Human Brain Organizes Emotion Concepts Using Grid-Like Prefrontal Neural Coordinates

Human emotional experiences encompass a diverse spectrum of everyday events, ranging from calm conversations to tense encounters, yet behavioral research has long established that people systematically organize these complex states within a continuous two-dimensional affective space defined by valence and arousal. Although this circumplex structure has guided affective science for decades, the specific neural architecture supporting abstract emotion concepts has remained unresolved. In a newly published neuroimaging investigation in PLOS Biology, researchers Yumeng Ma and Philip A. Kragel from the Department of Psychology at Emory University report that the human ventromedial prefrontal cortex organizes emotion knowledge using coordinate principles characteristic of grid-like coding. By recording functional magnetic resonance imaging data while participants tracked dynamic transitions across affective space conveyed through distinct stimulus modalities, the investigators observed robust hexadirectional modulation in the ventromedial prefrontal cortex that operated across sensory formats and reflected an abstract metric organization independent of overt behavioral choices.

To navigate dynamic real-world environments adaptively, the human brain organizes accumulated knowledge into structured geometric frameworks that support forward-looking prediction, relational inference, and flexible decision-making across varied behavioral domains. Converging neuroscientific findings have previously established that diverse forms of knowledge, including representations of physical space, visual objects, social networks, odors, economic choice options, and sequential action plans, are encoded in map-like coordinate systems within medial temporal and prefrontal networks. Embedding discrete experiences within continuous coordinate dimensions allows geometric relationships among diverse environmental states to be maintained, thereby facilitating mental simulation beyond immediately available sensory inputs. Information regarding emotional events is exceptionally vital for organismic survival and social welfare, and psychological research across decades indicates that individuals consistently structure emotion concepts according to the affective circumplex, reliably situating category instances along orthogonal axes of pleasantness and physiological activation across vocalizations, facial cues, language, and subjective self-reports.

Despite the empirical ubiquity of circumplex geometry in behavioral evaluations, contemporary brain theories have struggled to define the precise representational format through which affect is organized. Psychological constructionist frameworks have proposed that default network regions integrate long-term memory with bodily states and momentary goals to assemble situated conceptual models, identifying the ventromedial prefrontal cortex as a central nexus for prospective valuation and emotional self-report. However, these formulations have not clarified whether affective space is implemented as a metric coordinate system or why dimensional geometry arises with such consistency across individuals and tasks. Drawing inspiration from spatial cognition, where entorhinal and prefrontal grid cells support navigation by firing in periodic six-fold rotational symmetry across continuous environments, the Emory University research team hypothesized that ventromedial prefrontal cortex and connected structures might encode trajectories across emotional space using analogous hexadirectional coordinate codes.

The researchers conducted a functional magnetic resonance imaging experiment involving twenty healthy, English-fluent adult volunteers recruited from the Atlanta, Georgia area, in accordance with protocols approved by the Institutional Research Board at Emory University under Protocol Number 2025P010826.

To rigorously test whether putative grid-like coding represents abstract emotional meaning rather than idiosyncratic sensory properties, the experimental paradigm incorporated two perceptually distinct stimulus modalities featuring dynamic affective transitions: photographic face morphs and sequential text phrases. Facial stimuli were constructed using seventy-eight photographs of human models selected from the Karolinska Directed Emotional Faces database, depicting twenty-six distinct individual identities across seven basic emotion categories comprising happiness, surprise, anger, fear, sadness, disgust, and neutral expression. Pairs of facial photographs depicting identical identities across differing emotional categories were rendered into continuous dynamic transitions using Abrosoft FantaMorph software, creating eighty-four unique face trajectories that each unfolded over five seconds. In parallel, linguistic stimuli were developed across eight everyday environmental settings: shopping, park, forest, school, food, animal, weather, and social contexts. An extensive pool of candidate phrases was generated using ChatGPT-4o, validated in an independent ten-participant pilot study, and refined into one hundred twenty-six final phrases, yielding one24 distinct sequential phrase trajectories where a starting phrase appeared for 2.5 seconds followed by an ending phrase for 2.5 seconds.

Functional neuroimaging was structured into four separate scanning runs, divided evenly between face morph and phrase sequence modalities, totaling 36.5 minutes of active task acquisition within an approximately one-hour session on a 3-Tesla Siemens MAGNETOM Prisma scanner. Within each run, trials were grouped into block-wise pseudo-randomized sequences of stable identities or contextual settings to reduce low-level perceptual volatility, with blocks averaging 4.73 trials across the task. On each individual scanning trial, participants observed a five-second affective transition, followed by a jittered fixation interval lasting between one and five seconds, and then had three seconds to register whether the stimulus had increased, decreased, or remained unchanged along one specifically queried dimension. The queried dimension, whether valence or arousal, was divided evenly across trials and disclosed only upon the onset of the response screen, compelling participants to continually monitor trajectories along both affective axes simultaneously without knowing which evaluation would be requested.

Following scanning, participants completed comprehensive self-paced behavioral evaluations outside the scanner, providing continuous ratings of valence and arousal alongside seven distinct emotion categories for every individual face and phrase stimulus. Multidimensional scaling performed on cosine distance matrices of group-average category ratings revealed an optimal two-dimensional configuration that strongly converged with direct valence and arousal coordinates under Procrustes alignment, yielding a disparity of 0.20 and a permutation p-value below 0.0001. During scanning, participant choice accuracy aligned robustly with post-scan coordinates: logistic regression confirmed that as the absolute distance along the probed dimension increased, behavioral agreement with consensus ratings rose substantially, with a regression slope of 0.9137, a standard error of 0.1212, and a z-score of 7.540, whereas distance along the unprobed dimension demonstrated no significant relationship, confirming that participants accurately tracked affective transitions across both axes.

Functional imaging utilized an advanced multiband multi-echo echo-planar imaging sequence, capturing fifty-six interleaved axial slices at an isotropic voxel size of 2.5 millimeters with a repetition time of 1,250 milliseconds and three distinct echo times of 15, 37.16, and 59.3 milliseconds, while structural T1-weighted MPRAGE data were acquired at 1.0 millimeter isotropic resolution. Preprocessing was executed using fMRIPrep version 25.0.0 based on Nipype 1.9.2, incorporating intensity non-uniformity correction with N4BiasFieldCorrection in ANTs 2.5.4, brain tissue segmentation through FSL fast, nonlinear volumetric normalization to the MNI152NLin2009cAsym standard space template, B0 distortion correction with FSL topup, and rigorous head-motion correction via FSL mcflirt.

To evaluate whether multivoxel activity patterns exhibited hexadirectional modulation, the authors conducted cross-validated multivariate pattern analyses within an a priori ventromedial prefrontal cortex region of interest centered at Montreal Neurological Institute coordinates [6, 44, −10], defined by a five-millimeter spherical radius containing sixty-nine voxels. For each participant, functional runs were divided into independent training and testing sets: single-trial beta weights from the training set were fitted against sine and cosine functions of six-fold trajectory angles to estimate voxel-wise coefficients and compute participant-specific grid orientations. In held-out test data, pattern correlations between trial pairs whose trajectory vectors were aligned with estimated grid axes were compared against pairs offset by thirty degrees. Response patterns in the ventromedial prefrontal cortex exhibited significantly greater multivoxel pattern similarity for aligned-aligned pairs (mean correlation = 0.0801, standard deviation = 0.0262) than for aligned-misaligned pairs (mean correlation = 0.0729, standard deviation = 0.0262), generating a significant difference of Δr = 0.0072, with a 95% confidence interval of [0.0031, 0.0115], an effect size of d = 0.7820, and a p-value of 0.0014.

The specificity and analytical robustness of this prefrontal hexadirectional signal were confirmed through multiple control procedures and periodicity contrasts. When the researchers tested alternative rotational symmetries, including three-fold, four-fold, five-fold, seven-fold, and eight-fold periodicities, none produced comparable pattern similarity differences, and direct comparisons showed that six-fold modulation significantly exceeded these control periodicities with a differential effect of Δr = 0.0088, a 95% confidence interval of [0.0046, 0.0133], d = 0.9223, and p = 0.0005. Furthermore, to verify that pattern similarity was not an artifact of trial count asymmetries, a subsampling routine matched pair counts across 10,000 iterations, confirming an identical modulation of Δr = 0.0072. Multiple linear regressions controlling for potential confounding factors—including starting and ending valence, arousal levels, trajectory distance, shared emotion categories, stimulus identity, environmental setting, and motor choice—demonstrated that hexadirectional modulation remained highly robust (Δr = 0.0070, 95% confidence interval [0.0033, 0.0109], d = 0.8416, p = 0.0008). In contrast, anatomical masks of the hippocampus and entorhinal cortex derived from the Julich-Brain Cytoarchitectonic Atlas failed to show comparable grid-like modulation.

Crucially, the coordinate code in the ventromedial prefrontal cortex generalized across differing sensory inputs, confirming that the representation operates at an abstract conceptual level rather than reflecting low-level visual or linguistic processing. When the investigators estimated grid orientation exclusively using phrase sequences and tested pattern modulation on face morph trials, significant hexadirectional tuning was recovered (Δr = 0.0117, 95% confidence interval [0.0013, 0.0224], d = 0.5061, p = 0.0234), with the reciprocal cross-modal analysis from faces to phrases demonstrating a matching significant effect (Δr = 0.0072, 95% confidence interval [0.0013, 0.0135], d = 0.5495, p = 0.0196). When affective space was alternatively derived using multidimensional scaling of category ratings, cross-modal generalization remained significant in the phrase-to-face direction (Δr = 0.0123, d = 0.6242, p = 0.0096), although the reverse face-to-phrase mapping in that specific configuration did not reach significance (Δr = 0.0002, p = 0.4693), demonstrating general cross-modal consistency while highlighting subtle geometric dependencies across analytical space definitions.

The study also revealed a clear functional dissociation within the ventromedial prefrontal cortex between grid-like coordinate geometry and evaluative decision-making signals. Comparisons of hexadirectional modulation across trials with choices consistent versus inconsistent with normative consensus ratings showed no meaningful difference (Δr differential = 0.0049, p = 0.1220, Bayes factor BF10 = 0.4477), indicating that coordinate grid strength is independent of immediate behavioral responses. In contrast, multivariate partial least squares encoding models demonstrated that two-dimensional Euclidean distance through affective space significantly predicted prefrontal activity (r = 0.0127, p = 0.0021, accounting for 38.09% of the noise ceiling), outperforming unidimensional valence or arousal models. Trial-wise expression of this distance code directly predicted behavioral consistency during scanning (beta = 0.0788, z = 2.388, p = 0.0169), demonstrating that the ventromedial prefrontal cortex simultaneously houses an invariant metric map of emotion space alongside evaluative distance signals that actively support choice behavior.

The authors explicitly delineate several methodological and conceptual boundaries of their findings, noting that functional neuroimaging cannot establish the causal necessity of the ventromedial prefrontal cortex in maintaining affective geometry. Additionally, although two-dimensional Euclidean distance accounted for decision-related signals, the experimental design was not structured to dissociate alternative metric frameworks, such as city-block distance, nor did in-scanner tasks require simultaneous mathematical integration of both dimensions. Nevertheless, by demonstrating that abstract emotion concepts are mapped into hexadirectional coordinates, the investigation bridges affective psychological theory and cognitive spatial neuroscience, revealing that the human brain structures its internal world of feeling using universal spatial principles.

Subject of Research: Grid-like neural representations of affective space in human prefrontal cortex

Article Title: Emotion knowledge is structured in a grid-like coordinate system in the human ventromedial prefrontal cortex

Article References: Ma, Y., & Kragel, P. A. (2026). Emotion knowledge is structured in a grid-like coordinate system in the human ventromedial prefrontal cortex. PLOS Biology, 24(11), e3003795. https://doi.org/10.1371/journal.pbio.3003795

Image Credits: AI Generated

DOI: 10.1371/journal.pbio.3003795

Keywords: Neuroscience, Ventromedial Prefrontal Cortex, Emotion Knowledge, fMRI, Affective Circumplex, Grid Cells, Cognitive Maps, Multivoxel Pattern Analysis, Emotion, knowledge, structured, grid-like