Left Ventrolateral Prefrontal Cortical Activity During Reward Expectancy Predicts Mania Risk Up to One Year Post Scan

Abstract

Introduction

Identification of neural markers associated with risk for manic symptoms is an important challenge for neuropsychiatric research. Previous work has highlighted the association between predisposition for mania/hypomania and elevated reward sensitivity. Elevated activity in the left ventrolateral prefrontal cortex (L vlPFC) during reward expectancy (RE) is associated with measures predictive of risk for manic/hypomanic symptoms. However, no studies have examined this relationship longitudinally. The goal of this study was to identify a neural marker associated with longitudinal risk for manic/hypomanic symptoms.

Methods

We used a card guessing functional magnetic resonance imaging (fMRI) paradigm to examine RE-related L vlPFC activity. One hundred and three young adults who were either healthy or experiencing psychological distress completed a single baseline fMRI scan and self-report measures of manic/hypomanic symptoms. Self-report measures were repeated up to two follow up visits over one year.

Results

We identified a significant positive relationship between baseline RE-related L vlPFC activity and MOODS Manic Domain scores up to one-year post scan. This relationship was specific to manic symptoms and was not present for MOODS depression-related domains.

Limitations

This study was not designed to predict conversion to bipolar disorder, but rather the more proximal construct of lifetime risk for mania/hypomania.

Conclusions

RE-related L vlPFC activity may serve as an important marker of risk for future manic/hypomanic symptoms and may also be a potential target for intervention.

1. Introduction

Accurate diagnosis of bipolar disorder (BD) in young adulthood is a significant challenge due to overlap in initial presentation with other mood disorders. Given that hypomanic/manic symptoms differentiate BD from other mood disorders, identification of biomarkers reflecting neural processes associated with the future development of these symptoms can aid in early identification of BD risk.

Others have highlighted the importance of elevated reward sensitivity in individuals with and those at risk for developing BD (Bart et al., 2021; Johnson et al., 2019), and there is a growing literature indicating that individuals with BD show abnormal activity in the fronto-striatal reward network during reward processing (Shi et al., 2020; Dutra et al., 2017; Mason et al., 2014; Nusslock et al., 2012; O’Sullivan et al., 2011). We have more specifically reported that individuals with BD show elevated activity in components of the distributed reward network, including bilateral ventral striatum (VS) and left ventrolateral prefrontal cortex (L vlPFC), during reward expectancy (RE, Manelis et al., 2016; Chase et al., 2013; Nusslock et al., 2012). We have also shown that elevated L vlPFC during RE characterizes young adults at risk for future BD, as measured by the Moods Spectrum (MOODS) questionnaire (Edmiston et al., 2020), a lifetime count measure of tendencies or symptoms associated with predisposition to mania and depression (Dell’Osso et al., 2002). The L vlPFC is associated with evaluation of stimuli denoting future potential rewards, including RE (Boorman et al., 2016). Thus, elevated RE-related L vlPFC activity may reflect heightened valuation of potential future rewards associated with heightened reward sensitivity, and thus mania and BD risk. It remains unclear, however, whether L vlPFC activity predicts risk for mania longitudinally, as our previous findings were in a cross-sectional rather than a longitudinal study. To our knowledge, no longitudinal study has examined the contribution of RE-related activity to mania risk over time.

We aimed to determine the extent to which activity in components of the distributed reward network predicted future mania risk, as measured by the MOODS Manic Domain over the course of one year after the neuroimaging scan (Dell’Osso et al., 2002). We recruited a sample of young adults from the community who were either healthy or seeking treatment for psychological distress, as the latter is often an early manifestation of future psychiatric illnesses. Thus, our recruitment strategy allowed us to examine young adults across a range of potential future mania risk. We hypothesized that elevated RE-related L vlPFC activity would predict future mania risk as measured by the MOODS, specifically the MOODS Manic Domain, over one year.

2. Methods

2.1. Participants

One hundred and three right-handed young adults (79 female) ages 18–25 who were either healthy or seeking treatment for psychological distress were recruited from the Pittsburgh community from an ongoing study (R37MH100041; Table 1). This sample included a subsample of 67 participants from a previously published cross-sectional dataset, now with completed 6- and/or 12-month longitudinal follow up assessments; as well as 36 participants who were not included in previous samples from a second funding period.

Table 1:

Sample Demographics

	Gender (F/Total)	Education (Some college or more/Total)	Mean(SD) Age	Mean(SD) FWD	Median(Range) Maximum MOODS Manic	Median(Range) Maximum MOODS Manic Cognition	Median(Range) Maximum MOODS Manic Energy	Median(Range) Maximum MOODS Depression	Median(Range) Maximum MOODS Depressive Cognition	Median(Range) Maximum MOODS Depressive Energy
Funding Period 1	53/67	54/67	21.89(2.15)	0.21(0.076)	13(2–27)	8(0–22)	6(0–12)	20(3–25)	16(1–23)	7(0–9)
Funding Period 2	26/36	32/36	23.71(3.10)	0.081(0.041)	4(0–21)	3(0–13)	1(0–9)	7(0–20)	2(0–23)	1(0–9)

The Structured Clinical Interview for DSM-5 was administered to assess for exclusion criteria, which were identical to those in previous reports, and included psychosis, lifetime alcohol/substance use disorder, and illicit substance use, excepting cannabis, in the past three months (Edmiston et al., 2020; Chase et al., 2017). Cannabis was allowed, given its common usage in young adults. Past psychotropic medication usage was allowed if the participant had been unmedicated for at least three months or was currently taking medication for less than two weeks.

The University of Pittsburgh Institutional Review Board approved this study. Participants gave written informed consent following a description of all procedures.

2.2. Assessments

As a measure of tendency to experience manic/hypomanic symptoms, participants completed the MOODS (Dell’Osso et al., 2002). To extend our previous findings and given our a priori interest in assessing predictors of manic symptoms associated with reward sensitivity, we used the MOODS Manic Domain, which assesses the affective components of mania/hypomania, and is the MOODS Domain most closely associated with affective constructs, such as reward sensitivity, that are implicated in BD risk. To confirm the specificity of findings, we also examined the relationship between RE-related L vlPFC activity and the five additional mood-related MOODS Domains: Manic Energy, Manic Cognition, Depressive, Depressive Energy, and Depressive Cognition Domains. As the MOODS is a lifetime count-based measure, we used the maximum score across the baseline, 6- and 12-month follow-up visits as our measure of interest. This approach allowed us to best represent the maximum number of endorsed items over the one-year follow-up period. All 103 participants had baseline and 6-month data; 66 participants also had 12-month data.

2.3. MRI data acquisition

Data were collected using a 3.0 Tesla PRISMA (n = 72) or Siemens Trio 2 (n = 31) MRI scanner at the University of Pittsburgh Magnetic Resonance Research Center. Acquisition parameters were identical for both scanners from the first funding period of the study. Data acquisition from the second funding period employed slightly different parameters (Supplement).

2.4. FMRI task

As in previously published studies, participants completed a 16-minute (two eight-minute blocks) card guessing task designed to examine neural activity during anticipation of reward (Chase et al., 2017). Participants were instructed to indicate via button press whether they thought a subsequently revealed card would have a value greater than or less than five. A 2–6 s reward expectancy cue was presented, denoting the trial type. The four trial types were: win (win or no change, indicated by a deck of cards and an up arrow); loss (loss or no change, indicated by a deck of cards and a down arrow); mixed win and loss trials with win or loss outcomes (indicated by a deck of cards and both up and down arrows); or neutral trials with no change (indicated by a deck of cards and no arrows, Supplement).

2.5. fMRI data preprocessing

FMRI data were preprocessed using previously reported methods and Nipype (Edmiston et al., 2020; Gorgolewski et al., 2011, Supplement).

2.6. fMRI data first-level modeling

The canonical hemodynamic response function was convolved to each regressor (reward expectancy, outcome expectancy, prediction error) using SPM8 to generate first-level general linear models for each participant (Supplement).

Our parametric regressor of interest was the RE condition, which corresponded to the 2–6 s period following participant choice but before the outcome of the trial was revealed. Reward expectancy was weighted to represent the expected value for the four possible trial conditions: +0.5 for win trials (50 % chance of winning $1), −0.375 for loss trials (50 % chance of losing $0.75), 0 for neutral trials, and +0.125 for mixed trials (50/50 chance of winning $1 or losing $0.75, i.e., 50/50 chance of a net gain of $0.25).

2.7. Second-level modeling

Mean BOLD signal was extracted from our a priori reward network region of interest (ROI), the L vlPFC, derived from an activation likelihood estimation meta-analysis of studies showing increased reward-related activity in this region (Fig. 1a, Chase et al., 2017). We extracted BOLD signal from the R vlPFC, the contralateral homolog of L vlPFC, as a control region, as we previously showed no relationships between RE-related activity in R vlPFC and mania risk as measured by the MOODS (Cassano et al., 2009). For completeness, we also extracted BOLD signal from L and R ventral striatum, as secondary regions of interest. One participant was excluded from the R and L VS analyses due to >30 % signal drop out in the VS. As data were collected using two scanners, we used COMBAT, a Matlab-based program that reduces site- or scanner-based variability, while preserving variability due to participant characteristics (Yu et al., 2018).

Fig. 1.

The left ventrolateral prefrontal cortex region of interest was derived from an activation likelihood estimation meta-analysis of reward-related activity (1a). Reward expectancy-related left ventrolateral prefrontal cortex activity is positively correlated with MOODS Manic Domain scores over the course of one year (1b).

We used a general linear model with the maximum MOODS Manic Domain score as the dependent variable, and with RE-related L vlPFC activity as the independent variable. Age, gender, educational attainment, and framewise displacement (FWD) were covariates in the model. Given our single a priori hypothesis, we used a statistical threshold of p < 0.05.

2.8. Secondary analyses

To confirm specificity to the L vlPFC, we repeated analyses with L and R VS and R vlPFC ROIs. We performed analyses with RE-related activity in all four ROIs using the five remaining MOODS domains as dependent variables, to determine if any significant findings were specific to mania/hypomania per se or related to mood-related psychopathology generally. For all the above secondary analyses, results were considered significant at p = 0.0022, corrected for 23 comparisons (3 ROIs in 6 MOODS Domains, plus the L vlPFC ROI in 5 MOODS Domains) in SPSS according to the Benjamini and Hochberg (1995) False Discovery Rate method (FDR; Supplement).

We performed additional analyses to test for effects of baseline diagnosis on RE-related L vlPFC BOLD activity. We also assessed for effects of changes in medication over the course of the study (Supplement).

3. Results

RE-related L vlPFC BOLD activity was significantly associated with maximum longitudinal MOODS Manic Domain score (B = 4.995, p = 0.037, Fig. 1b).

Secondary regression findings with RE-related activity in the R vlPFC and VS ROIs indicated a significant relationship between the R VS and maximal longitudinal MOODS Manic Domain score (B = 6.764, p = 0.025). However, this finding was not robust to the removal of one influential outlier (B = 5.929, p = 0.06).

There were no significant relationships between RE-related L vlPFC activity and maximum longitudinal MOODS Manic Energy (B = 1.787, p = 0.147), Manic Cognition (B = 1.123, p = 0.558), Depressive (B = 2.585, p = 0.337), Depressive Energy (B = 1.66, p = 0.147), or Depressive Cognition (B = 0.976, p = 0.716) Domain scores. No relationships between RE-related L and R VS and R vlPFC activity and these other MOODS Domains survived correction for multiple comparisons (Supplemental Table 1).

4. Discussion

Our previous work highlighted RE-related L vlPFC activity as a marker of risk for mania/hypomania (Edmiston et al., 2020). This study determined the contribution of RE-related L vlPFC activity to predicting manic/hypomanic symptom risk measured longitudinally. We demonstrated a relationship between RE-related L vlPFC activity and the MOODS Manic Domain measured over 12 months. This study extends our previous findings, demonstrating that RE-related L vlPFC activity predicts mania/hypomania risk measured up to one-year post-scan.

We did not identify any significant relationships between RE-related activity in other reward network regions and the MOODS Manic Domain; nor were there any significant relationships among RE-related L vlPFC activity, or other reward network ROI RE-related activity, and MOODS depression-related domains. These findings indicate that RE-related L vlPFC activity specifically predicts future mania/hypomania risk measured over one year and does not represent a marker of risk for future mood-related psychopathology generally. The fact that RE-related L vlPFC activity specifically predicted future MOODS Manic, but not MOODS Manic-Energy or Manic-Cognition, Domain scores suggest that RE-related L vlPFC activity is a risk marker for the affective component of mania/hypomania, likely reflecting the reward sensitivity component of manic/hypomania (Mason et al., 2012).

4.1. Limitations

The goal of the study was to extend our previous findings highlighting RE-related L vlPFC activity as a potential biomarker of future BD risk indexed using a continuous measure of future mania/hypomania risk, rather than predict conversion to BD. As such, this study was not designed to assess for factors associated with conversion to BD, and only two participants converted to BD during the study (Supplement). However, this study was designed to study lifetime risk for mania/hypomania, not proximal conversion to BD. This study used the MOODS, a lifetime measure designed to identify and assess spectrum traits associated with mood disorders. Studies designed to examine the utility of the MOODS, as well as neuroimaging biomarkers, in predicting conversion to BD are an area for future work. Another limitation of this study was the use of two scanners, which we accounted for by using COMBAT. The sample was 76.7 % female, consistent with the population of treatment-seeking adults recruited from the community. Thus, we were unable to test for gender moderation effects.

4.2. Conclusions

The present study indicates that elevated RE-related L vlPFC activity is associated with future mania/hypomania, measured longitudinally. This longitudinal finding provides further support for the role of the L vlPFC in encoding the value of future potential rewards, with greater RE-related activity in this region likely reflecting heightened sensitivity to reward. Our finding also highlights the specificity of RE-related L vlPFC activity as a biomarker of risk for manic/hypomanic symptoms over time. The L vlPFC is a promising target for intervention to reduce or prevent the onset of manic/hypomanic symptoms. Studies using neuromodulatory interventions to target the L vlPFC (Bertocci et al., 2021), as well as studies with participants who convert to BD, can help to determine if these markers can indeed predict transition to a BD diagnosis or be modulated to reduce such risk.