Deficits in Frontoparietal Activation and Anterior Insula Functional Connectivity During Regulation of Cognitive-Affective Interference in Bipolar Disorder

Abstract

Objectives:

Bipolar disorders (BD) are characterized by emotion and cognitive dysregulation. Mapping deficits in the neurocircuitry of cognitive-affective regulation allows for potential identification of intervention targets. The current study used functional MRI data in BD patients and healthy controls during performance on a task requiring cognitive and inhibitory control superimposed on affective images, assessing cognitive and affective interference.

Methods:

Functional MRI data were collected from 39 BD patients and 36 healthy controls during performance on the Multi-Source Interference Task overlaid on images from the International Affective Picture System (MSIT-IAPS). Analyses examined patterns of activation in a priori regions implicated in cognitive and emotional processing. Functional connectivity to the anterior insula during task performance was also examined, given this region’s role in emotion-cognition integration.

Results:

BD patients showed significantly less activation during cognitive interference trials in inferior parietal lobule, dorsomedial prefrontal cortex, anterior insula, mid-cingulate, and ventrolateral prefrontal cortex regardless of affective valence. BD patients showed deviations in functional connectivity with anterior insula in regions of the default mode and frontoparietal control networks during negatively valenced cognitive interference trials.

Conclusions:

Our findings show disruptions in cognitive regulation and inhibitory control in BD patients in the presence of irrelevant affective distractors. Results of this study suggest one pathway to dysregulation in BD is through inefficient integration of affective and cognitive information, and highlight the importance of developing interventions that target emotion-cognition integration in BD.

Introduction

Bipolar disorder (BD) is a debilitating illness that often leads to chronic functional impairment and subsequent poor quality of life. Individuals diagnosed with BD struggle with affect lability,^1–5 emotion dysregulation,^6–9 cognitive dysfunction,^10–15 and behavioral impulsivity.^16–19 Whereas emotion dysregulation and cognitive dysfunction in bipolar disorder have been relatively well studied,^20–26 the interaction of these two areas of functional impairment is less well understood, as is the relation to behavioral impulsivity. The ability to adaptively integrate affective and cognitive information and modulate behavior accordingly is a crucial aspect of healthy functioning. Understanding deficits in the interaction of emotion, cognition and behavior in BD patients relative to healthy individuals at the level of neurocircuitry provides important clues to the pathophysiology of BD, with the potential for delineating viable targets for rehabilitative intervention.

The simultaneous processing of cognitive and affective information for behavioral response selection has been shown to recruit distributed networks of cortical regions including dorsal anterior cingulate (dACC), dorsomedial prefrontal cortex (dmPFC), dorsolateral prefrontal cortex (DLPFC), supplemental motor area (SMA), ventrolateral prefrontal cortex (VLPFC), insula and regions of the parietal cortex.^27–31 The dACC and anterior insula, extending to the VLPFC (fronto-insular pole), represent key regions within the salience network implicated in monitoring, detecting and regulating salience signaling from limbic and striatal areas.³² The DLPFC, dmPFC, rostral VLPFC and superior and inferior parietal lobule (SPL, IPL) represent key regions within the fronto-parietal executive control network implicated in the regulation of attention towards or away from salient cues and the maintenance of goal-directed behavior and response selection.^{33, 34} Cognition-emotion integration requires the dynamic coordination and flexible switching between these functional networks.³⁵ A growing literature has identified the anterior insula as a key structure modulating switching between functional networks in service of regulatory goals.^36–38 Examining responses along this broader neurocircuitry in BD patients and the functional connectivity of these network regions to the anterior insula during the simultaneous processing of cognitive and affective information could provide clues to sources of regulatory dysfunction in BD.

Only a handful of studies to date have explicitly examined the integration of cognitive and affective processing in BD. These studies implicate disruptions in broader fronto-parietal and fronto-striatal networks in BD patients relative to controls.^39–41 For example, using an emotional Stroop task, in which participants must indicate the ink color of emotional words, Lagopoulus et al⁴⁰ found healthy controls primarily recruited fronto-parietal control regions including DLPFC and IPL to complete the task, whereas BD patients primarily recruited subcortical limbic regions including amygdala, hippocampus and regions of the caudate. This suggests greater regulatory control during task performance in healthy controls versus greater processing of affective salience in BD patients. Using a similar emotional Stroop task, Mahli et al⁴¹ found significantly weaker fronto-striatal recruitment in BD patients relative to controls, including weaker VLPFC, thalamus, caudate and putamen activation, suggestive of weaker regulatory control of behavioral response selection. BD patients were also significantly slower to respond to stimuli, suggesting impaired attentional control. In a more recent study, Favre et al³⁹ used a modified emotional Stroop task that superimposes emotion words (happy, fear) on images showing happy or fearful facial expressions.⁴² BD patients evidenced significantly reduced DLPFC activation and slower reaction time in response to word and facial expression incongruence relative to healthy controls. Further, functional connectivity analyses showed significantly stronger positive correlations between DLPFC and default mode network regions (DMN; subgenual anterior cingulate cortex, posterior cingulate cortex) in BD patients during incongruence, in contrast to significantly stronger negative DLPFC-DMN functional correlations in healthy controls, suggesting BD patients may have greater difficulty switching from internally focused processing to task-related processing in the presence of competing cognitive-affective demands. Collectively, these studies suggest BD patients exhibit reduced regulatory control and increased salience processing during cognitive tasks that involve the simultaneous processing of affective information.

In the current study, we sought to add to the existing literature on cognitive and emotion integration in BD by examining the ability to engage in a complex cognitive task and modulate behavior accordingly while simultaneously screening out distracting and irrelevant affective information, a process that is crucial for adaptive regulation. In this design, participants are not asked to engage directly with affective stimuli, but instead affective stimuli serve as a background distractor. This is relevant to clinical presentations of BD, wherein patients struggle to override affective information (positive or negative) and inhibit emotion-driven behavioral responses (i.e. impulsivity), in order to remain focused on goal directed or adaptive behaviors. To assess this process, we used a modified version of the Multi-Source Interference Task (MSIT; a well-established paradigm eliciting cognitive interference)^{43, 44} by superimposing task presentations onto affective images selected from the International Affective Picture System (IAPS).⁴⁵

The MSIT was developed to assess cognitive interference, or the ability to overcome task-irrelevant stimuli that distracts from task-relevant stimuli, requiring cognitive control to resolve. Participants are required to override a prepotent motor response by utilizing cognitive information (selective attention, visuospatial processing, decision making), to screen out irrelevant information and inhibit automatic behavior. A recent meta-analysis of MSIT studies in healthy controls found consistently increased activation of the dACC, dmPFC and SMA during interference trials across studies⁴⁶. These findings have been suggested to relate to conflict monitoring (dACC) and motor planning (dmPFC, SMA) in the context of the inhibition of prepotent motor responses in order to successfully overcome interference and select the correct task response. Additionally, increased activation of the right insula, right VLPFC and putamen during interference trials was found across studies, suggesting increased processing of the saliency of interference. Abnormalities in dACC and dmPFC activation during interference trials has been shown across multiple clinical populations including schizophrenia,^{47, 48} major depressive disorder,⁴⁹ post-traumatic stress disorder,⁵⁰ obsessive compulsive disorder,⁵¹ attention deficit hyperactivity disorder ADHD,⁵² and chronic pain.⁵³ A previous study using the MSIT (without affective stimuli) in BD patients found reduced dACC and mid-cingulate activation in BD patients relative to healthy controls.⁵⁴ In addition, BD patients evidenced increased response time and greater errors of omission (missed response) during trials requiring inhibitory control, suggestive of slower cognitive control processing, and greater errors of commission (incorrect response) during control conditions, indicative of greater impulsive responding. Thus, clinical populations including BD patients consistently show abnormal recruitment of fronto-parietal executive control regions during the cognitively demanding interference trials of the MSIT relative to healthy controls, with concomitant behavioral effects evidenced in BD.

Relevant to the current study, there are a few existing studies that have investigated the combination of the MSIT with irrelevant emotional distractors. In a study of healthy females, Jasinska and colleagues⁵⁵ found the additional presence of negatively-valenced distractors (irrelevant flankers depicting facial expressions of fear) during interference trials significantly reduced dACC, DLPFC and insula/VLPFC activation relative to neutral trials, and significantly increased activation in these same regions during the less cognitively demanding non-interference trials. Similar findings for positively valenced trials were found, although to a lesser effect than negative valenced trials, with no significant differences between negative and positive valenced trials. A study of schizophrenic patients using a similarly adapted MSIT paradigm superimposed on irrelevant affective images found reduced DLPFC activation relative to healthy controls during all interference trials (negative or neutral), with strongest effects during negatively valenced interference trials.⁴⁷ These studies suggest the inclusion of irrelevant emotional stimuli has a deleterious effect on trial performance on the MSIT, which may be further accentuated in clinical populations.

In the present study, we used a similarly modified MSIT-IAPS task to examine the ability of BD patients to overcome cognitive interference and inhibit automatic behavioral responses while affective systems are activated and competing for limited resources. We hypothesized BD patients would demonstrate significantly reduced recruitment of frontoparietal control and striatal regions relative to controls during interference trials on the modified MSIT-IAPS task in the context of negative and positive valenced images. Additionally, we hypothesized BD patients would evidence weaker functional connectivity between anterior insula and regions of the executive control network, and stronger functional connectivity between anterior insula and salience network regions during performance on cognitive interference trials in the presence of affective distractors. Thus, we hypothesized BD patients would exhibit under-reliance upon neurocircuitry related to regulatory control and over-reliance upon salience circuitry while attempting to regulate cognition and inhibit behavior in the context of irrelevant affective information. Further, we hypothesized BD patients would evidence slower reaction times and decreased task accuracy in the context of emotional stimuli relative to neutral stimuli.

Methods

Participants

Study participants were 39 individuals with DSM-IV bipolar I disorder, (mean age 36.69 ± 12.91; 16 female; Table 1) recruited through the Dauten Family Center for Bipolar Treatment Innovation at the Massachusetts General Hospital, and 36 healthy control participants (mean age 36.69 ± 12.91; 18 female). All participants provided written informed consent prior to participation, in accordance with the guidelines of the Internal Review Board of the Massachusetts General Hospital. Eligible participants were English speakers, 18–65 years old. Patient participants were required to meet DSM-IV criteria for bipolar I disorder, current mild to moderate depressive symptoms, and be stable pharmacotherapy (no dose change within the past 8 weeks). Exclusion criteria included diagnosis of rapid cycling bipolar disorder subtype or current mixed episode, current or past psychosis or schizophrenia spectrum disorder, or lifetime substance/alcohol use disorder. Additional exclusion criteria for all participants (patients and controls) included presence of serious medical conditions, history of neurological disorder, history of moderate to severe head trauma, as well as standard fMRI contraindications (e.g. pregnancy, claustrophobia, non-removable metallic objects).

Table 1.

Study demographics.

	BD (n= 39)	HC (n = 36)
Measure	n (%)	n (%)	Sig.
Gender			0.44
Male	23 (59.0)	18(50.0)
Female	16(41.0)	18(50.0)
Age	36.69 ± 12.92	34.69 ± 12.64	0.33
Race			0.49
White	35 (89.7)	30(83.3)
Black	1 (2.6)	3 (8.3)
East Asian	1 (2.6)	3 (8.3)
Other or unreported	2(5.1)	-
Ethnicity			0.60
Hispanic/Latino	1 (2.6)	2 (5.6)
Not Hispanic/Latino	38 (97.4)	34 (94.4)
Highest Level of Education			0.67
Doctorate	3 (8.3)	-
Masters	2 (5.6)	6(15.8)
Bachelors	20 (55.6)	23 (15.8)
Associate	5(13.9)	1 (2.6)
Some College	1 (2.8)	8(21.1)
High School/GED	5(13.9)	-
Clinical Measures
HAM-D-17	13.49(7.37)	1.08(1.65)	<0.001
YMRS	5.15(4.98)	0.56(1.16)	<0.001
BDI	19.28(11.53)	2.52 (4.79)	<0.001

Note. Non-parametric tests for significance tested using Mann-Whitney U test.

Clinical measures

The presence of current mood symptoms was assessed using well-validated clinician-administered and self-report measures. Current presence of depressive symptoms was measured using the Hamilton Depression Rating Scale, 17-item version (HAM-D-17)⁵⁶ and the Beck Depression Inventory (BDI-II).⁵⁷ Current manic or hypomanic symptoms were measured using the Young Mania Rating Scale (YMRS).⁵⁸

fMRI Paradigm

MSIT-IAPS task.

All participants completed an affective version of the MSIT (MSIT + IAPS, Figure 1) while undergoing fMRI. The MSIT paradigm was set up in a rapid-event related design. The MSIT incorporates aspects of well-established measures of cognitive interference (Stroop; Simon, and Eriksen Flanker tasks), and uses two different types of cognitive interference (spatial and flanker) to measure cognitive control.^{43, 44} Each trial of the MSIT was overlaid on an image from the International Affective Picture System (IAPS).⁴⁵ IAPS pictures were either neutral, positive, or negative valenced, counterbalanced between the control and interference conditions. During each trial of the MSIT, a three-digit number (comprised using the numbers 0, 1, 2, or 3) was presented for 1.7 seconds on the screen. Each set contains two identical distractor numbers and a target number that differed from the distractors. Participants report via a button press the identity of the target number that differs from the two distractor numbers (Figure 1). During Noninterference (control) trials, distractor numbers are always zeros, and the identity of the target number always corresponds to its position on the button response pad (100, 020, 003). By contrast, during Interference trials, distractor numbers are always numbers other than 0, and the identity of the target number is always incongruent with its position on the button response pad (e.g. 211, 232, 331, etc.). Trial stimuli were presented on the screen for 1.7s, followed by an inter-trial interval (ITI) fixation cross of varying lengths (Figure 1). The trial and ITI sequence was determined using Optseq2 (http://surfer.nmr.mgh.harvard.edu/optseq).⁵⁹ Trials were analyzed to examine the main effect of interference (All Interference – All Noninterference), the main effect of valence (All Negative – All Neutral; All Positive – All Neutral), and the interaction of interference and valence (Negative Interference – Negative Non-Interference; Positive Interference – Positive Non-Interference; Negative Interference – Neutral Interference; Positive Interference – Neutral Interference; Negative Non-Interference – Neutral Non-Interference; Positive Non-Interference – Neutral Non-Interference). Task behavioral performance was analyzed by calculating the average response time (reaction time, in milliseconds) and percentage of correct responses (accuracy) for each trial of interest.

Figure 1.

Illustration of MSIT-IAPS task trials, which consist of the Multisource Interference Task (MSIT) overlaid on images from the International Affective Picture Scale (IAPS). A) Example of a negatively valenced “Non-Interference” trial. The target number “2” is in the same position as the second finger on the number pad. B) Example of a positively valenced “Interference” trial. The target number “2” is in a different position than the second finger on the number pad. In the “Interference” trial, the prepotent response to press the first finger (matching the position of the target number) is inhibited in order to make the correct selection with the second finger. Trials not shown are “Positive Non-Interference,” “Neutral Non-Interference,” “Negative Interference,” “Neutral Interference.”

MRI scanning

MRI data were acquired using a 3.0-T whole-body scanner (Trio-System), equipped for echo planar imaging (Siemens Medical Systems, Iselin NJ) with a 3-axis gradient head coil. Head movements were restricted using foam cushions. Images were projected using a rear projection system and E-Prime stimulus presentation software was used to show the task stimuli (Psychology Software Tools; http://www.psychology-software-tools.mybigcommerce.com). Following automated scout and shimming procedures, two high-resolution 3D MPRAGE sequences (TR=2530ms, TE=3.39ms, flip angle=7^o, voxel size = 1.3×1.0×1.3 mm) were collected for positioning of subsequent scans. FMRI images (i.e. blood oxygenation level dependent signal or BOLD) were acquired using T2* -weighted sequence (27 axial slices aligned perpendicular to the plane intersecting the anterior and posterior commissures, 5 mm thickness, skip 1 mm, TE=30ms, TR=1600ms, flip angle = 90^o).

MRI Data Analysis

Functional Data were processed using SPM8 software (Wellcome Department of Cognitive Neurology, London, UK; www.fil.ion.ucl.ac.uk/spm ). Each individual’s FMRI images were slice time corrected using slice number 7 as a reference, motion corrected using 2^nd degree B-spline interpolation, co-registered to the T1 MPRAGE sequence and segmented into white, grey and CSF, spatially normalized to the standardized space established by the Montreal Neurologic Institute (MNI; www.bic.mni.mcgill.ca), resampled to 2mm³ voxels (anatomical images were resampled to 1mm³ voxels), and smoothed with a three-dimensional Gaussian kernel of 6 mm width (FWHM). All collected data had minimal head motion (< 3 mm linear movement in the orthogonal planes; <0.5 degrees radians of angular movement). The general linear model was applied to the time series⁶⁰, convolved with the canonical hemodynamic response function and a 128s high-pass filter. Serial autocorrelations were addressed with an AR(1) model. Movement parameters, derived from motion correction during preprocessing, were included in the model as regressors of no interest. For each subject, condition effects were modeled with the SPM canonical hemodynamic response function. For each subject (individual subjects’ analysis), condition effects were estimated at each voxel, and contrast images were produced for pre-specified contrasts of interest (see above). Individual subject contrast images were entered into a 2^nd Level random effects flexible factorial for group-level comparisons of contrast effects in a priori specified regions of interest (ROIs), selected based upon prior reported findings of cognitive control and emotion regulation (see Introduction). To control for residual depressive and manic symptoms, scores on the HAM-D-17 and YMRS were entered as covariates. Specific ROIs included IPL, DLPFC, VLPFC, dorsal and ventral medial PFC (dmPFC, vmPFC), cingulate cortex, anterior insula, caudate, hippocampus, and amygdala. ROI masks were created using Anatomical Automatic Labeling tool⁶¹ implemented in the WFU Pickatlas (http://www.ansir.wfubmc.edu, see Supplement, Table S1).^{62, 63} Contrast results were examined at a small volume corrected (SVC) threshold of p<.005. Significant SVC results with a family-wise error (FWE) corrected p-value less than .05 were followed up for further analysis. The linear time series from significant contrasts that survived this additional FWE correction were extracted for further analysis using MarsBar.⁶⁴

To assess functional connectivity differences as a function of task contrast, generalized PPI analyses⁶⁵ were conducted in SPM8. At the individual subject level, condition contrasts of interest were modeled for main effects of cognitive interference and valence (Interference – Non-Interference trials; Negative – Neutral trials; Positive – Neutral trials), as well as the interaction of cognition and valence (Negative Interference – Negative Non-Interference; Positive Interference – Positive Non-Interference; Negative Interference – Neutral Interference; Positive Interference – Neutral Interference). A volume of interest (VOI) in the right anterior insula was selected as the seed region. Center-of-sphere for the anterior insula seed VOI was defined based upon results of the group-level GLM (right anterior insula, x y z = 44 20 2; see Table 2). The first eigenvariate was extracted by creating a spherical VOI with a radius of 6mm for each condition contrast of interest. Resulting individual contrast images were then entered into a 2^nd Level GLM for group analysis in SPM. Functional connectivity between the anterior insula seed region and our a priori ROIs specified above was examined (IPL, DLPFC, VLPFC, dmPFC, vmPFC, cingulate cortex, caudate, hippocampus, and amygdala). Results of all group analyses were examined with a voxelwise threshold of p<.005. The linear time series from significant contrasts with a FWE corrected p-value less than .05 were extracted for further analysis using MarsBar.⁶⁴

Table 2.

Behavioral results.

Trial	BD	HCs			95% CI
	EMM (SE)	EMM (SE)	Mean Diff.	Sig.a	Lower	Upper
Reaction time (ms)
All Interference	868 (20)	820 (21)	48.0	0.17	−20.8	116.7
All Non-interference	659 (20)	609(21)	49.0	0.16	−20.3	118.4
All Neutral	762 (20)	712(21)	50.4	0.15	−18.0	118.9
All Positive	762 (20)	713 (20)	48.9	0.15	−17.6	115.5
All Negative	765 (20)	719(21)	46.1	0.18	−21.4	113.6
Neutral Interference	870(21)	819(22)	51.1	0.16	−20.3	122.5
Positive Interference	868 (20)	824 (21)	43.8	0.20	−23.8	111.5
Negative Interference	865 (20)	816(21)	49.0	0.16	−19.4	117.3
Neutral Non-interference	655 (20)	605(21)	49.8	0.16	−19.6	119.2
Positive Non-interference	657 (20)	603(21)	54.1	0.13	−15.3	123.5
Negative Non-interference	664(21)	621 (21)	43.2	0.22	−26.8	113.3
Accuracy (%)
All Interference	82.3 (2.5)	91.3(2.6)	−8.9	0.04	−17.5	−0.4
All Non-interference	96.5(1.0)	99.5(1.0)	−6.0	0.07	−6.3	0.2
All Neutral	90.0(1.5)	95.6(1.6)	−5.6	0.04	−10.8	−0.4
All Positive	89.8(1.6)	95.7(1.7)	−5.9	0.04	−11.5	−0.4
All Negative	88.3 (1.8)	94.8(1.9)	−6.5	0.04	−12.5	−0.4
Neutral Interference	83.5 (2.4)	91.7(2.5)	−8.2	0.05	−16.3	0.0
Positive Interference	83.1 (2.6)	91.6(2.7)	−8.4	0.06	−17.2	0.4
Negative Interference	80.3 (2.8)	90.5 (2.9)	−10.2	0.03	−19.6	−0.9
Neutral Non-interference	96.5 (1.0)	99.5(1.0)	−3.0	0.07	−6.2	0.2
Positive Non-interference	96.5 (1.0)	99.9(1.0)	−3.4	0.05	−6.7	−0.1
Negative Non-interference	96.4(1.1)	99.1(1.1)	−2.7	0.15	−6.4	1.0

Note: Results of MANCOVA, controlling for residual depression and mania symptoms. EMM (SE) = estimated marginal means and standard errors, derived from omnibus MANCOVA model.

^aBonferroni corrected.

Behavior Analysis

Behavioral performance on the MSIT-IAPS task was assessed by calculating the mean reaction time to response score (reaction time) and percentage of accurate responses made (accuracy). To assess differences in task performance between groups, we conducted a multivariate analysis of covariance (MANCOVA) controlling for baseline symptom severity (HAM-D-17, BDI-II, YMRS) and medication load (see below). Follow up pairwise t-tests were conducted and Bonferroni corrections were applied to control for multiple comparisons.

To assess the relationship between task-related regional activation and functional connectivity and behavioral performance, a series of bivariate, 2-tailed Pearson’s correlations were conducted in SPSS (version 20.0; IBM) between behavioral response time and accuracy scores and extracted linear time series from GLM and gPPI modeled contrasts. To correct for multiple comparisons, 95% confidence intervals for the results of correlational analyses were calculated using an iterative bootstrap method (resample and replace, 1000 samples) in SPSS.

Medication effects

The effects of psychotropic medications (medication load) on task performance and MRI results were assessed using an established approach by Phillips and colleagues.^66–68 For each participant, the dose of each class of medication (antidepressants, mood-stabilizers, antipsychotics, and anxiolytics) is coded as absent (0), low (1), or high (2) using the dosing guidelines, and a participant’s total medication load is reflected as a sum of these scores (range = 0–8). To assess the effects of medication, individual medication load scores were regressed on measures of task behavior (reaction time, accuracy) and significant findings from the ROI contrast analyses and gPPI results.

Results

Demographics

Complete participant demographics are presented in Table 1. No significant differences were found between healthy control and BD patient groups in age, sex, ethnicity, race, or years of education. BD participants on average scored in the mild to moderate range on measures of depressive symptoms (HAM-D-17; BDI), and in the range of remission on a measure of manic symptoms (YMRS). Healthy controls were asymptomatic across all measures. Medication load for BD patients ranged from 0–8, with a mean of 2.94 (±2.20), suggesting low to moderate medication load on average.

Behavioral Results

Table 2 shows complete behavioral results across all trial types. Data from one BD participant failed to record during scanning, therefore results presented are for the remaining 74 participants. Results from the omnibus MANCOVA controlling for residual depression and mania symptoms showed a significant difference between groups for accuracy (F(2, 37) = 2.39; p = .03, η² = .22) but not reaction time (F(2, 37) = 2.06; p = .16, η² = .03). Averaging across all trial types, BD participants committed significantly more response errors relative to healthy controls, even after controlling for residual symptoms (Mean_diff (SE) = −5.99 (2.75), p = .03 Bonferroni corrected). Looking at specific trial types, BD participants committed significantly more errors across all three valence presentations regardless of interference [all neutral: Mean_diff (SE) = −5.58 (2.62), p = .04 Bonferroni corrected; all negative: Mean_diff (SE) = −6.48(3.04), p = .04 Bonferroni corrected; all positive: Mean_diff (SE) = −5.91(2.79), p = .04 Bonferroni corrected]. The strongest effect was seen during negative valenced Interference trials (Mean_diff (SE) = −10.24 (4.71), p = .03 Bonferroni corrected).

fMRI Results

Effect of Cognitive Interference (Interference vs. Non-Interference trials).

Results of Interference vs. Non-Interference trials are summarized in Table 3. Across all Interference – Non-Interference trials regardless of valence (main effect Interference vs. Non-Interference), bipolar patients showed significantly less activation of right dmPFC, right IPL, right anterior insula, bilateral mid cingulate and right VLPFC, and significantly greater activation of right DLPFC and right posterior cingulate cortex (PCC) relative to HC participants (Table 3, Figure 2). Bipolar patients also evidenced significantly less activation of left IPL [t(365) = 3.23, p=.001 (p = .30 FWE)], left insula [t(365) = 3.47, p<.001 (p = .18 FWE)], and left VLPFC [t(365) = 3.55, p<.001 (p = .30 FWE)], but these did not survive corrections for multiple comparisons. No effects surviving uncorrected or corrected significance thresholds were found when examining only Negative Interference – Negative Non-Interference trials or only Positive Interference – Positive Non-Interference trials.

Table 3.

fMRI contrast results.

		MNI Coordinatesa
Region	BA	x	y	z	Vol (mm3)	t−score	p (FWE)
Interference > Non-interference
HCs > BD
R. dmPFC	6	2	4	62	1224	5.57	<.001***
R. IPL	40	46	−44	38	248	5.17	<.001***
R. Anterior Insula	13/47	44	20	2	1000	3.53	0.05**
L. Mid Cingulate	8/32	−14	14	46	888	4.7	0.01**
R. Mid Cingulate	8/32	12	16	44	1216	3.78	0.07**
R. VLPFC	44	48	4	20	1128	3.78	0.07**
BD>HCs
R. DLPFC	9	44	22	30	848	4.10	0.05*
R.PCC	23	4	−28	−38	912	3.76	0.05*

Note. Table shows family-wise error (FWE) corrected p-values, controlling for residual depression and mania symptoms. BD = bipolar patients. HC = healthy controls. DLPFC = dorsolateral prefrontal cortex. dmPFC = dorsomedial prefrontal cortex. IPL = inferior parietal lobule. PCC = posterior cingulate cortex. VLPFC = ventrolateral prefrontal cortex.

^*uncorrected p <.01.

^**uncorrected p <.001.

^***uncorrected p<.0001.

Figure 2.

Regions of significantly greater activation in healthy controls relative to bipolar patients during Interference > Non-Interference contrast conditions. Regions include (a) right dmPFC (x=2 y=2 z=60); (b) right inferior parietal lobule (x=46 y=−44 z=38); and (c) right anterior insula (x=44 y=20 z=2). Images displayed at p <.005 uncorrected, whole brain.

Effect of Valence (Negative, Positive vs. Neutral trials).

Across all Negative – Neutral trials regardless of interference, BD participants showed significantly greater DLPFC and VLPFC activation, but these results did not survive FWE correction for multiple comparisons [left DLPFC: t(365) = 3.59, p <.001 (p = .41 FWE); right DLPFC: t(365) = 3.56, p <.001 (p = .44 FWE); left VLPFC: t(365) = 3.22, p = .001 (p = .60 FWE); right VLPFC: t(365) = 3.07, p = .001 (p = .75 FWE)]. No effects surviving uncorrected or corrected significance thresholds were found for Positive – Neutral trials.

Functional connectivity (gPPI).

Results of gPPI analysis are summarized in Table 4. Because we did not find any significant effects of positive valence trials, gPPI analyses included only negative and neutral valence trials. Significant differences in functional connectivity were found in BD patients relative to controls during Negative Interference – Negative Non-Interference trials between the right anterior insula seed region and bilateral dorsal anterior cingulate (dACC; Figure 3a), left dmPFC, bilateral DLPFC, bilateral vmPFC, bilateral IPL, bilateral caudate, and bilateral hippocampus (Table 4). Across all anterior insula seed- ROI target pairs, BD patients evidenced positively correlated activation, whereas healthy controls evidenced negatively correlated activation (See supplement, Figure S1).

Table 4.

fMRI gPPI results.

		MNI Coordinatesa
Region	BA	x	y	z	Vol (mm3)	t−score	p (FWE)b
Interference > Non-interference
HCs > BD
No significant results at FWE <.10
BD > HCs
R. Hippocampus		34	−26	−14	56	3.73	0.05***
Negative > Neutral
HCs > BD
No significant results at FWE <.10
BD > HCs
L. IPL		−30	−50	46	147	4.06	0.03***
R. dACC		14	28	28	48	4.75	0.002***
Negative Interference > Negative Non-Interference
HCs > BD
No significant results at FWE <.10
BD>HCs
R. Hippocampus		34	−22	−14	217	5.06	<0.000
L. Hippocampus		−34	−38	−4	56	4.27	0.008
R. dACC	32	12	28	20	548	5.03	<0.000
L. dACC	10	−12	52	2	83	4.71	0.002
L. dmPFC		−14	44	26	814	5.28	0.001
R. Caudate		14	14	14	274	4.69	0.001
L. Caudate		−16	22	10	132	4.48	0.003
L. DLPFC		−32	2	50	158	5.05	0.003
R. DLPFC		32	26	34	1416	4.85	0.006
R. IPL		56	−44	46	271	4.46	0.007
L. IPL		−40	−36	38	173	4.16	0.02
R. vmPFC		16	52	−2	100	4.37	0.004
L. vmPFC		−12	52	−2	34	3.96	0.02
Negative Interference > Neutral Interference
HCs > BD
No significant results at FWE <.10
BD>HCs
R. dmPFC		26	6	44	147	5.13	0.002
L. dmPFC		−12	14	48	333	4.93	0.005
R. Hippocampus		26	−18	−14	107	4.54	0.003
L. SPL		−30	−50	44	329	5.73	<0.000
L. IPL	40	−44	−34	36	124	4.05	0.03
R. dACC		14	28	28	48	4.75	0.002

Note. Table shows family-wise error (FWE) corrected p-values. BD = bipolar patients. HC = healthy controls. dACC = dorsal anterior cingulate. DLPFC = dorsolateral prefrontal cortex. dmPFC = dorsomedial prefrontal cortex. IPL = inferior parietal lobule. vmPFC = ventromedial prefrontal cortex.

Figure 3.

Functional connectivity of right anterior insula and right dACC in bipolar patients relative to healthy controls during Negative Interference > Negative Non-Interference trials. a) Right anterior insula – right dACC functional connectivity maps showing dACC max voxel (x=12 y=28 z = 20). Results displayed at p < .005 uncorrected. b) Correlations between reaction time during negative interference trials and right anterior insula – right dACC functional connectivity. dACC = dorsal anterior cingulate. c) Correlations between reaction time during negative interference trials and right anterior insula – right DLPFC functional connectivity. DLPFC = dorsolateral prefrontal cortex.

Effects of Medications

Medication load in BD patients significantly negatively predicted bilateral anterior insula activation (left: β = −.49, p = .002; right: β = −.37, p = .03). No other significant correlations between medication load and regional activation, functional connectivity, or task performance were found.

Relationship to Behavior

Activation in the left IPL during Interference trials was significantly negatively correlated with reaction time for BD patients only (r = −.33, p =.04, Bootstrap 95% CI: −0.70, −0.08), such that greater activation of the left IPL corresponded to faster reaction time. No other significant relationships between task performance and ROI activations were found. For comparisons with gPPI results, functional connectivity between the right anterior insula and right dACC during Negative Interference > Negative Non-Interference trials was significantly positively related to reaction time for BD patients only, such that stronger positive functional connectivity was related to slower reaction times (r =.35 p =.03, Bootstrap 95% CI: 0.12, 0.65; Figure 3b). Functional connectivity between right anterior insula and right DLPFC during Negative Interference > Negative Non-Interference trials was significantly negatively associated with response accuracy in BD patients only, such that stronger positive functional connectivity was related to less response accuracy (r = −.43, p =.009, Bootstrap 95% CI: −0.66, −0.09; Figure 3c). For healthy controls only, right anterior insula and right hippocampus functional connectivity during Negative Interference > Negative Non-Interference trials was significantly negatively correlated with reaction time (r = −.52, p = .002, Bootstrap 95% CI: −0.77, −0.20) and significantly positively correlated with response accuracy (r = .45, p = .008, Bootstrap 95% CI: 0.15, 0.73), such that stronger negative insula-hippocampus functional connectivity was related to slower reaction times and greater accuracy (see Supplement, Figure S2). The same pattern was found for right anterior insula – right vmPFC functional connectivity for healthy controls only (reaction time: r = −.43 p =.01, Bootstrap 95% CI: −0.65, −0.18; accuracy: r = .43, p =.01, Bootstrap 95% CI: 0.19, 0.64; Supplement, Figure S3).

Discussion

In this study, we sought to examine the integration of cognition, emotion, and behavioral inhibition in patients with BD using a complex cognitive task superimposed on irrelevant affective stimuli. Consistent with our first hypothesis, BD patients showed decreased regional activations relative to healthy controls across a distributed set of fronto-parietal executive control regions during cognitive interference, including right IPL, right dmPFC/bilateral mid-cingulate, and right VLPFC. These effects were found across all interference trials regardless of the valence of affective stimuli, and are consistent with a previous study using the original (non-affective) MSIT in BD. However, in addition to reduced recruitment of cognitive control, decreased activation of the right anterior insula (salience network) and increased activation of both the posterior cingulate cortex (default mode network), and right DLPFC (executive control network), was also found in BD patients relative to healthy controls across interference trials. This suggests the addition of irrelevant affective information further impairs BD performance on a complex cognitive task, potentially through increased self-monitoring (posterior cingulate), increased compensatory control (DLPFC) and decreased recruitment of a key region implicated in flexibly switching between affective and cognitive processing modes (anterior insula).

Consistent with our second hypothesis, BD patients showed stronger functional connectivity between anterior insula and the dACC, a region of the salience network, with BD patients evidencing positive functional connectivity and healthy controls evidencing negative functional connectivity. Further, this connectivity was significantly positively related to reaction time in BD patients, such that stronger positive functional connectivity was associated with slower (less efficient) reaction times. However, contrary to hypotheses, stronger positive functional connectivity was also found between anterior insula and several fronto-parietal executive control network regions (bilateral IPL, DLPFC, dmPFC) relative to healthy controls, and also in the same direction (positive connectivity in BD patients, negative connectivity in healthy controls; see Supplement, Figure S1). Further, these functional connectivity effects were found only during negatively valenced trials, and most strongly during negative interference trials. Collectively, these results are interesting, in that, prima facie, they seem counterintuitive. One might conclude positive functional connectivity between regions to imply increased efficiency, or to be related to increased causal (i.e. inhibitory) connectivity. However, an alternate interpretation of increased positive functional connectivity strength argues for decreased efficiency, insomuch as prolonged concurrent activation of two regions might imply failure of one region to exert effects on the corresponding region. Indeed, increased positive anterior insula-DLPFC functional connectivity, a key node of the frontoparietal control network, was significantly associated with decreased rather than increased response accuracy. Hence, increased positive functional connectivity strength, which is a measure of correlated activity over time, might imply less efficient regulation of otherwise correlated regions, with detrimental behavioral consequences. A more precise understanding of the clinical implications of positive versus negative functional connectivity patterns is a research area in particular need of further attention and would help clarify the results presented here. Overall, results from the current study suggest the ability to adaptively utilize frontoparietal control systems rather than salience or default mode systems in order to disengage from irrelevant emotional stimuli and meet task demands is compromised in BD patients.

Our third hypothesis, that BD patients would evidence slower reaction times and decreased task accuracy in the context of emotional stimuli relative to neutral stimuli, was only partially supported. After correction for multiple comparisons, we found no significant differences in reaction times across all task conditions between BD patients and healthy controls. This is in contrast to the findings of Gruber et al.,⁵⁴ using the original MSIT task, in which BD patients exhibited significantly slower reaction times across interference trials relative to healthy controls. The difference in findings may be accounted for by the presence of irrelevant emotional stimuli in the modified task, which may affect performance in healthy control subjects and reduce differences in reaction time between healthy controls and BD patients below a statistically significant threshold. However, in the current study, behavioral effects were calculated controlling for residual symptoms of depression and mania, and more stringent corrections for multiple comparisons were applied, which may also account for the differences between studies. Indeed, in the absence of these controls, BD patients evidenced significantly slower reaction times relative to healthy controls across all task conditions (all p’s <.05). Therefore, it is unclear whether the presence of emotional distractors, the presence of residual mood symptoms, or the interaction of both ultimately impacts performance speed in BD patients.

We did find significant differences in response accuracy across all negative and neutral interference trials, which is in keeping with previous findings of reduced accuracy during interference.⁵⁴ Interestingly, we did not find a significant difference in accuracy during positively valenced interference trials, but we did find a significant difference in accuracy during positively valenced non-interference trials. This may be due to the salience of interference trials themselves – the more difficult and challenging interference trials may evoke distracting negative affect regardless of the presence of irrelevant affective stimuli, and this negative affect may override the distracting effect of positively valenced stimuli. In the absence of this additional cognitive challenge (i.e. non-interference trials), the presence of positive stimuli may have a stronger distracting effect. Overall, future replication studies are needed to disambiguate these collective behavioral effects.

Unlike previous studies using emotional Stroop-like tasks to measure interference,⁴⁰ we did not find significantly greater regional limbic (amygdala, hippocampus) activation in BD patients relative to controls across any comparisons. However, the current study did find significant differences in functional connectivity patterns between anterior insula and limbic regions (hippocampus, caudate) in BD patients relative to healthy controls during negatively valenced interference trials. As was the case with anterior insula connectivity to salience network and fronto-parietal executive control network regions, BD patients again evidenced positively correlated functional connectivity between anterior insula and these regions, whereas healthy controls evidenced negatively correlated functional connectivity (see Supplement, Figure S1). In addition, consistent with results found by Favre and colleagues³⁹ reported above, BD patients evidenced positively correlated functional connectivity between anterior insula and the ventromedial PFC, a key region of the default mode network, whereas healthy controls exhibited negative functional connectivity. These results suggest regional limbic reactivity to irrelevant emotional stimuli may not differentiate BD patients from controls per se, but rather increased synchronous engagement of salience and default mode network regions may compromise disengagement from irrelevant affective stimuli.

A few additional outcomes of the current study merit mention. First, results from the current study contribute to a growing literature implicating the IPL in BD.^69–71 Deviations in IPL activation between BD patients and healthy controls during cognitive interference emerged as one of the strongest regional activation effects, alongside dmPFC activation. Further, decreased functional connectivity between anterior insula and the IPL was significantly associated with deficits in behavioral task performance. These findings are intriguing in light of recent research implicating dysfunctional IPL recruitment as a potential genetic risk factor for BD.⁷¹ Specifically, in a study of BD patients and their first degree relatives utilizing the Stroop task, Pompeii and colleagues found abnormalities in IPL activation distinguished BD and their unaffected relatives from controls.⁷⁰ Data from our own lab show abnormal resting-state functional connectivity between right anterior insula and the IPL in BD patients distinguishes them from unipolar depressed and healthy controls after controlling for medication effects.⁶⁹ As a key node in the fronto-parietal executive control network, the IPL has been shown to play a major role in emotion-cognition integration and is an important regional hub within an integrative multi-network system implicated in orienting attention and regulating cognitive resources in response to salient stimuli.^72–74 The IPL also shares strong functional connections to anterior insula, VLPFC, and dmPFC and is strongly activated in tasks requiring simultaneous processing of interoceptive or affective cues and cognitive demands.^{37, 75} Thus, the decreased recruitment of IPL and corresponding deficits in task performance found in the current study further implicates this region in the dysfunctional cognitive-affective integration and regulation seen in BD.

Second, whereas behavioral and neural responses to negative valenced stimuli differentiated BD patients from healthy controls, we did not find effects for positive valenced stimuli. This may be due to the nature of the positive stimuli used in the present study. Specifically, dysfunction related to positive valenced stimuli in BD primarily relates to over approach to appetitive stimuli in the context of increased arousal. The positively valenced stimuli in the present study (positive images) may have failed to elicit a similar autonomic response in BD patients; therefore, meaningful neural differences in the context of positive stimuli may not have been detected.

There are limitations to the interpretation of the study results presented here. First, BD patients include a heterogeneous mix of BD symptom presentations, ranging from mildly to moderately depressed. In addition, the current study included only BD-I patients. Therefore, it is unclear whether differential results would be found when examining BD-I versus BD-II patients or in the absence of residual depressive symptoms. Second, the current study relied on a more conservative ROI approach. Examining task effects in whole-brain functional networks would help to clarify results presented here. Third, as mentioned above, affective images have somewhat limited ecological validity in eliciting emotional responses, and the stimuli here was more effective at eliciting reactions to negative rather than positive stimuli. Future studies using more ideographic or personally relevant stimuli may improve the ecological validity of the affective component of the task.⁷⁶

Conclusions

Understanding the neural correlates of cognition-emotion integration in BD is important not only for the purposes of shedding light on the pathophysiology of BD but also to aid in the identification of potential treatment targets. The current study contributes to existing studies of cognition-emotion integration in BD by examining behavioral and neural responses to a task requiring attention to ongoing task goals and the inhibition of prepotent behavioral responses in the context of irrelevant emotional stimuli. Results of our analyses found BD patients less able to recruit regions implicated in cognitive control in order to accomplish task goals, and concurrent deficits in task performance. Further, deviations in functional connectivity between the anterior insula and regions of all three major functional networks (salience, default mode, executive control) suggest overall inefficiency in recruiting frontoparietal executive control networks to disengage from irrelevant salient stimuli. Results from the present study add to existing literature suggesting a particular role of the IPL in cognitive-affective integration and cognitive control in BD. Future studies are needed to understand the relationship between patterns of functional connectivity identified in the current study and broader intra- and inter-network connectivity in order to identify potential targets for rehabilitative intervention, such as non-invasive neuromodulation.