Treatment resistant recurrent unipolar and bipolar depression: associative learning abnormalities

TY - JOUR

T1 - Treatment resistant recurrent unipolar and bipolar depression

T2 - associative learning abnormalities

AU - Suveges, Szabolcs

AU - Wang, Yuxi

AU - Tolomeo, Serenella

AU - Gilbertson, Tom

AU - Steele, Douglas

PY - 2025/10/1

Y1 - 2025/10/1

N2 - Severe and enduring psychiatric illness affects about 3% of the UK population and is associated with significant disability and a substantial reduction in average life expectancy. Two types are treatment-resistant recurrent unipolar depression and treatment-resistant bipolar depression, the latter being the depressed phase of bipolar disorder. Different phenotypes and different responses to antidepressant medications suggest different neural abnormalities. As bipolar depression can be clinically indistinguishable from unipolar depression yet require different treatments, it is important to develop objective ways to discriminate these two illnesses. Here, we used reinforcement learning drift diffusion models of decision-making, and event-related fMRI acquired during a reward gain and loss avoidance task, to investigate patients with treatment-resistant recurrent unipolar depression and bipolar depression, in long-term General Adult Psychiatry follow-up. We tested the null hypothesis that both unipolar and bipolar depressive illnesses show similarly blunted reward learning signals, and increased loss avoidance learning signals, with similar psychomotor slowing. Consistent with our null hypothesis, we found abnormally slowed decision-making for both depression types, with individual patient reinforcement learning drift diffusion model parameter estimates correlating with depression severity. For unipolar depression, we found blunted outcome and value signals for positive feedback, and increased signals for negative feedback. However, in contrast to our null hypothesis, bipolar depression was associated with preserved striatal reward prediction error signalling, and an absence of hippocampal and lateral orbitofrontal enhanced encoding of loss events, which was present for unipolar depression. Overall, both treatment-resistant recurrent unipolar depression and treatment-resistant bipolar depression showed a similar pattern of neural abnormality compared with controls for the lateral orbitofrontal cortex reward value signal and the amygdala loss value signal. However, the illnesses also differed significantly, particularly with regard to hippocampal, striatal and lateral orbitofrontal function, potentially allowing objective discrimination. Using a support vector machine with the results of our neuroimaging analyses, it was also possible to differentiate the two depression types with an accuracy of 74.3%. Further studies of currently ill patients with severe and enduring illness are indicated.

AB - Severe and enduring psychiatric illness affects about 3% of the UK population and is associated with significant disability and a substantial reduction in average life expectancy. Two types are treatment-resistant recurrent unipolar depression and treatment-resistant bipolar depression, the latter being the depressed phase of bipolar disorder. Different phenotypes and different responses to antidepressant medications suggest different neural abnormalities. As bipolar depression can be clinically indistinguishable from unipolar depression yet require different treatments, it is important to develop objective ways to discriminate these two illnesses. Here, we used reinforcement learning drift diffusion models of decision-making, and event-related fMRI acquired during a reward gain and loss avoidance task, to investigate patients with treatment-resistant recurrent unipolar depression and bipolar depression, in long-term General Adult Psychiatry follow-up. We tested the null hypothesis that both unipolar and bipolar depressive illnesses show similarly blunted reward learning signals, and increased loss avoidance learning signals, with similar psychomotor slowing. Consistent with our null hypothesis, we found abnormally slowed decision-making for both depression types, with individual patient reinforcement learning drift diffusion model parameter estimates correlating with depression severity. For unipolar depression, we found blunted outcome and value signals for positive feedback, and increased signals for negative feedback. However, in contrast to our null hypothesis, bipolar depression was associated with preserved striatal reward prediction error signalling, and an absence of hippocampal and lateral orbitofrontal enhanced encoding of loss events, which was present for unipolar depression. Overall, both treatment-resistant recurrent unipolar depression and treatment-resistant bipolar depression showed a similar pattern of neural abnormality compared with controls for the lateral orbitofrontal cortex reward value signal and the amygdala loss value signal. However, the illnesses also differed significantly, particularly with regard to hippocampal, striatal and lateral orbitofrontal function, potentially allowing objective discrimination. Using a support vector machine with the results of our neuroimaging analyses, it was also possible to differentiate the two depression types with an accuracy of 74.3%. Further studies of currently ill patients with severe and enduring illness are indicated.

KW - bipolar depression

KW - recurrent depressive disorder

KW - treatment-resistant illness

KW - RLDDM

KW - event related fMRI

KW - event-related fMRI

U2 - 10.1093/brain/awaf280

DO - 10.1093/brain/awaf280

M3 - Article

C2 - 40755210

SN - 0006-8950

VL - 148

SP - 3705

EP - 3717

JO - Brain

JF - Brain

IS - 10

ER -