Decreased brain network global efficiency after attachment memories retrieval in individuals with unresolved/disorganized attachment-related state of mind

The main aim of the study was to examine how brain network metrics change after retrieval of attachment memories in individuals with unresolved/disorganized (U/D) attachment-related state of mind and those with organized/resolved (O/R) state of mind. We focused on three main network metrics associated with integration and segregation: global (Eglob) efficiency for the first function, local (Eloc) efficiency and modularity for the second. We also examined assortativity and centrality metrics. Electroencephalography (EEG) recordings were performed before and after the Adult Attachment Interview (AAI) in a sample of 50 individuals previously assessed for parenting quality. Functional connectivity matrices were constructed by means of the exact Low-Resolution Electromagnetic Tomography (eLORETA) software and then imported into MATLAB to compute brain network metrics. Compared to individuals with O/R attachment-related state of mind, those with U/D show a significant decrease in beta Eglob after AAI. No statistically significant difference among groups emerged in Eloc and modularity metrics after AAI, neither in assortativity nor in betweenness centrality. These results may help to better understand the neurophysiological patterns underlying the disintegrative effects of retrieving traumatic attachment memories in individuals with disorganized state of mind in relation to attachment.


Materials and methods
Participants. The enrolment procedure lasted from October 2017 to May 2018 and was performed by posting advertisements around the university campus. Participants aged between 18 and 30 years were included in the study. Exclusion criteria were: (i) left-handedness, (ii) history of neurological and/or psychiatric disorders, (iii) head trauma, (iv) having taken Central Nervous System active drugs in the two weeks prior to experiment participation. Socio-demographic data, and inclusion/exclusion criteria were assessed using a checklist of dichotomous items.
We chose a sample of 50 individuals which is almost double the number used in a previous EEG study combining cortical connectivity with AAI 20 and which showed acceptable power even to detect large size differences. Given the low prevalence (i.e., approximately 17%) of U/D individuals in student samples 13 and the time and expense associated with administering the AAI [51][52][53] , 146 participants were pre-screened for quality of parenting using the Measure of Parenting Style (MOPS 54 ), a self-report questionnaire assessing mother' and father' perceived parental styles (i.e., indifference, over-control, and abuse) during their first 16 years of life. According to the empirical data showing the association between dysfunctional parenting and unresolved/disorganized attachment 55 , in order to identify two groups with a similar ratio of U/D and O/R participants, of the 146 individuals pre-screened, we enrolled for the experimental session (i.e., EEG recordings and AAI) a convenience sample of 50 participants (36 females, mean age = 22.62 ± 2.41 years) scoring in the lowest (N = 25) and highest (N = 25) portions of the MOPS scores distribution. Specifically, the 146 participants were divided into two groups based on the median scores of the MOPS maternal and paternal scales (6 and 5, respectively). Sixty-three participants (i.e., 43.2%) who scored both ≥ 6 and ≥ 5 were classified as having highly dysfunctional parenting. In contrast, eighty-three (i.e., 56.8%) individuals were classified as having low dysfunctional parenting. Finally, participants from the lowest (i.e., N = 25) and highest portions (i.e., N = 25) of the MOPS score distribution were randomly www.nature.com/scientificreports/ selected. These participants are the same involved in our previous EEG study 56 . All participants provided their written informed consent. The study was conducted according to the Helsinki declaration standards and the research project was approved by the European University's ethics review board.
Experimental procedure. The experiment was conducted on two different days. On the first day, after providing written informed consent, participants completed an inclusion/exclusion criteria checklist (already used in 57 ) and a battery of self-report questionnaires in their Italian version, including: the Brief Symptom Inventory  (BSI 58 ), the Trait-version of the State-Trait Anxiety Inventory (STAI-T 59 ), and the Beck Depression Inventory-II  (BDI-II 60 ). On the second day of the experimental procedure, each participant underwent the AAI 61 . For each subject, two eye-closed resting-state (RS) EEG recordings have been performed prior and after AAI (i.e., pre-AAI and post-AAI). The whole experimental procedure was carried out in the Cognitive and Clinical Psychology Laboratory of the European University of Rome. The BSI 62 is a 53-item inventory (Likert scale ranging from 0 = not at all to 4 = extremely) assessing general psychopathological symptoms experienced in the last 7 days. This inventory provides 9 symptomatologic dimensions and a general measure of psychological distress (i.e., Global Severity Index, GSI). The Cronbach's alpha in the current sample was 0.94 for the GSI.
The STAI-T 63 is a widely used self-report inventory to evaluate trait-anxiety composed of 20 items rated on a 4-point Likert scale (1 = almost never; 4 = almost always), with total score ranging 20-80 (higher scores reflect higher levels of trait anxiety). The Cronbach's alpha in the current sample was 0.92 for the STAI-T total score.
The BDI-II 64 is a self-report assessing depressive symptoms in the 2 weeks prior to administration. It is composed by 21 items on which responses are rated on a 4-point Likert scale which ranges from 0 to 3 (0 = not at all; 3 = extremely), such that higher scores reflect higher depressive symptoms (total score range = 0-63). In the present sample, the Cronbach's alpha for the total score was 0.93.

Adult Attachment Interview (AAI).
The AAI 61 is a semi-structured interview which requires the subject to retrieve memories of attachment experiences occurred during childhood (i.e., emotional and relational) triggering the activation of attachment system 10,41,61 . This clinical interview has been developed to evaluate the attachment related current state of mind respect to attachment experiences with caregivers 9 . Previous rigorous works (i.e., meta-analyses, psychometric studies) on the AAI demonstrated its good psychometric properties (i.e., stability, predictive and discriminant validity) in both non-clinical and clinical samples 65,66 . Accordingly, in the current study, the AAI has been used both for classifying individual's attachment-related state of mind in terms of U/D vs. O/R, and to trigger the activation of the attachment behavioral system. The AAI was administered by a trained clinical psychologist in a quiet and comfortable room, and the average time for each interview was approximately 1 h and 30 min.
In the current study, AAI transcripts were coded by means of the Main et al. 67 coding system by an experimenter certified (R.B.A) as reliable by Mary Main and Erik Hesse, University of Berkeley, California. As in previous works (e.g. 68 ), 20% of the AAIs (N = 10) were double-coded by another certified AAI coder (L.P.) who was also trained and certified as reliable by Mary Main and Erik Hesse. Inter-rater agreement for the two-way analysis (U/D vs. O/R) was 100%, kappa = 1.00; inter-rater agreement for the four-way analysis (Free/Autonomous, Dismissing, Entangled/Preoccupied, Unresolved and/or Cannot Classify) was 90%, kappa = 0.80. All coders were blind to all other measures and participant information. Disagreements between coders for the two-way analysis were resolved through conferencing until consensus was reached.
EEG recordings and processing. The EEG data, acquisition procedure, and processing methods used in the current study are described in Adenzato et al. 56 . For each participant, eye-closed RS EEG recordings have been performed before and immediately after (i.e., 8-10 min for each person who takes part to the experiment) the interview, corresponding to pre-AAI and post-AAI condition, respectively. RS EEG was performed in University EEG Lab allowing rest condition (i.e., comfortable armchair, semi-darkened and isolated from noise room). Considering the possible effects of alcohol and/or caffeine on EEG traces, participants were asked not to drinking alcohol or caffeine before EEG recording session (i.e., 4/6 h before).
EEG recordings were performed with 31 standard scalp electrodes according to the 10-20 system [for detailed information about recording sites, reference electrodes, impedances, sampling frequency, band-pass filters see 56]. Artifact rejection procedure was described in detail in a previous work of our team 57 . For each participant and for each condition (i.e., before and after AAI), we consider at least 180 s of artifact-free EEG trace even if not consecutive. The Micromed System Plus digital EEGraph (Micromed© S.p.A., Mogliano Veneto, TV, Italy) was used for all EEG recordings.

Functional connectivity and network construction.
In the current study, the Fast Fourier Transform (FFT) was computed by means of the exact Low-Resolution Electromagnetic Tomography (eLORETA) software for frequency decomposition of the EEG trace considering the following frequency bands: delta (0.5-4 Hz); theta (4.5-7.5 Hz); alpha (8-12.5 Hz); beta (13-30 Hz); gamma (30.5-60 Hz). EEG connectivity analysis was performed using the eLORETA software, a validated tool for localizing brain electric activity based on multichannel surface EEG recordings 69 , which is characterized by satisfactory localization agreement with different multi-modal imaging techniques (e.g. [70][71][72]. The regularization parameter for eLORETA was α ≥ 0. According to previous studies [73][74][75] , by means of eLORETA and according to the Montreal Neurological Institute (MNI) atlas 76 , we computed connectivity across all 84 Regions of Interest (ROIs), 42 for each hemisphere, in order to have a complete view of the brain connectivity 73 . Specifically, we chose the "ROI-maker#2 method" available in the eLORETA software selecting 42 distinct Brodmann areas (BAs) for the left and right hemispheres. www.nature.com/scientificreports/ As recommended 77,78 , the "single nearest voxel" option (i.e., each ROI consisted of a single voxel, the one that is closest to the center of mass of the ROI) was chosen and the neuroanatomical MNI coordinates were computed by the eLORETA software (the table of all considered ROIs with the relative coordinates is reported in Supplementary Table 1). Then we computed the lagged phase synchronization (LPS) to obtain connectivity matrices. The LPS is a measure of "similarity of two time series by means of the phases of the analyzed signal" 79 and its values range 0-1, respectively for no synchronization and maximum one. Connectivity matrices were used for network construction: an 84-by-84 functional network was constructed for each frequency band for each subject. More in detail, we imported connectivity weighted and undirected matrices for each subject and for each frequency band into MATLAB. These matrices reflected connectivity among 84 ROIs which represented the nodes of the network and the edges of the network were weighted by LPS connectivity. Figure 1 shows an overview of the EEG recording, processing, and network analysis procedure.

Network analysis.
After network construction, we used a topological threshold, namely "ECOfilter" 80 in order to obtain binarized matrices on which we computed graph theoretical metrics. Specifically, as principal analyses, we computed (i) E glob as a measure of functional integration 32 , (ii) E loc and (iii) modularity (Q) as measures of network functional segregation 32,36 . In addition, as exploratory analyses, we computed (iv) assortativity as a measure of resilience 32,49 , and (v) betweenness centrality (BC) as a measure of node importance 32,50 . E glob estimates the efficiency of distant information exchanging in the network. It is inversely correlated with the topological distance between nodes and is described as the numerical average of the inverse shortest path length between all pairs of nodes in the network 28,32 . Therefore, higher E glob values reflect higher capability of parallel information exchanging and higher integration of processes 81 . E loc is a measure related to the clustering coefficient and described as the E glob calculated on the neighborhood of nodes 28,32 . In the context of complex functional networks such as the brain, E loc is considered a measure of www.nature.com/scientificreports/ functional segregation 32 . In the current study, the local efficiency of the entire network (E loc ) was computed as the average of the local efficiency over all nodes in the network. Modularity (Q) is a metric that indicates the extent to which it is possible to subdivide the network into nonoverlapping and defined groups of nodes [29][30][31] , and it is considered a global measure of network segregation 32 .
Assortativity is a measure of the extent to which network nodes with many connections are more likely to be linked to nodes with low (disassortative networks) or many (assortative networks) connections 49 . Accordingly, highly assortative networks seems to be more resilient, while disassortative ones (e.g., biological ones) less, thus being more vulnerable 49 . In an assortative network "the nodes (…) many connections tend to be connected to other nodes with many connections" [49; p. 208,701-1]. Thus, the assortativity coefficient (A c ) 49 can be seen as "a correlation coefficient between the degrees of all nodes on two opposite ends of a link" ( 32 ; p. 1065) with a positive A c indicating a more resilient network and a negative one indicating a more vulnerable system 32,49 .
Lastly, the BC is a network metric characterizing the importance of nodes in bridging different parts of the network 50 . Mathematically, it is "the fraction of all shortest paths in the network that pass through a given node" [32; p. 1064]. Accordingly, a high BC is often associated to nodes with a function in connecting distant network's elements 32 . Specifically, in the current work, this metric has been computed as the average of all the nodal BC values. Thus, higher BC values indicate that, on average, more nodes in the network are part of more shortest paths. Conversely, lower BC values indicate that lesser nodes in the network are part of a lower number of shortest paths.

Statistical analyses.
According to a previous study 19 two groups were defined on the basis of their state of mind organization in relation to attachment: individuals with organized state of mind in relation to attachment (O/R), and individuals with disorganized organized state of mind in relation to attachment (U/D). Differences between groups for clinical and sociodemographic variables were investigated using independent t tests and chisquared (χ 2 ) tests for dichotomous and dimensional measures, respectively.
To investigate the effect of retrieval of attachment memories on brain network metrics across frequency bands in U/D vs O/R participants, we used analysis of covariance (ANCOVA). Specifically, the brain network metrics for each frequency band at T1 (i.e., post AAI) were considered as dependent variables, with group (i.e., O/R vs U/D participants) as a between-subject factor. Network metrics for each frequency band at T0 (i.e., pre AAI) were included as covariates to control for baseline EEG values (see for example [82][83][84]. Due to the different network properties (i.e., small-worldness) showed by males and females in EEG RS networks 85 , we also added gender as a further covariate. Finally, in the case of differences in age and measures of general psychological distress (GSI), depression, or anxiety, we add these variables as additional covariates. E glob , E loc , and Q ANCOVAs were performed separately, applying a formal Bonferroni correction for each family of comparisons by dividing the limit of significance by the number of comparisons (i.e., five frequency bands: delta, theta, alpha, beta and gamma). Thus, the threshold level for significance was p = .05/5 = .01. To determine the effect size, Cohen's d ppc2 values were calculated ( 86 ; equation number 8). Similarly, also for A c and BC the same statistical analyses have been performed. All statistical analyses were performed using SPSS (version 18.0).

Results
AAI coding shows that, of the total sample (i.e., n = 50), n = 29 individuals were classified as Organized/Resolved (O/R = 58%; 22 females) and n = 21 individuals were classified as having Unresolved/Disorganized (U/D = 42%; 14 females) state of mind in relation to attachment. Indicators of state-of-mind disorganization in relation to attachment or unresolved state of mind were associated with experiences of abuse and/or loss in relation to primary attachment figures. In the U/D group, n = 19 interviews were classified primarily as Unresolved (U), and n = 2 were classified as Cannot Classify (CC). In the O/R group, n = 8 were classified primarily as Free/Autonomous, n = 16 as Dismissing, and n = 5 as Entangled/Preoccupied. Sociodemographic and clinical data for both groups are shown in Table 1. The groups did not differ with respect to gender and education level, but they did differ with respect to age (p = .022). The groups did not differ with respect to other clinical variables such as general psychological distress (GSI), depression (BDI) and trait anxiety (STAI-T). As expected, the U/D group had a higher MOPS total score compared to the O/R group (p = .009).  Fig. 2 shows the averaged network matrices in the two conditions considered (i.e., pre-AAI and post-AAI) for both groups (i.e., O/R and U/D). The connectivity weighted matrices for both groups, conditions, and all frequency bands are shown in Supplementary Fig. 1. As regarding brain network measures, ANCOVA analysis (Table 2) shows that groups significantly differ in E glob metric post-AAI in beta frequency band [O/R group = .152 ± .040 (M ± SD); U/D group = .125 ± .031 (M ± SD); F T1 (1;49) = 7.60; p = .008]. ANCOVA analysis showed no significant differences between groups for either E loc (Table 3) and the Q metric (Table 4). We also did not find statistically significant differences between groups for the A c and BC metrics. Only a tendency near to significance has emerged for the BC metric in the beta frequency band. Full data on A c and BC results are presented in Supplementary Tables 2 and 3.

Discussion
The aim of the current study was to examine changes in network metrics across frequency bands following exposure to AAI in individuals with organized and disorganized attachment-related state of mind. To our knowledge, this is the first study to examine attachment-related brain network metrics following activation of affective and cognitive processes associated with attachment.
Two main results have emerged. Partly in agreement with our hypothesis, the results show that, after exposure to AAI, compared with individuals with O/R state of mind, the individuals with U/D showed reduced global efficiency in the beta frequency band. There were no statistically significant differences in the other frequency bands examined. Contrary to expectations, there were no significant differences in local efficiency and modularity when comparing the groups after AAI. Similarly, no statistically significant differences emerged for the metrics of assortativity and betweenness centrality, suggesting that activation of attachment memories is more likely to affect cortical connectivity for the integrative network metric than for the metrics of segregation, resilience, and centrality. These results appear particularly interesting given the absence of differences in clinical variables (i.e., BDI, STAI-T, and GSI scores). In other words, our results suggest that U/D state of mind in absence of psychiatric  www.nature.com/scientificreports/ diagnosis or psychopathological conditions could affect integration but not segregation and alter optimal mental functioning based on the balance between segregation and integration of information 33-35 . In agreement with previous studies 20, 56 , these results suggest that AAI is a very demanding task that requires highly integrated mental activity in terms of cognitive and affective processes and an increase in global efficiency that may be associated with increased widespread functional connectivity in healthy individuals. Since global efficiency is considered a measure of network integration 27 , our results may shed light on pathogenic functional network processes related to activation of attachment memories in individuals with U/D state of mind without psychiatric disorder.
It has been reported in the literature that high-frequency oscillations (i.e., beta and gamma oscillations) and, in particular, dynamic, large-scale brain networks operating at these frequencies appear to be involved in higherorder cognitive processes, including sensorimotor integration, consciousness, attention, memory, and affective processes [42][43][44][45][46] . Indeed, it is well known that high globally efficient information exchange across all the network and functional connectivity would promote highly integrated mental processes such as emotional awareness 87 and emotion regulation 44,88 , executive cognitive functions 81 , social cognition and mentalization 89 and self-related mental processes such as autobiographical memory, self-consciousness and its properties of self-coherence, self-continuity, self-embodiment, and sense of agency 88,90 . According to this literature, our finding seems to suggest that the decrease in cortical global efficiency observed in the U/D group may reflect state-dependent impairments in higher-order affective, cognitive, and metacognitive processes triggered by traumatic attachment memories of AAI and affecting mentalization, autobiographical memory, and emotion regulation. Consistently, U/D state of mind is associated with high-level alterations in integrative mental functioning, leading to impairments in cognitive, meta-cognitive and emotional regulation 11,20,56,91 . Indeed, individuals with U/D attachment-related state of mind show signs of impaired mental integration when discussing potentially traumatic experiences (e.g., transient alterations in consciousness) 9,92 .
From a clinical point of view, several authors hypothesized that the repeated threatening and contradictory interactions of attachment disorganization could lead to a simultaneous and competitive activation of the attachment and defensive systems at the same moment towards the same person 10,16,93 , resulting in an intersubjective process that hamper the integrated and organized development of mental functions 15,94 . Therefore, it has been suggested that, in individuals with U/D states of mind, activation of attachment memories may act as a traumatic trigger for disintegrative processes that impair optimal functioning of higher-order mental functions in ways that are not different from other trauma-related pathogenic processes involving executive and regulatory functions, self-related processes, social cognition, and mentalization 11,20,56,90,95 .
The present results seem to be in agreement with previous studies reporting altered integration between cortical and cortical-subcortical structures in clinical samples 1,20,23 . However, there are some differences worth investigating.
For example, a previous EEG study on functional connectivity 20 reported that dissociative patients had lower functional connectivity among widespread cortical regions in high frequency bands after exposure to AAI, compared to controls. This suggests that this pattern may reflect poor integrative mental processes required during retrieval of attachment-related memories, which are known to be impaired in these patients, but also more generally in individuals with a disorganized state of mind in relation to attachment 9,10 . Nevertheless, in our study we did not observe changes in segregation metrics, that we expected to be indicative of dissociation. It could be argued that our sample in the present study did not show clinical symptoms of dissociation and that the possible relationship between segregation changes and dissociation should be investigated in clinical samples with compartmentalization symptoms 96 . Indeed, Liotti 97 hypothesized that the threatening and paradoxical intersubjective experience of disorganized attachment may constitute only a predisposition to later dissociation regarded as compartmentalization of self-experiences and that other subsequent and long-lasting traumatic experiences after early disorganized attachment are necessary for clinically relevant dissociative symptoms to develop. This hypothesis has been supported by controlled longitudinal studies 98,99 . We, therefore, argue that in our non-clinical sample disorganized attachment-related memories temporarily impaired global integrative capacity without affecting the more stable segregation functioning that likely plays a role in dissociative disorders 11 .
Although potentially interesting, our study has some limitations related to EEG, such as the small number of electrodes considered and the intrinsic limit of EEG, which is known to have poor spatial resolution. Future studies should overcome these limitations and investigate integrative/disintegrative processes and, more generally, the topology of functional brain networks related to attachment by considering a broader range of network metrics, including nodal ones to gain a more comprehensive insight into the functioning of large-scale networks in individuals with different attachment-related states of mind after activation of the attachment system. Moreover, future studies should either replicate or discard these data in larger and clinical (e.g., patients with borderline personality disorder) samples. In conclusion, even if there are some aspects limiting the generalizability of these findings, to our knowledge, this is the first study exploring network metrics such as those of efficiency following exposure to AAI in individuals with disorganized and organized state of mind in relation to attachment. Our results show that, compared to individuals with O/R attachment-related state of mind, those with U/D have lower global efficiency after exposure to AAI suggesting poor network integration. These data may help to better understand the neurophysiological patterns underlying the disintegrative effects of retrieving traumatic attachment memories in individuals with disorganized state of mind in relation to attachment.

Data availability
The data that support the findings of this study are available from the first author upon request.