Abstract
During learning of a sensory discrimination task, the cortical and subcortical regions display complex spatiotemporal dynamics. During learning, both the amygdala and cortex link stimulus information to its appropriate association, for example, a reward. In addition, both structures are also related to nonsensory parameters such as body movements and licking during the reward period. However, the emergence of the cortico-amygdala relationships during learning is largely unknown. To study this, we combined wide-field cortical imaging with fiber photometry to simultaneously record cortico-amygdala population dynamics as male mice learn a whisker-dependent go/no-go task. We were able to simultaneously record neuronal populations from the posterior cortex and either the basolateral amygdala (BLA) or central/medial amygdala (CEM). Prior to learning, the somatosensory and associative cortex responded during sensation, while amygdala areas did not show significant responses. As mice became experts, amygdala responses emerged early during the sensation period, increasing in the CEM, while decreasing in the BLA. Interestingly, amygdala and cortical responses were associated with task-related body movement, displaying significant responses ∼200 ms before movement initiation which led to licking for the reward. A correlation analysis between the cortex and amygdala revealed negative and positive correlation with the BLA and CEM, respectively, only in the expert case. These results imply that learning induces an involvement of the cortex and amygdala which may aid to link sensory stimuli with appropriate associations.
Significance Statement
Levitan and Gilad study neuronal dynamics in the cortex and amygdala as mice learn a sensory discrimination task. They found that amygdala areas display opposing responses only after the mice learn the task. In contrast cortical areas display responses both before and after learning. These results imply that learning induces an involvement of amygdala and cortical areas in order to optimally link sensory stimuli with appropriate associations.
Introduction
Somatosensation is often used to guide animals toward an important goal. In doing so, animals must learn to associate specific stimuli with certain outcomes, for example, a reward or punishment, and eventually act accordingly (Helmchen et al., 2018; Petersen, 2019). Learning, like other cognitive phenomenon, requires a recruitment of neuronal populations distributed across several brain regions which become dynamically and functionally connected through learning (Poort et al., 2015; Makino et al., 2017; Gilad et al., 2020; Gilad and Helmchen, 2020; Lin et al., 2021; Esmaeili et al., 2022; Roy et al., 2022). In rodents, during learning of whisker-related sensory discrimination task, cortical and subcortical regions modulate their sensory evoked responses (Komiyama et al., 2010; Wiest et al., 2010; Le Merre et al., 2018; Gilad et al., 2020; Gilad and Helmchen, 2020; Esmaeili et al., 2022). In the posterior cortex, the primary somatosensory cortex (barrel cortex; BC), the secondary sensory cortex (S2), and regions in the associative cortex (rostrolateral; RL; Hovde et al., 2018; Lyamzin and Benucci, 2019) process sensory and associative information, enabling the cortex to discriminate between different stimuli and to associate them with the right context (Romo and Salinas, 2003; Erlich et al., 2011; Harvey et al., 2012; Raposo et al., 2014; Yang et al., 2015; Kwon et al., 2016; Jeong et al., 2017; Gilad et al., 2018). Once the stimulus information is integrated, information flows to regions in the frontal cortex, where the primary motor cortex (M1) and secondary motor cortex (M2) coordinate the appropriate motor program (Miyashita and Feldman, 2013; Sachidhanandam et al., 2013; Guo et al., 2014; Goard et al., 2016; Makino et al., 2017; T-W. Chen et al., 2017; Gilad and Helmchen, 2020; Esmaeili et al., 2021).
But the cortex does not do this by itself. In order to properly associate specific stimuli with certain outcomes, the cortex interacts with several subcortical brain regions, where one important area is the amygdala (Mcdonald, 1998; Kayyal et al., 2019; Fu et al., 2020; Hintiryan et al., 2021; Mahmood et al., 2023). The amygdala is a central part of a principal pathway connecting the limbic system to the cortex, with multiple subdivisions, among them the basolateral amygdala (BLA) and central/medial amygdala (CEM). Anatomically, both structures receive direct projections from the sensory and associative areas in the posterior cortex (Mcdonald, 1998; Fu et al., 2020; Hintiryan et al., 2021), and their projections innervate effector regions in the motor area in the anterior cortex and regions in the brainstem, midbrain, thalamus, and hypothalamus (Mátyás et al., 2014; Levitan et al., 2020; Steinberg et al., 2020). The amygdala, through its circuitry, encodes and distributes valence of the incoming sensory stimulation and is instrumental for goal-directed learning (Samuelsen et al., 2012; Kim et al., 2017; Lichtenberg et al., 2017; Beyeler et al., 2018; Reed et al., 2018; Levitan et al., 2020; Steinberg et al., 2020). Indeed, previous studies indicate that learning could be reflected in the functional connectivity between the amygdala and regions such as orbitofrontal cortex and gustatory cortex (Grossman et al., 2008; Gutierrez et al., 2010; Samuelsen et al., 2012). However, simultaneous recording of the amygdala with the somatosensory and associative cortex is scarce, although predicted to be important for somatosensory processing (Gothard and Fuglevand, 2022; Martin et al., 2023). Here, we used longitudinal wide-field calcium imaging to examine activity of neurons through somatosensory regions in the posterior cortex together with fiber photometry to record neuronal activity in the amygdala as mice gradually learn a go/no-go texture discrimination task. We found that both the amygdala and cortex develop discriminating responses following learning, increasing in CEM, while decreasing in BLA during the sensation period. In expert mice, amygdala and cortical responses precede body movement initiation which eventually led to a decision to lick for a reward. In addition, the cortex displayed a negative and positive correlation with the BLA and CEM, respectively, emerging during the early sensation period and only after learning. We suggest that in order to learn a certain association, the amygdala is recruited and interacts with cortical areas to integrate a proper association and guide appropriate action.
Materials and Methods
Animals
A total of n = 10, 8–16-week-old male C57BL/6 mice were used in this study. All experiments were approved by the Institutional Animal Care and Use Committee (IACUC) at the Hebrew University of Jerusalem, Israel (Permit Number: MD-20-16065-4).
Surgery
To enable simultaneous recording of the whole posterior cortex and amygdala, we combined wide-field imaging with fiber photometry (Fig. 2A). Mice were anesthetized with 2% isoflurane (in pure O2) and body temperature was maintained at 37°C. We applied local analgesia (lidocaine 1%), exposed and cleaned the skull, and removed some muscles to access the entire dorsal surface of the both hemispheres. On the left hemisphere, we drilled a 5 mm craniotomy which exposed the whole posterior cortex including the barrel cortex (BC), secondary somatosensory cortex (S2), association areas including rostrolateral cortex (RL), and somatosensory forelimb cortex (FL). We then superficially injected (microinjector, World Precision Instruments) 200 nl of an AAV virus pAAV.Syn.GCaMP6f.WPRE.SV40 (AAV9; from Addgene) to nine cortical sites spread across the craniotomy. Next, we implanted a cover glass and applied transparent dental cement to fix it to the skull (Tetric EvoFlow T1). On the right hemisphere, we drilled a small hole (−1.6 mm posterior to bregma; 3.2 mm lateral to midline) and then injected the same virus with a 55° angle and depth of 7.9 mm targeting the amygdala of the left hemisphere (under the imaged cortex). Using the same coordinates, we then inserted a 400 mm optical fiber with an attached cannula (CFMC14L10; Thorlabs; 10 mm long) and chronically fixed the fiber position using dental cement. This procedure localized the fiber tip in either the central or medial amygdala (pooled together as CEM; −1.3 mm posterior to bregma, 2.2 mm lateral to midline, 4.6 mm under the pia; 5 mice) or the BLA (−1.6 mm posterior to bregma, 3.4 mm lateral to midline, 4.6 mm under the pia; 5 mice; Fig. 1A). Success rate was ∼70% and enabled simultaneous imaging of both the posterior cortex and amygdala.
Amygdala targeting and behavioral paradigm. A, A calcium indicator (GCaMP6f) was injected to the cortex and amygdala in the same mouse. Right, Coronal slices displaying virus spread in the cortex and amygdala along with fiber tips in CEM or BLA. B, Fiber tip locations in CEM (red) and BLA (pink) for each mouse. C, Imaging and behavioral setup for combining wide-field imaging and fiber photometry. D, Trial structure for the go/no-go whisker dependent task. E, Performance (d’) as a function of session number for each mouse. Dashed black line depicts the learning threshold (d’ = 1.5).
Texture discrimination task
Mice were trained on a go/no-go discrimination task (Fig. 1C,D; J. L. Chen et al., 2013; Gilad and Helmchen, 2020) using a data acquisition interface (USB-6008; National Instruments) and custom-written LabVIEW software (National Instruments). Each trial started with an auditory cue (stimulus cue; 2 beeps at 2 kHz, 100 ms duration with 50 ms interval), signaling the approach of either two types of sandpapers (grit size P100: rough texture; P1200: smooth texture; 3 M) to the mouse's whiskers as “go” or “no-go” textures (Fig. 1C; pseudorandomly presented with no more than three repetitions). Sandpapers were mounted onto panels attached to a stepper motor (T-NM17A04; Zaber) mounted onto a motorized linear stage (T-LSM100A; Zaber) to move textures in and out of reach of whiskers. The texture stayed in touch with the whiskers for 2 s, and was then moved out after which an additional auditory cue (response cue; 4 beeps at 4 kHz, 50 ms duration with 25 ms interval) signaled the start of a 2 s response period. The stimulus and response cues were identical in both textures. The interval between the trials was 6 s (8 s from response to next cue). A water reward (∼3 μl) was given to the mouse for licking for the go texture after the response cue (“Hit”), that is, for the first correct lick during the response period (Fig. 1D; licks were detected using a piezo sensor). Punishment with white noise was given for licking for the no-go texture (“false alarms”; FA). Licking before the response cue was neither rewarded nor punished. Reward and punishment were omitted when mice withheld licking for the no-go (“correct-rejections”, CR) or go (“Misses”) textures.
Training and performance
Mice were first handled and accustomed to head fixation before starting water scheduling. Before imaging began, mice were conditioned to lick for reward after the go texture (P100 grit size). Imaging began only after mice reliably licked for the response cue (typically after the first day; 200–400 trials). On the first day of imaging, mice were presented with the “go” texture and after 50 trials the “no-go” texture was gradually introduced (starting from 10% and increasing by 10% approximately every 50 trials) until reaching 50% probability for the no-go texture by the end of the day. Most mice reached expert level (i.e., a learning threshold of d’ > 1.5) after 3–4 d (Fig. 1E). Imaging and monitoring of body movements was performed continuously throughout the whole learning process up to 14 d. An effort was made to maintain a constant position of the texture and cameras across imaging days in order to maintain similar stimulation and imaging parameters. In this study we divided the data into naive (i.e., days before reaching the learning threshold) and expert (i.e., days after reaching the learning threshold) cases.
Wide-field cortical imaging and fiber photometry
On the left hemisphere, we used a wide-field approach to image large parts of the posterior cortex while mice learned to perform the task (Gilad et al., 2018; Abdelfattah et al., 2022; Gilad, 2024). A sensitive CMOS camera (Hamamatsu Orca Flash 4.0) was mounted on top of a dual objective setup. Two objectives (Navitar; top objective: D-5095, 50 mm f0.95; bottom objective inverted: D-2595, 25 mm f0.95) were interfaced with a dichroic (510 nm; AHF; Beamsplitter T510LPXRXT) filter cube (Thorlabs). Blue LED light (Thorlabs; M470L3) was guided through an excitation filter (480/40 nm BrightLine HC), a diffuser, collimated, reflected from the dichroic mirror, and focused through the bottom objective ∼100 µm below the blood vessels. Green light emitted from the preparation passed through both objectives and an emission filter (514/30 nm BrightLine HC) before reaching the camera. The total power of blue light on the preparation was <5 mW, i.e., <0.1 mW/mm2. At this illumination power, we did not observe any photobleaching. Data was collected with a temporal resolution of 20 Hz and a spatial sampling of 512 × 512 pixels, resulting in a spatial resolution of ∼20 μm/pixel. For fiber photometry of the amygdala, a 400 μm optical fiber (M79L01; Thorlabs) was connected to the implanted cannula using a mating sleeve (ADAF1; Thorlabs), and the cable end was connected to an optical setup. This setup was similar to the wide-field setup in terms of the imaging camera, dichroic and filters, differing only in the excitation source (473 OBIS laser; Coherent; output power of ∼0.8 mW) and shaping of the excitation path to enlarge the laser beam onto the optical fiber.
Control for non-calcium dependent signals
The data collected in this study used a single wavelength (473 nm) to image calcium dynamics as in previous studies (Gilad et al., 2018, 2020; Gilad and Helmchen, 2020; Gallero-Salas et al., 2021; Marmor et al., 2023). This protocol may additionally collect noncalcium-dependent signals, such as hemodynamic signals, which may affect the results. To control for this, we trained three additional mice using an interleaved imaging protocol of 473 and 405 nm (isosbestic) excitation lights. These mice were implanted with a 5 mm window over the posterior cortex similar to the mice in this study. Correcting for noncalcium signals (473 light minus 405 light within each trial) maintained the results in all cortical areas, including differences between naive and expert cases and movement-related effects (Fig. 2D–F).
Amygdala display divergent responses. A, Left, Schema of the preparation enabling simultaneous imaging of the posterior cortex and amygdala. Right, Example activity maps averaged during the sensation period displaying activity in BC, RL, S2, and FL. B, Grand average responses in the cortex and amygdala during Hit (blue for cortex; red for amygdala) or CR (black) trials for naive (left) and expert (right) cases. Shaded area depicts the sensation period. C, Grand average responses averaged during the sensation period in each area during Hit and CR trials for naive (left) and expert (right) cases. Error bars depict SEM across recording sessions (n = 22 and 38 for naive and expert cases; 10 mice for cortex, 5 mice for BLA, 5 mice for CEM). D–F, Control experiments for cortical signals. D, Schematic illustration of the control experiments where the cortical window was excited with interleaved blue (473 nm) and control (405 nm) lights. E, Example response in BC for Hit and CR of an expert mouse for the 473 nm light (left), 405 nm light (middle), and corrected signal (473 signal minus 405 signal). Error bars depict mean ± SEM over trials (n = 94 and 89 trials for Hit and CR trials, respectively). F, Grand average responses averages during the sensation period (−1 to 1 s relative to texture stop) in Hit and CR trials, naive and expert cases, for each of the three mice separately. Error bars depict mean ± SEM over trials (range for number of trials from 118 to 719). ***p < 0.001; n.s., not significant.
Monitoring of body movement
In parallel to imaging the cortex and amygdala, we tracked movements of the mouse's body during the task, from naive to expert (Fig. 1C). The mouse was illuminated with a 940 nm infrared LED, and a camera monitored the movements of the mouse at 30 Hz (the imaging source; DMK 22BUC03; 720 × 480 pixels). To synchronize between the body camera and neuronal camera, we first detected the texture stop in the body camera in every trial which was clearly visible throughout the whole recording session. Next, we aligned all movement vectors to texture stop, which also corresponds to 3 s after the start of neuronal recordings. We used movements of both forelimbs and the head/neck region to assess body movements and reliably detect large movements (Fig. 3A; see below, Data analysis). Body movement was extracted by calculating the frame-to-frame correlation within each area (forelimb or head–neck region) as a function of time (across the whole movie during the session) and subtracting the result from 1 (1 minus frame-to-frame correlation). The general body movement was then averaged across both regions. We note that the selected regions were not contaminated by the movement of the texture. This was done as a function of time for each trial. In addition, tongue movement, indicating licking, was also detected (taking the lower jaw area in the movie; Matteucci et al., 2022) and analyzed separately from the body movement (Fig. 3B).
Neuronal responses in the cortex and amygdala are affected by movement. A, Bottom left, Example trials showing the body movement of the mouse in a passive (gray; not moving during the sensation period; shaded area) or active (green; moving during the sensation period) Hit trial. Dashed gray line depict movement threshold. Bottom right, Movement probability during the trial in a naive (gray) and expert (green) example recording session. B, Activeness (defined as the probability of movement during the sensation period) based on body (light green; left) or tongue (dark green; right) movements in naive and expert cases. C, Movement onset (defined as the first time frame exceeding movement threshold; relative to texture stop) based on body (light green; left) or tongue (dark green; right) movements in naive and expert cases. D, Neuronal response in the cortex and amygdala divided into active (colored red or blue traces) or passive (gray traces) Hit trials in naive (left) and expert (right) cases. Error bars depict SEM across recording sessions. E, Grand average responses in the cortex and amygdala averaged during the sensation period for active and passive trials in naive (left) and expert (right) cases. Error bars depict SEM across recording sessions (n = 15 and 28 for naive and expert cases; 8 mice for cortex, 4 mice for BLA, 4 mice for CEM). F, Example trial displaying body (light green; bottom) and tongue (dark green; top) movements. Dashed gray line indicates movement threshold. Arrows indicate movement onset. Body movement precedes tongue movement. G, Body movement onset versus tongue movement onset for each recording session. *p < 0.05; **p < 0.01; ***p < 0.001; n.s., not significant.
Data analysis
Data analysis was performed using Matlab software (MathWorks). Wide-field fluorescence images were sampled down to 256 × 256 pixels, and pixels outside the imaging area were discarded. This resulted in a spatial resolution of ∼40 μm/pixel and was sufficient to determine cortical borders, despite further scattering of emitted light through the tissue and skull. Each pixel and each trial the ΔF / F was calculated by dividing the raw signal to the baseline signal several frames before the stimulus cue (frame 0 division; −2.2 to 2.1 s relative texture stop). Fiber photometry signals underwent the same preprocessing protocol. In this study, we focused mainly on the sensation periods ranging from −1 to 1 s relative to texture stop. In this period, we observed an initiation of neuronal responses and body movements. To define cortical areas, we first registered cortical activity maps onto a 2D top view mouse atlas using skull coordinates and functional activity patch in the BC [during sensation; ©2004 Allen Institute for Brain Science. Allen Mouse Brain Atlas. Available from: http://mouse.brain-map.org/ (Oh et al., 2014; Musall et al., 2019; Gilad and Helmchen, 2020)]. Within the atlas borders, we defined four cortical areas of interest, barrel cortex (BC), secondary somatosensory cortex (S2), rostrolateral association cortex (RL; part of posterior parietal cortex), and somatosensory forelimb cortex (FL), with some manual modifications within these borders to fit the functional activity for each mouse (Fig. 2A).
Movement analysis
To quantify body movements, first a binary movement vector was defined for each trial by crossing a fixed threshold (Fig. 3A; mean ± 2SD of baseline movement throughout the session). This resulted in a movement probability vector for each recording session. Activeness was defined as the probability of movement (either body or tongue) during the sensation period (Fig. 3B; −1 to 1 s relative to texture stop). Activeness was calculated for each recording session separately. Movement onset was defined as the first time frame exceeding the movement threshold in each trial (Fig. 3A,F). To link neuronal responses to movement, we first separated Hit trials into active or passive, based on the binary movement vector. Hit trials in which the mouse moved for at least 0.5 s within the sensation period (−1 to 1 s relative to texture stop) were defined as active trials and otherwise as passive trials. Next, we plotted and compared neuronal responses between active and passive trials (Fig. 3D,E). To quantify the temporal relation between body movements and neuronal responses, we next aligned neuronal responses in each area and Hit trial to the movement onset (Fig. 4). We then averaged responses across trials to obtain a movement-triggered average response for each area separately (Fig. 4B). The latency of neuronal responses (in relation to movement onset) was defined as the first time frame exceeding mean ± 2SD of baseline response (Fig. 4B; −1 to −0.5 relative to movement onset). A similar analysis was done for tongue movement instead of body movement (Figs. 3B,C, 4E).
Neuronal responses precede movements in expert mice. A, Example trials aligning body movement (green) with neuronal responses in the cortex (blue) and amygdala (red). Movement onset is marked with a dashed green line. Latency of neuronal responses is marked with an arrow. B, Movement-triggered average in an example recording session in the BC, CEM, and BLA. Dashed black lines indicate mean ± 2SD of baseline activity. Latency of response (first time frame exceeding mean ± 2SD of baseline activity) is marked with an arrow. Error bars depict SEM across trials (n = 79 and 64 for CEM and BLA, respectively). C, Grand average of movement-triggered average in the cortex (BC, RL, S2, and FL) and amygdala (CEM and BLA). Error bars depict SEM across recording sessions (n = 15 and 28 for naive and expert cases; 8 mice for cortex, 4 mice for BLA, 4 mice for CEM). D, Latency in the cortex (BC, RL, S2, and FL) and amygdala (CEM and BLA) during the naive (left) and expert (right) cases. Error bars as in C. E, Similar plot as D but for tongue movement. Latency for each area in naive and expert cases. Error bars as in D. *p < 0.05; **p < 0.01; n.s., not significant. Signed-rank test.
Correlation analysis
To study the relationship between the whole posterior cortex and amygdala, we calculated a seed correlation map. To do this, we correlated (Pearson's coefficient) the response in an amygdala area (either CEM or BLA) with each pixel in the posterior cortex during the early sensation period (−1 to 0 s relative to texture stop). This analysis was done for each Hit trial separately, and example averages across trials are shown in Figure 5B. Each pixel depicts the correlation coefficient (r) between the amygdala seed area and that specific pixel. This analysis was done separately for expert and naive cases (Fig. 5C). A similar analysis was done for a later sensation period (Fig. 5D,E; −0.5 to 0.5 s relative to texture stop).
The CEM and BLA are positively and negatively correlated with the cortex, respectively. A, Example trials displaying responses in amygdala and cortical areas. During the sensation period (shaded area) BLA displays opposite dynamics as the cortex (left) whereas the CEM display similar dynamics as the cortex (right). B, Example CEM and BLA seed maps. Colors depict correlation coefficient (r) between the seed area (BLA or CEM) and each pixel in the posterior cortex during the sensation period. C, Correlation coefficient between the seed area (CEM or BLA) and different cortical areas (BS, RL, S2, and FL) during the sensation period for the naive and expert cases. Error bars depict SEM across recording session (n = 11 and 14 for naive and expert cases; 6 mice for cortex, 3 mice for BLA, 3 mice for CEM). D, Example trial displaying BLA and BC responses. Early (orange; −1 to 0 s relative to texture stop) and late (purple; −0.5 to 0.5 s relative to texture stop) sensation periods are marked above. Early period displays high correlation between areas. E, Correlation coefficient between the CEM (left) and BLA (right) to different cortical areas (BC, RL, S2, and FL) during early (orange) and late (purple) sensation periods. Error bar depicts as in C. F, The full correlation matrix between cortical areas during the early sensation period for the naive (left) and expert (right) cases. In general, we found positive corticocortical correlations during the naive case which are increased in the expert case. *p < 0.05; **p < 0.01; ***p < 0.001; n.s., not significant. Signed-rank test.
Statistical analysis
A Wilcoxon signed-rank test was used to compare a population's median to zero (or between two paired populations). For nonpaired populations, we used a ranked sum test to compare between medians. We pooled recording sessions from all mice in CEM and BLA separately (exact numbers in the appropriate figure legends). Multiple group correction was used when comparing between more than two groups.
Histology
Mice were given an overdose of Pental and were perfused transcardially with phosphate-buffered saline (PBS) followed by 4% paraformaldehyde (PFA) in PBS. Brains were postfixed for 12–24 h in 4% PFA in PBS and then cryoprotected for >24 h in 30% sucrose in PBS. Then, 100 mm coronal slices of the entire brain were made using a freezing microtome (Leica SM 2000R), incubated for 15 min in 2.5 mg/ml of DAPI (4′,6-diamidino-2-phenylindole), mounted onto glass slides, and imaged using an Olympus IX-81 epi-fluorescent microscope with a 4× and 10× objective lens (0.16 and 0.3 NA; Olympus). Fiber tracks and GCaMP6f expression were detected in all mice and were mainly localized to the CEM or BLA (Fig. 1A,B).
Data and code availability
All data reported in this paper will be shared by the lead contact upon request.
Results
Learning-related dynamics in the amygdala and posterior cortex during sensation
Our main goal is to study the learning-related spatiotemporal dynamics between the cortex and amygdala underlying sensory discrimination. In the cortex, we focused on posterior areas, including the barrel cortex (BC), secondary somatosensory cortex (S2), forelimb cortex (FL) as well as an associative region, rostro lateral cortex (RL). In the amygdala, we targeted either the BLA or the CEM. To target the amygdala and cortex in the same mice, we combined wide-field calcium imaging to examine neuronal population responses across the posterior dorsal cortex (Gilad and Helmchen, 2020), together with fiber photometry (Gilad et al., 2020), to record neuronal population dynamics in the amygdala (Figs. 1A, 2A). In short, first we injected a calcium indicator to areas in the cortex and amygdala in wild-type mice (see Materials and Methods). Then we implanted a cranial window over the cortex (5 mm over the posterior part) and inserted an optical fiber from the other hemisphere at a 55° angle to reach either the BLA or CEM (Figs. 1A,B, 2A).
After recovery and handling, 10 mice were trained to either lick (go stimuli) or reject licking (no-go stimuli) upon whisker touch with a coarse or smooth surface sandpaper (Fig. 1C,D; Materials and Methods). In Hit trials, mice were rewarded with a water drop for correctly licking the go texture. Trials where mice did not lick for the go stimuli were defined as Miss trials. Mice were punished with white noise for falsely licking for no-go texture (FA; false alarm). Withholding licking upon no-go stimuli was labeled as correct rejection (CR). Mice were trained daily with a single session per day that consisted of ∼350 trials, until they reached expert level (Fig. 1E; defined as exceed a d-prime of 1.5). Importantly, we continuously recorded neuronal dynamics in the cortex and amygdala as mice learned the task across 5–14 d (Fig. 1E).
We analyzed the temporal dynamics of amygdala and cortical activity with an emphasis on the sensation period, when texture reaches the whiskers (−0.5 to 0.5 s relative to texture stop; Gilad and Helmchen, 2020). Figure 2A shows cortical activation maps during sensation for Hit trials in the expert case. Somatosensory-related regions show activity patches in BC, RL, and to some extent S2 and FL. Next, we compared Hit versus CR responses in naive and expert cases (Fig. 2B). During the naive case, responses in the cortical areas (BC, RL, S2, and FL) and amygdala (BLA and CEM) were similar for Hit and CR. In contrast, in the expert case, cortical areas display enhanced activity in Hit compared with CR trials and also compared with the naive case. Notice that differences between Hit and CR trials appear a second before texture stop, since the mouse actively whisks and touches the texture as it approaches (first touch ranged from −1.2 to −0.2 relative to texture stop), obtaining information on texture type (Gilad et al., 2018; Gilad and Helmchen, 2020). In the amygdala during the naive case, both CEM and BLA were relatively nonresponsive during the sensation period. In the expert case, we observed a divergent response specifically in Hit trials, in which CEM displays an enhancement whereas BLA displays a decreased response (Fig. 2B). Averaging across mice and during the sensation period, we found that these differences in the cortical and amygdala areas between Hit and CR trials are statistically significant only in the expert case and not in the naive case (Fig. 2C; p = 2.75 × 10−10, 2.01 × 10−06, 6.77 × 10−11, 5.83 × 10−6, 5.03 × 10−6, 1.25 × 10−4 in expert case for BC, RL, S2, FL, CEM, and BLA respectively; p = 0.67, 0.51, 0.6, 0.2, 0.91, 0.92 for the naive case; signed-rank test; Bonferroni corrected). To control for possible contamination from noncalcium-dependent signal, we trained three additional mice using an interleaved imaging protocol of 473 and 405 nm (isosbestic) excitation lights. These mice were implanted with a 5 mm window over the posterior cortex similar to the mice in this study (Fig. 2D). Correcting for noncalcium signals (473 light minus 405 light within each trial) maintained the results in all cortical areas (Fig. 2E,F; for expert case p < 0.001 for all areas in each mouse separately; for naive case p > 0.05 for all areas in each mouse separately). Taken together, these results indicate that cortical and amygdala dynamics are made prominent as the mouse gains expertise.
Effect of body movement on learning-related neuronal responses
In previous studies, it has been shown that body movements of the mouse during the task have a substantial effect on cortical (Gilad et al., 2018; Musall et al., 2019; Stringer et al., 2019; Gilad and Helmchen, 2020; Gallero-Salas et al., 2021) and subcortical (Steinmetz et al., 2019; Gilad et al., 2020) neuronal responses. During learning, mice tend to exert more body movement and licking as they become experts (Gilad and Helmchen, 2020). Thus, we next examined the effect of body movements on learning-related modulations in the cortex and amygdala. We calculated the body movement of the mouse for each trial and divided trials based on whether the mouse moved (i.e., active trial) or not (i.e., passive trial) during the sensation period (Fig. 3A; see Materials and Methods). To further quantify this, we calculated a binary movement vector (either moving or not for each time frame; exceeding ±3SD from baseline) and plotted the movement probability for naive and expert cases (Fig. 3A). We then quantified activeness for each mouse as the mean probability of movement during the whole sensation period (−1 to 1 relative to texture stop). We found that mice were significantly more active as they gained expertise, reaching levels of ∼70% activeness in the expert case, compared with ∼30% in the naive case (Fig. 3B; p = 9.58 × 10−8; signed-rank test). This result was similar when calculating activeness based only on the tongue movement (i.e., licking; dark green; p = 2.09 × 10−8; signed-rank test). We also found a positive correlation between activeness and performance (r = 0.67; p = 5.48 × 10−7). In addition, the onset of body movement (i.e., the first time frame exceeding threshold) was significantly lower in expert compared with naive mice, where expert mice started to move as the texture approached their whiskers (Fig. 3C; p = 7.21 × 10−9 for body; p = 4.58 × 10−8 for tongue; signed-rank test). Movement onset was also negatively correlated with performance (r = 0.63; p = 0.008).
Next, we divided Hit trials into active and passive trials and plotted responses in the cortex and amygdala for naive and expert cases (Fig. 3D; Materials and Methods). In the cortex, both active and passive trials displayed responses in naive and expert cases. During the sensation period, responses in active trials were significantly higher in most areas compared with passive trials, in the expert but not in the naive (Fig. 3E; p = 0.0013, 0.08, 0.0022, 0.038, 0.004, 0.0004 in expert case for BC, RL, S2, FL, CEM, and BLA, respectively; p = 0.066, 0.02, 0.13, 0.062, 0.31, 0.17 for the naive case; signed-rank test; Bonferroni corrected). In the amygdala, neuronal responses during the sensation were present mostly in active trials specifically during the expert case. During the sensation period, active compared with passive trials displayed significantly higher responses in the CEM and significantly lower responses in the BLA, specifically in the expert case and not in the naive (Fig. 3E; p < 0.01; signed-rank test; Bonferroni corrected). This analysis revealed that neural responses, especially in the amygdala, are at least partially affected by body movements of the mouse during expert task performance. We further found that body movements preceded tongue movement, specifically in expert sessions (Fig. 3F; −0.46 ± 0.08 ms for body movement and −0.13 ± 0.08 ms for tongue movement, relative to texture stop; p = 0.0053 for differences; signed-rank test). There was also a positive correlation between body and tongue movement onsets (Fig. 3G; r = 0.72; p = 1.02 × 10−5). This suggests that as mice become experts, they adopt an uninstructed and mostly stereotypic movement sequence starting with body movement as the texture approaches the whiskers followed by a tongue movement toward the lick detector to collect the upcoming reward. This stereotypic movement pattern may aid in successfully performing the task, i.e., making a correct association.
To further investigate the link between neuronal responses and body movements, we analyzed their temporal relationships. Example single Hit trials of body movements along with responses in the cortex and amygdala clearly show that the latter precede the former (Fig. 4A). To quantify this, we first aligned neuronal responses in each area and each Hit trial to the onset of movement. Next, we averaged the aligned neuronal responses across Hit trials to obtain a movement-triggered average response (see Materials and Methods). Figure 4B displays movement-triggered averages from the BC, CEM, and BLA in an example recording session. We defined the latency of neuronal response (relative to movement onset) as the first time frame exceeding mean ± 2SD of the baseline response (−1 to −0.5 s relative to movement onset). Latency was found to be several hundreds of milliseconds before movement onset in both the cortex and amygdala. Neuronal modulation before movement onset was also observed at the grand average level across mice mainly in the expert case and to a much lesser extent in the naive case (Fig. 4C). The latency across mice was significantly negative, i.e., preceded movement onset by ∼200 ms, in all cortical (BC, RL, S2, FL) and amygdala (CEM and BLA) areas only in the expert case (Fig. 4D; p = 0.0003, 6.24 × 10−5, 0.006, 0.0015, 0.048, 0.036 in expert case for BC, RL, S2, FL, CEM, and BLA respectively; signed-rank test; Bonferroni corrected). In the naive case, latencies were not significantly different than 0 (i.e., occurred around movement onset; p = 0.26, 0.94, 0.87, 0.77, 0.31, 0.96 in the naive case for BC, RL, S2, FL, CEM, and BLA, respectively; signed-rank test; Bonferroni corrected). We performed a similar analysis on the tongue movement (instead of body movement) and found latencies in the expert case to be significantly lower by ∼200 ms compared with body movement (Fig. 4E). These results identify that cortex and amygdala areas initiate neuronal responses before body and tongue movements but only after a correct association between stimulus and reward is learned.
Correlation between the amygdala and cortex emerges following learning
Next, we investigated the relationship between the cortex and amygdala. Single trial examples show that responses in the CEM are positively correlated with cortical areas, especially during an early sensation period in which the whiskers first touch the texture and neuronal responses initiate (Fig. 5A; −1 to 0 s relative to texture stop). In contrast, responses in the BLA are negatively correlated with cortical areas. To continue this line, we calculated a seed correlation map by correlating responses during the early sensation period in the CEM or BLA (i.e., seed area) with each pixel in the posterior cortex, resulting in a map of Pearson's correlation values (see Materials and methods). Example seed correlation maps in Figure 5B show that the CEM is positively correlated with most cortical areas, especially the BC and RL. In contrast, the BLA is negatively correlated with large parts of the posterior cortex. Averaged across mice and in the expert case, during the sensation period, the CEM is significantly and positively correlated with the BC, RL, S2, and FL, whereas th BLA is significantly and negatively correlated with the BC, RL, S2 and FL (Fig. 5C; p = 0.0012, 0.0017, 0.0037, 0.004 for CEM seed with BC, RL, S2, and FL, respectively; p = 0.0052, 0.0057, 0.0066, 0.0038 for BLA seed with BC, RL, S2, and FL respectively; signed-rank test; Bonferroni corrected). In the naive case, correlation values in both t CEM and BLA seed areas were not significantly different from 0 (Fig. 5C; p = 0.36, 0.95, 0.91, 0.64 for CEM seed with BC, RL, S2, and FL, respectively; p = 0.75, 0.89, 0.92, 0.86 for BLA seed with BC, RL, S2, and FL, respectively; signed-rank test). In addition, correlation between the amygdala and cortex at a later sensation period (−0.5 to 0.5 s relative to texture stop) decreased in magnitude and in some cases were not significantly different than 0 (Fig. 5D,E; p = 0.04, 0.011, 0.0052, 0.2 for CEM seed with BC, RL, S2, and FL, respectively; p = 0.034, 0.045, 0.016, 0.13 for BLA seed with BC, RL, S2, and FL, respectively; signed-rank test; Bonferroni corrected), suggesting that the amygdala–cortex relationship emerge early during the sensation period where association is thought to occur. Finally, we also investigated the correlation between different cortical areas in naive and expert cases. We found that corticocortical correlations are significantly positive during the naive case (p < 0.01 for all pairs; signed-rank test; Bonferroni corrected) and further increase in the expert case (Fig. 5F; p < 0.001 for all pairs; signed-rank test; Bonferroni corrected). These results emphasize that corticocortical interactions, unlike amygdala–cortical interactions, are present prior to learning and also display learning-related effects.
Discussion
In this study, we addressed the amygdala–cortex relationships as mice learn a sensory discrimination task. By combining wide-field cortical imaging with fiber photometry, we were able to simultaneously record neuronal populations from both large parts of the posterior cortex and different amygdala areas (BLA or CEM). We found the following: (1) the amygdala and cortex developed discriminating responses to the reward-associated stimuli (between Hit and CR) post-learning. (2) These responses preceded a stereotypic movement execution during early sensation involving body movement followed by licking for a reward. (3) Widespread positive (CEA) and negative (BLA) correlations with the posterior cortex were present during early sensation and emerged only after learning. We suggest that learning recruits amygdala and cortical ensembles that drive goal-directed behavior.
The amygdala is dynamically regulated in texture discrimination learning
Previous studies have highlighted the role of the amygdala in sensory discrimination learning, using vision, auditory, taste, and olfactory sensory systems (Gutierrez et al., 2010; Samuelsen et al., 2012; Lichtenberg et al., 2017; Beyeler et al., 2018; Reed et al., 2018; Steinberg et al., 2020). However, its relationship with the posterior cortex (primarily sensory cortex) during learning a texture discrimination task has not been explored. The amygdala is a principal pathway, connecting the limbic system to the cortex. The BLA and CEM receive sensory and associative information from the posterior cortex and their neurons project to many effector brain regions in the anterior cortex and subcortex, among them the gustatory cortex, orbitofrontal cortex, and the nucleus accumbens, which are instrumental for reward-related licking behavior (Gutierrez et al., 2010; Samuelsen et al., 2012; Lichtenberg et al., 2017; Beyeler et al., 2018; Reed et al., 2018; Steinberg et al., 2020). Thus, the positioning of the amygdala at the crossing point of associative learning and sensorimotor transformation implies that amygdala areas have a pivotal role in learning a texture discrimination task.
Here, we show that amygdala neurons in the CEM and BLA were nonresponsive to the incoming texture prior to learning, i.e., in the naive case. In contrast, as learning occurred and an appropriate association was made between a certain texture and an upcoming reward, the CEM and BLA developed robust and discriminating responses. Importantly, amygdala responses were strongly associated with active sensation, i.e., trials in which the mouse rigorously moved and whisked toward the incoming texture. In addition, amygdala response latencies preceded movement onset, specifically in the expert case. These results suggest that amygdala areas have a leading role in associating a certain stimulus with a specific outcome. Interestingly, amygdala subregion responses were opposite, while BLA neurons decreased their activity to the incoming texture and CEM neurons showed increased activity. These response patterns resemble previous findings showing increased activation of BLA inhibitory neurons and decreased activation of BLA excitatory neurons in auditory discrimination learning (Zhang and Li, 2018), as well as reduction in BLA excitatory input to the nucleus accumbens during reward-seeking behavior (Reed et al., 2018). Suppression in the BLA during sensory association could also be explained by an active silencing of a Pavlovian fear circuit (Davis, 1992; LeDoux, 2000; Ehrlich et al., 2009; Duvarci and Pare, 2014) which may interfere with the positive process of acquiring a reward. On the contrary, enhancement of CEM responses may send information to midbrain areas and be linked to action execution which are part of the sensorimotor transformation (Fadok et al., 2018; Gothard, 2020). Taken together, amygdala early dynamics display divergent responses only after learning which imply its crucial role in associating a specific stimulus with a certain reward.
Amygdala dynamics precede movement initiation and licking for a reward
Classically, the amygdala is thought to be strongly linked to valence, i.e., how a certain stimulus is good (e.g., related to a reward) or bad (e.g., related to a punishment; Weiskrantz, 1956; Blanchard and Blanchard, 1972; LeDoux et al., 1990; O’Neill et al., 2018). But a majority of studies, using, for example, Pavlovian conditioning paradigms, are accompanied with a change in a motor output such as licking or freezing. Despite this co-occurrence, a direct link between amygdala responses and motor execution is mostly overlooked in many amygdala studies. In cortex, it is now well established that body movements in mice have a profound effect on neuronal responses in many cortical areas, well beyond the motor cortex (Gilad et al., 2018; Musall et al., 2019; Stringer et al., 2019; Salkoff et al., 2020; Gallero-Salas et al., 2021). Here, we show that neuronal responses in the BLA and CEM are strongly related to body movement, since most neuronal responses are present almost exclusively during active (i.e., rigorously moving) trials. Are amygdala responses solely related to body movements? First, amygdala responses are less present in active trials during the naive case, i.e., before a proper association was made (Fig. 3E). Second, responses in the amygdala precede body movement initiation by ∼200 ms and initiate early during the sensation periods (Fig. 4). Taken together, it seems that both the BLA and CEM are modulated before body movements which are part of a stereotypic pattern present only in expert mice. This pattern starts with a modulation in the amygdala and cortex, followed by movement initiation toward the incoming texture (after ∼200 ms) which results in an active sensation period. Then, the tongue is taken out toward the lick port (after ∼400 ms), and reward comes out only 2 s after the texture stops. Therefore, we found that amygdala responses encode the motivational aspect of a conditioned stimulus and are indirectly related to movement, licking, and reward. In contrast, cortical responses were only partially related to movement, sensation, and reward and seem to have a more broader response profile. These results further emphasize the role of the amygdala in encoding valence but also highlight the importance of relating other parameters (such as body movement and licking) to neuronal responses.
Cortical–amygdala interactions mediate learning
Neuronal responses in the posterior cortex have differences and similarities with amygdala responses. Unlike in the amygdala, cortical responses are present also in the naive case during the sensation period. In addition, different cortical areas such as the BC display substantial responses also in passive trials in which the mouse quietly waited for the texture to approach its whiskers. Thus, it seems that cortical responses are at least partially related to sensory parameters regardless of cognitive effects such as learning or active sensing (Feldmeyer et al., 2013; Petersen, 2019). In addition, cortical areas that were measured in this study (i.e., BC, RL, S2, and FL) all displayed an increase in response to the stimulus and also in active trials (compared with passive trials), implying that these cortical regions belong to the same subnetwork that processes the incoming stimulus.
One of the main similarities between the cortex and amygdala is that both structures display responses that precede movement initiation by ∼200 ms. The CEM is positively correlated with cortical responses, implying that the CEM and posterior cortex are part of the same subnetwork underlying sensorimotor integration and may involve other brain areas such as the frontal cortex, striatum, and anterior thalamus (T-W. Chen et al., 2017; Peters et al., 2021; Inagaki et al., 2022). In contrast, we found that the BLA was negatively correlated with cortical areas, which may indicate that the BLA receives an inhibitory effect from one of the nodes in the cortical network, further implying that the BLA does not belong to the cortical network. Instead, the BLA may belong to a different subnetwork, involving other areas such as the medial geniculate thalamus and nucleus accumbens, and may be responsible for aversive learning (Davis, 1992; Ehrlich et al., 2009; Duvarci and Pare, 2014; Mátyás et al., 2014; Cambiaghi et al., 2016). This alternative network which is suppressed during reward-related associative learning could emerge in aversive scenarios where certain stimuli may indicate a punishment of some sort. In this case, it could be that sensory integration may quickly bypass the standard cortical network through an alternative processing route involving the medial geniculate thalamus and the limbic system (LeDoux et al., 1983; Iwata et al., 1986; Taylor et al., 2021; Khalil et al., 2023). Thus, we propose the existence of two opposing brain-wide subnetworks that operate in a pull–push fashion and may facilitate learning.
Limitations of the study
This study measures neuronal population dynamics from the cortex and amygdala, overlooking single cell dynamics. Thus, the measured bulk signals could miss out important and locally intermixed subnetworks that average out at the population level (Harvey et al., 2012; J. L. Chen et al., 2013; Gründemann et al., 2019). In addition, our injected virus targets all neuronal subtypes and does not allow to further dissect the exact neurophysiological mechanisms, e.g., inhibitory subtypes. We emphasize that this work outlines interesting hubs within brain-wide subnetworks which will guide future studies which may (1) use neuronal specific indicators with the same imaging system or (2) combine GRIN lens (Gründemann et al., 2019; Taylor et al., 2021) with multiarea two photon imaging (Sofroniew et al., 2016) or (3) implement neuropixel probes to target single cells in multiple areas (Steinmetz et al., 2021). Another concern is that the measured bulk signal may reflect axons originating from long-distance areas (e.g., measurement in the amygdala could pick up axonal signals originating from the cortex), thus potentially contaminating the measured signal and bias the correlation analysis. Although axonal imaging was performed in several studies (Petreanu et al., 2012; Roth et al., 2016; Pardi et al., 2020), it has also been reported that the calcium indicator used in this study diffuses poorly to distal axons (Broussard et al., 2018). In addition, we report that neuronal dynamics in the amygdala are similar in cases where there is no injection of a calcium indicator in the cortex (Fig. 6). These results imply that neuronal dynamics observed in this study are rather local, reflecting cell bodies and local dendrites.
Amygdala responses without cortical injections. A, A coronal slice example from one mouse, in which we injected the calcium indicator only in the CEM area, but not in the cortex. Fiber is localized in the CEM. B, Responses in CEM for an expert mouse for Hit (red) and CR (black) trials. CEM dynamics are comparable with CEM responses with cortical injections (Fig. 2B). Error bars indicate mean ± SEM over trials (n = 93 and 89 trials for Hit and CR trials, respectively).
Another limitation of this study is that it only measures from a limited part of the brain (amygdala and posterior cortex), whereas other brain areas may be crucially involved in learning a sensory discrimination task, for example, hippocampus, thalamus, striatum, and more. It is further possible that amygdala learning-related modulation may have a subcortical origin rather than the posterior cortex. Therefore, it is of interest to expand our simultaneous observation to many more brain areas, with emphasis on subcortical areas. Indeed, a growing number of studies implement large-scale simultaneous recording using either multiple high-density probes (Steinmetz et al., 2019), multifiber photometry (Sych et al., 2019), or combining deep neuronal recordings with wide-field cortical imaging (Xiao et al., 2017; Peters et al., 2021; Pedrosa et al., 2022). In addition, this study is correlative (i.e., observational) and future studies may further probe the causal relationship between the amygdala and cortex. In summary, this study explores the cortex–amygdala relationships highlighting complex and dynamic subnetworks during learning that merit further dissection and expansion.
Footnotes
We thank Kelly Ohayon and Elad Avidan for help with data analysis. This project is funded by the European Union (ERC Starting Grant, MESO-AG, 101040378) and a Hebrew University Start-up Grant.
The authors declare no competing financial interests.
- Correspondence should be addressed to Ariel Gilad at ariel.gilad{at}mail.huji.ac.il.