Relations among maternal withdrawal in infancy, borderline features, suicidality/self-injury, and adult hippocampal volume: A 30-year longitudinal study

Abstract

The hippocampus plays an important role in stress regulation and has been the focus of research regarding the effects of early life stress on brain development. Much of this research has focused on severe forms of early adversity, particularly maltreatment. However, a handful of studies are now examining the effects of more subtle variations in quality of early caregiving on hippocampal development. In addition, both early caregiving and hippocampal volumes have been linked to psychopathology, particularly borderline personality disorder (BPD) and its associated features, such as suicidality. In the context of a 30-year longitudinal study, we assessed associations between maternal withdrawal in infancy, hippocampal volume, and BPD features in adulthood. Hippocampal volume was assessed among 18 adults (29.33 ± 0.49 years) assessed for caregiving quality at 18 months (M=18.55 months, SD=1.21 months) and followed longitudinally to age 29. Left hippocampal volume in adulthood was associated with maternal withdrawal in infancy, but not by other components of disrupted parenting. Other risk factors, including maternal psychosocial risk and severity of maltreatment in childhood, were not significantly related to left hippocampal volume. Left hippocampal volume was further associated with increased BPD features and suicidality/self-injury. In addition, left hippocampal volume partially mediated the association between early maternal withdrawal and later suicidality/self-injury. Results point to the importance of quality of early care for hippocampal development and suggest that the first two years of life may be an early sensitive period during which intervention could have important consequences for long-term psychological functioning into adulthood.

1. Introduction

Early life stress is linked to a range of severe physical and mental health difficulties [1,2]. Early life stress, including disruptions in early caregiving and maltreatment, has been particularly strongly associated with the development of borderline personality disorder (BPD) features [3,4]. In addition, converging medical and social science research documents the role of early life stress in altering multiple neurobiological and physiological systems, which impact health across the lifespan. In particular, early life stress may result in increased risk for psychopathology through epigenetic modifications to genes involved in glucocorticoid regulation, which can alter key neurodevelopmental processes such as neurogenesis, synaptogenesis and myelination [5].

Given the high density of glucocorticoid receptors in the hippocampus, this region plays an important role in stress regulation and is strongly impacted by environmental stress [6,7]. As such, the hippocampus has been the focus of much work regarding the effects of early life stress on both BPD features [8–13] and brain development [e.g., 14, 15].

Early life stress is a complex construct which encompasses several levels of contextual stress, such as childhood abuse and neglect [16], witnessing domestic violence [17], and problematic parenting characteristics, including low parental warmth, insensitivity, and parental mental health difficulties [18,19]. A large body of work has shown that early life stress, in the form of both early disruptions in caregiving quality and later maltreatment, is related to an increased incidence of BPD features and suicidality/self-injury in adulthood [20–22]. For example, using retrospective interview measures, Bandelow et al. (2005) found associations between childhood traumatic events and increased BPD features [20]. Other work has found similar associations between maltreatment or family instability and increased BPD features [23–25]. In addition, Lyons-Ruth et al. (2013) prospectively assessed early maternal withdrawal, as well as other aspects of maternal interaction, at 18-months infant age, and found that early withdrawal, in particular, was associated with increased BPD features and suicidality/self-injury in young adulthood [22]. This association was independent of the additive effect of severity of maltreatment during childhood.

A number of studies have also related alterations in structure and function of the hippocampus to adult BPD and to suicidality/self-injury, as well as to other adult psychiatric disturbances, including PTSD and depression [8]. However, studies examining associations between BPD and hippocampal volume have yielded variable results, with some studies finding that individuals with BPD have smaller hippocampal volumes [11–13] and others finding no significant relation between BPD and hippocampal volume [9,10]. In addition, some of the studies that did not find significant associations between BPD and hippocampal volume did find trends indicating that individuals with BPD and PTSD had slightly larger hippocampi than controls [9,10].

While suicidality/self-injury is one of nine possible criteria for a diagnosis of BPD, because of the complex criteria for a BPD diagnosis, individuals with BPD may or may not evince recurrent suicidality or self-injury [26]. In addition, suicidality or self-injury often occurs in the absence of a diagnosis of BPD [27]. Two studies have found that psychiatric patients with depression who previously attempted suicide had smaller hippocampal volumes compared to depressed patients who did not attempt suicide [28,29]. In contrast, a third study found no differences in hippocampal volume when comparing a group of suicidal patients to non-suicidal patients with depression and to healthy controls [30]. Finally, Altshuler et al. found enlarged parahippocampal areas (areas immediately surrounding the hippocampus) in the brains of individuals who died by suicide compared to healthy controls [31]. Interestingly, McGowan et al. found epigenetic modifications in the hippocampi of suicide victims who had also experienced childhood abuse, in the form of decreased levels of glucocorticoid receptor mRNA, which, in turn, are related to increased HPA activity [32]. Thus, given the serious public health consequences of suicidality/self-injury, in the current work both overall extent of BPD features and the specific endorsement of suicidality/self-injury were assessed in relation to hippocampal volumes.

Studies have also examined the direct associations between a variety of early life experiences and hippocampal volume in children and adults. Much of this research has focused on severe forms of early adversity, particularly maltreatment [for reviews see [33,34]]. Within maltreatment research, the direction of effect varies depending on the age of the sample and the type of maltreatment. Tupler and De Bellis found enlarged hippocampal white matter volumes among maltreated children and adolescents [25]. Similarly, Whittle et al. found that childhood maltreatment was associated with larger baseline hippocampal volumes in early adolescence, but that prior childhood maltreatment then predicted reduced growth of the hippocampus from early to late adolescence [36]. In contrast, some studies of early life stressors, including trauma, maltreatment, maternal depression, and institutional rearing [37–40], have failed to find significant associations between these stressors and hippocampal volume in childhood and adolescence. Other studies have reported reduced volumes as a result of stress [41,42]. Finally, in adult samples, childhood abuse and neglect has been associated rather consistently with reduced hippocampal volumes [8,34].

In addition to the effects of severe deprivation and maltreatment, a handful of studies have demonstrated the effects of less dramatic variations in quality of caregiving on brain development, and hippocampal development in particular, from infancy to adolescence. In a sample of 6-month-old infants, lower levels of maternal sensitivity were concurrently associated with larger bilateral hippocampal volumes [43]. Similarly, less maternal positivity observed at home at one year infant age (i.e., maternal positive affect and attitude in interaction with the infant) was associated with larger bilateral hippocampal volume at child age 10 years [44]. Finally, Rao and colleagues showed that less maternal nurturance at age 4 years (but not 8 years) was associated with larger left hippocampal volume at age 14 [45]. However, Lyons-Ruth et al. found that overall attachment disturbance at 18 months of age, which included infant attachment disorganization and overall maternal disrupted interaction, was related to increased left amygdala volume but was not related to hippocampal volumes [46].

In contrast, Kok et al. found that higher parental sensitivity between one and 4 years of age was related to larger total gray matter volume but was not related to regional brain volumes [47]. In addition, higher levels of maternal support in early childhood (3–5 years) were associated with larger bilateral hippocampal volume at school age (7–13 years old), but only among children who were not depressed [48]. Taken together these studies suggest that early parenting behaviour is associated with hippocampal volume in childhood, but that other, as yet unidentified, intrinsic or contextual factors may influence the direction of effects.

Developmental timing is one factor that likely contributes to the varying effects of early life stress. The hippocampus undergoes particularly rapid development during the first two years of life, finally reaching peak volume at approximately 9–11 years of age [49,50]. Importantly, hippocampal volume has been found to be relatively stable in early adulthood. For example, in a sample of 20- to 85-year-olds, Sullivan et al. did not find significant correlations between hippocampal volumes and age, despite age-related decline in temporal volumes [51]. Other work has found volumetric changes in hippocampal volume, but not until after age 30 [52], age 40 [53], or age 60 [54]. This suggests that hippocampal volume is likely to be relatively stable between ages 20 and 30 (the assessment time points for psychopathology and brain volume assessed here). Given the early growth of the hippocampus, as well as its important role in hypothalamic-pituitary-adrenal (HPA) axis activity, the hippocampus might be differentially susceptible to environmental stress during different developmental windows. Thus, the developmental timing of stress is likely an important factor affecting hippocampal growth, given the long trajectory of hippocampal development [34].

Consistent with this hypothesis, Andersen et al. assessed the relation between self-reported exposure to childhood sexual abuse and hippocampal volume in young women [15]. They found that bilateral hippocampal volume was most strongly negatively associated with sexual abuse between 3–5 years of age and was also negatively associated with sexual abuse between 11–13 years of age, but not during other developmental stages. A second study by Teicher et al. found that decreased male hippocampal volume in adulthood was associated with neglect, but not abuse, up to age 7 [14]. In contrast, decreased female hippocampal volume in adulthood was associated with abuse, but not neglect, at older ages. Further, the aforementioned studies that reported an association between negative parenting behavior and larger hippocampal volume has most often assessed parenting behavior in infancy and early childhood [43,44,47]. This suggests that early parenting might have distinct effects on hippocampal development compared to later maltreatment. Given these developmental trajectories, there is a critical need for longitudinal studies to prospectively examine how early parenting quality is related to later brain development.

In addition, the types of measures used to assess early life stress may contribute to the differential findings regarding the effects of stress on the hippocampus. The majority of studies of maltreatment have been conducted using retrospective measures of abuse or neglect. Although advanced self-report measures, such as the Maltreatment and Abuse Chronology of Exposure Scale (MACE; [55]), are able to capture the timing of adverse experiences during childhood, direct observational measures of the quality of the caregiving environment are particularly important for evaluating the quality of care received in infancy, which is difficult to assess by self-report. As noted above, a handful of studies have used observational measures of the quality of parent-child interaction, particularly in infancy and early childhood, to prospectively assess the quality of early care [43–48]. However, no study to date has prospectively observed parenting quality in infancy in relation to later hippocampal development in adulthood.

Many individuals with BPD or suicidality/self-injury have experienced a number of forms of early life stress, including abuse, neglect, parent-child role reversal, and parental abandonment [12,13,24,56]. Thus, the type of life stress experienced may also be important to differentiating effects on hippocampal development. As reviewed above, studies of childhood maltreatment generally show an association with decreased adult hippocampal volume, while studies of maladaptive parenting in infancy or early childhood more frequently show an association with increased hippocampal volume in childhood. Thus, careful differentiation of the types and timing of stressors to which the child must adapt will be important in future studies of hippocampal volume.

Finally, research has yet to examine models of mediating mechanisms that may best describe the relations among early life stressors, hippocampal volume, and BPD or suicidality/self-injury. For example, individuals who experience early disruptions in caregiving may develop atypical hippocampal volumes, which, in turn, may increase the likelihood of developing the emotional dysregulation indexed by BPD features and suicidality/self-injury in adulthood. Alternatively, individuals who experience early disruptions in caregiving may develop the emotional dysregulation indexed by increased BPD features and suicidality/self-injury, which, in turn, may increase the likelihood of atypical hippocampal development.

The current study drew its participants from a prospective longitudinal study that extended from the first year of life until 30 years of age. This prospective longitudinal study explored developmental trajectories toward maladaptation in a low-income cohort stratified for exposure to poor care in infancy. Previous work with this longitudinal cohort indicated that maternal withdrawal in infancy was associated both with disorganized attachment behavior in infancy [57] and with increased BPD features and presence of suicidality/self-injury at age 20 [22]. Further, the link from maternal withdrawal to adult outcomes was independent of variance accounted for by income, gender, and severity of childhood maltreatment.

In addition, a prior study using the current MRI subsample [46] assessed the hypothesis derived from rodent studies that disturbances in the early mother-infant relationship would be associated with stress-related increases in amygdala volume that would endure into adulthood [58–60]. Results from this work [46] confirmed that overall attachment disturbance in infancy, indexed by a combination of infant attachment disorganization and overall maternal disrupted interaction, was associated with left amygdala volume in adulthood, but not right amygdala volume. This index of overall attachment disturbance was not significantly related to hippocampal, caudate, or thalamic volumes in adulthood.

However, the previous longitudinal work on BPD features, cited above, found a specific role for early maternal withdrawal in pathways to BPD and suicidality. Given the complementary body of studies relating BPD and suicidality to hippocampal volumes, the present report assessed whether hippocampal volume might play a role in the pathways from early maternal withdrawal to later BPD and suicidality. In the current study, we first aimed to confirm in this subsample that maternal withdrawal in infancy continued to relate significantly to later BPD features and suicidality. Second, we evaluated whether early maternal withdrawal, in particular, would be prospectively associated with hippocampal volume in young adulthood. However, given the variability of findings in the literature, the direction of the association to hippocampal volume in adulthood was unclear. Third, we predicted that alterations in hippocampal volume would also be associated with BPD features and suicidality/self-injury. Our fourth aim was to evaluate whether hippocampal volume was better viewed as a mediator or as an outcome of the relation between early maternal withdrawal and later BPD features and suicidality/self-injury.

2. Materials and methods

2.1. Participants

Participants were 18 young adults (M=29.33, SD=0.49 years) first recruited as infants (M=8.5, SD=5.6 months) as part of a longitudinal study of the effects of social risk factors on child development [61]. This larger longitudinal study consisted of 76 families who were at or below federal poverty levels. Specifically, half of the families in the cohort were referred to the study by clinical service providers because of concerns about the quality of care for the infant. The remaining families were low-income controls from the same communities, matched to the high-risk families on socioeconomic status (i.e., education and income) (see [61]). The relations between infancy risk factors and maladaptive developmental outcomes from infancy to adulthood have been demonstrated in multiple reports from this larger sample (e.g. [61–64],).

At age 29, 33 participants were relocated and screened for inclusion in the current study. Exclusion criteria included MRI safety criteria, no reported substance abuse in the past six months, and no significant medical or neurological conditions. Eighteen met inclusion criteria and participated in the current MRI study. All participants except one were right-handed. Seventy-two percent of participants in the MRI study had been referred for parent-infant clinical services during the first 18 months of life, suggesting high levels of early social and environmental risk.

Participants in the MRI study were representative of the larger longitudinal cohort. They did not differ from the remainder of the cohort in demographic characteristics: family income (η = 0.21, p = 0.13), male gender (φ = −0.136, p = 0.23), mother single parent (φ = −0.02, p = 0.87), mother high school education only (φ = −0.03, p = 0.85), ethnicity (φ = 0.05, p = .65), severity of childhood maltreatment (η = 0.10, p = 0.46); or extent of Axis I or Axis II psychopathology on the SCID in adulthood (η = 0.05–0.03, p = 0.75 −0.85). Participants in the current study were marginally more likely than those in the larger longitudinal study to have been referred for clinical services in infancy (φ = 0.22, p = 0.09).

2.2. Measures

2.2.1. Disrupted mother–infant communication (18 months)

The Atypical Maternal Behavior Instrument for Assessment and Classification (AMBIANCE; [57]) was used to code disrupted maternal communication from videotapes of the 25-minute Strange Situation Procedure. Frequencies of five dimensions of disrupted maternal communication were coded: 1. Affective communication errors, defined as contradictory affective signals to the infant (e.g., using a sweet voice with a derogatory message) or inadequate or inappropriate responses to the infant’s signals (e.g., failing to comfort a distressed infant); 2. Roleconfusion, defined by the mother calling the infant’s attention to herself in ways that override or ignore the infant’s cues (e.g., asking the infant for a kiss when the infant is distressed); 3. Frightened/disoriented behavior, defined by fearful, hesitant, or deferential behavior towards the infant (e.g., hesitating before responding to infant; tense body postures) or by disoriented behavior (e.g., flat or odd affect in interaction; frenetic or uncoordinated overtures toward the infant); 4. Negative-intrusive behavior, defined as harsh or critical behavior (e.g., pulling the infant by the wrist; mocking or teasing the infant; attributing negative affect to the infant); 5. Withdrawing behavior, defined by creating physical or emotional distance from the infant (e.g., standing across the room while interacting; interacting silently). Reliability between two naive coders (n=15) for the five dimensions ranged from r_i=0.73–0.84. Validity and stability of the AMBIANCE coding system have been demonstrated by meta-analysis [65].

2.2.2. Family income and maternal psychosocial risk in infancy

Sociodemographic and psychosocial risk factors in infancy were coded from maternal interviews during infancy to provide measures of family income and maternal psychosocial risk. Psychosocial risk was coded positive if any of the following three risk factors were present: 1) mother had a history of psychiatric hospitalization, 2) mother had a history of state protective service involvement for maltreating a child, and 3) mother reported clinically significant depressive symptoms on the CES-Depression scale, using the established cut point of ≥ 16 [66].

2.2.3. Severity of childhood maltreatment

Severity of exposure to childhood maltreatment was assessed in adulthood using the Maltreatment and Abuse Chronology of Exposure Scale (MACE; [55]). Participants reported on exposure to ten types of maltreatment during each year of childhood from ages 6–18, including childhood sexual abuse, parental verbal abuse, parental non-verbal emotional abuse, parental physical abuse, witnessing of intra-parental violence or violence toward siblings, peer verbal abuse, peer physical abuse, parental emotional neglect, and parental physical neglect. The MACE has excellent reliability [55] and demonstrated validity in relation to a range of adult neurobiological outcomes [15,67].

2.2.4. Borderline personality disorder features and suicidality/self-injury

Number of borderline personality disorder features and presence of recurrent suicidality/self-injury were assessed during adolescence using the Structured Clinical Interview for DSM-IV-R (SCID) for Axis II, administered by trained clinicians [68]. Both SCID features and diagnoses have shown adequate reliability [69,70].

2.2.5. Imaging data acquisition and processing

Data were collected using a 3 T TIM Trio scanner (Siemens AG, Erlangen, Germany) with a 32-channel head coil using a T1-weighted Magnetization Prepared Rapid Gradient Echo (MPRAGE) pulse sequence (TE = 2.25 ms; TR = 2100 ms; FA = 12; FOV = 256 mm; slice number = 128; voxel size = 1.0 × 1.0 × 1.3 mm; slice thickness = 1.33 mm) in the sagittal plane (scan duration: 6 min). Gray matter volume (GMV) in the hippocampus, caudate and thalamus and total grey matter volume (TGMV) were assessed using FreeSurfer 6 (http://surfer.nmr.mgh.harvard.edu) [71]. This program includes robust tools for subcortical volume analyses [72,73]. Voxels within subcortical regions were labeled using an elaborate process based on both a subject-independent probabilistic atlas derived from a hand-labeled training set and on subject-specific measures [71]. This procedure labels the brain in a manner that is statistically indistinguishable from those provided by experienced manual raters [71]. Automated segmentation eliminates differences between studies and facilitates replication by other investigators [67]. The authors (CMA, PP) visually inspected all T1-weighted and automated images. Manual adjustments were not required.

2.3. Statistical analysis

Hypothesis testing of relations between neuroimaging measures and aspects of developmental history and adult psychopathology was conducted using partial correlations controlling for relevant covariates (SPSS 24). Our primary hypotheses were related to early maternal withdrawal, psychopathology, and hippocampal volume. As control analyses, we also analyzed associations between disrupted maternal interaction and control brain regions not thought to be as central as the hippocampus in regulating the HPA axis stress response, including the thalamus and caudate. Mediation analyses were conducted in Mplus (Version 8). All analyses controlled for significant covariates. FIML and bootstrapping using 2000 samples was estimated in Mplus for all mediation models to account for missing data, non-normality, and small sample size [75]. As recommended by Preacher and Hayes (2008), bias-corrected confidence intervals (CIs) were used to determine significance. CIs that do not contain zero are significant at p < .05 or greater [76].

3. Results

3.1. Descriptive data and covariates

Descriptive data for sample characteristics, disrupted maternal interaction scores, extent of BPD features, incidence of suicidality/self-injury, and regional brain volumes are displayed in Table 1. Of the five interaction scores, role confusion had a significant outlier and was log-transformed to minimize skew. All other study variables (maternal interaction, brain regions, and psychopathology) did not have significant outliers and had acceptable skew and kurtosis [77].¹ Maternal behavioral coding was missing from two participants (n = 2, 11%). There was no missing data for demographic, brain volume, or psychopathology variables. Based on Little’s Missing Completely at Random (MCAR) test, the missing maternal interaction data were deemed to be missing completely at random (χ²(6) = 6.876, p = 0.332), and thus are appropriate for the use of FIML. Demographic variables, as well as TGMV, were explored as potential covariates associated with independent variables (i.e., maternal interaction, psychopathology variables) and dependent variables (i.e., hippocampal, caudate and thalamic volumes). TGMV (right r = .75, p < .001; left r = .71, p < .001) and male gender (right r = .76, p < .001; left r = .76, p < .001) were significantly correlated with hippocampal volumes. Race (white = 0, black = 1; r = .63, p < .01) was associated with maternal withdrawal. There were no significant covariates related to suicidality/self-injury. Thus, all partial correlations below are adjusted for gender, and race and, when relevant, TGMV.

Table 1. Sample Characteristics.

	Mean ± SD or N (%)
Adult characteristics
Age (years)	29.3 ± 0.5
Female	10 (57%)
White	14 (78%)
Single	11 (61%)
College degree	2 (11%)
Psychosocial risk factors in infancy
No risk	6 (33%)
Maternal depression only	7 (39%)
Depression and/or psychiatric hospitalization,	5 (28%)
maltreatment petition
Disrupted maternal interaction
Role confusion	5.1 ± 7.1
Affective communication errors	5.0 ± 4.5
Negative-intrusive behavior	3.8 ± 4.1
Fearful-disoriented behavior	3.0 ± 2.7
Withdrawal	3.5 ± 3.0
Adult symptomatology
Borderline personality disorder features	1.4 ± 1.9
Suicidal or self-injurious behavior	5 (28%)
Brain region volumes
Total grey matter volume	676573.8 ± 85853.8
Hippocampus Left	3481.7 ± 389.3
Hippocampus Right	3504.8 ± 356.7
Caudate Left	3653.9 ± 511.7
Caudate Right	3595.7 ± 560.6
Thalamus Left	6466.3 ± 920.6
Thalamus Right	7026.5 ± 943.4

3.2. Early maternal withdrawal in relation to BPD features and Suicidality/Self-injury in adulthood

In previous work with the larger cohort from which these participants were drawn [22], maternal withdrawal in infancy was significantly associated with BPD features and suicidality/self-injury in adulthood. In the present subsample, maternal withdrawal continued to be significantly related to BPD features (r_p = .68, p = .007) and to suicidality/self-injury (r_p = .88, p < .001).

3.3. Maternal withdrawal in infancy in relation to hippocampal volumes

In accord with hypotheses, maternal withdrawal was also strongly correlated with left hippocampal volume in adulthood (r_p = .64, df = 11, p < .05), such that greater withdrawal was related to en larged left hippocampal volume. The association between maternal withdrawal and adjusted right hippocampal volume was not significant (r_p =.33, p = .26). In addition, the stronger left than right hippocampal association was reflected in a significant hippocampal laterality index (r_p = .62, p < .03). The associations between maternal withdrawal and hippocampal volume are depicted in Fig. 1. None of the other dimensions of disrupted maternal communication were significantly correlated with hippocampal volumes (Table 2). Specificity of the effect of early maternal withdrawal on the hippocampus compared to the caudate and thalamus was further demonstrated, in that offspring caudate or thalamic volumes were not significantly related to early maternal withdrawal (caudate: left r_p = .14, p = .66; right r_p = .29, p = .34; thalamic: left r_p = .31, p = .29; right r_p = .40, p = .18), nor to any other aspects of early maternal disrupted communication (Table 2).

Fig. 1.

Left and right hippocampal volume in adulthood in relation to maternal withdrawal behavior in infancy. Associations are adjusted for total grey matter volume, gender, and race. Left (rp=.64, p < .05) but not right (rp = .33, p = .26) hippocampal volume was related to maternal withdrawal behavior.

Table 2. Partial Correlations.

	1	2	3	4	5	6	7	8	9	10	11	12	13
1. Maternal withdrawal	—
2. Maternal role reversal	-.33	—
3. Maternal communication errors	.56*	.18	—
4. Maternal negative-intrusive behavior	.07	.56*	.73	—
5. Maternal disorientation	-.13	.72**	.28	.62*	—
6. Left hippocampus	.64*	-.45	.30	-.10	-.40	—
7. Right hippocampus	.34	-.04	.09	-.01	.05	.77**	—
8. Left caudate	.14	-.54ŧ	.02	-.08	-.13	.24	.18	—
9. Right caudate	.29	-.58*	-.11	-.26	-.23	.31	.27	.91**	—
10. Left thalamus	.31	.07	-.06	-.01	.07	.01	.05	-.16	-.02	—
11. Right thalamus	.40	.21	.02	.06	.22	.13	.37	-.08	.06	.81**	—
12. Suicidality/self-injury	.88**	-.38	.52ŧ	.03	-.35	.58*	.30	-.04	.08	-.03	-.01	—
13. BPD features	.68**	-.26	.32	.13	-.21	.55*	.28	-.16	-.10	.48ŧ	.39	.58*	—

Note: BPD = Borderline personality disorder. All correlations control for race, gender, and, when relevant, total gray matter volume. ŧ p < .10, *p < .05, ** p < .01.

3.4. Other childhood risk factors in relation to hippocampal volume

We also examined whether the relation between early maternal withdrawal and offspring left hippocampal volume could be accounted for by other infancy or childhood developmental risk factors. Neither left nor right hippocampal volumes were significantly correlated with family income (left r_p = -.13, p = .66; right r_p =- .15, p = .61) or maternal psychosocial risk (history of psychiatric hospitalization, involvement with state protective services, or high depressive symptoms; left r_p=.04, p= .90; r_p = -.04, p = .89). In addition, childhood maltreatment was not significantly associated with left hippocampal volume. However, there were significant partial correlations between right hippocampal volume and severity of exposure to maltreatment at ages 5 (r_p = .59, p < .02), 6 (r_p = .64. p < .01), 13 (r_p = .62, p < .02) and 14 (r_p = .58, p < .04).

3.5. BPD features and Suicidality/Self-injury in relation to hippocampal volumes

Extent of BPD features and incidence of suicidality/self-injury were also associated with increased left hippocampal volume (BPD features: r_p=.55, p<.05; suicidality/self-injury: r_p=.58, p<.05. Right hippocampal volume was not significantly associated with BPD features or suicidality/self-injury (Table 2). The hippocampal laterality index was also associated with BPD features (r_p=.61, p < .02, controlling for gender and race), indicating a stronger relation of BPD features to left hippocampus than right hippocampus.

3.6. Left hippocampal volume as mediator of the relation between early maternal withdrawal and BPD features and suicidality/self-injury in adulthood

Given the associations between maternal withdrawal and left hippocampal volume, and between left hippocampal volume and BPD features and suicidality/self-injury, as well as the direct relations between maternal withdrawal and both aspects of psychopathology, a further important question was whether left hippocampal volume mediated the relations between early withdrawal and later BPD features or suicidality/self-injury. Conversely, it could also be the case that the emergence of BPD features or suicidality/self-injury was the mediating factor linking early withdrawal to later enlarged left hippocampal volume.

To explore these potential mechanistic models, mediation analyses were conducted. The first mediation model assessed whether left hippocampal volume mediated the relation between maternal withdrawal and later BPD features, controlling for covariates (gender, ethnicity, and total gray volume). Left hippocampal volume was not found to be a significant mediator of the effect of maternal withdrawal on BPD features (indirect effect: β = .11, SE = .15, CI {-.04, .41}).

Regarding suicidality/self-injury, in a similar mediation model, controlling for covariates (gender, ethnicity, and total gray volume) and accounting for other effects in the model (Fig. 2), maternal withdrawal was not significantly related to left hippocampal volume (a path: β = .31, SE = .20, CI {-.10, .58}), but left hippocampal volume was related to suicidality/self-injury (b path: β = .33, SE = .14, CI {.10, .55 }), and maternal withdrawal was associated with suicidality/self-injury (c path: β = .71, SE = .17, CI {.30, .90}) (Fig. 2). There as a significant indirect effect of maternal withdrawal on suicidality/self-injury through left hippocampal volume (indirect effect: β = .10, SE = .09, CI {.01, .30}). In addition, the total effect (the sum of the direct and indirect effects) was also significant (β = .81, SE = .11, CI {.57, .95}; Fig. 2). Thus, this model confirmed that the association between maternal withdrawal and suicidality/self-injury, was accounted for, in part, through enlargement of the left hippocampus.

Fig. 2. Mediation model for effect of maternal withdrawal on suicidality/self-injury through left hippocampal volume.

Note: The values in the figure are coefficients (with bootstrapped standard error) for each path. Path a models the effect of maternal withdrawal on left hippocampal volume. Path b models the effect of left hippocampal volume on overall suicidality/self-injury. Path c’ models the direct effect of maternal withdrawal on suicidality/self-injury. Path c models the total effect of maternal withdrawal on suicidality/self-injury (c’ + ab). Indirect effect: β = .10, SE = .09, p < .05 [95% CI = .01, .30]. *p < 0.05.

3.7. Alternative mediation models

We also assessed alternative mediation models in which the development of BPD features or suicidality/self-injury was the mediating mechanism linking early withdrawal to hippocampal volume, rather than vice versa as tested above. In these models, early withdrawal contributes to the emergence of psychopathology, which then contributes to enlarged hippocampal volume, rather than hippocampal volume contributing to increased psychopathology.

In regard to BPD features, with relevant covariates, the indirect effect of maternal withdrawal being associated with left hippocampal volume through BPD features was not significant (indirect effect: β = .12, SE = .21, CI {-.11, .57}). Thus, BPD features did not mediate relations between early maternal withdrawal and adult left hippocampal volume. In addition, after controlling for relevant covariates, suicidality/self-injury did not mediate the association between maternal withdrawal and left hippocampal volume (indirect effect: β = .12, SE = .32, CI {-.10, .92}). Taken together, the two sets of mediation results indicate that the development of suicidality/self-injury in response to early withdrawal is, in part, mediated by increased hippocampal volume, and not vice versa. That is, in the context of maternal withdrawal, hippocampal volume contributes to increased suicidality/self-injury, but suicidality/self-injury does not contribute to enlarged left hippocampus.

Given the small sample size, we should note that the initial correlations in Table 2 between the five aspects of parenting, the two hemispheres of the hippocampus, and the two aspects of psychopathology (nine variables) do not reach the Bonferroni correction p-value of .0055. However, the Bonferroni is widely noted to be too conservative [78], and all mediation models were run with bootstrapping to 2000 samples and were robust to that criterion.

4. Discussion

A large emerging literature supports a view of the hippocampus as both highly stress sensitive and as modifiable in response to environmental influences [79]. Translational studies of rodents [80–83] and non-human primates [84,85], as well as studies of human children and adults [8] point to this conclusion. The current study prospectively assessed disruptions in caregiver-infant interactions in infancy in relation to hippocampal volume and BPD features, including suicidality/self-injury, in adulthood. There was notable specificity in the findings, such that early maternal withdrawal, but not other aspects of interaction, was significantly related to enlarged left hippocampal volume. Left hippocampal volume, but not right hippocampal volume, was further related to BPD features and suicidality/self-injury in adulthood. Notably also, maternal withdrawal was not significantly associated with altered volumes of brain regions less central to regulation of the HPA axis response, including the caudate and thalamus. Nor could the association between left hippocampal volume and maternal withdrawal be accounted for by low SES in infancy, maternal psychosocial risk in infancy, or later childhood maltreatment. Therefore, these results converge with translational work that points to the potential for early caregiving disruptions to impact hippocampal development [43–45,86].

In contrast to most other developmental studies of hippocampal volume, this cohort was recruited in the first year of life and, for half (52.6%) of the sample, recruitment was based on the concerns of community providers about the quality of care for the infant. Families of concern were socioeconomically matched to other low-income families whose care was not of concern. Thus, the parenting risk indexed by the laboratory withdrawal scores was serious enough to create concerns among community providers who interacted with the families.

Finally, prior work with the larger at-risk cohort from which this sample was drawn had shown that early maternal withdrawal was associated with the serious difficulties in emotional regulation indexed by BPD features and suicidality/self-injury in adulthood [22]. That finding from the larger cohort remained significant in this smaller subsample. Further, results of the two mediation models indicated that left hippocampal volume partially mediated the association between early maternal withdrawal and later adult suicidality/self-injury. Thus, an enlarged left hippocampus appears to play a role in accounting for the association between early maternal withdrawal and the later severe emotional dysregulation indexed by suicidality/self-injury. In contrast, BPD features more generally were best modelled as separately associated with both early maternal withdrawal and with enlarged left hippocampal volume (Table 2). These findings indicate the functional significance of both hippocampal volumes and qualities of early parenting for emotional regulation in adulthood.

These findings also add to the literature documenting altered hippocampal volume in individuals displaying BPD features or suicidality/self-injury [8,11,13,87]. However, it needs to be pointed out that the relation between BPD features and enlarged left hippocampal volume is somewhat at odds with neuroimaging studies in individuals with BPD. As Teicher and Samson (2016) summarized, seven of 16 studies of individuals with BPD reported significant findings either for the right hippocampus only or for both the right and left hippocampus [8]. Further, all identified studies that were based on individuals with BPD and provided separate right/left hippocampal results had significant volumetric differences for the right hippocampus. In contrast, all of the adult hippocampal morphometry studies that found significant findings only in the left hippocampus were based on samples without BPD (either with depression or no psychopathology). Importantly, the current community sample differs from clinical BPD samples in that, as expected, few individuals reached full criteria for a diagnosis of BPD (5.6%, N=1). Individuals who meet full criteria for BPD may have experienced more varied and severe forms of life stress over multiple developmental periods than the community sample assessed here. Given these prior studies showing left/right volumetric differences [8], it is possible that one neurobiological contributor to BPD might lie in the relative size of the left versus right hippocampus, which may include, in some studies, contributions of an enlarged left hippocampus related to early rearing conditions, and, in other studies, contributions of a reduced right hippocampus related to more severe forms of maltreatment at later ages.

Notably, the emerging body of studies on early caregiving quality and hippocampal development has more frequently found larger hippocampal volumes in childhood and adolescence [43–45] associated with less optimal caregiving quality, including lower levels of parental positivity, nurturance, and sensitivity in infancy and early childhood, similar to the results here. It should be noted that sensitivity is a broad construct comprised of different maternal behaviours, including withdrawal, nonsynchronous responses, difficulty soothing the infant, and intrusive behavior [89]. In contrast, some other research has found that decreased hippocampal volume is associated with less optimal parenting quality [48]. In addition, as reviewed earlier, some studies in childhood have failed to find significant differences in hippocampal volume in relation to more general indices of early quality of care [38,39].

The overall picture that emerges is that decreased hippocampal volume has more often been associated with severe forms of childhood maltreatment at later ages [33,34], while increased hippocampal volume has been associated more consistently with less maternal support and sensitivity in early childhood [43–45]. Thus, these literatures suggest that the type and timing of deviation in caregiving (i.e., early caregiving versus later maltreatment) might also contribute to differences in hippocampal development.

In relation to the laterality finding, we have proposed previously that different types of survival threats, as well as their developmental timing, may be important to understanding responses of the right and left amygdala [46]. This hypothesis is also relevant to consideration of potential laterality in regard to the left and right hippocampus. Briefly, the human infant, due to its very long period of dependence on the caregiver for survival, must respond very differently to survival threats associated with caregiver withdrawal and neglect, compared to threats of attack and physical harm. Threats of attack have been underscored in theories of stress, with fight, flight, or freeze being documented as the prepotent defensive responses to such threats across species. However, these responses would be quite maladaptive as responses to the threat posed by caregiver absence or neglect, and caregiver absence or neglect is arguably the most significant threat to survival during the early years of human life. The infant needs to respond to threat of withdrawal/neglect by increasing their calling for the parent and their proximityseeking to the parent, that is by ‘seeking and squeaking’, not by fight, flight or freeze. The current work suggests that the left/right hippocampus, might be involved in differentiating between these two classes of threat and mobilizing the different forms of response needed to meet the threat and enhance survival.

The possibility that withdrawing versus threatening/abusive parental stances over time may have different effects on hippocampal development warrants consideration because such differences are consistent with a much broader literature exploring how the child adapts to particular contingencies in the primary attachment relationship over time. In brief, research on infant attachment [88,90] has documented that infants adapt their attachment behavior to a range of caregiving contingencies by the end of the first year of life. In the face of aversive responses by the caregiver, such as negative-intrusive behavior or discomfort with close contact, infants may decrease their signaling for comfort and contact and turn attention away from the caregiver (hypoactivation of attachment signals relative to normative infant behavior; [88,91–93]). Conversely, in the face of inconsistent responding, infants may increase their signaling to the caregiver (hyperactivation of attachment bids in relation to normative infant behavior; [88,91,93,94]). Subsequent work has further found that more pronounced forms of caregiver withdrawal to the infant’s attachment bids are associated both with odd, out-of-context, conflicted behavior (disorganized behavior), and with concomitant hyperactivation, that is, with the infant’s concerted attempts to approach the caregiver, despite signs of conflict (termed Disorganized-Approach behavior [57,95]). Thus, the current finding of a further relation between early maternal withdrawal and enlarged left hippocampus suggests that the left hippocampus may be involved in the infant’s hyperactivation of attachment bids to more extreme caregiver withdrawal in the service of ensuring caregiver availability. Notably, in the present results, the same participants exhibited both increased left hippocampal volumes associated with early maternal withdrawal and right hippocampal volumes associated with severity of maltreatment later in childhood.

This hypothesis is also consistent with the motivational hypothesis emerging from EEG studies that the left hemisphere generally is more involved in motivating approach behavior and the right hemisphere is more involved in motivating avoidance behavior [96] and extinguishing the fear response [97]. Thus, the right hippocampus may be differentially responsive to the forms of anger and threat of attack that are associated with fight, flight, or freeze [98–100], while the left hippocampus may be differentially responsive to parental emotional unavailability or threats of abandonment, enlarging in relation to the degree of hypervigilance and proximity-seeking needed to monitor and engage the caregiver. Taken together, if a continuing feature of the child’s environment is the need to heighten approach behaviors toward the caregiver in order to obtain a modicum of response, this might have a particular developmental impact on the left hippocampus compared to the right.

One alternative hypothesis to consider is that the left hippocampus might confer greater sensitivity to caregiving conditions more generally, with enlarged left hippocampus enhancing the effects of caregiving variations. This is an important possibility and consistent with the moderation effects found by Schreiber et al. [101] and Whittle et al. [102]. However, the few studies that report a differential relation to the left hippocampus show some specificity in the association of increased left hippocampal volume with caregiver unavailability. For example, Rao et al. found that less maternal nurturance at age 4 predicted greater left hippocampal volume at age 14 [45]. In addition, among those with larger left hippocampus, but not right, lower family connectedness (i.e., feelings of closeness, love, and support) was related to higher depression in adolescence [101]. Other work has shown that youth who have a history of caregiver deprivation and emotional neglect demonstrated greater left hippocampal activation while processing threatening stimuli [103]. Other related research, suggests, in contrast, that smaller left hippocampi might be a predisposing factor for psychopathology. This work has shown that smaller left hippocampus in combination with a variation in the 5-HTTLPR serotonin transporter gene increases risk for the onset of major depression [104], and smaller hippocampus increases risk for developing PTSD after experiencing a trauma [105].

However, the literature relevant to differentiating conceptual models of hippocampal laterality is still quite slim. Evaluation of broader models will require additional studies that parse the qualities of parenting and maltreatment in more detail. Most research has typically assessed global indicators of parenting in relation to hippocampal volume, such as overall severity of childhood maltreatment (e.g., [33]), overall parental sensitivity (e.g. [43],), or overall disrupted communication in childhood [46]. Thus, to better identify the critical aspects of care that influence hippocampal development, it will be important in future studies to include assessments that differentiate degrees of neglect/lack of nurturance from degrees of negative parenting/abuse across infancy and childhood, because these different parenting stances may have very different biobehavioral consequences over the course of development.

Despite a range of suggestive findings, it is important to consider the multiple factors that contribute to structural volume (e.g., size of neurons and glial cells, density, vascularity) [106]. Based on randomized animal models, there are several possible neurobiological mechanisms that could account for the association between early maternal withdrawal and increased left hippocampal volume, including both early stress-related overproduction of dendritic branches and/or spines in the hippocampus and later inadequate pruning. In addition, because hippocampal volume was assessed in adulthood, non-linear developmental change in hippocampal volumes, specifically, high rates of change in infancy and early childhood [49,50], along with stressful circumstances that were not measured here, may have influenced the relation between quality of early care and left hippocampal volume. Although we examined associated risk factors prominent in the literature, including maternal psychosocial stress and depression, low SES, and childhood maltreatment, other factors, at different time points across development, may have been involved in mediating the observed relations.

The present findings must be considered in relation to the findings of Lyons-Ruth et al., which used the same sample [46]. Parallel findings across these studies indicate that both the left amygdala and left hippocampus are affected by early parenting quality, while the right hippocampus and right amygdala are affected by maltreatment later in childhood. However, the parent-child relationship qualities that affected amygdala volume were broader than those that influenced the hippocampus, comprising both overall infant disorganized attachment behavior and overall maternal disrupted communication. In the prior study [46], because of the focus on amygdala rather than hippocampus, only overall attachment disturbance was assessed in relation to hippocampus and not found to be significant, therefore maternal behavior was not further decomposed into the five dimensions and assessed in relation to hippocampal, caudate, or thalamic volumes. The focus on maternal withdrawal specifically in the current paper stemmed from previous findings in the larger longitudinal study on the specific link between maternal withdrawal and later BPD and suicidality/self-injury [22]. Notably, in the current work, maternal behavior showed similar specificity in relation to left hippocampal volume, in that only withdrawal was related and not other dimensions of maternal disrupted communication. In summary, the findings of these two studies indicate that infant attachment disorganization and overall disrupted maternal interaction are associated with amygdala volume, while maternal withdrawal is specifically associated with hippocampal volume. Further work is needed to replicate these findings and also to understand in more detail why withdrawal is more specifically related to left hippocampus, BPD, and suicidality/self-injury, while broader aspects of attachment disturbance are related to left amygdala volumes.

Given the small sample size of the present study, this must be considered a pilot study, and confirmation of these longitudinal relations need to be confirmed in larger and more diverse samples before we can be confident of their validity. An additional limitation of the study is that the key variables were assessed at single time points, thus it is possible that dissimilar effects would be evident at ages before and after the assessment points of this study. In particular, it should be noted that lifetime prevalence of BPD features and suicidality/self-injury were assessed by the SCID at age 20, while regional brain volumes were not assessed until age 30. However, as reviewed earlier, there appears to be minimal change in hippocampal volume over early adulthood from ages 20 to 30 [51–54], the assessment time points for psychopathology and brain volume assessed here. Clearly, future work including assessments of both psychopathology and regional brain volumes at both ages is needed, so that direction of effect can be more conclusively established. In addition, studies of the effects of trauma occurring in adulthood (combat, disasters, violence), especially in the absence of a history of childhood maltreatment and disrupted parenting, on structure and function of left and right hippocampus are needed. Given the limited assessments of this study, we are unable to evaluate whether the nonsignificant associations between early maternal behavior and the right hippocampus are due to the effects of trauma experienced at later points in childhood and adulthood. Future longitudinal studies that assess parenting behavior, later maltreatment and trauma, regional brain volumes, and psychopathology at multiple stages of development will be needed to better understand the developmental trajectories associated with both atypical regional brain volumes and serious emotional dysregulation in adulthood.

5. Conclusions

The prospective longitudinal design of this study and the scope of assessments from infancy to adulthood provide a unique window into the potential developmental implications of early disruptions in caregiving for both brain and behavior. The results suggest that early maternal withdrawal may have long-term effects on later emotional dysregulation in adulthood, in part by promoting increased volume in the left hippocampus. These findings also suggest that early childhood may be one developmental window within which clinical intervention could preempt some of the serious long-term consequences of early disrupted maternal care.