Neural correlates of fears of abandonment and rejection in borderline personality disorder

Borderline personality disorder (BPD) is a prevalent and disabling Background: psychiatric condition commonly associated with early life adversity. Social difficulties are a prominent symptom of BPD, particularly a fear of abandonment and rejection. There has recently been a growing interest in the neural basis of these social symptoms and their relationship to early experience. In Methods the current study, we examined social brain function and learning in BPD using functional MRI. Participants with BPD (n=20) and healthy controls (n=16) completed a computerized parametric social exclusion task (the “Cyberball” task). Brain activation was compared between groups and related to social symptom status and experiences of childhood trauma. Additional analyses were conducted using a reinforcement learning model treating social inclusion as a rewarding event. Participants with BPD demonstrated a group Results: effect of decreased right temporoparietal junction (TPJ) activation (p < 0.013, FWE-corrected). Increased fear of abandonment in BPD was associated with reduced inclusion-related activation of the inferior frontal gyrus (p = 0.003, FWE-corrected). Across all participants, TPJ inclusion-related activation was modified by prior experience of childhood physical neglect (p < 0.001, FWE-corrected). Reinforcement learning modelling revealed decreased midbrain responses to social inclusion in BPD participants (p = 0.028, FWE-corrected within midbrain mask), with decreased anticipatory midbrain activation in anticipation of social inclusion specifically associated with fears of abandonment (p = 0.019, FWE-corrected within a midbrain mask). findings demonstrate alterations in social brain function and Conclusions: social reinforcement learning in BPD, which are influenced by both early life experience and symptom status.


Introduction
Borderline personality disorder (BPD) is a common and serious disorder associated with significant impact on the individual, their families and a wide range of healthcare providers [1][2][3] . Social difficulties are a core feature of BPD 4 , and include marked fears of abandonment and social rejection 3,5-8 . Indeed these factors underlying their characteristically disturbed relationships have been proposed to be a phenotype of this personality disorder 9 . Both genetic and early environmental factors have been shown to contribute risk for developing BPD symptoms 3,10,11 , with evidence that early environmental factors, such as childhood adversity 11 , play a particularly prominent role in the development of BPD 12 and interpersonal symptoms in particular 5,10,13 .
Fear of abandonment is one of the core symptoms of BPD recognized in both Diagnostic and Statistical Manual of Mental Disorder (DSM)-5 6 and International Statistical Classification of Diseases and Related Health Problems (ICD)-10 14 . Fearing abandonment is one of the most stable features of the disorder, and is associated with negative core beliefs 7,15 and an increased propensity for self-injurious behavior [16][17][18] . The central nature of this symptom is further confirmed by formal testing, in which people with BPD show heightened sensitivity to rejection and abandonment in questionnaire measures 19 , and increased psychophysiological reactions to scripts relating to abandonment 20 .
The marked fear of abandonment seen in people with BPD is reflected in the disorganised attachment patterns observed in these patients 5,21-23 . Abnormalities in the development of secure attachment patterns are considered to underlie the development of persistent fear of rejection seen in people with BPD 5,6,24 . The development of stable attachment relationships requires the experience-dependent maturation of specific brain circuitry mediating key aspects of social function and learning, such as mentalizing 5,24 . Such regions include the superior temporal cortex, temporoparietal junction (TPJ), and insula and inferior frontal gyrus (IFG) 25 . Theoretical analyses of BPD have suggested that altered social learning during development, potentially secondary to childhood adversity, may lead to the emergence of key symptoms of BPD, such as fear of abandonment 5,24 .
Within BPD, several imaging studies have assessed the processing and regulation of emotion, self-injury, cognitive disturbance and impulsivity (see reviews 26,27). However, there has been an increasing interest in imaging the neurobiology mediating interpersonal difficulties. The behavioural and neural correlates of social rejection and social learning can be studied experimentally using computer games 28,29 . A commonly applied task is the "Cyberball" game, in which the participant takes part in a computerised ballthrowing game with two on-screen figures, where the degree to which the participant is included or excluded from the game is systematically varied 29 . Imaging studies using the Cyberball task have revealed a network of brain regions that are sensitive to the degree of inclusion and exclusion, including frontal and temporal brain regions implicated in social cognition 30-33 . It has also been shown that social inclusion events produce reward learning signals in a range of brain regions, including the inferior frontal cortex and posterior temporal areas that are known to be involved in social behavior 31,34 .
In the current study, we built on previous studies of BPD using the Cyberball task 33,35-40 . Ruocco et al. demonstrated that people with BPD demonstrate increased medial prefrontal cortex (PFC) activation during social exclusion 40 . Using functional MRI (fMRI), Domsalla and colleagues found that people with BPD exhibited generalised hyperactivation of dorsal medial PFC, anterior cingulate, precuneus and superior/inferior parietal lobules, and that this activity was not modulated by social inclusion, as was seen in controls 33 . We extended this work by examining the relationship between Cyberball brain activation and (a) BPD symptoms of social difficulty, named fears of abandonment and unstable relationships; and (b) reported early life adversity. We also analysed the results in terms of social reward learning models and demonstrate altered reward learning in BPD consistent with previous models of the disorder 24 . We hypothesised that within people with BPD, social inclusion-evoked brain activation would demonstrate associations with previous childhood adversity, and their current interpersonal symptoms. These associations would likely be demonstrated in areas mediating mentalizing, namely the superior temporal cortex, insula and IFG.

Methods and materials Participants
A total of 20 people with borderline personality disorder (BPD) were recruited from outpatient and support services from around Edinburgh, Scotland. Diagnoses were confirmed using the Structured Clinical Interview for DSM-IV (SCID-II; http://www.scid4. org/). Current symptoms were assessed using the Zanarini Rating Scale for Borderline Personality Disorder (ZAN-BPD) 41 . Adverse childhood events were assessed using the Childhood Trauma Questionnaire (CTQ 42 ). In total, 15 BPD participants were receiving antidepressant medication and 12 were taking antipsychotic medication. A total of 20 age-and sex-matched controls were recruited from the community; however, 4 were excluded due to technical issues during scanning, leaving 16 controls. Exclusion criteria for all participants included pregnancy, MRI contraindications, diagnosis of a psychotic disorder, previous head injury or current illicit substance dependence. Controls met the additional criteria of no personal or familial history of major mental illness. Ethical approval was obtained from the Lothian National Health Service Research Ethics Committee (09/S1101/49), and all participants provided written informed consent before taking part. Demographic details are given in Table 1.

Experimental task
Participants performed the Cyberball social exclusion task 43 during fMRI, adapted from a previous implementation by Kumar et al. 2009 31 . The task involves playing "catch" with two computercontrolled players, during which the participant can be systematically included or excluded from the game. We used this task as it assesses neural responses to social exclusion, is known to activate a range of social brain regions 30 , and is amenable to reinforcement learning modelling 31 . The task was modified such that inclusion was varied parametrically over four levels: 0%, 33%, 66% and 100%, achieved by arranging the task into blocks of nine throws, respectively involving zero, one, two or three throws to the participant (Supplementary material; Supplementary Figure 1). Here, 100% inclusion means the degree to which the participant was included was equal to that of the other two players, with each receiving three throws per nine-throw block. Mean block duration was 24s (in part dependent on participants' reaction times), with onsets denoted by the appearance of the cartoon figures following rest, and offsets by the conclusion of the final throw animation. Blocks were randomized, and interleaved with 13s rest blocks. Within blocks, throwing events were jittered to permit event disambiguation for reinforcement learning analysis. Reaction times were analysed within a repeated measures ANOVA (within-subjects: inclusion level; between-subjects: group). Response bias, that is the propensity to favour one computer player over the other, was also assessed (Supplementary methodology; Supplementary Results).
At the first-level, each level of inclusion was modelled using separate regressors. Motion parameters were included as nuisance regressors. T-contrasts were generated for each inclusion level > rest, and taken into a random-effects second-level full factorial design matrix, modelling inclusion and diagnostic group. Both the main effects and the interaction were evaluated.
The relationships between current social symptoms and fMRI response to increasing inclusion were examined within a second level multiple regression of the first-level parametric contrast of increasing inclusion (100% > 66% > 33% > 0%), by the ZAN-BPD Total Disturbed Relationships sub-score items. Focusing on social interaction, this comprises items six (frantic efforts to avoid real or imagined abandonment; Z6) and nine (unstable and intense interpersonal relationships; Z9). It has been theorized that several characteristics of borderline psychopathology, especially those concerning interpersonal relationships, are secondary to changes in self-other representations 44 . Any assessment of one's relationships with others implicitly involves the consideration of self 24 , and could therefore be modulated by its disturbance, outside the boundaries of any one particular personality disorder 45 . Therefore, we also included ZAN-BPD item four (identity disturbance: markedly and persistently unstable self-image or sense of self; Z4) in this single symptom regression analysis. The impact of past childhood adversity on fMRI responses to increasing inclusion were also assessed within a separate second level multiple regression incorporating the subscales of the CTQ (physical/emotional abuse/neglect, and sexual abuse).
We also adopted a reinforcement learning approach, examining how those with BPD learn to anticipate the rewarding outcome of social inclusion. Reinforcement learning models describe mathematically the learning processes that occur during conditioning and reward learning and closely relates to the observed changes in dopaminergic neuron firing seen in animals 46 . Previous studies have demonstrated that reinforcement learning approaches are applicable to social learning tasks, including the Cyberball task 34,47 . A temporal difference reinforcement learning model represented the task as an instance of classical conditioning, where those moments when the ball was received by the participant being modelled as a rewarding unconditioned stimulus (outcome), and moments when the ball was held by the computer-controlled characters as a conditioned stimulus (anticipation). fMRI regressors describing the anticipatory and outcome phases of each ball throw were modulated by a parametric regressor modelling the prediction error at these moments, using MathWorks Matlab R2014a (http://www.mathworks.com). The outcome phase represents trial-by-trial responses to inclusion (reward=1) and exclusion (reward=0), whereas anticipation captures learning regarding potential inclusion (see Supplementary methodology for more information). As well as group comparisons, patients' contrasts were also regressed by the three social symptoms (Z4, Z6, Z9).
For all fMRI analyses, a voxel-wise threshold of p < 0.001 was applied, and only those regions additionally achieving a clusterwise FWE-corrected significance of p <0.05 were reported. As temporal difference models have been shown to provide a good account for midbrain activation during reward learning 48,49 , the reinforcement learning analyses were corrected within an anatomically-defined mask comprising the substantia nigra and ventral tegmental area (Supplementary methodology).

Behaviour and psychopathology
There were no performance differences between the groups as assessed by reaction times across the task (F(1,34) = 0.21; p =0 .65), indicating matched engagement with the task (Supplementary Figure 2). Participants with BPD had significantly greater Zanarini and CTQ scores than controls, as expected (Table 1).

Main effect of inclusion on brain activation
Examining the main effect of increasing inclusion on brain activation across the groups revealed robust activation of several brain regions known to be engaged during social cognition, including the IFG, inferior parietal lobule/TPJ, anterior cingulate cortex, caudate and thalamus (Supplementary Figure 3).
Main effect of group on brain activation Across the task as a whole, participants with BPD showed lower activation of the right TPJ and left middle occipital cortex bilaterally ( Figure 1).

Figure 1. Controls > BPD across all levels of inclusion.
The behavioural personality disorder (BPD) group showed diminished temporoparietal junction activation across the Cyberball task as a whole (Table 2). Activation map displayed at a voxel-wise uncorrected threshold of p < 0.001; however, only clusters achieving a FWE-corrected significance of p < 0.05 are reported. Bar chart demonstrates R temporoparietal junction activation (TPJ) at each level of inclusion for both groups.
levels of reported childhood physical neglect were associated with a decreased inclusion response within bilateral superior temporal cortex, and middle anterior cingulate cortex (Table 2; Figure 2). These effects remained significant even when restricted to the BPD group only (p < 0.002). No effects were seen for the other CTQ subscales or total score.

Group x inclusion interaction
No regions showed a significant group by inclusion interaction.

Relationship of childhood trauma to brain activation
We next investigated whether experiences of childhood trauma correlated with neural responses to increasing inclusion. Higher

Relationship of social symptoms in BPD to brain activation
Analysis of the effect of social symptoms of BPD on brain activation was restricted to the BPD group, due to a floor effect for these symptoms in the control group. This analysis revealed that increasing fear of abandonment was associated with a reduced response to inclusion within the right IFG, and left postcentral gyrus within the BPD group (Table 2; Figure 2).
Temporal difference modelling of inclusion events as rewarding We next used temporal difference modelling to examine neural responses to individual events of inclusion and exclusion, with inclusion events treated as rewarding. Compared to controls, participants with BPD showed significantly less activation in the midbrain during trials on which they were included, suggesting that their experience of inclusion was not imbued with the same salient or rewarding character as that felt by controls (Figure 3a; MNI -9 -13 -14; Z = 3.26; p = 0.028, FWE-corrected within the midbrain mask). There was no association between midbrain outcome responses and social symptom status, suggesting that decreased midbrain response to inclusion may represent a stable feature of the condition. We further examined brain activation during the anticipation phase of each trial. In participants with BPD, increasing fear of abandonment was associated with reduced midbrain activation, suggesting a diminished ability to learn to anticipate inclusion (Figure 3b; MNI 6 -13 -14; Z = 3.55; p = 0.019, FWE-corrected within the midbrain mask). There were no associations between midbrain activation and childhood adversity.
Correlation with medication and depressive symptoms Neither depressive symptoms nor medication status correlated with activation of the R IFG, TPJ or midbrain (Supplementary results; Supplementary Figure 4).

Discussion
In healthy individuals, social inclusion in the Cyberball task activated a network of brain regions including the inferior frontal gyrus (IFG) and temporoparietal junction (TPJ). Participants with BPD demonstrated a number of differences in brain activation during the Cyberball task. Firstly, participants with BPD showed reduced activation of the TPJ across all levels of inclusion. Secondly, individuals with BPD showed altered activation of the right IFG in relation to a key symptom of the disorder -fear of abandonment. Thirdly, TPJ, insula and anterior cingulate cortex (ACC) responses in both groups were modified by prior experiences of childhood physical neglect. Finally, participants with BPD showed a decreased reward response to social inclusion in the midbrain, and decreased anticipatory midbrain responses to social inclusion in relation to symptoms of fear of abandonment. Overall these results demonstrate sustained alterations in the TPJ and midbrain in BPD with additional modulation of RIFG function in relation to the key symptom of fear of abandonment. Differences in TPJ function were in part related to adverse childhood experience.
There was a main effect of group within the right TPJ, with participants with BPD demonstrating reduced activation. The difference was not attributable to behavioural performance, which was matched across groups; however there was no significant task x group interaction in this region, meaning this could not be firmly attributed to altered responses to inclusion per se. Notably these findings are consistent with those reported in a previous study of BPD 33 . However, the TPJ is known to play a key role in social behavior 25,50-53 and has been implicated in social reward learning 34,47 . The roles of posterior temporal regions in social function are complex, but are believed to sub-serve core components of reciprocal social interaction, imitation and relating self to others 25,51,52,54-57 . The TPJ has in particular been implicated in considering the internal states of others 58 , and in learning social rules about how much to trust another individual 34 . Individuals with BPD can have difficulty in developing trust in relationships and show a tendency to rate others as untrustworthy on formal testing 59,60 . They have also demonstrated reduced activation in nearby posterior superior temporal structures during a cognitive empathy 61 and attribution of emotional empathy tasks 62 . Decreased activation of TPJ in individuals with BPD is thus likely to reflect an impaired ability to predict and trust other's social intentions 59 .
We also found that in BPD activation of the right IFG, a core component of the social brain network, was related to the symptom of fear of abandonment. The IFG is known in both primates and humans to contain neurons that are involved in social learning and show activation which "mirrors" responses seen in others, in particular encoding the intention of the other's action 63-67 . This role of the IFG in interpreting and responding to the actions of others extends beyond motor actions to recognizing and empathizing social and emotional responses 66,67 . As such this region has been considered to form a key part of the brain mechanisms required for reciprocal social learning during development, developing an "internal working model" of social interaction, which governs expectations about the outcomes of social encounters 5,24,47,67 . It is known that people with BPD demonstrate a bias in attributing malevolent intentions to others 68 , and reduced IFG activation when attributing emotional intentions 62 .
We additionally sought to determine whether experience of childhood adversity might contribute to the development of altered social brain responses across both groups. Regression of childhood trauma questionnaire scores against brain activation during the Cyberball task revealed that greater experiences of physical neglect during childhood were associated with reduced responses to social inclusion within left TPJ, right insula, and ACC. Physical neglect represents a definitive breakdown of normal reciprocal social interactions during childhood. These results therefore support the view that impaired nurturing from caregivers during childhood alters one's understanding of the likely outcomes of social interactions 5,24 . As the negative relationship was also seen in insula, this suggests that physical neglect also impacts on the normal development of one's interoceptive sensations in the context of social interactions 69,70 . Correlations with brain activation during the Cyberball task were not seen for the other four CTQ measures. This could however be due to a lack of power to detect an effect in this sample or be because of resistance -conscious or otherwise 71,72 -to the recall of physical or sexual abuse 73 .
The IFG and TPJ have both been implicated in mediating the brain's responses to the rewarding nature of social inclusion and interaction. Specifically these regions show temporal patterns of activation during social learning, which are consistent with those seen to a range of other rewarding stimuli 34,47 . The underlying source of these reward-related brain signals is believed to be the firing of dopaminergic neurons in midbrain regions, which have been shown to be phasically responsive to reward learning in animal models and in human imaging studies 46,49 . Here we examined the response of the midbrain to social inclusion events in BPD by modelling inclusions events as rewarding. This analysis showed that participants with BPD showed decreased midbrain activation to the reward of social inclusion than control participants, suggesting that people with BPD have a diminution of positive reward responses to social experience. In addition, participants with BPD with greater fear of abandonment also showed decreased activation of the midbrain in response to the anticipation of social inclusion. Phasic dopamine release has been argued to be more representative of the expectation-outcome mismatch i.e. reward prediction error, rather than simple hedonistic reward itself 75 . It has been suggested that the pervasive rejection felt by people with BPD -even when being fully included -may be due to people with BPD biasing their perceptions to conform to their expectations 19,33 , in which case, a dopaminergic "prediction error" signal is unlikely to be generated. This apparent selective blindness to inclusion in people with BPD is additionally supported by evidence from EEG studies 39 . Indeed it seems that only during conditions of extreme over-inclusion do people with BPD report reduced negative emotions during social interaction 76 .
A number of important limitations to the current study should be noted. Firstly, we did not directly measure the participants' subjective experience of the different levels of social inclusion during the task. Previous behavioural studies have shown that individuals with BPD show greater feelings of exclusion during the Cyberball task 35 . However we chose not to directly assess responses to the changing task contingencies in the present study, as doing so would necessarily interrupt the flow of the task and alert the participants to the covert changes in the degree of inclusion, whilst post-task analysis would be confounded by the differing degrees of inclusion/exclusion present across the task. Secondly, participants with BPD in the current study had a number of comorbid conditions and many were on some form of medication. However, we chose to include participants with typical outpatient presentations of BPD, increasing the generalisability of our findings. Notably, we did not identify any consistent relationship between the differences in social brain activation seen in the study and the most common comorbid conditions, such as depression, or treatment with either antidepressant or antipsychotic medications. Thirdly, our results appear to differ from those of Kumar et al. 31 , who also adopted a reinforcement learning approach to Cyberball data, but did not find TD-associated midbrain activation. This could be because their data was acquired at 1.5T or because the randomized nature of our inclusion blocks induced more abrupt violations of expectation. Our studies are qualitatively comparable, although we introduced a degree of jitter in considering the anticipation of inclusion, whereas their implementation was fixed, albeit on a similar time scale.

Conclusions
The current findings illuminate the neural basis of key social symptoms in borderline personality disorder and their relationship to developmental experience. We show that participants with BPD have differential activation of brain regions mediating social interaction, including the TPJ and midbrain. TPJ activation was also found to be related to childhood physical neglect. Furthermore, individuals with BPD experiencing a high fear of abandonment had altered brain responses during social learning in key brain regions mediating social interactions and reward, including the IFG. These results provide a framework for the further translational modelling of disrupted social reward learning in BPD. Future therapies for BPD may aim to reprogram this altered circuitry by training people with these symptoms to learn that social interaction can have sustained positive outcomes.

Consent
All participants provided written informed consent before taking part. Written informed consent for publication of the participants' anonymised details and data were obtained from the participants.
Author contributions JH conceived the study. LR, DS and JH designed the experiment. LR, MP, KN and JH carried out the research. LR and JH prepared the first draft of the manuscript. All authors were involved in the revision of the draft manuscript and have agreed to the final content.

Competing interests
No competing interests were disclosed. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.