|Home | About | Journals | Submit | Contact Us | Français|
The fragile X mental retardation gene (FMR1) contains a CGG repeat sequence in its 5′ untranslated region that can become unstable and expand in length from generation to generation. Alleles with expanded repeats in the range of ~55–199, termed premutation alleles, are associated with an increased risk for fragile-X-associated primary ovarian insufficiency (FXPOI). However, not all women who carry the premutation develop FXPOI. To determine if additional genes could explain variability in onset and severity, we used a random-effects Cox proportional hazards model to analyze age at menopause on 680 women from 225 families who have a history of fragile X syndrome and 321 women from 219 families from the general population. We tested for the presence of a residual additive genetic effect after adjustment for FMR1 repeat length, race, smoking, body mass index, and method of ascertainment. Results showed significant familial aggregation of age at menopause with an estimated additive genetic variance of 0.55–0.96 depending on the parameterization of FMR1 repeat size and definition of age at menopause (P-values ranging between 0.0002 and 0.0027). This is the first study to analyze familial aggregation of FXPOI. This result is important for proper counseling of women who carry FMR1 premutation alleles and for guidance of future studies to identify additional genes that influence ovarian insufficiency.
The X-linked fragile X mental retardation gene (FMR1) contains a highly polymorphic CGG repeat in the 5′ untranslated region of exon 1 [Ashley et al., 1993; Verkerk et al., 1991]. The most common alleles contain roughly 30 repeats [Fu et al., 1991]. When mutated, this triplet repeat becomes unstable and expands with transmission from one generation to the next. To date, expansion of this repeat has been linked to several clinical disorders.
First, expansions of 200 repeats or more, termed full mutations, are associated with aberrant hypermethylation which results in a loss of FMR1 expression [Bell et al., 1991; McConkie-Rosell et al., 1993; Sutcliffe et al., 1992]. The subsequent loss of the protein product, FMRP, results in a mental retardation syndrome, fragile X (FXS) [Pieretti et al., 1991]. Males with full mutations tend to be more severely affected compared to females with full mutations due to the X-linked nature of the gene.
Second, repeats in the range of about 55–199, as defined by the American College of Medical Genetics [Sherman et al., 2005] and termed premutation alleles, remain unmethylated and are associated with increasing levels of the FMR1 transcripts and decreasing levels of FMRP [Kenneson et al., 2001; Primerano et al., 2002; Tassone and Hagerman, 2003; Tassone et al., 2000a,b]. Older males (>50 years) who carry the premutation are at an increased risk for fragile-X-associated tremor/ataxia syndrome (FXTAS), while premutation females are at an increased risk of fragile-X-associated primary ovarian insufficiency (FXPOI) [Hagerman et al., 2003; Sherman, 2000]. Neither of these disorders is seen in conjunction with FXS and is therefore thought to be caused by a toxic gain-of-function effect of increased levels of the expanded FMR1 transcripts.
Roughly 20% of women who carry premutation alleles experience premature ovarian failure, clinically defined as the cessation of menstrual periods before the age of 40 [Sherman, 2000]. This risk is about 20 times higher than that seen in the general population, with the highest risk being for premutation alleles in the range of about 80–100 repeats [Allen et al., 2007; Sullivan et al., 2005]. In the study of Allen et al., the mean age at menopause for carriers with 59–79 repeats was 48.5±0.7, for those with 80–100 repeats was 44.9±0.6, and those with >100 repeats was 47.5±1.2. These reduced menopause ages compared with 52.3±0.5 found among non-carriers. In addition, premutation females who are still cycling have increased levels of follicle stimulating hormone, an indicator of reduced ovarian function, compared to non-carriers [Hundscheid et al., 2001; Murray et al., 1999; Sullivan et al., 2005; Welt et al., 2004] and altered cycle characteristics [Allen et al., 2007; Welt et al., 2004]. Thus, it is recommended that the term “ovarian insufficiency” be applied to this condition to capture the observation that women who carry the premutation have traits associated with reduced ovarian function [Abrams, 2007; Welt, 2007].
The effect of the FMR1 premutation on reducing age at menopause is the most common single major gene effect known to date. However, not all women who carry a premutation allele experience FXPOI. Presently, neither the etiology of FXPOI nor the cause of the variation in phenotype is understood. We know that repeat size variation within the premutation allele range explains a significant proportion of the variation in FXPOI [Allen et al., 2007; Ennis et al., 2006; Sullivan et al., 2005]. Another likely source of phenotype variation is background genes that, in conjunction with the effects of the FMR1 alleles, could either decrease or increase the age of onset and severity of FXPOI. Large heritability estimates for the age at menopause support this hypothesis. For example, estimates of heritability of age at natural menopause range from 31 to 85% depending on the type of sample (twins, mother–daughter, sister pairs) and other sample attributes (e.g., age structure, geographical region) [de Bruin et al., 2001; Murabito et al., 2005; Snieder et al., 1998; Treloar et al., 1998; van Asselt et al., 2004; Vink and Boomsma, 2005].
This is the first study to examine heritability of age at menopause, a reflection of the level of severity of ovarian insufficiency, in the presence of a single gene effect, the FMR1 premutation, known to influence ovarian function. Here, we ask if there are residual additive genetic effects that influence onset and severity of ovarian insufficiency after adjusting for the effects of FMR1 repeat length. Our study was based on 230 families with a history of FXS as well as 219 families from the general population. For analysis, we used a random-effect version of the Cox proportional hazards model that allows for shared additive genetic effects within families [Pankratz et al., 2005]. This model takes variable age at onset into account and includes censored measurements, thus maximizing the information derived from our study population.
Study participants were ascertained from families with a history of FXS to enrich the sample with varying repeat lengths as well as from the general population (see Sullivan et al.  for a review of ascertainment protocols). Once a proband was identified, female relatives were also invited to participate in the study. Participants were aged 18–92 and had English as their primary language. All participants were asked to complete a reproductive history questionnaire and provide a biological sample, either buccal or blood, for repeat-length determination. For our analysis, we had reproductive history information from 680 women from 225 FXS families and 321 women from 219 general population families. The protocols and consent forms for ascertainment were approved by the Institutional Review Board at Emory University.
All study participants were asked to complete a reproductive history questionnaire. Depending on the participant’s availability, the questionnaire was administered in person, over the telephone, or through the mail. The questionnaire included items regarding demographic information, such as date of birth and ethnicity, as well as information on potential confounders such as body mass index (BMI) and smoking history. The bulk of the questionnaire pertained to menstrual cycle history and hormone medication use.
Menopause is defined as the cessation of menses for at least one year. However, pinpointing the exact age at which menopause occurs can be problematic due to the duration of the transition as well as the common use of hormone medication at the start of menopause symptoms which might allow the woman to continue cycling. Thus, age at menopause for each study participant was determined in three ways using the same strategy outlined in Sullivan et al.  (Table I). The first definition of age at menopause uses the self-reported age at menopause without taking into account hormone use. This definition scheme allowed us to define data points for 197 participants and censor data points for 804 participants. The second definition of age at menopause defines age at menopause as either the self-reported age at menopause or the age at which the participant started using hormone replacement therapy (HRT). This definition scheme was the most liberal and allowed us to define data points for 262 participants and censor data points for 739 participants. The third definition of age at menopause is the most conservative, with participants being censored at the start of HRT or oral contraceptive use. This definition scheme allowed us to define data points for 161 participants and censor data points for 840 participants.
All participants were asked to provide a biological sample, either buccal or blood. DNA was extracted using the Qiagen QiAmp DNA Blood Mini Kit (Valencia, CA). A fluorescent-sequencer method is used to determine FMR1 CGG repeat length [Meadows et al., 1996]. Briefly, fluorescent-labeled primers are used to polymerase chain reaction amplify across the repeat region and the resulting product is run on an automated sequencer. Repeat lengths up to 90 can be determined with this method. In the event that a single band was detected in females, indicating either homozygous status or the presence of a larger repeat band from the second allele, an alternative polymerase chain reaction-based, hybridization technique was used [Brown et al., 1993]. For heterozygous females, the CGG repeat length from the larger repeat allele was used in subsequent statistical analyses.
We conducted statistical analyses using a random-effect version of the Cox proportional hazards model that allows for shared effects due to polygenes [Pankratz et al., 2005]. For the ith subject, we defined the hazard function for menopause at time t as
where λ0 represents an unspecified baseline hazard function, Xi represents a design vector for the fixed effects of FMR1 repeat length and additional confounders with related parameter vector β, and bi represents the subject’s random effect due to shared polygenes within the family. We assume that the additive genetic random effects among all subjects follow a multivariate normal distribution with mean zero and the variance–covariance matrix Σ, which we model using
where is the variance due to additive genetic effects and Φ is the kinship matrix (which depends on the familial relationships among subjects). Therefore, the introduction of bi allows for correlation in age of menopause among related subjects, with such correlation assumed to be due solely to polygene effects.
Using this model scheme, we tested the hypothesis that there are no additional polygene effects beyond FMR1 repeat size major gene effect that contribute to age at menopause ( ) by maximizing the likelihoods under both the null and alternative hypotheses. We then construct a likelihood-ratio statistic, which asymptotically follows a distribution under H0 [Self and Liang, 1987]. We rejected H0 assuming a significance threshold of 0.05.
We analyzed the data using age at menopause as the outcome variable, based on the three definitions outlined in Table I. For FMR1 repeat length, we modeled the predictor as either a continuous variable or a four-level categorical variable (non-carriers = <59 repeats, low premutation group = 59–79 repeats, middle premutation group = 80–100 repeats, and high premutation group = 101–199 repeats). This second approach accounts for the nonlinear relationship between age at menopause and repeat length [Allen et al., 2007; Sullivan et al., 2005]. These definitions, although somewhat arbitrary, were based on the risk to expand to the full mutation in one generation [Nolin et al., 2003]. We also used these to be comparable to our previous studies. To determine which parameterization of FMR1 repeat length best fit the data, we calculated the Akaike information criterion (AIC) and compared the models for each definition of age at menopause.
We initially examined potential confounders to assess significant differences among repeat classes using analysis of variances for continuous variables and χ2 analysis for dichotomous variables (Table II). Twenty-four participants did not self-report race, 28 were missing information on history of smoking, and 17 were missing BMI data. The mean age at time of interview differed significantly among repeat classes, although it was not a confounder. This finding is expected as younger participants are likely to have higher repeat sizes due to the expansion bias from parent to offspring. Race and ascertainment source also differed significantly among repeat groups, with the higher repeat groups being composed mostly of Caucasians from families with a history of FXS. Thus, we included race and ascertainment as potential confounders in the model. Smoking and BMI were not significantly different between classes. However, as these factors are known to affect age at menopause, we kept them in the model. Thus, within the random-effects Cox proportional hazards model, we adjusted for confounders consisting of race (dichotomous variable: 0 = non-White, 1 = White), history of smoking in “packyears” (continuous variable calculated using period of time the subject reported smoking in years multiplied by the number of cigarette packs smoked a day), BMI (continuous variable), and method of ascertainment (0 = ascertained from families with a known history of FXS, 1 = ascertained from families in the general population).
All analyses were run using the “kinship” statistics package [Therneau and Atkinson, 2007] in R 2.4.1.
We initiated our studies by confirming the association between FMR1 repeat size and age at menopause after adjusting for confounders, including race and ascertainment site, and for covariates known to affect age at menopause, namely smoking and BMI. Using the random-effect Cox proportional hazards model, we found statistically significant evidence for FMR1 repeat size measured as a continuous variable (Table III). With each increase in a single repeat, there was an increased risk of earlier age at menopause by approximately 1%. This finding was consistent across the various definitions of age at menopause. We also examined models that adjusted for repeat length as a four-level categorical variable with the expectation that this parameterization may fit the data better due to the nonlinear association between repeat length and age at menopause. As expected, we found that the mid-size premutation group had the highest risk of earlier age at menopause (about four times that of the non-carrier group), and low- and high-repeat size groups had an increased risk of about two times the non-carrier group (Table IV).
Next, we examined the variation in age of menopause due to residual additive genetic effects, after adjusting for the major gene effect due to FMR1 repeat length and confounders. Independent of our definition of age at menopause, we found significant evidence of an additive genetic component for age at menopause using either the continuous FMR1 predictor (Table III) or the categorical FMR1 predictor (Table IV). We found that the estimated variance due to additive genetic effects changed based on the definition of age at menopause and on the coding of the repeat-length covariate (continuous vs. indicator variables). Using self-reported age at menopause (which does not incorporate hormone use), we estimated the additive genetic variance component to be 0.55 (P = 0.0221) using the continuous FMR1 repeat-length variable (Table III) and 0.64 (P = 0.0040) using the categorical FMR1 variable (Table IV). Interestingly, the use of the definitions of age at menopause that incorporate information about the use of HRT increased the estimate of the additive genetic variance component. When the definition of age at menopause included the age at the start of HRT, the additive genetic variance component was estimated to be 0.82 (P = 0.0005) using the continuous FMR1 repeat-length variable (Table III) and 0.75 (P = 0.0002) using the categorical FMR1 variable (Table IV). Likewise, when we censored age at menopause at the initiation of HRT, the estimated additive genetic variance component increased to 0.96 (P = 0.0027) using the continuous FMR1 variable and 0.87 (P = 0.0023) using the categorical FMR1 variable.
We compared the AIC values of the models that included FMR1 repeat length as a continuous variable and as a repeat-length group categorical variable to determine which fitted the data best. For each definition of age at menopause, the categorical parameterization was better. The difference in AIC values for the model including the continuous vs. categorical variables was 18.52, 21.99, and 19.04 for the three definitions, respectively.
The Cox regression model does not allow for an estimate of heritability. Therefore, familial aggregation of age at menopause was quantified using measures described by Pankratz et al. . In particular, we can determine the estimated range of the hazard ratio for age of menopause among subjects by exponentiating the square root of the additive genetic variance estimate. In the case of FMR1 repeat length as categorical variables for the most “conservative” definition of age at menopause (i.e., censoring at start of HRT use), the estimate of the additive genetic variance component is 0.87, giving a subject-specific hazard ratio based on genetic relationships, that is, on average, 2.54 times larger or smaller than the overall hazard ratio for age at menopause. This indicates a substantial familial aggregation of age at menopause due to shared polygenes, even after adjusting for the major effect of the FMR1 repeat length.
In this study, we found significant evidence for an additive genetic component for age at menopause after adjusting for the influential effects of the FMR1 premutation allele. We took a comprehensive approach to maximize the information from a sample of 1001 women from 444 families using a random-effect version of the Cox proportional hazards model allowing for shared poly-genes. We were also able to include information on other sources of variation including race, history of smoking, BMI, and ascertainment source.
In our previous studies [Allen et al., 2007; Ennis et al., 2006; Sullivan et al., 2005], a significant nonlinear association between repeat size and ovarian insufficiency was identified. The repeat size alleles that carry the highest risk are those in the mid-range of ~80–99, not the highest premutation repeat sizes (i.e., 100–200 repeats). Those carriers with 80–99 repeats compared to noncarriers have increased rates of infertility, a seven-year reduction in mean age at menopause, and a consequently increased prevalence of premature ovarian failure (32 vs. 1% in the general population) that initiates at younger ages. Carriers of both smaller and larger premutation repeat sizes also suffer from ovarian insufficiency, but not to as great an extent. To incorporate this nonlinear effect, we parameterized repeat length as a four-level categorical using repeat size groups and found that it fit the data better than did the continuous variable. We suggest that the estimates of the additive genetic component are more accurate when the FMR1 major gene effect is adjusted using the indicator variables.
One of the important limitations of this study concerns the reliability of self-reported age at menopause. Pinpointing an event that has a long transition period and is somewhat ambiguous depending on symptoms is difficult. This is particularly true for a cross-sectional survey. Moreover, many women are prescribed hormone medication as soon as symptoms of menopause occur and may continue to cycle until medication is stopped. Some women on HRT may have had the ability to continue cycling naturally and some may not; HRT use masks this distinction. This, of course, complicates the ability to define a specific menopausal age. However, if it is measured similarly among those with and without the premutation, analyses will identify important patterns. To address the possible effects of the use of HRT in defining age at menopause, we conducted the analyses using the three definitions of menopausal age that incorporate hormone medication use. In all models, estimates of the additive genetic variance component were significant and ranged from 0.64 to 0.87, assuming a categorical modeling of FMR1 repeat length. The highest estimates of the familial genetic component were obtained using the most conservative definition of age at menopause that censored data at the initiation of HRT. It is possible that other nongenetic factors account for the increased estimate of familial aggregation when using this definition. For example, the use of HRT for menopausal symptoms may be a shared phenomenon within families.
This is the first study to analyze familial aggregation of ovarian insufficiency among families with a history of FXS. Overall, the models outlined in this study provide significant evidence that the onset of FMR1-associated ovarian insufficiency, as marked by age at menopause, is controlled in part by additive genetic effects. This finding is important for two reasons. First, the average age at menopause within a family should be taken into consideration in addition to repeat size when counseling a woman who carries the premutation. These two attributes should help to determine the woman’s risk for ovarian insufficiency that may interfere with fertility. Second, this evidence for additional genes that influence age of menopause motivates the next stage to research: identification of additional genes that are involved in ovarian function.
We thank the women who took the time and effort to participate in this study as well as Terry Therneau for his assistance in the statistical analysis. We also thank the members of the Fragile X Research Team for their help in conducting this project.
Contract grant sponsor: National Institutes of Health; Contract grant number: HD29909.