Despite growing concerns about validity, the NIMH Research Domain Criteria (RDoC) framework plays a key
role in organizing basic, translational, and clinical research. RDoC’s approach to fear and anxiety is categorical:
threat is either acute or potential; engages either the Amygdala or the bed nucleus of the stria terminalis (BST);
and elicits either fear or anxiety. Recent work casts doubt on this binary perspective, spurring the development
of alternative approaches. Dimensional models posit that threat responses vary along a smooth continuum of
perceived danger—from absolutely safety to on-going attack. Danger perceptions are thought to emerge from
parametric estimates of threat proximity, probability, and certainty, which are computed in weakly segregated
cortico-subcortical circuits. To date, there have been no systematic, well-powered efforts to computationally
implement these competing models and compare their validity. Furthermore, while both models highlight the
importance of threat uncertainty, they do not specify which kind. Computational psychiatry recognizes 2
mathematically distinct kinds of uncertainty: Risk and Ambiguity. Which of these is more relevant to threat
reactivity and how they map onto the underlying neurobiology is unknown. To address these fundamental
questions, we will recruit a racially diverse community sample enriched for elevated fear/anxiety symptoms. Two
parametric threat-anticipation paradigms will allow us to simultaneously probe circuits sensitive to categorical
(RDoC) and dimensional variation in threat for the first time. Smartphone phenotyping will assess real-world
threat exposure, uncertainty, and distress. A1. We will test a series of competing predictions about the
architecture of threat-sensitive brain circuits. We will use theory-driven computational modeling to go beyond
binary threat categories; identify regions sensitive to risk, ambiguity, and other dimensional facets of threat; and
explore trial-by-trial relations with signs and symptoms of fear and anxiety. A2. RDoC implies that Acute and
Potential Threat are represented in different patterns of brain activity; indeed, this was the major rationale for
creating separate RDoC constructs. Dimensional models predict substantial similarities. Multivoxel machine-
learning approaches provide a rigorous means of adjudicating these claims and clarifying the importance of the
Amygdala, BST, and other regions. A3. Fusing the fMRI and smartphone data-streams will enable us to establish
the relevance of specific facets of threat and specific brain regions to real-world distress. We will also explore
relations between neuroimaging metrics and fear- and anxiety-related diagnoses, symptoms, and traits.
Significance. Extreme fear and anxiety are leading causes of human misery and morbidity. This project will
provide a potentially transformative opportunity to develop the first computationally grounded model of fear and
anxiety. It will help adjudicate on-going theoretical debates, validate a new conceptual approach for use with
other read-outs and species, set the stage for new kinds of translational models and clinical studies, prioritize
new targets for neuromodulation and other therapeutics development, and guide the development of RDoC 2.0.
