Participants and Procedures
The sampling plan for this research was designed to include a sufficient number of participants from the full CGG repeat range to evaluate genotype-phenotype associations. The number of CGG repeats in FMR1 in the human population is not evenly distributed across the CGG range and is highly polymorphic (46, 47). The peak value for CGG repeats is 30, with >90% of individuals having fewer than 40 repeats and the lowest number of repeats ever reported being 6 (7, 46, 48, 49).
Four categories along the CGG repeat range (below FXS) have been described in the literature: premutation, intermediate zone, normal, and low zone (9, 50-54). However, precisely defining the number of repeats in each category is challenging due to both scientific and technical factors. The published guidelines provided by the American College of Medical Genetics and Genomics for defining normal and mutation categories in FMR1 (52) note that the borders of the various categories are approximate. “Each definition may change with increased empirical data and research” (p. 578), and there is an acceptable margin of error of several CGG repeats at the borders of the categories. For this reason, in the current research, we investigate the phenotypic associations of variation in CGG repeats by treating repeat number as a continuous variable.
Participants included 1275 mothers with CGG repeats ranging from 18 to 123.
The majority of these participants (n = 1152) were drawn from the Marshfield Clinic Personalized Medicine Research Project (PMRP) (55), a 20,000-person population-based biobank. Individuals enrolled in this biobank in the early 2000s, and provided written informed consent to provide researchers with access to their DNA and electronic health records, and to be contacted for additional data collection. Per IRB, research results were not returned to participants, nor were the results entered into their medical record or provided to health care personnel. Over half of the PMRP members were female (n = 11,556) and DNA was available for 99.7% of them.
For a previous investigation (56), the DNA samples of all PMRP members were screened for FMR1 CGG repeats. This screening made it possible to select participants for the present study from across the CGG repeat range, with adequate numbers of individuals at the lower and higher ends of the range. Based on previous research defining the number of CGG repeats that can be considered expansions (53, 57-61), we invited all PMRP females who had at least 41 CGG repeats on one allele to participate in the present study. Similarly, we invited all those who had at least one allele below the normal range (defined here as below 26 CGG repeats; (50, 54) to participate. Additionally, based on a power analysis, a random sample of females with normal-range CGGs was selected for inclusion in the present research. Thus, by design, the recruited sample included all females in the population biobank who had expanded or low numbers of CGGs, and a random sample of females in the normal range. The response rate of the recruited females was 77.4%. We further restricted the current analysis to data obtained from mothers who had at least one biological or adopted child.
The CGGs of the participants from PMRP ranged from 18 – 100 repeats. Of note, 44 of these participants had CGG repeats in the premutation range (55+ CGG repeats). To extend the range of FMR1 CGG repeats, clinically-ascertained mothers of children diagnosed with FXS were included in the present analysis (n = 123, with 67-123 CGGs). Participants from the clinically-ascertained samples were recruited from fragile X clinics, via local media, newsletters, brochures, and disability registries (62, 63). All participants (PMRP and clinically-ascertained) completed a questionnaire that provided information on whether they had a child with a developmental or mental health condition (see Table 1), as well as all other non-genetic measures for the current study.
The Institutional Review Boards at the University of Wisconsin-Madison and the Marshfield Clinic approved all procedures and all participants signed informed consents.
Measures
Stress.
Perceived Stress Scale. The Perceived Stress Scale (PSS) (64) is a 10-item, self-report measure that quantifies an individual’s appraisal of stressful experiences from the past month. Examples include “In the last month, how often have you felt difficulties were piling up so high that you could not overcome them?” and “In the last month how often did you feel nervous or stressed?” Each item was scored on a scale of 0 (never) – 4 (very often); four positively-stated items were reverse-coded. The total score represents a sum across items; higher numbers indicate a greater degree of perceived stress (i.e., subjective stress). Age- and gender-based norms were previously established (n = 1406 females; M = 13.7, SD = 6.6) with a Cronbach α coefficient of .78 (65). Higher PSS scores have been linked to poorer health including greater risk of developing depressive symptoms following life events and vulnerability to the common cold (65).
Life Events. Participants reported life events (positive and negative) that they personally experienced during the past year (adapted from Abidin’s Parenting Stress Index; (66). Participants selected events from a list of 22 items, such as divorce, going into debt, and the birth of a child. Higher scores indicate a greater number of personal life events.
Parenting Status. Participants reported whether their child had a developmental or mental health condition (0 = no, 1= yes).
Cognitive function.
Executive Function. Participants completed the Behavior Rating Inventory of Executive Function-Adult Version (BRIEF-A) (67, 68), a well-validated self-report measure of executive function in daily life for adults. The BRIEF-A consists of 75-items that yields an overall raw score of executive function (Global Executive Composite; GEC), made up of two indices: Behavior Regulation Index (BRI) and Metacognitive Index (MI). Participants indicated the extent to which they experienced problems across nine domains: Inhibit, Shift, Emotional Control, Self-Monitor, Initiate, Working Memory, Plan/Organize, Task Monitor, and Organization of Materials, which together comprise the GEC (see Table 2 for definitions). Each item was rated from 1 (never) to 3 (often). Raw scores for each domain were converted into t-scores, with higher scores suggestive of greater executive difficulties in daily life. T-scores that exceeded 65 on any domain indicated clinically-significant executive dysfunction in that area. In order to ensure that respondents did not indicate excessively negative self-perception about their own executive function, the Negativity scale was examined to ensure that no participant met or exceeded a total score of six (67).
The BRIEF-A was previously standardized on a representative population sample of 1136 adults with Cronbach α coefficients ranging from .93-.96 and test-retest reliability ranging from .93-.94 across domains, with utility demonstrated in both clinical and non-clinical samples (67-70). The BRIEF-A has been shown to correlate significantly with direct-assessment measures of executive function (e.g., go/no go and trail making tests) in healthy adults (71) and in individuals with disorders associated with executive dysfunction (72-74). The present study used the GEC t-score as the indicator of executive functioning.
The BRIEF-A was standardized on participants ages 18-90. Since there were 12 participants in the present sample who were over the age of 90, we checked all findings excluding participants over age 90, which did not change results. Therefore, the findings reported below include all participants.
Self-reported Memory Problems. Participants answered the question: Do you have problems with memory? This question was rated as 0 (no problems with memory), 1 (undiagnosed problems with memory), or 2 (diagnosed memory problems). Only 20 mothers (1.6% of the sample) reported diagnosed memory problems. To reduce skewness, all “2” responses were collapsed to “1”.
FMR1-related variation. DNA samples were obtained from cheek swabs and blood samples from all participants, and were analyzed for CGG repeats in FMR1. Assays were completed at the Wisconsin State Laboratory of Hygiene and the Rush University Medical Center Molecular Diagnostics Laboratory, using procedures described previously (56, 63, 75, 76). Mothers who were mosaic for the full mutation were excluded from the present analyses.
The DNA of the clinically-ascertained mothers of children with FXS was obtained from cheek swabs, and CGG repeat length was assayed in the laboratory of Elizabeth Berry-Kravis, MD, PhD at Rush University. The DNA of the PMRP members was obtained from blood samples, tested either in the Rush laboratory or in the Wisconsin State Laboratory of Hygiene under the supervision of Mei Wang Baker, MD.
Statistical Analyses
Statistical analyses were performed using IBM SPSS Statistics, version 26 (77). Descriptive statistics and Pearson correlations among all study variables are presented in Table 3. Maternal age and education were controlled in all subsequent analyses. To control for potential effects of the second FMR1 allele (as females carry two X chromosomes), the “shorter” allele (i.e., the allele with the lower number of CGG repeats) was included as a covariate in all regression analyses.
For the domain of executive function, the primary analysis involved three hierarchical regressions (one for each type of stressor) that assessed the key prediction that stress and FMR1 CGG repeat length would each uniquely contribute to self-reported executive function difficulties. For memory problems, three logistic regressions (one for each type of stressor) were completed to test if stress and FMR1 CGG repeat length would predict the likelihood of a self-reported memory problem. For both executive function and memory problem models, maternal age, education, the number of CGG repeats on the shorter allele, and each stress measure were entered into the first block; CGG repeat length (on the long allele) was entered into the second block. In the third block, a separate term (CGG squared) was included in all regression models to evaluate potential curvilinear CGG effects within the sample. A significant curvilinear effect would suggest that components of the CGG distribution (e.g., premutation expansions) would potentially be driving the CGG effect, whereas if the curvilinear effect is not significant, this would suggest that the CGG effect is linear across the full CGG range. Following the approach of Hunter et al. (16), a Bonferonni correction was used to adjust for multiple testing for each indicator of cognitive functioning, with the alpha level set at p = .016 (0.05/3).