Dentate gyrus and CA3 activity mediates light-tone second-order conditioning expression in mice.

Dentate gyrus and CA3 activity mediates light-tone second-order conditioning expression in mice - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice iScience . 2026 Mar 30;29(4):115522. doi: 10.1016/j.isci.2026.115522 Search in PMC Search in PubMed View in NLM Catalog Add to search Dentate gyrus and CA3 activity mediates light-tone second-order conditioning expression in mice Marc Canela Marc Canela 1 Cell-Type Mechanisms in Normal and Pathological Behavior Research Group, Neuroscience Research Program, Hospital del Mar Research Institute, Barcelona, Spain Find articles by Marc Canela 1 , Jordi Bach Adell Jordi Bach Adell 1 Cell-Type Mechanisms in Normal and Pathological Behavior Research Group, Neuroscience Research Program, Hospital del Mar Research Institute, Barcelona, Spain Find articles by Jordi Bach Adell 1 , David Roura Coll David Roura Coll 1 Cell-Type Mechanisms in Normal and Pathological Behavior Research Group, Neuroscience Research Program, Hospital del Mar Research Institute, Barcelona, Spain Find articles by David Roura Coll 1 , Jose Antonio González-Parra Jose Antonio González-Parra 1 Cell-Type Mechanisms in Normal and Pathological Behavior Research Group, Neuroscience Research Program, Hospital del Mar Research Institute, Barcelona, Spain Find articles by Jose Antonio González-Parra 1 , Julia S Pinho Julia S Pinho 1 Cell-Type Mechanisms in Normal and Pathological Behavior Research Group, Neuroscience Research Program, Hospital del Mar Research Institute, Barcelona, Spain Find articles by Julia S Pinho 1, 2 , Arnau Busquets-García Arnau Busquets-García 1 Cell-Type Mechanisms in Normal and Pathological Behavior Research Group, Neuroscience Research Program, Hospital del Mar Research Institute, Barcelona, Spain Find articles by Arnau Busquets-García 1, 3, ∗ Author information Article notes Copyright and License information 1 Cell-Type Mechanisms in Normal and Pathological Behavior Research Group, Neuroscience Research Program, Hospital del Mar Research Institute, Barcelona, Spain ∗ Corresponding author abusquets@researchmar.net 2 Present address: Gulbenkian Institute for Molecular Medicine, Oeiras 2780-156, Portugal 3 Lead contact Received 2025 Sep 1; Revised 2025 Oct 16; Accepted 2026 Feb 11; Collection date 2026 Apr 17. © 2026 The Author(s) This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). PMC Copyright notice PMCID: PMC13091456  PMID: 42006365 Summary Second-order conditioning (SOC) enables animals to form predictions about their environment without direct reinforcement. Although the neural basis of first-order conditioning (FOC) is well characterized, the circuits supporting SOC recall remain unclear. Here, we investigated the brain regions and cell types involved in SOC retrieval in mice and addressed the challenge of quantifying brain-wide neural activity. Using a light-tone SOC paradigm in TRAP2:Ai14 mice, we tagged neurons active during SOC recall by expressing tdTomato fluorescent protein. We quantified the fluorescent cells in brain slices with CellRake, a self-developed Python package. We applied generalized linear models that revealed that activity in the dentate gyrus (DG) and cornu ammonis 3 (CA3) subfields of the dorsal hippocampus was associated with SOC-related behavioral responses. Chemogenetic inhibition of CaMKII + neurons in these regions reduced SOC recall, supporting a causal contribution of DG/CA3 circuits. These findings identify the dorsal hippocampus as a key substrate for retrieving indirectly learned associations. Subject areas: Neuroscience, Behavioral neuroscience, Systems neuroscience Graphical abstract Open in a new tab Highlights • Mice increase immobility in first- and second-order conditioned responses • CellRake accurately quantifies fluorescent cells in brain sections • Immobility in second-order recall correlates with dorsal hippocampal activation • DG/CA3 CaMKII+ inhibition impairs second-order memory recall Neuroscience; Behavioral neuroscience; Systems neuroscience Introduction The brain’s ability to encode, store, and recall memories 1 remains one of the least understood parts of human biology. Indeed, the brain works differently depending on the type of memory, such as episodic, procedural, and semantic memories. 2 , 3 Among these, a large portion is unconscious, mainly operating through associative mechanisms. 4 , 5 In such cases, a specific cue, such as a smell, sound, or place, can bring up the memory of another stimulus or trigger a behavioral response. 6 These associative processes include various subtypes corresponding to different conditioning paradigms. 4 In first-order conditioning (FOC), an animal learns to associate a neutral stimulus, such as a light or tone, with a biologically significant unconditioned stimulus (US), such as food or an electric footshock. After repeated pairings, the neutral cue becomes a conditioned stimulus (CS) and alone can trigger a conditioned response similar to the US. 7 , 8 In contrast, higher-order conditioning forms an indirect link between a specific CS and the US, rather than a direct association. This type of learning includes two primary forms: sensory preconditioning (SPC) and second-order conditioning (SOC). In SPC, the process begins by pairing two neutral stimuli, such as a light and a tone. After the animal establishes an association between the stimuli, one of them is then associated with the US. As a result, the cue directly paired with the US becomes the first-order conditioned stimulus (CS 1 ), while the other becomes the second-order conditioned stimulus (CS 2 ). In SOC, the sequence reverses: A neutral stimulus first pairs with the US, then this CS 1 links to another neutral stimulus. Remarkably, in both SPC and SOC, CS 2 can trigger conditioned responses, even though CS 2 lacks a direct association with the US. 5 , 9 , 10 Therefore, this ability to increase the set of stimuli an animal can respond to makes higher-order learning crucial for survival. 10 , 11 However, when these processes are disrupted, they may contribute to disorders in which individuals maintain persistent fear responses through mechanisms such as SOC. 12 , 13 Thus, understanding how the brain encodes and retrieves such complex associations is vital for explaining both adaptive behavior and the neural basis of cognitive disorders. Although researchers have well characterized the neural basis of FOC, 7 the brain mechanisms supporting SPC and SOC remain less understood. In fact, there is more research on SPC than SOC, mainly because it addresses how the brain links two initially neutral stimuli, an associative process that FOC and SOC fail to tackle. As a result, numerous investigations have mapped some circuits involved in SPC, highlighting important regions such as the retros-plenial cortex (RSC), 14 the orbitofrontal cortex (OFC), 5 the perirhinal cortex (PRh), 15 the dorsal hippocampus, 16 the amygdala, especially the basolateral amygdala (BLA), 17 , 18 as well as the nu-cleus accumbens (NAc) and ventral tegmental area (VTA). 19 On the other hand, some research has begun to show causal involvement of specific brain regions involved in SOC, including the BLA, 18 , 20 the lateral hypothalamus (LH), 21 and the parabrachial nucleus (PBN). 22 However, few studies have investigated the brain substrates of SOC recall in mice and whether these circuits are shared with FOC expression. To investigate the brain circuits underlying associative learning and other behaviors, researchers often analyze fluorescent activity markers in brain slices. 23 , 24 , 25 , 26 However, manual anal-ysis of fluorescence images is slow, prone to error, and susceptible to observer bias and fatigue. These factors reduce reproducibility and make manual methods less reliable for large-scale studies. On the other hand, automated tools provide better accuracy, consistency, and scalability. 27 , 28 Yet they also come with significant limitations: Many require extensive customization to accommodate different experimental protocols and tissue types, demand substantial computational power, and rely heavily on labor-intensive image labeling for training. Additionally, pre-trained models often struggle to generalize across datasets, creating inefficiencies that slow research progress. To address these issues, this research presents CellRake, a new analytical tool. In this study, we combined behavioral, imaging, and computational approaches to investigate the neural substrates of recall in SOC. Using TRAP2 +/− ;Ai14 +/− (hereafter, called TRAP2:Ai14) mice, we genetically tagged neurons activated during CS 1 and CS 2 presentation in a validated light-tone SOC paradigm. 29 , 30 We quantified behavioral responses using DeepLabCut, 31 Keypoint- MoSeq, 32 and DeepOF. 33 Our multimodal approach revealed that glutamatergic neurons in the DG and cornu ammonis 3 (CA3) are engaged during CS 2 recall, and chemogenetic inhibition confirmed a causal contribution of dorsal hippocampal excitatory neurons to SOC expression. Results Second-order conditioning expression in TRAP2:Ai14 mice To investigate the neural substrates underlying the recall of first- and SOC, we used TRAP2 +/− ;Ai14 +/− mice (hereafter, called TRAP2:Ai14), which enable the permanent genetic labeling of neurons activated during a specific time window, via CreER expression driven by the Fos promoter. 29 This method enabled us to selectively tag neurons activated during the presentation of conditioned stimuli, specifically a tone (CS 2 ) and a light (CS 1 ). We employed a previously validated SOC protocol in our laboratory, 30 consisting of five stages on separate days: habituation, pairing a CS 1 (e.g., a light) with a foot shock, pairing a CS 2 (e.g., a tone) with CS 1 , and conducting two independent probe tests for both conditioned stimuli. To label neurons active during the display of CS 2 , we administered 4-hydroxytamoxifen (4-OHT) immediately after this probe session, which triggered tdTomato expression in active neurons exposed to CS 2 . One week later, after enough time for tdTomato to express, we performed the CS 1 recall test and collected the brains for cFos immunohistochemistry, allowing us to identify neurons active during the CS 1 presentation. This experimental design enabled us to map and quantify brain-wide activity patterns associated with each stage of SOC and relate them to behavioral responses ( Figure 1 A). Figure 1. Open in a new tab Second-order conditioning recall induces robust behavioral changes in TRAP2:Ai14 mice (A) Schematic representation of the light-tone behavioral SOC protocol using TRAP2:Ai14 an-imals. Mice underwent habituation (Stage 1), first-order conditioning with light and footshock (Stage 2), second-order conditioning with tone paired with light (CS 1 ) (stage 3), and independent recall sessions for tone (CS 2 ) (stage 4) and light (CS 1 ) (stage 5). We tagged active neurons during CS 2 recall using 4-hydroxytamoxifen (4-OHT), waited one week, and collected the brains after the CS 1 recall. (B and C) Outcomes from Hotelling’s T 2 analysis on the Δ metrics, shown as the negative logarithm of the p -value for each tested experimental condition: CS 1 males ( p < 0.0001, n = 12), CS 2 males ( p = 0.0210, n = 12), CS 1 females ( p < 0.0001, n = 11), and CS 2 females ( p = 0.0167, n = 11). (D and E) Transition dynamics measured through the Manhattan distance (MD) across conditions: CS 1 males ( p < 0.0001, n = 12), CS 2 males ( p = 0.000114, n = 12), CS 1 females ( p = 0.000162, n = 11), and CS 2 females ( p = 0.0424, n = 11). (F) Heatmap of average syllable Δ vectors shows upregulation (red) or downregulation (blue) in syllable abundance. (G) Differences of average Δ values in syllable 0 between sexes: CS 1 ( p = 0.616, males: n = 12, females: n = 11), CS 2 ( p = 0.999, males: n = 12, females: n = 11). (H) Differences of average Δ values in syllable 2 between sexes: CS 1 ( p = 0.999, males: n = 12, females: n = 11), CS 2 ( p = 0.997, males: n = 12, females: n = 11). (I) Correlation of the percentage of time doing huddling with the percentage of time in syllable 2 (r = 0.924, p < 0.0001, n = 146). (J and K) Huddle percentage computed through DeepOF across conditions: CS 2 males ( p = 0.00195, n = 12), CS 1 males ( p = 0.0005, n = 12), CS 2 females ( p = 0.0167, n = 11), and CS 1 females ( p < 0.0001, n = 11). (L and M) Behavioral response ratios (post/pre-immobility). The darker line indicates a ratio of 1, corresponding to the brink between increase and decrease. We established a threshold of 1.5 (thinner line) to classify animals as high- or low-performers for subsequent neural analyses. P-values are denoted as p < 0.05 (∗), p < 0.01 (∗∗), and p < 0.001 (∗∗∗). The data are presented as mean ± SD. For this experiment, we used adult male and female mice to replicate the behavioral findings reported in our recent work. 30 In brief, we recorded the animals’ behavioral responses and estimated their pose using DeepLabCut, 31 tracking 14 body parts per frame. Then, we analyzed the resulting trajectories with the unsupervised Keypoint-MoSeq algorithm, 32 which decomposes behavior into stereotyped motifs known as behavioral syllables. Finally, we analyzed the syllables using a dual approach, 30 focusing on the syllable abundance and examining the transition patterns between syllables. For the abundance, we measured the amount of time each animal spent in each syllable and calculated the change in time between the post-stimulus and pre-stimulus periods. These differences, called Δ values, show whether a syllable’s usage increased (Δ > 0) or decreased (Δ < 0) after stimulus presentation. To determine if these changes indicated a real behavioral modulation, we used a multivariate Hotelling’s T 2 test to see if the set of Δ values differed significantly from zero. In males ( Figure 1 B), the Δ values were markedly different from zero in response to CS 1 ( p < 0.0001) and CS 2 ( p = 0.0210). Similarly, in females ( Figure 1 C), the Δ values also significantly differed from zero in response to CS 1 ( p < 0.0001) and CS 2 ( p = 0.0167), showing that both conditioned cues caused consistent behavioral changes in both sexes. To analyze syllable transition, we calculated transition matrices between syllables to examine the dynamics of behavior. 30 We assessed the change by measuring the Manhattan distance (MD) between different time points (before and during the stimulus presentation) and derived an empirical p -value through bootstrapping. This dynamic analysis confirmed Hotelling’s T 2 test results. In males ( Figure 1 D), both CS 1 (MD = 75, p < 0.0001) and CS 2 (MD = 65, p = 0.000114) showed significant shifts in their transitions. Similarly, females ( Figure 1 E) also exhibited a significant shift to CS 1 (MD = 65, p = 0.000162) and CS 2 (MD = 53, p = 0.0424). We next aimed to identify a behavioral metric that we could correlate with neural activity. To this end, we analyzed the Δ values across syllables, each representing a stereotyped movement pattern, such as forward, backward, stop, or turn ( Figure 1 F). The value given to each syllable is proportional to its abundance, with lower numbers being the most abundant ones. Importantly, our results replicated previous findings involving two highly abundant syllables, 0 and 2. 30 For syllable 0 ( Figure 1 G), which corresponds to walking forward, the most common movement, we observed a downregulation in response to CS 1 , with a mean Δ value of approximately −50 and no significant sex differences ( p = 0.616). In contrast, syllable 0 showed upregulation to CS 2 , with a mean Δ value of approximately +50, again without significant differences between sexes ( p = 0.999). These finding on syllable 0 agrees with the fact that responses to CS 1 and CS 2 are essentially distinct, and not one a weaker version of the other. 30 On the other hand, for syllable 2, representing immobility, both CS 1 and CS 2 evoked strong upregulation, with mean Δ values exceeding +700 for CS 1 and +150 for CS 2 ( Figure 1 H). These effects were consistent across sexes, with no significant differences for either CS 1 ( p = 0.999) or CS 2 ( p = 0.997), indicating that CS 1 and CS 2 cause an increase in immobility. While acknowledging the complexity of behavior and the detailed information obtained by each syllable, it is essential to select a metric that effectively summarizes behavioral change. Given that the highest Δ scores for both CS 1 and CS 2 are for syllable 2, which represents immobility-like behaviors, we chose this type of response for downstream modeling of brain activity. Specifically, we focused on an immobility-related measure, huddling, which we can automatically quantify using DeepOF, 33 and we previously confirmed that it strongly correlates with manual freezing measurements. 30 Indeed, huddling also strongly correlates with measurements of syllable 2 (r = 0.924, p < 0.0001) ( Figure 1 I). The analysis of huddling displays significant changes in response to CS 1 and CS 2 in both males ( Figure 1 J) and females ( Figure 1 K). Furthermore, a closer examination of individual values for our immobility measure revealed variability in performance. By calculating the behavioral response ratio between post-stimulus and pre-stimulus periods, we found that the lowest response ratio to CS 1 was 1.5, indicating at least a 50% increase in immobility after the first-order stimulus. Because the distribution of ratios was continuous rather than bimodal, we used this value as a biologically motivated threshold to classify animals as high- or low-performers for CS 2 in males ( Figure 1 L) and females ( Figure 1 M). Overall, using TRAP2:Ai14 mice, we confirmed the effectiveness of our SOC protocol. Behavioral analysis with DeepLabCut and Keypoint-MoSeq showed significant and consistent changes in syllable expression after CS 1 and CS 2 presentations in both sexes. Notably, immobility-related behavior (syllable 2) exhibited strong upregulation, leading us to select immobility as a reliable behavioral measure for further analysis alongside neural activation. CellRake accurately quantifies fluorescent markers in histological brain sections After establishing a behavioral metric to correlate behavioral responses with brain activity, we quantified this activity by counting tdTomato- and cFos-positive cells across various brain regions. Due to the large number of animals from both sexes and the involvement of multiple brain areas, manual counting would be tedious and prone to bias. To address this, we developed CellRake, a Python package that leverages several basic Python libraries, including NumPy 34 for numerical operations, pandas 35 for data manipulation, scikit-image 36 for advanced image analysis techniques, scikit-learn 37 for predictive data analysis, and Matplotlib 38 for plotting. Relying solely on these well-known, basic packages creates a robust environment that enhances CellRake’s usability, reliability, and maintainability. CellRake uses a two-step workflow (segmentation followed by classification) to detect cells expressing positive fluorescent markers such as tdTomato and cFos ( Figure 2 A). To assess Cell- Rake’s performance throughout this process, we used a benchmark dataset of 153 images of the BLA from 36 male mice, since this brain region is a key player in higher- order conditioning 17 , 18 , 39 and its cellular architecture tends to be more uniform. 40 All images show nuclear cFos labeling and cytoplasmic tdTomato expression resulting from TRAP2:Ai14-mediated genetic recombination. Figure 2. Open in a new tab Overview and performance evaluation of CellRake’s cell detection pipeline (A) Schematic representation of CellRake in two phases: segmentation of candidate blobs and machine learning-based classification to identify true positives. (B) Comparison of CellRake’s segmentation (second panel) against standard thresholding methods (Otsu, Multi-Otsu, Li, Yen), highlighting its ability to preserve candidate blobs across heterogeneous intensity backgrounds. The scalebar corresponds to 200 μm. (C and D) Classifier results for cFos- and tdTomato-labeled cells. Left: example of nuclear cFos and cytoplasmatic tdTomato staining, along with the identified true positive blobs. Right: PCA of 22 extracted features using two components: PC1 enriched in color intensity features and PC2 enriched in size features. The actual markers (in red) are separate from the false-positive blobs (in blue). The scalebars correspond to 200 μm. (E) Feature analysis comparing tdTomato + ( n = 990) and cFos + ( n = 12642) cells: eccentricity ( p = 0.545), area ( p < 0.0001), perimeter ( p < 0.0001), intensity ratio ( p < 0.0001), intensity difference ( p < 0.0001), mean intensity ( p < 0.0001), SD intensity ( p < 0.0001). (F and G) Feature importance analysis in the cFos and tdTomato classifiers. CellRake primarily relies on intensity-based features, whereas size and shape play a minor role. (H–J) Correlation of CellRake’s automated counts with manual ground truth for cFos (r = 0.944, p < 0.0001, n = 153), tdTomato (r = 0.879, p < 0.0001, n = 153), and their colocalization (r = 0.912, p < 0.0001, n = 153). (K and L) Example of cFos (green) and tdTomato (red) markers, along with the identified cells in blue, for the dentate gyrus (DG, left) and dorsal periaqueductal gray (PAGd, right). These images are automatically generated by CellRake just for visual inspection, thus lack a scalebar. P-values are denoted as p < 0.05 (∗), p < 0.01 (∗∗), and p < 0.001 (∗∗∗). The data are presented as mean ± SD. Comparing CellRake’s initial segmentation method with other thresholding techniques, our approach preserves the integrity of detected regions, enabling robust detection even in images with variable intensities or heterogeneous backgrounds. A visual inspection of images with non-uniform surroundings shows that CellRake consistently retains candidate blobs under challenging conditions. In contrast, traditional thresholding methods such as Otsu, Multi-Otsu, Li, and Yen often produce incomplete or fragmented segmentations ( Figure 2 B). Once segmentation is complete, we independently trained two random forest classifiers, one for each marker (cFos and tdTomato). For each candidate blob, CellRake extracts 22 features, including brightness, area, perimeter, and eccentricity. These features are then used as input to the classifier, which evaluates each blob as either a true or false positive. On the training set, for the cFos-labeled images, the classifier achieved an ROC AUC of 0.987, an average precision (AP) of 0.976, a precision of 0.971, a recall of 0.913, and an F1 score of 0.941. On the test set, the model exhibited strong generalization performance, with a ROC AUC of 0.993, AP of 0.979, precision of 0.963, recall of 0.883, and F1 score of 0.922 ( Figure 2 C). In contrast, the classifier trained on tdTomato-labeled images achieved near-perfect metrics on the training set: ROC AUC of 0.999, AP of 0.999, precision of 0.992, recall of 0.983, and an F1 score of 0.987. These high values remained on the test set, with a ROC AUC of 0.9998, AP of 0.9996, precision of 0.992, recall of 0.994, and F1 score of 0.993 ( Figure 2 D). Feature analysis using CellRake confirmed that both markers share a similar elliptical shape (eccentricity = 0.65, p = 0.545), though tdTomato + markers are 33.4% larger ( p < 0.0001) and have a 13% longer perimeter ( p < 0.0001). Regarding intensity, tdTomato markers display significantly higher contrast with a 34% greater mean intensity ratio ( p < 0.0001) and a 52% larger cell-to-background difference ( p < 0.0001). At the same time, cFos + cells show 23% higher mean absolute intensity ( p < 0.0001) and reduced variability ( p < 0.0001) ( Figure 2 E). These morphological and intensity features serve purely as computational descriptors that enable accurate classification and are not intended to convey direct biological meaning, but instead to capture image-level properties that differentiate the two labeling methods. Examining which features drive the classifier’s decision-making, we found that CellRake heavily prioritizes intensity-based features across marker types. For cFos, it relies more on features such as the standard deviation of blob intensity and mean ratio, which better capture dynamic intensity patterns ( Figure 2 F). In contrast, for tdTomato, where the contrast with the background is high, features such as the difference between the blob and background intensities dominate the decision-making process ( Figure 2 G). Interestingly, shape and size features, such as area and perimeter, play a negligible role in both cases. This indicates that the classifier primarily relies on intensity metrics to distinguish true positives from background noise. To evaluate the global performance of CellRake, we compared the classifier’s output counts for cFos and tdTomato with manual counts and the colocalization output. The results indicate that CellRake can detect and colocalize cFos + and tdTomato + cells in all three cases, with automatic counts closely matching the ground truth. For cFos, the mean classified positives (82.627) closely resemble the ground truth (79.725), with a strong and significant correlation (r = 0.944, p < 0.0001) ( Figure 2 H). Similarly, for tdTomato, the average classified positives (6.471) closely align with manual counts (5.569), as evidenced by a strong and significant positive correlation (r = 0.879, p < 0.0001) ( Figure 2 I). The colocalization analysis also achieves outstanding results, with a high and significant correlation (r = 0.912, p < 0.0001) and the average number of colocalized cells identified by CellRake (1.431) closely approximating the manual value (1.699) ( Figure 2 J). Although we trained and evaluated CellRake using images from the BLA, its feature set, which includes global morphological and intensity descriptors, captures properties of fluorescently labeled cells that are consistent across different brain regions. Therefore, the algorithm is not limited to region-specific architecture and can be applied to various areas. This generalization is evident when applying our classifier to other regions with significantly different cellular density and background structure, such as the dentate gyrus (DG, Figure 2 K) and dorsal periaqueductal gray (PAGd, Figure 2 L). Overall, evaluating CellRake on a benchmark dataset of cFos- and tdTomato-labeled images, the classifier achieved high performance across all metrics (ROC AUC, AP, precision, recall, and F1), with notably higher evaluation values for tdTomato detection than for cFos. We also observed that intensity features play a key role in driving classification decisions. Importantly, CellRake’s automatic cell counts showed strong, significant correlations with manual counts. DG and CA3 link behavioral performance to neural activation during second- order memory recall After benchmarking CellRake, we used it to quantify the number of positive cells across different brain regions in TRAP2:Ai14 mice. These regions included the primary visual cortex (V1), retro-splenial cortex (RSC), DG, CA3, paraventricular nucleus of the thalamus (PV), medial orbitofrontal (MO) area, prelimbic cortex (PL), cingulate cortex (CG), ventral periaqueductal gray (PAGv), PAGd, primary auditory area (AU1), PRh, and BLA. For each region, we analyzed 4–6 brain slices per animal, from both males and females, stained for tdTomato ( Figure 3 ) and cFos ( Figure 4 ). Figure 3. Open in a new tab Representative images for all 13 brain areas of both sexes for the tdTomato marker The analyzed brain areas are the primary visual cortex (V1), retrosplenial cortex (RSC), dentate gyrus (DG), cornu ammonis 3 (CA3), paraventricular nucleus of the thalamus (PV), medial or-bitofrontal area (MO), prelimbic cortex (PL), cingulate cortex (CG), ventral periaqueductal gray (PAGv), dorsal periaqueductal gray (PAGd), primary auditory area (AU1), perirhinal cortex (PRh), and basolateral amygdala (BLA). The white dotted line indicates the contour of each region. Scale bars, 200 μm. Figure 4. Open in a new tab Representative images for all 13 brain areas of both sexes for the cFos marker The analyzed brain areas are the primary visual cortex (V1), retrosplenial cortex (RSC), dentate gyrus (DG), cornu ammonis 3 (CA3), paraventricular nucleus of the thalamus (PV), medial or-bitofrontal area (MO), prelimbic cortex (PL), cingulate cortex (CG), ventral periaqueductal gray (PAGv), dorsal periaqueductal gray (PAGd), primary auditory area (AU1), perirhinal cortex (PRh), and basolateral amygdala (BLA). The white dotted line indicates the contour of each region. Scale bars, 200 μm. After obtaining cell counts for each slice, we then assessed the relationship between these neural activation profiles and behavioral performance using a linear model. Specifically, we fit-ted separate generalized linear mixed models (GLMMs) using immobility ratios (i.e., the ratio of huddling before and after the presentation of CS 1 or CS 2 ). We used individual brain regions for each model, except for the DG and CA3, which we combined into a single model with a fixed effect, given their close anatomical proximity and the fact that the DG encloses part of the CA3. Overall, these separated models enabled us to evaluate the contribution of each region independently, thereby preventing multicollinearity and enhancing model interpretability. We used the raw, un- normalized counts without averaging, and included the logarithm of the ROI area as an offset and the individual IDs as random effects to account for repeated measures. We also included other fixed effects, such as sex and the behavioral performance category (high-vs. low-performers), the latter only for tdTomato count, since all animals were high-performers for cFos. Initially, we fitted a GLMM assuming a Poisson distribution because it is suitable for discrete data, such as counts. However, the tdTomato and cFos data showed overdispersion, with the variance more than twice the mean, violating the Poisson assumption that the mean equals the variance. To address this, we used a Negative Binomial distribution, which is robust to overdispersion. When fitting a GLMM for tdTomato labeling, which marks CS 2 -evoked neural activity in TRAP2:Ai14 mice, we found evidence of zero inflation. In other words, the number of zero counts was higher than expected under standard count models, indicating an excess of brain slices with no tdTomato + cells. To account for this, we incorporated different zero-inflation structures into the models to capture the likelihood of excess zeros. We then compared these models using the Akaike Information Criterion (AIC) to determine the best approach for handling zero inflation in our count data. When comparing the effects of behavioral response ratios on tdTomato counts across brain regions, only the dorsal DG and CA3 showed a significant coefficient ( p = 0.00389) ( Figure 5 A), with no significant differences across sex ( p = 0.09). Between the two brain regions, DG exhibited significantly higher tdTomato + cell counts than CA3, with more than twice the expected number of labeled cells ( p < 0.0001). Notably, the interaction between the immobility ratio and the behavioral performance category (high- and low-performers) was also significant ( p = 0.0429), indicating that performance level differently influences the neural activation patterns associated with behavioral responses. For high performers, a one-unit increase in the response ratio yields a 35.5% increase in expected tdTomato + cell counts, holding other variables constant. In contrast, for low performers, this effect not only disappeared but also slightly reversed, with a one-unit increase in the behavior ratio corresponding to a 28.2% decrease in expected cell counts, assuming all other variables unchanged ( Figures 5 B and 5C). Figure 5. Open in a new tab Quantification and modeling of tdTomato and cFos counts during memory recall (A) Significance of the generalized linear mixed models (GLMM) coefficients of the tdTomato + counts in each brain region by the immobility ratio: primary visual cortex (V1), retrosplenial cortex (RSC), dentate gyrus (DG), cornu ammonis 3 (CA3), paraventricular nucleus of the thalamus (PV), medial orbitofrontal area (MO), prelimbic cortex (PL), cingulate cortex (CG), ventral periaqueductal gray (PAGv), dorsal periaqueductal gray (PAGd), primary auditory area (AU1), perirhinal cortex (PRh), and basolateral amygdala (BLA). Only the hippocampal DG and CA3 showed a significant coefficient ( p = 0.00389, n = 23) for the CS 2 recall. (B and C) tdTomato + cells in the CA3 and DG for each immobility ratio, separated by performance ( p = 0.0429, n = 23). In DG and CA3, high-performing mice exhibited increased tdTomato + cell activation with immobility, whereas low-performing mice showed an opposite relationship. (D) Heatmaps of mean density of tdTomato + cells separated by sex and brain region. (E) Significance of the GLMM coefficients of the cFos + counts in each brain region by the immobility ratio. The V1 ( p = 0.0405, n = 23) and PAGv ( p = 0.00902, n = 23) yielded significant coefficients. (F and G) cFos + cells in the V1 and PAGv for each immobility ratio. Both areas exhibit a positive trend, with males having higher values than females in both regions (V1: p = 0.00793, n = 23; PAGv: p < 0.0001, n = 23). (H) Heatmaps of mean density of cFos + cells separated by sex and brain region. P-values are denoted as p < 0.05 (∗), p < 0.01 (∗∗), and p < 0.001 (∗∗∗). To assess differences in labeling across brain regions, we normalized raw counts by the area of each region of interest (ROI) and calculated the average cell density per region for each animal. Overall, females exhibited similar or slightly higher tdTomato levels than males in most areas, with the most pronounced difference in the PV, where females showed nearly twice the density (78 cells/mm 2 ) compared to males (39 cells/mm 2 ). Additional regions with no-table female-dependent increases included the RSC, DG, MO, PAGd, and PAGv. In contrast, only a few regions, such as the CG, showed higher counts in males, though the effects were modest. Overall, many areas showed consistently low tdTomato counts in both sexes ( Figure 5 D). Regarding the cFos effects, indicating immediate-early gene expression after CS 1 exposure, the two brain regions that showed a significant effect of immobility ratios were the PAGv ( p = 0.00902) and the V1 ( p = 0.0405) ( Figure 5 E). In both areas, higher immobility ratios corresponded with increased expected numbers of cFos + cells. Specifically, in V1, a one-unit increase in the immobility ratio led to a 29.1% increase in cFos + cell counts according to the model ( Figure 5 F), while in PAGv, the same increase in the ratio corresponded to a 24.2% rise in cFos labeling, holding other variables constant ( Figure 5 G). Additionally, both regions showed a notable effect of sex: The model estimated that males had approximately 91.5% more cFos + cells than females in V1 ( p = 0.00793), and nearly four times more in PAGv ( p < 0.0001). In cFos labeling, the sex differences showed an opposite pattern compared to tdTomato. Males generally had higher cFos counts across nearly all regions, especially in the MO (460 vs. 190), CG (510 vs. 161), and V1 (203 vs. 95). In contrast, females exhibited relatively higher cFos activation only in the DG and PAGd, indicating region-specific sex differences in stimulus-induced activation ( Figure 5 H). Overall, the linear models on CS 2 -evoked activity showed that behavioral performance significantly shaped neural activation in the dorsal DG and CA3. In high performers, increased immobility corresponded to higher tdTomato + counts, while in low performers, the same trend reversed. These findings suggest that successful memory recall of second-order conditioned memories may involve dorsal hippocampal circuits, specifically DG and CA3. Inhibiting DG/CA3 CaMKII + cells impairs second-order conditioning To gain a deeper understanding of the roles of DG and CA3 in SOC, we aimed to identify cells activated by CS 2 exposure and determine whether the tdTomato + cells were glutamatergic or GABAergic. To achieve this, we employed an immunohistochemical approach to manually colocalize tdTomato signal with antibodies against CaMKII, a marker of glutamatergic neurons, and GAD67, a marker of GABAergic neurons ( Figure 6 A). Figure 6. Open in a new tab Role of CaMKII + cells in DG/CA3 during SOC recall (A) Immunohistochemical colocalization of tdTomato (red) with CaMKII + cells (magenta) and GAD67 + cells (green) in DG and CA3, showing that most tdTomato + cells are excitatory neurons expressing CaMKII. Dashed lines outline the hippocampal DG. Scale bars, 200 μm. (B) Proportion of CaMKII + cells out of CaMKII + and GAD67 + cells for each immobility ratio, separated by performance ( p = 0.033, n = 23). High-performers maintain a stable CaMKII + ratio of 70%, whereas low-performers show an increasing trend with immobility. (C) Experimental design schema. Male mice received bilateral DG/CA3 injections of AAV-CaMKII-hM4Di-mCherry (DREADD group) or control virus AAV-CaMKII-mCherry. After 4 weeks, mice underwent the behavioral SOC protocol, in which they received an injection of the DREADD agonist J60 (0.1 mg/kg, i.p.) 1 h before CS 2 recall testing. (D) Verification of viral expression: Representative images of mCherry fluorescence (top) indicating viral spread, and P-PDH α 1 immunostaining (bottom) to measure neural inactivity. Scale bars, 400 μm. (E) Quantification of P-PDH α 1 intensity ratio (DG/CA3 vs. surrounding tissue) shows significant inhibition of neural activity in the DREADD group treated with J60 compared to controls with J60 ( p = 0.0030, n = 3), DREADD with saline ( p = 0.0356, n = 4), and controls with saline ( p = 0.0513, n = 3). (F and G) Behavioral effects of chemogenetic inhibition on immobility (huddle) during the SOC recall. The control group shows normal increases in immobility to both CS 2 ( p = 0.0443, n = 15) and CS 1 ( p = 0.0033, n = 15). The DREADD group exhibits an impaired immobility response to CS 2 ( p = 0.55, n = 15) but not to CS 1 ( p = 0.0078, n = 15). P-values are denoted as p < 0.05 (∗), p < 0.01 (∗∗), and p < 0.001 (∗∗∗). The data are presented as estimated intensity ratios with their 95% confidence interval on E, and mean ± SD on panels F and G. We analyzed the results by fitting a GLMM under a binomial distribution, controlling for the same explanatory variables as the previous linear models. When examining the proportion of CaMKII + cells across experimental conditions, we observed no significant effects for sex ( p = 0.731) or between brain regions (i.e., DG and CA3, p = 0.210). However, group-specific effects of the immobility ratio on CaMKII + proportions showed a significant interaction ( p = 0.033). Among low performers, each one-unit increase in the immobility ratio resulted in a 4.37-fold increase in the expected proportion of CaMKII + cells. This sharp rise, however, led to saturation as the response ratio continued to increase. High performers maintained a relatively stable CaMKII + proportion of 80% across the entire range of immobility ratios, with no significant modulation ( Figure 6 B). Therefore, this result suggests that the majority of labeled cells are excitatory neurons. By combining the CaMKII + cell proportions with the tdTomato + counts, we suggest that CaMKII + cells in the dorsal hippocampus, particularly within the DG/CA3 regions, may contribute to modulating CS 2 memory recall. To test its causal involvement, we employed a chemogenetic approach using designer receptors exclusively activated by designer drugs (DREADDs). We bilaterally injected male mice in the dorsal hippocampus, spreading across DG/CA3 with either AAV-CaMKII-hM4Di-mCherry (hereafter, DREADD group) or AAV-CaMKII-mCherry (hereafter, control group). This chemogenetic approach enabled selective inhibition of CaMKII + cells in the targeted hippocampal regions during the retrieval of CS 1 and CS 2 memories. After four weeks to ensure adequate viral expression, the animals underwent the standard behavioral protocol used to induce and measure first- and second-order conditioned responses. On the recall day, mice received a 0.1 mg/kg intraperitoneal injection of the DREADD agonist JHU37160 dihydrochloride (J60) 41 1 h before the CS 2 exposure test ( Figure 6 C). To confirm proper viral expression in the targeted cells, we assessed the localization of the viral constructs in both DREADD and control groups ( Figure 6 D). In addition, to validate the DREADD-dependent inhibition of hippocampal circuits, we divided each group into two subgroups: One received a saline injection, and the other received J60 to modulate the DREADD receptors (i.e., inhibiting cell activity). One hour after injection, we perfused the mice and processed the brain tissue to immunolabel phospho-pyruvate dehydrogenase (P-PDH α 1), which inversely correlates with neural activity. 24 cFos expression typically reflects neural activation and may be less reliable for indicating decreases in activity following DREADD-mediated inhibition. To quantify changes in neural inhibition, we calculated the ratio of the P-PDH α 1 signal in the targeted (DG/CA3) region relative to the surrounding brain tissue. We analyzed this metric using a linear mixed-effects model (LMM) with fixed effects for brain region, viral construct, injection type, and their interaction. We included mouse identity as a random intercept to account for repeated measures across areas. The model showed that mice expressing the inhibitory DREADD and injected with J60 had a significantly higher P-PDH α 1 intensity ratio, about 1.6 times, indicating decreased neural activity in the targeted areas. In contrast, the same mice injected with saline had a significantly lower ratio of only 1.2 ( p = 0.0356). The control groups with J60 and saline also displayed baseline activity, with ratios of 1.1 and 1.3, respectively ( Figure 6 E). These results confirm the functional expression of the viral constructs and the effectiveness of the chemogenetic manipulation in CaMKII + hippocampal neurons ( Figure 6 D). After validating our chemogenetic approach, we evaluated behavioral responses in both DREADD and control groups using huddling, which is the same immobility measure that showed significant modulation in response to CS 1 and CS 2 in TRAP2:Ai14 mice. The control group exhibited significant behavioral changes in response to both CS 2 ( p = 0.0443) and CS 1 ( p = 0.0033) ( Figure 6 F). In contrast, in the DREADD group, the inhibition of DG/CA3 selectively disrupted the response to CS 2 ( p = 0.55), while the response to CS 1 remained unaffected ( p = 0.0078) ( Figure 6 G). Overall, these findings suggest that CaMKII + cells in the DG/CA3 are essential for recalling second-order, but not first-order, conditioned memories. Discussion By combining pose tracking, genetic tagging, automated image analysis, and chemogenetic manipulation, this study investigates the brain substrates underlying SOC and FOC expression in a light-tone SOC protocol in mice. Using TRAP2:Ai14 male and female mice, we permanently labeled neurons activated during memory retrieval in SOC and FOC and identified brain-wide activity patterns. 29 Behavioral analysis using DeepLabCut 31 and Keypoint-MoSeq 32 revealed consistent changes in syllable usage, particularly increased immobility, in response to both CS1 and CS2, establishing huddling 33 as a reliable and straightforward behavioral marker for SOC recall. Modeling brain activity showed that CS 2 -induced behavioral responses could explain the activity of specific excitatory neurons (CaMKII + cells) in the DG/CA3 regions of the dorsal hippocampus. Notably, chemogenetic inhibition of these cells impaired the recall of SOC memories without affecting FOC recall, underscoring the crucial role of hippocampal excitatory neurons in the expression of SOC. The data-driven analysis of TRAP2:Ai14 animals confirms our previous findings on SOC behavioral responses in wild-type C57BL/6J mice 30 : (1) behavioral responses to CS 1 and CS 2 in our SOC paradigm remain unaffected by sex, (2) responses to CS 2 are generally less intense than those to CS 1 , and, most importantly, (3) CS 1 and CS 2 evoke intrinsically distinct behavioral patterns. Rather than simply mirroring first-order behavioral responses, SOC elicits unique behavioral profiles for each conditioned stimulus. In particular, as the syllable analysis shows, responses to CS 1 mainly consist of immobility, whereas CS 2 evokes a more diverse behavioral repertoire, including both passive behaviors, such as immobility, and active responses, such as exploration and escape-like movements. However, to correlate brain activity alongside behavioral performance, we relied on immobility (i.e., huddling) as our primary metric. Immobility emerged as the most robust and reliable behavioral indicator of associative memory during SOC, with mean Δ values between 3 and 14 times higher than other behavioral syllables. Although linear models that include the full syllable quantification might achieve higher coefficients of determination, we deliberately prioritized interpretability over marginal improvements in model fit. Including all syllables leads to severe multicollinearity, weakening the stability, and reliability of lin-ear estimates. 42 Even when applying techniques to address this problem, such as dimensionality reduction and regularization, models with multiple features are more difficult to interpret biologically. Conversely, one metric offers a clear, meaningful, and reproducible measure that we can strongly associate with neural activity. 43 , 44 By focusing on this well-defined feature, we maintain interpretable and actionable results, particularly for posterior manipulations (e.g., chemogenetic approaches), where a solid behavioral anchor is essential for assessing causal effects. 45 After establishing a behavioral metric, we measured neural activity patterns in SOC (tdTomato + ) and FOC (cFos + ), which showed sex-dependent differences. CS 2 -evoked tdTomato labeling was generally higher in female mice, whereas cFos expression following CS 1 display was more pro-nounced in males. These differences may reflect sex-specific strategies or patterns of brain circuit engagement during recall, in line with prior reports indicating that female rodents often recruit broader neural networks for associative learning. 46 However, several methodological factors could also contribute. We cannot entirely exclude methodological confounds, such as sex differences in 4-OHT metabolism 47 or baseline expression of immediate-early genes. 48 , 49 We went without monitoring the estrous cycle, and hormonal fluctuations are known to affect hippocampal plasticity. 50 , 51 We therefore interpret these sex differences with caution, and future studies incorporating estrous monitoring and additional controls may determine whether these differences reflect true biological dimorphism or are due to technical variability. Moreover, we must note that tdTomato + counts are generally lower than cFos + ones. Although density calculations from small sampled volumes can introduce variability, normalizing by ROI area is essential for comparing regions and animals with different sampling extents. Such sparse labeling is an inherent feature of TRAP2-based tagging, as only neurons with sufficiently strong and sustained cFos promoter activation are permanently labeled, resulting in generally lower tdTomato counts. Additionally, the 4-OHT dose used in our experiments (25 mg/kg) is lower than the doses employed in some other studies, which can reach up to 100 mg/kg and may therefore yield denser labeling. 52 Finally, our quantification is based on 3–5 sections at 30 μm per region, even though each region spans many more rostro-caudal levels. We believe that a complete reconstruction across the entire extent of each structure would naturally yield much higher overall cell counts. Therefore, we must interpret these tdTomato labeling results with caution, acknowledging that the sparse tagging reflects methodological limitations rather than a lack of neural activity. Regardless of sex, we found that specific brain regions were strongly activated upon the presentation of CS 2 . These regions include the PV, PAGv, and PAGd, which are structures commonly linked to emotional and survival-related processes. The PV is involved in arousal, attention, and stress regulation, 53 , 54 , 55 indicating that SOC recall may engage circuits related to emotional salience and autonomic control, even without a direct pairing with the US. Similarly, the PAG, known for its role in fear, defensive behavior, and pain modulation, 56 , 57 showed strong activation, supporting the idea that CS 2 can trigger internal fear-related states. Future experiments will interrogate the potential causal involvement of these brain regions, which strongly activate during light-tone associations, in SOC formation. During the presentation of the first-order stimulus (CS 1 ), cFos activation in the PAGv and V1 significantly correlated with better behavioral performance, specifically immobility. The PAGv is classically associated with freezing and passive defensive responses to aversive stimuli, 7 , 58 , 59 while activation in V1 may reflect enhanced visual sensory input during light presentation. 59 , 60 These findings support the idea that CS 1 recall engages established defensive and sensory circuits. In contrast, during the presentation of the second-order stimulus (CS 2 ), tdTomato expression in the DG and CA3 significantly correlated with greater immobility responses. Notably, the DG facilitates pattern separation, transforming similar inputs into distinct representations, while CA3 supports pattern completion, enabling the retrieval of complete memories from partial or ambiguous cues. 61 , 62 , 63 Therefore, these findings suggest that DG and CA3 activity may help facilitate the inference of threat from indirect associations, such as CS 2 , which was never directly paired with the US. Importantly, chemogenetic inhibition confirmed a causal role for excitatory neurons in these hippocampal regions during the expression of CS 2 responses, while leaving CS 1 unaffected. This difference indicates that dorsal hippocampal excitatory activity is necessary for recalling second-order, but not first-order, associations. While CS 1 memory relies on a direct US pairing, typically supported by subcortical circuits such as the amygdala, PAG, and hypothalamic, 7 , 8 our SOC paradigm seems to require an additional inference step (i.e., linking CS 2 to CS 1 , and then to the US 5 , 11 , 64 ). However, a limitation of our method is that the chemogenetic manipulation affects all CaMKII- expressing neurons in the brain region, rather than selectively targeting the activated ensembles identified in our labeling experiments. Therefore, the causal evidence might reflect the role of a broader hippocampal excitatory population rather than just the neurons engaged during SOC recall. In this context, these hippocampal excitatory neurons may mediate complex associative learning, enabling animals to infer threat from stimuli that are only indirectly linked to the US. 61 , 62 , 63 Thus, while DG and CA3 activity may be dispensable for direct associations (FOC), they appear critical for higher-order associations (SOC). Previous studies support this interpretation, at least in the context of SPC. For example, the dorsal hippocampus is necessary to encode associations between two neutral stimuli during preconditioning, but not for learning a neutral cue association with the US. 16 Similarly, type 1 cannabinoid receptors (CB1R) in hippocampal GABAergic neurons are both necessary and sufficient for SPC, but not for direct conditioning. 65 Although some studies have also implicated the dorsal hippocampus in FOC with light or tone stimuli, 66 , 67 , 68 the overall body of evidence supports the idea that higher-order conditioning engages dorsal hippocampal mechanisms specifically tuned to inferring relationships between indirectly associated stimuli. Throughout our study, we analyzed tdTomato and cFos expression using our self-developed CellRake package, a scalable and accurate Python tool for the automated detection of fluorescently labeled cells. We benchmarked CellRake by training separate models for tdTomato and cFos. Although both classifiers performed well, the tdTomato model consistently outperformed the cFos model across all metrics, including computing speed. This difference likely reflects the intrinsic properties of the markers: cFos, a nuclear marker, produces dimmer, smaller, and often overlapping signals, while tdTomato, a cytoplasmic marker, generates larger, more uniformly labeled cells with higher contrast against the background. Indeed, the nature of each fluorescent marker can also explain these differences, since tdTomato is triggered by genetic recombination, whereas cFos depends heavily on antibody-based labeling. However, despite these differences, CellRake achieved strong agreement with manual counts of both markers, and its design could enable easy adaptation to other markers or experimental conditions. This tool also offers several advantages over existing software. Unlike broader platforms such as QuPath, 69 CellProfiler, 70 Ilastik, 71 or Cellpose, 28 which provide extensive but often complex functionalities, CellRake is designed for a single task, resulting in a distribution size of under 30 KB and a significantly lower learning curve. Its intuitive design allows both advanced users and beginners to work efficiently from day one. The annotation process is fast and precise, relying on Python tools without the need for external programs such as ImageJ, 72 thereby eliminating time-consuming manual steps, such as drawing regions of interest around blobs, required for training the model. Additionally, CellRake runs efficiently on standard hardware using scikit-learn models 37 rather than GPU-intensive deep learning methods, making it a more accessible and practical choice for researchers without high-performance computing resources. Taken together, our findings identify excitatory neurons in the DG/CA3 as key substrates supporting SOC expression. Future research should investigate how hippocampal excitatory and inhibitory circuits interact across the different stages of aversive SOC paradigms, whether these mechanisms apply to other types of second-order associations (e.g., using appetitive cues), and how they may become dysregulated in models of psychiatric disorders. Overall, by combining advanced behavioral measurement, temporal cell tagging, scalable histological analysis, and targeted circuit inhibition, this study advances our understanding of how the brain learns from indirect experiences. Limitations of the study Although our behavioral analyses leveraged advanced pose estimation and unsupervised syllable decomposition, we ultimately relied on immobility (huddling) as the primary metric to relate neural activity to performance. While this choice improved interpretability, it may fail to fully capture the multidimensional nature of second-order conditioned responses. In addition, we performed cell quantification on a limited number of rostro-caudal sections per region, which may not reflect the complete anatomical distribution of activated neurons. Furthermore, the chemogenetic manipulation targeted broad CaMKII + populations within DG/CA3 rather than the specific neuronal ensembles identified during SOC recall. Finally, potential sex-dependent effects should be interpreted cautiously, as we did not monitor the estrous cycle. Resource availability Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Dr. Arnau Busquets Garcia ( abusquetsresearchmar.net ). Materials availability This study did not generate new materials. Data and code availability • All data generated in this study are available in a Zenodo repository under the following https://doi.org/10.5281/zenodo.19075438 . • All original code of CellRake has been deposited at github.com/marccanela/cellrake and it is publicly available under the https://doi.org/10.5281/zenodo.19002677 . • Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Acknowledgments We want to thank the personnel of the Animal Facility of the Parc de Recerca Biomedica de Barcelona (PRBB) for mouse care. We thank all the members of our lab for valuable discussions during the development of the project. This work was supported by the Generalitat de Catalunya (SGR-00022) from the Departament d’Economia i Coneixement de la Generalitat de Catalunya (Spain) and from the European Research Council under the European Union’s Horizon 2020 research and innovation program (grant agreement no. [948217]). The project that gave rise to these results received the support of a fellowship from 'la Caixa' Foundation (ID 100010434). The fellowship code is LCF/BQ/DR22/11950014. Author contributions A.B-G. and M.C. contributed to the conception of the project. M.C., J.B-A., D.R-C., and J.G-P performed and analyzed all experiments. J.P. helped in setting up the SOC protocol. A.B-G. and M.C. wrote and revised the manuscript. All authors approved the final version of the manuscript. Declaration of interests The authors declare no competing interests. STAR★Methods Key resources table REAGENT or RESOURCE SOURCE IDENTIFIER Antibodies Rabbit monoclonal anti-cFos SynapticSystem #226008 Rabbit monoclonal anti-CaMKII Abcam #EP1829Y Mouse monoclonal anti-GAD67 Sigma Aldrich #MAB5406 Rabbit monoclonal anti-P-PDHα1 Cell Signaling Technology #37115 Donkey monoclonal anti-rabbit Alexa Fluor 488 Jackson ImmunoResearch #711-545-152 Donkey monoclonal anti-mouse Alexa Fluor 647 Jackson ImmunoResearch #712-605-150 Bacterial and virus strains AAV-CaMKII-mCherry Addgene viral prep #114469-AAV5 AAV-CaMKII-hM4Di-mCherry Addgene viral prep #50477-AAV2 Chemicals, peptides, and recombinant proteins JHU37160 dihydrochloride (J60) Hello Bio #HB6261 4-hydroxytamoxifen (4-OHT) Merck #H6278 Kolliphor oil Merck #C5135-500G Deposited data Raw and analyzed data This paper https://doi.org/10.5281/zenodo.19075438 Experimental models: Organisms/strains Fosˆtm2.1(icre/ERT2)Luo/J The Jackson Laboratory #030323 B6.Cg-Gt(ROSA)26Sorˆtm14(CAG-tdTomato)Hze/J The Jackson Laboratory #007914 C57BL/6J mice Charles River Laboratories N/A Software and algorithms CellRake This paper https://doi.org/10.5281/zenodo.19002677 Open in a new tab Experimental model and study participant details Animal models We used male and female TRAP2+/-; Ai14+/- mice (hereafter referred to as TRAP2:Ai14) to identify brain regions involved in SOC recall. Animals were 8–10 weeks of age at the time of the experiments. We generated TRAP2:Ai14 mice by crossing two homozygous mouse lines: TRAP2 mice (Fosˆtm2.1(icre/ERT2)Luo/J; The Jackson Laboratory, USA; stock 030323), which express CreERT2 under the control of the Fos promoter, and Ai14 tdTomato reporter mice (B6.Cg-Gt(ROSA)26Sorˆtm14(CAG-tdTomato)Hze/J; The Jackson Laboratory, USA; stock 007914). The resulting offspring were double heterozygous for Fos-CreERT2 and the Ai14 re- porter cassette, as previously described. 29 For experiments involving inhibition of CaMKII cells in the dorsal hippocampus, we used male C57BL/6J mice aged 8–10 weeks. These animals were obtained from Charles River Laboratories (Spain). We housed all mice under controlled environmental conditions, with four males or five females per cage, sexes separated. We main- tained animals on an inverted 12-h light/dark cycle (lights off at 7:30 a.m.) and conducted all experimental procedures during the dark phase under red light. We provided food and water ad libitum . We maintained ambient temperature (22 ± 1 °C) and relative humidity (50 ± 5%) through- out the study. All experimental procedures were approved by the institutional Animal Care and Use Committee (protocol ABG-19-055) and by the Generalitat de Catalunya (authorization ref. 10784), and were conducted in accordance with applicable national and institutional guidelines for the care and use of laboratory animals. To inhibit CaMKII + cells in the dorsal hippocampus, we used male C57BL/6J mice, aged 8-10 weeks, obtained from Charles River Laboratories (Spain). We housed all mice in groups of four males or five females, never mixing both sexes. We kept the animals in an inverted 12-hour light cycle (lights turned off at 7:30 a.m.) and conducted the experiments in darkness under red light. We provided food and water ad libitum and controlled the temperature (22 ± 1°C) and humidity (50 ± 5%). The appropriate animal care and use committee of our institution and the Generalitat de Catalunya approved all the procedures stated. Method details Conditioning chambers We performed the behavioral protocol in Imetronic conditioning boxes (France) provided with LEDs, speakers, and metallic grid floors. We used colored white noise (CWN, 57.1 dB) as the background and a white LED and a 3000 Hz sound (59.3 dB) as neutral conditioned stimuli (CS 2 and CS 1 , respectively). The unconditioned stimulus (US) was an electric foot shock (2-second, 0.6 mA). We alternated the context of the conditioning chamber across sessions by changing wall patterns (blank, striped, or spotted), floor textures (grid, plastic, or sawdust bedding), and olfactory cues(70% ethanol; VWR Chemicals, ref. 20821.33 or CR36; Jose Collado, ref. 12- 20/40-02020). Second-order conditioning protocol We used the same light-tone behavioral protocol for mouse light-tone second-order conditioning (SOC) described and validated in our previous work. 30 Briefly, it consisted of five stages: ha- bituation to the conditioning chambers (day 1), pairing a light (CS 1 ) with an electric shock (day 2), pairing the previously conditioned CS 1 with a novel tone (CS 2 ) (day 3), and an independent probe test for the CS 2 and the CS 1 (day 4). Using this behavioral protocol, which involves observ- ing specific CS 1 - and CS 2 -mediated behavioral responses, we can study the brain substrates of SOC. Pose estimation and behavioral analysis We followed the methodological framework described in our previous work. 30 Briefly, the proce- dure involved recording videos of mice during recall stages and analyzing them with DeepLabCut 3.0, 31 focusing on 14 body parts. To identify the behavioral structure, we applied Keypoint- MoSeq, 32 an unsupervised clustering tool that captures the frequency and timing of behavioral motifs. In parallel, we used the supervised DeepOF pipeline 33 to extract huddling behavior, which acts as a proxy for fear-related responses. 30 We divided the animals into high- and low-performing groups based on their huddling re- sponses to CS 1 and CS 2 . We calculated a response ratio by dividing the immobility percentage during the stimulus period by that of the preceding baseline period. To define the threshold, we selected the lowest response ratio observed for CS 1 , which was 1.5. Animals with a response ratio below 1.5 were classified as low performers, while those with ratios above this threshold were considered high performers. Drug preparation and administration We prepared 4-hydroxytamoxifen (4-OHT; Merck, #H6278) at 20 mg/mL in ethanol by gently shaking it at 37°C for 5 minutes. We then aliquoted the solution and stored it at –20°C. Before use, we mixed 4-OHT with Kolliphor oil (Merck, Cat #C5135-500G) by shaking at 37°C for 5 minutes. We remove ethanol using a Savant DNA120 SpeedVac concentrator, yielding a final concentration of 10 mg/mL. Finally, we added 0.1 M phosphate-buffered saline to adjust the solution to a final dose of 25 mg/kg. We immediately administered the 4-OHT by intraperitoneal injection just after assessing the mice’s response to the tone (CS 2 ). We prepared JHU37160 dihydrochloride (J60; Hello Bio, #HB6261) in sterile saline to achieve a final concentration of 0.1 mg/kg. We administered J60 or saline intraperitoneally 1 hour before the probe test, during which CS 2 exposure occurred. Induction of tdTomato expression in TRAP2:Ai14 Immediately after assessing the mice’s response to the tone (CS 2 ), we injected a solution of 25 mg/mL 4-hydroxytamoxifen (4-OHT) intraperitoneally at a ratio of 0.1 mg/kg. The flow of 4-OHT induces the expression of the tdTomato marker protein in the cells activated during the CS 2 - recalling session. We then waited 1 week before assessing the recall of the light (CS 1 ) to allow sufficient time for tdTomato to express. Then, we evaluated the response to light, and 90 minutes after stimulus presentation, we prepared the mice for perfusion. Perfusion and sectioning of brains One week after 4-OHT administration and CS 2 exposure, we assessed recall of the light (CS 1 ) to allow sufficient time for tdTomato to express. We prepared the mice for perfusion 90 minutes after displaying the light stimulus. Following a modified protocol from, 73 we transcardiacally perfused the mice. Briefly, we anesthetized them using a ketamine (100 mg/kg) and xylazine (10 mg/kg) mixture and then perfused them with a 4% paraformaldehyde (PFA) solution in 0.1 M PB, pH 7.4. We stored the brains individually in 4% PFA-filled tubes at 4°C and, after 24 hours, transferred them to a 30% sucrose solution for cryoprotection at the same temperature. After three days, we froze the brains and cut coronal sections at a thickness of 30 μm using a Leica CM 1950 cryostat at -20°C. We stored the slices in cryoprotectant (30% ethylene glycol and 20% glycerol in 0.02 M PB) until their usage. Immunofluorescence staining We followed an immunohistological protocol as previously described 74 , 75 to detect cFos, CaMKII, GAD67, and Phospho-Pyruvate Dehydrogenase (P-PDH α 1) in perfused brain slices. We rinsed the collected brain sections three times with 0.1 M phosphate buffer (PB) for 5 minutes each. The tissue was then permeabilized with a 0.2% Triton X-100 solution (in PB) for five 5-minute intervals and blocked with 5% normal donkey serum (NDS) for one hour at room temperature. After blocking nonspecific binding, we incubated the slices overnight at 4°C with the primary antibody at 1:1000 in the remaining blocking solution. The primary antibodies were rabbit anti- cFos (SynapticSystem, #226008), rabbit anti-CaMKII (Abcam, #EP1829Y), mouse anti-GAD67 (Sigma Aldrich, #MAB5406), or rabbit anti-P-PDH α 1 (Cell Signaling Technology, #37115). The next day, we rinsed the slices again with a 0.2% Triton X-100 solution three times, fol- lowed by a 2-hour incubation at room temperature in the dark with the secondary antibody so- lution at a 1:1000 dilution. The secondary antibodies used were donkey anti-rabbit AlexaFluor 488-conjugated (Jackson ImmunoResearch, #711-545-152) or donkey anti-mouse AlexaFluor 647-conjugated (Jackson ImmunoResearch, #712-605-150). After incubation, we rinsed the sections five times with a 0.5% Triton X-100 solution, air-dried them on microscope slides, and mounted them with Fluoromount-G containing DAPI for nuclear staining. We stored the slides at 4°C and protected them from light. Imaging of brain slices We imaged the stained brain sections using a Nikon Eclipse Ni-E fluorescence microscope with NIS-Elements software. At 10X magnification, we captured four to six images of each animal’s region of interest in the brain. In TRAP2:Ai14 mice, we selected 13 brain regions: primary visual cortex (V1, Bregma = -3.64), retrosplenial cortex (RSC, Bregma = -1.82), dentate gyrus (DG, Bregma = -1.82), cornu ammonis 3 (CA3, Bregma = -1.82), paraventricular nucleus of the thalamus (PV, Bregma = -1.82), medial orbitofrontal area (MO, Bregma = 2.46), prelimbic cortex (PL, Bregma = 1.70), cingulate cortex (CG, Bregma = -0.22), ventral periaqueductal gray (PAGv, Bregma = -4.60), dorsal periaqueductal gray (PAGd, Bregma = -4.60), primary auditory area (AU1, Bregma = -3.64), perirhinal cortex (PRh, Bregma = -1.82), basolateral amygdala (BLA, Bregma = -1.82). CellRake’s segmentation algorithm CellRake software is available in the following GitHub repository: github.com/marccanela/cellrake . It begins the image analysis by processing each grayscale fluorescence image with a multistep segmentation pipeline. The purpose of segmentation is to identify well-defined regions of in- terest (ROIs) that correspond to positive signal detections. First, it detects potential blob-like structures using a Laplacian of Gaussian (LoG) filter, 76 which enhances local maxima in in- tensity and captures structures of varying sizes. These detections serve as seeds for further processing. CellRake extracts a circular area centered on each detection for every candidate region and applies Otsu’s method 77 to perform adaptive thresholding around the specific blob. This localized approach accounts for background intensity heterogeneity. Following threshold- ing, CellRake uses morphological operations to eliminate small holes within blobs and discard irregular or fragmented detections based on compactness and area constraints. The algorithm then applies a marker-based watershed algorithm to separate overlapping signals. Finally, Cell- Rake transforms the final segmented shapes into polygonal contours, clips them to remain within image bounds, and stores them in a structured dictionary format for downstream classification. CellRake’s classification algorithm We implemented a semi-supervised learning pipeline based on label spreading to classify seg- mented blobs. First, CellRake extracts 22 morphological and intensity-based features for each blob identified during segmentation. These features serve as input to a k-means clustering al- gorithm (with a default of 10 clusters) to facilitate stratified manual labeling. The user manually labels 10 representative blobs from each cluster as positive or negative. Given the typical class imbalance, which has fewer positives, we apply the Synthetic Minority Oversampling Technique (SMOTE) to oversample the minority class. CellRake then applies a K- Nearest Neighbors (KNN) kernel to spread labels across the dataset. To ensure high-confidence labels, we compute the entropy of each blob’s posterior class probabilities and retain only predic- tions with low entropy (less than 0.025). CellRake then uses these confidently labeled samples to perform a train-test split for supervised model training and evaluation. Finally, it outputs five performance metrics: area under the receiver operating characteristic curve (ROC AUC), aver- age precision (AP), precision, recall, and F1 score. CellRake’s colocalization of markers Identifying colocalizations (i.e., markers that overlap significantly in the same image) is often valuable in experimental designs with multiple markers. CellRake supports this by using images that were previously analyzed for each marker individually. After independent analyses, CellRake calculates the intersection of all the detected markers in each image. In this case, two markers are considered colocalized if the overlap area constitutes at least 80% of the smaller area. Stereotaxic surgeries We anesthetized mice via intraperitoneal injection of ketamine (75 mg/kg) combined with medeto- midine (1 mg/kg). We provided analgesia through a subcutaneous injection of meloxicam (5 mg/kg) administered before surgery and for two consecutive days postoperatively. After anes- thetization and confirmation of the absence of a paw pinch reflex, we secured the animal’s head in a stereotaxic frame and aligned it using bregma and lambda coordinates. We then performed two small craniotomies to enable bilateral infusion of the viral vector solution into the DG/CA3 from the dorsal hippocampus. The stereotaxic coordinates used were: anteroposterior (AP) -2 mm, mediolateral (ML) ±1.5 mm, and dorsoventral (DV) -2.1 mm. We injected 500 nL of either AAV-CaMKII-hM4Di-mCherry (Addgene viral prep #50477-AAV2, titer ≥7 × 1,012 vg/mL) or AAV-CaMKII-mCherry (Addgene viral prep #114469-AAV5, titer ≥7 × 1,012 vg/mL). After the injections, we sutured the incision site and allowed the mice to recover for 3 to 4 weeks before initiating the behavioral protocol. Quantification and statistical analysis To analyze the behavioral data, we used the same approach described in our recent work. 30 In brief, for each behavioral syllable obtained from Keypoint-MoSeq, 32 we calculated the difference between the time spent before and after stimulus presentation, known as Δ values. To as- sess whether syllable usage changed significantly after cue presentation, we applied Hotelling’s T 2 test on the multivariate distribution of Δ scores per sex and stimulus condition. We further analyzed behavioral dynamics using transition matrices, in which we computed Manhattan dis- tances (MD) between the pre- and post-stimulus periods. We assessed statistical significance using bootstrap-based empirical p-values. To assess significant differences between conditions in continuous univariate variables, we used two-tailed t-tests (paired or unpaired, depending on the design) and one-way or two-way ANOVA for comparisons involving multiple groups. We verified normality assumptions using the Shapiro–Wilk test and homogeneity of variances using Levene’s test. When normality was vio- lated, we used nonparametric tests, such as the Wilcoxon signed-rank test or the Mann–Whitney U test, as appropriate. To validate CellRake, we compared automated counts against manual annotations. We first assessed bivariate normality using the Henze-Zirkler test for multivariate normality. We used Pearson’s correlation for normally distributed data and Spearman’s correlation otherwise. To test the relationship between brain activation and behavior, we fitted generalized linear mixed models (GLMMs) separately for each brain region. We modeled cell counts (cFos or tdTomato) as the response variable with log-transformed ROI area as an offset, and animal ID as a random intercept to account for repeated measures. Fixed effects included sex, behavioral performance group, and immobility ratio, with interaction terms when appropriate. Initially, we assumed a Poisson distribution; however, due to overdispersion, we refitted the models with a Negative Binomial distribution. For the tdTomato data, which displayed excess zeros, we tested multiple zero-inflated models and selected the best-fitting one based on the Akaike Information Criterion (AIC). To determine the cellular identity of activated neurons, we used binomial GLMMs to model the probability that tdTomato + cells are CaMKII + across brain regions and performance groups. Fixed effects included immobility ratio, sex, and their interaction, with animal ID as a random effect. To confirm effective neural silencing via DREADDs, we analyzed P-PDH α 1 intensity ratios (targeted vs. surrounding tissue) using a linear mixed-effects model (LMM) with fixed effects for region, viral construct, injection type, and their interactions, as well as mouse identity as a random intercept. All statistical comparisons used the following significance thresholds: p < 0.05 (∗), p < 0.01 (∗∗), and p < 0.001 (∗∗∗). Published: March 30, 2026 Footnotes Supplemental information can be found online at https://doi.org/10.1016/j.isci.2026.115522 . Supplemental information Document S1. Table S1 mmc1.pdf (80.2KB, pdf) References 1. Poo M.m., Pignatelli M., Ryan T.J., Tonegawa S., Bonhoeffer T., Martin K.C., Rudenko A., Tsai L.H., Tsien R.W., Fishell G., et al. What is memory? the present state of the engram. BMC Biol. 2016;14 doi: 10.1186/s12915-016-0261-6. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 2. Squire L.R., Dede A.J.O. Conscious and unconscious memory systems. Cold Spring Harb. Perspect. Biol. 2015;7 doi: 10.1101/cshperspect.a021667. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 3. Baddeley A.D., Hitch G.J., Allen R.J. From short-term store to multicomponent working memory: The role of the modal model. Mem. Cogn. 2018;47:575–588. doi: 10.3758/s13421-018-0878-5. [ DOI ] [ PubMed ] [ Google Scholar ] 4. Gostolupce D., Lay B.P.P., Maes E.J.P., Iordanova M.D. Understanding as- sociative learning through higher-order conditioning. Front. Behav. Neurosci. 2022;16 doi: 10.3389/fnbeh.2022.845616. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Holmes N.M., Fam J.P., Clemens K.J., Laurent V., Westbrook R.F. The neural substrates of higher-order conditioning: A review. Neurosci. Biobehav. Rev. 2022;138 doi: 10.1016/j.neubiorev.2022.104687. [ DOI ] [ PubMed ] [ Google Scholar ] 6. Yonelinas A.P., Ranganath C., Ekstrom A.D., Wiltgen B.J. A contextual bind- ing theory of episodic memory: systems consolidation reconsidered. Nat. Rev. Neurosci. 2019;20:364–375. doi: 10.1038/s41583-019-0150-4. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 7. Kandel E.R., Schwartz J.H., Jessell T.M. 6th ed. McGraw-Hill; 2023. Principles of Neural Science. [ Google Scholar ] 8. Domjan M.P. 7th ed. Cengage Learning; 2014. The Principles of Learning and Behavior. [ Google Scholar ] 9. Gostolupce D., Iordanova M.D., Lay B.P.P. Mechanisms of higher-order learning in the amygdala. Behav. Brain Res. 2021;414 doi: 10.1016/j.bbr.2021.113435. [ DOI ] [ PubMed ] [ Google Scholar ] 10. Fam J., Chieng B., Westbrook R.F., Laurent V., Holmes N.M. Second-order fear conditioning involves formation of competing stimulus-danger and stimulus-safety as- sociations. Cerebr. Cortex. 2022;33:1843–1855. doi: 10.1093/cercor/bhac176. [ DOI ] [ PubMed ] [ Google Scholar ] 11. Gewirtz J.C., Davis M. Using pavlovian higher-order conditioning paradigms to investigate the neural substrates of emotional learning and memory. Learn. Mem. 2000;7:257–266. doi: 10.1101/lm.35200. [ DOI ] [ PubMed ] [ Google Scholar ] 12. Beckers T., Hermans D., Lange I., Luyten L., Scheveneels S., Vervliet B. Understanding clinical fear and anxiety through the lens of human fear conditioning. Nat. Rev. Psychol. 2023;2:233–245. doi: 10.1038/s44159-023-00156-1. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 13. Wessa M., Flor H. Failure of extinction of fear responses in posttraumatic stress disorder: Evidence from second-order conditioning. Am. J. Psychiatry. 2007;164:1684–1692. doi: 10.1176/appi.ajp.2007.07030525. [ DOI ] [ PubMed ] [ Google Scholar ] 14. Robinson S., Todd T.P., Pasternak A.R., Luikart B.W., Skelton P.D., Urban D.J., Bucci D.J. Chemogenetic silencing of neurons in retrosplenial cortex disrupts sen- sory preconditioning. J. Neurosci. 2014;34:10982–10988. doi: 10.1523/JNEUROSCI.1349-14.2014. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 15. Nicholson D.A., Freeman J.H., Jr. Lesions of the perirhinal cortex impair sen- sory preconditioning in rats. Behav. Brain Res. 2000;112:69–75. doi: 10.1016/s0166-4328(00)00168-6. [ DOI ] [ PubMed ] [ Google Scholar ] 16. Pinho J.S., Ramon-Duaso C., Manzanares-Sierra I., Busquets-Garcia A. Dorsal hippocampus mediates light–tone associations in male mice. eLife. 2025;14 doi: 10.7554/elife.105863. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 17. González-Parra J.A., Acciai V., Vidal-Palencia L., Canela-Grimau M., Busquets- Garcia A. Projecting neurons from the lateral entorhinal cortex to the basolateral amygdala mediate the encoding of incidental odor–taste associations. Proc. Natl. Acad. Sci. 2025;122 doi: 10.1073/pnas.2502127122. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 18. Holmes N.M., Parkes S.L., Killcross A.S., Westbrook R.F. The basolateral amygdala is critical for learning about neutral stimuli in the presence of danger, and the perirhinal cortex is critical in the absence of danger. J. Neurosci. 2013;33:13112–13125. doi: 10.1523/JNEUROSCI.1998-13.2013. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 19. Sharpe M.J., Chang C.Y., Liu M.A., Batchelor H.M., Mueller L.E., Jones J.L., Niv Y., Schoenbaum G. Dopamine transients are sufficient and necessary for acquisition of model-based associations. Nat. Neurosci. 2017;20:735–742. doi: 10.1038/nn.4538. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 20. Parkes S.L., Westbrook R.F. The basolateral amygdala is critical for the ac- quisition and extinction of associations between a neutral stimulus and a learned danger signal but not between two neutral stimuli. J. Neurosci. 2010;30:12608–12618. doi: 10.1523/JNEUROSCI.2949-10.2010. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 21. Sharpe M.J., Batchelor H.M., Mueller L.E., Gardner M.P.H., Schoenbaum G. Past experience shapes the neural circuits recruited for future learning. Nat. Neurosci. 2021;24:391–400. doi: 10.1038/s41593-020-00791-4. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Park S., Zhu A., Cao F., Palmiter R.D. Parabrachial Calca neurons mediate second-order conditioning. Nat. Commun. 2024;15 doi: 10.1038/s41467-024-53977-w. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 23. Dragunow M., Faull R. The use of c-fos as a metabolic marker in neuronal pathway tracing. J. Neurosci. Methods. 1989;29:261–265. doi: 10.1016/0165-0270(89)90150-7. [ DOI ] [ PubMed ] [ Google Scholar ] 24. Yang D., Wang Y., Qi T., Zhang X., Shen L., Ma J., Pang Z., Lal N.K., McClatchy D.B., Seradj S.H., et al. Phosphorylation of pyruvate dehydrogenase inversely associates with neuronal activity. Neuron. 2024;112:959–971.e8. doi: 10.1016/j.neuron.2023.12.015. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 25. Garden G.A., Campbell B.M. Glial biomarkers in human central nervous sys- tem disease. Glia. 2016;64:1755–1771. doi: 10.1002/glia.22998. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 26. Masuda T., Sankowski R., Staszewski O., Prinz M. Microglia heterogeneity in the single-cell era. Cell Rep. 2020;30:1271–1281. doi: 10.1016/j.celrep.2020.01.010. [ DOI ] [ PubMed ] [ Google Scholar ] 27. Shariff A., Kangas J., Coelho L.P., Quinn S., Murphy R.F. Automated image analysis for high-content screening and analysis. SLAS Discovery. 2010;15:726–734. doi: 10.1177/1087057110370894. [ DOI ] [ PubMed ] [ Google Scholar ] 28. Stringer C., Wang T., Michaelos M., Pachitariu M. Cellpose: a generalist algorithm for cellular segmentation. Nat. Methods. 2021;18:100–106. doi: 10.1038/s41592-020-01018-x. [ DOI ] [ PubMed ] [ Google Scholar ] 29. DeNardo L.A., Liu C.D., Allen W.E., Adams E.L., Friedmann D., Fu L., Guenthner C.J., Tessier-Lavigne M., Luo L. Temporal evolution of cortical ensembles promoting remote memory retrieval. Nat. Neurosci. 2019;22:460–469. doi: 10.1038/s41593-018-0318-7. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 30. Canela-Grimau M., Pinho J.S., Busquets-Garcia A. Profiling mouse behavior with computational tools to assess age-dependent differences in associative learning. Cell Rep. Methods. 2025;5 doi: 10.1016/j.crmeth.2025.101144. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 31. Mathis A., Mamidanna P., Cury K.M., Abe T., Murthy V.N., Mathis M.W., Bethge M. Deeplabcut: Markerless pose estimation of user-defined body parts with deep learning. Nat. Neurosci. 2018;21:1281–1289. doi: 10.1038/s41593-018-0209-y. [ DOI ] [ PubMed ] [ Google Scholar ] 32. Weinreb C., Pearl J.E., Lin S., Osman M.A.M., Zhang L., Annapragada S., Conlin E., Hoffmann R., Makowska S., Gillis W.F., et al. Keypoint-MoSeq: Parsing behavior by linking point tracking to pose dynamics. Nat. Methods. 2024;21:1329–1339. doi: 10.1038/s41592-024-02318-2. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 33. Bordes J., Miranda L., Reinhardt M., Narayan S., Hartmann J., Newman E.L., Brix L.M., van Doeselaar L., Engelhardt C., Dillmann L., et al. Automatically annotated motion tracking identifies a distinct social behavioral profile following chronic social defeat stress. Nat. Commun. 2023;14 doi: 10.1038/s41467-023-40040-3. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 34. Harris C.R., Millman K.J., van der Walt S.J., Gommers R., Virtanen P., Cournapeau D., Wieser E., Taylor J., Berg S., Smith N.J., et al. Array programming with NumPy. Nature. 2020;585:357–362. doi: 10.1038/s41586-020-2649-2. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 35. McKinney W. Vol. 445. 2010. Data structures for statistical computing in python; pp. 51–56. (Proceedings of the 9th Python in Science Conference). [ Google Scholar ] 36. Van der Walt S., Schönberger J.L., Nunez-Iglesias J., Boulogne F., Warner J.D., Yager N., Gouillart E., Yu T., scikit-image contributors scikit-image: image processing in python. PeerJ. 2014;2 doi: 10.7717/peerj.453. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 37. Pedregosa F., Varoquaux G., Gramfort A., Michel V., Thirion B., Grisel O., Blondel M., Prettenhofer P., Weiss R., Dubourg V., et al. Scikit-learn: Machine learning in python. J. Mach. Learn. Res. 2011;12:2825–2830. [ Google Scholar ] 38. Hunter J.D. Matplotlib: A 2d graphics environment. Comput. Sci. Eng. 2007;9:90–95. [ Google Scholar ] 39. Parkes S.L., Westbrook R.F. Role of the basolateral amygdala and NMDA receptors in higher-order conditioned fear. Revneuro. 2011;22:317–333. doi: 10.1515/RNS.2011.025. [ DOI ] [ PubMed ] [ Google Scholar ] 40. Yang Y., Wang J.Z. From structure to behavior in basolateral amygdala- hippocampus circuits. Front. Neural Circuits. 2017;11 doi: 10.3389/fncir.2017.00086. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 41. Bonaventura J., Eldridge M.A.G., Hu F., Gomez J.L., Sanchez-Soto M., Abramyan A.M., Lam S., Boehm M.A., Ruiz C., Farrell M.R., et al. High-potency ligands for dreadd imaging and activation in rodents and monkeys. Nat. Commun. 2019;10 doi: 10.1038/s41467-019-12236-z. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 42. Otse E.J., Obunadike G.N., Abubakar A. Linear regression approach to solv- ing multicollinearity and overfitting in predictive analysis. Journal of Science Research and Reviews. 2025;2:108–117. [ Google Scholar ] 43. Trott J.M., Hoffman A.N., Zhuravka I., Fanselow M.S. Conditional and uncon- ditional components of aversively motivated freezing, flight and darting in mice. eLife. 2022;11 doi: 10.7554/elife.75663. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 44. Totty M.S., Warren N., Huddleston I., Ramanathan K.R., Ressler R.L., Oleksiak C.R., Maren S. Behavioral and brain mechanisms mediating conditioned flight be- havior in rats. Sci. Rep. 2021;11 doi: 10.1038/s41598-021-87559-3. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 45. Whissell P.D., Tohyama S., Martin L.J. The use of dreadds to deconstruct behavior. Front. Genet. 2016;7 doi: 10.3389/fgene.2016.00070. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 46. Mitchell J.R., Vincelette L., Tuberman S., Sheppard V., Bergeron E., Calitri R., Clark R., Cody C., Kannan A., Keith J., et al. Behavioral and neural correlates of diverse conditioned fear responses in male and female rats. bioRxiv. 2024 doi: 10.1101/2024.08.20.608817. Preprint at. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 47. Chen W., Xiao Y., Chen J., Liu J., Shao J., Li T., Zhu Y., Ma J., Gao Y., Wang J., et al. Sex-related pharmacokinetic differences and mechanisms of metapristone (ru486 metabolite) Sci. Rep. 2017;7 doi: 10.1038/s41598-017-17225-0. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 48. Parel S.T., Peña C.J. Genome-wide signatures of early-life stress: Influence of sex. Biol. Psychiatry. 2022;91:36–42. doi: 10.1016/j.biopsych.2020.12.010. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 49. Swilley C., Lin Y., Zheng Y., Xu X., Liu M., Jarome T., Hodes G.E., Xie H. Sex linked behavioral and hippocampal transcriptomic changes in mice with cell-type spe- cific egr1 loss. Front. Neurosci. 2023;17 doi: 10.3389/fnins.2023.1240209. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 50. Sheppard P.A.S., Choleris E., Galea L.A.M. Structural plasticity of the hip- pocampus in response to estrogens in female rodents. Mol. Brain. 2019;12 doi: 10.1186/s13041-019-0442-7. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 51. Wolcott N.S., Redman W.T., Karpinska M., Jacobs E.G., Goard M.J. The estrous cycle modulates hippocampal spine dynamics, dendritic processing, and spatial coding. Neuron. 2025;113:2297–2309.e7. doi: 10.1016/j.neuron.2025.04.014. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 52. Jahn H.M., Kasakow C.V., Helfer A., Michely J., Verkhratsky A., Maurer H.H., Scheller A., Kirchhoff F. Refined protocols of tamoxifen injection for inducible dna re- combination in mouse astroglia. Sci. Rep. 2018;8 doi: 10.1038/s41598-018-24085-9. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 53. Penzo M.A., Gao C. The paraventricular nucleus of the thalamus: an integra- tive node underlying homeostatic behavior. Trends Neurosci. 2021;44:538–549. doi: 10.1016/j.tins.2021.03.001. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 54. Kirouac G.J. Placing the paraventricular nucleus of the thalamus within the brain circuits that control behavior. Neurosci. Biobehav. Rev. 2015;56:315–329. doi: 10.1016/j.neubiorev.2015.08.005. [ DOI ] [ PubMed ] [ Google Scholar ] 55. Kirouac G.J. The paraventricular nucleus of the thalamus as an integrating and relay node in the brain anxiety network. Front. Behav. Neurosci. 2021;15 doi: 10.3389/fnbeh.2021.627633. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 56. Kim E.J., Horovitz O., Pellman B.A., Tan L.M., Li Q., Richter-Levin G., Kim J.J. Dorsal periaqueductal gray-amygdala pathway conveys both innate and learned fear responses in rats. Proc. Natl. Acad. Sci. USA. 2013;110:14795–14800. doi: 10.1073/pnas.1310845110. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 57. Zhang H., Zhu Z., Ma W.X., Kong L.X., Yuan P.C., Bu L.F., Han J., Huang Z.L., Wang Y.Q. The contribution of periaqueductal gray in the regulation of physiological and pathological behaviors. Front. Neurosci. 2024;18 doi: 10.3389/fnins.2024.1380171. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 58. Silva B.A., Burns A.M., Gräff J. A cfos activation map of remote fear memory attenuation. Psychopharmacology. 2018;236:369–381. doi: 10.1007/s00213-018-5000-y. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 59. Roy D.S., Park Y.G., Kim M.E., Zhang Y., Ogawa S.K., DiNapoli N., Gu X., Cho J.H., Choi H., Kamentsky L., et al. Brain-wide mapping reveals that engrams for a single memory are distributed across multiple brain regions. Nat. Commun. 2022;13 doi: 10.1038/s41467-022-29384-4. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 60. Clawson B.C., Pickup E.J., Ensing A., Geneseo L., Shaver J., Gonzalez-Amoretti J., Zhao M., York A.K., Kuhn F.R., Swift K., et al. Causal role for sleep-dependent reactivation of learning-activated sensory en- sembles for fear memory consolidation. Nat. Commun. 2021;12 doi: 10.1038/s41467-021-21471-2. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 61. Besnard A., Sahay A. Adult hippocampal neurogenesis, fear generalization, and stress. Neuropsychopharmacology. 2016;41:24–44. doi: 10.1038/npp.2015.167. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 62. Lee H., Wang Z., Tillekeratne A., Lukish N., Puliyadi V., Zeger S., Gallagher M., Knierim J.J. Loss of functional heterogeneity along the ca3 transverse axis in aging. Curr. Biol. 2022;32:2681–2693.e4. doi: 10.1016/j.cub.2022.04.077. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 63. Rolls E.T. The mechanisms for pattern completion and pattern separation in the hippocampus. Front. Syst. Neurosci. 2013;7 doi: 10.3389/fnsys.2013.00074. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 64. Rescorla R.A. Simultaneous second-order conditioning produces s-s learning in conditioned suppression. J. Exp. Psychol. Anim. Behav. Process. 1982;8:23–32. [ PubMed ] [ Google Scholar ] 65. Busquets-Garcia A., Oliveira da Cruz J.F., Terral G., Pagano Zottola A.C., Soria-Gómez E., Contini A., Martin H., Redon B., Varilh M., Ioannidou C., et al. Hippocampal CB1 receptors control incidental associations. Neuron. 2018;99:1247–1259.e7. doi: 10.1016/j.neuron.2018.08.014. [ DOI ] [ PubMed ] [ Google Scholar ] 66. Schmajuk N.A., DiCarlo J.J. Stimulus configuration, classical conditioning, and hippocampal function. Psychol. Rev. 1992;99:268–305. doi: 10.1037/0033-295x.99.2.268. [ DOI ] [ PubMed ] [ Google Scholar ] 67. Klee J.L., Souza B.C., Battaglia F.P. Learning differentially shapes prefrontal and hippocampal activity during classical conditioning. eLife. 2021;10 doi: 10.7554/eLife.65456. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 68. Berger T.W., Alger B., Thompson R.F. Neuronal substrate of classical condi- tioning in the hippocampus. Science. 1976;192:483–485. doi: 10.1126/science.1257783. [ DOI ] [ PubMed ] [ Google Scholar ] 69. Bankhead P., Loughrey M.B., Fernández J.A., Dombrowski Y., McArt D.G., Dunne P.D., McQuaid S., Gray R.T., Murray L.J., Coleman H.G., et al. Qupath: Open source software for digital pathology image analysis. Sci. Rep. 2017;7 doi: 10.1038/s41598-017-17204-5. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 70. Stirling D.R., Swain-Bowden M.J., Lucas A.M., Carpenter A.E., Cimini B.A., Good- man A. Cellprofiler 4: improvements in speed, utility and usability. BMC Bioinf. 2021;22 doi: 10.1186/s12859-021-04344-9. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 71. Berg S., Kutra D., Kroeger T., Straehle C.N., Kausler B.X., Haubold C., Schiegg M., Ales J., Beier T., Rudy M., et al. ilastik: interactive machine learning for (bio)image analysis. Nat. Methods. 2019;16:1226–1232. doi: 10.1038/s41592-019-0582-9. [ DOI ] [ PubMed ] [ Google Scholar ] 72. Schneider C.A., Rasband W.S., Eliceiri K.W. Nih image to imagej: 25 years of image analysis. Nat. Methods. 2012;9:671–675. doi: 10.1038/nmeth.2089. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 73. Gage G.J., Kipke D.R., Shain W. Whole animal perfusion fixation for rodents. J. Vis. Exp. 2012;65:3564. doi: 10.3791/3564. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 74. Perrin-Terrin A.S., Jeton F., Pichon A., Frugière A., Richalet J.P., Bodineau L., Voituron N. The c-FOS protein immunohistological detection: A useful tool as a marker of central pathways involved in specific physiological responses In Vivo and Ex Vivo. J. Vis. Exp. 2016;110:53613. doi: 10.3791/53613. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 75. Li J., Bai Y., Liang Y., Zhang Y., Zhao Q., Ge J., Li D., Zhu Y., Cai G., Tao H., et al. Parvalbumin neurons in zona incerta regulate itch in mice. Front. Mol. Neurosci. 2022;15 doi: 10.3389/fnmol.2022.843754. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 76. Lindeberg T. Scale selection properties of generalized scale-space interest point detectors. J. Math. Imaging Vis. 2012;46:177–210. [ Google Scholar ] 77. Otsu N. A threshold selection method from gray-level histograms. IEEE Trans. Syst. Man Cybern. 1979;9:62–66. [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Document S1. Table S1 mmc1.pdf (80.2KB, pdf) Data Availability Statement • All data generated in this study are available in a Zenodo repository under the following https://doi.org/10.5281/zenodo.19075438 . • All original code of CellRake has been deposited at github.com/marccanela/cellrake and it is publicly available under the https://doi.org/10.5281/zenodo.19002677 . • Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Articles from iScience are provided here courtesy of Elsevier ACTIONS View on publisher site PDF (18.8 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top

Related documents

• Image 3_Ric-8A controls Cited2 subcellular localization and transcriptional programs during neural crest development.tif#367158
• Structure and organization of AMPA receptor-TARP complexes in the mammalian cerebellum#236585
• Hierarchically and Temporally Distinct Ca(2+) Signaling Across Thalamic Nuclei During Nociceptive Stimulation.#636515
• Processing of sub- and supra-second intervals in the primate brain results from the calibration of neuronal oscillators via sensory, motor, and feedback processes#738295
• Ric-8A controls Cited2 subcellular localization and transcriptional programs during neural crest development.#415414