Placental telomere length shortening is not associated with severe preeclampsia but the gestational age

Overview of the study cohort

In Table 1, we present the phenotypic characteristics of 95 pregnancies with severe PE and 129 healthy controls that are matched in criteria including maternal age, height, BMI, blood type, ethnicity and weight gain between the severe PE cases and controls. There is no significant difference in the above-mentioned factors between the severe PE cases and controls, confirming the success of matching criteria in the study design. We excluded samples from smokers and samples with neonatal malformations. All patients with gestational diabetes mellitus (GDM) were found in the case group, so we also removed GDM samples from the analysis. From the correlation heatmap of all phenotypic variables (Supplementary Figure 1), significantly more mothers from the severe PE group have other pre-existing conditions including history of anemia (p = 0.0046) and chronic hypertension (p = 0.0013) or comorbidities like intrauterine growth restriction (IUGR) (p=4.04e-7) and ruptured membrane (p=0.0089). More control samples have macrosomia (p=0.031) than severe PE cases. As expected, the severe PE group has a smaller delivery gestational age (p < 4.05e-23): the average and standard deviation of the delivery gestational age in severe PE cases is 35.28 weeks +/- 3.01 weeks, compared to controls 39.33 weeks +/- 0.85. The distribution of absolute TL is right-skewed (Supplementary Figure 2), similar to observations of TL in other tissues [20]. We ln-transformed TL similar to others [21, 22], so that the TL is approximately Gaussian distribution and suitable for the downstream linear regression analysis. All the TL in the remaining report refers to ln-transformed TL unless noted otherwise. The average Ln-transformed TL in severe PE cases is 5.53, significantly higher than the average value of TL in controls 5.41 (p = 0.018) (Figure 1A).

Placental TL is not associated with severe PE but with gestational age

Although placental TL is longer in the severe PE group from the exploratory analysis (Table 1), such difference may be due to other confounding factors, such as delivery gestational age. To explore potential confounders of PE, we examined the effect of all other clinical variables on PE using logistic regression and reported the p-values of top variables in Figure 1B and Supplementary Table 1. Indeed, the P-values show that delivery gestational age has the strongest relationship with severe PE (p=4.25e-08, Figure 1B). Gestational age-adjusted TLs in each race, blood type, maternal age interval, and infant sex subgroup show no significant difference between severe PE and controls (Supplementary Figure 3). Moreover, when examining the contribution of all 25 phenotypic variables to TL, other demographic variables present no significant difference among racial variation, delivery gestational age shows the highest and most dominant proportion variance of 4.18% explained according to the ANOVA test, much higher than that of 2.09% from the 2nd ranked Latin ethnicity. Severe PE is not ranked among the top 10 variables (Figure 1C). Thus, the exploratory analyses coherently show the important relationship between gestational age and TL change in the third trimester.

To quantitatively delineate the relationship between TL and phenotypic factors including severe PE, we then constructed a full linear regression model by regressing placental TL on all 25 demographic and clinical variables and significant pairwise interactions among them, using both severe PE and control samples (n=224). This full model shows no significant difference (P = 0.462) in TL between severe PE patients and controls (Supplementary Table 2). We next determined the optimal model (Table 2), by applying the stepwise selection algorithm with the lowest Akaike Information Criterion (AIC) on the full model. Again, severe PE was excluded from the optimal model due to its lack of contribution to model fitness. Instead, placental TL shows a negative association with gestational age (beta = -0.03, p = 9.94e-4) according to the optimal model (Figure 1D and Table 2). The consistent decrease in TL is correlated with the active cell divisions in the placenta through the last trimester of pregnancy.

To better control the confounding effect of delivery gestational age on TL, we focused on the subset of severe PE patients (n=18) with comparable delivery gestational age (gestational age greater or equal to 38 weeks) to the control samples (n=128). Boxplot shows no significant difference (P =0.73) in TL between full-term severe PE and controls (Figure 1E). The linear regression model of TL on all 26 phenotypic variables on this subset (Supplementary Table 3) again shows a lack of association between severe PE and TL (P =0.65). Additionally, we performed a gestational week stratified (weeks 38, 39 and 40+) comparison between full-term severe PE samples (n = 18) and their 1-on-1 matched controls (n = 18) based on delivery gestational age, ethnicity, maternal age and BMI. Again, boxplots show no significant difference (P = 0.35, P= 0.28, and P= 0.88) of TL between cases and controls in all gestational week strata (Figure 1F). The optimal model using only severe PE cases still shows that delivery gestational age is negatively (beta = -0.029, p= 0.047) associated with TL (Table 3 and Figure 1D). In all, our analyses from many different perspectives all show that the shorter telomeres observed in severe PE patients are mostly associated with the gestational ages rather than severe PE itself.

Racial disparity is present in placental TL

Besides gestational age, the optimal model also revealed significant associations of placental TL to other demographic or clinical variables (Table 2). Among them, race shows significance second to the delivery gestational age. In particular, Latino mothers have longer placental TL than non-Latino mothers (beta = 0.281, P = 3.97e-3) in combined severe PE and control samples, after adjusting for all other confounders with linear regression (Figure 2A). To further decouple the effect of severe PE on race-associated TL differences, we separated the severe PE cases and controls and built optimal models within each category, respectively.

Interestingly, the trend of longer placenta TL among Latino mothers still exists among severe PE patients (Table 3 and Figure 2B), though slightly above the statistical significance threshold (beta =0.264, P= 0.112). On the other hand, among control samples, Latino is no longer associated with TL (Table 4 and Figure 2C). This indicates that severe PE samples are the main contributors to the observed positive association of Latino ethnicity to TL overall. We also observe some other apparent racial associations in severe PE and control samples, respectively. Among severe PE patients, Caucasian mothers have shorter TL in the optimal model (beta = -0.319, P= 0.009), and a similar shorter TL trend exists with African American mothers among the healthy controls (Table 4 and Figure 2C), although not statistically significant (beta = -0.444, P= 0.093). We also compared maternal age, gestational age, maternal education level, parity and gravida among different ethnic groups (Supplementary Figure 4). No significant difference was found, suggesting low confounding effects from these variables for observed ethnicity association to TL. While PE itself does not appear to be associated with telomere length in each ethnicity statistically, the trends of difference of TL between PE vs controls are different among the ethnicities after adjusting for gestational age, the most major determinant of TL (Supplementary Figure 3). This suggests that there is interaction between ethnicity and PE. Overall, the data show that there exist racial differences in placenta TL, and such differences may be dependent on the severe PE status.

Biospecimens

Placenta samples from 224 women (95 with severe PE and 129 full-term healthy controls), who delivered their babies in Hawaii between January 2006 and June 2013 were collected and analyzed in this study. Samples were delivered at Kapiolani Women and Children’s Hospital from and stored at the University of Hawaii Biorepository (HiBR). HiBR is one of the largest research tissue repositories in the pacific region, containing specimens from more than 9250 mother-child pairs at the time of sample collection. Severe PE was defined as sustained pregnancy-induced hypertension with blood pressure higher than 160/110 mmHg and urine protein and/or organ dysfunction according to the diagnosis of OBGYNs in Kapiolani Medical Center. For each severe PE case, a matched healthy control sample was identified using mother age, prepregnant body mass index (BMI) and ethnicity group as the matching criteria. All other clinical information was recorded and analyzed for potential confounding factors. Samples from smoking are removed due to the small size. Samples with gestational diabetes and neonatal malformations (which only occurred in the case group) were excluded from the data set, to avoid potential sources of confounding.

TL measurement

Placenta samples were obtained from the intermediate portion of the placenta after maternal and fetal membranes were trimmed away. The placenta tissue samples were stored at -80° C. The genomic DNA extracted from placenta tissue was performed using AllPrep DNA/RNA/Protein Mini Kit (Qiagen, USA) according to the manufacturer’s instructions. Briefly, 30mg of frozen placenta tissue samples were weighted in a 50 ml Falcon tube (VWR, USA) and mixed with 600 μl buffer RLT (Qiagen, USA) supplemented with 1% β-mercaptoethanol (Sigma, USA). The tissue was then homogenized with Tissue Ruptor II (Qiagen, USA) for 30 seconds at medium speed. The lysate was transferred to a 2 ml microcentrifuge tube and centrifuged for 3 minutes at 13,000g at 4° C. Then the supernatant was pipetted into an AllPrep DNA spin column (Qiagen, USA) and processed following Kit’s instruction. The DNA concentration and purity were evaluated by NanoDrop2000 spectrophotometer (Thermo Scientific, Waltham, USA) and the DNA integrity was investigated by electrophoresis in 1% agarose gel with ethidium bromide (Sigma, USA).

The average TL of placenta tissue samples was measured using the Absolute Human TL Quantification qPCR assay Kit (AHTLQ, ScienCell Research Laboratories, USA). Each sample was tested in duplicate with StepOne™ Real Time System (Applied Biosystems, USA) using 96-well reaction plates. The melting curve was observed to evaluate the specificity and efficiency of the qPCR reaction. The Comparative ∆∆Cq (Quantification Cycle Value) method was applied to calculate the relative average TL of target tissue samples [32]. The coefficient of variation (CV), the ratio of standard deviation over mean, is used as the metric to measure performance precision among all technically duplicated samples in qPCR. The maximum CV among samples is 0.105, smaller than the recommended cutoff threshold 0.11. A 100 bp region single copy reference (SCR) primer set on chromosome 17 was used as the reference. The absolute TL is calculated as the relative TL multiplied by the SCR TL.

Demographical, clinical and physiological data

A total of 26 demographical, clinical and physiological variables obtained from the University of Hawaii HiBR biobank were used in this study: mother’s age, height, pre-pregnancy BMI, parity, gravida, maternal education level high school, college, masters and unknown with college as the reference level, weight gain during pregnancy, blood type A, B, O, AB with AB as the reference level, African American race, Asian race, Caucasian race, Latino race and Pacific Islander with Asian race as the reference level, history of asthma, history of anemia, chronic hypertension, macrosomia, vaginal bleeding, IUGR, membrane ruptured, babies’ sex, gestational age at delivery, and sample group. Categories with no more than 5 patients (eg. amnion infection) were too low in frequency to draw any sounding conclusion, thus we removed them from our analysis. The missing values were imputed with the predictive mean match (PMM) algorithm from R package “mice” [33].

Statistical analysis

Two-sided Welch t-tests and Chi-squared tests were used to compare numeric and categorical characteristics in case and control groups. Logistic regression was performed to measure the correlations between severe PE and other factors. The proportional variance explained (PVE) values of all variables were calculated using eta-values from the 1-way ANOVA result. The PVE value describes the proportion of total variance in data that can be explained by a certain variable, demonstrating its importance [34]. The absolute TL was ln-transformed to reduce skewness and ensure normality. TL refers to ln-transformed placental TL unless stated otherwise.

A multiple linear regression model was constructed by regressing TL of 224 samples on all 26 variables and the most significant interaction between gestational age and infant sex. We refer to this model as the full model. To select the most relevant variables and improve model fitness, we applied the stepwise selection algorithm on the full model to search for the optimal model with the lowest Akaike Information Criterion (AIC), by adding or removing one variable at a time. This stepwise selected model is referred to as the optimal model, including 7 variables: mother’s age, gestational age at delivery, baby’s sex, blood type O, Latino ethnicity, Pacific Islander ethnicity and parity. To assess the impact of severe PE without bias from different gestational ages, we stratified the TL of severe PE patients exceeding 38 weeks of gestation (n = 18) and a one-on-one matched control subset (n=18) by gestational week at birth. Each sample from the control subset was matched to one full-term severe PE case based on the maternal age, race, gestational age and BMI. We then compared the TL in each gestational week group with a t-test.

Two multiple linear regression models were constructed on severe PE cases and controls respectively by regressing TL on all other variables. To select the most relevant variables and improve model fitness, we applied the stepwise selection algorithm on these two models again to search for the optimal models with the lowest Akaike Information Criterion (AIC), by adding or removing one variable at a time until AIC cannot be further reduced.

All analyses were conducted with different packages in R software version 4.0.2 [35]. “tidyr”, and “dplyr” were used in data cleaning [36, 37]; “mice” was used to impute missing values [33]; “ggplot2”, “gridExtra”, “ggsignif” and “corrplot” were used in visualization [38–41]. All R script used in this project is in repository.