmon and Nelson, 1937), administered at age 17, were available for all WLS respondents. As noted, IQ scores ranged from 61 to 145 (mean = 103.3, SD = 15.0). The test measures a wide variety of mental abilities, including verbal, spatial, and numerical knowledge and reasoning, the composite of which reflects generalized intellectual functioning. Reliability coefficients for the Henmon-Nelson are consistently around 0.90 (Buros, 1940). The scores on the original version of the test show good predictive validity in forecasting academic success in college (Drake and Henmon, 1937), and scores on a later revision were correlated 0.83 with IQs from an individually administered test of intelligence (Wagner, 1981). Within the WLS sample, Henmon-Nelson IQs were correlated 0.60 with high school grade rank, which is consistent with the usual associations found between IQ and school achievement (Gregory, 2004).

Other demographic characteristics used to describe the sample for the current study were years of education completed by the respondent, household income of the respondent, and marital status (1 = currently married, 0 = not currently married). These demographic characteristics reflect status as of the 2011 round of data collection.

We compared those with normal and low-normal CGG repeats with respect to outcome variables in the following domains: cognition, mental health, cancer, and having children with disabilities. Cognition was measured in 2011 by the HUI Mark 3 cognition score (Feeny et al., 1996), which has six levels of cognitive functioning based on respondents’ answers to two questions: “how would you describe your ability to remember things?” and “ how would you describe your ability to think and solve day to day problems?” The six levels of cognitive score are: 1 = able to remember most things, think clearly and solve day to day problems; 2 = able to remember most things, but have a little difficulty when trying to think and solve day to day problems; 3 = somewhat forgetful, but able to think clearly and solve day to day problems; 4 = somewhat forgetful, and have a little difficulty when trying to think or solve day to day problems; 5 = very forgetful, and have great difficulty when trying to think or solve day to day problems; and 6 = unable to remember anything at all, and unable to think or solve day to day problems. The majority (60.8%) were in the least impaired category of this measure, and less than 1% classified themselves in the most impaired category.

Mental health was measured by anxiety, depression, and alcohol symptoms. Anxiety was measured by the Spielberger Anxiety Index (Spielberger et al., 1970), which is a summary score of seven items asking the number of days during the past week respondents felt each emotional state: calm, tense, being at ease, worrying over possible misfortune, nervous, jittery, and relaxed. For depression, respondents reported whether since the previous WLS interview (an average of 6 years previously), they experienced a period of 2 weeks or more when nearly every day they felt sad, blue, depressed, or when they lost interest in most activities, such as work, hobbies, or things they usually liked to do for fun (1 = yes, 0 = no). Episodes of depression associated with physical illness, medications, and alcohol or drug use were not included in this measure. For alcohol symptoms, respondents reported whether they had to drink more to get the same effect as when they first started drinking (1 = yes, 0 = no). This is a standard item from the DSM used to determine whether an individual has developed a tolerance for alcohol and the effect refers to the feeling the individual is trying to achieve by drinking (typically a feeling of intoxication).

Cancer (for women) included lifetime history of breast cancer (coded 1 if yes and 0 if no) and uterine cancer (similarly coded).

For women, having children with disabilities was coded 1 if the respondent had a biological child with a developmental disability or mental health condition, and 0 if not. For men, the analysis of this variable was restricted to fathers of daughters, as it is only daughters who inherit a father’s X chromosome. Respondents who had children with disabilities were identified through a series of screener questions asked during the 2004/06 and 2011 rounds of data collection. The screener consisted of a maximum of 31 questions that began by asking parents if any of their children (living or deceased) had an intellectual or developmental disability and the specific diagnosis. If the parent indicated a specific condition (e.g., Down syndrome, fragile X syndrome, autism), or used terms such as developmental disability, mental retardation, or cognitive disability, he or she was coded as having a child with a disability. In a small number of cases (5.4%), the parent did not know the specific diagnosis given to his or her child, but indicated that the child had difficulties in school. In such cases, branching follow-up questions asked if the child was below-average in intelligence, attended special education classes, and/or had difficulty performing activities of daily living. If so, he or she was classified as having a disability. Other childhood disabilities mentioned by the respondents were noted, such as attention deficit hyperactivity disorder, seizure disorder, etc., and included in this measure of having a child with a disability. Parents also reported whether a child had been diagnosed by a health professional with schizophrenia, bipolar disorder, or a major clinical depression (that required hospitalization or limited the ability of the person to carry on activities of daily living). Suicide in a child was based on parental report of cause of death. Other mental health diagnoses mentioned by parents, such as anxiety disorders, were noted and included in this measure of having a child with a disability.

For each outcome variable, we estimated a regression model, including controls for age and sex, and the category of CGG repeat (normal or low-normal). We controlled for age and sex because of known associations between these variables and several of the outcome variables (e.g., cognition, health) and because of the wide age range of the sample (46 to 92 in 2011). We used the generalized estimating equation (GEE; Diggle et al., 2013) approach to regression modeling, and report regression coefficients (for continuous outcome variables) or odds ratios (for binary outcome variables) with robust standard errors and 95% confidence intervals based on clustering at the level of sibling pair. We also tested the Sex X CGG repeat category interaction effect. We report coefficients or odds ratios for age and sex from the models without the interaction. If the interaction was not significant, we also report the CGG repeat effect from this same model. If the interaction was significant, it is reported, and separate CGG effects are reported for males and females. Use of the GEE modeling approach was motivated by the study design, which as noted earlier, included 1311 sibling pairs whose endpoints on the outcome variables might be correlated.

Due to variable amounts of missing data across outcome variables, sample sizes for males in the low-normal CGG repeat category ranged from 304 to 341, and in the normal CGG repeat category ranged from 2418 to 2690. For females, sample sizes in the low-normal CGG repeat category ranged from 42 to 46, and in the normal CGG repeat category from 2213 to 2404.

In the aggregate, participants in the present study with low-normal CGG repeat lengths did not differ significantly from those who had normal length CGG repeats with respect to any of the background demographic characteristics. As shown in Table 1, their fathers and mothers had similar levels of education (between 9 and 11 years). When in high school, the two groups of participants had equivalent standardized IQ score measurements (between 103 and 106). The two groups had almost identical years of education (13–14 years). They did not differ in their number of biological children (ranging from 2.5 to 2.8).

At an average age of 71, those who had low-normal numbers of CGG repeats did not differ from those with normal numbers of CGG repeats in current marital status (with 83 to 84% of males and 64 to 66% of females currently married) or household income.

Table 1 also shows the means and standard deviations for the outcome variables, broken down by low-normal and normal CGG categories. Table 2 presents the results of the regression models that tested whether and how those in the low-normal CGG repeat category differed from those in the normal category with respect to the outcome variables. All models adjusted for age. In models in which the Sex X CGG repeat category interaction term was non-significant, the regression models also adjusted for sex. In models with a significant interaction term, differences in the outcome variables were reported separately by sex.