Association of Maternal Stress and Social Support During Pregnancy With Growth Marks in Children’s Primary Tooth Enamel

Key Points Question Can exfoliated primary teeth be used to identify children exposed to psychosocial risk and protective factors during their prenatal and perinatal life? Findings This cohort study of 70 children from the Avon Longitudinal Study of Parents and Children birth cohort found evidence that teeth may record early-life events. Children exposed to prenatal maternal depression or anxiety had wider neonatal lines, a marker of enamel growth, and children with perinatal maternal social support had narrower neonatal lines. Meaning Exfoliated primary teeth, specifically neonatal line width, may be associated with prenatal and perinatal life experiences.

In the current study, we considered a wide range of previously studied perinatal physiological factors as potential covariates. Because of the rich phenotype data collected by the ALSPAC, we were able to explore the associations between most factors identified from prior studies (eTable 1) and NNL width. Specifically, we tested 12 maternal exposures, four delivery related exposures, and 13 child exposures. We were unable to assess the following exposures due to data unavailability (i.e., the following previously examined factors from the literature were either not collected in ALSPAC or were not available to us at the time we completed the data analysis): use of antispasmodics during pregnancy, maternal age at pregnancy, number of children in family, past miscarriage, use of pregnancy sustaining medication, child blood transfusions, and child hypocalcemia. Further, the following perinatal factors were excluded because the cell size of being exposed was under 5: maternal diabetes, glycosuria, maternal distress during labor, vomiting during pregnancy, and child feeding problems. Additionally, we did not include maternal antidepressant use as a covariate, because none of the mothers in our sample reported use of psychopharmacological treatment. Considering that the number of preterm births in our analytic sample was low (n=3), we modeled gestational age as a continuous variable (unit: week). The assumption about a linear relationship between gestational age and NNL width was consistent with prior studies 21,22 . Overall, we made an effort to consider all available perinatal factors previously investigated to be related to NNL widths and thus possibly confounding the effects of maternal factors in our main analyses.
To study the NNL width as a biomarker for psychosocial exposures in psychiatry and pediatrics, we selected covariates by examining a comprehensive list of previously examined perinatal factors capturing maternal exposures, delivery related factors, and child exposures. Specifically, we estimated the unadjusted associations between each of the factors and NNL widths at the cuspal (eTable 4), middle (eTable 5), and EDJ portions (eTable 6). Out of many tested associations, only the following three factors showed a significant association with NNL widths at the middle portion at p<0.05: maternal iron supplements during pregnancy (yes/no, β = 2.28, p=0.01), maternal prepregnancy obesity (yes/no, β = -3.91, p=0.048), and gestational age (weeks, β = -0.56, p=0.02) (eTables 4-6).
Associations between the three selected covariates and the maternal psychosocial factors we considered are shown in eTable 8. Mothers who reported taking iron supplements during pregnancy were more likely to be exposed to some maternal psychosocial stressors, especially lifetime psychiatric problems (χ2 test p=0.04). However, caution is required when interpreting these results, due to very small cell counts.
Genotyping procedure and quality control Blood samples were collected from ALSPAC mothers during routine antenatal care and subsequent clinic visits. DNA was extracted via a phenol-chloroform, salting-out or guanidine hydrocholoride extraction methods 23 .

Polygenic risk score (PRS) generation and selection
To construct PRS for major depressive disorder in mothers, we used the most recent summary statistics from the Psychiatric Genomics Consortium GWAS of major depressive disorder (MDD), wave 2 (PGC-MDD2; 43,204 cases; 95,680 controls), and UK Biobank (127,552 cases; 233,763 controls) 26,27 . Additional genomic quality control steps preceding score construction included: removing imputed SNPs with low imputation quality (metric score <0.8), and any SNP with MAF <1%, call rate <95%, or HWE p<1e-6. Prior to calculating the PRS, we pruned the SNPs based on linkage disequilibrium patterns using p-value informed clump-based method in PLINK v1.90 (with r 2 threshold set to 0.25 within a 250kb window). To create the PRS, we summed the number of risk alleles (0, 1, or 2) for each SNP at a given p-value threshold, weighted by the log odds ratio reported for MDD in the GWAS by PGC 27 . In the ALSPAC cohort, we selected independent subsets of SNPs from GWAS summary data at a significance threshold of pT <0.001, which explains the most phenotypic variance in the discovery sample, meaning the GWAS performed by Wray et al as shown in another more recent GWAS of depression 28 . All PRS were standardized prior to analysis; thus, PRS values can be interpretated on the scale of each one-unit standard deviation difference in the genetic score of ALSPAC mothers.

Sensitivity analysis
We performed two sets of sensitivity analyses. Because maternal genetic liability for depression could be linked to both maternal psychopathology and potentially tooth-based markers, leading to a spurious association between maternal psychosocial stress and NNL widths, we performed an analysis controlling for maternal polygenic risk scores (PRS) for depression in the subsample of children with maternal genotype data available (n=54). PRS values were derived, as described above. Results from these analyses are shown in eTable 7 and eFigure 6.
To further examine the relative contributions of each maternal psychosocial factor on NNL widths, we additionally performed a mutually adjusted regression analysis. By mutually adjusted regression, we mean an analysis where multiple exposures were included in the same regression model. In the mutually adjusted model, we included the following three psychosocial factors simultaneously, in addition to the covariates: maternal severe lifetime depression history, maternal depression or anxiety at gestational 32 weeks, and high social support at 8 weeks postpartum. Of note, because any lifetime psychiatric problem and severe lifetime depression history were highly correlated, we only included maternal severe lifetime depression history. The remaining three variables were not highly collinear, as indicated by variance inflation factors (VIFs) under 2 in the mutually adjusted models (VIFdepression=1.19, VIFanxiety or depression=1.11, VIFsupport=1.05). Results from these analyses are shown in eTable 9. eResults Correlations between maternal psychosocial factors Psychosocial factors were weakly to strongly correlated, ranging from 0.02 to 0.81(eFigure 1). The maternal psychopathology measures clustered together, with the highest correlation occurring between the reports of severe lifetime depression history and any psychiatric problems (rtetrachoric = 0.82). In other words, measures of "trait" (or history of) psychopathology before pregnancy were correlated with the "state" (or symptoms of) psychopathology during pregnancy.
The lack of correlation between types of psychosocial factors indicated that they likely captured separate domains of experiences during pregnancy, warranting their inclusion in the analyses. Social support, as a protective factor, was generally negatively associated with psychosocial stress, as expected.

Power analysis
Before performing any empirical analysis, we assessed the statistical power a priori, by considering a range of R 2 values and varying sample sizes. With the sample size of the current study (n=70), our study would be sufficiently powered (i.e., with power above 80%) to detect an effect of R 2 = 0.1 or larger in a simple linear regression model (eFigure 5). To detect smaller effects, such as R 2 < 0.05, a sample size of 200 or higher would be required.

Sensitivity analysis
Results from the mutually adjusted model suggest that severe lifetime depression history and high social support postpartum were still associated with NNL width at the EDJ portion (eTable 9). Specifically, the positive effect estimate associated with severe lifetime depression history was twice as large as the negative effect estimate corresponding to having higher social support after birth (βdepression=3. 18 Note. Bolded values indicate a nominally significant association between a maternal psychosocial factor and NNL widths, after accounting for covariates as well as two other psychosocial factors tested.