🧬 biology

Personality Heterogeneity Shapes the Neural Organization of Conditioned Threat Learning

This study demonstrates that individual differences in neuroticism shape the neural organization of conditioned threat learning by reducing the brain's ability to differentiate threat from safety signals, a pattern that specifically links neural threat representations to anxiety vulnerability within high-neuroticism individuals.

Original authors: Mohammed Milad, Kai Zhang, Zhenfu Wen, Lihan Cui, Zakiya Atiyah, Isabel Moallem, Hani Meelad, Joe Borrelli, Isabella Mason

Published 2026-07-03

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Mohammed Milad, Kai Zhang, Zhenfu Wen, Lihan Cui, Zakiya Atiyah, Isabel Moallem, Hani Meelad, Joe Borrelli, Isabella Mason

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

The Big Idea: It's Not About Volume, It's About Tuning

Imagine your brain's reaction to danger is like a radio. For a long time, scientists thought that people who get anxious easily (those with high "neuroticism") were just like radios turned up to maximum volume. They thought these people simply heard every threat louder and reacted more strongly.

This study suggests that's not quite right. Instead of just being "louder," the brains of people with high neuroticism are like radios that are out of tune. They struggle to tell the difference between a song that is a warning signal and a song that is safe background music.

The Experiment: A "Scary" Video Game

The researchers put 249 people into an MRI machine and played a simple "video game" of fear.

The Setup: Participants saw colored shapes on a screen.
The Threat (CS+): One color was sometimes followed by a mild, uncomfortable electric shock.
The Safety (CS-): Another color was never followed by a shock.
The Goal: The brain learns to fear the "threat" color and relax when it sees the "safety" color.

Before the game, the researchers asked everyone to fill out a personality survey. They didn't just look at one trait; they looked at the whole picture of a person's personality. Using a smart computer algorithm (like a sophisticated sorting machine), they grouped the participants into two main teams based on their personality profiles:

The Low-Neuroticism Team: People who generally feel calm and don't worry much.
The High-Neuroticism Team: People who are naturally more sensitive to negative feelings and perceive the world as more threatening.

What They Found: The "Tuning" Difference

When the researchers looked at the brain scans, they found a surprising difference between the two teams.

1. The "Volume" Myth is Wrong
The High-Neuroticism team didn't have brains that were just "exploding" with fear. In fact, their brains were surprisingly quiet when they saw the threat.

2. The "Differentiation" Problem
The real issue was clarity.

The Low-Neuroticism Team: Their brains were excellent at drawing a bright, clear line between "Danger!" and "Safe." When they saw the threat color, specific parts of their brain lit up strongly. When they saw the safe color, those parts stayed calm. They knew exactly which was which.
The High-Neuroticism Team: Their brains were muddy. When they saw the threat, the "danger" lights didn't turn on as brightly as they should have. Because the "danger" signal was weak, it looked very similar to the "safety" signal. They couldn't tell the difference as clearly.

The Analogy: Imagine you are trying to spot a red car in a parking lot.

Low-Neuroticism: You see a bright, glowing red car next to a dull gray car. Easy to tell apart.
High-Neuroticism: The red car is dim, and the gray car is slightly reddish. You stare at them and think, "Is that one dangerous or not?" Your brain is confused because the signals are too similar.

Where in the Brain?

The study found this "muddy signal" happened in three specific control centers of the brain:

The vmPFC and sgACC: These are like the brain's "judges" that decide how scary something is.
The PCC: This is like the brain's "context librarian" that helps you remember if a situation is usually safe or dangerous.

In the High-Neuroticism group, these judges and librarians didn't get a strong enough signal to make a clear decision.

The Symptoms: A Hidden Connection

The researchers also checked how these brain patterns related to real-life anxiety.

State Anxiety (How you feel right now): Both groups felt about the same level of nervousness during the test.
Trait Anxiety (Your personality baseline): The High-Neuroticism group had much higher scores on long-term anxiety and sensitivity to fear.

The Big Surprise:
When the researchers tried to predict who would have high anxiety symptoms based only on their brain scans, it only worked for the High-Neuroticism team.

For the Low-Neuroticism team, their brain activity during the test didn't predict their anxiety levels at all.
For the High-Neuroticism team, the "muddier" their brain signals were, the higher their anxiety symptoms were.

It's as if the brain scan is a "diagnostic tool" that only works if you are already in the "High-Neuroticism" category. For the calm group, the brain scan doesn't tell you much about their anxiety because their brain and anxiety aren't linked in the same way.

What They Did NOT Find

Behavior: Even though the High-Neuroticism group's brains were confused, they didn't act differently. They didn't report feeling more scared than the other group when asked. Their brains were struggling internally, but their outward behavior looked the same.
Future Cures: The paper does not claim that fixing these brain areas will cure anxiety. It only identifies where the difference happens.

The Takeaway

This study tells us that personality isn't just a "volume knob" for fear. Instead, personality acts like a filter. For people with high neuroticism, this filter makes it hard for the brain to clearly separate "danger" from "safety." This confusion in the brain's wiring is what links their personality to their anxiety symptoms, but only for them, not for everyone.

Technical Summary: Personality Heterogeneity Shapes the Neural Organization of Conditioned Threat Learning

Problem Statement

The relationship between individual differences in personality structure and the neural mechanisms of fear conditioning remains a critical, unresolved question in precision psychiatry. While high emotional sensitivity (particularly Neuroticism) is a robust predictor of psychiatric vulnerability, existing neuroimaging literature has yielded contradictory findings regarding how personality shapes threat processing. Traditional approaches often operationalize affective sensitivity as an "exaggerated" neural response to danger. However, adaptive fear learning relies not merely on threat detection, but on the accurate differentiation between threat and safety signals.

Furthermore, it is unclear whether the relationship between neural threat representations and anxiety-related symptoms is uniform across individuals or dependent on latent personality heterogeneity. Current methods frequently rely on single-trait correlations or arbitrary score-splitting (e.g., median splits), which fail to capture the multidimensional nature of personality and may obscure distinct neurobiological subtypes. This study addresses whether personality-defined heterogeneity shapes the neural organization of conditioned threat learning and whether the coupling between neural threat representations and anxiety symptoms is personality-dependent.

Methodology

Participants and Design

The study utilized a multi-site cohort of 249 participants (ages 18–60) recruited from three independent acquisition sites (NYU Grossman School of Medicine and two cohorts at UTHealth Houston). Participants underwent a Pavlovian fear conditioning paradigm during functional MRI (fMRI) acquisition. The paradigm involved pairing visual conditioned stimuli (CS+) with aversive electric shocks while maintaining an unpaired safety stimulus (CS−). Only data from the fear conditioning phase (Day 1) were analyzed.

Personality Profiling and Stratification

To address the limitations of single-trait analysis, the authors employed a data-driven, unsupervised approach:

Instrument: The NEO Five-Factor Inventory-3 (NEO-FFI-3), covering Neuroticism, Extraversion, Openness, Agreeableness, and Conscientiousness.
Clustering: A Gaussian Mixture Model (GMM) was applied to the standardized five-factor profiles. Model selection via Bayesian Information Criterion (BIC) identified a two-cluster solution as optimal.
Subgroups: This process stratified participants into two latent subgroups: a Low-Neuroticism group ( $n=154$ ) and a High-Neuroticism group ( $n=95$ ). The separation was primarily driven by divergent Neuroticism scores.
Validation: Robustness was confirmed through subsampling stability, random-seed consistency (Adjusted Rand Index, Normalized Mutual Information), and split-half reproducibility analyses.

Neuroimaging and Analysis

Acquisition: 3T MRI scanners (Siemens Prisma/MAGNETOM) with varying protocols harmonized using CovBat to mitigate site effects.
Neural Decoding: A previously validated distributed neural threat decoder (based on a 24-node threat-circuit atlas) was used to estimate "fear probability" scores for CS+ and CS− cues across conditioning blocks.
Univariate Activation: General Linear Models (GLM) compared early conditioning contrasts (CS+ > CS−) between subgroups, focusing on the ventromedial prefrontal cortex (vmPFC), subgenual anterior cingulate cortex (sgACC), and posterior cingulate cortex (PCC).
Dynamic Functional Connectivity (dFC): A jackknife correlation procedure quantified trial-wise connectivity fluctuations. Analyses focused on threat–safety connectivity differentiation ( $\Delta FC = FC_{CS+} - FC_{CS-}$ ) within the 24-node threat-circuit atlas and a 442-region whole-brain parcellation.
Symptom and Behavioral Measures: Anxiety sensitivity (ASI-3, ASI-Total), trait anxiety (STAI-Trait), and state anxiety (STAI-State) were assessed. Subjective fear ratings were collected pre- and post-conditioning.
Predictive Modeling: Cross-validated linear regression models tested whether fear-learning-related neural features predicted anxiety symptom severity within each subgroup.

Key Results

1. Distinct Neural Organization in High-Neuroticism Individuals

Contrary to the hypothesis that high neuroticism reflects exaggerated threat reactivity, the High-Neuroticism subgroup exhibited a reduced capacity to differentiate conditioned threat from safety signals.

Neural Decoding: During early acquisition, the Low-Neuroticism group showed significantly higher fear probability expression to CS+ cues compared to the High-Neuroticism group. The High-Neuroticism group showed attenuated neural expression of conditioned threat, while responses to safety cues remained comparable across groups.
Regional Activation: The Low-Neuroticism group demonstrated greater threat-related activation in the vmPFC, sgACC, and PCC during the CS+ > CS− contrast. The High-Neuroticism group showed significantly reduced activation in these regions.
Dynamic Functional Connectivity: The Low-Neuroticism group exhibited stronger threat–safety connectivity differentiation. Significant subgroup differences were observed in connections including PCC–centromedial amygdala (cmAMY), dorsal anterior cingulate cortex (dACC)–orbitofrontal cortex (OFC), and precentral–middle frontal gyrus (MFG). Whole-brain analyses revealed differences in interactions involving the Dorsal Attention Network (DAN), Cerebellum Network (CEM), Frontoparietal Control Network (CON), and Sensorimotor Network (SMN).

2. Symptom Profiles and Behavioral Dissociation

Trait vs. State: The High-Neuroticism subgroup showed significantly elevated scores on trait-related measures (ASI-3, ASI-Total, STAI-Trait) compared to the Low-Neuroticism group. However, state anxiety (STAI-State) did not differ significantly between groups.
Behavioral Performance: Despite pronounced neural differences, there were no significant differences in subjective fear ratings (behavioral fear learning) between the two subgroups. Both groups successfully acquired fear associations behaviorally.

3. Personality-Dependent Brain–Symptom Coupling

Predictive modeling revealed a critical dissociation:

High-Neuroticism Subgroup: Fear-learning-related neural activation patterns significantly predicted anxiety sensitivity (ASI-3 and ASI-Total) scores. Stronger threat–safety activation contrasts were associated with greater symptom severity.
Low-Neuroticism Subgroup: No significant predictive relationship was found between neural features and symptom severity.
Full Sample: No significant prediction was observed when analyzing the entire cohort without stratification, suggesting that personality structure masks or modulates the brain–symptom relationship.

Significance and Claims

The paper argues that personality-defined heterogeneity is not merely a source of noise but a fundamental organizer of the neurobiology of threat learning. The authors claim:

Mechanism of Vulnerability: High Neuroticism is characterized not by an "exaggerated volume" of threat response, but by a reduced capacity to differentiate threat from safety at the neural level. This is anchored in reduced engagement of the vmPFC, sgACC, and PCC, and weaker dynamic coordination within distributed threat-processing networks.
Trait vs. State: The neural differences observed are associated with stable, trait-like vulnerability (anxiety sensitivity) rather than transient state fluctuations or overt behavioral responses.
Precision Psychiatry Implications: The relationship between neural threat representations and anxiety symptoms is personality-dependent. Neural markers of fear learning predict anxiety vulnerability only within high-neuroticism individuals. Consequently, incorporating multidimensional personality profiling is essential for identifying biologically meaningful subgroups and developing stratified, personalized interventions.
Neural Targets: The study identifies the vmPFC, sgACC, and PCC, along with their network dynamics, as convergent regions associated with personality-dependent threat learning, suggesting them as potential targets for future neuromodulation studies (e.g., TMS, tDCS, neurofeedback).

The authors maintain modesty regarding causal claims, noting that the study was conducted in a non-clinical population and focused solely on the acquisition phase, leaving the generalizability to clinical anxiety disorders and the role of fear extinction for future investigation.