Developmental psychopathology has long recognized the importance and complexities of two or more disorders co-occurring (e.g., Caron & Rutter, Reference Caron and Rutter1991). In this paper, we shift the focus from comorbidity of disorders to possible co-occurrences among vulnerabilities to the development of psychopathology. Building on a long and honored history (Garmezy, Reference Garmezy1971; Masten & Garmezy, Reference Masten, Garmezy, Lahey and Kazdin1985; Murphy & Moriarty, Reference Murphy and Moriarty1976; Werner & Smith, Reference Werner and Smith1977), the concept of vulnerability to psychopathology fits well with an ontogenic process perspective (Beauchaine & McNulty, Reference Beauchaine and McNulty2013). Better understanding of vulnerability has the potential to reveal key mechanisms in the development of psychopathology (Ingram & Price, Reference Ingram, Price, Ingram and Price2010). In this paper, we build on the literature on specific individual vulnerabilities by proposing and testing a model of co-occurrence between infant vulnerabilities at two levels of analysis, each of which has been implicated in the intergenerational transmission of risk for psychopathology in infants born to women at elevated risk for perinatal depression, and examining the roles of prenatal and postnatal depression exposure on their co-occurrence. We further consider continuities and discontinuities, over the first year of infants' lives, in the patterns of co-occurrence among vulnerabilities, and the potential role of mothers' prenatal and postnatal depressive symptom levels in those patterns of continuities and discontinuities of co-occurrence. Specifically, we examine levels of prenatal depressive symptoms in mothers as a potential moderator in the co-occurrence between two vulnerabilities and in the continuity and discontinuity of co-occurrence and also examine questions of dose and timing, in terms of potential exposure to mothers' depression prenatally, in the postpartum, or in both developmental time periods.

At the psychophysiological level of analysis of vulnerability, asymmetry in frontal EEG activation during a resting baseline has been well established as an index of individual differences in emotion regulation beginning in infancy, with numerous studies finding that greater relative left frontal activation is associated with approach behavior/positive affect and greater relative right frontal activation is associated with withdrawal behavior/negative affect (e.g., Fox, Calkins, & Bell, Reference Fox, Calkins and Bell1994; Fox & Davidson, Reference Fox and Davidson1984). In particular, measures of resting frontal EEG asymmetry are thought to reflect traitlike tendencies and styles of processing affective information (Coan & Allen, Reference Coan and Allen2004). Prenatal and postpartum depression in women have both been found to be associated with the pattern of right frontal EEG asymmetry in their infant offspring, with moderate effect sizes (as meta-analytically reviewed by Thibodeau, Jorgensen, & Kim, Reference Thibodeau, Jorgensen and Kim2006); and that pattern in infants has been found to be prospectively associated with early indices or precursors of psychopathology, such as more stable observed behavioral inhibition over the first 4 years of life (Fox, Henderson, Rubin, Calkins, & Schmidt, Reference Fox, Henderson, Rubin, Calkins and Schmidt2001). Overall, resting frontal EEG asymmetry scores, beginning within the first month of life, have been found to be at least moderately stable over the course of infancy (Bell & Fox, Reference Bell, Fox, Dawson and Fischer1994; Field, Hernandez-Reif, & Diego, Reference Field, Hernandez-Reif and Diego2006; Fox, Bell, & Jones, Reference Fox, Bell and Jones1992), although stability between late infancy and 24 months of age is low (Howarth, Fettig, Curby, & Bell, Reference Howarth, Fettig, Curby and Bell2015).

At the behavioral level of analysis of vulnerability to the later development of psychopathology in infants, most attention has focused on trait-level negative affectivity (NA) temperament. As conceptualized by Derryberry and Rothbart (Reference Derryberry, Rothbart, Kalverboer and Gramsbergen2001), temperament NA refers to infants' tendencies to engage particular reactive processes in the face of negative emotions such as frustration or fear. Although temperament is understood to reflect behavioral tendencies to regulate emotion and physiological processes, it is typically measured in relation to patterns of behavior. Both prenatal and postpartum depression in mothers have been associated with infants' greater NA (Davis et al., Reference Davis, Snidman, Wadhwa, Glynn, Schetter and Sandman2004, Reference Davis, Glynn, Schetter, Hobel, Chicz-Demet and Sandman2007; Huot, Brennan, Stowe, Plotsky, & Walker, Reference Huot, Brennan, Stowe, Plotsky and Walker2004; Rouse & Goodman, Reference Rouse and Goodman2014). Conversely, NA in infants has been prospectively associated with higher levels of internalizing and externalizing problems among toddlers and preschool-aged children (Gartstein, Putnam, & Rothbart, Reference Gartstein, Putnam and Rothbart2012). Especially striking in terms of infant vulnerability to psychopathology, the specific infant temperament tendency to express negative emotions has been found to predict behavior problems and psychopathology as far forward as middle childhood (Bates, Bayles, Bennett, Rige, & Brown, Reference Bates, Bayles, Bennett, Ridge and Brown1991; Sayal, Heron, Rowe, & Ramchandani, Reference Sayal, Heron, Maughan, Rowe and Ramchandani2014) and even adolescence (Guerin, Gottfried, & Thomas, Reference Guerin, Gottfried and Thomas1997; Teerikangas, Aronen, Martin, & Huttunen, Reference Teerikangas, Aronen, Martin and Huttunen1998). Moreover, the temperament construct of NA can be reliably identified as early as 3 months of age (Gartstein & Rothbart, Reference Gartstein and Rothbart2003). Although NA has been found to be relatively stable over periods of 3 months within infancy, it generally increases over infants' first year of life (Bridgett et al., Reference Bridgett, Gartstein, Putnam, McKay, Iddins and Robertson2009; Gartstein & Rothbart, Reference Gartstein and Rothbart2003; Rothbart, Reference Rothbart1986). Despite this shift toward increasing NA over infancy, infant NA was found to be strongly associated with NA in preschool-aged children (Putnam, Rothbart, & Gartstein, Reference Putnam, Rothbart and Gartstein2008).

Given the strength of the evidence for each of these psychophysiological and behavioral levels of analysis of vulnerabilities in the intergenerational transmission of risk from depression in mothers, there are important reasons to consider how they may be associated with each other. The Goodman and Gotlib (Reference Goodman and Gotlib1999) model for the transmission of risk for the development of psychopathology from depression in mothers to their children argues for the importance of multiple domains of infant functioning to index vulnerabilities. Moreover, it is increasingly understood not only that risk factors, such as depression in mothers, are likely to be associated with heterogeneous vulnerabilities but also that no single vulnerability in infants is likely to explain the development of psychopathology (Cicchetti & Rogosch, Reference Cicchetti and Rogosch1996). In this paper, we consider the importance of understanding co-occurrences across psychophysiological and behavioral levels of analysis of vulnerabilities to the development of psychopathology in infants of mothers with elevated risk for perinatal depression given their history of depression prior to pregnancy.

In a seminal paper on domains of emotion regulation, Bauer, Quas, and Boyce (Reference Bauer, Quas and Boyce2002) called for investigation of how multiple systems relate to each other, albeit with a specific focus on biological systems, as being promising in enhancing prediction of later psychopathology. Consistent with that understanding, support has been emerging for the importance of studying co-occurrences across multiple levels of analysis of vulnerabilities (e.g., Obradović, Reference Obradović2016; Rotenberg & McGrath, Reference Rotenberg and McGrath2016). Systems working in sync with each other, which is sometimes referred to as coupling or coherence, may suggest more adaptive emotion regulation processes. For example, DiPietro, Costigan, and Pressman (Reference DiPietro, Costigan and Pressman2002) and DiPietro, Costign, Shupe, Pressman, and Johnson (Reference DiPietro, Costigan, Shupe, Pressman and Johnson1998) have shown that greater coupling of fetal heart rate and body movement is associated with a higher level of integration of the central nervous system and, later, with better infant state regulation; that such coupling increases with gestational age; and that lower coupling is found in fetuses of lower socioeconomic status women and those with higher daily stress levels.

Coupling of vulnerabilities in infants may similarly be suggestive of more adaptive regulation. Beauchaine (Reference Beauchaine2015) and Marsh, Beauchaine, and Williams (Reference Marsh, Beauchaine and Williams2008) argue that dysynchrony across levels of analysis, in particular between behavioral (e.g., observed sadness) and physiological (e.g., respiratory sinus arrhythmia) systems, may be a marker of emotion dysregulation. Similarly, Nigg (Reference Nigg2006) suggested that if physiological and temperamental domains do not coregulate, there is increased risk for the development of psychopathology. That is, it may be most adaptive when the systems at different levels of analysis work in sync with each other, even if both reflect a vulnerability; a mismatch (lower coherence or less association) between vulnerabilities may reflect greater risk for the development of psychopathology. Specifically, temperament NA and right frontal EEG asymmetry are measures of highly related systems, with both being understood to reflect tendencies toward behavioral withdrawal (e.g., Fox, Reference Fox1991; Rothbart & Derryberry, Reference Rothbart, Derryberry, Lamb and Brown1981). Thus, examining these two levels of analysis of vulnerability allows for consideration that depression exposure may be associated with dysynchrony between these two systems, suggesting emotional dysregulation.

In general population samples, not taking mothers' depression into account, studies of associations between infants' EEG and temperament (psychophysiological and behavioral levels of analysis of vulnerabilities to the later development of psychopathology, respectively) have yielded mixed findings. However, consistent with the longstanding understanding that resting frontal EEG asymmetry is an early psychophysiological marker of NA in infancy and that temperament is understood to reflect biologically based (i.e., nervous system) dispositions (e.g., Fox, Reference Fox1991; Rothbart & Derryberry, Reference Rothbart, Derryberry, Lamb and Brown1981), infants' resting relative right frontal EEG asymmetry has been found to be associated with higher maternal-rated fear in 9-month-old infants (Schmidt, Reference Schmidt2008), with greater distress in response to maternal separation in a small sample of 10-month-olds (Davidson & Fox, Reference Davidson and Fox1989), and with more inhibited behavior at 14 months among the subset of infants who had exhibited high levels of reactivity to novel sensory stimuli at 4 months of age (Calkins, Fox, & Marshall, Reference Calkins, Fox and Marshall1996). Resting relative right frontal EEG asymmetry and fear temperament (a component of NA), however, were not significantly correlated in 10-month-olds; moreover, neither 10-month-olds' EEG asymmetry predicted fear in 24-month-olds nor did 10-month-olds' fear predict EEG asymmetry at 24 months of age (Howarth et al., Reference Howarth, Fettig, Curby and Bell2015). The mixed support for associations between these two levels of analysis of vulnerability (psychophysiological [EEG] and behavioral [NA]) suggests a possible role of a moderator variable, in that for different subsets of infants, the direction of the association may differ. We explored the role of exposure to maternal prenatal depression as a possible moderator in that prenatal depression may disrupt the expected coherence and be associated with lower levels of association between the two levels of analysis of vulnerabilities. We further considered the possibility that the ongoing development of these systems may reveal reorganization, potentially yielding changes in coherence over the course of infancy.

How might exposure to mothers' depression during pregnancy be expected to be associated with the extent to which vulnerabilities co-occur? Heritability is one consideration in that, rather than inheriting a risk for depression per se, infants of depressed mothers are understood to inherit a predisposition to depression, including either trait negative affectivity (temperament) or relative right frontal resting EEG asymmetry or both. General tendencies at each of these two levels of analysis may be what is inherited, with potentially different genetic factors accounting for each. These behavioral and psychophysiological dispositions with which infants of depressed mothers might be born would then be expected to play out over time through complex Gene × Environment correlations (Rothbart, Sheese, Ryeda, & Posner, Reference Rothbart, Sheese, Rueda and Posner2011). In families in which the mother continues to experience depression postnatally, even if episodically, and the child is exposed to the stressors known to be associated with depression in mothers (Hammen, Reference Hammen, Goodman and Gotlib2002), the child's particular genetic vulnerabilities may be potentiated, which we would expect to be in the form of emotion dysregulation. Gene × Environment evocation likely would also be at play given that infants born with these behavioral and psychophysiological tendencies would influence their environments by, for example, adding to the challenge of a depressed mother to engage in sensitive, responsive care (Lenroot & Giedd, Reference Lenroot and Giedd2011).

In addition to heritability, another consideration in how mothers' prenatal depression might be associated with lower co-occurrence among vulnerabilities relative to vulnerabilities in infants of mothers low in depression is fetal programming. Fetal programming refers to influences on a fetus's development of neuroregulatory systems. As such, mothers' depression during pregnancy may influence fetal development in ways that are expressed as offspring vulnerabilities at multiple levels of analysis (see Lewis, Austin, Knapp, Vaiano, & Galbally, Reference Lewis, Austin, Knapp, Vaiano and Galbally2015, for a recent review of this literature). The influences on fetal neuroregulatory systems would be expressed in infants as behavioral traits, such as NA, and activity of brain systems, such as are reflected in frontal EEG asymmetries (Calkins, Reference Calkins1994). The major challenge for this conceptual model has been with the null or inconsistent finding on particular mechanisms that might explain what it is about fetal exposure to the mother's depression that matters for babies. Current considerations include cortisol or other stress hormones (Zijlmans, Riksen-Walraven, & de Weerth, Reference Zijlmans, Riksen-Walraven and de Weerth2015), epigenetic influences, and placental programming. Overall, the understanding is that depression during pregnancy represents multiple risks to offspring. However, given that most studies examined a single system or level of analysis, there has not been the opportunity to consider the extent to which prenatal depression exposure might disrupt the coherence of systems at different levels of analysis. For example, developmental plasticity may explain how the fetus may adapt to some aspects of its intrauterine environment, but systems may be differentially plastic and fetal programming effects may be highly specific (Lewis et al., Reference Lewis, Austin, Knapp, Vaiano and Galbally2015).

Moreover, it is important to understand how co-occurrences across levels of analysis of vulnerability may change or reorganize postnatally over the course of infant development. For example, there is theoretical support for developmental changes in the correlation between behavioral and psychophysiological markers of vulnerability (Beauchaine, Reference Beauchaine2001); in other words, as normative behavioral changes take place across development, the relation between psychophysiological and behavioral levels of analysis may also change. Further, it has been suggested that behavioral traits and neural systems associated with emotion regulation may change at differing rates across infant development (Kagan & Snidman, Reference Kagan and Snidman2004). As such, pairs of vulnerabilities may continually adapt, reorganize, and differentiate from each other over time (Nigg, Reference Nigg2006), with the latter point being most central to our premise. Nevertheless, there is little knowledge of how systems at different levels of analysis may relate to each other differently at different points over the course of infancy. In particular, it is essential to understand the extent to which prenatal exposure to mothers' depressive symptom levels might explain patterns of association among vulnerabilities not only early in infant development but also over the course of infants' development.

A highly practical concern is that it is not currently known to what extent the pattern of vulnerabilities seen in young infants foretells their vulnerabilities later in infancy. Might we expect infants to recover from vulnerabilities associated with fetal exposures? Although there was insufficiently strong justification for a specific hypothesis, contributing to this understanding was one of the aims of this study. Further, given that postnatal depression is associated with mothers showing less of the sensitive responsiveness that infants need for their developing emotion regulation (Calkins, Reference Calkins1994; Kopp, Reference Kopp1989), we also take postnatal depression exposure into account by considering the added role of postnatal depression and the relation of both dose and timing of exposure to maternal depression to continuity/discontinuity of the co-occurrences among the vulnerabilities.

Method

Participants

This study takes advantage of a prospective, longitudinal investigation of a unique sample of women at high risk for perinatal depression and their infants as part of a larger study, Perinatal Stress and Gene Influences: Pathways to Infant Vulnerability. Specifically, we focused on the subset of participants (n = 234) who completed at least one infant laboratory visit at 3, 6, or 12 months of age. There were eight sets of twins in the sample, yielding a sample size of 242 total infants. Women were recruited during pregnancy through a women's mental health program in a psychiatry department. The main inclusion criterion, indexing risk for perinatal depression, was women having met DSM-IV diagnostic criteria for depression at some point in their lifetimes. Reflecting the expected high rates of comorbidity between depression and anxiety, the women varied in terms of which disorder had been their primary lifetime disorder. For 82% of the women, major depressive disorder or depressive disorder not otherwise specified was their primary lifetime diagnosis; of these women, 30% also met criteria for panic disorder, 16% for obsessive compulsive disorder, 29% for generalized anxiety disorder, and 17% for social anxiety disorder. For the remaining 18% of the women, their primary lifetime diagnosis was an anxiety disorder (panic disorder, obsessive compulsive disorder, generalized anxiety disorder, or social anxiety disorder); all of these women also met criteria for major depressive disorder or depressive disorder not otherwise specified. Further inclusion criteria included less than 16 weeks pregnant measured from last menstrual period at intake and between ages 18 and 45. Exclusion criteria included active suicidality or homicidality, psychotic symptoms, bipolar disorder, schizophrenia, or currently active eating disorder, active substance use disorder within 6 months prior to last menstrual period or positive urine drug screen, illness requiring treatment that can influence infant outcomes (epilepsy, asthma, or autoimmune disorders), and abnormal thyroid stimulating hormone or clinically significant anemia.

Participating mothers ranged from 20.7 to 44.5 years of age at delivery (M = 33.84 years, SD = 4.49). Most (88%) were married. On average, mothers had completed 16.51 years (SD = 2.01) of education. Nearly half (43%) were primiparous. Total Hollingshead scores for women and their partners were on average 50.71 (SD = 8.98), indicating middle to upper middle class. The majority of the mothers were European American (89%), with the remaining 9% being African American, 1% Asian, and <1% each Native American and multiracial. Of the infants, 48% were female and 52% were male; the majority (77%) were born at term. Similar to findings with other samples of pregnant women with histories of major depressive episodes (Goodman & Tully, Reference Goodman and Tully2009), this sampling strategy in the current sample yielded 82 (34%) mothers who had at least one major depressive episode during the prenatal period and 55 (23%) who had at least one episode during the postpartum year.

Procedure

Data were collected as part of a larger, longitudinal study examining pathways to infant vulnerability to the development of psychopathology. All women participated in an informed consent procedure, and all procedures were approved by the Emory University Institutional Review Board. Data including depression symptom levels were collected from the women at multiple time points throughout pregnancy and the first 12 months postpartum. During pregnancy, women completed an average of 5.48 Beck Depression Inventory scales (BDI; Beck, Ward, Mendelsohn, Mock, & Erbaugh, Reference Beck, Ward, Mendelsohn, Mock and Erbaugh1961), with a range from 1 to 13 times (SD = 1.69). During the postpartum year, women completed an average of 5.74 BDI scales, with a range of 1 to 18 times (SD = 2.60). Mothers and infants participated in lab visits at infant ages 3, 6, and 12 months. During these visits, infants' EEG was collected during a 3-min baseline. Women also completed the Infant Behavior Questionnaire—Revised (IBQ-R) as a measure of infant temperament at each of the three infant ages.

At the visits when the infants were 3, 6, and 12 months of age, prior to the baseline segment, an EEG cap was secured to the infant's head while a research assistant manipulated toys in order to distract the infant. The baseline segment was designed to keep the infant quiet and alert and minimize eye movements and gross motor movements. Infants sat on their mothers' laps, and a research assistant blew bubbles for the infants to watch (Dawson, Panagiotides, Klinger, & Spieker, Reference Dawson, Panagiotides, Klinger and Spieker1997). Mothers were instructed not to talk to their infants during the EEG recording.

Measures

Maternal depression: BDI

The BDI (Beck et al., Reference Beck, Ward, Mendelsohn, Mock and Erbaugh1961) is a 21-item self-report measure of depression symptom severity in the past week, with each item rated on a 4-point scale, ranging from 0 to 3. The score is a sum across items, with higher scores indicating greater severity of depressive symptoms. Scores of 0–9 indicate no depression, 10–18 indicate mild–moderate depression, 19–29 indicate moderate–severe depression, and 30–63 indicate severe depression (Beck et al., Reference Beck, Ward, Mendelsohn, Mock and Erbaugh1961). The BDI has been found to be both a valid and a reliable measure of depression severity, with an especially high degree of content validity and internal consistency reliability including during pregnancy (Beck et al., Reference Beck, Ward, Mendelsohn, Mock and Erbaugh1961; Ji et al., Reference Ji, Long, Newport, Na, Knight and Zach2011). For pregnancy, we calculated area under the curve (AUC) scores for each woman to represent the overall depression severity level across the pregnancy, standardized to a 40-week pregnancy. Although analyses were conducted using total AUC scores, both AUC scores and mean BDI scores during pregnancy are shown in Table 1 for ease of interpretation. For the postpartum, analyses were conducted using AUC scores calculated for each woman to represent the overall depression severity level during the first 12 months postpartum. For descriptive purposes, these AUC scores were divided by number of weeks (52) in order to determine the average weekly BDI score. Descriptive statistics for the total AUC scores and the weekly AUC scores are shown in Table 1. Because both the prenatal and the postpartum AUC variables were significantly skewed (D = 0.12 and 0.14, respectively, ps < .001), these variables were transformed using a square root transformation.

Table 1. Descriptive statistics of maternal depressive symptom levels

Note: Depression symptoms were measured with the Beck Depression Inventory; AUC, area under the curve.

Behavioral level of analysis of infant vulnerability: Temperament

Mothers completed the 191-item scale of infant temperament, the IBQ-R (Gartstein & Rothbart, Reference Gartstein and Rothbart2003), which is factor analytically based on Rothbart and Derryberry's (Reference Rothbart, Derryberry, Lamb and Brown1981) definition of temperament. Mothers rate the items regarding the infant's behavior during the past week in a variety of domains on a 7-point scale, from 1 (never) to 7 (always). The questionnaire yields 14 scales, with 10 to 18 items per scale and scale scores being the mean of items on that scale. Scales cluster into three overarching factor scores: orienting/regulatory capacity, surgency/extraversion, and NA. The variable of interest in this study, NA, is calculated as the mean of four scales: falling reactivity, fear, frustration/distress to limitations, and sadness, with a possible range of 1 to 7. Higher scores indicate higher levels of NA, which is conceptualized as poorer emotion regulation.

The IBQ-R is a reliable and valid index of infant temperament (Gartstein & Rothbart, Reference Gartstein and Rothbart2003) and has been proven to be resistant to the potentially biasing influence of parental depression (Gartstein & Marmion, Reference Gartstein and Marmion2008). For the current sample, the α coefficient for NA at 3 months was 0.71 and for the scales were as follows: falling reactivity (0.83), fear (0.89), frustration/distress to limitations (0.71), and sadness (0.86). At 6 months, the α coefficient(s) for NA was 0.76 and for the scales were 0.88 for falling reactivity 0.88 for fear, 0.82 for frustration/distress to limitations, and 0.82 for sadness. Finally, the α coefficient(s) for NA at 12 months was 0.87 and for the scales were 0.84 for falling reactivity, 0.89 for fear, 0.82 for frustration/distress to limitations, and 0.84 for sadness.

Descriptive statistics for the infant temperament (NA) scores can be found in Table 2. Of note, there was a restricted range of NA scores in this sample, with most infants falling at a moderate level of NA. This pattern is consistent not only with the sample on which the IBQ-R was developed (Gartstein & Rothbart, Reference Gartstein and Rothbart2003) but also with similar population samples (Rouse & Goodman, Reference Rouse and Goodman2014). The mean NA score in our sample was somewhat higher compared to Gartstein and Rothbart's normative data and essentially the same as the mean NA score for 3-month-olds in the Rouse and Goodman sample. The NA variable was significantly skewed at 3 months of age (D = 0.08, p < .001); a square root transformation was used for all of the NA scores.

Table 2. Descriptive statistics of infant variables

^a Negative affectivity was measured by the Infant Behavior Questionnaire—Revised.

Psychophysiological level of analysis of infant vulnerability: EEG

The resting EEG recordings were made from 16 left and right scalp sites: frontal pole (Fp1, Fp2), medial frontal (F3, F4), lateral frontal (F7, F8), central (C3, C4), anterior temporal (T3, T4), posterior temporal (T7, T8), parietal (P3, P4), and occipital (O1, O2), referenced to Cz. EEG was recorded using a stretch cap (Electro-Cap, Inc., Eaton, OH) with electrodes in the 10/20 system. After the cap was placed on the infant's head, a small amount of abrasive gel was placed into each recording site and the scalp gently rubbed. Following this, conductive gel was placed in each site. Electrode impedances were measured and accepted if they were below 5K ohms. The electrical activity from each lead was amplified using separate James Long Company Bioamps (Caroga Lake, NY) and bandpassed from 1 to 100 Hz. Activity for each lead was displayed on the monitor of the acquisition computer. The EEG signal was digitized online at 512 samples per second for each channel so that the data were not affected by aliasing. The acquisition software was Snapshot-Snapstream (HEM Data Corp., Southfield, MI), and the raw data were stored for later analysis.

Infant EEG data were examined and analyzed using EEG Analysis System software developed by James Long Company (Caroga Lake, NY). The data were rereferenced via software to an average reference configuration. Average referencing, in effect, weighted all the electrode sites equally and eliminated the need for a noncephalic reference. Active (F3, F4, etc.) to reference (Cz) electrode distances vary across the scalp. Without the rereferencing, power values at each active site may reflect interelectrode distance as much as they reflect electrical potential. The average reference configuration requires that a sufficient number of electrodes be sampled and that these electrodes be evenly distributed across the scalp. Currently, there is no agreement concerning the appropriate number of electrodes (Davidson, Jackson, & Larson, Reference Davidson, Jackson, Larson, Cacioppo, Bernston and Tassinary2000; Hagemann, Naumann, & Thayer, Reference Hagemann, Naumann and Thayer2001; Luck, Reference Luck2005), although the 10/20 configuration that we used does satisfy the requirement of even scalp distribution.

The average reference EEG data were artifact scored for eye blinks using Fp1 and Fp2 (Myslobodsky et al., Reference Myslobodsky, Coppola, Bar-Ziv, Karson, Daniel and van Praag1989), with a peak to peak criterion of 100 uV or greater. Artifacts associated with gross motor movements over 200 uV peak to peak were also scored. These artifact-scored epochs were eliminated from all subsequent analyses. The data then were analyzed with a discrete Fourier transform using a Hanning window of 1-s width and 50% overlap. Power was computed for the 6 to 9 Hz frequency band. Infants and young children have a dominant frequency between 6 and 9 Hz (Bell & Fox, Reference Bell, Fox, Dawson and Fischer1994; Marshall, Bar-Haim, & Fox, Reference Marshall, Bar-Haim and Fox2002), and this particular frequency band has been correlated with patterns of emotion reactivity and emotion regulation during infancy (Bell & Fox, Reference Bell, Fox, Dawson and Fischer1994; Buss et al., Reference Buss, Schumacher, Dolski, Kalin, Goldsmith and Davidson2003; Dawson, Reference Dawson1994) and early childhood (Fox et al., Reference Fox, Henderson, Rubin, Calkins and Schmidt2001). The power was expressed as mean square microvolts and the data transformed using the natural log to normalize the distribution.

Frontal EEG asymmetry values were computed by subtracting ln power at left frontal (F3) from ln power at right frontal (F4). In infants and young children, power in the 6 to 9 Hz band has been shown to be inversely related to cortical activation during emotion reactivity and regulation (Fox, Reference Fox1994). Thus, a negative asymmetry score reflects greater relative right frontal activation (conceptualized as poorer emotion regulation), whereas a positive asymmetry score reflects greater relative left frontal activation. Descriptive statistics on the infant frontal EEG asymmetry values can be found in Table 2. The frontal EEG asymmetry values of the infants in our study are similar to those of infants in other studies focused on depressed and nondepressed mothers (e.g., Dawson, Klinger, Panagiotides, Spieker, & Frey, Reference Dawson, Klinger, Panagiotides, Spieker and Frey1992; Jones, Field, & Almeida, Reference Jones, Field and Almeida2009; Jones, Field, Fox, Davalos, & Gomez, Reference Jones, Field, Fox, Davalos and Gomez2001). The EEG variable was significantly skewed at 3 and 6 months of age (D = 0.07, p < .02 and D = 0.09, p < .001, respectively); therefore, this variable was transformed using a square root transformation at all three ages.

Data analytic strategy

Preliminary analyses

Data were checked for outliers, and descriptive statistics were run using IMB SPSS Statistics 22.

Hypothesis testing

We addressed the primary aim of this study, to examine the moderating role of mothers' depressive symptoms on the co-occurrence among vulnerabilities, as well as on the continuity/discontinuity among vulnerabilities over the course of infancy, using multilevel modeling (MLM) conducted in SPSS. Given that the association between infant frontal EEG asymmetry and NA is understood to be bidirectional (Howarth et al., Reference Howarth, Fettig, Curby and Bell2015), yet MLM requires specifying one variable as a predictor, two models were run: frontal EEG asymmetry predicting NA and NA predicting frontal EEG asymmetry. The independent variable (one of the two infant vulnerabilities) was centered at infant age 3 months. Maternal prenatal depression was indexed by AUC, and square root transformation was performed to correct the skewness of this index. This index was then grand mean centered to be included in the analyses. The three-way interaction (Infant Vulnerability × Age × Depression) was estimated along with all the lower order terms of these three variables included in the model to provide unbiased estimates of the three-way interaction.

Next we tested the hypothesis that with higher levels of prenatal depression, infant frontal EEG asymmetry and NA would be less in sync with each other: that is, a positive association between NA and frontal EEG asymmetry, such that higher infant NA is associated with lower relative right frontal EEG asymmetry. This hypothesis would be supported by, first, a significant three-way interaction between one of the infant vulnerabilities, age, and prenatal depression grand mean centered in predicting the other of the infant vulnerabilities. A planned contrast, examining the two-way interactions (Infant Vulnerability × Age) predicting the other infant vulnerability, tested separately for infants of mothers who were above or below the median split in terms of prenatal depressive symptoms, would reveal that the association between frontal EEG asymmetry and NA differs based on infant age and level of prenatal depression. The latter analysis also enabled us to examine our exploratory hypothesis regarding the role of infant age in potential changes in the associations between NA and EEG over infancy, in relation to levels of mothers' prenatal depression symptoms. Marginal R ² from the multilevel model was calculated in SAS Version 9.4 (SAS Institute, Cary, NC) to index the variance explained by the fixed factors (Engert, Plessow, Miller, Kirschbaum, & Singer, Reference Engert, Plessow, Miller, Kirschbaum and Singer2014; Nakagawa & Schielzeth, Reference Nakagawa and Schielzeth2013).

In order to examine the impact of dose and timing of exposure to mother's depression, we ran MLM in SPSS. To examine dose, a categorical variable was created in which women were coded as having exceeded the BDI cutoff (≥10) for clinically significant levels of depressive symptoms during either the prenatal or the postpartum period, during both the prenatal and the postpartum period, or during neither the prenatal nor the postpartum period. To examine timing, a categorical variable was created in which women were coded as having exceeded the BDI cutoff (≥10) for clinically significant levels of depressive symptoms during only the prenatal period, during only the postpartum period, or during both the prenatal and the postpartum period. For both dose and timing, we followed up significant models from the primary hypothesis testing (frontal EEG asymmetry predicting NA and NA predicting frontal EEG asymmetry). The independent variable (one of the two infant vulnerabilities) was centered at infant age 3 months. The three-way interaction (Infant Vulnerability × Age × Depression Dose/Timing) was estimated along with all the lower order terms of these three variables included in the model to provide unbiased estimates of the three-way interaction. Support for the association between infant frontal EEG asymmetry and NA differing based on dose/timing of depression and age would be found first with a significant three-way interaction between one of the infant vulnerabilities, age, and depression group in predicting the other of the infant vulnerabilities. Second, post hoc comparisons between all three groups would reveal significant differences between the depression groups, which could be further explored with the two-way interaction (Infant Vulnerability × Age) predicting the other infant vulnerability separately for each of the depression groups.

Results

Preliminary analyses

Tests of normality and outliers

As is expected with longitudinal designs, there were missing data points. MLM has been found to be a useful strategy for analyzing between- and within-subject predictors in the context of missing data (e.g., Zautra et al., Reference Zautra, Davis, Reich, Nicassio, Tennen and Finan2008). The following describes missing data patterns.

A total of 220 women and their infants (91% of the total sample) completed the lab visit at infant age 3 months, and 22 (9%) missed the visit. At infant age 6 months, 209 women and their infants (86%) participated in the lab visit, 22 (9%) missed the visit, and 11 (5%) were dropped from the study prior to the visit. Finally, at infant age 12 months, 166 women and their infants (69%) completed the visit, 44 (18%) missed the visit, and 32 (13%) were dropped from the study prior to their visit. A total of 147 women and their infants (61%) completed all three visits; a total of 206 women and their infants (85%) completed at least two visits.

Of the 220 infants and their mothers who participated in the 3-month lab visit, 200 infants (91%) had usable resting EEG data; data for 9 infants (4%) were unable to be edited due to too much artifact, data for 7 infants (3%) were excluded for an insufficient quantity (less than 10 s) of artifact-free data, data for 2 infants (1%) were excluded as outliers (±3 SD from the mean), data for 1 infant (<1%) was unusable due to technical problems, and 1 infant (<1%) had no EEG data collected. In terms of infant NA data, 215 infants (98%) had IBQ-R data, and for 5 infants (2%), the IBQ-R was not collected. A total of 196 infants (89%) had usable data for both variables at 3 months.

Of the 209 infants and their mothers who participated in the 6-month lab visit, 194 infants (93%) had usable resting EEG data; data for 7 infants (3%) were unable to be edited due to too much artifact, data for 3 infants (1%) were excluded for an insufficient quantity (<10 s) of artifact-free data, data for 1 infant (<1%) was excluded as an outlier (±3 SD from the mean), data for 2 infants (1%) was unusable due to technical problems, and 2 infants (1%) had no data collected. In terms of infant temperament data, 200 infants (96%) had IBQ-R data, and for 9 infants (4%), the IBQ-R was not collected. Altogether, a total of 186 (89%) infants had usable data for both variables at 6 months.

Finally, of the 166 infants and their mothers who participated in the 12-month lab visit, 145 infants (87%) had usable resting EEG data; data for 7 infants (4%) were unable to be edited due to too much artifact, data for 4 infants (2%) were excluded for an insufficient quantity (<10 s) of artifact-free data, data for 2 infants (1%) were excluded as outliers (±3 SD from the mean), data for 2 infants (1%) were unusable due to technical problems, and 6 infants (4%) had no data collected. In terms of infant temperament data, 162 infants (98%) had IBQ-R data, and for 4 infants (2%), the IB-R was not collected. Altogether, a total of 140 infants (84%) had usable data for both variables at 12 months.

Identification of potential control variables

In order to test for possible confounding variables, we tested associations between demographic variables (maternal age, infant's gestational age at birth, infant gender, and antidepressant usage during pregnancy or the first year postpartum) and the two infant variables (NA and frontal EEG asymmetry) at each infant age. Maternal age was only associated with infant's NA at 6 months of age at a trend level, r (198) = –.14, p = .05; all other associations between maternal age and infant variables at 3, 6, and 12 months of age were not significant (rs < .10, ps > .19). Infant's gestational age at birth was not significantly associated with either infant variable at 3, 6, or 12 months of age (rs < .13, ps > .10). The only sex difference in infant variables was in infant frontal EEG asymmetry at 6 months, t (192) = 2.10 p = .04, in which case, female infants had greater relative right frontal EEG asymmetry than male infants. There were no other sex differences in the infant variables (ts < 2.10, ps > .05). The number of prenatal weeks that infants were exposed to maternal antidepressant use was not significantly associated with any infant variables at any age (rs < .11, ps > .19), nor was the number of postpartum weeks that mothers were taking antidepressants (rs < .15, ps > .06). Given this pattern of findings, no control variables were included in analyses.

Descriptive analyses

Results of Pearson product moment correlations (see Table 3) revealed no significant concurrent associations between the infant frontal EEG asymmetry and NA variables at 3, 6, or 12 months of age. In terms of prospective associations, there was strong evidence of stability of infant NA over time and evidence for small, but significant stability of infant frontal EEG asymmetry, albeit only from 6 to 12 months of age; there were no significant prospective associations between infant frontal EEG asymmetry and NA.

Table 3. Concurrent and prospective associations among infant variables

^a Negative affectivity was measured by the Infant Behavior Questionnaire—Revised.

**p < .01.

Hypothesis testing

The results of MLM analyses for NA predicting frontal EEG asymmetry indicated some support for our hypothesis that with higher levels of prenatal depression, infant frontal EEG asymmetry and NA would be less in sync with each other: that is, a positive association between NA and frontal EEG asymmetry. These results yielded interesting findings regarding how the pattern of association between NA and EEG might change between early in infancy to later in the course of infant development (see Table 4). Specifically, the three-way interaction between NA, age, and prenatal depression grand mean centered in predicting frontal EEG asymmetry yielded a standardized estimate of 0.004 (SE = 0.002, p = .06). Although not statistically significant, it was associated with a marginal R ² = .080. This pattern of results did not change when adding postnatal depression grand mean centered into the model as a covariate. Given the effect size, we probed the three-way interaction with planned comparisons, using a median split to create a categorical two-level variable on mothers' prenatal depression symptom level. The two-way interaction (NA × Age) predicting infant frontal EEG asymmetry was tested separately for infants of mothers who were above or below the median split in terms of prenatal depressive symptoms.

Table 4. Multilevel models of infant negative affectivity (NA), age, and maternal prenatal depression predicting infant frontal EEG asymmetry

Note: NA was measured by the Infant Behavior Questionnaire—Revised.

^a Depression symptoms were measured with the Beck Depression Inventory and calculated as area under the curve scores for the prenatal period (unless otherwise specified), then the state parameter was grand mean centered.

†p < .10. *p < .05.

For infants of mothers who were high in prenatal depressive symptoms (above the median split), there was a significant two-way interaction of infant NA and age predicting infant frontal EEG asymmetry scores (β = 0.06, SE = 0.02, p = .01). Specifically, at infant age 3 months, there was a negative association between infant NA and frontal EEG asymmetry, such that higher NA was associated with greater relative right frontal EEG asymmetry (more negative EEG asymmetry scores; see Figure 1). By contrast, a positive association was present at infant age 12 months, with higher NA associated with lower relative right frontal EEG asymmetry (more positive EEG asymmetry scores). That is, with high levels of prenatal depression, infants shifted from having higher NA associated with greater right frontal asymmetry at 3 months of age to having higher NA associated with greater left frontal asymmetry at 12 months of age.

Figure 1. (Color online) The interaction of infant negative affectivity and age predicting infant EEG asymmetry scores among infants of mothers with high prenatal depression symptoms (above median split).

For infants of mothers who were below the median split in terms of prenatal depressive symptoms, there was no significant interaction between infant NA and age in predicting infant frontal EEG asymmetry (β = –0.02, SE = 0.02, p = .38). Therefore, a follow-up model was run in which the interaction term was removed, in order to investigate the main effects of age and NA on frontal EEG asymmetry in the context of low maternal prenatal depressive symptoms. Results from this model showed no support for NA as a main effect in the prediction of frontal EEG asymmetry, but a significant main effect of age (β = 0.004, SE = 0.002, p = .04) was present; for infants of mothers with low levels of depressive symptoms during the prenatal period, frontal EEG asymmetry scores increase with age, toward greater relative left frontal asymmetry.

By contrast, in the model of frontal EEG asymmetry predicting NA, results did not provide evidence to support the hypothesis (see Table 5). Specifically, the interaction between age, frontal EEG asymmetry, and prenatal depression grand mean centered was not significant (β = –0.001, SE = 0.002, p = .57). When removing the nonsignificant three-way interaction term from the model, none of the two-way interactions were significant. In a final model removing the nonsignificant two-way interaction terms, there were two significant main effects. First, there was a significant main effect of age (β = 0.01, SE = 0.001, p < .001); for the sample as a whole, regardless of maternal depressive symptoms, with increasing age, NA increased, as had been shown descriptively in Table 2. Second, there was a significant main effect of prenatal depression (β = 0.004, SE = 0.001, p < .001); for the sample as a whole, and regardless of age, higher levels of prenatal depression were associated with higher infant NA. This pattern of results did not change when adding postnatal depression grand mean centered into the model as a covariate.

Table 5. Multilevel models of infant frontal EEG asymmetry, age, and maternal prenatal depression predicting infant negative affectivity

Note: Negative affectivity was measured by the Infant Behavior Questionnaire—Revised; AUC, area under the curve.

^a Depression symptoms were measured with the Beck Depression Inventory and calculated as AUC scores for the prenatal period (unless otherwise specified), then the state parameter was grand mean centered.

**p < .01.

Dose of exposure to mother's depression

In order to determine whether this pattern of findings changed based on infants' exposure to differing doses of maternal depression, MLM analyses were rerun using a categorical variable in which women were coded as having exceeded the BDI cutoff for clinically significant levels of depression (≥10) during either the prenatal or the postpartum period, during both the prenatal and the postpartum period, or during neither the prenatal nor the postpartum period. Within the sample, 49 women (20%) exceeded the cutoff during either the prenatal or the postpartum period, 118 (49%) exceeded the cutoff during both the prenatal and the postpartum period, 51 (21%) did not exceed the BDI cutoff in either the prenatal or the postpartum period, and information was missing for the remaining 24 participants (10%). Consistent with the findings from the primary analyses, results of MLM analyses for NA predicting frontal EEG asymmetry indicated some support that differing doses of depression were associated with changes in co-occurrence among the vulnerabilities over time. Specifically, the three-way interaction between NA, age, and the categorical depression variable in predicting frontal EEG asymmetry yielded a standardized estimate of 0.03 (SE = 0.02, p = .09).

We then probed the three-way interaction running post hoc comparisons between all three groups (three comparisons). When comparing the group that experienced elevated depression symptom levels either pre- or postnatally to the group that experienced no perinatal depression, the three-way interaction was no longer significant, with a standardized estimate of –0.07 (SE = 0.05, p = .15). That is, there were no significant differences between infants of women with clinically significant depression in either the prenatal or the postpartum period and infants of women with no clinically significant levels of depression in the prenatal or the postpartum period in terms of the co-occurrence of infant EEG and NA over time.

Similarly, the three-way interaction when comparing the group that experienced no perinatal depression to the group that experienced elevated depression symptom levels both during the pre- and the postnatal periods was not significant, with a standardized estimate of 0.03 (SE = 0.02, p = .13). That is, once again, there were no significant differences between infants of women without depression in either the prenatal or the postpartum period and infants of women with depression in both the prenatal and the postpartum periods in terms of the co-occurrence among infant EEG and NA over time.

When comparing the group that experienced elevated depression symptom levels either pre- or postnatally to the group that experienced elevated depression symptom levels both during the pre- and the postnatal periods, the three-way interaction was significant, with a standardized estimate of 0.12 (SE = 0.04, p = .004). Therefore, the two-way interaction (NA × Age) predicting infant frontal EEG asymmetry was tested separately for infants of mothers who had elevated depression symptom levels in either the prenatal or the postpartum period, and for infants of mothers who had elevated depression symptom levels in both the prenatal and the postpartum periods (see Figure 2).

Figure 2. (Color online) The interaction of infant negative affectivity and age predicting infant EEG asymmetry scores among infants of mothers with clinically significant depressive symptom levels in both the prenatal and the postpartum periods (both) or significant depressive symptoms in either the prenatal or the postpartum period (either).

For infants of mothers who experienced elevated depression symptom levels either pre- or postnatally, the two-way interaction of infant NA and age predicting infant frontal EEG asymmetry scores was of marginal significance (β = –0.07, SE = 0.04, p = .06). Specifically, at infant age 3 months, there was a positive association between infant NA and frontal EEG asymmetry, such that higher infant NA was associated with lower relative right frontal EEG asymmetry (more positive EEG asymmetry scores). By contrast, this association was negative at infant age 12 months, such that higher infant NA was associated with greater relative right frontal EEG asymmetry (more negative EEG asymmetry scores). That is, with depression exposure in either the prenatal or the postpartum period, infants shifted from having higher NA associated with lower relative right frontal asymmetry at 3 months of age to having higher NA associated with greater relative right frontal asymmetry at 12 months of age.

By contrast, this pattern was reversed for infants of mothers who experienced elevated depression symptom levels both during the pre- and the postnatal periods, with the two way-interaction of infant NA and age predicting infant frontal EEG asymmetry scores being significant (β = 0.05, SE = 0.02, p = .03). Specifically, at infant age 3 months, there was a negative association between infant NA and frontal EEG asymmetry, such that higher NA was associated with greater relative right frontal EEG asymmetry (more negative EEG asymmetry scores). By contrast, a positive association was present at infant age 12 months, with higher NA associated with lower relative right frontal EEG asymmetry (more positive EEG asymmetry scores). That is, with depression exposure in both the prenatal and the postpartum period, infants shifted from having higher NA associated with greater relative right frontal asymmetry at 3 months of age to having higher NA associated with lower relative right asymmetry at 12 months of age.

Timing of exposure to mother's depression

In order to also determine whether the pattern of findings from the primary hypotheses changed based on differing timing of maternal depression, MLM analyses were rerun using a categorical variable in which participants were coded as having exceeded the BDI cutoff for clinically significant depression levels (≥10) during the prenatal period only, during the postpartum period only, or during both the prenatal and the postpartum period. Within the sample, 20 women (8%) exceeded the BDI cutoff in the prenatal period, 29 (12%) exceeded the cutoff in the postpartum period, and 118 (49%) exceeded the cutoff in both the prenatal and the postpartum period. Similar to the primary analyses, results of MLM analyses for NA predicting frontal EEG asymmetry indicated support for differing timing of depression being associated with changes in co-occurrence among the vulnerabilities over time. Specifically, the three-way interaction between NA, age, and the categorical depression timing variable in predicting frontal EEG asymmetry yielded a standardized estimate of 0.06 (SE = 0.03, p = .03).

We then probed the three-way interaction using post hoc comparisons between all three timing groups. When comparing the group that experienced prenatal depression only to the group that experienced elevated depression symptom levels both pre- and postnatally, the three-way interaction was no longer significant, with a standardized estimate of 0.04 (SE = 0.03, p = .19). That is, there were no significant differences between infants of women with clinically significant depression only in the prenatal period and infants of women with clinically significant levels of depression in both the prenatal and the postpartum periods in terms of the co-occurrence of infant EEG and NA over time.

Similarly, the three-way interaction when comparing the group that experienced prenatal depression only to the group that experienced postnatal depression only was not significant, with a standardized estimate of –0.08 (SE = 0.08, p = .33). That is, once again, there were no significant differences between infants of women with depression in the prenatal period only and infants of women with depression in the postpartum period only in terms of the co-occurrence of infant EEG and NA over time.

However, when comparing the group that experienced postnatal depression only to the group that experienced elevated depression symptom levels both pre- and postnatally, the three-way interaction was significant, with a standardized estimate of 0.16 (SE = 0.05, p = .002). Therefore, the two-way interaction (NA × Age) predicting infant frontal EEG asymmetry was tested separately for infants of mothers who had elevated depression symptom levels in only the postpartum period and for infants of mothers who had elevated depression symptom levels in both the prenatal and the postpartum periods (see Figure 3).

Figure 3. (Color online) The interaction of infant negative affectivity and age predicting infant EEG asymmetry scores among infants of mothers with clinically significant depressive symptom levels in both the prenatal and the postpartum periods (both) or significant depressive symptoms in the postpartum period only (postpartum only).

For infants of mothers who experienced elevated depression symptom in only the postnatal period, the two-way interaction of infant NA and age predicting infant frontal EEG asymmetry scores was significant (β = –0.11, SE = 0.05, p = .03). Specifically, at infant age 3 months, there was a positive association between infant NA and frontal EEG asymmetry, such that higher infant NA was associated with lower relative right frontal EEG asymmetry (more positive EEG asymmetry scores). By contrast, this association was negative at infant age 12 months, such that higher infant NA was associated with greater relative right frontal EEG asymmetry (more negative EEG asymmetry scores). That is, with depression exposure in the postpartum period only, infants shifted from having higher NA associated with lower relative right frontal asymmetry at 3 months of age to having higher NA associated with greater relative right frontal asymmetry at 12 months of age.

By contrast, this pattern reversed for infants of mothers who experienced elevated depression symptom levels during both the pre- and the postnatal periods, with the two way-interaction of infant NA and age predicting infant frontal EEG asymmetry scores being significant (β = 0.05, SE = 0.02, p = .03). Specifically, at infant age 3 months, there was a negative association between infant NA and frontal EEG asymmetry, such that higher NA was associated with greater relative right frontal EEG asymmetry (more negative EEG asymmetry scores). By contrast, a positive association was present at infant age 12 months, with higher NA associated with lower relative right frontal EEG asymmetry (more positive EEG asymmetry scores). That is, with depression exposure in both the prenatal and the postpartum periods, infants shifted from having higher NA associated with greater relative right frontal asymmetry at 3 months of age to having higher NA associated with lower relative right asymmetry at 12 months of age.