19 mothers with BPD and 30 HC mothers (Table 1) participated in the study that was approved by the ethics committee of the Charité – Universitätsmedizin Berlin (approved 27.08.2013, EA2/097/13). All 49 participated in the fMRI study and had sufficient data quality after fMRI quality control (Supplementary Material 1 contains quality control criteria). One mother in the BPD group and one in the HC group did not participate in the mother-child interaction task due to time constraints unrelated to the study design, reducing sample sizes in these sub-analyses by one per group.

Mothers from both groups and their biological children aged 5–12 years (M = 8.31, SD = 1.97) were required to live together and be proficient in German. In the BPD group, inclusion criteria were a current BPD diagnosis (per the International Personality Disorder Examination interview, IPDE) (36) and remission from mood episodes as per a score of < 8 in the Hamilton Depression Scale (37, 38).

In the HC group, mothers were included if they reported no current or lifetime axis I or II disorders in the IPDE or the Mini International Neuropsychiatric Interview (MINI) (39).

Mothers were excluded in case of: acute suicidality, being in an emotional crisis or hospitalized, neurological diseases, current alcohol or drug dependence, a history of schizophrenia or manic episodes per the MINI, anxious-avoidant or antisocial personality disorder per the IPDE, use of benzodiazepines within the past six months (40), change of psychotropic medication two weeks before study entry.

Children were excluded if they had a diagnosis of autistic disorder (based on DSM-IV criteria) or an IQ score below 70 in the Culture Fair Intelligence Tests (41).

In the BPD group, current and lifetime comorbid disorders were assessed using the MINI (39). Of the mothers with BPD, n = 17 reported a history of previous depressive disorders, n = 4 reported current comorbid anxiety disorders, n = 1 a current obsessive compulsive disorder, n = 3 reported current substance abuse, and n = 2 reported a current eating disorder. N = 5 mothers with BPD reported no history of psychiatric hospitalizations, n = 4 reported one hospitalization, n = 10 reported two or more hospitalizations.

Mothers with and without BPD showed overall mostly non-hostile parenting behavior with average ratings between five and six, indicating mostly covert signs of anger, impatience, or boredom (e.g., slightly raising one’s voice, rolling eyes, snapping at the child or responding irritably), but the HCs were rated significantly more non-hostile than mothers with BPD (MHCs = 5.93 ± 0.96; t = 2.6, p = .01; MBPD = 5.08 ± 1.31). This was in line with the results from the total UBICA sample reported in Kluczniok et al. (32).

In the following, ANOVA results for RTs and hit rates of the affect recognition task are reported (Figure 1, Table 2). In two individuals of the BPD and one of the HC groups, hit rates were 0% in at least one condition. In a sensitivity analysis, we excluded these individuals and repeated the ANOVA analyses, but the results remained identical in interpretation. Therefore, we report the results with the full sample.

None of the correlations between hit rates or RTs in the affect recognition task of sad faces with maternal non-hostility were significant after correcting for multiple comparisons (p values >.16), disconfirming Hypothesis 2.

For the HC > BPD contrast of own > unknown sad child faces, we observed a non-significant (ROI p_FWE = .28) activation within the left dmPFC (x = -12, y = 50, z = 30; t = 3.6), disconfirming Hypothesis 3. In the reversed BPD > HC contrast of own > unknown sad children, no group differences in ToM network activation emerged.

In mothers with and without BPD, associations between higher non-hostility scores and brain activation while viewing sad faces of their own vs. unknown children were significant for the left TPJ (x = -50, y = -44, z = 28; T = 4.52; ROI p_FWE = .03), right TPJ (x = 52, y = -44, z = 22; T = 4.44; ROI p_FWE = .04), and left dmPFC (x = -16, y = 58, z = 20; T = 4.39, ROI p_FWE = .05), confirming Hypothesis 4 (Figure 2).

To explore if the association between ToM network activation and maternal behavior was driven by differences in positive parenting behavior, we regressed maternal sensitivity during the mother-child interaction on ToM network activation but found no significant relationship (T < 3.69, ROI p_FWE >.23).

There were no activations significantly associated with maternal non-hostility in the precuneus or PCC.

Mothers with BPD showed a reduced ability to understand children’s facially expressed emotions. Stepp et al. (62) suggest that mothers with BPD have difficulties understanding and regulating their own and their children’s emotions, possibly leading to higher ambiguity in emotional social interactions. Problems in dealing with other individuals’ emotional expressions is further implied by studies showing that individuals with BPD have difficulties in understanding the feelings of others (63) and find emotionally loaded social situations more overwhelming than HCs (64). The findings of the present study support this view. Possibly, in social interactions with their children, mothers with BPD find it harder than HCs to understand their child’s feelings, causing higher ambiguity and possibly, higher parenting stress (65). As indicated by an interaction between group and emotion showing difficulties of mothers with BPD in identifying sad faces, this might be most prevalent in situations where their children are sad.

However, we found no association between affect recognition performance and maternal non-hostility. This absence of a direct correlation suggests that the basic ability to recognize a sad face may be too distant from the complex demands required for displaying non-hostile parenting behavior during real-life interaction. In a study by Kiel et al. (66), mothers with BPD showed the same negative affect towards their child’s distress as HCs in the strange situations test but were less likely to engage in positive soothing behavior. Therefore, the difference may lie not in emotion recognition itself but in the downstream expression, regulation, or utilization of emotions to maintain positive behavior.

Future studies should employ more faceted, naturalistic affective ToM tasks. Further, studying mother-child social interactions might benefit from employing more emotionally salient stimuli such as infant and child crying, as these are expected to elicit stronger maternal emotional responses (67, 68).

Our results suggest that a neural correlate of non-hostile parenting behavior is increased activation of the mothers’ ToM network when viewing their own sad children. This association was found specifically for maternal non-hostility, not maternal sensitivity, highlighting the role of ToM processes, specifically affective ToM, in reducing dysfunctional parenting behavior. The regions found to be associated with non-hostility, the TPJ and dmPFC, have previously been shown to be associated with social cognition in close relationships (69). TPJ activation was, for example, shown to be positively related to self-reported parental empathy, meaning empathy with the participants’ children versus empathy in general (70). A dismissive attachment style was associated with lower TPJ activation during the reading the mind in the eyes test, an affect recognition task often implicated in affective ToM (71). Wever et al. (72) found the TPJ and dmPFC to be associated with empathy with the participants’ own versus unfamiliar children. The dmPFC is thought to play an important role in mentalizing processes (73), possibly enabling individuals to represent the intentions of others (74).

Together with the previous literature, our results suggest that mothers who respond in a non-hostile manner when interacting with their children show stronger activation of the TPJ and dmPFC when seeing their child displaying sadness. The role of these brain regions in maternal care is suggested (75). Further, this is compatible with the notion that interpersonal problems in mother-child relationships of mothers with BPD are partly explainable by maternal affective ToM.

We found no significant group differences between mothers with BPD and HCs in ToM network activation while seeing the sad faces of own vs. unknown children. Activation of the ToM network might be related to BPD symptoms in ways not covered by a group contrast. For example, when comparing brain activation upon seeing their own children in mothers with early life maltreatment versus non-maltreated mothers, Neukel et al. (47) reported heightened activation in ToM-related regions of the brain. Boehme et al. (76) found increased social anxiety to be related with increased TPJ and dmPFC activation in self-referential thinking. Beeney et al. (77) also reported greater activation in BPD compared to HCs in the TPJ, dmPFC, precuneus, and cingulate cortex when participants were asked to evaluate themselves. Possibly, in the present study, different symptom constellations in our BPD patients might have led to different, counteracting alterations in the ToM network activation. Alternatively, the absence of a group difference might be attributable to the small sample size, as group analyses are especially prone to power issues (78).

The association between ToM network activation upon seeing own sad children and maternal non-hostility implies affective ToM as a potential intervention target in parents at risk for dysfunctional parenting behavior. Interventions targeting social abilities are being developed for parents and their families (79). For example, Volkert et al. (80) present a parent training directed at severely mentally ill parents that constitutes an adaptation of the lighthouse parent training against childhood maltreatment (81) interlaced with mentalization-based psychotherapy (82). Mentalization is conceptualized as the ability to represent the mental states of oneself and others and understanding current feelings and behaviors as an expression of these mental states (82).

Previous studies demonstrate that by psychotherapeutic interventions, specific neural processes can be strengthened (31, 83). For instance, in patients with bipolar disorder, Meyer et al. (31) demonstrated that a psychotherapeutic intervention with a practical focus on social functioning and perspective-taking increased the neural activation in the ToM network post-therapy. Pre- and post-intervention fMRI studies are needed to better understand the role of the ToM brain network in improving parent-child interactions.

The major limitation of the present study is its small sample size, especially in the BPD group, which renders the results preliminary and limits the generalizability of the findings. Further, it is possible that fluctuating mood states in the mothers with BPD influenced the association between ToM network activation and maternal non-hostility. Repeated measurements are warranted in future research, as they would allow for stronger conclusions regarding causality, which the current cross-sectional data do not permit.

The strengths of the study lie in its integration of different data sources (observational, experimental, neuroimaging) and the use of stimuli based on images of the participants’ own children. Pooling BPD and HC groups for the regression had the advantages of increasing the variance in maternal non-hostility observed in interactions and improving the power over studying the groups separately but prevented us from detecting group-specific effects.

As described in Table 1, the BPD sample of the current study reported low intake of medication relative to many other patients with BPD (84), a possible consequence of the psychotherapeutic as opposed to pharmacological treatment approach of our clinic, where many of the patients were recruited, and of the exclusion of patients with current depression scores above a clinical cut-off from participation. This low intake of medication has the advantage of reducing the effect of the pharmacotherapy on brain activation.

Mothers with BPD showed difficulties recognizing facial emotion expressions compared to HCs. Maternal non-hostile behavior was related to higher bilateral TPJ and left dmPFC activation, indicating that mothers who showed more non-hostility while playing with their children employed their ToM network more strongly when seeing their own sad children. This finding underscores the relevance of affective ToM in parenting behavior and may suggest that targeting affective ToM therapeutically could benefit parent-child interaction quality.