The RDoC Positive Valence Systems (PVS) encompass motivational processes underlying normal reward-
guided behavior and its alterations in many mental disorders. Yet, the theoretical links between the PVS
constructs of Reward Responsiveness, Learning, and Valuation remain under-specified. Hence, our goal is to
unify them under a new model of computational reinforcement learning with momentum dynamics wherein
momentum reflects whether recent outcomes have generally exceeded or fallen short of our expectations,
signaling an improving or worsening reward rate. Momentum is closely linked with mood and our model offers
new insights into the interplay of mood and reward learning. Thus, we are seeking to provide a mechanistic
account of transdiagnostic mood dynamics and affective instability (AI), a dimension of psychopathology seen
in depression, anxiety, eating and personality disorders, and suicidal behavior. While ecological momentary
assessment (EMA) studies of AI have shown how mood changes over time in mental illness, to date we have
no formal model that can explain why it changes thus. On the other hand, lab-based experimental studies have
used tools from cognitive neuroscience to explore potential neural mechanisms of affective instability. Though
promising, lab studies are too brief to capture the temporal dynamics of AI in psychopathology, which typically
unfold over hours or days. Here, we overcome the limitations of EMA and laboratory studies to date by bringing
together key elements of both within a framework grounded in reinforcement learning and dynamical systems
theory. To this end, we will combine mood tracking with learning experiments carried out in daily life over 4
weeks, concurrently recording neurophysiological signals via wearable heart rate and electroencephalography
sensors. We have shown that this platform captures the behavioral and physiological effects of positive and
negative outcomes, and that physiological learning signals predict day-to-day changes in subjects’ mood. We
will use this platform to examine PVS constructs and AI in individuals sampled from the community (Sample 1,
n = 300) and in a clinical sample of individuals with borderline personality (Sample 2, n = 150) recruited from
two ongoing studies in Pittsburgh and State College, PA. In our earlier study, mood induction that impacted
reward valuation also impacted striatal reward responsiveness. Here, we will investigate the cortico-striatal
substrates of momentum dynamics in relation to real-life mood fluctuations by combining mobile longitudinal
assessment with model-based fMRI. During the scan, subjects will choose between experimental stimuli they
previously encountered in different moods. This will allow us to examine how mood impacts neural valuation
and learning signals and how learning signals shape future mood. Our interdisciplinary team has expertise in
computational modeling of mood and its integration with EMA, physiology and imaging (Eldar), computational
model-augmented functional imaging and EMA in clinical populations (Dombrovski, Hallquist), and
neuroimaging methods (Hallquist).
