A coordinate-based meta-analysis of human amygdala connectivity alterations related to early life adversities

By affecting core neurobiological systems early in development, early life adversities (ELAs) might confer latent vulnerability to future psychopathologies. This coordinate-based meta-analysis aims to identify significant convergent alterations in functional connectivity of the amygdala related to ELAs across resting-state and task-based fMRI-studies. Five electronic databases were systematically searched until 22 October 2020, retrieving 49 eligible studies (n = 3162 participants). Convergent alterations in functional connectivity related to ELAs between the amygdala and the anterior cingulate cortex (ACC) and left hippocampus were found. Sub-analyses based on hemisphere and direction showed that connectivity seeded in the right amygdala was affected and, moreover, revealed that connectivity with ACC was decreased. Analyses based on paradigm and age showed that amygdala-ACC coupling was altered during resting state and that amygdala–left hippocampus connectivity was mostly affected during task-based paradigms and in adult participants. While both regions showed altered connectivity during emotion processing and following adverse social postnatal experiences such as maltreatment, amygdala-ACC coupling was mainly affected when ELAs were retrospectively assessed through self-report. We show that ELAs are associated with altered functional connectivity of the amygdala with the ACC and hippocampus. As such, ELAs may embed latent vulnerability to future psychopathologies by systematically affecting important neurocognitive systems.

prioritized investigating the effects on amygdala coupling to specific regions of interest and thereby potentially overlook broader effects on a whole-brain level (e.g.Ref. 3 ).In this study, we therefore conducted a quantitative summary of individual findings on a whole-brain level to capture these heterogeneous findings.By pooling data from multiple studies and analysing coordinates of altered connectivity, we aimed to comprehensively evaluate the variability and convergence of findings across different regions to identify consistent patterns of affected amygdala connectivity in relation to ELAs.Following current recommendations for coordinate-based Figure 1.ALE analyses for functional alteration in connectivity with the amygdala as seed-region revealed two significant clusters of convergence, with the first within the (A) ACC (BA24), with peaks at [− 6, 36,  8], ALE = 0.0180 and at [2, 38, 10], ALE = 0.0167, and the second cluster within the (B) left hippocampus ([− 30, − 26, − 10], ALE = 0.0236).The ALE map was computed in MNI152 and thresholded (p cluster-level < 0.05, p uncorrected < 0.001).L = left, R = right.ALE analyses were conducted using GingerALE 3.0.2(https:// www.brain map.org/ ale/) and results were visualized using Mango 4.1 (https:// mango viewer.com/).www.nature.com/scientificreports/meta-analyses (CBMAs) [34][35][36] , this study is the first to decipher the overall effect of ELAs on amygdala network connectivity.In this context, we consider ELAs more broadly as developmental risk factors acting early in life and therefore included both prenatal exposures (e.g.substance exposure) and postnatal experiences (e.g.childhood maltreatment or poverty).In line with our previous meta-analyses on neural alterations related to ELAs 4,5 , we performed a coordinate-based meta-analysis using activation likelihood estimation (ALE; for an explanation, see Ref. 37 ) 36,38,39 to consolidate this yet inconclusive literature and assess robust effects of altered amygdala connectivity across samples and analytic approaches.Based on the seed-based amygdala connectivity literature (for a recent review, see Ref. 3 ), we hypothesize that particularly the amygdala-PFC circuit and amygdala-hippocampus circuit are affected in relation to ELAs.As such, these functional connectivity phenotypes might serve as stratification markers for early detection and treatment of the potentially long-lasting effects of ELAs throughout life.

Results
The global ALE analysis on the overall effect of ELAs was based on 45 experiments (3162 participants) and revealed convergence within the anterior cingulate cortex (ACC; L ACC: ALE-value: 0.0180, Z-score: 3.99; R ACC: ALE-value: 0.0167, Z-score: 3.78) (Fig. 1A, Table 1) and the left hippocampus (ALE-value: 0.0236, Z-score: 4.76) (Fig. 1B, Table 1).Additional sub-analyses were performed if a sufficient number of experiments (n ≥ 17) was available 35 .Separating the analyses by paradigm (resting-state or task-based) indicated convergent altered coupling of the amygdala with the ACC during resting-state (25

Discussion
Our meta-analysis provides a comprehensive overview of neuroimaging studies investigating the association of ELAs with brain connectivity using the amygdala as seed-region.We have demonstrated robust evidence for decreased amygdala-ACC and altered amygdala-hippocampus connectivity in connection with ELAs, that support a certain level of equifinality of ELAs in these neural adaptations 40 .As such, this study complements previous meta-analyses reporting neural alterations associated with ELAs 4,5,41 and offers more insight into the mechanisms of how ELAs might become embedded into the human brain.The overall analysis revealed significant alterations in functional connectivity related to ELAs between the amygdala and the ACC, as well as between the amygdala and the left hippocampus-which is in line with previous studies [3][4][5]18,[42][43][44] . Subsequet sub-analyses were applied to further specify these results by restricting them based on paradigm, direction, hemisphere, age, disease status, ELA-subtype or -assessment.
For the ACC, these sub-analyses revealed a predominant decrease in amygdala-ACC connectivity mainly arising from the right amygdala during resting-state related to postnatal, social ELAs in healthy participantsespecially when these ELAs were retrospectively assessed through self-report.This has several implications.First, www.nature.com/scientificreports/as the ACC can be recruited to inhibit negative emotional processing in the amygdala 14 , decreased connectivity between the amygdala and ACC might indicate that the effectivity of this emotional stress resolution is affected in relation to ELAs-which may result in altered stress reactivity 22,[45][46][47] and memory extinction 48 .Second, as the right amygdala is mainly responsible for global, dynamic detection of (negative) emotional stimuli due to a faster habituation rate [49][50][51][52] , this further suggests that ELAs are related to altered automatic, emotional stress responses, which may lead to a more sustained emotional responding 49,53 .This aspect is corroborated by the fact that altered amygdala-ACC connectivity associated with ELAs primarily arises during the resting-state paradigm, a measure of intrinsic brain connectivity 29 .Overall, the altered resting-state connectivity pattern seems to be consistent across developmental stages, which is in line with previous studies 23 .Moreover, alterations in functional connectivity between the amygdala and ACC, as well as the hippocampus for that matter, are particularly associated with postnatal social ELAs, such as maltreatment.This intuitively makes sense: severe and prolonged trauma experiences can have severe (neural) consequences, and are correlated in space and time-a line of reasoning that is supported by literature 3,54 .It must however be taken into account that most of the studies investigated postnatal social ELAs and that the planned analysis for either socio-economic ELAs or prenatal exposures could not be performed due to an insufficient number of experiments.Lastly, the observation of altered amygdala-ACC connectivity primarily in individuals who self-report ELAs, might reinforce the hypothesis that such alterations contribute to enhanced cognitive biases that intensify the subjective evaluation of ELAs.This complements recent findings on different risk pathways for prospective and retrospective assessments 55 , and a superiority of the impact of subjectively experienced ELA-burden on the development of psychopathology 56 .
For the hippocampus, the sub-analyses further specified the results to an altered connectivity between right amygdala-left hippocampus.This alteration was predominantly observed in task-based experiments involving healthy adults who experienced postnatal, social ELAs.Interestingly, contributors to the task-based effect mainly employed an emotion processing paradigm, which further implicates this link between amygdala and hippocampus in emotional memory processing 21 .It also raises the assumption that ELAs might foster vulnerability to future psychopathologies via increased emotional memory consolidation 21,22 .The observation that this altered connectivity in relation to ELAs mainly arises in adult participants parallels previous literature showing hippocampal alterations in adults, but not in children 57 .Given the positive correlation between ELAs and future stress throughout life 58 , it might be that this cumulative effect of stress only manifests itself in adulthood-probably mediated by chronic stress-induced hippocampal glucocorticoid exposure 33,47,[59][60][61] .
Multiple theoretical frameworks have tried to capture the range of relationships between ELAs and neural adaptations.These include the latent vulnerability framework 2 , implicating that changes in neurocognitive systems in relation to ELAs, reflecting an adaptation to these negative early environments, alter one's vulnerability to future mental health problems.In later life, exposure to stress or challenge might unveil these vulnerabilities, thereby manifesting as clinical symptoms.Furthermore, the allostatic load model 56 suggests that intense and enduring exposure to adversities can disrupt the body's ability to maintain homeostasis, resulting in a dysregulation of the stress response 56 .In addition, the cumulative stress model 62 emphasizes the accumulation of stressors over time, implying that the combined effect of multiple stressors can have a significant impact on health outcomes.Overall, these models are not mutually exclusive in conceptualizing the complex and currently incompletely understood nature and consequences of ELAs, and instead might influence and complement each other.For example, chronic stress exposure during childhood can alter the allostatic load of physiological systems, such as amygdala-hippocampal coupling.These changes may act as latent vulnerabilities, thereby modifying responses to future stressors 2 and in turn impacting the allostatic load even more.Another conceptualization is the dimensional model of adversity, implicating that different stressors, such as threat-and deprivation-related ELAs, might act on qualitatively different mechanisms to increase the risk for specific psychopathologies 20,[63][64][65] .While intriguing, this framework remains to be elucidated by future meta-analyses, given that most of the included studies encompass multi-faceted adversities with only few studies making use of this suggested dimensional approach 63 .Importantly, in order to establish a clear, coherent and consistent model to conceptualize the neural adaptations in relation to ELAs in its entirety, the understanding of the exact mechanisms by which ELAs impact the brain should be advanced in future neurodevelopmental studies.
It is important to highlight the heterogeneous nature of ELAs.It comprises many exposures that are often interlinked with each other and share commonalties 3,66 .As such, the ultimate effect of ELAs reflects an intricate interplay between the exposure and one´s characteristics, such as genetic make-up and personality factors, and its socioenvironmental embedding, e.g.social support 67 .So while we aimed to stratify the results on amygdala connectivity as much as possible, we must acknowledge that-by following stringent guideline criteria for inclusion 34,35,68 -the number of available experiments was insufficient for further identification of potential moderators, such as sex, other subtypes including psychopathology, timing of ELAs, prenatal ELAs or any specification of direction within the sensitivity analyses.Concerning prenatal ELAs, while their exclusion yielded no significant result in the pooled analysis, further direction-, task-, and sample-specific sub-analyses do reveal the same convergence clusters in the ACC and the hippocampus (Table 1), supporting the robustness of the presented results.We also acknowledge that the effects-including the potential teratogenic effects 69 -of prenatal substance exposure, such as marijuana or cocaine, on the developing brain may differ from those of other ELAs, like maltreatment or poverty.However, as both types of ELAs are linked to changes in similar brain areas 4,70,71 , as well as to altered amygdala connectivity 72 , this seems to suggests some level of equifinality.Future studies should further look into this.Of note, a potential limitation might be that no correction for multiple testing across all sensitivity analyses was performed, as it was considered as too conservative for this purpose.
Furthermore, due to the heterogeneity and interrelatedness of the different types of ELAs included in this meta-analysis, our findings may rather point to the amygdala as being a nosologically unspecific network hub targeted by many kinds of adversities with effects being present independent of specific samples.Thus, its affected connections to the hippocampus and ACC reflect alterations suggestive of transdiagnostic phenotypes that may imply a latent vulnerability signature, which unfolds during system-challenging stressful situations 2 .This understanding is well in line with therapeutic studies that show neurotrophic changes in the amygdala following electroconvulsive therapy 73,74 across (patient) samples or other methodological considerations 74 .In light of this, we speculate that these recalibrations of amygdala connectivity as reported here might represent shared mechanisms of ELAs, that may be considered as a transdiagnostic risk correlate.
Given the recent discussion on different risk pathways dependent on ELA assessment 55,75 , the number of studies permitted a separate analysis on subjective and retrospectively assessed ELAs.However, the specification of neural embedding of ELAs that were either objectively or prospectively assessed was not possible and should therefore be focused in future meta-analyses.As the convergent effects reported in our work were mainly driven by findings from healthy participants and are thus not confounded by psychopathology, they should be assumed to reflect latent vulnerability signatures as similar alterations have been reported in clinical populations 76 .However, we cannot rule out that these altered neural phenotypes in relation to ELAs might be unrelated to psychological functioning altogether nor that they might reflect compensatory mechanisms supporting adaptive functioning later in life.After all, such recalibrated responses may be either adaptive or maladaptive based on environmental conditions 6 .
Previous studies indicate that amygdala-prefrontal connectivity develops with age, with the occurrence of a valence shift in task-related amygdala-prefrontal connectivity around the age of 10 years 77 .We therefore ensured that our result of decreased amygdala-ACC coupling still holds when participants under 10 years of age are excluded (see Table S5)-even though it must be noted that the literature is not consistent in reporting such agerelated alterations of functional connectivity [78][79][80] .To further investigate the normative developmental pattern of this circuit, as well as to clearly delineate whether the neural recalibration associated with ELAs-mostly seen in children and adolescents-presents a delay of maturation or an acceleration, longitudinal studies are warranted 5 .
Taken together, our current meta-analysis provides robust evidence for decreased amygdala-ACC and altered amygdala-hippocampus connectivity in relation to ELAs.These results are in line with previous research (for a recent review, see Ref. 3 ) and fits well within the theoretical framework of latent vulnerability 2 .This inherent neural plasticity to environmental exposures also holds a promise, as it might potentially enable normative recalibration and thereby promote resilience 67,[81][82][83] .While initial evidence does exist in relation to the reversibility of the structural and functional alterations associated with ELAs 4,67,84 , future (longitudinal) studies should examine this neural malleability in light of potential therapeutic interventions.

Methods and materials
This meta-analysis was preregistered with PROSPERO (CRD42018107773) and integrates all neuroimaging studies on the relation between ELAs and task-specific and resting-state brain connectivity using the amygdala as seed region.The study was conducted according to the PRISMA guidelines and current consensus guidelines for CMBAs [34][35][36] .
All screening, evaluation and data extraction procedures were performed by two independent authors (EK, NH) to reduce the chances of selection bias, and disagreements were resolved by consensus.The review protocol and data can be accessed upon request.

Search strategy and study selection
Relevant articles published until October 2020 were identified through a comprehensive literature search using five databases (EMBASE, MEDLINE, PsychINFO/PsychARTICLES, Scopus and Web of Science).Search strategies were composed of the search terms 'neuroimaging' or 'MRI' AND 'preterm birth' , 'prenatal exposure' or 'adverse childhood experience' with associated synonyms, using the keywords appropriate to each individual database (for full search terms, see Supplementary Table 1).In line with the definition of ELAs as deviations from the expected environment that require adaptation 85 and that brain development does not start at birth but at conception 86 , we conceptualized ELA in this manuscript as developmental risk factors acting early in life and therefore included both pre-and postnatal exposures and specified the analyses in further steps (see below).Additional articles were identified by reference tracking of all included studies and consultation of relevant review articles 17,19,71,87,88 .
Studies were selected if (1) peer-reviewed, original articles were published in English language; (2) human brain connectivity was measured using the amygdala as seed-region; (3) prenatal exposures and/or postnatal experiences were assessed; and (4) whole-brain results with stereotactic coordinates were reported or if not, provided by the authors.As such, from the 7195 unique publications that were initially identified, 119 publications were included in the qualitative synthesis (Fig. 2; a detailed overview of the included and excluded studies is in Table 2 and Supplementary Table 2, respectively).

Data extraction and quality assessment
The following variables were extracted: bibliographic information, sample characteristics (e.g.sample size, age and sex) and methodological specifics regarding ELA assessment and fMRI procedures (e.g.acquisition, paradigm and analyses) (see Table 2).Criteria for quality assessment were based on best-practice guidelines 21,22 .Studies passed the stringent quality assessment if (1) sample characteristics were properly described; (2) ELA subtype and assessment was reported; (3) details about fMRI paradigm (resting-state or task-based) and acquisition (scanner, settings and seed-region) were provided; and (4) details about fMRI processing (motion correction), analyses (whole-brain and where applicable, contrast(s) of interest) and results (coordinates of peak foci of activation in MNI or Talairach space) were provided.When details were not explicitly mentioned in the article itself, corresponding authors were contacted.www.nature.com/scientificreports/organized by subjects to limit within-subject effects as well 38 .Full-width at half maximum (FWHM) was subjectbased 89 and the modeled activation maps were computed using as more conservative mask size.Cluster-level interference thresholding was used based on uncorrected p-values (p < 0.001), with a cluster-level family wise error corrected threshold of p < 0.05 and 1000 permutations.Results were visualized using Mango 4.1.A global ALE analysis was performed across all experiments (n = 49 studies) to assess the relation between ELAs and functional brain connectivity irrespective of direction, hemisphere or paradigm.Regarding the latter, the joint analysis across heterogeneous designs, i.e. task-dependent and task-independent, reflects alterations of network connectivity across several mental states that are internally and externally determined.In a further step, we therefore specified the analyses into task-based (n = 22 studies) and resting-state (n = 27 studies) to delineate whether reported changes reflect common disturbances in neural mechanisms or paradigm-specific effects 90 .Additional sub-analyses based on direction (decrease: n = 31 studies, increase: n = 34 studies), hemisphere (left amygdala seed: n = 37 studies, right amygdala seed: n = 33 studies), age (adults: n = 23 studies, children: n = 26 studies), and disease status (healthy: n = 31 studies) were performed to further specify the effect (see Supplementary Table 4 for included studies per analysis).Moreover, separate sub-analyses on social ELAs (n = 35 studies) only, such as maltreatment, trauma, violence exposure as well as negative parenting, and on all postnatal ELAs excluding prenatal adversities were carried out (for a complete overview, see Supplementary Tables 3 and 4).Also, given recent evidence of a stronger link of subjective when compared to objective reports 75 and that prospectively and retrospectively assessed populations may follow different risk pathways 55 , such sub-analyses were added as well (subjective self-report: n = 29 studies and retrospective report: n = 34 studies).
Subsequently, the resulting ALE clusters were functionally decoded using all eligible BrainMap experiments (n = 19,044) on healthy subjects coded in terms of all behavioral domains (cognition, action, perception, emotion, and interoception) and paradigm classes to avoid preselection bias [91][92][93][94] .Of note, this database reflects activity and not connectivity, and emotion regulation is not included.For functional characterization, we considered forward inference using a binomial test (significant at p < 0.05) that determines in which domains and classes the probability of finding activation in the respective cluster [P (Activation|Task)] was significantly higher than the overall chance, i.e., across the entire database [P(Activation)].

Figure 2 .
Figure 2.This meta-analysis was conducted according to current consensus guidelines for CBMAs and the PRISMA guidelines.Five databases were systematically searched from 2001 (which was chosen as the earliest date for studies with sufficient quality) to 22 October 2020, retrieving 49 eligible studies.

Table 1 .
Results of the individual ALE analyses.L = left, R = right; X, Y, Z coordinates in MNI152 space.ACC, anterior cingulate cortex; ALE, activation likelihood estimation; ELAs, early life adversities.*Theseanalyses could not be performed due to an insufficient number of experiments based on current CBMA guidelines (≥ 17 experiments)34-36 .