Hypothalamic CRH neurons represent physiological memory of positive and negative experience

Behavioral responses to salient stimuli are accompanied by changes in internal physiological state, yet how the brain encodes and recalls such physiological memories is poorly understood. The study investigates whether the same or distinct neural architectures underlie external (behavioral) and internal (physiological) memories, focusing on corticotropin-releasing hormone (CRH) neurons in the paraventricular nucleus (PVN) of the hypothalamus that control the HPA axis and corticosterone (CORT) release. CRH^PVN neurons bidirectionally respond to aversive (increase) and rewarding (decrease) stimuli, detect predatory threats, and are required for anxiety states; repeated optogenetic manipulation can form contextual memory, suggesting a potential role in storing salient experiences. While associative threat memories often recruit engrams in cortex, amygdala, and hippocampus, species lacking these structures can still form associative memories, and hypothalamic oxytocin neurons have exhibited engram-like contextual fear memory. This work asks how CRH^PVN neurons encode contextual memories of negative and positive valence and how such encoding relates to behavioral expressions of memory.

Animal model: CRH-IRES-Cre mice crossed with Ai14 (tdTomato) or Ai148 (GCaMP6f) lines (Jackson Laboratories) were used. Mice (6–8 weeks at surgery) were housed on a 12:12 light:dark cycle and randomly assigned to experimental conditions. All procedures were approved by the University of Calgary Animal Care and Use Committee. Stereotaxic procedures: For fiber photometry, Cre-dependent AAV9-CAG.Flex.GCaMP6s was injected into PVN (AP −0.7 mm, L −0.3 mm, DV −4.5 mm), followed by implantation of a 400 µm optical fiber dorsal to PVN. For miniscope imaging, a GRIN lens was implanted dorsal to PVN (AP −0.7 mm, L −0.2 mm, DV −4.2 mm). Animals recovered for 2–4 weeks prior to experiments. Histology confirmed expression and implant locations. Recording approaches: CRH^PVN calcium activity was recorded in freely moving mice using fiber photometry (dual-excitation 465/470 nm and 405 nm isosbestic channels; TDT or Doric systems; ~30 µW at patch cord tip) and head-mounted miniscopes for single-cell resolution imaging. Behavioral paradigms: - Neutral context (novel environment) exposure to assess anticipatory activity without overt threat or appetitive stimuli. - Aversive conditioning: 10 foot shocks (FS) delivered in a novel context (Pre), with re-exposure to the same context 24 h later (Post) to test contextual recall; repeated re-exposures assessed extinction and persistence. - Appetitive conditioning: After a Pre exposure to a novel context, a small hazelnut spread (Nutella) sample was presented; mice were re-exposed to the context 24 h later and across additional days without the reward to assess recall of positive valence. Endocrine measurements: Plasma corticosterone (CORT) was measured upon re-exposure sessions to compare endocrine responses with neural and behavioral measures. Data analysis: Photometry signals were motion/bleach corrected using polynomial detrending and 405 nm regression; ΔF computed and z-scored. Miniscope data were processed with MIN1PIPE and custom alignment (stochastic descent of merged two-channel images) to track individual neurons across days. Principal component analysis (PCA) on time-aligned data matrices characterized low-dimensional state-space shifts. Affinity propagation clustering identified subpopulations (Weak, Intermediate, Strong) based on Pre activity. Statistical tests included paired t-tests, linear regression, ANOVA/mixed-effects models as appropriate (see figures and Supplementary Table 1 for details). Computational modeling: Networks of leaky integrate-and-fire neurons with a calcium-like indicator were simulated. Each neuron received context, foot-shock, and Nutella-like inputs. A two-factor learning rule updated context weights for aversive memory: recruitment toward a ceiling (Ω − Σω_i) multiplied by a plasticity gate triggered when a slow calcium-like variable exceeded threshold during FS. A simpler one-factor delayed rule modeled appetitive memory, scaling weights based on Nutella input without a shock-dependent gate. Simulations reproduced empirical increases (aversive) or decreases (appetitive) in anticipatory responses and PCA state-space shifts. Key parameters are listed in Table 1 of the paper.

- Neutral (novel) context evokes sustained anticipatory activation of CRH^PVN neurons: - Population increase relative to home cage (photometry), persistent during exposure (p = 0.0001, n = 10 mice, t = 6.340, paired two-tailed t-test). Miniscope single-cell activity also increased (p = 1.057×10^−11, n = 96 cells, N = 4 mice, t = 10.56). - PCA revealed an expansion and shift in activity state space versus home cage. - Aversive contextual memory (foot shock, FS): - Re-exposure (Post, 24 h) produced a stronger increase in CRH^PVN activity than Pre (photometry: p = 0.0079, n = 9 mice, t = 3.514; miniscope: p = 6.634×10^−16, n = 115 cells, N = 5 mice, t = 9.409). - CORT was higher than after neutral re-exposure; immediate post-FS activity increases were not apparent, indicating learning-related changes. - Control neutral-context re-exposure (no FS) showed no change (population: p = 0.7772, n = 9 mice, t = 0.2927; single-cell: p = 0.5760, n = 96 cells, N = 4 mice, t = 0.5612). - Pre vs Post single-cell activity showed weak correlation (r = 0.1865, p = 0.0458), indicating a subpopulation drives the increase. - Data-driven clustering based on Pre activity identified Weak, Intermediate, and Strong cells. Weak cells showed a dramatic Post increase (p = 2.952×10^−6, n = 37, t = 8.602), while Strong cells did not change (p = 0.9748, n = 17, t = 0.0320); Weak and Strong responses converged on Post. - During FS, Weak cells had stronger responses than Strong cells (p = 0.0156, t = 2.501). FS response magnitude predicted the Post increase for Weak cells (r = 0.5922, p = 0.0001) but not for Strong cells (r = 0.0924, p = 0.7244). - Divergence of internal (CRH^PVN/CORT) and external (behavior) responses across repeated aversive recalls: - Freezing during 3-min re-exposures declined across Post_1–Post_3 (extinction: n = 15 mice, p = 5.638×10^−10, F(2,28) = 50.09). - CORT remained stable (n = 8 mice, p = 0.3655, F(2,14) = 1.082). - CRH^PVN population response remained invariant (photometry: Post_1–Post_3, p = 0.2722, F(2,26) = 1.369); single-cell means unchanged (miniscope: p = 0.1003, n = 115 cells, N = 5 mice, F(2,228) = 2.323). - Single-cell activity was stable and predictive across days (r = 0.7329, p < 1×10^−7). PCA showed a preserved shift in state space across recalls. - Appetitive contextual memory (hazelnut spread): - Access to hazelnut spread acutely decreased CRH^PVN activity; 24 h later, re-exposure to the associated context produced significantly lower activity than Pre, with high correlation between Pre and Post. - Clustering revealed Strong cells exhibited a pronounced drop in activity upon recall, whereas Weak cells responded similarly to Pre. - The blunted response persisted over at least two additional recall days; single-cell responses were stable and predictable; PCA confirmed stability. - Computational modeling: - A two-factor local learning rule (recruitment toward a ceiling plus an FS-gated plasticity term) reproduced the aversive-memory increase, convergence of Weak and Strong pools, and PCA shifts. - A simpler one-factor rule captured the appetitive-memory decrease without a shock-dependent gate. Overall, CRH^PVN neurons encode contextual memories of negative and positive valence with distinct learning rules, and aversive physiological memory outlasts behavioral extinction.

The study shows that hypothalamic CRH^PVN neurons form associative contextual memories of both negative and positive experiences via local learning mechanisms distinct from classical engram recruitment in cortical and limbic structures. For negative valence, a two-factor rule preferentially recruits previously weakly responsive neurons based on their low Pre activity and strong shock-evoked responses, increasing the pool of strongly responsive cells during recall. For positive valence, a simpler one-factor rule scales down activity across the population during contextual recall, with Strong cells showing the largest decreases. These local hypothalamic mechanisms produce enduring internal state representations that persist over multiple days. Importantly, internal measures (CRH^PVN activity and corticosterone) remain stable across repeated aversive recalls, while freezing behavior extinguishes, revealing a dissociation between physiological state encoding and overt behavioral expression. This dissociation cautions against relying solely on behavior as a proxy for internal stress states and underscores the hypothalamus as a memory substrate for physiological states.

This work identifies CRH^PVN neurons as a hypothalamic substrate for enduring physiological memories of both aversive and appetitive experiences. Negative experiences induce a two-factor, locally implemented learning rule that recruits previously weak cells to strengthen anticipatory responses, whereas positive experiences engage a simpler one-factor rule that scales down activity upon contextual recall. These physiological memories persist despite extinction of overt defensive behavior, highlighting a dissociation between internal state encoding and external behavior. The findings advance understanding of how internal states are remembered and represented in the brain and emphasize the importance of measuring physiological signals alongside behavior.