Associations between measures of oestrogen exposure and severity of Fuchs endothelial corneal dystrophy

Study design

Clinic-based cross-sectional study of measures of oestrogen exposure in individuals with mild and severe FECD. Strengthening the Reporting of Observational Studies in Epidemiology cross-sectional reporting guidelines were used.15

Participant recruitment

Participants were recruited from the cornea subspecialty clinic of the principal investigator (PI) of this study (SPP) at the Ross Eye Institute, University at Buffalo, between July 2019 and November 2021. All participants were >55 years old, and women were postmenopausal. Individuals with a history of or signs of bilateral complex cataract surgery or other intraocular surgery, other than uncomplicated cataract surgery with posterior chamber intraocular lens (completed >1 year prior to enrolment), were excluded from the study. Only one eye of a patient was needed to meet the eligibility criteria for that individual to participate in the study. The study protocol included matching participants with mild and severe FECD for age (±5 years) and sex. However, these criteria could only be loosely maintained after March 2020 due to the COVID-19 pandemic. To reduce the need for repeat visits to the clinic, interested individuals were recruited at the time of routine clinic visits, thus matching by age and sex was not as rigorous after the first 10 participants. Because we were looking to identify measures of oestrogen exposure relating to disease severity, not presence of disease, we excluded those with absence of FECD.

Data collection

Each enrolled participant was administered a questionnaire (online supplemental appendix 1) assessing demographic and anthropometric indices, as well as smoking history, medical history, family history, ultraviolet light exposure and reproductive history (women only). Ultraviolet light exposure score was calculated based on participant responses to survey questions “how often do you wear a sun-blocking hat when outside” and “how often do you wear contact lenses, glasses, or sunglasses when outside”. Participant responses ‘always’, ‘sometimes’ or ‘never’ corresponded to scores of 2, 1 and 0, respectively, and responses to the two questions were summed (possible scores 0–4). For women, reproductive lifespan was calculated as the interval between age at menarche and menopause. Estimated lifetime oestrogen exposure was calculated in months as reproductive lifespan – duration (in months) breastfeeding – (1.5)(#pregnancies without breastfeeding) + months of postmenopausal hormone replacement therapy. If women did not breastfeed following a pregnancy, 1.5 months were subtracted to approximate the typical delay before ovulatory cycles resume.

Each participant had an eye examination by slit-lamp biomicroscopy by the PI of the study (SPP). FECD was graded using a modified Krachmer grading scale,16 and participants with mild (grade 1–2; non-confluent guttae) or severe (grade 5–6; confluent guttae>5 mm diameter with or without corneal oedema) FECD were included in the study. While primary grading of mild FECD was by slit-lamp biomicroscopy examination, presence of guttae was confirmed by at least one classic gutta observed by specular microscopy (Konan Specular Microscope X, Konan Medical) in any one of the five imaging fields of the microscope. In addition, the central field was required to have less than 50% of area occupied by guttae to be considered mild FECD. There were no imaging criteria for grading of severe FECD because it was easily observed and graded by slit-lamp biomicroscopy. For all participants, corneal thickness was measured (DGH pachymeter, DGH Technology, Inc), and specular microscopy (Konan Specular Microscope X) and corneal Scheimpflug tomography (Pentacam HR, Oculus) imaging were obtained. Tomographic features of advanced FECD in the central 4 mm of the cornea were evaluated on the ‘4 Map Refractive’ output of the Pentacam HR according to established criteria.17 Patients had blood drawn for laboratory testing (Kaleida Health, Department of Pathology and Laboratory Medicine, Buffalo, New York, USA) to determine serum total estradiol, serum-free estradiol (liquid chromatography tandem mass spectrometry; equilibrium dialysis) and sex hormone binding globulin (SHBG; electrochemiluminescence immunoassay). Patients also underwent bone densitometry testing using quantitative calcaneal ultrasound (Achilles Express, GE Healthcare) with measures of broadband ultrasound attenuation (BUA) and speed of sound (SOS), measures which were averaged from readings of both right and left calcaneal bones.

Blood was drawn for genotyping to determine trinucleotide repeat expansion length in the TCF4 gene that is commonly associated with FECD. TCF4 repeat length was considered expanded if >50 repeats were present. Genomic DNA was extracted from buffy coat preparations using the Puregene Blood Core Kit (QIAGEN, Germany, Cat. #158023). DNA concentration was quantified using a NanoDrop ND-1000 Spectrophotometer (NanoDrop Technologies, USA). DNA was subjected to PCR (Platinum PCR Supermix, Invitrogen, Carlsbad, California, USA) using primers flanking the trinucleotide repeat of the TCF4 gene (F: 5′-aatccaaaccgccttccaagt-3′; R: 5′-caaaacttccgaaagccatttct-3′). PCR products were separated by electrophoresis on agarose gels (Lonza, Basel, Switzerland) to make initial determination of allele sizes. Amplified PCR products were also sent to Azenta (Burlington, Massachusetts, USA) for fragment length analysis, with resulting data analysed in Peak Scanner software (Applied Biosystems, Foster City, California, USA).

For five samples with inconclusive data on allele size by PCR, Southern blot was performed according to published protocols.18 Briefly, DNA was digested with EcoRI (New England Biolabs, Ipswitch, Massachusetts, USA) and resulting fragments were separated overnight by agarose gel electrophoresis. The gel was incubated in denaturation solution (0.5 N NaOH, 1.5 M NaCl) and washed in neutralisation buffer (0.5 M Tris–HCl, 1.5 M NaCl). The DNA was transferred to a nylon membrane (Roche, Basel, Switzerland) and UV crosslinked. A 392 bp ³²P-labelled DNA probe designed against the TCF4 sequence adjacent to the CTG repeat was incubated with the membrane to detect the TCF4 sequence and associated CTG trinucleotide repeat expansion. The membrane was washed and imaged using a GE Typhoon phosphorimager (Boston, Massachusetts, USA).

Statistical analyses

Prior to the initiation of this study, which was prior to the COVID-19 pandemic, effect size calculations had been performed for each measure of oestrogen exposure for an anticipated enrolment of 65 pairs of matched participants. The effect size calculations are presented in online supplemental appendix 2.

Descriptive summaries and exploratory statistical comparisons of potential risk factors associated with FECD severity, by severity category (mild vs severe; tables 1 and 2) and by sex (table 3), were conducted using χ² tests or Fisher’s exact tests for categorical risk factors, depending on whether all categories had    or not, respectively, and t-tests for quantitative risk factors. Due to the small sample sizes and the exploratory nature of these comparisons, raw p-values, calculated as descriptive measures, are reported in tables 1–3 without adjustment for multiplicity or inflation in frequentist type I error probabilities. Our main inferential analyses focused on Bayesian posterior probability-based uncertainty quantification to infer the associations of various risk factors with FECD severity, after controlling for sex differences, TCF4 genotypes and multiple confounding covariates. Specifically, we used a set of Bayesian logistic regression models to predict FECD severity (mild=0, severe=1) based on linear three-way interactions of each oestrogen exposure measure (   measures with a separate model for each measure), sex (for measures applicable to both sexes) and TCF4 genotype as covariates. Additionally, the models included effects of confounding covariates (race, smoking, current body mass index (BMI), BMI age 18, current height, height age 18, current weight and weight age 18). To address the high dimensionality of the explanatory variables relative to the moderate sample sizes, we employed weakly informative priors (online supplemental appendix 2) and fitted the models using Markov chain Monte Carlo computational methods.19 The weakly informative priors for model parameters enable shrinkage of the estimates (posteriors) of small effects towards zero. This approach provides implicit control over the frequentist type I error across data replicates.20 21 However, we emphasise the Bayesian nature of our inferences from the main statistical analyses, evaluated through posterior probabilities conditioned on the observed dataset (as opposed to sampling variability over data replicates).

From the fitted models, we inferred ORs of having severe FECD for a 1 SD increase in the associated oestrogen exposure measure, separately for each sex and TCF4 genotype-level combinations to understand sex and genotype-specific effects of the exposure on FECD severity. Statistical significance was determined at 95% (Bayesian posterior) probability levels.

There were a few missing data points in our analysis. One participant declined to provide a blood sample (missing total estradiol, free estradiol, SHBG and TCF4 repeat status). Another participant declined to disclose weight and BMI current and at age 18. TCF4 analysis was inconclusive for a third participant. Furthermore, serum total estradiol concentration in some patients was low and left-censored. The missing observations for the covariates and the censored observations for serum total estradiol concentration were multiply imputed (n=100 imputed data sets)22 23; and inferences from the imputed data sets were obtained by combining the posterior draws from all imputed data sets prior to computing the (mixture) posterior summaries (posterior median and 95% equi-tailed credible intervals). Detailed description of the statistical methods can be found in online supplemental appendix 2. The computations involving multiple data imputations and Bayesian model fitting were performed using statistical software R and probabilistic programming language Stan.

Characteristics of the study population by FECD severity

There were 43 participants in this study, of whom 23 had severe FECD. Mean age was similar between the mild and severe FECD groups (mean±SD: 66.8±9.2 and 71.4±8.8 years, respectively). The racial mix of the participants (88.4% white participants and 11.6% black participants) was representative of our local population in Erie County, New York, USA. Characteristics of the study population are shown in table 1.

Additional testing was performed to objectively measure FECD severity and reinforce FECD severity categorisation. These analyses included one eye of each participant: the eligible eye where only one eye met study criteria and the right eyes only where both eyes were eligible. (Analyses with left eyes showed similar results.) Pachymetry and Scheimpflug imaging corneal tomographic signs of advanced FECD were more frequently seen in the severe compared with mild groups (table 1).17 TCF4 repeat expansion was also more frequently seen in participants with severe FECD.

Multiple additional factors were compared between the study groups (table 1). Significantly more participants with severe disease had hypertension and pseudophakia. While pseudophakia may impact corneal oedema, it is not known to influence the extent of guttae as used for differentiation of the mild and severe FECD groups in this study. Serum-free estradiol was significantly higher in the severe FECD group compared with the mild FECD group. Reproductive history (women only) showed fewer total number of pregnancies in those with severe FECD (table 2). Oestrogen measures by FECD severity in women showed a statistically significant difference in bone density by SOS only. Because of a potential role of ultraviolet light exposure and FECD,24–26 we also compared ultraviolet light exposure scores by FECD severity in men (mild: 3.7±1.1; severe: 2.6±1.1; p=0.149) and women (mild: 2.1±0.7; severe: 2.5±1.1; p=0.166) and found no significant differences.

Characteristics of the study population by sex

We further characterised our study population by sex (table 3). There were 11 men and 32 women. Men were significantly taller and heavier, both currently and at age 18. Men also had significantly higher total and free estradiol, SHBG, and BUA values compared with women.

Factors predictive of FECD severity

We evaluated the interactions between sex, TCF4 genotype and each oestrogen exposure measure for predicting FECD severity while adjusting for confounding variables (race, smoking, current BMI, BMI age 18, current height, height age 18, current weight and weight age 18). Figure 1 visualises the findings with grouped forest plots (estimates and credible intervals are also provided in online supplemental appendix 2). For each oestrogen exposure measure except SOS, the association with severe FECD had a positive trend, with an increase in the exposure level resulting in a mostly positive (95% credible intervals) change in log odds of severe FECD. A few statistically significant (95% credible intervals excluding zero) associations were identified. Specifically, in women with the TCF4 repeat expansion present, SHBG demonstrated a significantly positive association with severe FECD. This association was not significant in women without the TCF4 repeat expansion. In addition, an increase in % free estradiol resulted in a significantly increased odds of severe FECD in women, but only in those without the TCF4 repeat expansion. A similar measure, free estradiol (in pg/mL) was also significantly positively associated with severe FECD, but only when the groups with and without TCF4 repeat expansion were combined, and only in women. Larger number of months of lifetime oestrogen exposure (measure for women only) was also associated with increased odds of severe FECD independent of TCF4 repeat expansion status. There were no significant associations between FECD severity and oestrogen measures for men.