The anterior cingulate cortex (ACC) is a critical hub implicated in the processing of aversive sensory states, including both pain and itch. While previous studies have established that the ACC contains segregated neuronal populations that respond to these different stimuli, and that full-trace classification using machine learning can distinguish between them, the precise nature of the underlying single-neuron calcium transients remained largely unresolved. This study, represents a detailed reanalysis of denoised in vivo calcium traces from stimulus-responsive ACC neurons, aiming to move beyond simple classification success and identify the specific, interpretable features within the neuronal response that carry this label-discriminative information.

The research team quantified several sophisticated, interpretable calcium-response features. These included basic metrics like peak rate and peak interval, as well as more complex kinetic measures such as the rise and decay area under the curve (AUC), and the corresponding rise and decay slopes. These features map onto fundamental dimensions of calcium dynamics: event frequency, event timing, the accumulation of calcium during the rise to a peak, the accumulation during the decay, and the steepness of those transitions. By rigorously analyzing these specific, quantifiable aspects of the calcium signal, the researchers sought to determine if the separation between itch- and pain-associated neurons was rooted in these specific temporal dynamics rather than in broader, less interpretable signal characteristics.

The application of trace-based stochastic gradient descent (SGD) classification proved successful in separating the two groups of neurons based on these response profiles. This success strongly indicates that the individual response profiles of the ACC neurons contain meaningful, label-discriminative information. Crucially, the study performed a series of control experiments to ensure this separation was biologically meaningful and not merely an artifact of data processing. For instance, classification performance was only weakly affected by neuron-wise z-scoring, a normalization technique that removes each neuron’s specific mean and variance scale. This finding suggests that the separation is not solely explained by simple, global amplitude scaling, but is embedded in the temporal structure of the response itself.

The study also addressed the inherent structural constraints of the dataset, a common issue in neuroscience where multiple neurons are sampled from the same animal. To test for mouse-related artifacts, the researchers employed a mouse-constrained shuffled-label control. In this control, labels were shuffled only within mice that contained both itch- and pain-associated neurons, thereby preserving mouse identity while disrupting the natural association between the response profile and the condition label. This control significantly reduced classification performance, providing strong evidence that the within-mouse label-associated structure is a genuine contributor to the classifier’s ability to distinguish between itch and pain, rather than simply reflecting shared mouse-level characteristics. Furthermore, the analysis revealed limited cross-animal consistency when comparing paired mouse summaries and mouse-exclusive validation, suggesting that while local organization is present, the classification boundaries do not generalize consistently across individual mice.

To visualize the neuron-level organization of these complex response profiles, researchers utilized Uniform Manifold Approximation and Projection (UMAP) on the full traces. This descriptive visualization provided insight into how the neurons clustered based on their entire response pattern. While the global label silhouette coefficients were low, indicating weak overall separation in the two-dimensional projection, a more detailed local-organization test provided a critical counterpoint. By projecting the data onto 50 principal components and comparing the excess same-label neighbors within the same mouse against label permutations, the researchers detected local label-associated organization. This finding is highly significant because it confirms that within the confines of a single mouse, the response profiles exhibit a meaningful, localized structure related to the stimulus condition, even if the patterns do not transfer reliably to other animals.

The detailed feature comparisons, conducted using two-sided Wilcoxon rank-sum tests with Benjamini–Hochberg correction, revealed specific directional differences between the two groups. Itch-associated neurons were characterized by a higher event frequency, meaning they exhibited more frequent calcium transients. In contrast, pain-associated neurons displayed longer peak intervals between events. Furthermore, pain-associated neurons tended to exhibit larger rise and decay AUC values and less negative raw decay slopes. These differences in peak rate, peak interval, and transient morphology provide interpretable correlates to the full-trace classification success reported in prior work, offering a mechanistic glimpse into how the ACC might be encoding these distinct aversive sensations at the cellular level.

The study’s methodological rigor is noteworthy, as it systematically evaluated the robustness of these findings across various data representations. The successful classification using raw denoised traces, as well as z-scored traces, strongly supports the conclusion that the separation is based on temporal features rather than simple amplitude scaling. The z-scoring process, which removes each neuron’s temporal mean and scales the trace by its temporal standard deviation, effectively neutralizes the influence of overall fluorescence scale and baseline offset. The persistence of classification performance after this transformation is a powerful indicator that the discriminatory information is encoded in the shape and timing of the calcium transient itself, which is precisely what the calculated AUC and slope features capture.

In summary, this research provides a nuanced understanding of how the ACC encodes the distinct experiences of itch and pain. While the findings establish small, specific differences in calcium-response profiles—such as higher event frequency in itch versus longer inter-event intervals in pain—the researchers also carefully delineated the boundaries of their conclusions. The evidence supports the existence of local label-associated organization within individual mice, suggesting that the neural population geometry carries meaningful information. However, the limited cross-animal consistency and the lack of consistent median shifts across paired mouse summaries define the scope of this work, emphasizing that the observed structure is robust within the context of a single animal’s response, while caution is warranted when extrapolating these patterns to the broader animal population.

Subject of Research: The study reveals that distinct calcium response features, such as peak rate and transient morphology, differentiate itch- and pain-associated neurons in the anterior cingulate cortex.

Article Title: Distinct calcium responses distinguish itch- and pain-associated neurons in the anterior cingulate cortex

Article References: Lee, C., & Ko, H.-G. (2026). Distinct calcium responses distinguish itch- and pain-associated neurons in the anterior cingulate cortex. Frontiers in Behavioral Neuroscience, 20, 1948042. https://doi.org/10.3389/fnbeh.2026.1948042

Image Credits: AI Generated

DOI: 10.3389/fnbeh.2026.1948042

Keywords: anterior cingulate cortex, calcium imaging, itch, pain, neuroscience, neural classification, stochastic gradient descent, calcium transients, Distinct, calcium, responses, distinguish