Introduction
A strong relationship exists between a person's social environment and their mental health (Meyer-Lindenberg & Tost, Reference Meyer-Lindenberg and Tost2012). Increased rates of psychiatric disorders are consistently observed in people who have suffered abuse, deprivation, ostracism and isolation (Van Os et al. Reference Van Os, Rutten and Poulton2008). For example, childhood abuse and neglect have each been associated with, and are predictive of, increased risk for major depressive disorder (MDD), post-traumatic stress disorder and substance abuse later in life (Widom et al. Reference Widom, Dumont and Czaja2007; Gilbert et al. Reference Gilbert, Widom, Browne, Fergusson, Webb and Janson2009). Similarly, risk for schizophrenia has been repeatedly associated with living in larger urban areas and with migrant status, possibly due to higher levels of social competition (the ‘social defeat’ hypothesis; Selten & Cantor-Graae, Reference Selten and Cantor-Graae2005).
Functional magnetic resonance imaging (fMRI) is increasingly being used to characterize how brain responses to social and emotional stimuli may be different in high-risk individuals (Harmon-Jones & Beer, Reference Harmon-Jones and Beer2012). Recent fMRI studies comparing individuals who have experienced higher versus lower levels of stress have identified significant differences in blood oxygen level-dependent (BOLD) response in brain regions involved in social and emotional processing. One difficulty for interpreting these results, however, is the range of social stressors considered (e.g. urban upbringing, childhood trauma) and cognitive tasks (e.g. social stress, emotion recognition) that have been used. As such, it remains unclear whether different social environmental stressors increase risk for mental illness through common neural mechanisms.
To address this issue, we performed a coordinate-based meta-analysis to examine whether any brain regions were consistently different in high-risk individuals. This type of meta-analysis examines shared BOLD response across independent studies by quantitatively identifying brain regions consistently associated with effects of interest (Turkeltaub et al. Reference Turkeltaub, Eden, Jones and Zeffiro2002; Laird et al. Reference Laird, Fox, Price, Glahn, Uecker, Lancaster, Turkeltaub, Kochunov and Fox2005; Eickhoff et al. Reference Eickhoff, Laird, Grefkes, Wang, Zilles and Fox2009; Wagner et al. Reference Wagner, Sebastian, Lieb, Tüscher and Tadić2014). We hypothesized that differences would be observed in brain regions associated with threat and negative affect in high-risk individuals compared with low-risk individuals.
Method
Using PubMed, we searched for fMRI studies reporting differences in BOLD response associated with exposure (in adulthood or childhood) to social environmental stress. Studies published until June 2015 were searched for with the following search term: (‘functional magnetic resonance imaging’ OR ‘functional MRI’ OR ‘fMRI’) AND (‘social stress’ OR ‘early life stress’ OR ‘developmental trauma’ OR ‘childhood trauma’ OR ‘childhood maltreatment’ OR ‘urban upbringing’ OR ‘urbanicity’ OR ‘social status’ OR ‘socioeconomic status’ OR ‘ethnic minority’ OR ‘ostracism’ OR ‘rejection’ OR ‘exclusion’) NOT Review. This resulted in 307 studies being identified in total, of which 42 were original studies that matched study criteria (a human fMRI study examining main effects on BOLD response of one or more of the social environmental stressors listed in the search term). An additional six studies were recommended by reviewers. Next, we reviewed the references from each of the papers identified. This additional search retrieved a further six studies matching criteria and with available information. In total, 54 studies meeting search criteria were retrieved. Fig. 1 lists the number of studies included and excluded in this meta-analysis, and the reasons for inclusion or exclusion.
Studies included and excluded from activation likelihood estimation meta-analysis.
Next, we used the activation likelihood estimation (ALE) method in GingerALE 2.3 (Laird et al. Reference Laird, Fox, Price, Glahn, Uecker, Lancaster, Turkeltaub, Kochunov and Fox2005; Eickhoff et al. Reference Eickhoff, Laird, Grefkes, Wang, Zilles and Fox2009; Eickhoff et al. Reference Eickhoff, Bzdok, Laird, Kurth and Fox2012; Turkeltaub et al. Reference Turkeltaub, Eickhoff, Laird, Fox, Wiener and Fox2012) to perform a meta-analysis to determine whether any specific brain regions were consistently associated with exposure to social environmental stress. Where cluster coordinates were presented in Talairach space, these were converted to Montreal Neurological Institute (MNI) space using GingerALE [‘Talairach to MNI (SPM)’ transform] to input into the meta-analysis.
Maps of altered BOLD response were created for each study by modelling individual coordinates as Gaussian functions. The width of each of these functions is calculated by GingerALE software based on each study's sample size, i.e. GingerALE will model coordinates as wider Gaussian functions for loci from larger studies. Next, the overlap between these maps was used to calculate an ALE map. The probability of finding a particular value within an ALE map across studies was used to create a p-value image, which was thresholded using a false discovery rate of p < 0.05 and cluster-thresholded using 1000 threshold permutations and a cluster-level threshold of p < 0.05. We performed an ALE analysis of studies examining adults (mean age ⩾18 years) and children and/or adolescents (mean age <18 years) separately, to examine effects of social environmental stress at different developmental stages. We also performed a contrast analysis to examine differences between environmental effects on children/adolescents versus adults. For this contrast analysis, we also used a false discovery rate of p < 0.05 and cluster-thresholded using 1000 threshold permutations.
Results
In total, 54 studies meeting search criteria were retrieved, including a total sample of 3044 participants (see Fig. 1 and online Supplementary Tables S1 and S2). Where studies or analyses used overlapping samples, we used the smaller sample to calculate the Gaussian functions, as this is a more conservative approach (S. Eickhoff, personal correspondence).
Studies varied in terms of social environmental stressors investigated and tasks used, though some studies reported similar stressors and tasks. For example, the most frequently examined stressor was ‘childhood trauma’ or ‘childhood maltreatment’ (12 studies), followed by ‘early life stress’ (five studies) and ‘socio-economic status’ (five studies). The most frequently used task was facial emotion recognition (22 studies).
Across all studies, 55.49% of participants were adults (mean age ⩾18 years) and 44.51% of participants were children and/or adolescents; 51.97% of participants across 52 of the studies were listed as male, and 48.03% of participants were listed as female (however, gender information for the final sample included in the analysis after quality control was not provided for two studies (Dannlowski et al. Reference Dannlowski, Stuhrmann, Beutelmann, Zwanzger, Lenzen, Grotegerd, Domschke, Hohoff, Ohrmann and Bauer2012, Reference Dannlowski, Kugel, Huber, Stuhrmann, Redlich, Grotegerd, Dohm, Sehlmeyer, Konrad and Baune2013; Hsu et al. Reference Hsu, Langenecker, Kennedy, Zubieta and Heitzeg2010). A total of 27 studies used negative emotional stimuli, seven used positive emotional stimuli, 12 used a mix of negative, positive and/or neutral emotional stimuli, seven used cognitive tasks, and one used a task containing both emotional and cognitive conditions of interest (see online Supplementary Tables S1 and S2).
ALE meta-analysis results – adult studies
We first performed an ALE meta-analysis that included all 34 identified adult studies (1703 participants), irrespective of participant status as a patient, healthy volunteer or combat veteran, and in the case of patients, irrespective of diagnosis. Of the participants included in this analysis, 27.95% were patients, 1.64% were combat veterans and 70.40% were healthy volunteers. This analysis identified eight separate clusters in which BOLD response differed in groups exposed to greater social environmental stress (see Table 1 and Fig. 2). Clusters incorporated the bilateral amygdala, left superior frontal gyrus, left precuneus, left putamen, left thalamus, left insula and left inferior frontal gyrus. The cluster showing differences in BOLD response across the largest number of separate empirical studies was located at the right amygdala (nine studies). In the right amygdala cluster, increased BOLD response in the risk group compared with the non-risk group was reported across all studies except one. The single study, by Boecker et al. (Reference Boecker, Holz, Buchmann, Blomeyer, Plichta, Wolf, Baumeister, Meyer-Lindenberg, Banaschewski and Brandeis2014), which showed decreased BOLD response may have differed from the other six studies by reporting decreased BOLD response in this region during reward anticipation (a happy face symbol indicating a reward), rather than presentation of overt emotional stimuli (emotional faces, social stress, pleasant music).
Brain regions showing differences in blood oxygen level-dependent response in groups exposed to social environmental stress – all adults. Activation likelihood estimation meta-analysis includes all adult studies irrespective of whether patient groups were included. Each two-dimensional axial slice is labelled with a Montreal Neurological Institute coordinate. Clusters are rendered on the ‘ch256’ brain template using MRIcroGL (http://www.mccauslandcenter.sc.edu/mricrogl/). Additional editing of the figure (e.g. changing the size/resolution) was performed using MS Paint and Paint.NET v3.5.10.
Brain regions showing differences in blood oxygen level-dependent response associated with social environmental stress – adult studies (ALE meta-analysis includes all studies irrespective of whether patient groups were included)
ALE, Activation likelihood estimation; MNI, Montreal Neurological Institute.
Given that 13 of the 34 studies reviewed included patient participants, and in one case, combat veterans, we re-ran the ALE meta-analysis on the 21 studies that only included healthy civilian participants (1099 participants). Clusters incorporating the amygdala, thalamus and insula, identified in the first part of our analysis, again showed significant differences in BOLD response (and in the same direction) based on a comparison of participants with a history of high versus low environmental stress (see Table 2 and Fig. 3). The most consistent differences in BOLD response were again observed for the right amygdala (seven studies).
Brain regions showing differences in blood oxygen level-dependent response in groups exposed to social environmental stress (healthy adults only). Activation likelihood estimation meta-analysis includes only studies that examined healthy volunteers. Each two-dimensional axial slice is labelled with a Montreal Neurological Institute coordinate. Clusters are rendered on the ‘ch256’ brain template using MRIcroGL (http://www.mccauslandcenter.sc.edu/mricrogl/). Additional editing of the figure (e.g. changing the size/resolution) was performed using MS Paint and Paint.NET v3.5.10.
Brain regions showing differences in blood oxygen level-dependent response associated with social environmental stress – adults (ALE meta-analysis includes only studies that examined healthy volunteers)
ALE, Activation likelihood estimation; MNI, Montreal Neurological Institute.
ALE meta-analysis results – children/adolescent studies
Analysis of 21 studies examining children/adolescents (mean age <18 years; 1341 participants) revealed six clusters showing significant overlap between studies, incorporating the bilateral amygdala, left superior temporal gyrus, right middle temporal gyrus, right cerebellum and right thalamus (see Table 3 and Fig. 4). The cluster showing differences in BOLD response across the largest number of separate empirical studies was located at the right amygdala (eight studies). Across all eight studies, the high-risk group showed increased amygdala BOLD response compared with the low-risk group.
Brain regions showing differences in blood oxygen level-dependent response in groups exposed to social environmental stress (children/adolescents). Activation likelihood estimation meta-analysis includes only studies that examined children/adolescents. Each two-dimensional axial slice is labelled with a Montreal Neurological Institute coordinate. Clusters are rendered on the ‘ch256’ brain template using MRIcroGL (http://www.mccauslandcenter.sc.edu/mricrogl/). Additional editing of the figure (e.g. changing the size/resolution) was performed using MS Paint and Paint.NET v3.5.10.
Brain regions showing differences in blood oxygen level-dependent response associated with social environmental stress – children/adolescents
ALE, Activation likelihood estimation; MNI, Montreal Neurological Institute.
ALE meta-analysis results – contrast between adult studies and children/adolescent studies
Contrast analysis comparing sets of foci for the adult studies and children/adolescent studies revealed no statistically significant differences.
Discussion
This study used ALE meta-analysis to investigate whether the neural effects of social environmental stress were consistent and reproducible based on fMRI studies to date. Based on this analysis, brain regions including the right amygdala showed a consistent pattern of increased BOLD response to emotional stimuli in groups with a history of social environmental stress across multiple studies and at multiple developmental stages.
Findings of increased BOLD response in the amygdala are consistent with fMRI studies reporting increased neural response of this region during negative affective states and in psychiatric illness compared with controls. For example, amygdala hyperactivity has previously been associated with trait anxiety (Etkin et al. Reference Etkin, Klemenhagen, Dudman, Rogan, Hen, Kandel and Hirsch2004; Sehlmeyer et al. Reference Sehlmeyer, Dannlowski, Schöning, Kugel, Pyka, Pfleiderer, Zwitserlood, Schiffbauer, Heindel and Arolt2011), faster processing of negative stimuli and decreased levels of psychological well-being (Van Reekum et al. Reference Van Reekum, Urry, Johnstone, Thurow, Frye, Jackson, Schaefer, Alexander and Davidson2007), depressive symptom severity (Gaffrey et al. Reference Gaffrey, Luby, Belden, Hirshberg, Volsch and Barch2011) and cognitive biases towards negative stimuli (Dannlowski et al. Reference Dannlowski, Ohrmann, Bauer, Kugel, Arolt, Heindel, Kersting, Baune and Suslow2007a , Reference Dannlowski, Ohrmann, Bauer, Kugel, Arolt, Heindel and Suslow b ). Increased amygdala BOLD response has also been reported in response to emotional stimuli in patients diagnosed with bipolar disorder, MDD, social anxiety disorder and borderline personality disorder when compared with healthy controls (Yurgelun-Todd et al. Reference Yurgelun-Todd, Gruber, Kanayama, Killgore, Baird and Young2000; Surguladze et al. Reference Surguladze, Brammer, Keedwell, Giampietro, Young, Travis, Williams and Phillips2005; Minzenberg et al. Reference Minzenberg, Fan, New, Tang and Siever2007; Evans et al. Reference Evans, Wright, Wedig, Gold, Pollack and Rauch2008).
Another cluster showing increased BOLD response across multiple adult studies with increasing stress incorporated the superior frontal gyrus/Brodmann area 8 (see Table 1). BOLD response in this region increases with increasing decision uncertainty, which may be higher in high-risk individuals during tasks involving social stimuli (Volz et al. Reference Volz, Schubotz and Von Cramon2005). Other brain regions highlighted by our analysis play an important role in processing social stimuli themselves, which might also be associated with chronic social stress. These include the precuneus, inferior frontal gyrus, superior temporal gyrus and middle temporal gyrus (see Tables 1–3) (Adolphs, Reference Adolphs2001; Spreng et al. Reference Spreng, Mar and Kim2009; Van Overwalle, Reference Van Overwalle2009). Differences in BOLD response of striatal regions were also observed in high-risk individuals across studies (see Tables 1–3). These regions, including the putamen and globus pallidus, are heavily influenced by dopamine, which is known to increase in response to psychological stress, and may account for some of these effects (Pruessner et al. Reference Pruessner, Champagne, Meaney and Dagher2004; Selten et al. Reference Selten, van der Ven, Rutten and Cantor-Graae2013).
The therapeutic effects of psychological and pharmacological interventions are hypothesized to be partially mediated via a normalization of brain response to emotional stimuli (e.g. of the amygdala; Fu et al. Reference Fu, Williams, Cleare, Brammer, Walsh, Kim, Andrew, Pich, Williams and Reed2004; DeRubeis et al. Reference Derubeis, Siegle and Hollon2008; Norbury et al. Reference Norbury, Taylor, Selvaraj, Murphy, Harmer and Cowen2009; Rawlings et al. Reference Rawlings, Norbury, Cowen and Harmer2010; Windischberger et al. Reference Windischberger, Lanzenberger, Holik, Spindelegger, Stein, Moser, Gerstl, Fink, Moser and Kasper2010). Our study suggests that having a history of prolonged exposure to a stressful social environment may have important neural effects on regions associated with responding to stressful stimuli and illness risk and may therefore potentially mediate these therapeutic effects. Consistent with this view, some studies have already suggested that antidepressant response in MDD patients may be mediated by a history of childhood trauma (e.g. Nemeroff et al. Reference Nemeroff, Heim, Thase, Klein, Rush, Schatzberg, Ninan, McCullough, Weiss and Dunner2003). Confirming whether categorizing patients according to early adversity can help predict response to treatment type or even modality (pharmacological versus psychological versus both) will be an important avenue for further study.
Although neurobiological effects of social stress were not the focus of this meta-analysis, it is interesting to speculate about the relationship between differences in neurochemical response, given the wealth of evidence that (social) threat-related processing in the brain results in increased downstream levels of glucocorticoids in the blood (Belmaker & Agam, Reference Belmaker and Agam2008). The amygdala, cingulate and hippocampus are particularly high in glucocorticoid receptors and are also associated with chronic cortisol secretion (Gold et al. Reference Gold, Drevets and Charney2002). Since social environmental stress causes lasting changes in hypothalamic–pituitary–adrenal (HPA) axis responsiveness to stress, including increased levels of cortisol, this could be one mechanism by which it affects BOLD response in these brain regions (Heim et al. Reference Heim, Newport, Heit, Graham, Wilcox, Bonsall, Miller and Nemeroff2000; Lee et al. Reference Lee, Geracioti, Kasckow and Coccaro2005; Belmaker & Agam, Reference Belmaker and Agam2008; Heim et al. Reference Heim, Newport, Mletzko, Miller and Nemeroff2008; Heim & Binder, Reference Heim and Binder2012).
Another complimentary mechanism by which social stress may affect BOLD response in these brain regions is through differences in immune function. Hormones increased by psychological stress (cortisol, adrenaline) have significant effects on the immune system, enhancing pro-inflammatory cytokine responses and pro-inflammatory gene expression (Eisenberger & Cole, Reference Eisenberger and Cole2012). Inflammation, in turn, is associated with hyperactivity of both the amygdala and anterior cingulate in response to emotional stimuli, neural responses that have been correlated in the same studies with social disconnection and mood deterioration, respectively (Harrison et al. Reference Harrison, Brydon, Walker, Gray, Steptoe and Critchley2009; Inagaki et al. Reference Inagaki, Muscatell, Irwin, Cole and Eisenberger2012).
With regard to possible gender effects, DeSantis et al. (Reference Desantis, Baker, Back, Spratt, Ciolino, Maria, Dipankar and Brady2011) report differing effects of early-life trauma on HPA axis functioning in males and females, and brain regions reported in this meta-analysis, such as the amygdala, are known to function differently in males and females in response to threatening social stimuli (Schneider et al. Reference Schneider, Peters, Bromberg, Brassen, Menz, Miedl, Loth, Banaschewski, Barbot, Barker, Conrod, Dalley, Flor, Gallinat, Garavan, Heinz, Itterman, Mallik, Mann, Artiges, Paus, Poline, Rietschel, Reed, Smolka, Spanagel, Speiser, Ströhle, Struve, Schumann and Büchel2011), suggesting that results may differ between males and females. This meta-analysis included studies that examined both males and females (44 studies), males exclusively (five studies) and females exclusively (three studies). However, only two of these studies reported different effects of social environmental stress between males and females (Felmingham et al. Reference Felmingham, Williams, Kemp, Liddell, Falconer, Peduto and Bryant2010, in which trauma-exposed women showed increased brainstem BOLD response compared with trauma-exposed men and Spielberg et al. Reference Spielberg, Galarce, Ladouceur, McMakin, Olino, Forbes, Silk, Ryan and Dahl2015, in which effects of socio-economic status on cingulate BOLD response were only observed in females). This research could therefore be extended by further studies comparing effects of social environmental stress between males and females, and more studies examining one gender specifically (to contrast in future meta-analyses).
A majority of the studies included investigated the effects of social stressors occurring in childhood (only 10 studies included social stressors experienced in adulthood). While this follows the widely held expectation that childhood adversity will have neurodevelopmental consequences, whether and how the duration and staging of these stressors mediated the effects observed was therefore not possible to evaluate in this meta-analysis. Answering this question will be an important priority for future meta-analyses as more studies examining effects of social stressors experienced in adulthood become available.
It is important to note that social environmental effects on BOLD response may be influenced by the types of tasks used (e.g. emotional versus cognitive tasks, positive emotion versus negative emotion). For example, Dannlowski et al. (Reference Dannlowski, Kugel, Huber, Stuhrmann, Redlich, Grotegerd, Dohm, Sehlmeyer, Konrad and Baune2013) report effects of childhood trauma on increased limbic BOLD response during sad face processing compared with happy face processing. Similarly, limbic hyper-responsiveness has consistently been observed in psychiatric disorders in response to negative stimuli, while responses to positive stimuli are not as well characterized (Rauch et al. Reference Rauch, Whalen, Shin, McInerney, Macklin, Lasko, Orr and Pitman2000; Siegle et al. Reference Siegle, Thompson, Carter, Steinhauer and Thase2007; Mothersill et al. Reference Mothersill, Morris, Kelly, Rose, Bokde, Reilly, Gill, Corvin and Donohoe2014). Given that a minimum of 15 studies should be included in each group in an ALE contrast analysis for valid results, we could not in this study compare different types of task (Wagner et al. Reference Wagner, Sebastian, Lieb, Tüscher and Tadić2014).
Our meta-analysis excluded studies where the neural effects of social stress were not presented as main effects, but only reported in interaction with other variables (e.g. genetic risk, oxytocin administration). Examining gene × environment interactions is clearly a priority for the field to determine the degree to which established environmental and genetic risk factors converge on the same neural circuits in psychiatric illness (Meyer-Lindenberg & Tost, Reference Meyer-Lindenberg and Tost2012). For example, Streit et al. (Reference Streit, Haddad, Paul, Frank, Schäfer, Nikitopoulos, Akdeniz, Lederbogen, Treutlein, Witt, Meyer-Lindenberg, Rietschel, Kirsch and Wüst2014) report that individuals who were both raised in an urban environment and who carried two copies of the neuropeptide S receptor 1 gene showed increased right amygdala BOLD response during stress processing, relative to individuals who grew up in highly urbanized environments with one or no copies of this variant. As further studies of the neural effects of gene by (social) environmental risk emerge, it will be interesting to determine how genetic background increases liability to, or resilience against, the neural effects of the early social environment.
Finally, although the focus of this meta-analysis was of BOLD response data during cognitive–emotional tasks, it is important to note that chronic environmental stress may also be associated with structural differences in the brain regions identified. For example, Dannlowski et al. (Reference Dannlowski, Stuhrmann, Beutelmann, Zwanzger, Lenzen, Grotegerd, Domschke, Hohoff, Ohrmann and Bauer2012) showed that childhood trauma was associated with decreased hippocampal and prefrontal grey matter volumes, while Tottenham et al. (Reference Tottenham, Hare, Quinn, McCarry, Nurse, Gilhooly, Millner, Galvan, Davidson and Eigsti2010) have shown that children adopted out of orphanages at older ages had larger amygdala volumes compared with early-adopted children and controls.
In conclusion, this meta-analysis examined fMRI studies of social environmental stress exposure in both adults and children/adolescents. Social environmental stress was found to be associated with altered BOLD response across a range of brain regions, and, of these, increased BOLD response of the right amygdala was a robust finding across a range of populations and based on response to a variety of stimuli. What remains unknown is whether social environmental stress has differing effects on treatment response in these brain regions.
Supplementary material
For supplementary material accompanying this paper visit http://dx.doi.org/10.1017/S0033291716000477
Acknowledgements
This work was supported by a Science Foundation Ireland (SFI) Research Investigator project award to G.D. (SFI: 12.IP.1359).
Declaration of Interest
None.
