Negative emotionality (NE) is derived from the temperamental dimensions of sadness, distress toward limitations, fear, and excessive reactions to minor changes, and reflects a generally stable tendency to show increased emotional reactivity toward negative situations (Gartstein & Rothbart, Reference Gartstein and Rothbart2003; Lemery, Goldsmith, Klinnert, & Mrazek, Reference Lemery, Goldsmith, Klinnert and Mrazek1999). NE is associated with the development of later problematic behavior and psychopathology (Eisenberg et al., Reference Eisenberg, Valiente, Spinrad, Cumberland, Liew and Reiser2009; Fox, Henderson, Rubin, Calkins, & Schmidt, Reference Fox, Henderson, Rubin, Calkins and Schmidt2001; Hyde, Mezulis, & Abramson, Reference Hyde, Mezulis and Abramson2008). For example, fearful temperament is associated with childhood anxiety disorders (Degnan, Alma, & Fox, Reference Degnan, Almas and Fox2010; Goldsmith & Lemery, Reference Goldsmith and Lemery2000), while NE is associated with depression (Phillips, Lonigan, Driscoll, & Hooe, Reference Phillips, Lonigan, Driscoll and Hooe2002) and maladjustment (Eisenberg et al., Reference Eisenberg, Valiente, Spinrad, Cumberland, Liew and Reiser2009). Understanding early influences of NE on socioemotional development (Davis et al., Reference Davis, Snidman, Wadhwa, Glynn, Dunkel Schetter and Sandman2004, Reference Davis, Glynn, Schetter, Hobel, Chicz-Demet and Sandman2007; Hayden et al., Reference Hayden, Dougherty, Maloney, Durbin, Olino and Nurnberger2007, Reference Hayden, Klein, Sheikh, Olino, Dougherty and Dyson2010) could inform efforts at prevention and early intervention. Recent contradictory findings about the role of genetic and prenatal adversity (Braithwaite et al., Reference Braithwaite, Ramchandani, O'Connor, van IJzendoorn, Bakermans-Kranenburg and Glover2013; Pluess et al., Reference Pluess, Velders, Belsky, van IJzendoorn, Bakermns-Kranenburg and Jaddoe2011) suggest the need for replication (Duncan, Reference Duncan2013) and modeling of genetic risk with multiple genes (Plomin, Reference Plomin2013). Accordingly, we present the findings from a study of the development of NE from 3 to 36 months of age from the interaction of prenatal maternal depression and a multilocus genetic profile.
The Role of Prenatal Maternal Stress
Prenatal maternal stress, measured in diverse ways, is associated with NE (Glover, Reference Glover2011; O'Connor, Heron, & Glover, Reference O'Connor, Heron and Glover2002). For example, higher prenatal maternal cortisol is associated with fussier behavior, more negative facial expressions, crying, as well as higher NE at 7 weeks of age (de Weerth, Hees, & Buitelaar, Reference de Weerth, Hees and Buitelaar2003). Symptoms of prenatal maternal anxiety and depression, loosely associated with the stress response, predict behavioral reactivity at 4 months of age (Davis et al., Reference Davis, Snidman, Wadhwa, Glynn, Dunkel Schetter and Sandman2004) and behavioral/emotional problems at 4 years of age (O'Connor, Heron, & Glover, Reference O'Connor, Heron and Glover2002). While most studies of prenatal maternal symptoms have examined prenatal anxiety (Glover, Reference Glover2011; Pluess et al., Reference Pluess, Velders, Belsky, van IJzendoorn, Bakermns-Kranenburg and Jaddoe2011), there is evidence for the specific effect of prenatal depression with outcomes reported as early as 2 months of age (Davis et al., Reference Davis, Glynn, Schetter, Hobel, Chicz-Demet and Sandman2007; Field, Reference Field2011; McGrath, Records, & Rice, Reference McGrath, Records and Rice2008). Specifically, Davis et al. (Reference Davis, Glynn, Schetter, Hobel, Chicz-Demet and Sandman2007) reported higher negative reactivity at 2 months and McGrath et al. (Reference McGrath, Records and Rice2008) reported more difficult temperament at 2 and 6 months. The association between prenatal maternal stress and NE is inconsistent though (Susman, Schmeelk, Ponirakis, & Griepy, Reference Susman, Schmeelk, Ponirakis and Gariepy2001), suggesting other factors may serve as moderators.
The Role of Genotype
Twin and genetic linkage studies supporting genomic influence for temperament (Bouchard, Reference Bouchard1994; Saudino, Reference Saudino2009) have recently been followed by studies of candidate genes. Genes in the serotonin cell signalling pathways have been of particular interest in the prediction of NE, given their role in regulating emotional responses (Sen, Burmeister, & Ghosh, Reference Sen, Burmeister and Ghosh2004) and their activity during the third trimester of pregnancy (Geva & Feldman, Reference Geva and Feldman2008). There has been considerable emphasis on a functional variation in the promoter region of the serotonin transporter gene because of its association with anxiety, depression, and affective regulation (Canli & Lesch, Reference Canli and Lesch2007; Hariri & Holmes, Reference Hariri and Holmes2006). The solute carrier family C6, member 4 gene (SLC6A4), which encodes for the serotonin transporter (5-HTT), contains a 43 base pair variable number tandem repeat polymorphism in the promoter region (5-HTT linked polymorphic region [5-HTTLPR]) that is coupled to transcriptional efficiency. The long compared to the short variant shows an increased basal transcription of 5-HTT messenger RNA (Canli & Lesch, Reference Canli and Lesch2007). Within the long genotype (Uher, Reference Uher2008), there is a functional polymorphism (A → G, rs25531; Hu et al., Reference Hu, Lipsky, Zhu, Akhtar, Taubman and Greenberg2006). The long-rs25531(A) (LA) variant has greater transcriptional efficiency and greater 5-HTT binding potential in humans (Hu et al., Reference Hu, Lipsky, Zhu, Akhtar, Taubman and Greenberg2006; Praschak-Rieder et al., Reference Praschak-Rieder, Kennedy, Wilson, Hussey, Boovariwala and Willeit2007), whereas the long-rs25531(G) (LG) variant has a functionally similar effect on 5-HTT messenger RNA expression as the short–short (SS) genotype. The frequency of the LG in Caucasians is not insignificant at 14% (Odgerell, Talatil, Hamilton, Levinson, & Weissman, Reference Odgerel, Talatil, Hamilton, Levinson and Weissman2013).
Carriers of the low expressing alleles (any short or LG [S/LG]) show enhanced processing of negative emotions (Pezawas et al., Reference Pezawas, Meyer-Lindenberg, Drabant, Verchinski, Munoz and Kolachana2005), positive stimuli, and general emotional processing (Canli et al., Reference Canli, Omura, Haas, Fallgatter, Constable and Lesch2005) that associates with structural differences in limbic brain regions (Hariri et al., Reference Hariri, Mattay, Tessitore, Kolachana, Fera and Goldman2002). Likewise carriers of the low expressing alleles (S/LG) have more depressive and anxious symptoms (Caspi et al., Reference Caspi, Sugden, Moffit, Taylor, Craig and Harrington2003), and more depressive and anxious symptoms relative to carriers of the LALA allele (Canli & Lesch, Reference Canli and Lesch2007; Gonda et al., Reference Gonda, Fountoulakis, Juhasz, Rihmer, Lazary and Laszik2009). There is also some evidence indicating that carriers of the low expressing short allele rate higher in NE than LL carriers (Auerbach et al., Reference Auerbach, Geller, Lezer, Shinwell, Belmaker and Levine1999; Hayden et al., Reference Hayden, Dougherty, Maloney, Durbin, Olino and Nurnberger2007, Reference Hayden, Klein, Sheikh, Olino, Dougherty and Dyson2010). For example, infants with the SS genotype are reported to rate higher in NE at 2 months (Auerbach et al., Reference Auerbach, Geller, Lezer, Shinwell, Belmaker and Levine1999), higher in fearful temperament during childhood (Hayden et al., Reference Hayden, Dougherty, Maloney, Durbin, Olino and Nurnberger2007), and higher in NE in the presence of low positive emotionality during childhood (Hayden et al., Reference Hayden, Klein, Sheikh, Olino, Dougherty and Dyson2010).
Functional variants in the dopamine receptor D4 gene (DRD4) are also associated with NE (Auerbach et al., Reference Auerbach, Geller, Lezer, Shinwell, Belmaker and Levine1999; Auerbach, Benjamin, Faroy, Geller, & Ebstein, Reference Auerbach, Benjamin, Faroy, Geller and Ebstein2001; Auerbach, Faroy, Ebstein, Kahana, & Levine, Reference Auerbach, Faroy, Ebstein, Kahana and Levine2001). There is a 48 base pair variable number tandem repeats in exon 3 of the DRD4 gene ranging from 2 to 11 copies. The long (6- to 10-repeat) alleles and specifically the 7-repeat (7R) allele are associated with lower dopamine receptor signalling and are identified as the susceptibility alleles. For example, the 7R allele is associated with approach behaviors (e.g., Zohsel et al., Reference Zohsel, Buchmann, Blomeyer, Hohm, Schmidt and Esser2014); and in a meta-analysis 7R was associated with more externalizing behavior in negative contexts, but was associated with less externalizing behavior in positive contexts (Bakermans-Kranenburg & van IJzendoorn, Reference Bakermans-Kranenburg and van Ijzendoorn2011). Reports of the main effect of 7R on NE are contradictory, with evidence that the 7R allele is associated with lower scores (Auerbach et al., Reference Auerbach, Geller, Lezer, Shinwell, Belmaker and Levine1999; Auerbach, Faroy, et al., Reference Auerbach, Faroy, Ebstein, Kahana and Levine2001; De Luca et al., Reference De Luca, Rizzardi, Buccino, Alessandroni, Salvioli and Filograsso2003) as well as higher scores (Auerbach, Benjamin, et al., Reference Auerbach, Benjamin, Faroy, Geller and Ebstein2001; Holmboe, Nemoda, Fearon, Sasvari-Szekely, & Johnson, Reference Holmboe, Nemoda, Fearon, Sasvari-Szekely and Johnson2011; Lakatos et al., Reference Lakatos, Nemoda, Birkas, Ronai, Kovacs and Ney2003) of NE and associated features. Auerbach et al. (Reference Auerbach, Geller, Lezer, Shinwell, Belmaker and Levine1999) report lower ratings of NE, distress toward limitations, and distress to novel stimuli on the Infant Behavior Questionnaire at 2 months in carriers of the 6 to 8 repeats (6-8R) allele than in carriers of the 2 to 5 repeats (2-5R) allele. At 12 months, infants with the 6-8R allele show less active resistance, and struggle less than infants with the 2-5R (Auerbach, Faroy, et al., Reference Auerbach, Faroy, Ebstein, Kahana and Levine2001). However, the reverse has also been reported. Infants with the 7R allele are higher on NE as measured by the Infant Behavior Questionnaire at 4 and 9 months of age (Holmboe et al., Reference Holmboe, Nemoda, Fearon, Sasvari-Szekely and Johnson2011), show less novelty preference at 12 months (Auerbach, Benjamin, et al., Reference Auerbach, Benjamin, Faroy, Geller and Ebstein2001), and show increased latency to accept a toy from strangers at 12 months (Lakatos et al., Reference Lakatos, Nemoda, Birkas, Ronai, Kovacs and Ney2003).
Consistent with evidence that multiple genes act in an additive fashion for the expression of a particular phenotype (Masarik et al., Reference Masarik, Conger, Donnellan, Stallings, Martin and Schofield2014; Plomin, Reference Plomin2013), there is evidence for the joint effect of 5-HTTLPR and DRD4 in the prediction of emotionally reactive behaviors (Auerbach et al., Reference Auerbach, Geller, Lezer, Shinwell, Belmaker and Levine1999; Ebstein et al., Reference Ebstein, Levine, Geller, Auerbach, Gritsenko and Belmaker1998; Holmboe et al., Reference Holmboe, Nemoda, Fearon, Sasvari-Szekely and Johnson2011; Lakatoes et al., Reference Lakatos, Nemoda, Birkas, Ronai, Kovacs and Ney2003). These findings are consistent with the overlapping role of the serotonin and dopamine systems in regulating behaviors related to NE such as approach and escape. DRD4 is highly expressed in the amygdala and the prefrontal circuits (Oak, Oldenhof, & Van Tol, Reference Oak, Oldenhof and Van Tol2000), whereas 5-HTTLPR significantly influences the amygdala and amygdala–prefrontal coupling (Hariri et al., Reference Hariri, Mattay, Tessitore, Kolachana, Fera and Goldman2002). Reports of the interaction effect of 5-HTTLPR × DRD4 also point to contradictory findings about which alleles increase the likelihood for NE. Ebstein et al. (Reference Ebstein, Levine, Geller, Auerbach, Gritsenko and Belmaker1998) report that the 5-HTTLPR SS genotype and the DRD4 2-5R allele are associated with lower orientation scores and reduced interactive behavior in 2-week-old neonates. Similarly, Auerbach et al. (Reference Auerbach, Geller, Lezer, Shinwell, Belmaker and Levine1999) find that SS and 2-5R are associated with higher NE and distress to limitations in 2-month-olds. Lakatos et al. (Reference Lakatos, Nemoda, Birkas, Ronai, Kovacs and Ney2003), however, find that SS and 7R are associated with increased stranger anxiety duration and latency to smile at 12 months, while Holmboe et al. (Reference Holmboe, Nemoda, Fearon, Sasvari-Szekely and Johnson2011) report reversed findings such that LALA and 7R are associated with higher NE at 4 and 9 months. A number of possible explanations for these divergences are suggested (Holmboe et al., Reference Holmboe, Nemoda, Fearon, Sasvari-Szekely and Johnson2011), including unmeasured and inconsistently present moderator such as adversity (e.g., Pluess et al., Reference Pluess, Velders, Belsky, van IJzendoorn, Bakermns-Kranenburg and Jaddoe2011; Smith et al., Reference Smith, Sheikh, Dyson, Olino, Laptook and Durbin2012) as well as heterogeneous developmental age. The overall pattern of findings suggests that the DRD4 2-5R allele is associated with higher NE in the first months of infancy (Auerbach et al., Reference Auerbach, Geller, Lezer, Shinwell, Belmaker and Levine1999; Ebstein et al., Reference Ebstein, Levine, Geller, Auerbach, Gritsenko and Belmaker1998) and the 7R allele is associated with higher NE later in the first year of life (Holmboe et al., Reference Holmboe, Nemoda, Fearon, Sasvari-Szekely and Johnson2011; Lakatos et al., Reference Lakatos, Nemoda, Birkas, Ronai, Kovacs and Ney2003). With the exception of one study (Holmboe et al., Reference Holmboe, Nemoda, Fearon, Sasvari-Szekely and Johnson2011), the 5-HTTLPR short and LG alleles seem consistently associated with higher NE.
Gene × Environment (G × E) Interactions
The interaction of the environment with genomic variants is consistent with findings in molecular biology that the activation of gene expression is contingent upon transcriptional signals that derive from the internal and the external environment (Meaney, Reference Meaney2009). G × E models used to investigate NE examine the continuum of environmental exposures from maternal preconception to postnatal periods. For example, infant carriers of the short allele of the 5-HTTLPR whose mothers experienced adversity during their childhood are reported to have higher NE at 18 and 36 months of age (Bouvette-Turcot et al., Reference Bouvette-Turcot, Fleming, Wazana, Sokolowski, Gaudreau and Gonzalez2016). Similar results are found in studies investigating G × Postnatal E models involving 5-HTTLPR. Carriers of the short allele with insecure attachment are reported to show elevated NE (Pauli-Pott, Friedl, Hinney, & Hebebrand, Reference Pauli-Pott, Friedl, Hinney and Hebebrand2009) and decreased emotion regulation at age 2 to 4 (Kochanska, Philibert, & Barry, Reference Kochanska, Philibert and Berry2009), while short carriers with low levels of social support show increased behavior inhibition at 7 years of age (Fox et al., Reference Fox, Nichols, Henderson, Rubin, Schmidt and Hamer2005). Similarly, Hayden et al. (Reference Hayden, Klein, Sheikh, Olino, Dougherty and Dyson2010) found that variants in the brain-derived neurotrophic factor gene and parental depression and marital discord interact to predict NE at 3 years of age.
Few studies though have examined the role of prenatal exposure, and to date, its effect on the development of NE is unclear. Pluess et al. (Reference Pluess, Velders, Belsky, van IJzendoorn, Bakermns-Kranenburg and Jaddoe2011) found that carriers of the 5-HTTLPR short allele who were exposed to higher levels of maternal prenatal anxiety have increased NE at 6 months. However, Braithwaithe et al. (Reference Braithwaite, Ramchandani, O'Connor, van IJzendoorn, Bakermans-Kranenburg and Glover2013) did not replicate this finding. Considering the emerging importance of models that include multilocus genetic profiles (Plomin, Reference Plomin2013), the inconsistent G × E findings may be in part explained by the use of models that include only a single genomic variant. A composite genetic factor, akin to the well-established factor for cumulative exposure to trauma (Sameroff, Gutman, & Peck, Reference Sameroff, Gutman, Peck and Luthar2003), has been used to test a (cumulative) G × E model (Belsky & Beaver, Reference Belsky and Beaver2011; Sonuga-Barke et al., Reference Sonuga-Barke, Oades, Psychogiou, Chen, Franke and Buitelaar2009). Belsky and Beaver (Reference Belsky and Beaver2011) found that the association between supportive parenting and regulation is stronger as the number of plasticity (or susceptibility) genes increases. Plasticity genes have been defined as variants in genes implicated in cellular responses to environmental signals (e.g., synaptic plasticity) and associated with increased biological sensitivity to environmental conditions (Boyce & Ellis, Reference Boyce and Ellis2005). There is now considerable evidence for the idea that variants of the 5-HTTLPR and DRD4 genotype serve as such plasticity/susceptibility genes (Belsky & Beaver, Reference Belsky and Beaver2011). A G × E model that includes a multilocus genetic profile with DRD4 and 5- HTTLPR would reflect evidence of joint influences of these two genotypes on NE and might determine if such an influence is stronger than that of only one gene. Further, because there has been variability in findings in the susceptibility alleles for both DRD4 and 5-HTTLPR at different times during development (Auerbach et al., Reference Auerbach, Geller, Lezer, Shinwell, Belmaker and Levine1999; Holmboe et al., Reference Holmboe, Nemoda, Fearon, Sasvari-Szekely and Johnson2011; Lakatos et al., Reference Lakatos, Nemoda, Birkas, Ronai, Kovacs and Ney2003), investigating DRD4 and 5-HTTLPR in a model with a multilocus genetic profile at multiple times during development would reflect whether the same alleles are susceptible across an age span.
Characterizing the Model of G × E
The diathesis–stress and the differential susceptibility models potentially characterize how genes associated with functional outcomes could, under conditions of prenatal depression, produce variation in NE. In the diathesis–stress model, carriers of the genotype that associate with an increased risk for disease (e.g., the S/LG for 5-HTTLPR), when exposed to prenatal depression, would have a greater likelihood of developing higher NE. Noncarriers would be insensitive to any environment with respect to NE, while in the absence of adversity, individuals with or without the “risk” variant would show comparable developmental outcome.
Unlike the diathesis–stress model, the differential susceptibility model allows for the possibility of positive outcomes as a function of the quality of the relevant environmental condition (Belsky & Pluess, Reference Belsky and Pluess2009; Ellis, Boyce, Belsky, Bakermans-Kranenburg, & Van IJzendoorn, Reference Ellis, Boyce, Belsky, Bakermans-Kranenburg and van IJzendoorn2011). The differential susceptibility model reframes risk as susceptibility in light of reanalyses of studies demonstrating that the same genotypes that confer a greater vulnerability under adverse conditions promote the development of phenotypes associated with resistance to mental disorders under more favorable conditions (Belskey & Pluess, Reference Belsky and Pluess2009; Pluess, Belsky, & Neuman, Reference Pluess, Belsky and Neuman2009). The differential susceptibility model suggests that “risk” genotypes are better considered “plasticity” or “susceptibility” genotypes, and that carriers are more susceptible to both adverse and enriched environments. To date, the prediction of NE from prenatal exposure and genetic risk is only explained by a diathesis–stress model (Pluess et al., Reference Pluess, Velders, Belsky, van IJzendoorn, Bakermns-Kranenburg and Jaddoe2011).
Purpose of the Study
In this study we aimed to investigate whether prenatal maternal depression, child 5-HTTLPR, and DRD4 genotype exert a joint influence on the development of early age NE and whether this association is stable over 3, 6, 18, and 36 months of life. Specifically, we examined: (a) the two-way interaction of prenatal depression and the 5-HTTLPR genotype predicting NE; (b) the two-way interaction of prenatal depression and DRD4 predicting NE; (c) the two-way interaction of a multilocus genetic profile consisting of DRD4 and 5-HTTLPR and prenatal depression in predicting NE, and whether the joint influence of 5-HTTLPR and DRD4 is more predictive than the individual effects of 5-HTTLPR or DRD4 alone in a G × E model; and (d) if the G × E model with the multilocus genetic profile is best explained by the diathesis–stress or the differential susceptibility model. We used confirmatory analysis of interaction models (Widaman et al., Reference Widaman, Helm, Castro-Schilo, Pluess, Stallings and Belsky2012), a novel statistical method favored over exploratory and conservative models, such as regression and simple slopes methods. Confirmatory models directly test four competing predictions (i.e., weak and strong versions of the diathesis–stress and differential susceptibility models) to identify which G × E model best explains the significant interaction findings (Appendix A, Figure A.1; Belsky, Pluess, & Widaman, Reference Belsky, Pluess and Widaman2013).
Method
Participants
The participants were a community-based sample of mother–infant dyads from Montreal, Quebec, and Hamilton, Ontario, who are part of the Maternal Adversity, Vulnerability and Neurodevelopment (MAVAN) Project. The MAVAN is an established community cohort that enrolled 578 mother–infant dyads between 2003 and 2009. Mothers were recruited from the general population at 13- to 20-weeks gestation during their routine ultrasound and were included in the study if they were at least 18 years old and fluent in either French or English. Participants were excluded if they experienced serious obstetric complications during pregnancy or during the delivery of their child; if their child had any congenital diseases ascertained using the Bayley Scales of Infant Development, Second Edition (Bayley, Reference Bayley1993); or if they delivered prematurely (before 37 weeks gestation). Dyads involved in the MAVAN have been assessed up to 72 months. Subjects recruited in Hamilton were oversampled for maternal prenatal distress.
Retention rates for the MAVAN subjects were 97.4% at 6 months, 84.04% at 18 months, and 80.5% at 36 months, reducing the total sample size to 464 dyads at 36 months. Compared to mothers who remained in the MAVAN study at all time points, mothers who left the study before reaching the 72-month time point did not differ significantly on measures of age at birth, depression, or education. Children whose mothers left the MAVAN study before reaching the 72-month time point did not differ significantly on outcomes of NE between 3 and 36 months; however, they had significant lower birth weights than children of mothers who stayed in the study at all time points. Greater details are available elsewhere (O'Donnell et al., Reference O'Donnell, Gaudreau, Colalillo, Steiner, Atkinson and Moss2014). The present study included a subsample of 179 mother–child dyads, with all measures complete including genotype (refer to Table 1 for sample characteristics). The reduction in sample size from 464 to 179 participants is explained as follows: 60 children were lost due to missing prenatal data (involved in the study prior to implementation of all study measures); 199 were missing genomic data (due to partial funding for genotyping); 21 were missing data on the Infant Behaviour Questionnaire or from the maternal depression rating scale; and 5 were outliers. A comparison of the study sample and the cohort sample indicated that this study sample had higher income and lower NE.
Table 1. Demographic characteristics of subjects from MAVAN included in 36-month analyses
Note: MAVAN, Maternal Adversity, Vulnerability, and Neurodevelopment project; CES-D, Center for Epidemiologic Studies Depressions Scale; S, short allele; L, long allele; 5-HTTLPR, serotonin transporter linked polymorphic region gene; DRD4, dopamine receptor D4 gene; 2-5R, two to five repeats; 6-8R, six to eight repeats.
a Significant difference between Montreal and Hamilton.
Procedure
The mothers were interviewed between 24 and 36 weeks of pregnancy, and the dyads were assessed at 3, 6, 12, and 18 months and yearly from 24 months onward. Maternal health and well-being were assessed each year using validated measures of maternal mental health, social and family functioning, and socioeconomic status (Kramer et al., Reference Kramer, Wilkins, Gouler, Seguin, Lydon and Kahn2009). The children were assessed with age-appropriate measures of temperament, socioemotional development, and psychopathology. Informed consent was obtained at the time of recruitment and at each time point of data acquisition. Ethics review board approval was obtained from the institution of each study site.
Measures
NE
NE at 3 and 6 months were obtained from the Infant Behaviour Questionnaire—Revised (IBQ-R; Gartstein & Rothbart, Reference Gartstein and Rothbart2003), a reliable and valid parent-completed measure of 15 scales of temperament (Parade & Leekes, Reference Parade and Leerkes2008). As per the recommendations of the authors, NE was extracted from our sample, given the younger age of our sample than that in the published sample (M. Gartstein, personal communication, May 13, 2012). Using promax oblique rotation, subscale loadings were determined using a 1% level of significance (subscales with loadings > 0.29; Stevens, Reference Stevens1986). At 3 and 6 months, 7 of the 15 subscales (activity level, distress to limitations, falling reactivity [negative loading], fear, sadness, cuddliness [negative loading], and soothability [negative loading]) substantially loaded on the NE factor; see Table A.1 for factor loadings). All subscales with significant loadings were transformed into z scores, aggregated and averaged to create a final NE factor as per the authors of the IBQ-R (Garstein & Rothbart, Reference Gartstein and Rothbart2003). There was good internal consistency (with Cronbach αs 0.74 at 3 months and 0.71 at 6 months).
NE at 18 and 36 months was obtained from the Early Child Behavior Questionnaire (ECBQ; Putnam, Gartstein, & Rothbart, Reference Putnam, Gartstein and Rothbart2006), a reliable and valid parent-completed measure of temperament. The ECBQ comprises 18 different scales measuring different temperamental dimensions and is considered an upward extension of the IBQ-R. It contains 11 scales that are similar in form to the IBQ-R (Garstein & Rothbart, Reference Gartstein and Rothbart2003; Putnam et al., Reference Putnam, Gartstein and Rothbart2006), and both instruments have yielded a similar three-factor temperament structure that includes NE with many consistent factor loadings across both measures (Putnam, Ellis, & Rothbart, Reference Putnam, Ellis and Rothbart2001). Convergent validity and structural continuity between the IBQ-R and the ECBQ has been demonstrated (Putnam, Rothbart, & Gartstein, Reference Putnam, Rothbart and Gartstein2008). NE was extracted as per the method for the IBQ-R and consisted of the same subscales (fear, frustration, motor activation, perceptual sensitivity, sadness, discomfort, shyness, and soothability [negative loading]) as per the original report (Putnam et al., Reference Putnam, Rothbart and Gartstein2008). Impulsivity (negative loading) only loaded at 18 months. Given the almost perfect correlation with and without impulsivity (r = .98447), it was omitted for the purposes of consistency (18 and 36 months) and in accord with the NE factor derived by Putnam et al. (Reference Putnam, Gartstein and Rothbart2006; see Table A.2 for factor loadings). Internal consistency was 0.76 and 0.75 at 18 and 36 months, respectively.
5-HTTLPR and DRD4 genotype
Child genotype was obtained with the use of buccal swabs, using the TaqMan methods on the ABI-7000 for single nucleotide polymorphism markers and ABI-3100 for repeat polymorphisms. To ensure a clear result, any ambiguous genotypes were discarded and the subjects were regenotyped until the results were unambiguous. Each 20th marker was regenotyped to check for error rates (0.5%).
We examined two categorizations of 5-HTTLPR. For the biallelic categorization, 5-HTTLPR was coded as (a) L/L, the highest expressing genotype or (b) S/S or S/L, the lowest expressing genotype. For the triallelic categorization it was coded as (a) LALA, the highest expressing genotype, (b) any LA (LA/S or LALG), or (c) no LA (S/S, S/LG, or LGLG; Hu et al., Reference Hu, Lipsky, Zhu, Akhtar, Taubman and Greenberg2006). The analyses with the biallelic categorization did not yield any significant findings. As such, only analyses using the triallelic categorization are reported here. Analyses with the triallelic categorization revealed comparable predictions in subjects with the presence of any LA allele, (i.e., LA/LA, LA/S, or LALG). As such, the 5-HTTLPR genotype variable was recoded as a dichotomous variable: any LA versus no LA. This facilitated the construction of a multilocus genetic profile score.
We also examined two categorization of DRD4. Specifically, DRD4 was coded as 6-8R or 2-5R, as per Auerbach et al. (Reference Auerbach, Geller, Lezer, Shinwell, Belmaker and Levine1999). DRD4 was also coded as 7R versus other genotypes, as per Holmboe et al. (Reference Holmboe, Nemoda, Fearon, Sasvari-Szekely and Johnson2011). Both categorizations yielded the same results. Findings are presented using the Auerbach et al. (Reference Auerbach, Geller, Lezer, Shinwell, Belmaker and Levine1999) classification (6-8R vs. 2-5R). For both the Montreal and Hamilton samples, the distribution of DRD4 conformed to the Hardy–Weinberg equilibrium (p = .58 and .95, respectively). Similarly, the genotype distribution for 5-HTTLPR conformed to the Hardy–Weinberg equilibrium for the Montreal (p = .60) and Hamilton (p = .06) samples. There were no gender differences for 5-HTTLPR or DRD4, χ2 (1) = 0.01, p = .92; χ2 (1) = 0.4, p = .52, respectively (Table 1).
Other plasticity genes that were examined in secondary analyses included dopamine receptor D2 (DRD2), dopamine transporter (DAT), catechol-O-methyltransferase (COMT), monoamine oxidase A (MAOA), and the variable number tandem repeats in the second intron of the SLC6A4 gene polymorphism (STin2). The following categorizations used for these genes were DRD2 coded as A/G or A/A versus G/G, DAT coded as 10/10 versus 9/9 or 9/10, COMT coded as G/G versus A/G or A/A, MAOA coded as any 4 (3/4, 4/4, or 4/5) versus any other combination (e.g., 3/3 or 5/5), and STin2 coded as any 10 (10/12 or 10/10) versus 12/12.
Multilocus genetic profile score
A genetic factor was obtained by summing the number of risk/susceptibility genotypes: no LA (S/S, S/LG, or LG/LG [S/LG]) for 5-HTTLPR and any 6-8R for DRD4 (Auerbach et al., Reference Auerbach, Geller, Lezer, Shinwell, Belmaker and Levine1999). The child's value ranged from 0 (no risk/susceptibility genotype) to 2 (both risk/susceptibility genotypes).
Prenatal depression
Maternal depressive symptoms were obtained with the Center for Epidemiologic Studies Depression Scale (CES-D; Radloff, Reference Radloff1977) at 24 to 36 weeks of pregnancy. Items on the CES-D are designed to measure symptoms of depression in community-based populations, and include 20 questions about mood, appetite, and sleep, rated on a Likert scale ranging from 0 to 3. Scores on the CES-D can range from 0 to 60 with a score of 16 or higher indicating the presence of a depressive illness. The CES-D has been validated in a sample of pregnant women (e.g., Field et al., Reference Field, Diego, Dieter, Hernandez-Reif, Schanberg and Kuhn2004). In the present study, scores were centered to facilitate interpretation of regression coefficients, with higher scores indicating more severe depressive symptoms. Internal consistency was 0.92.
Covariates
Child birth weight was obtained from the chart of the birthing unit. Other covariates were obtained from the Health and Well Being of Mothers and Their Newborns Questionnaire (Kramer et al., Reference Kramer, Wilkins, Gouler, Seguin, Lydon and Kahn2009) administered prenatally and at 6, 12, and 36 months postnatally. Postnatal depression was assessed with the CES-D at 6, 12, and 36 months. Maternal education, assessed prenatally, was dichotomized as university graduate or higher or others. The original categories (Table 1) were collapsed into two groups in light of small-sized categories.
Covariates were identified by preliminary analyses driven by theoretical conception. Variables were retained as covariates for the final analyses when they were associated with a predictor or the outcome. This included maternal postnatal depression, mother's age at birth, site, child socioeconomic status (represented by maternal education), child gender, and maternal 5-HTTLPR genotype in the models that also contained child 5-HTTLPR genotype. Variables considered as covariates but not retained for the final analyses were maternal DRD4 genotype, maternal alcohol consumption during pregnancy, and child birth weight. Postnatal depression was assessed with the CES-D at 6, 12, and 36 months and included in all analyses, except at 3 months when a measure of concurrent maternal depressive symptoms is unavailable.
Statistical analysis
Intraclass correlation, which depicts the proportion of variance in NE accounted for by site of recruiting, was 0.07 at 3 months and 0.00 at 6 months, 18, and 36 months (ps > .05). Given the negligible proportion of the total variance of NE explained by site level, site was not added as a fixed effect. However, with significant differences between both sites in prenatal CES-D score, t (197) = –2.22, p = .03, and on NE at 3 months, t (127) = –2.55, p = 0.01, site was entered as a covariate.
Outliers were assessed by examining values of studentized residuals with magnitudes greater than 2.8 (probability of .005) or with values greater than 2.0 (probability of .05) with a combined leverage larger then 2p/n (Hoaglin, & Welsch, Reference Hoaglin and Welsch1978). Five cases were removed. We corrected for heteroscedasticity with the use of consistent standard errors and p values applied to the graphical representations of each significant mixed model. All main predictor variables except for genotype were centered.
The data was analysed in two separate models to test the robustness of our findings: (a) a repeated-measure model (mixed model) predicting NE across all four time points simultaneously, and a (b) confirmatory analysis of interaction model to identify which type of G × E model best explains the interaction finding at each of the four time points.
Secondary analyses included exploration of other known plasticity genes, although lesser related to NE. The results revealed no significant main or cumulative effect. As such, findings are presented only for 5-HTTLPR and DRD4.
Mixed model for longitudinal data
We implemented mixed models for repeated measures with an unstructured covariance matrix to test the prediction of NE from prenatal depression and genotype. Child gender, site, maternal age, maternal 5-HTTLPR genotype, and education were entered as covariates. In a more conservative model, a measure of postnatal maternal depression (average score of maternal depression scores from 6 to 36 months) was entered. This is a more conservative model because some of the exposure to postnatal depression postdates the outcome under consideration, NE from 3 to 36 months.
Missing values were imputed for the mixed-model analysis with the multilocus genetic profile score using the multivariate imputation by chained equations algorithm (van Buuren & Groothuis-Oudshoorn, Reference van Buuren and Groothuis-Oudshoorn2011). The imputation model was used to explain the pattern of missing data and to obtain imputed values for these missing data.
As recommended by van Buuren, Boshuizen, and Knook (Reference van Buuren, Boshuizen and Knook1999), we included all variables used in the mixed-model analysis in addition to variables that explained a considerable amount of variance. No variable had to be removed because of excessive missing data. Imputed values were based on regression estimates. As recommended, five iterations of the algorithm were run. When the amount of missing values is high, it is recommended to run between 20 and 100 imputed data sets, so as a precaution, we created 50 imputed data sets. R software (version 3.1.0; The R Foundation for Statistical Computing, Vienna) was used to perform imputation.
Confirmatory regression
To specifically test the whether carriers of the 5-HTTLPR S/LG and DRD4 6-8R alleles were at risk (diathesis–stress) or susceptible (differential susceptibility) when exposed to prenatal depression, the regression was reparameterized using the following equation (Widaman et al., Reference Widaman, Helm, Castro-Schilo, Pluess, Stallings and Belsky2012):
$$Y = {\rm \beta} 0{\rm} + {\rm \beta} 1{\rm} \left( {{\rm CES-D} - C} \right){\rm} + {\rm \varepsilon}\comma $$
for carriers of nonrisk/susceptibility allele (any LA, 2-5R), and
$$Y = {\rm \beta} 0{\rm} + {\rm \beta} 2{\rm} \left( {{\rm CES-D} - C} \right){\rm} + {\rm \varepsilon}\comma $$
for carriers of risk/susceptibility allele (no LA, any 6-8R). The parameters in this equation were the intercept (β0), the slope for carriers of the nonrisk/susceptibility allele (β1), the slope for carriers of the risk/susceptibility allele (β2), and the crossover point between the two slopes (C). The diathesis–stress and differential susceptibility models both assume that carriers with no susceptibility alleles would not be influenced by the environment (prenatal depression); that is, β1 = 0. However, because the possibility remains that the environment exerts a slight effect even on noncarriers, the diathesis–stress and differential susceptibility models are further separated into two groups: a weak model (β1 ≠ 0 and β1 < β2) and a strong model (β1 = 0).
The Akaike information criteria with significance testing at a 95% confidence interval was used to determine which of the four models (i.e., weak vs. strong diathesis–stress, and weak vs. strong differential susceptibility) best fit the data at each time point. Only the strong diathesis–stress and strong differential susceptibility model testing are reported; however, all four models were tested. The models presented are calculated for average maternal age, nonuniversity maternal education, the nonsusceptible/risk genotype, and no prenatal or postnatal maternal depression (details available in Figure A.1).
Results
Descriptives
There was an almost equal distribution of males and females (Table 1). The mean age for mothers was 30, with half of the mothers highly educated. Specifically, 90 had a university degree or higher. Unstandardized prenatal depression scores ranged from 0 to 49 (M = 12.41, SD = 9.96, α = 0.92). Consistent with oversampling strategies, 26.26% of the women met the threshold for depression at 24 to 36 weeks of pregnancy. The mean prenatal CES-D score for mothers in Montreal (M = 10.6, SD = 7.91) was significantly lower than the mean in Hamilton (M = 14.93, SD = 11.9), t (177) = –2.94, p = .004. For the biallelic categorization of 5-HTTLPR, there were 143 infants with at least one long allele and 36 infants with only the short alleles. For the triallelic categorization of 5-HTTLPR, there were 129 infants with at least one LA allele and 50 infants with only the LG or short alleles. For DRD4, there were 69 infants with at least one 6-8R allele and 110 infants with two 2-5R alleles. Similar distributions have been found in other North American Caucasian samples for DRD4 (Chang, Kidd, Livak, Pakstis, & Kidd, Reference Chang, Kidd, Livak, Pakstis and Kidd1996) and 5-HTTLPR (Hu et al., Reference Hu, Lipsky, Zhu, Akhtar, Taubman and Greenberg2006). There was no significant association between maternal DRD4 genotype, prenatal depression, or NE. Maternal 5-HTTLPR genotype was associated with prenatal maternal depression and NE at 3 and 6 months. There was also no significant association between child 5-HTTLPR, DRD4 genotype, or the multilocus genetic profile score and prenatal depression or NE (refer to Table A.3 for correlation matrix). The demographic and socioeconomic distribution of women in this study was similar to that of women from the Generation R Study and the Avon Longitudinal Study of Parents and their Children, two comparable prenatal cohort studies (van Batenburg-Eddes et al., Reference van Batenburg-Eddes, Brion, Henrichs, Jaddoe, Hofman and Verhulst2013). Given the sample size and the complexity of the interaction model, we examined the number of subjects in all key cells (presence and absence of the susceptibility genes and prenatal depression dichotomized according to the established cutoff), and found an adequate number of participants in every risk group.
Prediction of NE from prenatal maternal depression and 5-HTTLPR from 3 to 36 months
Mixed-model analyses indicated prenatal maternal depression and 5-HTTLPR interacted to predict NE from 3 to 36 months (b = 0.016, SE = 0.01, p = .04; Table 2). Specifically, the association between prenatal depression and NE depended on the presence of child risk/susceptibility alleles (no LA). There was a main effect of child genotype and mothers genotype such that it predicted child's NE (b = 0.139, SE = 0.07, p = .04; b = –0.137, SE = 0.06, p = .03, respectively). Further, the absence of any university education in the mother was associated with an elevated NE score in the child (b = 0.197, SE = 0.08, p = .02), and the average score for postnatal maternal depression covariate (measured at 6, 18, and 36 months) was associated with elevated NE scores (b = 0.016, SE = 0.00, p < .001).
Table 2. The interaction of prenatal maternal depression and child 5-HTTLPR, DRD4, and a multilocus genetic profile score in the prediction of negative emotionality from 3 to 36 months (mixed models)
Note: 5-HTTLPR, Serotonin transporter linked polymorphic region gene; DRD4, dopamine receptor D4 gene; L, long allele; S, short allele. Multilocus genetic profile score –0, 1, or 2 risk/susceptibility genotype: no LA (S/S, S/LG or LG/LG) (5-HTTLPR); six to eight repeats (DRD4). In the model with 5-HTTLPR, McFadden's pseudo R 2 = .065, χ2 (9) = 112.16*** (Akaike information criterion = 920.8). (—) In the model with DRD4, McFadden's pseudo R 2 = .059, χ2 (9) = 112.49***. In the model with the multilocus genetic profile score, McFadden's pseudo R 2 = .066, χ2 (9) = 112.04*** (Akaike information criterion = –922).
†p < .10. *p < .05. **p < .01. ***p < .001.
Prediction of NE from prenatal maternal depression and DRD4 from 3 to 36 months
Mixed-model analyses indicated that prenatal maternal depression and DRD4 interacted to predict NE from 3 to 36 months (b = 0.015, SE = 0.01, p = .02; Table 2). The absence of any university education in the mother was associated with an elevated NE score in the child (b = 0.18, SE = 0.08, p = .03). Postnatal maternal depression (b = 0.018, SE = 0.00, p < .001) and mother's age at birth (b = –0.015, SE = 0.01, p = .03) were also significant predictors of NE.
Prediction of NE from prenatal maternal depression and the multilocus genetic profile from 3 to 36 months
Mixed-model analyses indicated that prenatal maternal depression and the multilocus genetic profile interacted to predict NE from 3 to 36 months (b = 0.013, SE = 0.00, p = .004; Table 2). Maternal education was a significant predictor of NE (b = 0.19, SE = 0.08, p = .02), as was postnatal maternal depression (b = 0.018, SE = 0.00, p < .001) and mother's age at birth (b = –0.017, SE = 0.01, p = .02). The parameter estimates for the interaction in the model with 5-HTTLPR and DRD4 were 0.016 and 0.015, respectively, while that for multilocus genetic profile score ranged from 0.013, for one susceptibility genotype, to 0.026, for two susceptibility genotypes. Similar results were found when the imputed values for missing data were used (for the interaction, b = 0.008, SE = 0.00, p = .04).
Further investigation of the change in the McFadden pseudo R 2 after entering the covariates, main effects, and interaction in three steps revealed that the fit of the model increased from 0.021 to 0.051 after the inclusion of postnatal depression and then increased to 0.066 after the inclusion of main effect and interaction effects of prenatal depression and the multilocus genetic profile.
These results are unique to prenatal depression as a model constructed to predict NE at 18 and 36 months from the interaction of postnatal depression (at 6 and 12 months), and the multilocus genetic profile score was not significant.
Confirmatory analyses found the best model to be the strong differential susceptibility model at 3 and 6 months, and diathesis stress at 36 months. At 3 months (Table 3, Figure 1), the interaction was significant, and the 95% interval for the crossover point fell within the range of the CES-D (between 8.35 and 31.74) and was significantly different from zero (b = 20.04, p < .001). We point out that the nonzero value of the crossover point indicates that carriers of at least one susceptibility allele (no LA or 6-8R) can have a better outcome with respect to NE than carriers of both the LA and the 2-5R allele when exposed to low levels of prenatal depression. At 6 months, the interaction was significant, and the 95% interval for the crossover point fell within the range of the CES-D (1.36 and 22) and was significantly different from zero (b = 11.68, p = .03). The estimates of the crossover points at 3 and 6 months (20.04 and 11.68, respectively) were close to the cutoff point for the presence of a depressive illness, indicating that children with one or two susceptibility alleles tended to be rated lower in NE than those with no susceptibility alleles if the mother was below the clinical cutoff score of 16. Further, these children had higher NE than those without the LA or the 6-8R genotype if the mother was above the clinical cutoff score of 16.
Figure 1. The interaction of prenatal maternal depression and child multilocus genetic profile score (5-HTTLPR and DRD4) in the prediction of negative emotionality at 3, 6, 18, and 36 months (confirmatory analyses). Multilocus genetic profile score = 0, 1, or 2 risk/susceptibility genotype: no LA (S/S, S/LG, or LG/LG; 5-HTTLPR); 6-8R (DRD4). The graphs depict models of strong differential susceptibility at 3 and 6 months and diathesis stress at 36 months.
Table 3. The interaction of prenatal maternal depression and child multilocus genetic profile score (5-HTTLPR and DRD4) in the prediction of negative emotionality at 3, 6, 18, and 36 months (confirmatory regression)
Note: 5-HTTLPR, Serotonin transporter linked polymorphic region gene; DRD4, dopamine receptor D4 gene; L, long allele; S, short allele. Multilocus genetic profile score –0, 1, or 2 risk/susceptibility genotype: no LA (S/S, S/LG or LG/LG) (5-HTTLPR); six to eight repeats (DRD4). At 3 months, R 2 = .24, F (8, 123) = 4.86***; at 6 months, R 2 = .18, F (9, 157) = 3.91***; at 18 months, R 2 = .20, F (9, 152) = 4.33***; and at 36 months, R 2 = .20, F (9, 169) = 4.61***.
a Strong differential susceptibility models are indicated by statistically significant cross over points at 3 and 6 months.
†p < .10. *p < .05. **p < .01. ***p < .001.
The interaction at 18 months was not significantly different from zero. At 36 months, the interaction was significant. However, the 95% interval for the crossover point was not significant from zero, indicating a model of diathesis stress.
Figure 1 depicts the G × E models for the prediction of NE (standardized) at 3, 6, 18, and 36 months. Carriers of the LA and 2-5R alleles were insensitive to prenatal depression exposure, with stable scores of NE throughout the period of study. Carriers of one susceptibility allele (No LA or 6-8R), however, had higher levels of NE as a function of exposure to greater levels of prenatal depression, while carriers of both alleles had even higher levels of NE. With lower prenatal depression, carriers of the susceptibility alleles had lower levels of NE than did noncarriers at 3 and 6 months. This effect is not significant at 36 months. From these results it appears that, as the child gets older, the advantaged conferred from having the susceptible genes and low prenatal exposure diminishes.
In both analyses, the model that included the multilocus genetic profile was the strongest predictor of NE, as confirmed from the parameter estimates, the Akaike information criteria model fit statistic, and McFadden pseudo R 2.
Discussion
The findings of our study suggest that prenatal maternal depression and a child multilocus genetic profile (no LA [only short or LG] for 5-HTTLPR and 6-8R for DRD4) interact to predict early age NE from 3 to 36 months of age. The interaction identified in the mixed-model analysis was replicated at 3, 6, and 36 months of age in confirmatory analyses. Specifically, exposure to prenatal maternal depression was associated with higher NE across the first 3 years of life when carriers had the S/LG allele of the 5-HTTLPR and the 6-8R allele of the DRD4 genotypes. These unique findings are strengthened by the design of the study (i.e., prenatal longitudinal data with multiple time points, refined functional genotyping of the long allele, and complimentary [mixed-model regression] and novel analyses [confirmatory analysis]).
Three findings stand out. First, the interaction between child 5-HTTLPR and DRD4 genotype and prenatal maternal depression predicting NE is strengthened when the genetic factor includes both alleles in the same model. There were separate contributions for the 5-HTTLPR and DRD4 genotypes, such that both moderated the relationship between prenatal maternal depression and NE in separate models. The model that included the multilocus genetic profile was the strongest predictor of NE. These results not only indicate the importance of 5-HTTLPR and DRD4 in predicting early NE in the presence of prenatal maternal depression but also show that multiple genes are likely involved in its development. Even exposure to low levels of depressive symptoms may influence differences in NE in children with the S/LG and 6-8R alleles because significant differences in NE were found between children with both susceptibility alleles and those with no susceptibly alleles even when their mother's depressive symptoms did not meet the cutoff for a diagnosis of prenatal depression. That the relationship between prenatal maternal depression and child NE was greatest in children with both susceptibility alleles also highlights the importance of considering genetic factors in main effect studies. As demonstrated by the small proportion of children in this study manifesting both alleles (10.06%), such findings may be influenced by a small number of highly susceptible children.
The association between prenatal maternal adversity and child development is consistent with the existing literature (e.g., Davis et al., Reference Davis, Glynn, Schetter, Hobel, Chicz-Demet and Sandman2007). For example, Pluess et al. (Reference Pluess, Velders, Belsky, van IJzendoorn, Bakermns-Kranenburg and Jaddoe2011) have reported that prenatal maternal distress and the biallelic 5-HTTLPR predict 6-month NE. Seeing as though we were only able to reproduce this finding when the triallelic categorization was used, questions remain about how divergences in categorization of 5-HTTLPR can influence the detection of an association. If the biallelic categorization identifies some subjects as long, when functionally as LG, they better resemble short, one wonders about the role of measurement error in the underestimation and inconsistencies of G × E findings involving 5-HTTLPR. The role of power and precision in G × E analyses is nicely demonstrated in the sensitivity analyses of Wong et al. (Reference Wong, Day, Luan, Chan and Wareham2003).
The joint effect of DRD4 and 5-HTTLPR is also consistent with the prior literature reviewed. Variants in the genotypes for both 5-HTTLPR (Hayden et al., Reference Hayden, Klein, Sheikh, Olino, Dougherty and Dyson2010; Hariri & Holmes, Reference Hariri and Holmes2006) and DRD4 (Belsky, Bakermans-Kranenburg, & van IJzendoorn, Reference Belsky, Bakermans-Kranenburg and van IJzendoorn2007) are associated with mood regulation and activation. Specifically, serotonin and dopamine have complementary and opposing action, with shared anatomy (e.g., direct projections from the 5-HT raphe nuclei to DA neurons in the substantia nigra; Dray, Gonye, Oakley, & Tanner, Reference Dray, Gonye, Oakley and Tanner1976) and neurophysiological activity (electrical stimulation of the raphe inhibits dopamine neurons in the substantia nigra, an effect mediated by serotonin; Tsai, Reference Tsai1989). The reward-processing circuits and adversity-processing circuits “compete to bias decision-making, motivation and well-being, by the opposite effect of dopamine and serotonin on the activation of each circuit” (Vadovicova & Gasparotti, Reference Vadovicova and Gasparotti2014, p. 7), such that drive, hope, and impulsivity work in opposition with negative affect, discomfort, depression, and worries. With several preclinical models revealing that dopamine and serotonin systems interact to determine corticolimbic responses to environmental adversity (e.g., Cools, Nakamura, & Daw, Reference Cools, Nakamura and Daw2011), we extend these findings to demonstrate that the two genotypes also interact with prenatal depression to predict NE. The finding that the S/LG and 6-8R alleles operate as the susceptibility risk/alleles is consistent with recent studies and meta-analysis (Bakermans-Kranenburg & van IJzendoorn, Reference Bakermans-Kranenburg and van Ijzendoorn2011; Holmboe et al., Reference Holmboe, Nemoda, Fearon, Sasvari-Szekely and Johnson2011). Nonetheless, we suggest that the presence of opposite findings opens the door for a more complex understanding of these genes and cell signalling systems. Finally, there are recent studies reporting that prenatal depression influences structural changes (Sandman, Buss, Head, & Davis, Reference Sandman, Buss, Head and Davis2015) and connectivity (Qiu et al., Reference Qiu, Anh, Li, Chen, Rifkin-Graboi and Broekman2015) of brain regions that influence emotion regulation. Because our findings show that prenatal depression influences the development of NE only in the presence of the DRD4 and 5-HTTLPR susceptibility genotypes, they may be the mechanisms through which these structural changes take place.
Second, the characterization of the interaction changes with the development of the child. A model of differential susceptibility characterizes our findings up to 6 months of age. This finding suggests that exposure to the prenatal environment might be moderated in a bidirectional manner by genotype, and that early in development S/LG and 6-8R carriers have the greatest capacity to benefit from positive environments. A model of diathesis stress characterizes the findings at 36 months where only exposure to higher levels of prenatal depression influenced higher levels of NE reported in S/LG and 6-8R carriers. Questions remain about how to explain divergences in the model across the life span such as developmental windows for positive outcomes and postnatal factors unmeasured in our model.
A key observation from our analyses is that findings were strengthened by the use of a confirmatory model. Confirmatory models maximize statistical power by aligning analyses with hypotheses of interest when two viable alternative models are possible, in this case diathesis stress and differential susceptibility. Our glovelike statistical analyses might explain the differences in our results from Braithwaite et al. (Reference Braithwaite, Ramchandani, O'Connor, van IJzendoorn, Bakermans-Kranenburg and Glover2013). Furthermore, because the model testing the multilocus genetic profile score was stronger than either model testing the separate contribution of 5-HTTLPR and DRD4, we wonder whether their discrepant findings might not also be related to the unmeasured role of DRD4.
Third, the effect of maternal depression on NE operates across the continuum of exposure, from the prenatal to the early postnatal period. Our findings indicate an increasing main effect of postnatal depression on NE from 6 to 36 months. This is consistent with all accounts of the importance of maternal mood in the development of the child (Goodman et al., Reference Goodman, Rouse, Connell, Broth, Hall and Heyward2011). However, postnatal maternal depression did not fully explain the association between prenatal depression and NE, even in a Postnatal × Genotype interaction model. That the prediction model for NE improved when both pre- and postnatal depression were included suggests the importance of depression across the continuum and its increasing influence when sustained from the prenatal to the postnatal period. We are currently further exploring these findings to gain a more complete understanding of its contribution on the development of NE (Gordon Green et al., Reference Gordon Green, Babineau, Bouvette-Turcot, Jolicoeur-Martineau, Minde and St-Andre2014).
Limitations
The design does not allow us to exclude that a G × E correlation (rGE) would better explain our G × E models. We found that mother's 5-HTTLPR genotype was correlated with prenatal maternal depression. Further, in the model examining the interaction of prenatal depression and 5-HTTLPR, mother's 5-HTTLPR genotype was a significant predictor of child NE. Because the interaction was still a significant predictor of NE above and beyond maternal genotype, it is likely that 5-HTTLPR and prenatal maternal depression contribute to the development of NE via both rGE and G × E pathways. The absence of a correlation between infant or mother DRD4 genotype and prenatal depression and of a confounding effect of maternal DRD4 genotype makes it unlikely that passive rGE factors are at play in the model examining the interaction of prenatal depression and DRD4. An evocative rGE remains possible although less likely with a consistent prediction starting at 3 months of age.
Our NE factors were obtained from parent-report measures. Although parent-report questionnaires benefit from a longer observation period (Rothbart, Reference Rothbart1981), parental mood may influence the ratings given to the child (Atella, DiPietro, Smith, & St. James-Roberts, 2009). Given that the IBQ-R and ECBQ specifically inquire about the frequency of observable behaviors, parent-reporting bias is minimized. The concern though that these two types of assessment might measure different aspects of temperament is supported by the low convergence rates between them (Seifer, Sameroff, Barrett, & Krafchuk, Reference Seifer, Sameroff, Barrett and Krafchuk1994). We tried to limit the effect of present parental mood on ratings of infant NE by controlling all our models for current maternal depressive symptoms.
There may be unmeasured confounds. Specifically, mothers who experience prenatal maternal depression may also be vulnerable to adverse environmental factors that could provoke a different type of stress experience by the fetus such as prenatal maternal anxiety. However, previous smaller analyses run by the authors of this investigation on the interaction of genetic susceptibility and anxiety and depression found that prenatal maternal depression had a stronger, separate effect on the development of infant NE than prenatal maternal anxiety (Gordon Green et al., Reference Gordon Green, Babineau, Bouvette-Turcot, Jolicoeur-Martineau, Minde and St-Andre2014), indicating that prenatal maternal depression makes a separate contribution. In addition, recent brain imaging studies with neonates reflect differential effects of prenatal maternal depression compared to depression on corticolimbic brain structures (Qiu et al., Reference Qiu, Anh, Li, Chen, Rifkin-Graboi and Broekman2015; Rifkin-Graboi et al., Reference Rifkin-Graboi, Bai, Chen, Hameed, Sim and Tint2013).
When compared to other genetic studies, the MAVAN has a relatively smaller number of subjects. However, our power is strengthened by the accuracy of our genotyping method, by the increased precision of functional genotyping from the triallelic categorization of 5-HTTLPR, and by the use of confirmatory models (Wong et al., Reference Wong, Day, Luan, Chan and Wareham2003).
Finally, we do not include data on prenatal antidepressant medication exposure. Community estimates of antidepressant use suggest that about 6% of our sample might have been exposed during pregnancy (Cooper, Willy, Pont, & Ray, Reference Cooper, Willy, Pont and Ray2007). There is a slight possibility that the association between prenatal depression and NE might be in part explained by the associated antidepressant exposure in a few cases. Even then, questions remain as to whether antidepressant exposure predicts developmental outcomes via direct causal processes or represents a marker of the severity for the associated prenatal depression (Weikum et al., Reference Weikum, Brain, Chau, Grunau, Boyce and Diamond2013).
Summary and implications
We report that the relation between prenatal depression and NE is better explained by the interaction of prenatal depression and genotype, such that infants with the susceptibility alleles of 5-HTTLPR and DRD4 will develop higher or lower NE depending on the severity of the exposure to prenatal maternal depression. The findings highlight the importance of considering multiple genes in a G × E model with a multilocus genetic profile including genes which in monogenic models seem to have a modest if any effect at all. The present study reports on two candidate genes consistent with preexisting literature. Although secondary analyses did not identify any other genes as significant predictors in the model, this does not preclude that other susceptibility genes could also be operating in the development of NE. For example, Hill et al. (Reference Hill, Breen, Quinn, Tibu, Sharp and Pickles2013) have identified MAOA as a moderator of adverse prenatal experiences and the development of NE at 5 weeks. It will be important for future studies to examine these and other candidate genes in greater detail.
Many studies have reported that the association between prenatal environmental exposure and child development is dependent on timing of gestation (e.g., Davis et al., Reference Davis, Glynn, Schetter, Hobel, Chicz-Demet and Sandman2007; Davis & Sandman, Reference Davis and Sandman2010; O'Connor, Heron, Golding, Beveridge, & Glover, Reference O'Connor, Heron, Golding, Beveridge and Glover2002b), although there is some evidence that behavioral and emotional outcomes are associated only with exposure during later gestation (Davis et al., Reference Davis, Glynn, Schetter, Hobel, Chicz-Demet and Sandman2007; O'Connor, Heron, Golding, Beveridge, & Glover, Reference O'Connor, Heron, Golding, Beveridge and Glover2002). Further examination of exposure to prenatal maternal depression earlier or later during pregnancy is needed to determine if similar associations may be found during different developmental periods.
A closer look at the moderating effect of the postnatal environment is also indicated. Boyce and Ellis (Reference Boyce and Ellis2005) postulate that susceptibly is heritable and influenced by stressful and protective environments. As such, NE is considered a susceptibility factor with positive and negative developmental outcomes documented for infants and children in the presence of certain adaptive or maladaptive environments (for a review, see Belsky & Pluess, Reference Belsky and Pluess2009). Our results reveal that NE is shaped in part by experiences during pregnancy. Consistent with the notion that exposure to stress during pregnancy functions as a primer to increase susceptibility (or vulnerability) to the later environment (Barker, Reference Barker2004; Grant, Sandman, Wing, Dmitrieva, & Davis, Reference Grant, Sandman, Wing, Dmitrieva and Davis2015; Pluess, Reference Pluess2015), such “prenatal programming” of postnatal susceptibility suggests the importance of close attention to the role of postnatal environment in the outcome of these children.
Finally, our findings underline the importance of identifying and treating prenatal depression (O'Connor, Monk, & Fitelson, Reference O'Connor, Monk and Fitelson2014). As noted above, the link between NE and later problematic behavior and psychopathology has been well established (e.g., Eisenberg et al., Reference Eisenberg, Valiente, Spinrad, Cumberland, Liew and Reiser2009). The moderating role of NE in treatment outcome aimed at improving internalizing and externalizing behavior (Blair, Mitchell, & Blair, Reference Blair, Mitchell and Blair2005) suggests that targeting contributing factors of NE could be effective in preventing the development of childhood psychopathology. The results from this study replicate previous findings that prenatal maternal depression has an influence on temperamental vulnerability in the offspring and supports the importance of prevention and early intervention of maternal depressive symptoms (O'Connor et al., Reference O'Connor, Monk and Fitelson2014).
Appendix A
Figure A.1. (Color online) An example representation of confirmatory analysis by Michael Pluess.
Table A.1. Infant Behaviour Questionnaire—Revised factor loadings at 3 months (N = 328) and 6 months (N = 418)
Table A.2. Early Child Behavior Questionnaire—Revised factor loadings at 18 months (N = 405) and 36 months (N = 370)
Table A.3. Spearman correlation matrix
