How do bipolar disease states affect positive and negative emotion processing? Insights from a meta‐analysis on the neural fingerprints of emotional processing

Functional magnetic resonance imaging studies on emotion processing in patients with bipolar disorder (BD) show hyperactivity of limbic‐striatal brain areas and hypoactivity in inferior frontal areas compared to healthy participants. However, heterogeneous results in patients with different disease states and different valences of emotional stimuli have been identified.


| INTRODUC TI ON
Bipolar disorder (BD) is a severe affective disorder with high levels of distress, severe impairment, and increased mortality. 1,2BD is characterized by alternating intense negative and positive disease states -referred to as depressive and (hypo-)manic episodes.It has thus been assumed that difficulties in emotion processing and regulation cause these affective extremes. 3[6][7] To understand the mechanisms of emotion processing changes in BD, their neural correlates have been subject to numerous functional magnetic resonance imaging (fMRI) studies. 3,8These studies identified aberrations in regions related to generating emotions, evaluating emotional stimuli, and signaling reward (limbic and striatal regions), as well as regions related to controlling and regulating emotional states (prefrontal regions, specifically the inferior frontal gyrus/ventrolateral prefrontal cortex 9,10 ).In these studies researchers had to face several challenges: First, patients with BD can experience (at least) three different disease states (euthymia, depression, or (hypo-)mania), which likely influence how they process emotional stimuli. 11,12In previous studies on emotion processing in patients with BD, these disease states have either not been differentiated or the investigations have been limited to one disease state, for example, bipolar depression (e.g., Ref. [13]).
However, comparing different disease states could help disentangle which alterations of emotion processing are a characteristic of the disease state (i.e., mania or depression) rather than a trait of BD itself.
Second, emotion valence may be an important contributor to differences in emotion processing in BD. 14 These differences could provide additional insight into valence-specific stimulus processing and how it promotes mood swings into depression or mania.For example, increased maintenance of positive emotion irrelevant of the context has been observed in BD 15 .Increased maintenance of positive emotion might promote arousal and could thus be conducive to the emergence of manic phases.However, previous neuroimaging studies either did not distinguish between different valences, focused on examining the processing of one specific emotion (e.g., fear), or examined the processing of positive or negative emotions without comparing them (e.g., Refs [13,16,17]).
Third, valence-specific processing may also switch between different disease states of BD-patients.For example, during mania BD-patients may show increased neural activity toward positive stimuli, reinforcing their manic disease state.During depression, BDpatients may show hyperactivity toward negative stimuli, promoting their depressed disease state.
To disentangle interactions of disease state and valencespecific neural processing in BD, it is thus necessary to compare positive and negative emotion processing during the different disease states of BD.At this stage, a meta-analysis could reveal conclusive correlates of emotion processing in BD and disentangle the interaction of disease state and valence-specific processing using subgroup analyses.Subgroup analyses look separately at regions that show activation differences depending on the patients' disease state and the valence of the stimuli.These analyses can also reveal subtle differences that would be missed in a whole-brain analysis and can, therefore, be helpful in examining the relationship between emotion processing and disease states in BD-patients.
[20][21][22][23] However, these have not addressed the relevance of disease states to the processing of positive and negative emotions.
Hence, in this meta-analysis, we aim to elucidate how disease states and valence of the emotional stimulus affect neural activity during emotion processing in patients with BD compared to healthy participants (HC): First, against the background of previous results of meta-analyses and review papers, we want to test the previous assumptions on emotion processing in BD-patients reporting a higher activity of emotion and sensory processing regions (e.g., amygdala) and a lower activity of regions associated with cognitive control and emotion regulation (e.g., inferior frontal gyrus) toward emotional stimuli. 4,8,9Second, given the amount of studies now available on emotion processing in BD-patients, we aim to track the effects of patients' respective disease state (manic, depressive, euthymic-Model 2), the valence of the stimulus (positive vs. negative-Model 3) as well as their interaction on emotion processing (Model 4-subgroup analysis).

| Literature search
The literature included in our meta-analysis was obtained by different search strategies.Most importantly, we performed keyword searches in Web of Science's Core Collection and Medline databases in November 2022.Comparable to other neuroimaging meta-analyses that investigate emotion processing, 24,25 we used a combination of the following search terms to identify studies of neural correlates of emotion processing in bipolar disorder: (i) "emotion" or "emotional" or "IAPS" or "emotion comprehension" or "emotion perception" or "affect comprehension" or "affect perception" or "facial expression" or "prosody" or "empathy" or "empathetic" or "altruism" or "sympathy" or "emotional contagion" or "compassion" and (ii) "neuroimaging" or "fmri" or "PET" and (iii) "bipolar disorder" or "bipolar" or "mania."Additionally, we performed backward and forward searches from relevant studies identified during our initial database search.Overall, this search strategy identified 1590 studies for more in-depth screening (see Figure 1, PRISMA flow diagram).PMH, KF, and LM screened the title and abstracts of 1590 studies.Studies not relevant to the research question (e.g., reviews, meta-analysis and animal studies) were excluded in a first brief screening of the title, leading to a sample of 919 studies for the detailed prescreening of the abstracts.This procedure resulted in a sample of 119 studies eligible for full-text screening.Two independent raters (KF and LM) evaluated these 119 studies using the following five criteria: (1) We included studies that reported results from participants with a diagnosis of bipolar disorder compared with healthy controls (e.g., excluding studies with participants at high risk for developing bipolar disorder) that (2) performed a task that could broadly be described as measuring "emotion processing" (see supplementary Table S1 for a description of tasks included).
The tasks included in our final sample could be described in terms of the following task categories: face matching, 26 emotion labeling, 27 observing emotional facial expressions/scenes (while performing another task, such as gender, color, or age identification 28 ; or passive observation 29 ); and more complex emotional-cognitive processing (emotional Go/NoGo task 30 ; emotional n-back task 31 ).Furthermore, studies had to adhere to the following criteria, as required for coordinate-based meta-analysis 32 : (3) reported coordinates had to correspond to standard stereotactic space (MNI, Talairach), (4) stem from whole-brain analysis, and (5) use a consistent threshold throughout the entire brain (excluding studies reporting small-volume corrected results).Applying these criteria, we obtained a sample of 31 studies which we included in our meta-analysis (patients with BD: n = 766, HC: n = 836).Interrater reliability was substantial (Cohen's kappa = 0.74).Discrepancies were discussed and solved via consensus.

| Meta-analysis method
We carried out effect size-based meta-analysis using the anisotropic effect size-based algorithm of seed-based d Mapping (AES-SDM 5.15, 32 www.sdmpr oject.com).Methodologically, SDM contains some features from other widely used methods for meta-analysis, such as Activation Likelihood Estimation (ALE 33 ), or Multilevel Kernel Density Analysis (MKDA 34 ), but extends them by allowing for an inclusion of effect sizes from the original studies. 32Specifically, effect size (Hedge's g) and variance maps for activation peaks derived from the reported effect size distribution are estimated based on each study's t-values and reported sample sizes.Within those effect size maps, effect sizes are estimated for all other voxels within a gray matter mask (where no initial peak activation was reported) depending on the distance to the closest activation peak using an unnormalized Gaussian kernel with FWHM of 20 mm (e.g., if a study reports an activation peak at the left precentral gyrus with MNI coordinates [−40, −2, 32], and an effect size of t = 2.28, an effect size map can be estimated for all other voxels in the brain depending on the distance of each voxel to the reported activation peak.For example, activation at the left postcentral gyrus with MNI coordinates [−40, −26, 52], which lies at an Euclidean distance of 31.24mm from our initial activation peak, would be assigned an estimated standardized effect size of Hedges g = 0.262).Statistical significance is determined by randomizing the location of the voxels within a whole-brain gray matter mask (we performed 100 randomizations), and results are reported in MNI space at the recommended statistical threshold of p < 0.005 uncorrected (voxel-level) and at a cluster extent of 10 voxels, which has been found to optimally balance sensitivity and specificity, and is approximately equivalent to a corrected threshold of p < 0.05 in a single study. 32

| Disease state-and valence-analysis (Models 2 and 3)
To further determine differences between subgroups and conditions within our main sample (differences in valence, and disease state), we calculated linear contrasts using SDM's linear model function, calculating the difference in effect size between meta-analyses.In order to obtain balanced sample sizes between subgroups and conditions, we randomly selected studies to include for group comparison in the linear contrast analysis (subgroup analysis: 10 studies investigating 178 euthymic patients, 8 studies investigating 131 depressed BD-patients, and 7 studies investigating 148 (hypo-)manic patients; condition-wise comparison: 12 studies investigating responses to emotionally positive stimuli in 778 BD and HC participants, 13 studies investigating responses to emotionally negative stimuli in 905 BD and HC participants).In addition to accounting for the within-study variance and sample size of each study by effect size, these linear models additionally weigh within-and between-study variance and sample size.Furthermore, since our data contained high variability in terms of participant's age, we also performed an additional meta-regression analysis using SDM's meta-regression function to account for the effect of age on our results (see Supplementary Analysis, Table S6), and included age as a covariate in the linear contrast analysis.

| Interaction of disease state and valence analyses (Model 4)
To identify the disease state-and condition-specific variations that underlie our main analysis pooled across groups and tasks within our database, we performed further detailed sub-analyses.That is, using the activation peaks across all studies as our ROI (activation peaks from the main analysis, see Table S2), we extracted effect size measures from individual studies included in our supplemental analyses, as well as effect size estimates from the linear models comparing valenceand disease state-specificity (e.g., anterior insula activation across bipolar patients in euthymic, depressed, or manic disease state, models 2 and 3).To further characterize disease-state-specific alterations in emotion processing, we furthermore extracted effect-size estimates from the disease-state analysis (model 2) restricted to valence ROIs (activation peaks from valence-specific analysis, contrasting negative > positive emotional processing, model 3, Table S4).Meta-analytic effect size estimates are SDM-z values (representing the standardized difference from the population mean).These estimates might point to activation differences between emotionally negative and positive stimuli, as well as disease states included in our literature database.
However, as they do not represent statistical difference measures per se, we show disease state as well as valence-specific differences in effect size estimates as descriptive and additional information, but do not infer statistically significant differences between them.For all ROIs from the main and valence-analysis in which the absolute effect size estimates differences between groups are greater than 0.75 (comparing effect size differences between positive and negative task stimuli, and paired comparisons between euthymic, depressed, and manic disease state), we show bar graphs illustrating the distribution of the effect-size estimate within each group.

| Heterogeneity and sensitivity analysis
To determine whether our observed activations showed variability among individual studies from our database, we performed heterogeneity analysis as implemented in SDM.This is performed by testing whether between-study variance for an area is larger than that which would be expected from sampling error alone (measures are reported in standardized z-values).Furthermore, we conducted systematic whole-brain jackknife sensitivity analysis to determine the robustness of our findings.Here, one study was removed from the overall sample and meta-analysis was performed on the remaining studies; this was performed for all studies.It is argued that if significant activation is observed across all or a combination of most studies, it can reasonably be concluded that these findings are highly replicable.

| Disease state analysis (Model 2)
Although limited in scope within our current meta-analytic sample due to small sample sizes (e.g., 167 patients in a (hypo-)manic disease state), we nevertheless want to highlight some striking activation differences that were observed between BD-patients in euthymic, depressed, and (hypo-)manic disease state.While we found some converging hyperactivation across all disease states (right Rolandic operculum, superior parietal gyrus, although lateralization was specific to disease state), BD-patients showed some disease-state-specific activation clusters.For example, converging hyperactivation in the right amygdala (contrast BD > HC) was observed in euthymic and depressed disease state, but not in manic BD-patients.Furthermore, we observed hyperactivation specific to one disease state (left thalamus activation in euthymic BDpatients, left activation of the inferior frontal gyrus in depressed BD-patients, left hippocampus activation in manic BD-patients; contrast BD > HC).A similar direction of results was observable for converging hypoactivation across studies (contrast BD < HC).More precisely, euthymic BD-patients were characterized by hypoactivation in the right cerebellum, and depressed BD-patients showed hypoactivation in the left amygdala (see Figure 2A "Disease Stateand Valence-Analysis (Models 2 and 3)", supplementary Table S3).

| Valence analysis (Model 3)
We were also interested in investigating whether there was a valence-specific effect observable in our data.To this end, we contrasted studies in which participants were presented with emotionally negative and/or positive stimulus material comparing the BD and HC groups.*For the contrast of emotionally negative > positive stimuli, we observed converging hyperactivation for BD-patients compared to HC at the anterior insula and paracingulate gyrus (contrast negative > positive, BD > HC).In the same valence-contrast (negative > positive, BD < HC), we observed converging hypoactivation for BD-patients compared to HC at the right superior frontal gyrus, cuneus, and anterior thalamic projections, as well as the left anterior insula (see Figure 2B "Disease State-and Valence-Analysis (Models 2 and 3)" and supplementary Table S4).

| Interaction of valence processing and disease state (Model 4)
To better understand the disease state-and valence-specific patterns of neural activation, we examined disease state-and valencespecific activation differences around activation peaks from the neural regions observed across our entire database (main analysis), as well as valence-specific activation differences.To this end, we (a) created brain masks from peak activations in our main analysis and extracted study-wise meta-analytic effect size estimates from the linear models investigating valence-and disease-state-specific effects.
To furthermore investigate whether there were disease-state-specific differences in the processing of emotionally positive versus negative emotions, we (b) repeated this analysis with brain masks from our valence-specific analysis (e.g., we extracted effect size estimates at the activation peak observed for emotionally negative vs. emotionally positive task stimuli from the disease-state-specific linear contrasts maps).Meta-analytic effect size estimates are displayed as bar graphs alongside the corresponding mean ROI activation map in Figure 3.
Supplementary Table S5 lists effect size estimates for disease state and valence where the absolute differences in SDM-z values between disease states and valence (positive vs. negative; pairwise comparison between euthymic, depressed, and manic disease state) is higher than 0.75 (for example, we included the effects from the left fusiform gyrus from the main analysis contrast BD < HC in the graph of our subgroup analysis, wherein the absolute difference in effect sizes between the depressed (SDM-z = −0.01)and manic (SDM-z = −1.233)patient group is larger than 0.75 (absolute difference: SDM-z = 1.223)).
Further characterizing our main analysis peaks (a), we found valence-specific differences in activation in the left anterior in-  S5).Furthermore, we found lower activation for participants in manic compared to euthymic and depressed disease state in the inferior frontal gyrus (SDM-z euthymic = −0.378,SDM-z depressed = −0.184,SDM-z manic = −1.193;specific to negative emotion processing) and anterior thalamus (SDM-z euthymic = 0.104, SDM-z depressed = 0.111, SDM-z manic = −0.975;specific to the contrast negative > positive emotion processing).

| Neurosynth decoding
The methodological details as well as the results of our neurosynth decoding can be found in the supplements (see also supplementary Figure S1, page 6 f.).

| DISCUSS ION
In this large-scale SDM meta-analysis specifically examining emotion processing in patients with BD, our particular focus was to elucidate how the valence of emotional stimuli and the disease state of the patient affect neural activity during emotion processing.
First, our meta-analysis corroborates previous results by showing hyperactivation (BD > HC) of regions associated with salience and emotion processing of emotional stimuli (the bilateral anterior insula, the posterior cingulate cortex, and the postcentral gyrus).In addition, our meta-analysis indicates hypoactivation (BD < HC) of regions associated with interoceptive processing and emotion regulation (the cuneus, right anterior insula, inferior frontal gyrus and middle temporal gyrus) during emotion processing of BD-patients in general. 3,9ese neural correlates of emotion processing in BD support the idea of a reduced control over an increased emotional reactivity leading to dysregulated/pathological affective states in BD (for a review see e.g., Refs [4,8]).Second, BD-patients showed similar converging activations during depression and euthymia, but a distinct pattern of neural activity during mania, revealing a higher involvement of regions associated with social and somatosensory processing during mania (e.g., supramarginal gyrus).Thereby, our meta-analysis extends earlier views on disease-state-specific processing 9 and bears insight into the dynamics of emotion processing during bipolar disease states.Third, positive and negative emotion processing also revealed a distinct pattern of results.While emotion processing circuits associated with salience processing showed an increased activity toward negative emotional stimuli in BD-patients (e.g., posterior insula), positive stimuli elicited elevated activity in distinct brain regions that are rather associated with sensory processing, emotion generation and perspective taking (e.g., calcarine fissure/precuneus, supramarginal gyrus).This hyperactivation of sensory processing regions might suggest that BD-patients are more responsive to emotional stimuli in general, which is in line with the enhanced emotional distractibility observed on the behavioral level in BD-patients. 58Finally, neural correlates of positive and negative emotions emerging during disease states were different, revealing reduced left anterior insula activation as a specific neural correlate of negative emotion processing in mania.The anterior insula specifically is responsible for interoception as a part of affective experiences 59 and might here reflect an altered interoceptive processing that has also been associated with depression, such as lower interoceptive accuracy. 60

| Disease-state-specific processing: Similarities and differences
Examining how disease states of BD impact emotion processing, we found that depressed and euthymic BD-patients hyperactivate the Rolandic operculum-an area that surrounds the insula and is involved in emotion processing as well as visceral-gustatory sensations. 61,62milarly, depressed and euthymic patients showed hypoactivation of the inferior frontal gyrus-a region involved in emotion regulationand middle occipital gyrus-a region involved in visual (face) processing. 63,64This implies that although euthymic and depressed BD-patients show reduced visual processing of emotional stimuli, they show heightened emotional reactivity in combination with a lower engagement of neural regions regulating emotional states.One could speculate if the reduced visual processing of emotional stimuli may also provide an attempt to reduce emotional reactivity when there is insufficient power to regulate emotions.Next to these similarities in emotion processing, there were also different changes for euthymic and depressed BD-patients.Euthymic patients showed a stronger activation of sensory and emotion processing areas (e.g., supramarginal gyrus and amygdala) and lower activation of the superior frontal gyrus-a region involved with working memory and cognitive control, further corroborating the idea of a lower cognitive control toward a heightened emotional reactivity in euthymic BD-patients. 65pressed BD-patients also show a stronger activation of attention-

| Positive and negative emotion processing
In the present meta-analysis, negative in comparison to positive stimuli lead to lower neural activation in regions associated with emotion and  F I G U R E 3 ROI-specific analysis for peaks from main analysis (A) and valence-specific analysis (B).We portray effect size estimates (SDM-z scores, the standardized difference in population means) extracted from the linear models contrasting valence of task stimuli (pos = positively valenced stimuli; neg = negatively valenced stimuli) and disease state of bipolar disorder (euth = bipolar patients in current euthymic disease state; depr = bipolar patients in current depressed disease state; manic = bipolar patients in current manic disease state) from main analysis peaks (3a), and the linear contrast comparing positive versus negative emotion processing (3b).In all panels red indicates hyperactivation (BD > HC) and blue indicates hypoactivation (BD < HC).Hypo-(blue cluster) and hyperactivation (red cluster) follow-up analysis of the anterior insula are reported in the top panel of 3a.

| Interaction of valence processing and bipolar disease state
Our supplemental analysis revealed specific activation of the anterior insula per group (see Figure 3 and Table S5 in the supplements).
While the anterior insula is a correlate of emotion processing in general, distinct disease-state-specific activation patterns have been detected.Here, patients with a manic disease state showed significantly lower anterior insula-activation compared to depressed and euthymic patients, suggesting lower anterior insula activation as a key correlate for the manic disease state in BD.One might thus look at left anterior insula activity as a correlate of swings into manic mood in euthymic and depressed patients.Interestingly, on a descriptive level the left anterior insula shows lower activity for the manic disease state when contrasting negative against positive emotion processing (see Figure 4).Hence, anterior insula activity during emotion processing may be a biomarker suitable to distinguish between disease states of BD-patients.Future studies may be focusing on the anterior insula and its role for relapse into and maintenance of pathological disease states.

| Clinical implications
Our meta-analysis confirmed common models of emotion processing in BD that assume increased limbic activity toward emotional stimuli and decreased activity of brain regions associated with cognitive control and emotion regulation, such as the prefrontal cortex, in comparison to HC (for a review, see Ref. [4]).These models could additionally be extended by findings on positive and negative emotion processing as well as emotion processing in different disease states.These findings on neurobiological emotion processing may provide clues to mechanisms that can be targeted in psychotherapy or psychopharmacological treatment of BD-patients.Psychotherapeutically, changes in emotional reactivity could be achieved in cognitive behavioral therapy through specific modification of emotion-triggering situations.Newer psychotherapeutic methods such as Emotion Regulation Therapy 66,67 target neural networks of emotion regulation and could also increase cognitive control over dysregulated emotional reactivity in BD-patients.Specific psychopharmacological treatments, for example with lithium, may also target dysregulated brain circuits of emotion processing. 68Despite recent evidence that lithium induces neuroplasticity in emotion regulation brain circuits in patients with bipolar disorder, 68 evidence linking changes in emotion processing circuits to mechanism-oriented treatments in BD is still lacking.Future studies should investigate if changes in emotion processing circuits can be specifically targeted by psychotherapeutic interventions in patients with BD.

| Summary
In line with previous models of emotion processing in BD, 9,10,69 the present meta-analysis corroborates alterations of brain function of emotion and sensory processing areas as well as areas of emotion regulation and cognitive control.
Taken together, previous BD-models assume an enhanced automatic processing of emotionally significant stimuli in general, which is reflected in our meta-analysis by a stronger activation of limbic as well as sensory processing regions during emotion processing in BD-patients.
On the other hand, there are functional impairments of automatic voluntary regulation and processing of emotional states, which are reflected in a reduced neural activation especially in prefrontal structures.
Our meta-analysis extends these models with valence-and diseasestate-specific correlates, yielding important information regarding the maintenance of disease states that approach the big question of how emotion processing contributes to mood swings in bipolar disorder.We found evidence that there is a distinct pattern of elevated activity toward positive emotions in patients with BD that may reflect enhanced emotional reactivity and enhanced maintenance of positive emotions in BD-patients.In addition, we found differences in disease-state-specific emotion processing indicating that disease states are maintained via specific alterations of emotion processing in the anterior insula.

| Limitations
The limitations of this meta-analysis are predominantly due to the studies analyzed and their design.First, the patient-samples are small and very heterogeneous (e.g., different age groups and disease states of patients are mixed), which resulted in small samples in our valence and disease state analysis.We addressed this limitation with a jackknife sensitivity analysis for the main analysis that replicated our results (see supplementary Tables S2-S6).The included studies also mixed patients of different age groups, which according to a meta-analysis comparing pediatric and adult BD show differences in neutral emotion processing. 23Therefore, age was accounted for in the analysis by ways of meta-regression for the main analysis, as well as inclusion as covariate in the linear models (see Table S6).Finally, only twelve studies examined unmedicated patients with BD.We, therefore, compared the results of these studies including unmedicated patients with the study results of the medicated patients.
Here, there was also no difference in the results (see Table S6).
Further adding to the results of previous meta-analyses, [18][19][20][21]23 we also aimed to clarify the further influence of explicitness of the emotion processing task on neural correlates of emotion processing in BD (see supplement methods and supplementary Table S7).

| CON CLUS IONS
This meta-analysis examined emotion processing in patients with BD.We provide detailed insights into emotion processing during F I G U R E 4 Anterior insula hypoactivation (BD < HC) as a correlate of negative emotion processing compared to positive emotion processing during mania.This is a descriptive image of the results on the interaction of valence and mood state on neural activation in bipolar disorder.Image created with biorender (biore nder.com).

3. 2 |
Main analysis (Model 1)Meta-analytic analysis of emotion processing in BD compared to healthy controls (HC) included comparison of 150 coordinates (BDpatients) and 117 coordinates (HC) in 1602 participants.Overall, our meta-analysis showed converging hyperactivation for BD-patients in bilateral anterior insula, left posterior cingulate cortex, and left postcentral gyrus compared with healthy controls (contrast: BD > HC).We observed converging hypoactivation for BD-patients compared with healthy controls in the cuneus, left anterior insula, right inferior frontal gyrus, middle temporal gyrus, and fusiform gyrus (contrast BD < HC).All results, including measures of heterogeneity and sensitivity, are reported in TableS2(see supplementary material), as well as Figure2(top panel), "Main Analysis: Patients with Bipolar Disorder > Healthy Controls (Model 1)".
sula (for contrast BD > HC: SDM-z positive = 0.074, SDM-z negative = 1.24; for contrast HC > BD: SDM-z positive = −0.673,SDM-z negative = −1.559),and the left posterior cingulate cortex (SDM-z positive = 1.270,SDM-z negative = 2.320), both specific to negative emotion processing, as well as disease-state-specific activation differences in the left fusiform gyrus (SDM-z euthymic = −0.004,SDM-z depressed = −0.01,SDM-z manic = −1.23),left anterior insula (SDM-z euthymic = −0.458,SDM-z depressed = −0.254,SDM-z manic = −1.105),and left middle temporal gyrus (SDM-z euthymic = −0.130,SDM-z depressed = −1.090,SDM-z manic = −0.298).Investigating disease-state-specific differences in the valence linear model peaks (b), we found lower activation for participants in manic disease state compared to euthymic and depressed disease state in the left anterior insula (anterior insula peak from contrast BD > HC: SDM-z euthymic = 0.142, SDM-z depressed = −0.121,SDM-z manic = −0.610;anterior insula peak from contrast BD < HC: SDM-z euthymic = −0.397,SDM-z depressed = −0.164,SDM-z manic = −1.138)which we have found to be specific to negative > positive emotion processing (cf.Table controlling and emotion-regulating areas (e.g., the inferior frontal gyrus and superior parietal gyrus) and lower activation of regions associated with contextual memory (parahippocampal gyrus), suggesting a dysregulated cognitive control system (up-and downregulation of the inferior frontal gyrus) during bipolar depression.Here, manic BD-patients showed stronger activation in areas associated with sensory processing (the Rolandic operculum) and attention-control (the superior parietal gyrus), but lower activation of emotion regulating areas (e.g., inferior frontal gyrus) and interoceptive processing (e.g., anterior insula).Overall, the results of the disease state analysis suggest a distinct pattern of neural activity during emotion processing in mania compared to euthymic and depressed mood: In mania, emotion recognition areas were hyperactivated.Euthymic and depressed BD-patients in turn showed hyperactivated areas of emotion regulation, cognitive control and attention.These distinct correlates of disease states may also indicate how mania and depression could be TA B L E 1 Study sample characteristics.
sensory processing of BD-patients (left anterior insula and right thalamus).Surprisingly, positive stimuli lead to higher activation in regions associated with sensory processing and emotion generation (supramarginal gyrus, calcarine fissure/precuneus).This result may also suggest an increased reactivity of the brain to positive stimuli in BD.An increased neural reactivity toward positive emotions is in line with earlier studies showing that patients with BD show elevated maintenance of positive in comparison to negative emotions15 on a behavioral level and may also become observable in the increased distractibility of patients with BD when emotions are presented during a mental arithmetic task.58

F I G U R E 2
Differences in neural activity during emotion processing between patients with bipolar disorder (BD) and healthy control participants (HC).Top panel: Main analysis comparing general effects of emotion processing in BD versus HC (Model 1).(A) Disease-statespecific analysis of differences in emotion processing in BD versus HC (Model 2).(B) Valence-specific analysis of differences in emotion processing in BD versus HC (Model 3).In all panels red indicates hyperactivation (BD > HC) and blue indicates hypoactivation (BD < HC).

controls Task characteristics Mean depression score Mania cut-off score Depression cut-off score N Age Sex (F/M) Task Category Valence of Stimuli
Abbreviations: ASRMS, Altman Self-Rating Mania Scale; BDI, Beck's Depression Inventory; CDI, Child Depression Inventory; CDRS-R, Children' Depression Rating Scale-Revised; HAM-D, Hamilton Depression Rating Scale; IDS, Inventory of Depressive Symptoms; K-SADS, Kiddie Schedule for Affective Disorders and Schizophrenia for School-Aged Children; MADRS, Montgomery-Asberg Depression Rating Scale; n.r., not reported; SCID, Structured Clinical Interview for DSM; YMRS, Young Mania Rating Scale.aAuthors reported age at first episode/mood symptom onset.bCollapsed across pediatric and adult bipolar disorder sample (mean illness duration pediatric sample: 4 years, mean illness duration adult sample: 20 years).