Neural basis of memory impairments and relation to functional disability in fully or partially remitted patients with affective disorders

substantial functional disability underwent neuropsychological assessment and functional magnetic resonance imaging (fMRI) scan during which they completed a strategic picture-encoding task. For comparison, 36 matched healthy controls underwent an identical test protocol. Patients showed encoding-related hypo -activity in the dPFC compared to controls. In patients, lower right dlPFC activity was associated with poorer overall functioning and more antipsychotic drug use. In conclusion, memory impairments were underpinned by failure to recruit the dPFC during task performance which was associated with impaired functioning in fully or partially remitted patients with affective disorders. This aberrant neurocircuitry activity has implications for the design of future pro-cognitive interventions that aim to improve not only cognition but also real-world functioning.


Introduction
Major depressive and bipolar disorders (MDD and BD) are often accompanied by cognitive impairment that persists even in euthymic phases Bourne et al., 2013). The impairment occurs across several cognitive domains, including episodic memory, which impedes patients' quality of life and daily functioning (Bonnín et al., 2010). In interaction with executive functions, episodic memory is involved in multiple daily life functions, including encoding and recall of tasks and information (Duff et al., 2005). Memory impairment can thus impede both occupational and social functioning (Tse et al., 2014). There is encouraging emerging evidence for cognitive benefits of cognitive training and cognitive remediation interventions across both MDD and BD according to recent systematic reviews and meta-analytic evidence (Miskowiak et al., 2022b;Woolf et al., 2022). Unfortunately, these studies also revealed a remarkable lack of transfer of cognitive gains to patients' daily life functioning (Woolf et al., 2022). This lack of transfer to real-life functioning is a key obstacle in current pro-cognitive intervention strategies that results from several methodological challenges in the field. One principal challenge is the current limited insight into the specific cognitive constructs and neurocircuitry abnormalities that give rise to functional impairment and may serve as treatment targets to aid functional recovery.
It is now well-established that cognitive impairment in affective disorders stems from disruptions of task-relevant neurocircuitry activity. Specifically, studies probing episodic memory encoding and retrieval found that patients display insufficient recruitment of prefrontal cortex (PFC) resources (Hall et al., 2010;Kassel et al., 2016;Oertel-Knöchel et al., 2013) as well as hippocampal hypo-activity (Kassel et al., 2016;Kelley et al., 2013;Meusel et al., 2013) and hyper-activity in the default mode network (DMN) (Oertel-Knöchel et al., 2014). However, findings are not uniform, with other studies showing encoding-related dorsal PFC (dPFC) hyper-activity (Dietsche et al., 2014;van Eijndhoven et al., 2011) or normal hippocampal activation (van Eijndhoven et al., 2012;Werner et al., 2009). The inconsistencies are likely partially attributable to differences in the cognitive status of patients across studies given the well-documented cognitive heterogeneity in affective disorders (Burdick et al., 2014;Jensen et al., 2016;Kjaerstad et al., 2021). We proposed that the conflicting findings may be reconciled in an integrative model of the association between task difficulty dPFC engagement based on the evidence from a systematic review of these studies (Miskowiak and Petersen, 2019). The model proposes a bell-shaped load-response curve, indicating increasing neuronal response with increasing task load followed by decreasing activity when the task exceeds cognitive abilities (Miskowiak and Petersen, 2019). In patients, there is a leftward shift in this load-response curve relative to healthy controls (HC). Hence, impaired performance in patients at high loads is associated with hypoactivity (reflecting reduced cognitive capacity) whereas preserved performance is associated with hyper-activity (reflecting increased cortical effort) (Miskowiak and Petersen, 2019). Accordingly, normalisation of aberrant (hyper-and hypo-activity) dPFC and of DMN hyper-activity are putative biomarkers of pro-cognitive effects of interventions (Miskowiak and Petersen, 2019;. This aligns with the most consistently observed neuronal changes in response to treatments targeting cognition in a new systematic review by the International Society of Bipolar Disorders (Miskowiak et al., 2022c).
In contrast, insights into the neurocircuitry origins of functional impairment are limited. The International Classification of Functioning, Disability and Health (ICF) is recommended as a conceptual model to describe functioning and disability within mental health in general and within bipolar disorder (Avila et al., 2010;Ayuso-Mateos et al., 2013;Ustün and Kennedy, 2009;WHO, 2001). Functional impairment thus refers to difficulties in managing daily life, e.g., performing adequately at work, managing household tasks and finances, or upholding friendships (Rosa et al., 2007). Functional impairment can stem from several factors, including cognitive impairment (Tse et al., 2014). One fMRI study of patients with schizophrenia found that deficient recruitment of dPFC during cognitive control of task-irrelevant negative information was associated with more distress in real-life socially stressful situations (Tully et al., 2014). Similarly, another study in schizophrenia found that dPFC dysfunction during a continuous performance task correlated with poorer daily functioning (Yoon et al., 2008). Consistently, we found that dPFC hypo-activity during attempts to downregulate negative emotions was associated with more functional disability in daily life in remitted BD patients (Kjaerstad et al., 2022). Together, these emerging findings suggest that dPFC plays a critical role in real-life functioning across these patient groups. Indeed, dPFC is involved in higher-order executive functions, including top-down regulation of cognitive processes, decision making, and working memory, all of which are utilised when managing complex daily-life situations (Emch et al., 2019;Krain et al., 2006;Niendam et al., 2012;Reber et al., 2002). Thus, failure to adequately recruit dPFC in these situations could ostensibly lead to functional disability and treatments that target dPFC deficits may therefore promote better functional outcomes. However, there is a need for large-scale fMRI studies to confirm this putative marker of functional impairment in affective disorders.
In the present fMRI study, we investigate the neural correlates of episodic memory impairments and the relation to functional disability in a large sample of fully or partially remitted patients with affective disorders. The aims were twofold: (I) to investigate differences in encodingrelated neural activity between cognitively and functionally impaired patients and HC, (II) to investigate the associations between task-related dPFC activity, memory performance, and daily functioning. The hypotheses were that (I) cognitively and functionally impaired patients would display hypo-activity in task relevant areas (dPFC and hippocampus), (II) lower encoding-related dPFC activity would correlate with poorer learning and memory performance and daily functioning.

Participants
The patients in the present study were recruited as part of a randomised controlled trial of the effect of erythropoietin (EPO) on cognitive impairment (PRETEC-EPO) (Petersen et al., 2018). Patients were recruited from psychiatric centres and consultant psychiatrists in the Capital Region of Denmark and through online advertisement on relevant websites from July 2017 to October 2021. Patients were assessed for eligibility according to the following criteria: diagnosis with BD or recurrent MDD according to ICD-10 criteria confirmed with the Schedules for Clinical Assessment in Neuropsychiatry (SCAN), full or partial remission (scores≤14 on Hamilton Depression Rating Scale (HDRS-17 (Hamilton, 1960)) and Young Mania Rating Scale (YMRS (Young et al., 1978). (scores≤7 were considered full remission), age between 18 and 65 years, fluency in Danish, and objective cognitive impairment defined as a total score ≤ 77 on the Screen for Cognitive Impairment in Psychiatry (SCIP) (Ott et al., 2016) or scores more than one standard deviation (SD) below the norm on ≥two subtests. Exclusion criteria were daily use of benzodiazepines (≥22.5 mg), estimated IQ < 70, dyslexia, diagnosis with schizophrenia or schizoaffective disorder, neurological disorder, alcohol or substance abuse present<three months before inclusion, and history of severe head trauma. The exclusion criteria for patients were: severe medical conditions (i.e., heart disease, diabetes, renal failure, insufficiently treated hypertension, malignancies/tumours, thromboses or epilepsy), first-degree relatives with thromboses or epilepsy, electroconvulsive therapy up to three months before inclusion, contraceptive medication, smoking, pregnancy, or breastfeeding.
Thirty-six matched healthy controls (HC) with no personal or firstdegree familial history of psychiatric disorders that underwent a similar assessment were included from the Bipolar Illness Onset (BIO) study .
The study was approved by the Danish Research Ethics Committee for the Capital Region of Denmark (protocol number H-16043370) and The Danish Data Protection Agency Capital Region of Denmark (protocol number RHP-2017-020). The study was carried out in accordance with the Declaration of Helsinki. All participants provided written informed consent before inclusion.

Neuropsychological tests
Patients were assessed with a cognitive battery of different domains, including verbal learning and memory, working memory and executive functions, and attention and psychomotor speed (see protocol (Petersen et al., 2018)). Verbal learning and memory was assessed with the Rey Auditory Verbal Learning Test (RAVLT) (Schmidt, 1996). The learning and memory domain composite score was calculated from the following subtests: total recall list I-V, immediate recall, delayed recall, and recognition. Alternate, matched forms of the RAVLT were used across three assessment points in a counter-balanced fashion to minimize learning effects. Verbal IQ was estimated using the Danish Adult Reading Task (DART) (Nelson and Willison, 1991).

Functional abilities
Functioning during the last 14 days was assessed with the clinicianrated interview Functioning Assessment Short Test (FAST). The FAST is an interviewer-administered instrument which is designed to be used by a trained clinician; the studied time frame refers to the last 14 days before assessment. It comprises 24 items, which are divided among 6 specific areas of functioning: 1) Autonomy refers to the capacity of the patient of doing things alone and take his/her own decisions. 2) Occupational functioning refers to the capacity to maintain a paid job, efficiency of performing tasks at work, working in the field in which the patient was educated and earning according to the level of the employment position. 3) Cognitive functioning is related to the ability to concentrate, perform simple mental calculations, solve problems, learn new information and remember learned information. 4) Financial issues involve the capacity of managing the finances and spending in a balanced way. 5) Interpersonal relationships refer to relations with friends, family, involvement in social activities, sexual relations and the ability to defend ideas and opinions. 6) Leisure Time refers to the capacity of performing physical activities (sport, exercise) and the enjoyment of hobbies. All of items are rated using a 4-point scale, 0 = no difficulty, 1 = mild difficulty, 2 = moderate difficulty and 3 = severe difficulty. The global score is obtained when the scores of each item are added up (Rosa et al., 2007). A FAST total score above 11 indicates functional impairment, with the thresholds of severity being: no impairment in functioning (scores between 0 and 11), mild impairment (scores between 12 and 20), moderate impairment (scores between 20 and 40) and severe impairment (scores between 40 and 72) (Bonnín et al., 2018).

Picture encoding fMRI task
The picture encoding fMRI paradigm was based on the picture encoding tasks from Hariri et al. (2003) and Miskowiak et al. (2007) and identical to the one used in our previous EPO-trial, where it was shown to engage dPFC at baseline . Patients were instructed to classify whether different pictures were of either indoor or outdoor scenes and to try to encode them. Pictures were neutral in valence and arousal and matched in complexity. Six picture blocks consisting of six serially presented pictures (3 s each) interleaved with a fixation cross (1 s) were presented. A fixation cross (24 s) was presented between blocks. The task was followed by a free recall test outside the scanner. Here, participants were asked to describe each picture that they remembered in as much detail as possible, and correctly recalled pictures were scored.

MRI data acquisition
Functional MRI data were acquired at the Copenhagen University Hospital, Rigshospitalet using a 3 Tesla Siemens Prisma scanner and a 64-channel head-neck coil. During the performance of the picture task, blood oxygen level dependent (BOLD) fMRI was acquired using a T2*weighted gradient echo spiral echo-planar (EPI) sequence with an echo time (TE) of 30 ms, repetition time (TR) of 2 s, and flip angle of 90 • . A total of 146 volumes were acquired, each consisting of 32 slices with a slice thickness of 3 mm with 25% gaps in-between, and a field of view (FOV) of 230 × 230 mm using a 64 × 64 grid. The BOLD images were registered to a T1-weighted structural images acquired during the scan (TR = 1900 ms; TE = 2.58 ms; flip angle = 9 • ; distance factor = 50%; FOV = 230 × 230 mm; slice thickness = 0.9 mm). A standard B0 field map sequence was also acquired with the same FOV and resolution as the fMRI sequence (TR = 400 ms; TE = 7.38 ms; flip angle = 60 • ) and used for geometric distortions correction of the BOLD images. The quality of the functional and structural images was checked by visual inspection.

Definition of the region of interest
Based on our a priori hypothesis, we constructed anatomical dPFC mask using FSLView 4.0.1 on a standard MNI template based on the Harvard-Oxford cortical structural Atlas probabilistic maps (Desikan et al., 2006) by including bilateral superior and medial frontal gyri and the superior portions of the anterior division of the cingulate gyrus and the frontal poles thresholded at 5%. The ventral border of the dPFC was defined by the plane separating the dorsal from the ventral regions of medial PFC (MNI z > 5), defined according to (Veit et al., 2010). Using a similar procedure, we also constructed a left and right mask for the hippocampi. We defined two spheric (8 mm radius) regions of interest (ROIs) for the right dorsolateral PFC (dlPFC) centred on the peak coordinates (x = 40, y = 34, z = 29) as used in .

fMRI pre-processing and subject-level analyses
The analysis of the fMRI data was performed using FEAT (Woolrich et al., 2001), part of FMRIB Software Library version (FSL) version 6.0.4 (Jenkinson et al., 2012). The pre-processing steps included non-brain removal, motion correction using rigid-body transformation, linear registration to the individual T1-weighted image and non-linear registration to the standard MNI (Montreal Neurologic Institute) space at 2 mm isotropic voxel size, and spatial smoothing using a 5 mm full-widthhalf-maximum gaussian kernel. A high-pass temporal filtering cut-off of 100 s was applied. Correction for geometric distortions related to the B0 field was performed based on the acquired B0 field map. All participants' registration to the MNI template and movement parameters were visually inspected.
At subject level, we modelled the picture encoding task using a block design, with the picture presentation (i.e., memory encoding) and fixation cross as events that were convolved with a double-gamma hemodynamic response function. In addition, the subject-level GLM included basic movement regressors (n = 6) and temporal derivatives of task regressors to model slice-timing effects. The picture memory encoding performance was assessed outside the scanner immediately after the MRI session by recording the number of correctly recalled pictures as described above.

fMRI group-level analyses
At the group level, we first confirmed task activations at baseline in the HC group by entering the picture encoding contrast from each individual in a one-sample t-test. We secondly investigated baseline differences between patients and HC. The significance level for clusters was set at p < .05 corrected for multiple comparisons using Gaussian Random Field (GRF) theory subsequent a cluster-forming threshold of z = 2.57 (p < .005). In SPSS v28 (IBM Corporation), we investigated if group differences remained significant after adjustment for mood symptoms using ANOVAs.
We further assessed group differences in the extracted mean percent BOLD signal change from a right dlPFC ROI and the hippocampi using independent sample t-tests or Mann-Whitney U tests according to normality distribution in SPSS using a α-level of 0.05.

Statistical analyses of demographic, clinical, and cognitive data
Patients' RAVLT raw scores were z-transformed using the means and SDs from the HC. These standardised scores were averaged into a verbal learning and memory composite. Data normality distributions were assessed with Shapiro-Wilk tests (Shapiro and Wilk, 1965) and visual inspections for each variable included in the analyses.
Patients and HC were compared on demographic variables, mood symptoms, correctly recalled pictures, verbal learning and memory performance, and FAST scores (total scores and subdomain scores) using independent samples t-tests or Mann Whitney U tests for continuous data and Pearson's Chi-square (χ2) tests for categorical data.
In the patient sample, we investigated correlations between extracted BOLD responses from the right dlPFC ROI, hippocampi, and dPFC clusters that showed differences between patients and HC and (i) verbal learning and memory composite, and (ii) level of functioning (FAST). For this we used either Pearson's correlations for normally distributed data or Spearman's Rho correlations for non-normally distributed data and implemented bootstrapping (no. of samples = 5000) to obtain 95% confidence intervals. For these analyses, α-level of significance was set at p < .01 to adjust for multiple comparisons based on the Bonferroni method for correlated measures with a mean r of 0.24 and n = 8 measurements (Sankoh et al., 1997).
As a sensitivity analysis, we wanted to control for mood symptoms (HDRS-17 and YMRS scores) in any significant correlations between BOLD response and functioning. For this, we used partial correlations and implemented bootstrapping (no. of samples = 5000). Further, we wanted to investigate if any correlations between BOLD response and functioning were independent of cognition. We therefore conducted multiple regression analyses with FAST total as dependent variable, BOLD response as predictor and verbal learning and memory as a covariate. We also subtracted the scores on the FAST cognitive functioning subdomain from the total FAST scores and investigated the correlations between this score and BOLD response with bootstrapping (no. of samples = 5000).
As post-hoc analyses within the group of patients we compared the extracted BOLD responses from the above-mentioned clusters between i) patients who did and did not take different classes of medication (antidepressants, antipsychotics, anticonvulsants, lithium, other medications, and polypharmacy (3 or more classes)), and ii) patients with MDD and BD, using independent samples t-tests or Mann Whitney U test. Further, patients with MDD and BD were compared on demographic, functional, cognitive, and clinical data to ensure homogeneity across diagnoses using independent sample t-tests or Mann-Whitney U tests according to normality distribution.
Data were analysed with SPSS v28 using a α-level of 0.05.

Demographic and clinical characteristics and group comparisons
Demographic, clinical, and cognitive comparisons between patients and HC are presented in Table 1. A total of 127 participants (n = 91 patients (n = 67 BD, n = 24 MDD), n = 36 HC) were included in this study. Patients and HC were matched for age, verbal IQ, and sex (ps ≥ 0.12) but had more subsyndromal mood symptoms (ps < 0.001), and fewer years of education (p = .003). Patients displayed baseline impairment in the verbal learning and memory composite (p < .001) but no difference in correctly recalled pictures after the scan compared to HC (p = .79). Moreover, patients presented with substantial functional deficits as measured with the FAST (total score and all subdomains) relative to HC (p < .001).

Neural activations during episodic memory encoding
During episodic memory encoding the HC activated two prefrontal regions in the bilateral dlPFC and dorsomedial PFC (dmPFC), respectively, within the dPFC mask ( Fig. 1 bottom, green regions). Patients showed lower encoding-related activity compared to HC in the dmPFC region engaged by the task as well as the left dlPFC ( Fig. 1 bottom, redyellow regions). This remained significant after adjustment for mood symptoms (ps ≤ 0.002). There were no differences in BOLD response variance between patients and HC (ps ≥ 0.11). Table 2 lists the clusters activated by the HC during task performance and clusters identified within the dPFC mask showing hypo-activity in patients. In contrast with hypothesis (I), there were no differences in task-related activations Notes: M = mean. SD = standard deviation. IQ = intelligence quotient. HDRS-17 = Hamilton Depression Rating Scale 17-items version. YMRS=Young Mania Rating Scale. FAST = Functioning Assessment Short Test. BD=Bipolar disorder. MDD = Major depressive disorder. Chi-square for categorical variables, Mann-Whitney for non-parametric data (median (IQR)), independent t-tests for normally distributed data (mean (SD)). Data was missing for 1 patient for educational years, 2 patients for DART, 1 patient for FAST total, 1 patient for employment status, 5 patients for depressive episodes, 3 patients for manic episodes, 7 patients for hypomanic episodes, 7 patients for mixed state episodes, and 15 patients for total number of episodes. **p < .01, ***p < .001 (two-tailed). 1 Employment status "Other" included: current sick leave, vocational rehabilitation programme participation, receiving early retirement pension, or working in flexible job. 2 In categories (0 = 0; 1 = 1-5; 2 = 6-10; 3 = 11-15; 4 = 16-20; 5 = 20+). 3 Comorbidities included ADHD/ADD (n = 6), anxiety disorders (n = 2), borderline personality disorder (n = 1), autism spectrum disorder (n = 2) in the a priori right dlPFC or hippocampi ROIs between the patients and HC (ps ≥ 0.10) (Fig. 1 top).

Associations between encoding-related activity and cognition and functioning
The statistical data from the correlation analysis between extracted encoding-related BOLD response from the right dlPFC ROI, the hippocampi, and the dPFC clusters where patients presented with hypoactivity compared to HC, and cognition, and level of functioning are presented in Table 3. Within the patient sample, there was a correlation between lower right dlPFC activity and poorer daily functioning (higher FAST total scores) (p = .01). There were no significant correlations between activity in the rest of the regions and either the verbal learning and memory composite or FAST scores (ps ≥ 0.23), in contrast with hypothesis (II).

Sensitivity analyses
The correlation between FAST total score and activity in the right dlPFC ROI remained significant when controlling for mood symptoms (r (86) = − 0.28, 95% CI [− 0.46, − 0.07], p = .009). In the multiple regression analysis, dlPFC activity was a significant predictor of FAST total scores, even when controlling for verbal learning and memory composite score (B = -4.89, p = .018). Finally, there was a significant correlation between dlPFC activity and FAST total scores with the cognitive functioning subdomain subtracted (r(88) = − 0.25, 95% CI [− 0.44, − 0.04], p = .016).  Table 2 Peak cluster activation during picture encoding, in patients with affective disorders and healthy controls (HC), in the dorsal prefrontal cortex volume of interest. Clusters significant at corrected p < .005 (z > 2.57).

Region
Hemisphere Notes: BA = Brodmann area. Cluster size = number of voxels in the significant cluster. p = corrected p-value of the cluster. x, y, z = MNI coordinates of cluster maxima. Z-stat = max statistical Z-values for voxel. dPFC=Dorsal prefrontal cortex; VOI=Volume of interest.

Post-hoc analyses within the patient group
Post-hoc analyses of the extracted BOLD signals from the dlPFC ROI, hippocampi, and the dPFC clusters showing hypo-activity in patients revealed no significant differences between patients with MDD and BD, respectively in any regions (ps ≥ 0.39). Patients treated with antipsychotics displayed hypo-activity in the right dlPFC ROI compared to patients not taking antipsychotics (t(89) = 2.88, p = .005) (for all other medication classes and polypharmacy ps ≥ 0.10). Patients with MDD and BD were comparable on gender, age, and education, although patients with MDD had higher verbal IQ than patients with BD (p = .019) (Supplementary Table 1). Patients with MDD and BD showed no differences across functional, cognitive, or clinical domains (ps ≥ 0.08), except for total number of episodes and medication use (ps ≤ 0.04) with more BD patients getting lithium, anticonvulsants, and antipsychotics and more MDD patients getting antidepressants in accordance with current treatment guidelines (Supplementary Table 1).

Discussion
This is the largest to date fMRI study of the association between neuronal activity during cognitive performance and functional impairments in affective disorders. We investigated encoding-related neural activity in 91 cognitively and functionally impaired, somatically healthy, fully or partially remitted patients with affective disorders (n = 67 BD, n = 24 MDD), and the association with cognition and daily functioning. We found that the cognitively and functionally impaired patients presented with encoding-related hypo-activity in the dPFC compared to HC (n = 36) at baseline in partial support of our hypothesis (I). Lower right dlPFC activity was associated with more functional impairment in partial support of our hypothesis (II). Further, patients taking antipsychotics displayed more dlPFC hypo-activity compared to patients who did not take antipsychotics. In contrast with our hypotheses, there were no differences in hippocampal activity between patients and HC and no correlations between neural activity and the verbal learning and memory composite.
Our main finding is consistent with the emerging evidence that highlights dPFC hypo-activity as a neuronal correlate of functional disability in daily life across affective disorders and schizophrenia (Kjaerstad et al., 2022;Tully et al., 2014;Yoon et al., 2008). This is the hitherto largest study to report this finding in patients with affective disorders across BD and MDD, albeit with a modest effect size (r = 0.27). It is well-known that dPFC hypo-activity is also associated with cognitive impairment across affective disorders and schizophrenia (Miskowiak and Petersen, 2019;Zarp Petersen et al., 2021). Together, this and previous studies suggest that the broad difficulties observed in patients with these disorders in both cognition and daily functioning may stem from shared abnormalities in neurocircuitry activity. Importantly, the association between dPFC hypo-activity and poorer daily functioning remained significant in the exploratory sensitivity analyses even when controlling for mood symptoms and cognitive performance as well as when the cognitive subdomain from FAST was removed. This suggests that intact dPFC activation is relevant not only for cognition specifically, but for overall daily functioning as well. Thus, task-relevant dPFC hypoactivity may be a marker of a general failure to recruit prefrontal resources in cognitively challenging situations across daily life situations.
Our finding of more dlPFC hypo-activity in patients treated with antipsychotics contrasts with previous findings in BD and schizophrenia patients that indicate that antipsychotics do not affect BOLD signal specifically (Hafeman et al., 2012) or even may have an ameliorating effect over time (Abbott et al., 2013). Antipsychotic medication is often given to patients with more severe illness burdens (Yatham et al., 2018). Our finding therefore may reflect that the more severely ill part of our sample has more hypo-activity in dlPFC. Indeed, post-hoc comparisons revealed that patients taking antipsychotics had more previous mood episodes in total (p = .009) and more previous hypomanic episodes (p = .001) than patients not taking antipsychotics (no differences in illness duration, hospitalisations, or depressive, manic, or mixed episodes). However, there is also evidence that points to cognitive performance being dependent on adequate levels of dopamine in the PFC with too little and too much dopamine attributing to cognitive impairment Schacht, 2016). As antipsychotic treatment down-regulates dopamine levels, those patients with illness-related cognitive impairment may therefore suffer from reduction in the recruitment of dPFC resources during memory encoding.
This study has implications for designs of future interventions that target cognitive and functional impairments in affective disorders. Previous pro-cognitive interventions have generally been characterised by a limited transfer between improved cognition and enhanced daily functioning (Miskowiak et al., 2022b;Woolf et al., 2022). This could in part be because training of cognitive skills using abstract computerised tasks may have little benefit in daily life, in which patients are provided with no instructions on how to handle and tackle ambiguous cognitive challenges or make complex decisions (Vita et al., 2021). Our demonstration of an association between lower dlPFC activity and more functional impairment suggests that in order for pro-cognitive interventions to aid real-world functioning, they should directly engage the dlPFC. This suggestion is in keeping with the critical involvement of this region across executive functions, including planning, initiation, and inhibition of actions, and flexibly shifting between tasks (Niendam et al., 2012) and metacognition (Vaccaro and Fleming, 2018). Such skills are of great importance when managing complex life situations, and therefore, a specific focus on strengthening dlPFC function and these executive skills in interventions could be of benefit to both cognition and real-life functioning. Indeed, it has been suggested that improved metacognition (i.e., the ability to monitor, understand, and manipulate one's own cognitive processes (Flavell, 1979)) is a driving mechanism behind transfer between improved cognition and enhanced functioning in cognitive remediation trials in affective disorders (Tsapekos et al., 2021). Thus, specific executive function exercises in adaptive cognitive training programmes, such as action-based cognitive remediation (ABCR), and/or metacognitive interventions could be examples of viable treatment strategies to engage dPFC Tsapekos et al., 2021). Importantly, enhanced metacognition and cognitive control may also aid in preventing relapse in affective disorders. In fact, dPFC hypoactivity was linked to an increased risk of relapse in BD (Kjaerstad et al., 2022), and we recently found in a systematic review that poorer executive functioning was a risk marker for (hypo)manic relapse (Miskowiak et al., 2022a). Strengths of this report include the large fMRI sample of patients with affective disorders and matched HC with comprehensive neuropsychological and neuroimaging data. It was a strength that patients were matched on IQ rather than educational attainment because patients often have lower educational attainment compared to HC with similar premorbid intellectual capacity due to long time periods with mood episodes and symptom instability which challenges their ability to sustain study activities (Sletved et al., 2021). Another strength is that our patient sample was somatically healthy and without comorbid alcohol or substance abuse disorder, allowing us to exclude the potential impact of somatic comorbid conditions on patients' cognitive impairments per se. Similarly, although functional and cognitive impairments are often most prominent in symptomatic phases, the inclusion of patients in full or partial remission at inclusion allowed us to investigate their neural correlates with minimal confounding effects of mood symptoms on cognition and task-related neural activity (Fernández-Corcuera et al., 2013). However, we did not include a requirement for length of remission prior to inclusion, which could be considered a limitation. Further, the extensive somatic exclusion criteria may also limit generalisability of our results. Neural activity may have been influenced by psychotropic medication use. However, as pharmacological interventions are the first line of treatment in affective disorders (Gabriel et al., 2020;Yatham et al., 2018) it is not representative to only include patients who did not take medications, and in this study 23% did not take medication. Importantly, there were no differences in BOLD response in any of the clusters or ROIs between patients who did or did not take medication (ps ≥ 52), and the correlation between functioning and dPFC activity was still significant even when removing patients who did not take medication from the analyses (p = .016). Finally, the picture encoding paradigm may not sufficiently tap into hippocampal memoryencoding processes , as there were no correlation between neuronal activity and learning and memory composite across the sample and no differences in hippocampal BOLD response data between the included patients with memory impairment and HC. Comparison with our previous study in a similar patient sample reveal lower post-scan recall scores in the current sample (mean ± SD, HC: 6 ± 3, patients: 6 ± 4) compared with our previous study (mean ± SD, patients: 8 ± 4) . Since different instructors were involved for the investigations, it is possible that the emphasis on the importance of the recall part after the scan was not the same. However, the picture encoding paradigm was primarily designed to investigate the neural basis of memory encoding and not optimized to measure recall performance. Therefore, the association between memory-related hippocampal response and functional impairment should be investigated further using paradigms with robust hippocampal engagement and the ability to detect behavioural differences.

Conclusion
In conclusion, we found that remitted patients with affective disorders and cognitive impairment presented with hypo-activity in the dPFC during encoding compared to HC. In the patient sample, lower activity in dPFC was associated with poorer overall functioning and more antipsychotic drug use. Our findings indicate that treatments that enhance dPFC activity in patients with affective disorders may show beneficiary effects on both cognition and daily functioning.

Ethical statement
The study was approved by the Danish Research Ethics Committee for the Capital Region of Denmark (protocol number H-16043370) and The Danish Data Protection Agency Capital Region of Denmark (protocol number RHP-2017-020). The study was carried out in accordance with the Declaration of Helsinki. All participants provided written informed consent before inclusion.

Role of funding source
The study is supported by the Lundbeck Foundation grant number: R215-20154121. The Lundbeck Foundation has not been involved in the design of the study, the data collection, analysis, or interpretation of data, the writing of this manuscript, or decision to publish.

Author contributions
KWM conceptualized the study, aquired funding, and supervised the investigation. KWM, JMac, PF, GMK, and LVK contributed to methodology. LVK, MV, and MBJ supervised the medical safety of all study participants. JZP was responsible for data curation. JMac, JMar, and JZP performed the formal analysis. JMac and JMar wrote the original draft. All authors reviewed, edited and approved the final version of the manuscript.

Declaration of Competing Interest
JMac, JMar, JZP, PMF, and MBJ report no conflicts of interest. MV has within the last three years received consultancy fees from Janssen Cilag and Lundbeck. GMK has received honoraria as a speaker for Sage Biogen and as a consultant for Sanos. LVK has within recent three years been a consultant for Lundbeck and Teva. KWM has received consultancy fees from Lundbeck and Janssen in the past three years.