FMR1 Genotype Interacts with Parenting Stress to Shape Health and Functional Abilities in Older Age

Abstract

This study investigated the association of genotype (CGG repeats in FMR1) and the health and well-being of 5628 aging adults (mean age = 71) in a population-based study. Two groups were contrasted: aging parents who had adult children with developmental or mental health disabilities (n = 785; the high-stress parenting group) and aging parents of healthy children who did not have disabilities (n = 4843; the low-stress parenting group). There were significant curvilinear interaction effects between parenting stress group and CGG repeats for body mass index and indicators of health and functional limitations, and the results were suggestive of interactions for limitations in cognitive functioning. Parents who had adult children with disabilities and whose genotype was two standard deviations above or below the mean numbers of CGGs had poorer health and functional outcomes at age 71 than parents with average numbers of CGGs. In contrast, parents who had healthy adult children and who had similarly high or low numbers of CGG repeats had better health and functional outcomes than parents with average numbers of CGGs. This pattern of gene by environment interactions was consistent with differential susceptibility or the flip-flop phenomenon. This study illustrates how research that begins with a rare genetic condition (such as fragile X syndrome) can lead to insights about the general population and contributes to understanding of how genetic differences shape the way people respond to environments.

Introduction

Trinucleotide repeat disorders are an intriguing and poorly understood group of conditions, including Huntington’s disease, spinal bulbar muscular atrophy, and myotonic dystrophy, that sometimes involve expansions within the coding region of a gene, resulting in insertion of polypeptide sequences, and other times expansions in the non-coding regions leading to altered expression levels or RNA toxicity. The full range of trinucleotide repeats has not been fully investigated in any of these disorders. Instead, the population has been categorized as ‘clinical’ (above a certain number) or ‘normal’ (below the same number) of repeats. Our focus is on the gene known as fragile X mental retardation 1 (FMR1), in which expansions above a critical threshold of a CGG triplet repeat (usually >200 CGG repeats) in the 5′ untranslated region cause fragile X syndrome (FXS), the most common inherited form of intellectual disability. FMRP, the protein produced by FMR1, normally regulates the translation of ~30% of all transcripts in the synaptic proteome, predicting a key and widespread role in functioning of the nervous system [Darnell et al., 2011].

Children inherit FXS from their mothers who most often are premutation carriers (defined as having between 55 and 200 CGG repeats in FMR1). Research conducted during the past 15 years has demonstrated that some premutation carriers, in addition to being at high risk of having a child with FXS, also have clinical symptoms, especially during older age, including a neurodegenerative motor and cognitive condition (Fragile X-associated Tremor and Ataxia Syndrome, FXTAS), and early reproductive aging among females (Fragile X-associated Primary Ovarian Insufficiency, FXPOI) [Hagerman & Hagerman, 2013]. RNA toxicity is believed to be responsible for the elevated level of neurological, motor, and reproductive symptoms observed in premutation carriers [Hagerman, 2013].

Some research has also implicated other health and mental health difficulties in premutation carriers such as anxiety and depression, headaches, muscle soreness and weakness, fatigue, joint pain, dizziness, constipation, diarrhea, and other symptoms [Bailey et al., 2008; Coffey et al., 2008; Smith et al., 2012]. Because most of the studies implicating these symptoms were conducted on mothers of children with FXS, however, there is controversy about whether these symptoms are the direct effect of the premutation, the result of exposure to the stress of parenting a child with FXS, or by an interaction between stress exposure and the biological effects of the premutation [Wheeler et al., 2014].

Research on premutation carriers who do not have children with disabilities is needed to determine the prevalence of such symptoms in the absence of chronic parenting stress caused by children’s disabilities. Recently, Gossett et al. [2016] compared premutation carriers who did not have children with disabilities (who were identified via family-wide genetic testing when a family member was diagnosed with FXS) and community controls, and found significantly elevated symptoms of anxiety, depression, interpersonal sensitivity, obsessive-compulsiveness, and somatization relative to controls during the previous week. Although these premutation carriers were not exposed to the stress of parenting a child with FXS, they were aware of their family’s genetic vulnerability and their own status as a premutation carrier, which may have led to some of their symptoms. Roberts et al. [2009] found that the onset of the mental health difficulties of premutation carrier mothers of children with FXS often preceded the birth of their child with FXS. Both of these studies were interpreted to suggest that there may be direct effects of the premutation, in addition to the reactive effects of stressful parenting. However, in a population-based study where premutation carriers were not aware of their genetic status, premutation carriers did not differ from controls in depression [Seltzer et al., 2012a], although they did have significantly higher rates of symptoms that might be indicative of FXTAS (numbness, dizziness/faintness) and FXPOI (younger age at last menstrual period).

Interest in variation in CGG repeats in FMR1 has extended to the “gray zone” (or “intermediate-length alleles”) of the FMR1 gene, defined as between 45 and 54 CGG repeats by the American College of Medical Genetics [Maddalena et al., 2001]. Other studies have defined the lower boundary of the gray zone by as few as 34 CGG repeats [Dombrowksi et al., 2002; Rousseau et al., 1995] and the upper boundary with as many as 60 CGG repeats [Bretherick et al., 2005; Hall et al., 2011; Nolin et al., 1996]. A small number of adults in the gray zone have been reported to have similar symptoms as the premutation, but milder in severity and considerably lower in frequency [Hall et al., 2012], although the evidence is inconsistent [Bretherick et al., 2005; Loesch et al., 2007]. Individuals with gray zone CGG repeats are currently considered to be normal, although research with this population in later life is very limited.

The modal number of CGG repeats in the human population is 30 [Chen et al., 2003; Fu et al., 1991], with reports of CGG repeats in humans as low as 6 [Fu et al., 1991]. There has been only limited interest in the normal range of the CGG repeat distribution, but the few available studies are informative. Chen et al. [2003] showed that synthetic human FMR1 promoter sequences transfected into cell lines were associated with reduced protein translation at both low and high numbers of CGG repeats (vs. the modal number of 30 CGGs). Nagamani et al [2012] reported that duplication and deletion of FMR1 can lead to overlapping clinical neurodevelopmental phenotypes. Ramocki and Zoghbi [2008] argued that there is a need for tight neuronal homeostatic control mechanisms for normal cognition and behavior, and that imbalances in homeostatic controls in multiple genes, including FMR1, might be responsible for neurodevelopmental and neuropsychiatric disorders.

These studies suggest that deviations in either direction from the mode of 30 CGG repeats may have similar associations with phenotypic characteristics. However, until recently, there has been very limited genetic data on the full range of CGGs in FMR1 in non-clinical human populations. Such data are needed to explore population-level genotype-phenotype associations in the full range of FMR1 CGG repeats currently considered normal. Fortunately, genetic data are increasingly being collected within population-level studies, including the Wisconsin Longitudinal Study (WLS) [Hauser et al., 1998], the source of data for the present research.

In a preliminary study that partially set the stage for the present investigation, we explored the effects of low numbers of CGG repeats (defined as two standard deviations below the mean ([i.e., 23 or fewer CGGs] [Mailick et al., 2014] using data from the WLS (n = 6747, described below). We found that men with low CGGs and women who were homozygous for low CGGs (i.e., low CGGs in FMR1 on both X chromosomes) had significant elevations (vs. controls) in symptoms associated with performance of motor activities and with memory in everyday life. We also found that women with low CGGs had elevated alcohol symptoms, but there was no elevation for either men or women in depression and anxiety. For both men and women, there was significant elevation in the likelihood of having a child with a developmental disability or mental health problem, again hinting that CGG repeat variation may be associated with brain development and functioning.

Not investigated in our earlier low-CGG study was how stress exposure (specifically, the stress of parenting a child with a disability) might interact with CGG repeat number to result in variation in phenotypic symptoms. Such an investigation could clarify the joint and interactive effects of genetic variation and stressful parenting. We also did not examine the association of CGGs, quantitatively measured, across the entire spectrum of FMR1 CGG repeats (not including FXS); rather, our analysis was a binary comparison (low vs. normal CGGs). A quantitative investigation could provide data directly relevant to Ramocki and Zoghbi’s theory of the need for tight homoeostatic control mechanisms for normal cognition and behavior. We also did not study the full population available in our prior study. For example, excluded from the study of low CGGs were women heterozygous with respect to CGG repeats (i.e., one low and one normal allele, or one expanded and one normal allele), which are common in the population.

The present study advances our prior research in three ways: It is an investigation of the association of quantitative variation across the range of CGG repeats and symptoms, including both low and high values of CGG (not including FXS); it includes males and females, and among the females it includes those who are heterozygous as well as those who are homozygous; and it explores gene by environment interactions by evaluating how stress exposure interacts with CGG repeat number across the CGG repeat range. For this analysis, we restricted the sample to parents of living children (non-parents and parents whose children were deceased were excluded) and we divided parents into two groups: those who had at least one child with a developmental or mental health problem and those had children without such disabilities (i.e., none of their children had these conditions). Our goal was to represent as broad a population in this analysis as possible.

Notably, data were collected when parents averaged 71 years of age and the children averaged age 40. Although all parents at any stage of life experience some child-related stress, parenting a son or daughter with a disability during older age is a unique measure of chronic stress exposure, spanning decades in the lives of the parents. Furthermore, in older age, parents whose children have disabilities continue to worry about and provide support to their adult child with the disability [Fingerman et al., 2009; Seltzer et al., 2011], whereas aging parents whose adult children are non-disabled expect to receive support from their children if such support is needed [Fingerman et al., 2011]. Furthermore, past research has shown that there are rippling effects of having a brother or sister with disabilities, affecting non-disabled siblings [Wolfe et al., 2014] and potentially the parent-child relationship. Thus, when parents are in their 70s, there may be greater divergence between the family context of parents who have an adult child with disabilities and parents of non-disabled adult children than there was when they and their children were young.

The overarching hypothesis of the present study, building on the previous research of Chen et al., Nagamani et al., and Romocki and Zoghbi, as well as our own prior studies [Mailick et al., 2014; Seltzer et al., 2012a; Seltzer et al., 2012b], is that having either low or expanded numbers of CGG repeats in FMR1 interacts with stress exposure to result in a similar phenotypic profile. When chronic stress exposure is high, we predict that there will be increased vulnerability at both the low and expanded ends of the CGG repeat distribution, as compared with those with low and expanded numbers of CGG repeats who do not have a child with disabilities and thus were not chronically exposed to parenting stress.

In previous research on FMR1 CGG expansions [Seltzer et al., 2012b], we observed that the FMR1 gene has the properties of a differential susceptibility gene. The differential susceptibility hypothesis [Belsky et al., 2007; Belsky et al., 2009] predicts that people with certain genotypes are more likely to manifest either poorer or better outcomes, depending on the nature of their environmental exposures -- hence the term differential susceptibility to the environment. As Belsky et al. [2009] noted in their review, “the very same individuals who may be most adversely affected by many kinds of stressors may simultaneously reap the most benefit from environmental support and enrichment (including the absence of adversity)” (p. 886). This was the pattern of results we found in our previous study, as in the absence of stress, premutation carriers with between 90 and 105 CGGs manifested the best health and mental health outcomes, whereas in the face of the highest level of stress, such individuals experienced the poorest outcomes. Therefore, we explored the differential susceptibility hypothesis in the present study, and examined how stress exposure interacted with CGG repeat number at the lower and higher ranges of the CGG distribution, and its effect on health and mental health outcomes.

Materials and Methods

Data and Sample Members

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 the cohort of high school seniors from that year [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 50s; in 2004 with 7,265 (80.0%) of the surviving respondents when they were in their mid-60s; 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, who were selected in either 1977 or 1994, and longitudinally followed in 1994 (for those selected in 1977), 2005, and 2011, with 5,823 siblings participating in one or more of these data collection points.

The WLS is a public-use data set. All of the original WLS participants were high school graduates, as were 93% of their siblings. Although this is a well-educated cohort, WLS participants ranged in IQ score 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 nearly 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); 19% had IQ scores of 115 or higher (1 SD above the mean). Reflecting Wisconsin’s population at the mid-20^th 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). Just over half (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 < .001), three points higher IQ scores (103.2 vs. 99.5, p, <.001), and higher high school grades (percentile rank of 54.6 vs. 46.7, p <.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, 6732 yielded sufficient DNA for the CGG repeat assay, of whom 5628 met the inclusion criteria for the present study (parents of living children). All participants provided informed consent under a protocol approved by the Institutional Review Board of the University of Wisconsin-Madison.

Determination of the FMR1 CGG Triplet Repeat Number

The number of FMR1 CGG repeats was determined for all samples using a PCR-based protocol that incorporated reagents developed and manufactured by Celera Corporation. The specific procedures we used were described in previous publications [Maenner et al., 2013; Seltzer et al., 2012a]. 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).

Identification of Parents Experiencing High Levels of Stressful Parenting

We conceptualized having a child with a disability as our indicator of chronic stress exposure. Of the 6732 cases (3263 males and 3469 females) whose saliva sample yielded sufficient DNA for the CGG repeat assay, the present study is based on 785 cases (344 males and 441 females) who had children with developmental or mental health disabilities, and 4843 comparison cases (2432 males and 2411 females) who reported that none of their children had disabilities. As noted earlier, non-parents were excluded. The participants who met the criteria for the present analysis included 926 sibling pairs.

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 had an intellectual or developmental disability or a mental health problem, and the specific diagnosis. (For a full description of the screener questions, see Wolfe et al., 2014). Table 1 lists the parent-reported diagnoses of the adult children included in the present analysis.

Table 1.

Type of Children’s Disabilities

	N
Developmental Diagnoses
Epilepsy	70 (8.9%)
Intellectual Disability	51 (6.5%)
Learning Disability	50 (6.4%)
ADHD	41 (5.2%)
Cerebral Palsy	24 (3.1%)
Autism Spectrum Disorder	23 (2.9%)
Down Syndrome	15 (1.9%)
Other Developmental Disability	41 (5.2%)
Mental Health Diagnoses
Bipolar	220 (28.0%)
Depression	91 (11.6%)
Schizophrenia	59 (7.5%)
Suicide	29 (3.7%)
Other Mental Illness	71 (9.1%)
TOTAL	785 (100.0%)

To confirm that having a child with a developmental or mental health disability is a valid indicator of stressful parenting, we compared the two groups of parents with respect to the health of the adult child with disabilities and the potential rippling effects on the relationship of parents with their non-disabled children. Indeed, the adult children of the two groups of parents differed in their overall health. Nearly 70% of the adult children with disabilities were rated as having poor or fair health (69.5%) versus only 5.7% of the non-disabled adult children in the comparison group (chi-square = 342.52, p < .001). Additionally, we found evidence of the rippling effects of stressful parenting, as parents of adults with disabilities had a more distant relationship with their non-disabled children than parents in the comparison group. Aging parents of adults with disabilities were significantly less likely to feel that they could ask their non-disabled child for help if they needed assistance while sick than parents whose children were all non-disabled (chi square = 67.29, p <.001). They were significantly less likely to feel that they could ask their non-disabled children for some money to help pay bills than those whose children were all non-disabled (chi square = 34.39, p<.001), and they were significantly less likely to feel loved and cared for by their non-disabled children (chi square = 62.20, p<.001). In addition, they lived at greater geographic distance from their non-disabled children than the comparison group (t = 4.33, p <.001). These differences provided supportive evidence of the operational definition of stressful parenting used in the present study.

Thus, the present study can be conceptualized as a study contrasting two groups of aging parents: (1) those whose adult children had disabilities and fair or poor health, who also had more distant relationships (instrumentally, affectively, and geographically) with their non-disabled children, and (2) those whose adult children were non-disabled, mainly healthy, and who had closer relationships with their children.

Measures

The outcome variables were taken from data collected by the WLS in 2011 when parents averaged 71 years of age. Data were collected during home visits and by self-administered questionnaires. The outcome variable domains included in this study were physical health, functional limitations, mental health, and cognitive functioning, and for women, age at menopause. These domains were selected because in past research they were identified as phenotypic characteristics associated with expanded CGG [Wheeler et al., 2014] and CGG [Mailick et al., 2014] repeat counts.

Physical health measures included Body Mass Index (BMI) and the total number of health symptoms. BMI was calculated based on reported weight and height, rounded to the nearest integer.The number of health symptoms was count of 16 symptoms (each coded as 1 = present or 0 = absent) that parents may have experienced in the past six months. The symptoms were: lack of energy, fatigue or exhaustion, headaches, dizziness or faintness, numbness, ringing in the ears, upset stomach, constipation, diarrhea, aching muscles, stiff/swollen joints, back pain or strain, chest pain, shortness of breath, excessive sweating, and skin problems.

Functional limitations were measured by two questions about the extent to which health problems limited parents’ activities: “how much does your health limit you in moderate activities such as moving a table, pushing a vacuum cleaner, bowling or playing golf?” and “how much does your health limit you in climbing several flights of stairs?” They were scored on a three point scale: 1 = not limited at all, 2 = limited a little, 3 = limited a lot.

Mental health was measured by standardized self-report measures of depressive symptoms and anxiety. Depressive symptoms were assessed by the Center for Epidemiological Studies-Depression Scale (CES-D) [Radloff, 1977]. For each of 20 depression symptoms, the parent was asked to indicate how many days in the past week the symptom was experienced (0 = never to 3 = 5 to 7 days). The total score is the sum of the ratings for the 20 items. Anxiety was measured by the Spielberger Anxiety Index [Spielberger et al., 1970], which is a summary score of 7 items asking the number of days during the past week (ranging from 0 to 7) respondents felt each emotional state: calm, tense, being at ease, worrying over possible misfortune, nervous, jittery, and relaxed (positive items were reverse coded). Items were summed with the total score ranging from 0 to 49.

Cognitive functioning was measured by two questions that are part of the Health Utility Index (HUI) Mark 3 cognition score [Feeny et al., 1996]. The first question was about memory, “During the past four weeks, how would you describe your ability to remember things?” and was scored on a four-point scale: 1 = able to remember most things, 2 = somewhat forgetful, 3 = very forgetful, 4 = unable to remember anything at all. The second question was about problem-solving ability, “During the past four weeks, how would you describe your ability to think and solve day-to-day problems?”, and was scored by a five-point scale: 1 = able to think clearly and solve problems, 2 = had a little difficulty, 3 = had some difficulty, 4 = had a great deal of difficulty, 5 = unable to think or solve problems. The two items were combined to form the HUI cognition score ranging from 1 = best (able to remember most things, think clearly and solve day to day problems) to 6 = worst (unable to remember anything at all, and unable to think or solve day to day problems).

Age at menopause was measured for mothers, who reported their age at last menstrual period (measured in 2004/05). Mothers were also asked whether their ovaries or uterus had been surgically removed prior to menopause, and those who answered affirmatively were not included in the analysis of age at menopause.

Statistical Analysis

For the females, the assay yielded CGG repeat data on the FMR1 gene on both X chromosomes. Because we did not have activation ratio data, one X chromosome was selected for analysis in the present study as follows. Although we considered alternative approaches [Kline et al., 2014; Gustin et al., 2015], we followed the approach of Hunter et al. [2012]. We selected the longer allele in mothers who had one expanded (i.e., > 40 CGGs) allele and one normal allele (n = 218) and in the one case who had two expanded alleles. Similarly, we selected the shorter allele in mothers who had one low allele (i.e., < 24 CGGs) and one normal allele (n = 553) and in mothers who had two low alleles (n = 41). Finally, we randomly selected one allele for analysis in the present study in mothers who had two normal alleles (between 24 and 40 CGG repeats, n = 2085), and also for those with one low allele and one expanded allele (n = 22).

The number of CGG repeats analyzed for the present study ranged from 8 to 134, but there were very few cases beyond 60 CGG repeats (15 cases in total, 4 who had a child with a disability and 11 in the comparison group). To handle this skewed distribution, CGG repeats longer than 60 were Winsorized (top-coded) as 60 in the analyses [Tukey, 1962].

For each of eight phenotypic outcome variables (Table 2), we estimated a regression model with CGG repeat count and an indicator variable for parenting stress (parenting a child with disabilities vs. comparison group) as the key predictors, and including controls for age, sex, and years of education of the respondent. To account for the dependency of observations among the sibling pairs in the sample, we used the Generalized Estimating Equation with the exchangeable correlation structure (GEE) [Diggle et al., 2013] approach to regression modeling, and report regression coefficients with robust standard errors based on clustering at the level of sibling pair. CGG repeats was centered at 30 repeats.

Table 2.

Descriptive Statistics of Study Variables

	Parents of children with disabilities	Parents in the comparison group	t-value / Χ2
	mean (s.d.) (n = 785)	mean (s.d.) (n = 4843)
Background Characteristics
Gender (1 = female) [%]	56.2%	49.8%	[Χ2 = 11.1***]
Age in years (in 2011)	71.2 (4.0)	71.0 (4.0)	1.00
Years of education	14.2 (2.6)	13.9 (2.4)	3.96***
Phenotypic Characteristics
BMI (kg/m2)	29.1 (5.9)	28.5 (5.3)	2.38*
# health symptoms	5.8 (3.6)	4.8 (3.4)	6.53***
Limitations in physical activitiesa	1.49 (.67)	1.43 (.63)	2.37*
Limitations in climbing several flights of stairsa	1.73 (.73)	1.62 (.70)	3.66***
Limitations in cognitive functioningb	1.9 (1.2)	1.8 (1.1)	2.31*
Depressive symptoms	8.6 (8.0)	7.5 (7.0)	3.60***
Anxiety symptoms	8.0 (8.4)	6.6 (7.2)	4.21***
Age at menopause (females)	50.3 (4.9)	50.8 (5.1)	−1.46
CGG Repeat Length c
CGG repeats: Males	mode/median = 30 mean = 29.0 (5.7) range = 13 – 58 (n=344)	mode/median = 30 mean = 29.4 (5.7) range = 8 – 79 (n=2432)	-1.44
CGG repeats: Females	mode/median = 30 mean = 29.0 (9.0) range =13 – 127 (n=441)	mode/median = 30 mean = 29.9 (7.2) range = 10 – 134 (n=2411)	0.51

Note

^a1 = not limited, 2 = limited little, 3 = limited a lot

^b1 = 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.… 6 = Unable to remember anything at all, and unable to think or solve day to day problems.

^cCGG repeats number before being Winsorized at 60.

Given the previously-reported curvilinear associations between CGG repeats in the premutation range and various outcomes [Allen et al., 2007; Ennis et al., 2006; Mailick et al., 2014; Roberts et al., 2009; Seltzer et al., 2012b; Tejada et al., 2008], we tested curvilinear effects of CGG repeats (quadratic term of CGG repeats). We also tested the linear and quadratic interactions between parenting stress (parents of children with disabilities vs. comparison group parents) and CGG repeat length. To test if the linear and/or the quadratic interaction term was significant, we first ran a regression model with both interaction terms (full model) and performed a joint test of whether either one or both of the pair of interaction terms are different from zero versus the null hypothesis that they are both simultaneously equal to zero. If this joint test (with 2 degrees of freedom) was significant (i.e., at least one of two interaction term was significantly different from zero), we retained both interaction terms in the final model and reported test statistics for each of the terms individually. If the joint test was not significant, we reported regression results from the main effects only model.

After modeling, plots to guide interpretation of the GEE analyses were conducted of the joint effects of CGG repeat length and parenting stress group on each outcome variable using component plus residual (CPR) plots [Mallows, 1986]. The plots display the effects of CGG and parenting stress, adjusted for other factors (age, sex, and years of education of the parent). In these plots, a full model is fitted to the data, including all adjustors, parenting stress group, linear CGG terms, and the linear interaction terms between CGG repeat number and parenting stress. The regression residuals are extracted for each observation and to each residual is added the joint parenting and linear CGG regression effects (3-term: parenting, linear CGG, and linear CGG-by-parenting interaction). These “component plus residual” terms are then scatterplotted versus CGG repeat, separately for each parenting group. The scatterplots are augmented by smooth non-linear curves (implemented with lowess) [Cleveland, 1981] for the mean response as a function of CGG repeat number for each parenting group. In the scatterplots, the CGG repeat values of individual data points were jittered slightly (adding a small amount of random noise) to avoid multiple points being plotted over one-another, and to provide a more representative display of the variation in data density.

Finally, we tested for differential susceptibility by following the steps outlined in Belsky et al. [2007]. Step 1 involves a statistical test for a cross-over interaction. Step 2 involves testing the independence of the susceptibility factor (CGG repeats in the present study) and the predictor (parenting a child with disabilities). Step 3 involves a test of the association between the susceptibility factor and the outcome variables. Step 4 involves comparison of the regression plot with prototypical displays in Belsky et al. [2007]. Step 5 involves replacing the susceptibility factors and outcomes with different susceptibility factors and outcomes to test the specificity of the model. For Step 5, we replaced the susceptibility factor of CGG repeat count with a different genetic susceptibility factor, namely APOE ε4 allele, which is associated with increased risk for neurodegenerative diseases and is available in the WLS data set.

We performed these tests to determine if differential susceptibility could be established for the patterns of data observed in present study. Belsky’s five tests assume linear interaction terms, but in the present analysis, our interaction term was curvilinear. Therefore, we created a dichotomous CGG-repeat length variable, contrasting a combined low and high group (CGG repeat length less than 24 or greater than 40) with the normal group (CGG repeat count of 24 to 40), following the approach we used previously [Seltzer et al., 2012b].

Results

Descriptive Findings

Figure 1 shows a histogram of the distribution of CGG repeats for alleles from men and women separately. The figure portrays the specific alleles that, for women, were analyzed for the present study (following the selection rules described above). The modal number of CGG repeats in this population-based sample was 30 and there were several other high-frequency genotypes, particularly 20 and 23 CGG repeats.

Figure 1.

Histogram of CGG Repeat Length Frequencies by Gender.

Table 2 shows the unadjusted means and standard deviations for the study variables, separately for parents of children with disabilities and parents in the comparison group. The two groups did not differ in age (mean age = 71 years), but there were more mothers among parents of children with disabilities than parents in the comparison group (56.2% vs. 49.8%). Additionally, parents of children with disabilities had more years of education than parents in the comparison group (14.2 years vs. 13.9 years).

Parents of children with disabilities had significantly poorer health and greater functional limitations than the parents in the comparison group, including higher BMI, more health symptoms, and a greater likelihood that their health problems limit their ability to perform physical activities (such as moving a table, pushing a vacuum cleaner, bowling or playing golf) and climbing several flights of stairs. Similarly, parents of children with disabilities had significantly poorer cognitive functioning and elevated levels of depressive and anxiety symptoms than parents in the comparison group.

Regarding CGG repeat length, mothers of children with disabilities ranged from 13 to 127 CGG repeats (mode = 30; mean = 29.0). Fathers of children with disabilities ranged from 13 to 58 CGG repeats (mode = 30; mean = 29.0). Comparison group mothers and fathers ranged from 10 to 134 (mode = 30; mean = 29.4) and from 8 to 79 (mode = 30; mean = 29.4) CGG repeats, respectively. Notably, there were mothers and fathers with both CGG expansions and low CGGs among those whose children had disabilities and those whose children did not have disabilities.

Multivariate Analyses

Table 3 presents the results of the regression models that tested whether the number of CGG repeats interacted with parenting stress group with respect to the outcome variables. The results of joint test of interaction showed that there were significant interaction effects between parenting stress group and CGG repeats for BMI, number of health symptoms, limitations in physical activities, and limitations in climbing several flights of stairs (p-values < .05). Also, the result was suggestive of significant interactions for limitations in cognitive functioning (p-value < .10). The results of full models showed that the curvilinear interactions for CGG repeats (CGG repeat squared X parenting stress group) were significant for these dependent variables.

Table 3.

Generalized Estimating Equation Analysis of Phenotypes by Parenting Stress Group and CGG Repeats: Testing joint null hypotheses that both interaction terms are not different from zero.

	BMI (kg/m2)	# health symptoms	Limitations in physical activities	Limitations in climbing several flights of stairs
Parental age	−.07 (.02)**	.00 (.01)	.02 (.00)***	.02 (.00)***
Education (in years)	−.23 (.03)***	−.10 (.02)***	−.03 (.00)***	−.04 (.00)***
Gender (female=1)	−1.04 (.17)***	.20 (.11)+	.01 (.02)	.10 (.02)***
Parenting stress group (comparison group=0)	.43 (.27)	.77 (.17)***	.05 (.03)	.09 (.03)**
CGG repeats- linear	.01 (.01)	.01 (.01)	.00 (.00)	.002 (.002)
CGG repeats- squared	−.002 (.001)*	−.002 (.001)*	−.00 (.00)	−.00 (.00)
CGG repeats- linear X parenting stress group	−.02 (.04)	−.05 (.03)	−.011 (.005)*	−.011 (.005)*
CGG repeats- squared X parenting stress group	.007 (.003)*	.005 (.002)**	.0006 (.0003)*	.0008 (.0004)*
Test: both interactions are zero (2df joint test)	Chi-sq=7.52* P=0.023	Chi-sq=7.73* P=0.021	Chi-sq=6.18* P=0.045	Chi-sq=6.67* P=0.036
	limitations in cognitive functioning	depression	anxiety	age (yrs) at menopause
Parental age	.02 (.00)***	−.06 (.031)	−.10 (.03)**	−.03 (.03)
Education (in years)	−.06 (.01)***	−.28 (.04)***	−.18 (.04)***	.19 (.06)***
Gender (female=1)	−.12 (.03)***	.86 (.21)***	1.09 (.22)***	--
Parenting stress group (comparison group=0)	.08 (.05)	0.84 (.36)*	1.02 (.37)**	−.45 (.41)
CGG repeats- linear	−.00 (.00)	−.01 (.02)	−.01 (.02)	.02 (.02)
CGG repeats- squared	−.00 (.00)	−.00 (.00)	−.00 (.00)	−.004 (.002)*
CGG repeats- linear X parenting stress group	−.01 (.01)	-	-	-
CGG repeats- squared X parenting stress group	.0012 (.0006)*	-	-	-
Test: both interactions are zero (2df joint test)	Chi-sq=4.76+ P=0.093	Chi-sq=3.28 P=0.194	Chi-sq=2.48 P=.290	Chi-sq=1.71 P=.426

^aCGG repeat length was Winsorized at 60, and centered at 30.

^*p < .05

^**p < .01

^***p < .001

For depressive symptoms, anxiety symptoms, and age at menopause, the joint test did not detect an interaction. Consistent with preliminary t-tests, there were main effects of parenting stress group and depressive symptoms and anxiety symptoms, such that parents of children with disabilities had significantly higher levels of depression and anxiety than the comparison group, regardless of CGG repeat number. For age at menopause, the curvilinear term of CGG repeat was significant, with earlier age at menopause evident for those with higher CGGs, but this did not interact with parenting stress. As CGG repeat length increased, particularly above 40 CGG repeats, age of menopause decreased.

The significant quadratic interaction effects are depicted in Figures 2 through 6. As described above, these plots display the variation in the dependent variable as a joint function of parenting stress group and CGG repeats, adjusting for age, years of education, and gender. Among those with modal numbers of CGG repeats (around 30), the plots generally show similarity between parents of children with disabilities and parents in the comparison group with respect to the dependent variables. However, at both ends of the CGG repeat distribution, parents of children with disabilities and parents in the comparison group diverged in their phenotypic characteristics.

Figure 2.

Component plus residual of BMI (kg/m²), and lowess curve fits of CGG repeats, adjusted for age, years of education, and gender.

Figure 6.

Component plus residual of limitation in cognition level, and lowess curve fits of CGG repeats, adjusted for age, years of education, and gender.

Differential Susceptibility Analyses

Visual examination of the plots of the interaction effects showed that the phenotypes of parents with CGG repeats at the extremes of the distribution were associated with stress exposure (i.e., stressful parenting). Those exposed to chronic stress had greater health and functional problems than those with the modal number of CGGs (30 repeats); conversely, those not exposed to chronic stress appeared to have better health and functional profiles than those with modal CGGs. This pattern may be indicative of differential susceptibility.

The differential susceptibility hypothesis predicts that people with certain genotypes are more likely to manifest either poorer or better outcomes, depending on their environmental exposure [Belsky et al., 2007; Belsky et al., 2009]. In the present study, parents who had low or high numbers of CGG repeats in FMR1 manifested either poorer or better outcomes, depending on their exposure to environmental stress, suggesting differential susceptibility.

We followed the five tests proposed by Belsky et al. [2007] to establish true differential susceptibility. The first of Belsky et al.’s tests is that there should be genuine interactions between the susceptibility factor (CGG repeat group in the present study) and the predictor (parenting stress group), with regression lines of the predictor for each subgroup of the susceptibility factor crossing each other. In the multivariate analysis using continuous CGG repeat data reported above, we showed that there were significant interaction effects between parenting stress group and CGG repeat length (curvilinear) with respect to five outcome variables (BMI, number of health symptoms, limitations in performance of physical activities, limitations in climbing several flights of stairs, and limitations in cognitive functioning). Using the dichotomous CGG repeat variable (low/high vs. normal), all of these interaction effects were significant (p<.05, see Table 4), except for BMI (where there was a trend-level interaction effect, p<.10). Graphical depiction of the interactions show a cross-over of the regression lines for low/high versus normal group for all of these outcome variables (see Table 4).

Table 4.

Test of Differential Susceptibility: Adjusted^a Means of Dependent Variables by CGG Repeat Length Groups and Parenting Stress Groups^b,^c.

	Plots of interactions: CGG length X parenting stress	Low/High CGG repeats		Normal CGG repeats		Interaction Coefficient
		Parents of children with disabilities	Comparison	Parents of children with disabilities	Comparison
BMI		29.8 [28.8, 30.7]	28.2 [27.8, 28.6]	29.1 [28.5, 29.6]	28.6 [28.4, 28.8]	1.13+ [−0.52, 2.31]
# health symptoms		6.25 [4.78, 6.87]	4.67 [4.43, 4.90]	5.68 [5.36, 6.00]	4.90 [4.78,5.03]	0.80* [0.06, 1.54]
Limitations in physical activities		1.58 [1.47, 1.69]	1.39 [1.35, 1.43]	1.48 [1.42, 1.54]	1.43 [1.41, 1.45]	0.13* [.0.001, 0.26]
Limitations in climbing several flights of stairs		1.84 [1.72, 1.96]	1.57 [1.53, 1.62]	1.71 [1.65, 1.78]	1.63 [1.61, 1.66]	0.19* [0.04, 0.33]
Limitations in cognitive functioning		2.11 [1.78, 1.85]	1.75 [1.68, 1.83]	1.87 [1.78, 1.97]	1.82 [1.78, 1.85]	.29* [0.06, 0.52]

Note

^aMeans are adjusted for age, sex, and years of education

^b95% confidence intervals are presented in the brackets

^cParenting stress group is defined as normal (24-41 CGGs) vs. low/high (<24 CGGs and > 41 CGGs)

The second step for establishing true differential susceptibility is to test for the independence of the susceptibility factor (CGG repeat group) and the predictor (parenting stress group). The partial correlation between the CGG repeat-length group and parenting stress group, net of the covariates, was not significant (r = .008, 95% CI [−.022, .038]), satisfying the second requirement.

The third step is to test the associations between the susceptibility factor and each outcome; if these associations are significant, that would rule out differential susceptibility. We examined biserial correlations between CGG repeat-length group and each outcome. Partial correlations, net of the covariates, between the CGG repeat-length group variable and each outcome variable were not significant (r ranging from −.003 to −.026, with the lowest lower confidence limit being −.056 and the highest upper confidence limit being .025), satisfying the third requirement.

The fourth step is to examine the regression plots with reference to the prototypical figures depicted in Belsky et al. [2007]. As shown in Table 4, regression lines of the two susceptibility subgroups - low/high CGGs vs. normal CGG repeats - crossed each other, indicating significant crossover interactions. The form of these interactions matches Figure 1a in Belsky et al. [2007], which is the specific form of interaction effects indicative of differential susceptibility.

The fifth step is to test the specificity of the effect, namely to demonstrate that the current model is not replicated with other outcomes and other susceptibility factors. Regarding other outcomes, we have already shown that the interactions between the square term of the CGG repeat length and parenting stress group were not significant for anxiety, depression, or age at menopause. These results supported the specificity of the model for other outcomes. Regarding the specificity of the effect for other susceptibility factors, we ran the same regression models for the five outcomes using the APOE ε4 allele, which is known to be associated with neurodegenerative disorders including Alzheimer’s disease [Bertram et al., 2008; Vemuri et al., 2010). The regression results showed that there were no significant interactions between parenting stress group and the presence of the APOE ε4 allele. These results support the specificity of the model for other susceptibility factors. Together, these results met the fifth and final requirement for establishing differential susceptibility, i.e., the specificity of effects.

Discussion

Very little attention has been focused on the phenotypic associations of the FMR1 gene across the full range of CGG repeats (not including FXS). The results of this study suggest that there may be much to be learned by treating FMR1 CGG repeats as a quantitative (rather than a categorical) variable to more fully probe genotype-phenotype correlations. Such analyses may reveal phenotypic variations within the range that is currently considered “normal”.

Gene by Environment Interaction: FMR1 CGG Variability by Environmental Stress

Although preliminary, the present study suggested that the association between genotype (measured by the number of CGGs in FMR1) and phenotype varied depending on the level of environmental stress exposure, a gene by environment interaction. The particular form of this interaction reflects what has been variably termed differential susceptibility [Belsky et al., 2007] or the flip-flop phenomenon [Lin et al., 2007]. Both of these conceptualizations, though drawing from different literatures, hypothesize that people with certain genotypes are more reactive to the environment whereas those who have other genotypes are less environmentally-reactive. According to Lin et al. [2007], “flip-flop associations may indicate heterogeneous effects of the same variant that are due to differences in genetic background or environment” (p. 531, emphasis added). As Belsky and Pluess [2009] note “some individuals are more affected than others by…environmental circumstances more generally. In particular, some individuals appear more susceptible to the adverse effects of unsupportive contextual conditions and the beneficial effects of supportive ones” (p. 907). FMR1 may be a gene that confers either advantage or disadvantage to individuals with low or high numbers of CGGs, depending on the level of environmental stress, while those with modal numbers of CGGs (around 30 CGGs) may thrive regardless.

In the present study, we measured environmental stress in the family context. We contrasted two groups: aging parents whose adult son or daughter had a range of developmental or mental health diagnoses (the high-stress parenting group) and aging parents whose children did not have disabilities (the low-stress parenting group). Among those who had around the modal number of CGG repeats, the two groups of parents did not diverge in health or functioning. However, parents who had either low or high numbers of CGG repeats and who experienced stressful parenting had greater health and functional limitations than the modal CGG group. In contrast, parents with either high or low numbers of CGGs whose adult children were non-disabled and healthy had fewer health and functional limitations than the modal group. These differences were evident at an average age of 71 and may reflect cumulative effects of stressful parenting across the life course, the poor health in midlife of their son or daughter with disabilities, and the rippling effects on their relationships with their non-disabled adult children. Parenting a child with disabilities across multiple decades is a uniquely valid indicator of stress exposure.

Whereas it is not difficult to understand why aging parents who have a 40-year old child with chronic disabilities experience their family context as stressful, in what sense can the comparison group (with non-disabled adult children) be conceptualized as having a supportive family context? Considerable past research has shown that adult children provide critically important emotional and instrumental support to their aging parents, resulting in better parental health and well-being in old age [Merz et al., 2009; Silverstein et al., 1994; Ward, 2008]. Indeed, in the present study, those with the very best outcomes are environmentally-susceptible aging adults who have healthy adult children (and hence can expect their adult child to provide support to them if needed). Those with the poorest outcomes in the present study have the opposite profile -- they are environmentally-susceptible aging adults who have adult children with disabilities and thus continuing need for support from them, even at the time when their own needs for support and care are likely to intensify.

Furthermore, these aging parents of adults with disabilities received less support from their non-disabled children than those in the comparison group, suggesting rippling effects of the child with disabilities on the family context. In contrast to the comparison group, they expected less support from their non-disabled children if they were to fall ill, they were less likely to feel able to ask their non-disabled children for a loan of money if needed, they felt less loved and cared for by their non-disabled children, and they lived at a greater geographical distance from their non-disabled children than those in the comparison group.

Differential Susceptibility and Flip Flop Phenomena associated with FMR1 Variability

The pattern of susceptibility or reactivity to stress in the family context that we observed in parents with low or expanded numbers of CGGs in the present study has been reported in earlier studies of parents with expanded CGGs. For example, in our earlier work with a different population than the present study [Seltzer et al., 2012b], we reported that FMR1 premutation carriers who were exposed to the highest levels of stressful life events in the previous year had the most elevated levels of anxiety and depression, and the most abnormal cortisol profiles, whereas those who were not exposed to any stressful life events the previous year had the lowest levels of these outcomes, but this was only true for those with certain numbers of (90-105) CGGs. We interpreted this as evidence of differential susceptibility, and it also would fit the definition of the flip-flop phenomenon.

Relatedly, Hunter et al. [2012] investigated how the CRHR1 (corticotropin-releasing hormone type 1 receptor) gene, which regulates the hypothalamic-pituitary-adrenal axis (HPA) and thus cortisol, interacts with environmental stress among women. Among female FMR1 premutation carriers who had a child with FXS, having increasing copies of one allele in the CRHR1 gene (the T allele, rs7209436) was associated with increasing social phobia scores, but among women who did not have a child with FXS, the effect was the opposite – having increasing copies of the T allele was associated with decreasing scores on the social phobia measure. The authors concluded that the T allele of rs7209436 is a risk factor for social phobia in FMR1 premutation carrier mothers who are exposed to chronic parenting stress, but it is a protective factor in women who do not have this type of stress exposure. This was interpreted by Hunter et al. [2012] as an example of the flip-flop phenomenon.

Future research with other populations should evaluate whether these findings are replicated. If so, then the FMR1 gene could be seen in a different light, namely that it is a gene that interacts with environmental stress to confer either advantage or disadvantage to those with high or low numbers of CGG repeats in addition to being the cause of FXS when the gene is

Limitations and Future Directions

The present study was not without limitations. There were a relatively small number of individuals at the low (<24 CGGs) and high (> 41 CGGs) ranges of CGG repeats (1300 and 311, respectively, out of a total of 5628 individuals in the present study), and thus larger samples would provide a more robust assessment of the patterns of gene by environment interactions portrayed here. Further, the phenotypic data were collected via survey methods, and warrant replication via direct clinical assessment. The Wisconsin Longitudinal Study represents a single birth cohort and is largely of European descent, limiting generalizability. For females in the study, it would have been advantageous to have access to activation ratio data in order to select the allele to analyze based on this biological factor. However, the WLS does not have such data and thus we had to select the allele to analyze based on past research in the FMR1 field [Hunter et al., 2010]. For all of these reasons, the present data should be viewed as preliminary and suggestive, warranting confirmation in independent samples.

Additionally, there was an inconsistency in findings between this study and our earlier research on low CGGs in FMR1 [Mailick et al., 2014]. In that study, we observed a significant association between having low CGGs and the probability of having a child disabilities, but in the present study we found no significant association (in step 2 of the tests for differential susceptibility). This apparent inconsistency might be due to methodological differences between the two studies (i.e., the earlier study treated low CGGs as a dichotomous variable [low versus normal CGGs] while this study treated CGGs as a continuous variable; the earlier study excluded women who were heterozygous for low or normal CGGs while this study included heterozygous women; and the previous study excluded those with gray zone or premutation expansions while the present study included the full range of CGGs in FMR1 (not including FXS). Nevertheless, the inconsistency in findings warrants careful consideration of explanations other than methodological differences.

One alternative explanation is that it is possible that there is a common factor that predisposes individuals to both the likelihood of having a child with a disability and having low or high numbers of CGG repeats. If such a common factor is discovered, that would cast the differential susceptibility interpretation into doubt. For example, it is possible that the number and location of AGGs in the CGG tract predisposes to both, but data on AGGs were not available in the present study. Although the substantial heterogeneity in the range of disabilities in the children of study participants argues in favor of stress exposure as an explanation rather than a common genetic factor, this is an important priority to address in future research.

Future research with larger samples of individuals with low or high numbers of CGG repeats should explore how additional family factors may influence the associations reported here, such as the number of children in the family, additional caregiving roles of the aging parents, and the severity of the child’s disability.

Conclusions

In conclusion, the present study is unique because it was based on genetic data obtained on a non-clinical population. It is unbiased in that neither study participants nor data collectors were aware of their FMR1 CGG repeats nor of the interest in linking this genotype to phenotypic characteristics. It is an example of how population genetics can advance understanding of individual-level health and functional limitations with implications for precision medicine. As genetic data are increasingly being collected in population-level (vs. clinical) studies, the opportunities for replication of the present research will increase. Future exploratory research on the full range (including low numbers) of repeats in other trinucleotide repeat disorders, such as Huntington’s disease, spinal bulbar muscular atrophy, and myotonic dystrophy, may also be productive. Finally, this study illustrates how research that begins with a focus on a rare genetic condition (such as FXS) can ultimately lead to insights about the general population and contributes to our understanding of how genetic differences between people shape the way they respond to their environments.

Figure 3.

Component plus residual of the number of health symptoms, and lowess curve fits of CGG repeats, adjusted for age, years of education, and gender.

Figure 4.

Component plus residual of limitation in physical activities, and lowess curve fits of CGG repeats, adjusted for age, years of education, and gender.

Figure 5.

Component plus residual of limitation in climbing flights of stairs, and lowess curve fits of CGG repeats, adjusted for age, years of education, and gender.