Prenatal famine exposure and estimated glomerular filtration rate across consecutive generations: association and epigenetic mediation in a population-based cohort study in Suihua China

Effect of prenatal famine exposure on kidney function in two cohorts

Table 1 summarized mean levels of study covariates at baseline and follow-up, respectively. At baseline, age, drink, and low working strength rate differed in F1 generation between two groups, whereas these covariates did not vary significantly in the F2 generation. And F1 and F2 generations in famine group had significantly higher FPG, 2-hPG, SBP, CRE and lower eGFR than those in the non-famine group, respectively. And the rates of eGFR<60 in the non-famine and famine groups were 12.76% and 23.21%, 14.70% and 16.74% in F1 and F2 generations, respectively. And then we applied the generalized linear mixed model to investigate the association between prenatal famine exposure and the risk of eGFR<60 in adulthood. After taking into account of the potential confounders, the odds ratios for eGFR<60 in two generations were exhibited in Table 2. In the F1 generation, the famine group had 1.80-fold (95% CI: 1.10, 3.36) higher risk compared with the non-famine group. In the F2 generation, however, parental exposure to famine was not shown to be associated with a higher rate of eGFR<60 which was probably due to the relatively young age in the F2 offspring at follow-up.

Moreover, Table 3 exhibited the results of linear mixed regression models which investigated the association of prenatal exposure to famine with ΔeGFR in two consecutive generations. The analyses results showed exposure to famine was associated with a β:-8.32 (-11.51, -5.12) lower ΔeGFR in the F1 generation and β:-6.11 (-11.88, -0.43) lower ΔeGFR in F2 generation after adjustment for age, sex, education, smoking, drinking, SBP, 2-hPG, working strength, daily energy intake, sport and BMI.

Effect of prenatal famine exposure on nutrition-sensitive DMSs in two consecutive generations

In the F1 generation, 961 differentially methylated sites (DMSs) (|Δβ|>0.05 and adjusted P<0.05) including 601 hypermethylated and 360 hypomethylated DMSs were identified. In the F2 generation, 503 DMSs (|Δβ|>0.05 and adjusted P<0.05) including 146 hypermethylated and 357 hypomethylated DMSs were identified. We constructed a genome-wide map of all autosomal DMSs (Figure 1). And only 19 of 961 F1 DMSs showed concordant alterations in F2 cohort, which may suggest that these DNA methylation alterations after prenatal famine exposure might be subject to intergenerational transmission. According to the annotation, 17 DMSs were physically located within 16 unique genes including genes with a critical role in kidney development (CUX1, PPARGC1A) as well as glomerular protection (ELMO1) and renal cell survival (AGTR1) (Supplementary Table 4).

Nutrition-sensitive DMSs correlated with ΔeGFR in two consecutive generations

Partial correlations revealed that among 19 methylated CpGs, 4 DMSs (cg09762515, cg13528513, cg17160506, cg20906621) were significantly correlated with ΔeGFR in F1 generation (Figure 2), 6 DMSs (cg13528513, cg13598358, cg17160506, cg20906621, cg21532512, cg23008606) were significantly relevant to ΔeGFR in F2 generation (Figure 3). Adjustment for key demographic factors such as age and sex overall resulted only in minor changes in the association between DMSs and ΔeGFR.

Key nutrition-sensitive DMSs were validated their differences in additional samples

Two DNA fragments from AGTR1 and one DNA fragment from PRKCA were analyzed, and the site-specific methylation level within each DNA fragment was quantified by targeted bisulfite sequencing to validate differential DNA methylation in an additional sample set. PCR primers were designed using the Methylation Primer software based on the bisulfite converted DNA. As shown in Supplementary Table 5, three DMSs were validated in 390 F1 participants and 210 F2 participants.

Mediation effects of methylation sites on the famine-ΔeGFR association

Mediation-analysis step 1 was conducted to investigate the mediator role of these three CpG sites in F1 generation in the correlation between F1 famine exposure and F1 ΔeGFR. As shown in Table 4, the total effect showed prenatal exposure to famine was associated with lower ΔeGFR in F1 generation after adjustment for parental characteristics. The mediation effects of methylation alteration at cg13528513, cg20906621 and cg17160506 were 23.20, 26.52 and 21.28 percent (P=0.031, P=0.016, P=0.009) of the total effect, respectively.

Mediation-analysis step 2 was performed to demonstrate the famine associated methylation variations could be transgenerational. As shown in Table 5, parental prenatal exposure to famine was associated with 1.6-2.0, 1.3-3.1 and 1.8-2.3 percent decrease in the methylation level at cg13528513, cg20906621 and cg17160506 in the F2 generation. Of the total effect, 28.66-29.31, 25.38-32.65 and 29.97-30.98 percent were mediated by methylation alterations at cg13528513, cg20906621 and cg17160506, respectively.

Mediation-analysis step 3 was implemented to the mediator role of these three CpG sites in F2 generation in the relationship between F1 famine exposure and F2 ΔeGFR. As shown in Table 6, our study uncovered that the association between parental prenatal exposure to famine and ΔeGFR in the offspring generation was significant. And the mediation effects of cg13528513, cg20906621 and cg17160506 on the ΔeGFR were significant (P=0.034, P=0.047, P=0.043).

The mediation-analysis in our study relied on the following negligible assumption of point identification. Since it is impossible to exclude the existence of missing variables in the relationship between confounding mediators and results, many studies do not satisfy the sequential ignorability hypothesis. Therefore, sensitivity analysis is recommended to determine to what extent the presence of confounders may affect the estimates. In our study, the results of sensitivity analysis showed that the proportion of residual variance (R^2*) and the proportion of original variance (R^2~) explained by ellipsis confusion were 0.040 ~ 0.160 and 0.033 ~ 0.112, respectively.

Participants and selection

The participants were recruited from the Suihua rural region of Heilongjiang Province, which is located in the northeast of China and suffered from severe famine from 1959 to 1961. At baseline, the participants were selected based on the household unit, including two generations (i.e. F1 parents and F2 offspring) in 2012. F1 parents born between Oct 1st 1959 and Sep 30th 1961 were classified as prenatal famine exposure group, and parents born between Oct 1st 1956 and Sep 30th 1958 and born between Oct 1st 1962 and Sep 30th 1964 were classified as the non-famine group. Correspondingly, F2 offspring were classified as famine and non-famine groups according to their parents' exposure status. And a total of 1173 households with F1 born between Oct 1st 1956 and Sep 30th 1964 were selected in the study at baseline. In 2017, the first follow-up survey was completed with a follow-up rate of 89.1%. 2909 participants, including 1881 F1 parent and 1028 F2 offspring with measurement of serum biochemical indexes were enrolled in this study. The Ethical Committee of Harbin Medical University approved the study protocol and written informed consent was obtained from all the participants. And this trial was registered at the Chinese Clinical Trial Registry as ChiCTR-ECH-13003644.

Questionnaire and anthropometric measurement

At baseline and follow-up, a standardized face-to-face interview by trained staff covered socio-demographic characteristics (e.g. age, sex, education level) and lifestyles (e.g. tobacco smoking, drinking, physical activity) and a food-frequency questionnaire (FFQ) which was used to collect dietary information on intake frequency and the average consumption of each food item in usual. And the availability and reproducibility of the FFQ were assessed in our previous study [19]. The daily energy intake of every participant was estimated using the Food Nutrition Calculator (V1.60; Chinese Center for Disease Control).

The physical examination was conducted at the same time as the questionnaire. Systolic and diastolic blood pressure were obtained with the participants in the sitting position. Two consecutive blood pressure measurements were conducted after five minutes of rest. We also measured participants' height without footwear and weight in light clothing and calculated the body mass index as weight divided by the square of height. Fasting blood samples and postprandial blood samples were collected using for laboratory measurements. Fast plasma glucose (FPG), 2-hour postprandial glucose (2-hPG), blood urea nitrogen (BUN), uric acid (UA), and serum creatinine (CRE) were determined using an automatic biochemical analyzer (Hitachi 7100; Hitachi Ltd). Hypertension was defined as SBP/DBP≥140/90 mmHg or current therapy for definite diagnosis of hypertension. T2D was defined as FPG≥7.0 mmol/L or 2-hPG≥11.1 mmol/L or current therapy for definite diagnosis of T2D. The eGFR values were calculated from CRE, age and sex by using the chronic kidney disease epidemiology collaboration equation (CKD-EPI equation):141×(CRE/0.9)^-0.411×(0.993) ^age (Male, CRE≤0.9); 141×(CRE/0.9)^-1.209×(0.993) ^age (Male, CRE>0.9); 144×(CRE/0.7)^-0.329×(0.993) ^age (Female, CRE≤0.7); 144×(CRE/0.7)-^1.209×(0.993) ^age (Female, CRE>0.7). And ΔeGFR was defined as eGFR at baseline - eGFR at follow-up.

Association between famine exposure and ΔeGFR

Continuous data were expressed as the mean ± standard deviation, and categorical data were presented as proportion. Association between prenatal exposure to famine and ΔeGFR, in both generations, was examined by using a linear mixed model. In this model, the family number was the random factor and famine exposure (yes or no) was the fixed factor. We adjusted for the following confounders: age (continuous), sex (male or female), education level (≤high school or > high school), energy intake (continuous), smoking (never, past, current), drinking (yes or no), working strength (low, medium, high), SBP (continuous), 2-hPG (continuous) and BMI (continuous).

Genome-wide DNA methylation profiling

Blood samples collected from 46 families in 2012 (92 adult F1 participants (n=46 famine; n=46 non-famine) and 46 F2 offspring (n=23 famine; n=23 non-famine)) were measured genome-wide DNA methylation profiling using the Illumina Infinium HumanMethylation850K BeadChip (Illumina Inc, USA). The basic characteristics of these subjects are listed in Supplementary Table 1. First, DNA was isolated from whole blood leukocytes using DNeasy Blood and Tissue Kit (Qiagen) and was stored at -80 °C for further research. Approximately, 500 ng of genomic DNA from each sample was utilized for sodium bisulfite conversion with EZ DNA methylation Gold Kit (Zymo Research, USA) leaving methylated cytosine unchanged and converting unmethylated cytosine to uracil. Then DNA samples were distributed uniformly in different arrays according to famine exposure status, generation and sex, to minimize batch (microarray chip) effects.

Data processing and statistical analysis of Illumina 850K methylation data

Raw genome-wide methylation data were processed using the R package minfi and assessed for basic quality control metrics. In each sample, probes with insignificant detection p-value above background control probes (detection p-value > 0.01) were removed. And probes with an SNP in the single base extension site were excluded. Also, we removed heterosomal probes from the raw methylation values and retained unique sequences for 755,464 autosomal probes which were available for analysis. A total of 138 samples passed quality control metrics and were further considered for statistical analyses. The HumanMethylation850k BeadChip contained two assay types (type I and type II) which utilized different probe designs. To correct distinct profiles produced by these two assay types, we performed a beta mixture quantile normalization method (BMIQ) on the normalized data to adjust the β-values of type II design probes into a statistical distribution characteristic of type I probes. Methylation β-value (ranging from 1 to 0, corresponding to fully methylated and completely unmethylated sites, respectively) represented the ratio of the signals of the methylated allele to the sum of the signals of the methylated and unmethylated alleles.

Our report implemented the comparison of famine and non-famine groups in two steps: (I) famine versus non-famine in F1 and (II) famine versus non-famine in F2. We performed a multivariable regression model to differentiate the role of prenatal malnutrition on methylation from the potential confounders of sex, smoking and BMI at these methylation sites. Once prenatal malnutrition in the F1 generation or parental prenatal malnutrition in F2 generation played a significant role in DNA methylation variation, p-values would be significant. Then the p-values would be done correction of FDR with the Benjamini-Hochberg method (adjusted p-value =p-value × Max rank/ rank of p-value). And the adjusted p-value <0.05 was considered statistically significant.

Targeted bisulfite sequencing analysis of methylation level for target CpG sites

The whole blood samples of 195 families (390 adult F1 participants (n=194 famine; n=196 non-famine) and 210 F2 offspring (n=83 famine; n=127 non-famine)) underwent targeted bisulfite sequencing analysis for target CpG sites. The basic characteristics of these subjects are listed in Supplementary Table 2. After PCR amplification of target CpG sites and library construction, the products were sequenced on Illumina MiSeq Benchtop Sequencer (CA, USA). All samples achieved a mean coverage of > 1000X. Tested CpG sites were named as its relative distance to the transcriptional start site (TSS). The methylation degree of each CpG site was calculated as the percentage of methylated cytosine relative to the total cytosine. PCR was performed to amplify the three targeted DNA sequences and the relative distance from the TSS, amplification primers, and product size of these CpG sites could be found in Supplementary Table 3. For targeted bisulfite sequencing assays data, the independent variables such as sex, smoking, BMI, incorporating prenatal malnutrition status would be included in the multivariable regression model. Likewise, adjusted p-value <0.05 was considered statistically significant.

Methylation/ΔeGFR correlations in F1 and F2 generation

Mediation models were performed to evaluate the role of DNA methylation as potential mediators of the association between famine and ΔeGFR in F1 and F2 generation. Famine or parental famine-exposure status as the predictor (X), CpG sites the mediators (M), ΔeGFR the outcome (Y). Then three steps mediation-analysis was performed to demonstrate that these methylation levels of CpG sites were the mediator of F1 famine exposure on F1 ΔeGFR (step 1), transgenerational (step 2), and the mediator of F1 famine exposure on F2 ΔeGFR (step 3). And mediation-analysis were performed using R package lavaan to obtain the total effects and mediation effects. Sensitivity analysis was completed within the function, and mediating effects were estimated to explore the robustness of the results in the case of violation of the following negligible assumption. The sensitivity parameters were residual variance proportion (R^2*) and original variance proportion (R^2~).

All statistical tests were performed using the R 3.4.4 software (The R Foundation for Statistical Computing) and SPSS Statistics v23.0 (IBM Corp. USA) and p-value < 0.05 was considered statistically significant.