Abstract
The hippocampus supports a multiplicity of functions, with the dorsal region contributing to spatial representations and memory and the ventral hippocampus (vH) being primarily involved in emotional processing. While spatial encoding has been extensively investigated, how the vH activity is tuned to emotional states, e.g., to different anxiety levels, is not well understood. We developed an adjustable linear track maze for male mice with which we could induce a scaling of behavioral anxiety levels within the same spatial environment. Using in vivo single-unit recordings, optogenetic manipulations, and population-level analysis, we examined the changes and causal effects of vH activity at different anxiety levels. We found that anxiogenic experiences activated the vH and that this activity scaled with increasing anxiety levels. We identified two processes that contributed to this scaling of anxiety-related activity: increased tuning and successive remapping of neurons to the anxiogenic compartment. Moreover, optogenetic inhibition of the vH reduced anxiety across different levels, while anxiety-related activity scaling could be decoded using a linear classifier. Collectively, our findings position the vH as a critical limbic region that functions as an “anxiometer” by scaling its activity based on perceived anxiety levels. Our discoveries go beyond the traditional theory of cognitive maps in the hippocampus underlying spatial navigation and memory, by identifying hippocampal mechanisms selectively regulating anxiety.
- anxiety
- anxiety levels
- rate remapping
- recruitment of neurons
- scaling of neural activity
- ventral hippocampus
Significance Statement
This study reveals how the ventral hippocampus (vH) functions as an “anxiometer,” tuning its activity to different anxiety levels. Using an adjustable linear track maze (aLTM) for mice, we demonstrated that vH activity scales with increased anxiety. By recording single-neuron activity and performing optogenetic manipulation of vH during the aLTM task, we identified key neuronal mechanisms for neuronal scaling during anxiety. Additionally, a linear classifier was used to highlight anxiety-related activity scaling. Our findings advance the understanding of hippocampal function beyond spatial navigation and memory, offering new insights into how the brain regulates anxiety at the neuronal level.
Introduction
Anxiety is an evolutionary conserved mental and emotional state, underscoring its crucial role in the survival of organisms facing perilous situations (Sylvers et al., 2011; Koskinen and Hovatta, 2023). Anxiety is characterized by the anticipation of potentially harmful events leading to an attentional bias toward threatening stimuli (Baldwin et al., 2010; Miceli and Castelfranchi, 2015; Parfitt et al., 2017). As anxiety prepares an animal or a human to face environmental challenges, the level of threat has to be carefully interpreted to support adaptive behavior. An altered perception of negative emotions (Feldborg et al., 2021) might contribute to persistent fear and/or chronic anxiety disorders which pose significant personal and societal burdens (Di Luca et al., 2011; Kavelaars et al., 2023), affecting >300 million people worldwide (Javaid et al., 2023). As treatments are available but generic and partially ineffective (Bandelow et al., 2017; Ansara, 2020; Garakani et al., 2020), there is an urgent need to understand the neurobiological basis of anxiety in more details. Previous studies have implicated areas of the limbic system including the amygdala, prefrontal cortex, and hippocampus in the generation of emotions (Shin and Liberzon, 2010). However, the precise brain region where variations in anxiety levels, and not just anxiety per se, are represented and computed remains elusive. We hypothesized that the hippocampus could play an important role in processing different anxiety levels as it is a key structure involved in learning and compares previously stored information (memories) with the current contextual information to generate cognitive maps (Katz et al., 2007; Duncan et al., 2012; Nomoto et al., 2022). Along the septotemporal or long axis of the hippocampus, differences in synaptic connectivity and molecular expression have been found, inferring functional subdivisions within the hippocampus (Cenquizca and Swanson, 2007; Fanselow and Dong, 2010). Studies investigating neural activity patterns and selective lesions of the ventral hippocampus (vH) provide evidence, from rodents to its homolog in humans (i.e., the anterior hippocampus), on the role of the vH in the processing of threatening stimuli (Bannerman et al., 2003; Bach et al., 2014; Ito and Lee, 2016; Bach et al., 2019). In rodents, anxiety has been frequently studied in mazes such as the elevated plus maze (EPM) which elicit innate anxiety of height and openness as the animals explore the open compartments of the mazes (Rodgers et al., 1997; Carobrez and Bertoglio, 2005; Cryan and Holmes, 2005; Li et al., 2024). vH anxiety neurons, i.e., pyramidal neurons that are activated in the open arms of the EPM, have been linked to anxiety-related behavior through the routing of anxiety-related information to the prefrontal cortex or the lateral hypothalamus (Adhikari et al., 2010; Ciocchi et al., 2015; Jimenez et al., 2018). If and how the activity level of vH neurons relates to the level of perceived anxiety on the EPM is unclear as the precise quantification of emotional states and incentives of animals on an EPM is difficult to assess due to the free exploration and trial-free nature of the task. Furthermore, the specific spatial compartments are inherently linked to the emotional states. Because of such spatial embedding of nonspatial information in many such tasks, there is an ongoing discussion on how much nonspatial information, such as emotions, are actually defined by a spatial framework or can be encoded independently in the hippocampus (Lisman et al., 2017; O’Keefe and Krupic, 2021).
To address the nature of anxiety coding in the vH, we developed a linear track maze (Malagon-Vina et al., 2023) that could be gradually adjusted to elicit different anxiety levels to an open and elevated compartment within the same spatial environment. With this fully automated adjustable linear track maze (aLTM), we could overcome some of the rigidities and difficulties of the EPM allowing us to define trials and achieve a scaling of anxiety levels while recording and manipulating the activity of neurons in the vH.
Materials and Methods
Subjects
Male C57BL6/J mice (N = 28) aged 4–6 months (Janvier Labs) were group housed with ad libitum access to food and water in a temperature-controlled room on a 12 h light/dark cycle. Following surgery, mice recovered for at least 21 d. Behavioral experiments were conducted in the light phase, and mice were food deprived to reach 90% and not <85% of the baseline weight (i.e., average weight of 3 consecutive days prior to the experiment day). All experimental protocols adhered to the guidelines set forth by the Animal Welfare Office at the University of Bern and received approval from the Veterinary Office of the Canton of Bern.
Surgical procedures
Mice were anesthetized with isoflurane (Attane, Provet, induction 4% in oxygen at a flow rate of 1 L/min in anesthetic box, maintenance of the anesthesia based on 1.5–2% of isoflurane in oxygen at a flow rate of 0.8 L/min during the entire surgery), and the body temperature was maintained at 37°C with a temperature controller (Harvard Apparatus). Analgesia was administered upon induction and if needed for 3 d postoperation (carprofen, 2–5 mg/kg, s.c.). The incision site was infiltrated with lidocaine (2%, Sintetica). Implants were secured to the skull using light-cured dental adhesive (Kerr Dental, OptiBond Universal) and dental cement (Ivoclar, Tetric EvoFlow). After surgery, mice were returned to their home cages for recovery and kept warm for 3 d by warming the cages through a heating pad (Medisana, HP625).
Virus injection and fiber implantation
For optogenetics experiments, mice were infused bilaterally in the vH with the following coordinates: AP, −3.08 mm; ML, ±3 mm; and DV, −3.8 mm. The viral solution (250 nl) was delivered via a glass micropipette (30–40 µm tip) attached by tubing to a Picospritzer III microinjection system (Parker Hannifin) at a rate of 25 nl/min. Bilateral infusion of either ssAAV5-CaMKIIα-eArchT3.0-EYFP or ssAAV5-CaMKIIα-EYFP (ETH Zurich Viral Vector Facility) was immediately followed by optic fibers implantation (200 µm core, 0.37 numerical aperture, Thorlabs) above the vH at the following coordinates: AP, −3.08 mm; ML, ±3 mm; and DV, −3.75 mm (Fig. 2A). The optic fiber implants were fixed to the skull with stainless steel screws, resin-based dental cement (A1 Tetric EvoFlow, Ivoclar Vivadent) and light-cured dental adhesive (Kerr Dental, OptiBond Universal). To ensure sufficient expression of the viral construct, we allowed a minimum of 5 weeks for the mice to recover before conducting the behavioral experiments.
Optogenetic manipulation
For the optogenetic inhibition of calcium-/calmodulin-dependent protein kinase II subunit α (CaMKIIα) expressing vH neurons during linear track maze navigation, a continuous light pulse was delivered when the mouse transitioned from the closed to the open part of the maze (Fig. 2C) via a patch cord coupled to a laser beam (Cobolt 06-DPL 561 nm, HÜBNER Photonics). The laser was triggered automatically when the mouse's body center was detected inside the second half of the transition and was turned off as the mouse's body center was detected in the openable compartment using the ANY-maze animal tracking software.
Silicon probe recordings
For silicon probe experiments, mice were injected with ssAAV5-CaMKIIα-eArchT3.0-EYFP or with ssAAV5-CaMKIIα-EYFP (ETH Zurich Viral Vector Facility) above the vH with the following coordinates, AP, −3.08 mm; ML, ±3 mm; and DV, −3.75 mm, and as described for the optogenetic experiments. At the same time, the mice were implanted with metal headplates on the skull above the cerebellum/left hemisphere. To ensure sufficient expression of the viral construct, we allowed a minimum of 5 weeks for the mice to recover before conducting further experiments. Then, they were habituated to head fixation on a treadmill over several days. The day before recordings, new craniotomies were done over the vH. Recordings were performed with 2 (shank) × 32 channels silicon probes with one optic fiber per shank above the recording sites (Takahashi 64 optoelectrodes, NeuroNexus). Neuronal units were recorded at a depth of ∼4 mm from the surface. During recordings, laser stimulations (Cobolt 06-DPL 561 nm, HÜBNER Photonics) were applied for 2 s for 20 times with 12 s break between laser stimulations. Neurons were clustered as described below, Spike detection and unit classification, and responses to time of laser application quantified in 1 s bins.
Surgery for in vivo electrophysiological recordings
Mice were implanted with a custom-made microdrive (Axona) containing eight independently moveable tetrodes made of four gold-plated (nanoZ, Multi Channel Systems) twisted tungsten microwires (12.7 µm inner diameter, California Fine Wire Company, impedances <150 kΩ). Microdrives were lowered to the vH. Implants were fixed to the skull with five miniature stainless steel screws (00–96 × 1/16, Bilaney Consultants) with two of them placed above the cerebellum and connected to the electrode interface board to serve as ground for the electrophysiological recordings.
aLTM
A home-made aLTM (Malagon-Vina et al., 2023; L80 cm × W10-7 cm × H12 cm; Movie 1) was built in order to create six different anxiety levels in mice: (1) a “no anxiety” (NA; Fig. 1A, top left) configuration in which the entire maze was surrounded by walls and the mice had no visual insight into the height of the maze; (2) a “very low anxiety” (VLA; Fig. 1A, top right) configuration in which the openable second half of the maze was opened and the mice could perceive the 20 cm height of the maze; (3) a “low anxiety” (LA; Fig. 1A, middle left) configuration in which the openable second half of the maze was opened and the mice could perceive the 70 cm height of the maze; (4) a “moderate anxiety” (MA; Fig. 1A, middle right) configuration where the maze was further elevated by a minilift (ESM100, Dahlgaard) to 120 cm from the ground; (5) the “high anxiety” (HA; Fig. 1A, bottom left) configuration in which the maze was still at 120 cm height but the open part was changed to a narrower width from 10 to 7 cm; and (6) the “very high anxiety” (VHA; Fig. 1A, bottom right) configuration with the maze further elevated to 170 cm with 7 cm width of the open part. Five milligram reward pellets were placed in distal locations on the openable part after each trial to motivate the mice. A custom-made automatic door (Stepper driver ROB-12859, SparkFun Electronics; Cypress Semiconductor; switch snap, C&K Switches) in the closed part was built in order to break stereotypic behaviors and create a safe environment for resting in-between trials without any interference from the experimenter (Fig. 1A).
Different configurations of the aLTM induce a scaling in anxiety behavior. A, aLTM's anxiety configurations. The yellow rectangle in the maze represents an automated door that closed after each trial for a random amount of time (between 10 and 40 s) to avoid stereotypic behavior. “R” indicates the reward delivery in the openable/open part of the aLTM. B, The number of trials (left) and percentage of time in the open/openable compartment (right) across anxiety configurations indicate gradually increasing behavioral anxiety from NA to VHA. Data represented as linear regressions with 95% confidence intervals calculated based on all behavioral sessions (n = 64 sessions from N = 16 mice). Mean ± SEM of all behavioral sessions for each anxiety configuration are also indicated (Extended Data Fig. 1-1).
Figure 1-1
Normalized anxiety behavioral read-outs. Number of trials normalized to trials taken in NA (left) and percentage of time in the openable compartment normalized to NA (right) across anxiety configurations indicate gradually increasing behavioral anxiety from NA to VHA. Simple linear regression, number of trials: F = 66.3, R2 = 0.164, p < 0.001; percentage of time F = 38.3, R2 = 0.102, p < 0.001; Spearman correlation: number of trials: r = -0.52, p < 0.001; percentage of time r = -0.36, p < 0.001. Data represented as linear regressions with 95% confidence intervals calculated based on all behavioral sessions (n = 64 sessions from N = 16 mice). Means ± SEM of all behavioral sessions for each anxiety configuration are also indicated. Download Figure 1-1, TIF file.
Each different anxiety configuration was presented twice with more anxiogenic or less anxiogenic configuration orders at 50 lux lighting conditions. The order of configurations was not conducted only in a linearly increasing manner to prevent learning or sensitization effects. In all experiments, each anxiety configuration was presented with only one anxiety-level difference from the previous or next configuration. For instance, the LA configuration was always preceded by either the MA or VLA configuration and was always followed by either the VLA or MA configuration. This design ensures that the observed neuronal activity can be more confidently attributed to the specific anxiogenic experience of each configuration, rather than to learning or sensitization effects. Scrambled configuration experiments were performed in the following order: LA–VLA–HA–NA–VHA–MA (Extended Data Fig. 1-1). For each experiment, the mouse was first placed into the closed compartment with the door closed for 300 s for habituation and rest. Then, the door was opened, and the mouse was free to initiate a trial or not. A trial is considered for every trajectory that the mouse performed going from the first half of the maze to the second half and came back to the first half. If the mouse explored the second half of the maze (the openable compartment) and returned to the closed part, the door would be closed for a random time of 10–40 s to break stereotypic locomotion and allow the experimenter to place the reward in the anxiogenic part of the maze. For each configuration, the duration that the mouse could run trials lasted 500 s. In a 2 min break between each configuration, the mice rested in the closed compartment with the door closed. In the novel configuration, a control NA session would be compared with a “novel” configuration (300 s rest + 500 s with trials) where the walls in the openable part would be replaced by walls decorated with stripes of a different color (yellow stripes) and different texture (tape; Fig. 5A, right).
For anxiety testing, every mouse had been challenged up to four times with an interval of at least 5 d between each experiment. Videotaping was carried out from above the maze with the behavior system ANY-maze (www.any-maze.com). Lighting was maintained at 50 lux in each configuration thanks to a dimmable light.
Histology
Mice were overdosed with a 5% ketamine and 2.5% xylazine solution (0.02 ml/g) and transcardially perfused with phosphate-buffered saline (0.1 M PBS, Roche), pH 7.4, followed by 4% paraformaldehyde (PFA, ROTI Histofix 4%, Carl Roth). For electrophysiology experiments, mice were anesthetized with isoflurane (as stated above) prior to perfusion, and electrolytic lesions were made at the tip of each recording tetrode by applying five times 30 µA for 2 s. Brains were kept in 4% PFA for 24 h postfixation. After a rinse in PBS for at least 24 h at 4°C, brains were sectioned coronally at a thickness of 50 µm using a vibratome (VT1000 S, Leica Microsystems) and preserved in 0.5% sodium azide PBS at 4°C. Lesions were located, and only tetrodes positioned specifically in the CA1 region of the vH were included for further analysis (Extended Data Fig. 3-1).
Immunohistochemistry
Free floating brain slices were blocked with 5% normal donkey serum (Abcam) in 0.1% Triton X-100 in PBS (PBS-T) for 30 min on a shaker at room temperature (25°C). Sections were incubated for 72 h at 4°C with an anti-GFP antibody (1:100, Abcam, ab13970) washed with PBS-T for 5 min (three times) and incubated for 2 h at room with secondary antibody (Alexa Fluor 488 1:100, Jackson ImmunoResearch Laboratories, 703545145). Slices were mounted onto slides (Thermo Fisher Scientific) and coverslipped with aqueous mounting medium (Aqua-Poly/Mount, PolyScience). Sections were imaged on a stereomicroscope (Leica M205 FCA, Leica) mounted with a monochrome camera (Leica DFC345 FX, Leica Microsystems) and a UV light source (X-CITE, Lumen Dynamics) or with a confocal microscope (LSM 880, Carl Zeiss) and stored at 4°C.
Behavioral quantification
For behavioral assessments, we measured the number of trials completed by the mice and the percentage of time spent within the openable section of the maze per presented configuration. These measurements were also normalized relative to the total number of trials completed and the duration of time spent in the openable maze section during the NA configuration. This normalization accounted for variations in locomotion and exploration among individual mice (Extended Data Fig. 1-1).
Behavior with electrophysiological recordings
After 21 d postsurgery, the mice were familiarized to the experiment room and acclimated to the handling procedures, which included cable tethering to the recording system. On the following days, tetrodes were lowered in a stepwise fashion while monitoring hippocampal electrophysiological hallmarks (i.e., sharp-wave–ripple and theta oscillations amplitude) to detect multiunit spiking activity in the stratum pyramidale of the vH. Electrophysiological signals were synchronized to behavior via TTL outputs sent from ANY-maze (at 30 Hz). Neurophysiological data were collected and analyzed from levels NA, VLA, LA, MA, and HA but not VHA due to the low sampling in the open part of this anxiety configuration. Our recordings were specifically from the CA1 region of the vH.
Spike detection and unit classification
The extracellular electrical signals from the tetrodes were amplified, filtered, and digitized with a headstage (Intan RHD amplifier, Intan Technologies). The signals were acquired using an Intan RHD2000 evaluation board (Intan Technologies) at a sampling rate of 20 kHz. Neuronal spikes were extracted off-line by either using Kilosort2 (Pachitariu et al., 2016; Stringer et al., 2019) or by detecting signal amplitudes five standard deviations above the root mean square of the digitally filtered signal (0.8–5 kHz) over 0.2 ms sliding windows. A principal component analysis was implemented to extract the first three components of spike waveforms of each tetrode wire (Harris et al., 2000; Henze et al., 2000). Spike waveforms were automatically clustered using KlustaKwik (http://klustakwik.sourceforge.net/). Individual units were manually refined in Klusters (Hazan et al., 2006). The complete and well-sorted single–unit dataset was included without applying any exclusion criteria.
Tuning and recruitment quantification
A neuron was classified as anxiety-tuned if its firing rate in the open part of the aLTM was significantly correlated (Spearman correlation analysis) with the linearly increasing anxiety configurations. Neuronal recruitment was assessed by comparing the neuron's firing rate in the open part of the maze to its firing rate in the closed part. The first configuration in which the neuron showed a statistically significant increased firing rate in the open part compared with the closed part [repeated-measure (RM) one-way ANOVA with Dunnett's multiple-comparison test or Friedman test with Dunn's multiple-comparison test] was used to classify that neuron as being recruited in that specific configuration.
Anxiety-level classification
For each recording session, we trained a linear classifier to identify the anxiety level based on the recorded neural activity. Since different neurons are detected during different recording sessions, classifier training and testing were performed separately for each session instead of combining data from different sessions and individuals.
For each recording session, we used the activity from all identified units (see above, Spike detection and unit classification). Data were preprocessed as follows. First, firing rates were obtained for each identified units in 200 ms bins using the detected spikes. For each neuron, z-scored neuronal activities were calculated to allow unbiased comparisons of firing rates. Each neuron was independently z-scored based on its activity in the entire recording session. Then, samples were defined by binning the activity of all neurons across time into nonoverlapping snippets of length N = 10 time steps (i.e., 2 s of recordings for 200 ms bins). Each sample is thus an m * N matrix, where m is the number of recorded neurons in the session, labeled by its corresponding anxiety level.
The goal of the classifier is to predict the anxiety level (labels) associated with these samples. Classification was performed using a linear support vector machine (SVM) trained via stochastic gradient descent. Implementation of this classifier was performed using the SGDClassifier function from the scikit-learn library (Pedregosa et al., 2011) with default parameters and with regularization parameter α = 10−7.
The training dataset for each session was balanced to ensure all five configurations have the same number of samples. This was achieved through oversampling. Oversampling was obtained by randomly duplicating samples from configurations with fewer samples than the most populated configuration.
For each experiment, a confusion matrix is obtained by training and testing the SVM via leave-one-out cross–validation (i.e., training the classifier on all samples except one, testing it on this held-out sample, and reporting the accuracy over all folds).
Next, for each session, the analysis was repeated M = 20 times after independently and randomly shuffling the labels. Results were then averaged over all M shuffles to obtain a chance-level confusion matrix for each session. Finally, the confusion matrices obtained for all sessions were compared with the chance-level confusion matrices: an ordinary one-way ANOVA was performed for each entry of the matrix to identify classification levels that are higher in the nonshuffled condition than in the shuffled condition. Results were Bonferroni’s corrected to account for the multiple testing (five significance tests are performed per ground-truth condition).
Calculations were performed on UBELIX (http://www.id.unibe.ch/hpc), the high-performance computing cluster at the University of Bern.
Statistical analysis
Analyses were conducted in GraphPad PRISM and MATLAB using custom scripts. All datasets were tested for normality using the D'Agostino and Pearson’s test, and if found to be normally distributed, we used parametric tests or otherwise the equivalent nonparametric tests. In case of multiple comparisons, significance was evaluated with a one-way or two-way RM ANOVA or Friedman's test followed by Dunn's, Dunnett's, or uncorrected Fisher's LSD post hoc tests. To evaluate scaling and linearity induced by the different anxiogenic configurations, we applied Spearman correlation coefficient and simple linear regression analyses. In order to perform these analyses, we assumed a mapping from anxiety conditions to anxiety levels (NA, 1; VLA, 2; …; VHA, 6). This (a priori arbitrary) mapping is justified a posteriori by the almost linear relationship between the number of trials (or time in the openable/open part) as a function of the anxiety levels (Fig. 1B).
All plots show the best-fit line with its R2 and p values and the 95% confidence bands. Both the Spearman correlation and the simple linear regression calculations were performed on the entire dataset of behavioral or electrophysiological individual values. However, to improve visualization, plots also present the mean ± SEM.
Data availability
The datasets generated during the current study are available from the corresponding authors upon request. MATLAB and Python files are available in the following GitHub repository: https://github.com/CarloCerquetella/aLTM.
Results
An aLTM to induce a scaling in anxiety behavior
The development of the aLTM allowed us to expose mice to six different anxiety configurations (Fig. 1A): at a height of 20 cm with the openable part either closed (NA) or open (VLA), with the openable part open at a height of 70 cm (LA), 120 cm (MA and HA), or 170 cm (VHA). Additionally, at the height of 120 cm, the floor width of the aLTM's open part could be adjusted from 10 to 7 cm, resulting in two different configurations (i.e., MA and HA), while it remained with a 7 cm floor width at 170 cm height (see Materials and Methods). To ensure a comparable motivation and prevent habitual behavior to the aLTM's open part across anxiety configurations, mice were food restricted and received small rewards (i.e., sugar pellets) in the aLTM's openable/open part. At the behavioral level, we observed that the trials taken and the time spent by mice (N = 16) in the openable part (including successive higher or lower anxiety configurations) linearly correlated with the six anxiety configurations (simple linear regression: number of trials, F = 66.20; R2 = 0.164; p < 0.001; percentage of time, F = 61.60; R2 = 0.154; p < 0.001; Spearman correlation: number of trials, r = −0.41; p < 0.001; percentage of time, r = −0.38; p < 0.001) and consequently resulted in six behaviorally scaling anxiety levels: NA–VLA–LA–MA–HA–VHA (Fig. 1B; Extended Data Fig. 1-1). A possible habituation/sensitization effect of the different anxiety configurations was further excluded by performing scrambled presentations of the different anxiety configurations, resulting in a linear scaling of behavioral anxiety when resorting the anxiety levels in an ascending order (Extended Data Fig. 1-1). In classical anxiety tests such as the EPM, fewer entries and reduced time spent in the open arms compared with the closed arms have been shown to be related to higher anxiety (Poulton and Menzies, 2002; Carobrez and Bertoglio, 2005). Also in our task, we found a progressive reduction in trials taken and time spent within the aLTM's open part, correlating with the increase in anxiety levels (Fig. 1B, left and right, respectively) and confirming that the anxiety configurations were inducing increasing anxiety in the open part of the maze.
vH inhibition impairs the scaling of anxiety behavior
Inhibition of vH activity has been shown to reduce anxiety behavior (Zhang et al., 2014; Padilla-Coreano et al., 2016; Maestas-Olguin et al., 2021). To explore the causal impact of vH activity in the aLTM, we applied an ArchT-mediated optogenetic intervention in vH CaMKIIα neurons, thereby predominantly targeting excitatory pyramidal neurons, during the transition of mice (N = 6 mice per group) from the closed to the open part of the aLTM (Fig. 2A–C; see Materials and Methods). With this approach, we aimed to influence vH activity primarily during the initial threat evaluation, and not during the complete exploration of the aLTM's open part, to ensure that the perceived anxiety level the mice are facing is not masked by vH optogenetic manipulation. We first validated that the optogenetic intervention inhibited the firing of vH neurons in vivo (Fig. 2D). Then, we checked the effect of this vH inhibition at the behavioral level. We found that the intervention in the transition zone of the aLTM led to a significant increase in trials and time in the aLTM's open part in the ArchT group compared with the control mice across anxiety-inducing configurations (Fig. 2E,F), indicating a reduction in anxiety-related behavior. Moreover, inhibition of the vH during the transition to the aLTM's open part disrupted the scaling of anxiety-related behavior expressed by unchanged trial numbers and time spent in the open part across anxiety levels in ArchT but not control mice (Fig. 2G,H; linear regressions of values of all sessions across all levels, F = 0.52; R2 = 0.00;, p = 0.475; F = 1.36; R2 = 0.021, p = 0.249 for ArchT mice trials and time spent; F = 10.50; R2 = 0.141; p = 0.002; F = 7.70; R2 = 0.107; p = 0.007 for control mice trials and time spent; Spearman correlation, r = −0.09; p = 0.492; r = −0.11; p = 0.372 for ArchT mice trials and time spent and r = −0.34; p = 0.005; r = −0.32; p = 0.009 for Ctr mice trials and time spent). Additionally, we could exclude an effect of the optogenetic silencing on locomotion by comparing the speed of ArchT and control mice. We also analyzed the mice' speed both across the entire maze and specifically within the openable part. Our results showed no statistical difference in locomotor activity with optogenetic silencing of the vH, indicating that the silencing does not produce locomotor changes per se (Extended Data Fig. 2-1). Collectively, this suggests that interference with vH activity can impair the perception of anxiety levels, without generally affecting locomotion.
vH inhibition impairs the scaling of anxiety behavior. A, Schematic representation of optic fiber implants (blue rectangles) and injection sites (green circles) of adeno-associated viruses (AAV5) expressing ArchT and eYFP reporter under the control of a CaMKIIα promoter. B, Image of anti-YFP immunolabeling in vH of ArchT mice with a magnification of vH somata on the right. vCA1 Pyr, ventral CA1 stratum pyramidale; dashed blue line, optic fiber. C, Schematic representation of optogenetic intervention in outbound trajectories indicating with a green rectangle the start of light application when the mice reach halfway from the door position to the open space and ending when the mice fully entered the open space. D, The raster plot of the activity of a representative neuron before, during, and after 2 s light delivery, repeated 20 times (left). Temporal dynamic (middle) of the averaged Z-scored firing rate of the entire recorded population in ArchT (green) and Ctr (black) mice (nCtr = 131 neurons; nArchT = 86 neurons). Green rectangles indicate the light delivery. The percentage distribution (right) of nonresponsive neurons (white), light-inhibited (green) and light-activated (red) neurons in the Ctr (top) and ArchT (bottom) group. ArchT, nNon-responsive = 29 neurons; nLight-inhibited = 56 neurons; nActivated = 1 neuron; Ctr, nNon-responsive = 123 neurons; nLight-inhibited = 8 neurons; nLight-activated = 0 neuron. Fisher's exact test p < 0.001. E, ArchT mice (N = 6) ran more trials in the open part of the maze across anxiety configurations (Mann–Whitney test p < 0.001) and (F) spent significantly more time in the open part than eYFP control (ctr) mice (N = 6; Mann–Whitney test p = 0.001). G, The gradually decreasing number of trials and (H) time spent in the aLTM's open part were disrupted in ArchT mice (right) but not control mice (left). Data represented as linear regressions with 95% confidence intervals calculated based on all behavioral sessions (n = 66 sessions). Mean ± SEM of all behavioral sessions for each anxiety configuration are also indicated (see Extended Data Fig. 2-1 for more details).
Figure 2-1
Optogenetic silencing does not produce locomotor changes. Average speed on the maze (left) and only when the mouse is on the openable part of the maze (right) per session. There is no significant effect of the optogenetic inhibition on the locomotor activity of the mice, Mann-Whitney test: p = 0.198, p = 0.347. (n = 24 sessions, N = 6 mice per group). Median of all behavioral sessions for each anxiety configuration is also indicated. Download Figure 2-1, TIF file.
vH activity scales with increasing anxiety levels
Tetrode implantations within the vH enabled the recording of 207 single units as mice navigated through the different anxiety configurations of the aLTM (Extended Data Fig. 3-1). Previous studies have highlighted that the vH exhibits increased activity in open anxiogenic compartments of mazes (Ciocchi et al., 2015; Jimenez et al., 2018; Forro et al., 2022; Volitaki et al., 2024). After behavioral confirmation of the scaling effect in implanted mice used for in vivo electrophysiology (Extended Data Fig. 3-2), we showed that individual neurons and vH population activity not only increased their activity in the open anxiogenic compartments, overrepresenting the anxiogenic part of the maze, but that the activity progressively increased from low to high anxiety configurations in the openable/open part of the maze (Fig. 3). We analyzed vH activity across NA to HA configurations (VHA was not included due to insufficient open part exploration needed for reliable activity analysis) whereby each level was represented by activity originating from increasing and decreasing anxiety-level presentations (see Materials and Methods). Simple linear regression analysis showed that the activity progressively increased from the NA to HA configuration in the openable/open part, but not in the closed part, in two single-neuron examples [F = 29.39; R2 = 0.406; p < 0.001 (Fig. 3A, left) and F = 64.80; R2 = 0.601; p < 0.001 (Fig. 3A, right) for the openable/open part; F = 0.64; R2 = 0.027; p = 0.432 (Fig. 3A, left) and F = 0.53; R2 = 0.023; p = 0.474 (Fig. 3A, right) for the closed part], and the same observation could be made in the entire recorded vH population (regression values of F = 43.00; R2 = 0.040; p < 0.001 and F = 0.49; R2 = 0.001; p = 0.485 for openable and closed part, respectively; Fig. 3B). Furthermore, the scaling of vH activity was not predominantly dependent on locomotion (Extended Data Fig. 3-2).
Scaling of vH activity during anxiety. A, Individual example neurons with increasing z-scored activity in the aLTM's open part across anxiety levels (linear regression on values of all sessions across NA to HA for Neuron 83 (left) and Neuron 102 (right). B, Heat map with z-scored firing rates show vH population activity progressively increasing in the aLTM's open part compared with the closed part (left). Heat maps depict the linearized z-scored firing rate activity of the entire recorded population (N = 6 mice; n = 207 neurons). Linear regression on all activity values of all sessions in the aLTM open part and in the closed part (right). Data represented as linear regressions with 95% confidence intervals based on all z-scored firing rate recorded values. Mean ± SEM of all behavioral sessions for each anxiety configuration are also indicated (see Extended Data Figs. 3-1, 3-2 for more details).
Figure 3-1
Histological validation. (A) Schematic representation of tetrode recordings sites indicated by electrolytic lesions (yellow circles). (B) Schematic representation of the optic fiber implants (red bars). Identifying mouse names are on top of every schematic representation. Download Figure 3-1, TIF file.
Figure 3-2
Anxiety behavior during aLTM navigation in mice used for electrophysiological experiments. Number of trials (left) and percentage of time in the open/openable compartment (right) across anxiety configurations indicate gradually increasing behavioral anxiety from NA to VHA in mice used for electrophysiological experiments. Simple linear regression: number of trials: F = 23.70, R2 = 0.098, p < 0.001; percentage of time F = 34.40, R2 = 0.136, p < 0.001; Spearman correlation: number of trials: r = -0.33, p < 0.001; percentage of time r = -0.38, p < 0.001. Data represented as linear regressions with 95% confidence intervals calculated based on all behavioral sessions (n = 22 sessions from N = 6 mice). Means ± SEM of all behavioral sessions for each anxiety configuration are also indicated. Download Figure 3-2, TIF file.
Tuning and reorganization of vH neuronal activity underlies scaling during anxiety
We hypothesized that two phenomena at the single-neuron level could contribute to the increased and scaled neuronal activity that we observed in the aLTM's open/openable part: first, a scaled activity tuning in which neurons could increase their activity with increasing anxiety levels, and second, a neuronal recruitment to the open part of the aLTM driven by different anxiety levels. To reveal if such a process might be contributing to the scaled representation of anxiogenic information in the vH population activity, we selected neurons that increased their activity linearly (significant Spearman correlation; see Materials and Methods) across each anxiety configuration, i.e., from NA to VLA, to LA, to MA, and to HA in the open part of the maze. We found 30 neurons (14.5% of the total population) whose firing rates increased consecutively across the anxiety levels in the aLTM's open, but not in the closed, part and thereby showed increased tuning with increasing anxiety (Fig. 4A,B; simple linear regression values, representative neuron F = 141.00; R2 = 0.979; p = 0.001 for openable; F = 2.65; R2 = 0.469; p = 0.202 for closed; all tuned neurons, F = 52.60; R2 = 0.150; p < 0.001 for openable; F = 0.51; R2 = 0.002; p = 0.478 for closed).
Tuning and reorganization of vH neuronal activity underlie scaling during anxiety. A, Linearized z-scored firing rate heat maps of a representative tuned neuron (left) and linear regressions between its z-scored firing rates and anxiety configurations (right). B, Linearized z-scored firing rate heat maps of all neurons showing an increase in activity at each anxiety configuration in the aLTM's open but not the closed part (left). Linear regressions between their z-scored firing rates and anxiety configurations (right). Data represented as linear regressions with 95% confidence intervals based on all recorded z-scored firing rate values (N = 6 mice; n = 30 neurons). Mean ± SEM for each anxiety configuration are also indicated. C, Linearized z-scored firing rate heat maps of a representative VLA recruited neuron (left) and z-scored firing rate differences between open to closed activities across levels in the representative VLA recruited neuron. D, Linearized z-scored firing rate heat maps of all VLA recruited neurons (left) and z-scored firing rate differences between open to closed activities across levels in all VLA recruited neurons show reorganized neuronal activity at VLA configuration that persists at higher anxiogenic configurations. VLA recruited neurons: Friedman test, p < 0.001; Dunn's multiple-comparison test; NA vs VLA, p < 0.001; NA vs LA, p < 0.001; NA vs MA, p < 0.001; NA vs HA, p < 0.001 (n = 92 neurons). E, Linearized z-scored firing rate heat maps of a representative LA recruited neuron (left) and z-scored firing rate differences between open to closed activities across levels in the representative neuron (right). F, Linearized z-scored firing rate heat maps of all LA recruited neurons (left) and z-scored firing rate differences between open to closed activities across levels in all LA recruited neurons show reorganized neuronal activity at LA configuration that persists at higher anxiogenic configurations. LA recruited neurons: RM one-way ANOVA, F(2.93,40.98) = 13.98; p < 0.001; Dunnett's multiple-comparison test, NA vs VLA, p = 0.360; NA vs LA, p < 0.001; NA vs MA, p = 0.002; NA vs HA, p = 0.003 (n = 15 neurons). G, Linearized z-scored firing rate heat maps of a representative MA recruited neuron and z-scored firing rate differences between open to closed activities across levels in the representative neuron (right). H, Linearized z-scored firing rate heat maps of all MA recruited neurons and z-scored firing rate differences between open to closed activities across levels in all MA recruited neurons show reorganized neuronal activity at MA configuration that persists at the higher anxiogenic configuration. MA recruited neurons: Friedman test, p = 0.007; Dunn's multiple-comparison test; NA vs VLA, p > 0.999; NA vs LA, p > 0.999; NA vs MA, p = 0.020; NA vs HA, p = 0.066 (n = 5 neurons). I, Linearized z-scored firing rate heat maps of a representative HA recruited neuron and z-scored firing rate differences between open to closed activities across levels in the representative neuron (right). J, Linearized z-scored firing rate heat maps of all HA recruited neurons (left) and z-scored firing rate differences between open to closed activities across levels in all HA recruited neurons show reorganized neuronal activity at HA configuration. HA recruited neurons, Friedman test, p < 0.001; Dunn's multiple-comparison test; NA vs VLA, p = 0.319; NA vs LA p > 0.999; NA vs MA, p = 0.236; NA vs HA, p < 0.001 (n = 11 neurons). N = 6 mice. Dotted line in the heat maps represents entrance location in the openable part. Error bars represent mean ± SEM (see Extended Data Fig. 4-1 for more details).
Figure 4-1
Adaptive scaling of vH neuronal activity indicates learning- and sensitization-independent effects. (A) Linear regressions between z-scored firing rates and anxiety configurations in every configuration when presented before a less anxiogenic configuration (NA was not preceded by any configuration) (left) or a more anxiogenic configuration (HA was not preceded by any configuration) (right). More anxiogenic order: open: R2 = 0.160, p < 0.001; closed: R2 = 0.001, p = 0.687; less anxiogenic order: open: R2 = 0.142, p < 0.001; closed: R2 = 0.002, p = 0.583. Data represented as linear regressions with 95% confidence intervals based on all recorded z-scored firing rate values (N = 6 mice, n = 30 neurons). Means ± SEM for each anxiety configuration are also indicated. (B) Z-scored firing rate differences between open to closed activities across levels in VLA recruited neurons show reorganized neuronal activity at VLA configuration that persists at higher anxiogenic configurations when every configuration was presented after a less anxiogenic (left) or a more anxiogenic (right) configuration. VLA recruited neurons: left: Friedman test, p < 0.001; Dunn's multiple comparisons test, NA vs VLA p < 0.001; NA vs LA p < 0.001; NA vs MA p < 0.001; NA vs HA p < 0.001; right: Friedman test, p < 0.001; NA vs VLA p < 0.001; NA vs LA p < 0.001; NA vs MA p < 0.001; NA vs HA p < 0.001 (n = 92 neurons). (C) Z-scored firing rate differences between open to closed activities across levels in LA recruited neurons show reorganized neuronal activity at LA configuration that persists at higher anxiogenic configurations when every configuration was presented after a less anxiogenic (left) or a more anxiogenic (right) configuration. LA recruited neurons: left: RM one-way ANOVA, F(2.36,33.07) = 8.63, p < 0.001; Dunnett's multiple comparisons test, NA vs VLA p = 0.684; NA vs LA p < 0.001; NA vs MA p = 0.019; NA vs HA p = 0.010; right: RM one-way ANOVA, F(3.16,44.28) = 6.06, p = 0.001; Dunnett's multiple comparisons test, NA vs VLA p = 0.145; NA vs LA p = 0.002; NA vs MA p = 0.025; NA vs HA p = 0.014 (n = 15 neurons). (D) Z-scored firing rate differences between open to closed activities across levels in MA recruited neurons show reorganized neuronal activity at MA configuration that persists at the higher anxiogenic configuration when every configuration was presented after a less anxiogenic (left) or a more anxiogenic (right) configuration. MA recruited neurons: left: Friedman test, p = 0.126; Dunn's multiple comparisons test, NA vs VLA p > 0.999; NA vs LA p > 0.999; NA vs MA p = 0.287; NA vs HA p = 0.646; right: Friedman test, p = 0.066; Dunn's multiple comparisons test, NA vs VLA p = 0.921; NA vs LA p > 0.999; NA vs MA p = 0.037; NA vs HA p = 0.287 (n = 5 neurons). (E) Z-scored firing rate differences between open to closed activities across levels in HA recruited neurons show reorganized neuronal activity at HA configuration when the HA configuration was presented after a less anxiogenic configuration (left) or when it was the first presented (right). HA recruited neurons: left: Friedman test, p = 0.011; Dunn's multiple comparisons test, NA vs VLA p = 0.018; NA vs LA p > 0.999; NA vs MA p = 0.236; NA vs HA p = 0.007; right: Friedman test, p < 0.001; Dunn's multiple comparisons test, NA vs VLA p = 0.900; NA vs LA p > 0.999; NA vs MA p = 0.04; NA vs HA p < 0.001 (n = 11 neurons). N = 6 mice. Error bars represent mean ± SEM. Download Figure 4-1, TIF file.
To investigate if changes in neuronal recruitment took place at any specific anxiety configurations, we quantified the neuronal activity based on the earliest significant increase in firing in the aLTM's open part compared with the closed part. Our results suggest that at each anxiety level, new neurons were recruited and maintained their activation to subsequent higher anxiety levels (Fig. 4C–J). Although major reorganization occurred at the first exposure to the aLTM's open part (in VLA), sequential recruitment occurred at any anxiety levels. This indicates that both enhanced tuning of neuronal responses and increased neuronal recruitment contribute to the scaling of neuronal activity in response to rising anxiety levels (Extended Data Fig. 4-1).
Novelty does not change vH population activity
To discern changes induced by novelty from the changes related to anxiety configurations, we replaced the openable compartment with a novel compartment where the sidewalls had a clearly different visual appearance and a different texture (Fig. 5A, right). During the exploration in the novel configuration, mice demonstrated a preference for the novel section of the maze (Fig. 5B, left) but showed no difference in trials taken compared with the control configuration (corresponding to a NA configuration; Fig. 5B, right). This shows that mice expressed a similar motivation to run trials and were able to detect and engage with the novel part of the maze (File, 2001).
Novelty does not change vH population activity. A, Control (left) and “Novelty” (right) configurations of the aLTM illustrating the visual differences of the novel part of the maze. B, Mice spent more time (left) in the novel part of the maze but did not engage in different number of trials (right) than in the control configuration. Wilcoxon test, p < 0.001. Paired t test, p = 0.392. C, Linearized firing rate heat maps (left) and z-scored firing rate (right) show no significant changes induced by the novelty and that activity levels did not reach values close to the maximum vH activations of the anxiety configurations. Two-way RM ANOVA interaction values, F(1,12) = 1.68; p = 0.219. Uncorrected Fisher's LSD: ctr, closed vs open p = 0.002; nov, closed vs open p = 0.070; closed, ctr vs nov p = 0.372; openable, ctr vs nov p = 0.373 (N = 6 mice; n = 207 neurons). The dotted line in the heat map represents the entrance location of the openable part. Error bars represent mean ± SEM.
Novelty can induce a spatial remapping of activity in the hippocampus (Leutgeb et al., 2005; Dong et al., 2021), but we observed that the novelty in the openable compartment, at least at the overall population level, did not lead to an increased vH activity compared with the control configuration (Fig. 5C), suggesting that changes in vH activity during aLTM navigation were mainly due to anxiety processing.
Decoding of anxiety levels by a linear classifier confirms the scaling of anxiety-related activity in the vH
Finally, to further validate that vH population activity encodes anxiety-related information and that this encoding scales with the actual perceived anxiety level, we measured the performance of a linear classifier trained to decode anxiety levels from recorded firing rates in the openable/open part of the maze (Fig. 6). First, we performed multiclass classification on all five anxiety levels. The values in the diagonal of the confusion matrix indicate, for each configuration, the proportion of samples which are correctly assigned to their ground-truth configuration. These test accuracies are significantly above the chance level for all anxiety configurations (Fig. 6A), which is a telltale sign that vH population activity encodes information about anxiety levels.
Decoding of anxiety levels by a linear classifier. A, Confusion matrix indicating the validation accuracy in balanced data obtained by oversampling actual data (left) and shuffled data (right). Ordinary one-way ANOVA, NA, F(9,210) = 85.56; p < 0.001; Bonferroni's multiple-comparison test, NAactual data–NAshuffled data p < 0.001; ordinary one-way ANOVA, VLA, F(9,210) = 12.24; p < 0.001; Bonferroni's multiple-comparison test; VLAactual data–VLAshuffled data p < 0.001; ordinary one-way ANOVA, LA, F(9,210) = 16.45; p < 0.001; Bonferroni's multiple-comparison test; LAactual data–LAshuffled data p < 0.001; ordinary one-way ANOVA; MA, F(9,210) = 5.65; p < 0.001; Bonferroni's multiple-comparison test; MAactual data–MAshuffled data p < 0.001; ordinary one-way ANOVA; HA, F(9,210) = 12.48; p < 0.001; Bonferroni's multiple-comparison test; HAactual data–HAshuffled data p < 0.001; B, Left, Linear classifier trained on NA and HA configurations for each experiment when the mouse was in the open/openable part of the maze (actual data, black, simple linear regression, F = 157.30; R2 = 0.593; p < 0.001). Linear classifier trained on balanced data from NA and HA configurations for each experiment with shuffled labels (gray) results in a flat line (shuffled data NA to HA linear regression, F = 0.46; R2 = 0.004; p = 0.502). The test set consists of all data from the VLA, LA, and MA configurations (VLA, LA, MA values). Results show that the decoder output monotonically decreases as the anxiety level increases (see Extended Data Fig. 6-1 for more details). Data represented as linear regressions with 95% confidence intervals calculated based on all behavioral sessions. Right, one-sided unpaired t test of individual anxiety configurations against shuffled data for the classifier performance, NAactual data–NAshuffled data Welch's correction p < 0.001; VLAactual data–VLAshuffled data Welch's correction p = 0.111; LAactual data–LAshuffled data Welch's correction p = 0.021; MAactual data–MAshuffled data Welch's correction p < 0.001; HAactual data–HAshuffled data Welch's correction p < 0.001; N = 6 mice; n = 22 sessions. Mean ± SEM of all behavioral sessions for each anxiety configuration are also indicated.
Figure 6-1
Decoding of anxiety levels using a linear classifier. (A) Confusion matrix indicating the validation accuracy. (B) Linear classifier trained on NA and HA configurations for each experiment when the mouse was in the open/openable part of the maze: held-out normalized firing rate data for these configurations are used as a validation set (HA and NA values) and shows that NA and HA configurations are correctly classified. The test set consists of all data from the VLA, LA and MA configurations (VLA, LA, MA values). Results show that the decoder output monotonically decreases as the anxiety level increases (left). Linear classifier trained on NA and HA configurations for each experiment with shuffled labels results in a flat line skewed upwards (right), meaning that the classifier has a classification tendency towards the NA category. This is likely due to the higher number of samples collected from the NA configuration than from other configurations. Data represented as linear regressions with 95% confidence intervals calculated based on all behavioral sessions (N = 6 mice, n = 22 sessions). Means ± SEM of all behavioral sessions for each anxiety configuration are also indicated. Download Figure 6-1, TIF file.
Representative aLTM trial. Representative example of a single aLTM trial. [View online]
Next, to verify that this encoding scales continuously with each anxiety level, we leveraged the performance of the classifier on unseen data. For a classifier trained solely on normalized firing rates from the two extreme conditions (i.e., NA and HA), the proportion of unseen data that are classified as NA (respectively as HA) can be interpreted as a measure of their similarity with the NA (respectively HA) configuration. For the vH to function as an anxiometer, we would thus expect the proportion of data being classified as NA to be the highest for actual NA data and to continuously decrease as the actual anxiety level goes to VLA, LA, MA, and ultimately HA. This is indeed what we observed (Fig. 6B, left), when the classifier was solely trained to discriminate NA and HA samples and its validation accuracy was assessed on held-out data from all five configurations. The leftmost value indicated the proportion of NA samples which were correctly classified as NA. Similarly, the rightmost value indicates the proportion of HA samples which were classified as NA. Expectedly, the proportion of samples classified as NA is decreasing when the ground-truth anxiety level increases. Then, we assessed how unseen VLA, LA, and MA samples would be sorted by the classifier only trained on NA and HA samples (second, third, and fourth values). The proportion of unseen data that are classified as NA decreases with the increasing anxiety level (Fig. 6B), which defines a continuously scaling “anxiety axis” in the vH population activity. Simple linear regression analyses confirmed the observed scaling phenomenon with F = 157.30; R2 = 0.593; p < 0.001. Furthermore, to verify that the scaling is not an artifact from the data or the classifier, we performed the classification after shuffling the data labels, to estimate the probability that a sample is attributed by chance to one of the two extreme configurations. As expected, this resulted in a flat line across all five anxiety configurations, contrasting with the previously observed scaling (Fig. 6B, left, gray, shuffled data NA to HA linear regression, F = 0.46; R2 = 0.004; p = 0.502; Fig. 6B, right).
Since the NA configuration is the only one with closed walls surrounding the entire platform, the vH neuronal activity of this configuration may stand out compared with the other more anxiogenic configurations. If this is the case, then training the classifier on VLA and HA (instead of NA and HA) might yield more robust results. Therefore, we conducted an additional analysis by training the classifier on VLA and HA and computing the fraction of unseen LA and MA samples that are assigned to the VLA configuration. The results demonstrate that the proportion of samples classified as VLA is higher for LA than for MA samples and the VLA to HA regression (Extended Data Fig. 6-1; VLA to HA linear regression, F = 265.50; R2 = 0.755; p < 0.001; shuffled data VLA to HA linear regression, F = 0.05; R2 = 0.001; p = 0.821) is stronger than the NA to HA regression (Fig. 6B; F = 157.30; R2 = 0.593; p < 0.001). Collectively, the decoding of anxiety levels from population activity confirms the scaling of anxiety-related activity in the vH.
Discussion
Humans and animals must strike a balance between minimizing exposure to possibly harmful events and exploring the environment to find rewards. Anxiety responses are controlled by the septohippocampal “behavioral inhibition” system (Gray, 1982), which is critical for the behavioral effects of anxiolytic drugs (McNaughton and Gray, 2000; McNaughton, 2006, 2024). Anxiety has been suggested to bias decision-making by increasing the prediction of negative future outcomes (Hartley and Phelps, 2012; Gagne et al., 2018). In our study, we investigated how anxiety representations in the vH of mice are tuned to different anxiety levels and might thereby provide a basis for decisions under approach–avoidance conflict. For this purpose, we developed a maze, the aLTM, on which mice could be exposed to six different anxiety configurations during navigation in the same spatial environment, while we recorded the activity of vH neurons.
We found that we could successfully induce different anxiety levels in mice and that anxiety-related activity in the vH progressively scaled with the different anxiety levels. Human neuroimaging studies have shown that greater hippocampal BOLD signals in the anterior hippocampus are associated with a greater probability of threat and higher anxiety during risk assessment (Bach et al., 2014; Odriozola et al., 2023). In rodents, vH activity is enhanced during anxiety encounters, predicts the exploration of anxiogenic locations, and has been shown to further increase with anxiety-associated head-dipping behavior (Jimenez et al., 2018). With our results, we provide evidence that behavioral scaling of anxiety within the same spatial and behavioral framework is supported by fine-tuning and scaling of vH activity, suggesting that the vH might act as an “anxiometer” in the brain. In contrast, previous work has shown that the basal amygdala encodes behavioral states such as freezing, head dips, entries, and exits to anxiogenic compartments, but does not represent the level of absolute perceived anxiety (Grundemann et al., 2019). Rather, our findings position the vH as a critical limbic region that functions as an “anxiometer” by scaling its activity based on perceived anxiety levels. To test the causal relationship between the scaling activity in the vH and anxiety experienced by the mice, we applied ArchT-mediated optogenetic intervention during the approach of the open compartment of the maze. In a study in humans, the decision to approach a reward relied on the hippocampus and not on the amygdala (Bach et al., 2019). In line with this finding, when we inhibited the vH during the approach of the open anxiogenic environment, the mice expressed reduced anxiety and a disruption in the scaling anxiety behavior in the open part of the maze indicating the vH's crucial role in approach–avoidance behavior. This suggests that while the vH plays a crucial role in processing and initially relaying the anxiogenic features, its involvement in maintaining the anxiety signal might be less critical once exploration has begun.
Tuning of neuronal activity has been extensively described in sensory areas, for example, in orientation-tuned neurons in the primary visual cortex that exhibit highest firing rates in response to specific orientation of a visual stimulus (Hubel and Wiesel, 1962; Niell and Stryker, 2008). Responses of dopamine neurons in the ventral tegmental area of the midbrain have been also described to scale to error prediction depending on the reward size (Eshel et al., 2016). The integration of sensory stimuli such as odors, tones, or speed of the animal has been shown in the hippocampus and entorhinal cortex to drive neural activity levels (Aikath et al., 2014; Kropff et al., 2015; Aronov et al., 2017; Iwase et al., 2020). How might individual neurons and population codes in higher cognitive areas such as the hippocampus achieve a tuning to internally generated emotional states such as anxiety? We described two processes that contributed to the scaling of anxiety-related activity in the vH, i.e., an enhanced tuning of neuronal responses and a neuronal activity recruitment to the open part of the aLTM driven by different anxiety levels. We hypothesize that these processes might be similar to the well described concepts of remapping in the hippocampus and rely on a spatial framework whereby neurons can either change their firing rates within the same location referred to as rate remapping or in a new environment randomly reconfiguring spatial codes during global remapping (Schlesiger et al., 2018). We argue that these processes might not be restricted to the encoding of space but to any information that is predominantly processed by the underlying hippocampal subregion (Eichenbaum, 2018). The recruitment of new vH neurons that were activated at higher anxiety levels might reflect a process similar to global remapping whereby neurons that were not responding in previous anxiety encounters were engaged and recruited at a specific anxiety “threshold,” not to a new environment as during global remapping but to a different anxiety level. vH circuits integrate a multiplicity of synaptic inputs, e.g., visual–spatial contexts and object information from the medial and lateral entorhinal cortex and anxiety-related information from the basolateral amygdala (Felix-Ortiz et al., 2013; Maestas-Olguin et al., 2021; Nguyen et al., 2023). We speculate that convergence of synaptic inputs and possibly inhibitory microcircuits might tune the scaling of rate remapping and gate the switching of newly recruited neurons during different anxiety encounters. Moreover, the increase in vH activity could indicate specific response to anxiety levels together with a general increase in arousal and emotional tone (Jimenez et al., 2018). Thus, future studies shall investigate whether vH activity can also scale across different negative or positive emotions using, for instance, fear or reward conditioning paradigms.
All changes of the anxiety configurations occurred within the same spatial environment; however, the configurational changes, even minor, were of physical nature (e.g., distance to the ground or removal of part of the maze floor). To exclude that the mice might perceive these as novel aspects of the environment, we introduced a novel configuration of the maze. Based on our results, we conclude that even if a remapping had occurred, it did not result in such prominent dynamic increases of neuronal activity as observed across anxiety configurations, and, therefore, novelty is unlikely to have driven the observed changes. The spatially independent tuning of activity also supports the concept that the hippocampus integrates and represents emotional information based on perceived anxiety levels.
Anxiety disorders in humans are characterized by an elevated perception of potential threats leading to excessive uncontrollable worries that patients overcome with difficulties (Bredemeier and Berenbaum, 2020; Mishra and Varma, 2023). In patients with obsessive–compulsive disorders (a form of generalized anxiety) and schizophrenia (with frequent emotional disturbances), hyperactivity of the vH has been reported (Wolff et al., 2018; McHugo et al., 2019). Also, patients with temporal lobe epilepsy and hyperexcitability of the vH can experience anxiety disorders (Navidhamidi et al., 2017; Wolff et al., 2018). We hypothesize that in some of these patients, the mechanisms to control and adjust the scaling of vH activity may be dysfunctional, possibly either shifted to more elevated levels or, instead of gradually scaling rather jump to higher levels, maladaptively influencing their decisions and behavior. Adaptive anxiety scaling, when disrupted, may thus contribute to the manifestation of anxiety disorders. Collectively, our findings provide insight into how the hippocampus could allow for a dynamic and flexible adjustment of anxiety responses according to accumulated experience and prevailing environmental conditions.
Footnotes
This work was supported by the following grants: an ERC starting grant (716761) to S.C., Swiss National Science Foundation (SNSF) professorship grants (170654, 206129) to S.C. and SNSF Grants (31003A_175644 and 310030_212247 to J.-P.P.; 31003A_175644 and P500PM_210800 to C.G.). We further thank Kaizhen Li for the help with histology and immunolabeling, the electronic workshop and members of the Ciocchi Laboratory for their valuable inputs on the project, and Jakob Jordan and Nicolas Deperrois for their fruitful discussions.
The authors declare no competing financial interests.
- Correspondence should be addressed to Carlo Cerquetella at carlo.cerquetella{at}unibe.ch or Stéphane Ciocchi at stephane.ciocchi{at}unibe.ch.
