Reduced theta-band neural oscillatory activity during affective cognitive control in bipolar I disorder

Abstract

Individuals with bipolar I disorder (BD) have difficulty inhibiting context-inappropriate responses. However, neural mechanisms of impaired cognitive control over impulsive behaviors, especially in response to emotion, are unclear. Theta-band neural oscillatory activity over midfrontal areas is thought to reflect cognitive control. The current study examined behavioral performance and theta-band activity during inhibition to affective stimuli in BD, relative to healthy control participants (HC). Sixty-seven participants with BD and 48 HC completed a Go/No-Go task with emotional face stimuli during electroencephalography (EEG) recording. Behavior was measured with reaction time, discriminability (d’) and response bias (β). Time-frequency decomposition of EEG data was used to extract event-related theta-band (4–7 Hz) neural oscillatory power and inter-trial phase consistency (ITPC) over midline fronto-central areas. Behavior and theta-band activity were compared between groups, while covarying for age. Participants with BD exhibited slower response execution times on correct Go trials and reduced behavioral discrimination of emotional versus neutral faces, compared to HC. Theta-band power and ITPC were reduced in BD relative to HC. Theta-band power was higher on No-Go trials than Go trials. The magnitude of differences in theta-band activity between Go/No-Go trial types did not differ between groups. Increased theta-band power was associated with faster response execution times, greater discrimination of differing facial expressions, and stronger tendency to respond both across the full sample and within the BD group. Attenuated midline fronto-central theta-band activity may contribute to reduced cognitive control and maladaptive behavioral responding to emotional cues in individuals with BD.

1. Introduction

Bipolar disorder (BD) is characterized by impulsive behavior, particularly in response to emotion (Johnson, 2018). Impaired cognitive control is a key process proposed to contribute to these behaviors (Johnson et al., 2018). Cognitive control processes involve adapting behavior flexibly and meeting goals with respect to the current context. This includes response inhibition, which refers to the ability to suppress a prepared or habitual response when it is inappropriate for a given context (Miyake et al., 2000). Individuals with BD show impaired inhibition of context-inappropriate responses on behavioral tasks (Gotra et al., 2020; Wright et al., 2014). These impairments often persist into remission from mood episodes and are associated with poor psychosocial functioning (Depp et al., 2012; Mora et al., 2013). Therefore, examining inhibitory control processes in response to emotion can advance our understanding of disease mechanisms of BD.

Cognitive control can be probed using a Go/No-Go response inhibition task, in which participants are instructed to respond as quickly as possible to frequent target stimulus (Go) and inhibit this prepared response when shown an infrequent non-target stimulus (No-Go). This engages both proactive and reactive cognitive control processes. Proactive control refers to a sustained process of maintaining task goals in anticipation of need for control (e.g., anticipating each upcoming trial could require response inhibition), while reactive control involves detecting and resolving interference following a stimulus (e.g., reacting to No-Go cues which signal the need to inhibit; Braver, 2012; Messel et al., 2021). Individuals with BD show impairments on affective and non-affective Go/No-Go tasks, including slowed reaction times and increased rates of failure both to execute responses on Go trials and to inhibit responses on No-Go trials (Wright et al., 2014). However, the neural underpinnings of these failures are not well understood. Electroencephalography (EEG) is well-suited to capture neural activity during rapid cognitive control processes engaged in a Go/No-Go task, even in the absence of a behavioral response (e.g., when successfully inhibiting a response).

One electrophysiological index of cognitive control is the N200 event-related potential (ERP) component (Folstein and Van Petten, 2008). Some evidence supports reduced N200 amplitudes in BD during response inhibition during non-affective tasks (e.g. letter stimuli) (Michelini et al., 2016; Van Voorhis et al., 2019). However, our initial examination of ERPs during an affective Go/No-Go task (using emotional face stimuli) in BD did not find such group differences (Menkes et al., 2022). The inconsistency of N200 findings may be due to methodological differences (e.g., task design, analytic approach). To further reveal neural dynamics that may be disrupted in BD during affective response inhibition, EEG data can be re-analyzed in the time-frequency domain. Time-frequency decomposition of EEG data allows for examining neural oscillations, which are thought to be more fundamental physiological mechanisms involved in various aspects of neural functioning relative to ERPs (Cohen, 2014). Neural oscillations are also studied more broadly across neuroscience, including in nonhuman animals (Cohen 2014). This makes oscillations more promising for translational work from other species to humans. Therefore, examining spectral features of neural oscillations that are not captured in ERPs can offer more insight into the neural underpinnings of cognitive control impairments in BD. These spectral features include stimulus-related changes in power and phase consistency, both of which are critical for coordinating neural communication both within and across brain regions (Wang, 2010). Theta-band (~4–7 Hz) activity over midline fronto-central scalp sites is of special relevance, as it is thought to reflect cognitive control processes (Cavanagh and Frank, 2014; McLoughlin et al., 2022) and the N200 ERP has been proposed to reflect changes in theta-band oscillatory activity over the medial frontal cortex elicited by response conflict (Cavanagh and Frank, 2014).

Recent findings show reduced theta-band power and ITPC—indicating reduced cognitive control—in schizophrenia, which is a clinically and etiologically related psychiatric disorder to BD (Boudewyn and Carter, 2018; Reinhart et al., 2015; Ryman et al., 2018). Fewer studies have examined theta-band activity in BD, though two provide support for reduced theta-band power in BD during auditory processing and visual face processing (Atagün et al., 2013; Lasagna et al., 2021). To our knowledge, no published studies have examined theta-band activity while engaging cognitive control in the context of emotion in BD. The current paper aims to characterize both power and phase aspects of theta-band neural oscillatory activity during an affective response inhibition task in BD and examine their relationship to behavior. If individuals with BD are particularly impaired in reacting to the need to inhibit, a greater theta-activity reduction following No-Go cues would be expected. However, our prior analyses of event-related potentials (ERPs) in BD did not indicate any No-Go specific impairments (Menkes et al., 2022). This may indicate that individuals with BD do not show a specific reactive control impairment, rather, they would be expected to exhibit reductions to overall theta-band activity reflecting broad cognitive control processes. This leads to the following hypotheses: (1) individuals with BD would show reduced theta-band power and ITPC (across Go and No-Go trials), relative to healthy control participants (HC), and (2) across all participants, reduced theta-band power and ITPC would be associated with poorer behavioral task performance.

2. Methods

2.1. Participants

The current sample consisted of 67 participants diagnosed with BD (all bipolar I disorder) and 48 HC participants. Demographic characteristics of the sample are displayed in Table 1.

Table 1.

Sample demographic and clinical characteristics in each diagnostic group.

	Healthy Controls (HC)		Bipolar I Disorder (BD)
	N = 48		N = 67
	M	SD	M	SD	t orχ2	p
Demographic
Age	40.479	14.247	41.478	11.150	0.421	.674
Sex (% F)	43.75%		52.24%		0.503	.478
Race (% W)	73.17%		89.23%		3.531	.060
Education (yrs)	16.000	2.571	15.379	2.306	−1.346	.181
Parental education (yrs)	15.000	3.845	15.061	3.105	0.084	.933
Estimated IQ	114.790	9.981	110.596	10.356	−1.505	.137
Clinical
BDI-II (0–63)	-		11.000	8.873	-
ASRM (0–20)	-		3.697	4.023	-
HAMD-17 (0–52)	-		9.188	8.491	-
YMRS (0–60)	-		4.912	6.201	-
Years since Illness Onset	-		24.612	12.767	-
Psychosis History	-		53.13%		-
Substance Abuse/Dependence History	-		53.03%		-
Medication (%)
Antipsychotic	-		38.81%		-
Lithium	-		34.33%		-
Anticonvulsant	-		46.27%		-
Antidepressant	-		47.76%		-
Benzodiazepine	-		20.90%		-

Notes: ASRM = Altman Self Rating Mania Scale; BDI-II = Beck Depression Inventory, HAMD-17 = Hamilton Depression Rating Scale; YMRS = Young Mania Rating Scale. Possible range of scores is in parentheses for all scales. Race was available for 41 HC and 65 BD, personal education for 47 HC and 66 BD, parental education for 49 BD and 42 HC, IQ estimate (via The Wechsler Abbreviated Scale of Intelligence) for 19 HC and 47 BD. BDI-II and ASRM were available for 64 and 66 BD, respectively. HAMD-17 and YMRS were available for 32 and 34 BD participants, respectively.

Participants aged 18—65 were recruited from the community via advertisements, local mental health clinics, and referrals from other researchers (e.g., Prechter Longitudinal Study of BD; McInnis et al., 2018). Exclusion criteria for all participants included substance abuse/dependence in the past 6 months (based on DSM-IV criteria) and lifetime history of neurological illness, closed head injury, or serious medical conditions with neurological sequelae. All participants had normal or corrected-to-normal vision (at least 20/30) according to a Snellen chart. Additional exclusion criteria for HC included personal history of a DSM-IV axis-I disorder or a history of psychotic or bipolar disorders among first-degree relatives. Diagnoses for BD participants were assessed with the Diagnostic Interview for Genetic Studies (DIGS; Nurnberger et al., 1994) and confirmed through a best estimate process in which at least two MD/PhD level clinicians reached consensus. For HC, psychiatric diagnoses were ruled out through a diagnostic interview based on DSM-IV criteria (either the DIGS or the Structured Clinical Interview for DSM-IV-TR; SCID-IV; First et al., 1996). 66 of the BD participants and all HC participants from the present sample were also included in a previous report on ERP analyses during current Go/No-Go task (Menkes et al., 2022). All participants completed informed consent and monetary compensation was provided for participation. The study was approved by the University of Michigan Medical School Institutional Review Board.

2.2. Additional Measures

Self-report symptoms of depression and mania were assessed using the Beck Depression Inventory-II (BDI-II; Beck & Steer, 1993) and Altman Self-Rating Mania Scale (ASRM; Altman et al., 1997). Additionally, clinicians assessed current mood symptoms in a subset of BD participants using the Hamilton Depression Rating Scale (HAM-D; Hamilton, 1967) and Young Mania Rating Scale (YMRS; Young et al., 1978). These scales were added part way through data collection and therefore were not available for earlier participants (details on availability of measures are in Table 1). The Wechsler Abbreviated Scale of Intelligence (WASI; Wechsler, 1999) vocabulary subtest provided an IQ estimate for participants that completed the assessment as part of the Longitudinal Study of BD (McInnis et al., 2018).

2.3. Experimental Task

EEG data was recorded while participants completed an affective Go/No-Go task (Figure 1), which was adapted from Schulz et al. (2007) and programmed for the present study in Eprime software (Psychology Software Tools). Stimuli consisted of grey-scaled faces with happy, sad, angry, and neutral expressions from 27 individuals from the NimStim set (Tottenham et al., 2009). In each block, participants were instructed to respond as quickly as possible to a face with a target expression (Go) and withhold responding to a face with a non-target expression (No-Go). Each block had one emotion face (happy, sad, or angry) as the target while neutral faces were non-targets, or vice-versa. There were six sets of Go/No-Go pairs: Go happy/No-Go neutral, Go angry/No-Go neutral, Go sad/No-Go neutral, Go neutral/No-Go happy, Go neutral/No-Go sad, and Go neutral/No-Go angry. In each block, 34 Go stimuli (68%) and 16 No-Go stimuli (32%) were presented for 150 ms on the center of the screen following 500 ms of fixation cross, with a maximum response time of 1500 ms. Stimuli order within blocks was pseudorandomized to not have two consecutive No-Go stimuli. Each block was repeated with the reverse order of stimulus presentation, yielding 12 blocks (600 trials) in total. The order of block presentation was counterbalanced across participants within each diagnostic group. Task duration was approximately 25 minutes.

Figure 1. Affective Go/No-Go experimental task design.

A trial of the happy go/neutral no-go block is depicted as an example.

2.4. Behavioral Task Performance

Behavioral performance during the affective Go/No-Go task was assessed with mean reaction time on correct Go trials and signal detection measures of d-prime (d’) and criterion (β), both calculated using formulas from Schulz et al. (2007). The d’ measures perceptual sensitivity to differing stimuli (in the present task, emotional versus neutral faces), with larger d’ indicating greater discrimination of target from non-target stimuli. β measures response bias, with negative β values indicating a stronger tendency to respond (Stanislaw and Todorov, 1999).

2.5. EEG Data Acquisition and Processing

EEG Acquisition.

EEG data were collected at 32 channels using the Brain Vision system (Brain Products, Gmbh) with Ag/AgCl electrodes positioned according to a modified 10–20 system. Vertical electrooculogram was measured with one electrode below the right eye. Data were sampled at 5,000 Hz and referenced to FCz during recording. Electrode impedances were kept at or below 10 kΩ.

EEG preprocessing.

EEG preprocessing was completed using scripts adapted from the EEGLAB 2021.0 toolkit for MATLAB R2021.a (Delorme and Makeig, 2004). Data was down-sampled to 500 Hz, filtered (0.1 Hz high-pass, 100 Hz low-pass, and 60 Hz line noise using the CleanLine plugin), and re-referenced to the average of all electrodes prior to visual inspection for removal of noisy electrodes and manual artifacts (e.g., muscular activity). An Independent Component Analysis (ICA)-based algorithm from EEGLAB was used on segmented data (100 ms pre-stimulus onset to 800 ms post-stimulus) to identify components corresponding to ocular and other non-neural artifacts. Subsequently, these non-neural ICA components were removed from data that was re-segmented into longer epochs (−1,600–2,400 ms). This longer epoch was used to minimize edge artifacts in the time window of interest that we would later extract spectral features from (−300–800 ms). Any previously removed noisy channels were then interpolated back into the data, followed by re-referencing again to the common average, baseline adjustment (−100–0 ms), and automatic artifact rejection of segments with ±100 μV or more in any channel. Finally, visual inspection and manual rejection were performed again to remove any remaining non-neural artifacts. Only participants with at least 50% of total task trials (i.e., 300 trials, when combining Go and No-Go) remaining after preprocessing were included for subsequent analyses. This led to excluding 2 BD participants and no HCs, yielding the analytic sample of 115 participants for time-frequency decomposition.

Time-frequency decomposition.

Following preprocessing, time-frequency decomposition was applied to cleaned EEG data on the longer epochs (−1,600–2,400 ms) for trials with correct behavioral responses (in which participants either correctly responded on Go trials or successfully withheld responses on No-Go trials). EEGLAB toolkits were used to conduct Morlet wavelet convolution (3–10 cycles) from 3 to 50 Hz. Based on visual inspection of spectrographs of 3–50 Hz activity for all participants to determine the window in which theta activity was most strongly represented (Figure 2), while still keeping within the traditional theta frequency-band boundaries (Cohen, 2014), the 4–7 Hz frequency range during the 150–400 ms time window was selected for analyses. This time-frequency window was used to extract event-related power and inter-trial phase consistency (ITPC) values (referred to as event-related spectral perturbations [ERSP] and inter-trial coherence [ITC] respectively, in EEGLAB; Makeig et al., 2004) for trials with correct behavioral responses. Power was standardized in decibels (dB), baseline adjusted (−300 to −50 ms), and averaged across correct trials, to capture post-stimulus changes in power relative to the baseline. ITPC measures the consistency of the phase of the waveform across trials at a given time point (ranging 0 to 1, with higher values indicating greater consistency). Scalp topography within this time-frequency window, collapsed across groups and trial types (Figure 2), shows that theta activity was most prominent at the midline over fronto-central areas (Cz electrode site). Therefore, Cz activity was used for all subsequent theta-band analyses.

Figure 2. Time frequency decomposition results and scalp topography of power, collapsed across both diagnostic groups and trial types.

Left: time-frequency decomposition results of total power (3–50 Hz activity at the Cz electrode site). The selected time-frequency window (150–400 ms, 4–7 Hz) is outlined in a black box. Right: scalp topography of total power in the selected time-frequency window. Mean activity in the selected time-frequency window at the Cz electrode site were used to extract theta power and ITPC values for subsequent analyses.

2.6. Statistical Analyses

Statistical analyses were conducted in R (version 4.1.2; R Core Team, 2021), using the car package (Fox and Weisberg, 2019) to compute type III sums of squares for one-way ANCOVAs, lmerTest package (Kuznetsova et al., 2017) with restricted maximum likelihood estimation for mixed-effects ANCOVAs, the base stats package for remaining analyses, and the ggpubr package for data visualization (Kassambara, 2020).

To verify groups were well-matched, BD and HC groups were compared on key demographic variables (age, sex, race, personal and parental education) using t-tests and chi-squared tests. Behavioral indices (d’, β, reaction time on correct Go trials) were compared between HC and BD using ANCOVA with age as a covariate, due to the wide age range of participants (18—65). Two separate mixed-effects ANCOVAs were used to determine whether theta power and ITPC are reduced in BD, relative to HC, with age as a covariate. Main effects of trial type (Go, No-Go) and diagnostic group (HC, BD) and their interaction effect on the dependent variables of power and ITPC were tested. These analyses were repeated with the addition of block type (emotion go/neutral no-go, neutral go/emotion no-go) and its interaction with group and trial type as independent variables, to explore whether theta-band activity was particularly reduced in BD when executing responses to emotion cues versus inhibiting responses to emotion cues. Significant group differences in theta power and ITPC were followed up with logistic regression analyses to explore the unique associations of power and ITPC with group membership (HC, BD). While including age as a covariate, power and ITPC (collapsed across Go and No-Go trials) were added to logistic regression models sequentially, to test if either one explained group differences above and beyond the other.

Bivariate Pearson correlational analyses were used to explore whether theta power and ITPC are associated with behavioral performance (reaction time, d’, β). Given evidence for effects of lithium on neural oscillatory activity (Atagün, 2016), we additionally explored possible effects of lithium medication by comparing theta activity between lithium users versus non-users within the BD group.

3. Results

HC and BD groups did not significantly differ on age, sex, race, personal or parental years of education, indicating groups were well-matched (Table 1).

3.1. Behavioral Task Performance

Table 2 displays descriptive statistics of behavioral affective Go/No-Go task performance. ANCOVA results are presented in Table 3. Reaction time on correct Go trials was slower for BD participants than HC (p = .005). Discriminability (d’) was also lower for BD than HC, indicating an illness-related impairment in distinguishing differing facial expressions (p = .044). Response bias (β) did not differ between groups, indicating no group differences in tendency to respond.

Table 2.

Descriptive statistics of behavior and theta-band activity during the affective Go/No-Go task in each diagnostic group.

	HC				BD
	M	SD	Min	Max	M	SD	Min	Max	Cohen’s d
Behavior
Reaction Time	521.371	67.723	408.580	696.787	568.651	95.518	405.366	857.903	.556
Discriminability (d’)	2.581	0.612	1.232	4.027	2.337	0.632	0.547	4.042	−.393
Response bias (β)	0.033	0.352	−0.606	1.013	0.045	0.414	−0.874	1.080	.030
Hit Rate	0.872	0.102	0.555	0.983	0.835	0.131	0.445	0.993	−.303
False Alarm Rate	0.112	0.073	0.005	0.344	0.135	0.099	0.010	0.526	.255
Number of Correct Trials †
Go Trials	330.104	49.119	222.000	398.000	308.194	61.992	142.000	404.000	−.384
No-Go Trials	159.813	18.216	120.000	204.000	153.612	24.399	78.000	198.000	−.281
Theta-Band Activity
Power (dB)
All Trials	2.772	1.902	−0.712	6.930	2.033	1.460	−0.988	6.976	−.445
Go Trials	2.660	1.888	−0.795	6.892	1.988	1.426	−0.947	7.099	−.411
No-Go Trials	2.932	1.977	−0.565	7.454	2.110	1.582	−1.075	6.704	−.467
ITPC
All Trials	0.574	0.150	0.306	0.866	0.523	0.130	0.162	0.755	−.364
Go Trials	0.575	0.152	0.293	0.862	0.525	0.132	0.186	0.780	−.351
No-Go Trials	0.580	0.147	0.283	0.875	0.528	0.131	0.133	0.741	−.378

Notes. ITPC = inter-trial phase consistency.

^†Number of correct trials indicates number of trials averaged together to compute theta-band activity measures (trials with correct behavioral responses that were remaining in cleaned data, following EEG data processing procedures). Hit rate = proportion of correct Go trials; False Alarm rate = proportion of incorrect No-Go trials. Cohen’s d was calculated to measure effect sizes.

Table 3.

Analysis of covariance results for behavioral performance and theta-band activity outcomes.

Outcome	Predictor	df	F	p
Behavior
Reaction Time	Age	1, 112	3.920	.050
	Group (HC, BD)	1, 112	8.388	.005
Discriminability (d’)	Age	1, 112	0.748	.389
	Group (HC, BD)	1, 112	4.158	.044
Response Bias (β)	Age	1, 112	5.377	.022
	Group (HC, BD)	1, 112	0.005	.943
Theta Activity
Power (dB)	Age	1, 112	0.180	0.672
	Group (HC, BD)	1, 112	5.641	0.019
	Trial Type (Go, No-Go)	1, 113	10.758	0.001
	Group X Trial Type Interaction	1, 113	1.543	0.217
ITPC	Age	1, 112	11.532	0.001
	Group (HC, BD)	1, 112	4.747	0.031
	Trial Type (Go, No-Go)	1, 113	0.817	0.368
	Group X Trial Type Interaction	1, 113	0.080	0.778

Notes. ITPC = inter-trial phase consistency.

3.2. Theta-Band Activity

Descriptive statistics of event-related power and ITPC values are shown in Table 2 and Figure 3. ANCOVA results are in Table 3.

Figure 3. Theta-band activity in each diagnostic group.

Left: time-frequency decomposition results (3–50 Hz at the Cz electrode site) for power and ITPC. Mean activity in the selected time-frequency window (150–400 Hz, 4–7 Hz; outlined with a black box) was used as theta-band activity data. Center: bar plots illustrating descriptive statistics of theta activity measures broken down by both diagnostic group and trial type. Error bars on bar plots indicate standard error of the mean.

Power.

Across all participants, theta power was higher during No-Go than Go trials (p = .001). Across trial types, BD participants showed lower theta power than HC (p = .019). There was not a significant interaction effect of group by Go/No-Go trial type, indicating the magnitude of difference in power between trial types did not differ between groups.

Inter-trial phase consistency (ITPC).

Theta ITPC did not differ between No-Go and Go trials. Theta ITPC was lower in BD participants than HC, across trial types (p = .031). The group by trial type interaction effect was also non-significant, indicating the magnitude of difference in ITPC between trial types did not differ between groups.

Unique associations of theta-band power and ITPC with group.

Logistic regression results are in Supplemental Table 1, and correlations between theta activity measures are presented in Table 4. Consistent with the ANCOVA results on group differences in theta activity, theta power and ITPC were associated with group membership when examined separately, each with age included as a covariate (p’s ≤ .031). Compared to these models, a logistic regression including both power and ITPC indicated neither variable explained unique additional information on diagnostic group differences beyond the other.

Table 4.

Correlations between all EEG and behavioral study measures.

Correlations between theta-band activity and behavior across all participants (N=115)
		Power						ITPC						Behavior
		All Trials		Go Trials		No-Go Trials		All Trials		Go Trials		No-Go Trials		RT		d’
Power	Go	.990	(<.001)
	No-Go	.976	(<.001)	.935	(<.001)
ITPC	All trials	.789	(<.001)	.806	(<.001)	.733	(<.001)
	Go	.772	(<.001)	.795	(<.001)	.707	(<.001)	.992	(<.001)
	No-Go	.802	(<.001)	.806	(<.001)	.765	(<.001)	.973	(<.001)	.938	(<.001)
Behavior	RT	.359	(<.001)	−.356	(<.001)	−.341	(<.001)	−.226	(.015)	−.212	(.023)	−.245	(.008)
	d’	.189	(.043)	.152	(.105)	.234	(.012)	.110	(.241)	.088	(.349)	.159	(.090)	.010	(.917)
	β	.241	(.009)	−.222	(.017)	−.257	(.006)	−.028	(.763)	−.033	(.726)	−.022	(.813)	.472	(<.001)	−.159	(.090)
Correlations between theta-band activity and behavior within HC participants (N=48)
		Power						ITPC						Behavior
		All Trials		Go Trials		No-Go Trials		All Trials		Go Trials		No-Go Trials		RT		d’
Power	Go	.993	(<.001)
	No-Go	.980	(<.001)	.950	(<.001)
ITPC	All trials	.849	(<.001)	.863	(<.001)	.800	(<.001)
	Go	.845	(<.001)	.862	(<.001)	.793	(<.001)	.994	(<.001)
	No-Go	.837	(<.001)	.846	(<.001)	.799	(<.001)	.979	(<.001)	.953	(<.001)
Behavior	RT	.353	(.014)	−.346	(.016)	−.343	(.017)	−.202	(168)	−.199	(.175)	−.189	(.199)
	d’	.016	(.914)	−.053	(.722)	.043	(.773)	−.075	(.615)	−.098	(.510)	−.016	(.915)	.126	(.395)
	β	.144	(.328)	−.134	(.365)	−.162	(.273)	−.030	(.841)	−.031	(.834)	−.021	(.889)	.596	(<.001)	−.056	(.703)
Correlations between theta-band activity and behavior within BD participants (N=67)
		Power						ITPC						Behavior
		All Trials		Go Trials		No-Go Trials		All Trials		Go Trials		No-Go Trials		RT		d’
Power	Go	.985	(<.001)
	No-Go	.969	(<.001)	.913	(<.001)
ITPC	All trials	.713	(<.001)	.733	(<.001)	.646	(<.001)
	Go	.680	(<.001)	.711	(<.001)	.599	(<.001)	.990	(<.001)
	No-Go	.754	(<.001)	.754	(<.001)	.714	(<.001)	.965	(<.001)	.920	(<.001)
Behavior	RT	.322	(.008)	−.328	(.007)	−.290	(.017)	−.189	(.125)	−.168	(.173)	−.225	(.067)
	d’	.309	(.011)	.276	(.024)	.337	(.005)	.201	(.103)	.180	(.146)	.241	(.050)	.036	(.775)
	β	.332	(.006)	−.305	(.012)	−.343	(.004)	−.024	(.846)	−.031	(.801)	−.020	(874)	.441	(<.001)	−.220	(074)

RT = mean reaction time on correct Go trials (i.e., correct hits), d’ = discriminability, β = response bias. Each table cell depicts: Pearson’s correlation coefficient (p-value). The p-values shown were not adjusted for multiple statistical tests. Correlations in each diagnostic group are also depicted in Supplemental Figure 1.

Effects of task block type on theta-band activity.

Briefly, neither theta-band power nor ITPC differed by block type (emotion go/neutral no-go, neutral go/emotion no-go), indicating no differences in theta activity when executing responses to emotion cues versus inhibiting responses to emotion cues. Additionally, group by block type interaction effects were also non-significant (see Supplemental Table 2 and 3 for full results).

Effects of lithium medication on theta activity.

Briefly, neither theta-band power nor ITPC differed by lithium use (see Supplemental Table 4 for full results).

3.3. Correlations Between Theta-Band Activity and Behavior

Results of correlation analyses are shown in Table 4. Across all participants, higher theta-band power and ITPC, regardless of trial type, were both associated with faster reaction times on correct hits (power: r ranging from −.341 to −.359; ITPC: r ranging from −.212 to −.245). Increased power on No-Go trials (r = .234) and all trials (r = .189) was also associated with increased discrimination (d’) of differing facial expressions. Power, regardless of trial type (r ranging −.222 to −.257), was associated with stronger tendency to respond (lower β). ITPC was not associated with discriminability or response bias (p’s > .09).

These correlations were largely similar in magnitude and direction when examined within each participant group, though there were differences in which correlations reached statistical significance in the smaller sample sizes (Supplemental Table 5 and Figure 1).

4. Discussion

In the current study, individuals with BD exhibited lower midline fronto-central theta-band event-related power and inter-trial phase consistency, slower reaction times in executing responses, and poorer behavioral discrimination of facial expressions than HC during an affective cognitive control task.

Our overall theta-band findings are consistent with prior findings that theta-band activity is reduced in BD during cognitive tasks (e.g., auditory processing; Atagün et al., 2013), and provide initial evidence of this attenuation during response inhibition to emotional stimuli. We observed the expected heightened theta-band power on No-Go trials across all participants (consistent with inhibition trials requiring increased control); however, the magnitude of this effect did not differ between groups—indicating the BD group showed the typical pattern of increased cognitive control engagement when inhibiting responses, despite overall reduced theta power. Similar to power, theta ITPC was reduced in BD across trial types. That is, abnormalities in both aspects of theta activity found in BD were not specific to inhibiting prepotent responses during No-Go trials. Rather, results indicate reductions to cognitive control processes, including both more proactive control processes involved throughout the task (i.e., anticipating each upcoming trial could be a No-Go cue) and reactive control to No-Go cues (Braver, 2012). It should be noted that ERP analyses (which reveal only phase-locked activity) published on this dataset (Menkes et al., 2022) did not find significant differences between HC and BD groups on N200 amplitudes—which is thought to also be an electrophysiological index of cognitive control (Folstein and Van Petten, 2008). The current findings of theta activity reductions in BD and their association with task performance suggest that non-phase-locked theta activity also contributes to cognitive control deficits, as well as affective response inhibition deficits, in BD. This supports the added value of time-frequency analyses in uncovering neural mechanisms that can be missed if examining only ERPs (Cohen, 2014).

To follow up findings of reduced power and ITPC in BD, we explored whether either aspect of theta-band activity explained unique additional information on group differences (HC, BD), above and beyond the other one. We did not find that either aspect contributed unique information to explain group membership, beyond the other. This is likely due to the high correlation between power and ITPC (Table 4) which suggests the reduction of these aspects of theta activity in BD tends to co-occur within individuals. Others have reported theta power abnormalities in the absence of ITPC abnormalities in other psychiatric groups (e.g., obsessive-compulsive disorder), or vice versa (e.g., attention-deficit/hyperactivity disorder; Groom et al., 2010; Suzuki et al., 2022), supporting that reductions in one does not necessarily also mean reductions in the other. Such knowledge of specific aspects of neural oscillatory activity implicated in a disorder is useful in developing targeted treatment. Given that both power and ITPC of theta activity are reduced in BD, non-invasive neuromodulation targeting both the amplitude and phase-timing of theta rhythms might help normalize neural dynamics and improve affective response inhibition. Recent advances in combining brain stimulation techniques (e.g., transcranial magnetic stimulation [TMS] and transcranial electrical current stimulation [tECS]) with specific pulse sequence/wave patterns and closed-loop features have shown promise in modulating short- and longer-term neural oscillatory dynamics in humans (Huang et al., 2005; Reinhart et al., 2015; Riddle and Frohlich, 2021). Further research should explore the feasibility and efficacy of these neuromodulation techniques to target theta-band activity to improve outcomes in BD.

Individuals with BD responded slower on successful Go trials than HC during the affective Go/No-Go task, as expected. As noted in our prior work, individuals with BD showed poorer behavioral discrimination of emotional versus neutral facial expressions (d’), but did not differ from HC on response bias (β) (Menkes et al., 2022). Increased theta-band power was associated with faster response execution times in the combined sample and within both BD and HC groups. This is consistent with prior findings regarding theta power during visual face processing in BD (Lasagna et al., 2021), as well as simulation and non-clinical studies indicating relationships between theta oscillations and processing speed/efficiency (Mueller et al., 2017; Smerieri et al., 2010; van Vugt et al., 2012). Across the combined sample and within only BD participants, increased power was also associated with greater discrimination (d’) of facial expressions and stronger tendency to respond. This faster response execution time, greater discrimination of differing facial expressions, and increased response tendency reflect generally adaptive behavior, given the task was to respond differentially based on facial expression, respond as quickly as possible to Go stimuli, and was designed to elicit a prepotent response tendency through more frequent Go cues than No-Go cues. Increased ITPC was related to quicker response execution times only in the combined sample, though this correlation was similar in magnitude within each of the HC and BD groups but did not reach statistical significance likely due to statistical power. Taken together, lower theta-band power (more so than ITPC) may be particularly relevant for less adaptative behavioral responding in BD.

4.1. Limitations

Our sample is largely white and highly educated, which limits generalizability. The current BD sample consists of individuals with long-standing bipolar I disorder (24.6 years since mood disorder onset, on average) and with minimal to mild mood symptoms at the time of participation. Thus, it is unclear how these findings may vary by severity of mood symptoms or generalize to individuals earlier on in the course of illness or with other BD subtypes. Additionally, we cannot fully disentangle the effects of medication versus bipolar illness on neural oscillatory activity. Although we did not observe any significant differences in either aspect of theta activity based on current lithium use in this study (Supplemental Table 4), prior research provides evidence that lithium treatment increases theta activity (Atagün, 2016). Thus, the current findings of reduced theta activity despite 34% of the BD sample taking lithium at the time of participation lends support for the hypothesized illness-related theta activity attenuations. Further research in earlier illness stages (with minimal confounds of illness chronicity and medication exposure) would help confirm our findings. Finally, we cannot conclude whether the effects reported in the current study are specific to the theta-band, or whether activity in lower frequencies may also be reduced in BD. Given the design of this task, our time-frequency analyses could not reliably extract activity of frequency lower than 3 Hz. This would require longer epochs than we analyzed, which could include activity related to the pre-trial fixation cross or the previous or subsequent trial. Future research should examine lower-frequency activity to clarify if delta-band activity also contributes to altered affective cognitive control in BD.

4.2. Conclusions

This study extended prior behavioral and neural findings of impaired cognitive control in BD by examining midline fronto-central theta-band oscillatory activity in emotional contexts. Our findings provide initial evidence of attenuated theta-band power and ITPC indexing cognitive control in response to emotional faces, despite intact increased engagement of cognitive control in reaction to the need to inhibit responses. Taken together with behavioral results, findings suggest attenuated theta-band power and inter-trial phase consistency may both contribute to reduced cognitive control and maladaptive behavioral responding to social-affective cues. Future studies should examine the relevance of theta-band activity for functional and clinical outcomes to inform neurobiologically-based interventions.