Borderline personality disorder (BPD) is characterized by affective instability, interpersonal dysfunction, suicidality, and self-harming behaviors (American Psychiatric Association, 2013). Although BPD is most often diagnosed in emerging adults, there is mounting evidence that symptoms often begin in adolescence and show continuity into adulthood (Cohen, Reference Cohen2008; Crowell, Beauchaine, & Linehan, Reference Crowell, Beauchaine and Linehan2009; Zanarini, Temes, Frankenburg, Reich, & Fitzmaurice, Reference Zanarini, Temes, Frankenburg, Reich and Fitzmaurice2018; Zelkowitz et al., Reference Zelkowitz, Paris, Guzder, Feldman, Roy and Rosval2007). This developmental course aligns with broader research on impulsivity and emotion regulation in adolescence (Steinberg et al., Reference Steinberg, Icenogle, Shulman, Breiner, Chein, Bacchini and Takash2018). In BPD, impulsive and self-destructive behaviors tend to occur in response to momentary emotional arousal and constitute a core symptom of the disorder (Brereton & McGlinchey, Reference Brereton and McGlinchey2020; Crowell et al., Reference Crowell, Beauchaine and Linehan2009; Linehan, Reference Linehan1993).
Difficulty controlling impulses in the face of negative emotions (e.g. self-injury after a break-up with a romantic partner) is referred to as negative urgency (Settles et al., Reference Settles, Fischer, Cyders, Combs, Gunn and Smith2012; Whiteside & Lynam, Reference Whiteside and Lynam2001). Positive urgency is a related construct (Cyders et al., Reference Cyders, Smith, Spillane, Fischer, Annus and Peterson2007; Cyders & Smith, Reference Cyders and Smith2008) that describes a tendency to act impulsively in order to enhance positive mood (e.g. engaging in risky sexual behavior with a stranger while intoxicated). Collectively, negative and positive urgency are referred to as emotion-related impulsivity and reflect a crucial intersection of emotion regulation capacities and impulsivity that is robustly associated with psychopathology (Johnson, Elliott, & Carver, Reference Johnson, Elliott and Carver2020). At the level of neural circuits, emotion-related impulsivity may reflect an imbalance between the inhibitory control functions of the ventromedial prefrontal and orbitofrontal cortex (vmPFC, OFC) and emotion-congruent response tendencies in the amygdala and ventral striatum (VS; Johnson et al. Reference Johnson, Elliott and Carver2020; Zorrilla & Koob, Reference Zorrilla and Koob2019). For example, even mild uncontrollable stressors can profoundly disrupt PFC functioning (Arnsten, Reference Arnsten2009), leading to a reliance on short-sighted, emotion-congruent behaviors that are immediately reinforcing, to the detriment of long-term wellbeing and safety.
Neuroimaging studies of emotion in adults with BPD consistently identify abnormalities in similar fronto-limbic circuits, including the amygdala, VS, medial PFC, and dorsal anterior cingulate cortex (dACC; Enzi et al. Reference Enzi, Doering, Faber, Hinrichs, Bahmer and Northoff2011; Herpertz et al. Reference Herpertz, Dietrich, Wenning, Krings, Erberich, Willmes and Sass2001; Minzenberg, Fan, New, Tang, & Siever, Reference Minzenberg, Fan, New, Tang and Siever2007; Salvador et al. Reference Salvador, Vega, Pascual, Marco, Canales-Rodríguez, Aguilar, Anguera and Pomarol-Clotet2016; Schulze, Schmahl, & Niedtfeld, Reference Schulze, Schmahl and Niedtfeld2016; Silbersweig et al. Reference Silbersweig, Clarkin, Goldstein, Kernberg, Tuescher, Levy and Stern2007). More specifically, several studies have noted a fronto-limbic imbalance in BPD: in response to a range of emotional stimuli, limbic regions are often more active whereas activity in prefrontal regions involved in emotion regulation (esp. mPFC and ACC) is blunted (Donegan et al., Reference Donegan, Sanislow, Blumberg, Fulbright, Lacadie, Skudlarski and Wexler2003; Herpertz et al., Reference Herpertz, Dietrich, Wenning, Krings, Erberich, Willmes and Sass2001; Kamphausen et al., Reference Kamphausen, Schröder, Maier, Bader, Feige, Kaller and Tüscher2013; Minzenberg et al., Reference Minzenberg, Fan, New, Tang and Siever2007; Soloff, White, Omari, Ramaseshan, & Diwadkar, Reference Soloff, White, Omari, Ramaseshan and Diwadkar2015). Fronto-limbic accounts of emotion dysregulation center on the amygdala, which is involved in detecting a threat, representing the emotional significance of stimuli (LeDoux, Reference LeDoux2007), and encoding and retrieving fear memories. Furthermore, emotional experience depends on interactions between the amygdala and mPFC and dACC, which are involved in emotional appraisal and regulation (Etkin, Egner, & Kalisch, Reference Etkin, Egner and Kalisch2011). Altogether, neuroimaging studies of BPD suggest a model of emotion dysregulation in which blunted activity in PFC and ACC and heightened activity in the amygdala and VS contribute to failures in inhibiting emotion-congruent responses (Schulze et al., Reference Schulze, Schmahl and Niedtfeld2016; Silbersweig et al., Reference Silbersweig, Clarkin, Goldstein, Kernberg, Tuescher, Levy and Stern2007). While these studies imply that effective connectivity (EC) of mPFC with amygdala and VS may play a key role in BPD, prior research has not tested this idea directly.
Preclinical research casts the amygdala within a larger circuit that includes the cortex, striatum, and midbrain. Together, this circuit is fundamental to the experience and expression of emotional behaviors (Cardinal, Parkinson, Hall, & Everitt, Reference Cardinal, Parkinson, Hall and Everitt2002). Further, nonhuman animal studies demonstrate that the central nucleus of the amygdala (CeN) and the basolateral amygdala (BLA) play dissociable roles in the generation of Pavlovian associations and retrieval of relevant unconditioned responses during instrumental learning (Cardinal et al., Reference Cardinal, Parkinson, Hall and Everitt2002; Cartoni, Balleine, & Baldassarre, Reference Cartoni, Balleine and Baldassarre2016). Broadly speaking, the phylogenetically old CeN is a major controller of the autonomic nervous system, having strong projections to the hypothalamus, periaqueductal gray (PAG), basal forebrain, and brainstem – brain regions contributing to autonomic arousal (Zahm, Jensen, Williams, & Martin, Reference Zahm, Jensen, Williams and Martin1999). In contrast, the phylogenetically newer BLA receives direct inputs from the sensory cortex and uses this information to construct an emotional representation of specific conditioned stimuli; it uses this information to influence the output of CeN (Cardinal et al., Reference Cardinal, Parkinson, Hall and Everitt2002; Zhang, Zhang, Holmes, & Pan, Reference Zhang, Zhang, Holmes and Pan2021). This body of research suggests that the expression of emotion-congruent behaviors depends on BLA's capacity to convey the emotional significance of specific conditioned stimuli to CeN, which projects to regions that affect arousal and initiate approach and avoidance behaviors (Cardinal et al., Reference Cardinal, Parkinson, Hall and Everitt2002).
Further, the amygdala and PFC functionally couple with the VS, forming fronto-striatal-limbic loops that support reward learning and motivation (Floresco, Reference Floresco2015). Studies of reward learning consistently find activity in VS in anticipation of reward (de la Fuente-Fernández et al., Reference de la Fuente-Fernández, Phillips, Zamburlini, Sossi, Calne, Ruth and Stoessl2002; Schott et al., Reference Schott, Minuzzi, Krebs, Elmenhorst, Lang, Winz and Bauer2008) and after the receipt of reward (Floresco, Reference Floresco2015; Pagnoni, Zink, Montague, & Berns, Reference Pagnoni, Zink, Montague and Berns2002). In studies of impulsivity, impaired dopaminergic functioning in VS discounts the value of future rewards, tilting choices toward more immediate rewards (Berridge, Reference Berridge2007; Gregorios-Pippas, Tobler, & Schultz, Reference Gregorios-Pippas, Tobler and Schultz2009; Martinez et al., Reference Martinez, Pasquereau, Drui, Saga, Météreau and Tremblay2020). Extant research suggests that VS function is impaired in individuals with BPD (Enzi et al., Reference Enzi, Doering, Faber, Hinrichs, Bahmer and Northoff2011; Sarkheil, Ibrahim, Schneider, Mathiak, & Klasen, Reference Sarkheil, Ibrahim, Schneider, Mathiak and Klasen2019; Silbersweig et al., Reference Silbersweig, Clarkin, Goldstein, Kernberg, Tuescher, Levy and Stern2007), though a fine-grained analysis of VS contributions to fronto-limbic abnormalities has not been a focus of the BPD neuroimaging literature.
Connectivity of both amygdala and VS show pronounced developmental changes in adolescence that are susceptible to stress (Fareri et al., Reference Fareri, Gabard-Durnam, Goff, Flannery, Gee and Tottenham2015; Gee et al., Reference Gee, Fetcho, Jing, Li, Glatt and Drysdale2016; Tottenham & Galván, Reference Tottenham and Galván2016), yet these connections have not been investigated in adolescents with borderline symptoms. This is surprising given that the emergence of BPD symptoms is associated with trauma, interpersonal discord, and chronic stress (Zanarini & Frankenburg, Reference Zanarini and Frankenburg1997; Zanarini, Gunderson, Marino, Schwartz, & Frankenburg, Reference Zanarini, Gunderson, Marino, Schwartz and Frankenburg1989). Importantly, during the transition from mid-adolescence to early adulthood, self-reported impulsivity shows marked mean-level decreases in the general populationFootnote †Footnote 1 (Harden & Tucker-Drob, Reference Harden and Tucker-Drob2011; Quinn & Harden, Reference Quinn and Harden2013). However, within-person changes in impulsive symptoms are heterogeneous. Some highly impulsive adolescents show relative stability in impulsivity or decrease only slightly during this period, potentially leaving these individuals vulnerable to persistent negative outcomes into their 20's (Burt, Sweeten, & Simons, Reference Burt, Sweeten and Simons2014; Quinn & Harden, Reference Quinn and Harden2013).
Altogether, while the fronto-limbic account has received attention in adults with BPD, little is known about fronto-limbic disturbances in adolescence, when the development of emotion-related impulsivity may crucially impact the onset and maintenance of BPD symptoms. In this resting-state fMRI study of adolescents and emerging adults with BPD symptoms, we examined fronto-limbic circuitry using EC analyses within a network neuroscience framework (Bassett & Sporns, Reference Bassett and Sporns2017). Furthermore, we tested how fronto-limbic connectivity related to within-person stability and change in emotion-related impulsivity over six- and twelve-month follow-up assessments. Motivated by evidence in a preliminary analysis that EC of R CeN differed substantially between groups, we tested whether R CeN connectivity underlies the relationship between clinical group membership (healthy control v. BPD) and within-person change in emotion-related impulsivity symptoms.
Methods and materials
Participants
Participants were 46 adolescents and emerging adults with BPD symptoms recruited from community and outpatient settings, as well as 44 sex- and age-matched healthy controls. All participants were screened using the Personality Assessment Inventory-Borderline scale (Morey, Reference Morey1991) with BPD participants screening ⩾ 30 and controls screening ⩽ 17. The average age was 20.53 years (range 13–30 years); 59 participants were female and 31 were male. Eight participants (n BPD = 6) were excluded from our analyses for having poor fMRI data quality due to excessive motion during the scan or failure to pass residual correlation checks (see online Supplementary Methods). See online Supplementary Table S1 for a complete demographic characterization of the final sample.
Procedure
Participants underwent two semi-structured diagnostic interviews to assess for psychopathology and personality disorder symptoms (First, Spitzer, Gibbon, & Williams, Reference First, Spitzer, Gibbon and Williams2002; Pfohl, Blum, & Zimmermann, Reference Pfohl, Blum and Zimmermann1997). Interviews were administered by two trained research assistants who were supervised by the senior author. Participants in the BPD group met diagnostic criteria for three or more of the DSM-IV-TR BPD symptoms, an empirically derived threshold for identifying clinically significant symptoms (Clifton & Pilkonis, Reference Clifton and Pilkonis2007). Exclusionary criteria for both groups included having a first-degree relative diagnosed with Bipolar I disorder or any psychotic disorder and a history of serious head injury or neurological disease. Control participants additionally had no history of psychiatric or substance abuse disorders.
In a separate session preceding the RS-fMRI scan and at six- and twelve-month follow-up visits, participants completed a battery of self-report questionnaires. We focus here on the UPPS-P Impulsive Behavior Scale (UPPS-P; Cyders et al. Reference Cyders, Smith, Spillane, Fischer, Annus and Peterson2007; Lynam, Smith, Whiteside, & Cyders, Reference Lynam, Smith, Whiteside and Cyders2006; Whiteside & Lynam, Reference Whiteside and Lynam2001), whose subscales measure (positive and negative) urgency, (lack of) premeditation, (lack of) perseverance, and sensation seeking. Internal consistency was excellent at baseline and follow-up assessments (α total = 0.95, α mean−subscales = 0.88). We were particularly interested in the negative and positive urgency subscales of the UPPS-P, given their relevance in psychopathology (Johnson et al., Reference Johnson, Elliott and Carver2020), though we compared our results with all UPPS-P scales to test specificity (online Supplementary Methods and Results). All study procedures were approved by the Institutional Review Boards of the University of Pittsburgh (PRO13010486).
MR data acquisition
Data were acquired using a Siemens 3T Tim Trio scanner with a 32-channel head coil at the University of Pittsburgh Medical Center. We collected five minutes of resting-state fMRI data at the end of a broader MRI protocol; subjects were asked to keep their eyes open and relax, but not fall asleep. We used a simultaneous multi-slice echo-planar sequence sensitive to BOLD contrast with scanning parameters: TR = 1.0 s, TE = 30 ms, flip angle = 55°, voxel size = 2.3 mm isotropic, 5× multiband acceleration. Participants completed a self-report questionnaire at the end of the protocol to determine if they fell asleep during the scan. No subjects were excluded for sleepiness.
RS-fMRI preprocessing
RS-fMRI preprocessing was conducted within FSL, NiPy, and AFNI (Cox, Reference Cox1996; Millman & Brett, Reference Millman and Brett2007; Smith et al., Reference Smith, Jenkinson, Woolrich, Beckmann, Behrens, Johansen-Berg and Matthews2004). Structural scans were registered to the MNI152 template (Fonov, Evans, McKinstry, Almli, & Collins, Reference Fonov, Evans, McKinstry, Almli and Collins2009) using affine and nonlinear transformations conducted in FSL. Functional image preprocessing included simultaneous 4-D motion and slice-timing correction (Roche, Reference Roche2011), brain extraction, alignment of subject's functional images to their anatomical scan using a boundary-based registration algorithm (Greve & Fischl, Reference Greve and Fischl2009), and a one-step nonlinear warp to MNI152 space that concatenated functional-to structural, structural-to-MNI152, and fieldmap unwarping transformations. To mitigate motion-related artifacts we used ICA-AROMA (Pruim et al., Reference Pruim, Mennes, van Rooij, Llera, Buitelaar and Beckmann2015), a data-driven classification algorithm that identifies and removes spatiotemporal components likely to reflect head movement. We note that ICA-AROMA is an empirically validated alternative to censoring (aka ‘scrubbing’) strategies that remove some volumes based on an a priori motion threshold (Ciric et al., Reference Ciric, Wolf, Power, Roalf, Baum, Ruparel, Satterthwaite and D2017). RS-fMRI data was not spatially smoothed for analysis (see online Supplementary Methods; Alakörkkö, Saarimäki, Glerean, Saramäki, and Korhonen, Reference Alakörkkö, Saarimäki, Glerean, Saramäki and Korhonen2017).
Analytic approach
Nodal parcellation and functional connectivity matrix generation
To define our regions of interest, we parceled voxels into functional regions (nodes) by combining leading cortical and subcortical parcellations (online Supplementary Methods) (Choi, Yeo, & Buckner, Reference Choi, Yeo and Buckner2012; Schaefer et al., Reference Schaefer, Kong, Gordon, Laumann, Zuo, Holmes and Yeo2018). We selected 19 fronto-striatal-limbic nodes, including portions of mPFC, OFC, and ACC as prefrontal nodes, and bilateral BLA, CeN, and VS as limbic nodes (Fig. 1, Table S2). Prior to computing connectivity between nodes, we averaged the time series for voxels with reliable signal in each node to obtain a single nodal time series. For each subject and node, we prewhitened time series with an Auto-Regressive Moving Average (4,2) model, retaining the residual time series for FC estimation (online Supplementary Methods).
Nodes without labels were not fit with CS-GIMME but depict medial prefrontal and ACC nodes that were dropped from our initial consideration set. Arrows reflect the directed influence of one node on another at the group and subgroup (BPD vs control) level. Solid lines denote edges that were estimated at the group (entire sample) level with the solid green edge denoting a significantly higher edge values amongst subjects in the control group. Dashed lines denote edges estimated at the subgroup (BPD vs HC) level, with red and green dashed lines denoting edges that were only estimated for subjects in the BPD and control group, respectively.
Finally, we computed Pearson correlations among the pre-whitened time series to yield a 19 × 19 adjacency matrix for each subject representing undirected functional connectivity amongst fronto-limbic regions. In order to remove unreliable edges from these matrices, we applied a minimal consensus thresholding procedure (de Reus & van den Heuvel, Reference de Reus and van den Heuvel2013). Specifically, we removed edges from all subjects that did not have a weight of r = 0.1 or higher in 25% or more of subjects. This resulted in the removal of 17 edges (10%) that were concentrated in OFC-subcortical and subcortical-subcortical connections.
Node selection: undirected analysis
The primary goal of our analyses was to examine EC using the Confirmatory Subgrouping Group Iterative Model Multiple Estimation algorithm (CS-GIMME; Gates & Molenaar, Reference Gates and Molenaar2012; Henry et al. Reference Henry, Feczko, Cordova, Earl, Williams, Nigg and Gates2019). However, given that GIMME conjointly estimates conditional relationships among all nodes, the number of free parameters increases exponentially as the number of nodes increases and parameter reliability decreases (Cassidy, Bowman, Rae, & Solo, Reference Cassidy, Bowman, Rae and Solo2017). To promote model convergence and reliable EC estimation, we performed a node selection analysis by fitting a single logistic ridge regression model predicting group status by all undirected edges, retaining nodes with edges that jointly predicted group status at the p < 0.01 level and were thus most likely to be implicated in EC (online Supplementary Methods and Results; Cule & De Iorio, Reference Cule and De Iorio2012; Moritz & Cule, Reference Moritz and Cule2018).
Effective connectivity network estimation and relations to group status and age
We retained the preprocessed time series of twelveFootnote 2 fronto-limbic nodes based on our node selection analysis (online Supplementary Table S3; online Supplementary Methods and Results). We estimated EC between these nodes using the CS-GIMME algorithm, a recent extension of the GIMME algorithm that reliably detects the presence and direction of edges in fMRI data at the individual, group, and sample levels.
After obtaining directed graphs from CS-GIMME, we investigated group differences in the role of individual nodes (nodal centrality) in fronto-limbic circuits. We calculated in- and out-degree centrality for each node that showed evidence of incoming or outgoing edges in the best-fitting CS-GIMME model (denoting the summed score of incoming and outgoing edges for each node; online Supplementary Methods). In order to identify which nodal centrality estimates best-differentiated groups, we entered all of these into a single logistic regressionFootnote 3 predicting group status:
where $k_1^{in}$
is in-degree for node 1 (Table 1).
Predicting stability and change in impulsivity
We tested for group differences in baseline levels of emotion-related impulsivity and within-person changes in emotion-related impulsivity symptoms over 6- and 12-month follow-up. We first fit a latent growth curve model (LGCM; Bollen and Curran, Reference Bollen and Curran2005), modeling latent intercept and slope terms for both negative and positive urgencyFootnote 4, which describe the baseline level and rate of within-person change, respectively, in negative and positive urgency scales. In this model, group, age, and their interaction were included as predictors of LGCM variables (Fig. 2).
(a) Conceptual diagram of latent growth curve model reported in Table 2. Solid green lines denote positive parameter estimates, demonstrating that group main effects are higher for both intercept terms (overall levels of ERI) in the BPD group. Solid red lines denote negative parameter estimates, demonstrating that age main effects on both intercept terms indicate decreasing levels of ERI in the total sample. Faded red lines denote “marginally” significant parameter estimates (0.05 < p < 0.10). Paths from age that intersect with paths from BPD to LGCM terms represent interactive effects (moderation) of age and group to predict LGCM terms. (b) Visual depiction of age × group interaction for each LGCM term, demonstrating significant group × age interactions for both NU and PU intercept terms and a marginally significant group × age interaction for NU slope.
We then tested if the group and group-by-age differences in the longitudinal model of emotion-related impulsivity were mediated by EC estimates of the R CeN (Fig. 3). As detailed below, the results from the joint logistic model indicated that R CeN played a particularly important role in differentiating BPD from HC participants (Table 1). Thus, we fit a combined dual-mediator LGCM, which tested the hypothesis that directed functional input and/or output of the R CeN mediated the relationship between group status and latent intercept and slope variables. In our final combined model, both in- and out-degree of the R CeN were mediators of LGCM variables for both negative and positive urgency. We allowed age to predict levels of directed centrality of R CeN, and to moderate the relationship between centrality estimates and LGCM variables (Hayes, Reference Hayes2015). As our final LGCMs only consisted of two models (one modeling the overall pattern of emotion-related impulsivity stability and change, and the second adding mediating roles of R CeN EC) a correction for multiple comparisons was not applied.
(a) Conceptual diagram of final latent growth curve model reported in Table 3. Solid green lines denote positive parameter estimates, demonstrating that main effects are higher for both baseline ERI variables and R CeNin in the BPD group. Solid red lines denote negative parameter estimates, demonstrating that R CeNin decreased with age, and that BPD was associated with lower R CeNout, which was in turn associated with lower (e.g. “more negative”) within-person change in PU. Faded red lines denote “marginally” significant parameter estimates (0.05 < p < 0.10). Paths from age that intersect with paths from R CeNin and R CeNout to LGCM terms represent interactive effects (moderation) of age on the second-stage mediation (i.e. from R CeN connectivity to LGCM terms). (b) Significant moderated mediation of from BPD to NU baseline levels through R CeNout, which was moderated by age. (c) Significant moderated mediation of from BPD to NU within-person change through R CeNout, which was moderated by age. (d) Visual split of moderated mediation into young participants (under 20 years old) and older participants (over 20 years old). Note most importantly the lower left panel, showing that amongst adolescents, participants with higher R CeNout had more negative levels of within-person change in NU over one year, demonstrating more rapid reductions in NU symptoms over the one-year follow-up.
Joint model: group differences in in- and -out-degree centrality
Note. Rows with missing values indicate that there was no group- or subgroup-level edge entering/exiting the node, and thus we elected to not compute estimates of centrality based only on individual-level edges. Results represent coefficients in a single model that were allowed to compete to explain variation in-group status. Note that positive coefficients represent that degree centrality estimates were higher in the BPD group. We also ran a supplementary set of analyses that fit separate models to each effect (online Supplementary Methods and Results, Table S5), which provides an intuition of the marginal group effects, whereas the joint model account the complex correlational structure of these connectivity scores (online Supplementary Fig. S3).
Results
Directed network estimation
A graphical depiction of the best-fitting CS-GIMME network is provided in Fig. 1; directed edge values are listed in online Supplementary Table S4. Effective connectivity estimates from CS-GIMME generated positive directed edges at both the group level (nine total) and the subgroup level (five total, two in the BPD group). We found a range of directed edges from the dmPFC to several mPFC nodes at the group level including left dACC, rmPFC, and vmPFC, as well as right dmPFC. We also found evidence at the group-level for EC between bilateral CeN, VS, and BLA nodes. In healthy controls, but not BPD participants, CS-GIMME detected edges from dACC to VS, vmPFC to the CeN, and CeN to BLA. Conversely in BPD subjects, the CS-GIMME algorithm found BPD-specific edges from bilateral BLA to CeN.
Group and age-related differences in fronto-limbic connectivity
In a joint model predicting group status from directed centrality estimates, we found a dissociation between in- and out-degree centrality in R CeN by group (Table 1). Subjects in the BPD group had significantly higher in-degree (t = 2.47, p = 0.02) and significantly lower out-degree (t = −3.82, p < 0.001) of R CeN. This effect did not differ by age: all age-related terms in the joint model were nonsignificant.
Associations with emotion-related impulsivity scales
As expected, negative and positive urgency subscale scores were significantly higher in the BPD group at baseline compared to controls (β NU = 0.88, p < 0.001; β PU = 0.53, p < 0.001; Table 2, Fig. 2). We also found group x age interactions in baseline emotion-related impulsivity scores: older controls had lower baseline negative and positive urgency scores than younger controls, whereas baseline negative and positive urgency scores in the BPD group did not differ substantially by age (β NU = 0.10, p = 0.01; β PU = 0.086, p = 0.01).
Parameter table: latent growth curve model of ERI scales and relations to BPD, Age, and their interaction
Note. Parameter table reflects parameter estimates estimated in and extracted from the lavaan R package (Rosseel, Reference Rosseel2012). Confidence intervals and corresponding p values were estimated based on standard errors derived from 1000 bootstrapped samples.
In a final combined LGCM, we tested if group differences in emotion-related impulsivity scores over the one-year follow-up period were accounted for by EC of R CeN at baseline. Results are detailed in Table 3 and visually depicted in Fig. 3. R CeN out-degree statistically mediated the association of group status with within-person changes in positive urgency (Indirect Effect = 0.08; p = 0.04), such that CeN out-degree was lower in the BPD group (β = −0.20 p < 0.001) and was related to sharper within-person declines in positive urgency (β = −0.35, p = 0.03). For negative urgency, we found that the influence of R CeN out-degree on baseline level and within-person change in negative urgency depended on age: Hayes' index = 0.06 (p = 0.01) and −0.02 (p = 0.01), respectively. In younger participants, higher R CeN out-degree was associated with higher baseline negative urgency and greater within-person decreases in negative urgency over the one-year follow-up whereas in older participants the opposite pattern was observed (Fig. 3b–d). We note that all effects held when participant-wise mean framewise displacement was included as a nuisance covariate in the analysis (see online Supplementary Results for details).
Parameter table: dual-mediator latent growth curve model includes mediation by R CeN in- and out-degree
Note. Parameter table reflects parameter estimates estimated in and extracted from the lavaan R package (Rosseel, Reference Rosseel2012). Confidence intervals and corresponding p values were estimated based on standard errors derived from 1000 bootstrapped samples. Moderated mediation estimates are based on the index described in (Hayes, Reference Hayes2015).
Although our emphasis on emotion-related impulsivity was informed by clinical theory and prior BPD research, we also note that in preliminary mixed-effects models, we found a significant group by age interactions in levels of negative and positive urgency but not other facets of impulsivity (online Supplemental Methods and Results; Table S6).
Discussion
In a sample of adolescents and emerging adults with BPD symptoms and matched healthy controls, we found that resting-state EC of amygdala subnuclei at study baseline accounted for group differences in initial levels and within-person change in emotion-related impulsivity over a one-year follow-up period. Specifically, input to R CeN (in-degree) was significantly higher in the BPD group, which largely reflected a heightened influence of BLA on CeN. Conversely, in the control group, higher levels of output of R CeN (out-degree) were primarily attributable to R CeN's directed input to BLA (Fig. 1, online Supplementary Fig. S3). Crucially, the output of R CeN statistically mediated the association between BPD symptoms and baseline emotion-related impulsivity, as well as within-person stability of emotion-related impulsivity in borderline adolescents. Our results indicate that CeN plays an important role in impulsive behaviors in response to intense emotions. Importantly, whereas we found evidence of directed connectivity from vmPFC to CeN in controls, this functional connection was not reliably observed in BPD participants. This suggests an altered integration of cortico-limbic signals in the CeN, which has important clinical implications regarding the developmental course of BPD from adolescence to emerging adulthood.
Fronto-limbic disturbances are frequently reported in fMRI studies of adults with BPD, with the strongest evidence for amygdala hyperactivity in the processing of emotions. Further, prior evidence suggests that the functional interaction between vmPFC and amygdala is suppressed in individuals with BPD when presented with emotional stimuli (Kamphausen et al., Reference Kamphausen, Schröder, Maier, Bader, Feige, Kaller and Tüscher2013; Silbersweig et al., Reference Silbersweig, Clarkin, Goldstein, Kernberg, Tuescher, Levy and Stern2007). We found that EC from vmPFC to CeN was positive in controls, but absent in the borderline group. This positive connection in our data is in line with other BOLD EC studies showing positive EC from vmPFC to amygdala (Nawa & Ando, Reference Nawa and Ando2019; Stein et al., Reference Stein, Wiedholz, Bassett, Weinberger, Zink, Mattay and Meyer-Lindenberg2007; Watanabe, Bhanji, Tanabe, & Delgado, Reference Watanabe, Bhanji, Tanabe and Delgado2019), though we note that animal studies of fear conditioning find that mPFC/infralimbic projections to the amygdala are primarily inhibitory (Quirk & Gehlert, Reference Quirk and Gehlert2006; Quirk, Likhtik, Pelletier, & Paré, Reference Quirk, Likhtik, Pelletier and Paré2003; Rosenkranz & Grace, Reference Rosenkranz and Grace2001, Reference Rosenkranz and Grace2002). One possibility is that at rest, positive EC between these regions reflects an increased capacity of vmPFC to control affective responding in CeN (Watanabe et al., Reference Watanabe, Bhanji, Tanabe and Delgado2019). This interpretation aligns well with evidence that humans with vmPFC lesions show potentiated amygdala responsivity (Motzkin, Philippi, Wolf, Baskaya, & Koenigs, Reference Motzkin, Philippi, Wolf, Baskaya and Koenigs2015). Additionally, preclinical research indicates that electrical stimulation of mPFC neurons suppress the activity of output neurons in CeN that control autonomic/emotional arousal (Quirk et al., Reference Quirk, Likhtik, Pelletier and Paré2003). If functional connections between vmPFC and CeN are absent or significantly weakened in young people with BPD, excitatory connections from BLA to CeN (Tye et al., Reference Tye, Prakash, Kim, Fenno, Grosenick, Zarabi and Deisseroth2011; Zhang et al., Reference Zhang, Zhang, Holmes and Pan2021) could exert heightened influence over CeN efferents that control arousal.
While CeN encodes the general affective and motivational significance of emotional events (Balleine & Killcross, Reference Balleine and Killcross2006; Corbit & Balleine, Reference Corbit and Balleine2005), BLA is involved in assigning emotional significance to sensory stimuli (Janak & Tye, Reference Janak and Tye2015; Namburi et al., Reference Namburi, Beyeler, Yorozu, Calhoon, Halbert, Wichmann and Tye2015). Importantly, BLA projections to CeN control fear acquisition in the presence of environmental stressors and the expression of anxious behavior (Ciocchi et al., Reference Ciocchi, Herry, Grenier, Wolff, Letzkus, Vlachos and Lüthi2010; Namburi et al., Reference Namburi, Beyeler, Yorozu, Calhoon, Halbert, Wichmann and Tye2015; Tye et al., Reference Tye, Prakash, Kim, Fenno, Grosenick, Zarabi and Deisseroth2011), while BLA projections to VS are preferentially activated during reward learning (Namburi et al., Reference Namburi, Beyeler, Yorozu, Calhoon, Halbert, Wichmann and Tye2015). This is consistent with the interpretation that heightened BLA-to-CeN EC in BPD adolescents could support a tendency to initiate impulsive behaviors in response to threatening environmental conditions. Alternatively, BLA-to-CeN EC in BPD may reflect a tendency to translate sensory signals with greater emotional significance (BLA; Jovanovic & Ressler, Reference Jovanovic and Ressler2010; Kaye, Reference Kaye2021) to a state of enhanced physiological arousal (CeN). Ultimately, such high arousal states could predispose adolescents toward making impulsive choices, either to enhance positive affect or to escape negative affect.
Crucially, differential EC between the R CeN and BLA was associated with self-reported emotion-related impulsivity levels at baseline and within-person change in emotion-related impulsivity over 12-month follow-up (Fig. 3). Adolescence is associated with increased levels of impulsivity, which decrease in the general population, yet adolescents in the BPD group demonstrated greater stability in emotion-related impulsivity symptoms. Within-person decreases in negative urgency in healthy adolescents (Fig. 3b, lower left) were associated with stronger EC from R CeN to R BLA at baseline. Most studies find that functional interactions between BLA and CeN reflect the influence of BLA activity on CeN, rather than the other way around (Cardinal et al., Reference Cardinal, Parkinson, Hall and Everitt2002; Tye et al., Reference Tye, Prakash, Kim, Fenno, Grosenick, Zarabi and Deisseroth2011). Thus, our finding that R CeN output mediates within-person decreases in negative urgency may reflect an increased capacity to resist BLA modulation in controls (a capacity that may be supported by signaling from vmPFC). Given that EC from CeN to BLA was not observed in the BPD group, we propose that BLA control of the CeN is one candidate mechanism for explaining the stability of emotion-related impulsivity in BPD. This proposal extends earlier findings of fronto-limbic abnormalities in BPD by illustrating a more anatomically nuanced account of intra-amygdala information flow than previously described in this population. Most importantly, we demonstrate that intra-amygdala EC in our data predicted the clinical course of emotion-related impulsivity symptoms up to one-year post-scan.
To our knowledge, this is the first study in humans to demonstrate tradeoffs in EC between BLA and CeN in any psychiatric population. Furthermore, we leveraged recent developments in EC estimation (Henry et al., Reference Henry, Feczko, Cordova, Earl, Williams, Nigg and Gates2019) that are well-suited for detecting group-specific connectivity patterns. Our findings have clear implications for future study in other disorders with heightened levels of emotion-related impulsivity including addiction (Zorrilla & Koob, Reference Zorrilla and Koob2019), anxiety disorders (Pawluk & Koerner, Reference Pawluk and Koerner2016), eating disorders (Wenzel, Weinstock, Vander Wal, & Weaver, Reference Wenzel, Weinstock, Vander Wal and Weaver2014), and PTSD (Weiss, Tull, Sullivan, Dixon-Gordon, & Gratz, Reference Weiss, Tull, Sullivan, Dixon-Gordon and Gratz2015). As such, it is worth noting that dissociations in functional (undirected) connectivity of BLA and CeN have been previously documented in human subjects with these disorders (Brown et al., Reference Brown, LaBar, Haswell, Gold, Beall and Morey2014; Frank, Shott, Riederer, & Pryor, Reference Frank, Shott, Riederer and Pryor2016; Shackman & Fox, Reference Shackman and Fox2016; Zorrilla & Koob, Reference Zorrilla and Koob2019).
A few limitations are worth noting. First, our RS-fMRI acquisition was cross-sectional. Though we demonstrated the ability of a cross-sectional measure of brain connectivity to explain within-person changes in emotion-related impulsivity over 12 months, longitudinal neuroimaging studies offer the opportunity to study developmental changes in EC between BLA and CeN. Second, while relating self-reported emotion-related impulsivity levels to measures of intrinsic EC is an important descriptive step in identifying candidate neural circuits implicated in personality pathology, highly stable levels of emotion-related impulsivity in the BPD group may be due to selection criteria for our study. That is, emotion-related impulsivity stability in the BPD group may reflect a byproduct of selecting participants with heightened BPD symptoms (including impulsivity). Third, our study collected five minutes of resting-state fMRI data, which is on the low end of acceptable acquisition length (Birn et al., Reference Birn, Molloy, Patriat, Parker, Meier, Kirk and Prabhakaran2013). That said, our fast TR (1 s) provided us with 300 time points per subject, allowing for reliable estimates of EC using GIMME, which has been validated with as few as 200 time points (Gates & Molenaar, Reference Gates and Molenaar2012). Finally, future studies should include clinical comparison groups to clarify this specificity/generality of our findings, as differences in fronto-limbic connectivity have been documented across many disorders.
We present evidence that EC between vmPFC, CeN, and BLA is altered in adolescents and emerging adults with BPD symptoms. One speculative interpretation is that vmPFC and BLA compete in modulating CeN activity, and differential contributions of these regions to CeN activity underlie stability and change in emotion-related impulsivity symptoms from adolescence through emerging adulthood. Further, our findings demonstrate altered fronto-limbic connectivity in adolescents with BPD symptoms, including a functional disconnection between vmPFC and CeN and tradeoffs in control between functionally distinct subnuclei of the amygdala, which may underlie impulsive behaviors in the face of strong emotions. We hope that future studies build on our results by examining biomarkers that can inform treatments for adolescents at high risk for negative BPD-related outcomes in early adulthood and beyond.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291722000101
Acknowledgements
This work was funded by the National Institutes of Mental Health (K01 MH097091 and R01 MH119399 to MNH). The funding agency had no role in the design and conduct of the study; the collection, management, analysis, and interpretation of the data; the preparation, review, and approval of the manuscript; or the decision to submit the manuscript for publication.
Conflict of interest
The authors have no financial or other conflict of interest to disclose.
