Spatiotemporally distinct dopamine domains in the hippocampus

Abstract

The neuromodulator dopamine plays a critical role in orchestrating behavior by signaling actions and rewards. Dopamine signaling in the hippocampus is necessary for the formation of long-term memories but the functional correlates, anatomical organization, and spatio-temporal scale of hippocampal dopamine release remain largely unknown. Using high resolution, two-photon imaging of optical sensors in the hippocampus of male and female mice, we identified dopamine release in two closely apposed spatial domains. In Pavlovian conditioning, there was no significant dopamine release in early training but over days to weeks of experience, dopamine transients appeared at rewards in the “deep” domain of CA1 (basal dendritic layer). Surprisingly these transients did not strongly encode reward. The act of stopping and licking, used to collect reward, was sufficient to produce deep dopamine transients even without the conditioned stimulus or reward. Similarly, hippocampal dopamine transients showed no reward prediction error at surprise rewards. In a spatial goal-directed task, deep dopamine domain transients at rewards persisted but dopamine ramps now appeared, ramping up in the superficial domain (cell body and apical dendrites) during reward approach while ramping down in the deep domain. Our results reveal anatomically segregated hippocampal dopamine release domains with dynamic signaling that is distinct from striatal dopamine. Rather than a unitary volume and function of dopamine, we identify multiple spatial domains and functional signals that likely play distinct and dissociable roles in hippocampal-dependent learning and memory.

Introduction

Dopamine has been intensively studied in the striatum where it is broadly associated with both rewards and actions (Schultz et al., 1997; Jin and Costa, 2010; Howe et al., 2013; Howe and Dombeck, 2016; da Silva et al., 2018; Engelhard et al., 2019; Mohebi et al., 2019; Kim et al., 2020). Functional differences in dopamine signals coarsely align with anatomy; movement correlated dopamine is more prevalent in dorsal striatum which is primarily innervated by afferents from substantia nigra pars compacta (SNc), while reward-related functions such as reward prediction error (RPE) and motivation are more prominent in the ventral striatum, which is primarily innervated by the ventral tegmental area (VTA). Far less is understood about dopamine's roles outside of the striatum, where lower dopamine levels make accurate measurement difficult.

Since the hippocampus is necessary for forming long-term declarative memories, dopamine has been hypothesized to play an important role within local circuits to aid memory formation by reinforcing reward-related behaviors. Pharmacological inhibition of hippocampal excitatory D1/D5 receptors impairs memory formation and retrieval (O’Carroll et al., 2006; Tang and Dani, 2009), and contextual or trace fear conditioning (Heath et al., 2015; Tsetsenis et al., 2021; Wilmot et al., 2023). Similar deficits in memory were seen in D1 knockouts (Ortiz et al., 2010; Sariñana et al., 2014). The inhibitory D2 receptor was also identified as necessary for spatial memory and intact synaptic plasticity using pharmacology (Caragea and Manahan-Vaughan, 2022) or knockout mice (Espadas et al., 2021). Thus, dopamine signaling is critical to hippocampal function. At a mechanistic level, numerous studies have found that dopamine is necessary for hippocampal-dependent learning, perhaps by stabilizing long-term plasticity or memory consolidation (Frey and Morris, 1997; Li et al., 2003; McNamara et al., 2014). In addition to VTA and SNc dopaminergic innervation (Gasbarri et al., 1994a, 1994b; McNamara et al., 2014; Adeniyi et al., 2020; Tsetsenis et al., 2021; Krishnan et al., 2022; Sayegh et al., 2024), locus coeruleus neurons also innervate the hippocampus, releasing both dopamine and norepinephrine (Kempadoo et al., 2016; Takeuchi et al., 2016; Heer and Sheffield, 2024; Wilmot et al., 2024). Functionally, both midbrain (McNamara et al., 2014; Tsetsenis et al., 2021) and locus coeruleus (Kempadoo et al., 2016; Takeuchi et al., 2016; Wilmot et al., 2024) hippocampal dopamine release contribute to distinct behaviors, although the exact functional role of different dopamine sources remain unclear.

A comprehensive picture of hippocampal dopamine function remains elusive since fundamental information is missing: when and where is dopamine released in the hippocampus? Here we address two questions. First, what are the functional correlates of dopamine release in the hippocampus? Does it represent rewards similar to dopamine in the ventral striatum? Another distinct possibility is that dopamine conveys movement-related signals as in the dorsal striatum, either as a general locomotion signal or marking specific actions. Movement is critical to hippocampal function as it defines network state, with periods of locomotion marked by place cell firing, while immobility switches the network to replay and consolidation. Furthermore, growing evidence identifies specialized neuronal circuits specifically active during periods of movement and immobility (Arriaga and Han, 2017; Dudok et al., 2021). Alternatively, the hippocampus could contribute to action selection, either by sending spatial or trajectory information to the striatum to inform appropriate choices or by playing a more direct role in gating or selecting action(Pennartz et al., 2004; Delcasso et al., 2014; Trouche et al., 2019; Barnstedt et al., 2024; Zutshi et al., 2025).

Second, what is the spatial scale of dopamine domains? Dopamine is a volume transmitter, with no pre- to postsynaptic alignment; however, the size of functional domains could vary by orders of magnitude, from single microns to millimeters (Liu and Kaeser, 2019). Defining the spatial scale of dopamine domains is critical for interpreting the results of fiber photometry experiments, which have no optical sectioning ability but instead take the mean of fluorescence over hundreds of microns.

Materials and Methods

Procedures were similar to those previously described (Arriaga and Han, 2017, 2019)

Animals

All experiments were approved by the Washington University Animal Care and Use Committee. Wild-type mice of both sexes on a C57BI/6J background were used. Mice were 2–3 months old at the time of surgery.

Viral injection and hippocampal window implantation

Mice were anesthetized with isoflurane, after which a 0.3 mm craniotomy was preformed above the left cortex. Viral vectors, including AAV-hsyn-GRAB_DA1h (2.0 × 10¹³, diluted 1:1 with PBS, mice 156 and 157), AAV-hsyn-GRAB_DA2h (2.6 × 10¹³, diluted 1:1–5 with PBS, mice 167, 168, 169, 171, 231, 232), AAV-hsyn-GRAB_DA2m (2.6 × 10¹³, diluted 1:1–5 with PBS, mice 220, 221, 222), GRAB-NE1h (2.6 × 10¹³diluted 1:1–5 with PBS, mice 223, 224, and 225),AAV-hsyn-GRAB_DA-mut (2.0 × 10¹³, diluted 1:1–5, mice 170, 179, 181, T10, and T11), and AAV2-CAG-Flex-eGFP-WPRE-bGH (8.2 × 10¹², diluted 1:10, mouse 158) were pressure injected (~50nL) through a beveled micro-pipette targeting dorsal CA1 (−2.4 mm posterior, −1.6 mm lateral, −1.3 mm ventral, relative to Bregma). Mice were water scheduled 2–3 weeks post virus injection. Afterward, an imaging cannula (2.8 mm diameter) was implanted above CA1 by aspirating the overlying cortex. Mice were allowed to recover at least two weeks after surgery before the commencement of behavioral training and imaging.

Mouse composition for various experiments were: Pavlovian, 4 GRAB-DA2h, 3 GRAB-DA2m, 5 GRAB-DAmut, 3 GRAB-NE1h; reward omission, 6 GRAB-DA2h; uncued UCS, 5 GRAB-DA2h; HRZ, 2 GRAB-DA1h, 6 GRAB-DA2h, 4 GRAB-DAmut, 1 EGFP; early Pavlovian lick vigor analysis, 4 GRAB-DA2h, 2 GRAB-DA2m, two recording sessions analyzed per mouse; late Pavlovian tongue appearance analysis, 4 GRAB-DA2h, 2 GRAB-DA2m, one recording session analyzed per mouse.

Imaging

Imaging was conducted using a laser scanning two-photon microscope (Neurolabware) equipped with an electrically tunable lens (ETL; Optotune, EL-10–30-NIR-LD), enabling rapid focal plane shifts. The microscope control and data acquisition were conducted using Scanbox (Neurolabware). All imaging was in the dorsal CA1 region of the hippocampus. The z-axis was scanned to obtain 3–4 axial planes, corresponding to different layers: Stratum Oriens (SO), Stratum Pyramidale (SP), Stratum Radiatum (SR), and Stratum Lacunosum Moleculare (SLM). Identification of specific strata was achieved through distinct features. SP was recognized by the presence of cell bodies, while SLM was identified based on depth and dark areas corresponding to blood vessels at the hippocampal fissure. Imaging depths for different strata were approximately as follows, relative to surface of the hippocampus: SO at −50 to −80 μm, SP at −110 to −130 μm, SR at −170 to −200 μm, and SLM at −300 to −350 μm. The field of view (FOV) in the x-y plane was approximately 350 × 350 μm. The total frame rate of imaging was 31.25 Hz, resulting in a per-plane sampling rate of 10.4 Hz for a three-plane recording and 7.8 Hz for a four-plane recording.

Laser was set at 920nm, with power ranging from 20 to 39 mW per plane after the objective. Higher powers were used for deeper planes for adequate fluorescence intensity. GRAB-DAmut, the dopamine insensitive control protein, had dimmer dopamine independent fluorescence, requiring higher powers in the range of 39 to 51mW to match pixel intensity with GRAB-DA mice. EGFP (Enhanced Green Fluorescent Protein) was used as a control and imaged at 9mW.

Mice were head-fixed above a cylindrical Styrofoam treadmill mounted on a 3D-printed base. Rotational velocity was tracked using a rotary optical encoder, and licking behavior was measured as a voltage change through the completion of an electrical contact circuit. The conditioned stimulus (CS) was provided by the click of a dummy solenoid located behind the mouse. Reward delivery, administered by a second solenoid, was sound-insulated and positioned several feet away outside the recording rig to ensure mice could not hear the reward solenoid activation. Testing showed that mice could not hear the water dispensing solenoid as they rarely licked unless cued by the CS solenoid.

Pavlovian conditioning

Mice underwent setup habituation by running on a treadmill without water reward delivery for 1–2 weeks. Following habituation, mice were introduced to water rewards as a conditioned stimulus (CS) and subsequent unconditioned stimulus (UCS) after a 0.5 second delay pair at random inter-trial intervals ranging from 15 to 45 seconds. At the start of every session there was designated no-stimulus period of 2 minutes. Water remained available on the lick spigot until consumed at any point during the behavioral session, with most mice quickly associating CS with UCS. Each session, lasting 20–24 minutes, included reward doubling trials (4 trials), where a second CS-UCS paring was added 0.5 seconds after the first. The experimental phase continued for 2–5 weeks.

Reward omission experiments, conducted in a 5 day block after Pavlovian conditioning in a subset of animals, involved two randomly inserted reward omissions in each session, with CS and no UCS. These would occur instead of a normal CS-UCS pair (15–45 s after the last pair, and 15–45 s before the next pair).

Then mice moved into a 5 day block of uncued water reward sessions (UCS with no prior CS). Four of these uncued UCS trials were inserted per session. To maximize the surprise of such delivery, uncued trials were inserted in-between normal pairs, with the added condition of at least 8 s apart from either cued pair. Only trials in which mice detected the water through spontaneous licking before the next CS-UCS pair spontaneous licking were used (~60%).

HRZ task

Following completion of the Pavlovian task, mice were transitioned to a virtual environment controlled by ViRMEn (Virtual Reality Matlab Engine), featuring a visual linear track with both local and distal landmarks, measuring either 180 (for 1h mice) or 270 cm in length. In this context, mice traversed the track, experiencing a brief dark interval of 2–4 seconds upon reaching the far end before initiating a new trial at track start. In the initial stages of training, water rewards were provided when the mouse licked the contact sensor and moved forward a minimum distance of 5 cm. Then rewards were delivered when the animals licked in a larger visual-uncued reward zone or broken into multiple smaller reward zones. Reward zones were shrunk or removed until a single 10 cm reward zone remained. After achieving success in 20–27 consecutive trials, three probe trials without rewards occurred, followed by the relocation of the reward zone. Each 20–27 trials of a specific reward zone period were defined as an epoch. These zones were chosen from early, middle, or late sections of the track, ensuring that sequential reward zones did not originate from the same section. The delineation of early, middle, and late sections was based on specific track position intervals (67–86, 101–120, 135–154 cm for the 180 cm track and proportionally scaled to the 270 cm track), and the center of the reward zone was randomly selected within these intervals. Mice could consistently complete 2–4 epochs per imaging session lasting 20–24 minutes.

Data Analysis

Data were analyzed using custom programs written in Matlab (MathWorks). Video Tiffs of fluorescence were motion correction using Suite2p software rigid and nonrigid registration and checked for motion artifact. The same approximate imaging region across days was identified using local landmarks such as blood vessels. Regions of Interest (ROIs) were drawn on the mean fluorescence from each plane, avoiding bright or dark areas and the edges of frames. ROIs were ~75 × 75 μm and encompassed multiple cells or dendritic trees. The mean fluorescence was subsequently extracted by calculating the average intensity across time for all pixels within each ROI. A Gaussian window with a duration of 0.5 seconds was applied to smooth the fluorescence data for visualization but not for quantification. A psuedocolored heat map was created by subtracting the average pixel intensity 0.5s before CS from the average pixel intensity 0.5s after and overlaid over the standard average pixel intensity. This heat map was also smoothed by a 20 pixel (20x and 20y) Gaussian window to show areas of fluorescent change.

Locomotion was measured with a rotary optical encoder (Yumo E6B2-CWZ3E) read by an Arduino Uno programmed using the Matlab rotary encoder library. Voltage output of the rotary optical encoder was converted to cm, such that virtual track distance corresponded traversal along the treadmill surface. Licks were defined by thresholding input voltage from the contact capacitive sensor circuit. Thresholds varied per mouse due to unique impedance, but remained reliable across sessions.

Locomotion traces were smoothed using a Gaussian window of 5 frames (0.15s). Locomotion stop events were identified as instances when the speed remained below 5 cm/s for longer than 1 second, while locomotion start events were triggered when speeds exceeded 5 cm/s after stopping. Stops were considered to have licks if at any point during the stop lick voltage exceeded threshold.

Triggered Event Analysis

Events such as CS were synchronized with the first subsequent per-plane imaging frame occurring at a frequency of 7.8–10.4 Hz. A time window of −5 to 5 seconds around each event was averaged to create a single trace per day and normalized by the baseline period (−5 to 0 seconds), except VR track behavior peri-CS which were normalized by −1 to −0.5 instead to avoid ramping effects on transient amplitude. Early Days in Pavlovian conditioning were defined as the first two days of imaging. Late Days were defined as the last four imaging sessions. These days are combined and averaged for plotting. Post-CS dopamine amplitudes were determined by averaging values within an early 0.5-second window (t=0–0.5s) and a 1-second Post-UCS window (t=0.5–1.5s) from CS (t=0).

Pre-start and post-stop dopamine amplitudes were measured within a 0.5-second window before and after t=0 respectively. Lick-triggered dopamine windows spanned 500ms centered on t=0 (−0.25 to 0.25s). To align CS-triggered behavior with spontaneous stop and lick or uncued reward events, we used a stop trigger, with an additional temporal offset. A straight CS to stop alignment is off because stopping after CS occurs with a delay. We offset stop traces by each mouse's average latency from CS to stopping by shifting the speed, lick rate, and dopamine backwards in time. The dopamine amplitude was analyzed in two windows. The first was always 0.5s long (t=0–0.5s after CS or equivalent trigger time). The duration of the second window was set to avoid confounding differences in locomotion. For spontaneous stop and lick events, the second window was 0.5s long (t=0.5–1s after CS or equivalent). For reward omit (CS with no UCS), the second window was 0.1s long (t=0.5–1.5s after CS or equivalent). Long reward omission trials were defined by the last lick during the stop occurring at least 0.8s after the first lick. Long spontaneous stop and lick events had a duration of >0.3s. Spontaneous stop and lick events had a shorter duration than reward omission because mice will wait longer to see if reward is delivered after a CS. Other window 2 durations (0.65–1.1s for reward omit and 0.2–0.0.5s for spontaneous stops) showed similar results. We also aligned by first rewarded lick. Here, window 1 was t=−0.5–0 relative to stop and t=0–1. The timing of first reward lick was shifted later so windows moved earlier to correspond to CS and reward collection.

Convolution Error and linear fit R² were measured from t= −0.5 to 2 seconds from the start of both unrewarded and rewarded bouts (see below). Ramping was quantified by applying a simple least squares linear fit to the −5 to 0 seconds period of the peri-CS traces.

Locomotion Correlation Analysis

Acceleration for speed was calculated using the diff of a 1 second Gaussian smoothed version of velocity. Velocity was then broken up into the following non-exclusive categories: stopped, stopping, slowing, starting, speeding up, maintaining speeds, and generally moving. Stopped is defined as previously mentioned as instances when the speed remained below 5 cm/s for longer than 1 second. Generally moving was then defined by all other time points that are at least 1 second away from stopped moments. Periods of deceleration were defined as stretches of time where acceleration was below −0.35 cm/s². Deceleration which ended at a point in which the mice were stopped where considered stopping. All other deceleration periods were than split into ones in which ended at a speed below 5 cm/s but for shorter than 1 second, or those which ended at a speed still higher than 5 cm/s. Acceleration was similar defined by a threshold of 0.13 cm/s² and categorized using identical starting conditions to the end of deceleration. Finally, maintaining speeds consisted of all of the remaining moments in the session. Locomotion correlation was calculated as a pearson correlation between dopamine and speed during these periods, the entire time series, or −1 to 1 seconds of stop or start triggered events.

Isolated, Unrewarded, and Rewarded Lick Bouts

Lick voltage was measured through a lick contact sensor and binarized into contact or no contact by thresholding the response. Consecutive frames of contact were treated as a single lick, and only the first frame of that contact was considered for triggered-events, count, convolution, and other analyses. Lick Bouts were defined by grouping any lick that occurred within 500 milliseconds of another lick. Isolated Licks were then identified as any lick which did not have another lick 500ms before or after. Rewarded Bouts were defined as lick bouts that contained within or first occurred after an UCS. Lick bouts or isolated licks that occurred within 2 seconds after a rewarded bouts were removed from analysis since it was unclear whether these events should be counted as rewarded or unrewarded. Any remaining lick bouts were labeled as unrewarded.

Generalized Linear Model (GLM)

To remove linear locomotion dependencies on dopamine signals, a GLM was applied to a 2-second gaussian smoothed version of fluorescence. Using the Matlab function fitglm and speed, velocity, acceleration, absolute acceleration, and interactions (products) of any permutation as inputs, a fit of fluorescence was created and subtracted from the raw data, creating a “GLM subtracted trace”. Additionally, all predictors were shifted anywhere from 0 to 3 seconds, using NaNs to pad for frames without predictor data, in 0.25 second increments to produce a total of 12 possible fits. The fit with the highest correlation to the original trace was chosen to be subtracted. The resulting GLM subtracted trace was then treated as fluorescence for convolution and bout analysis.

Lick Convolution and Initiation Convolution Models

For every session and plane a dopamine kernel was calculated to convolve with the lick binary signal. For the Lick Convolution model, this kernel was defined as the average unrewarded lick-triggered GLM subtracted trace average for that session from t=−2 to 2 seconds, normalized by subtraction the average from −2 to −1 seconds for each trigger. For the Initiation Convolution model kernel, the same window and normalization was applied instead to the average dopamine of both isolated licks and unrewarded lick bout starts. Kernels were then convolved using the Matlab function conv. For the Lick Convolution model, this kernel was convolved directly with the lick binary signal. For the Initiation Convolution model, this kernel was convolved with the start of every lick bout, essentially any initial lick that occurred with no lick 500ms prior. Licks occurring 2 seconds after a reward bout were convolved to make the final model traces, but not considered in either unrewarded or rewarded bouts/licks for the creation of kernels or goodness of fit.

An unrewarded or rewarded bout-triggered window of t= −0.5 to 2 seconds of dopamine and model traces was used to fit a linear model using fitlm and obtain an adjusted R² for quantification.

Face Video Analysis

Face video was collected at 60Hz under IR illumination and synchronized to two-photon imaging frames and behavior. Facial features of mice recorded during head-fixed behavior were tracked using DeepLabCut 2.2.2, an open-source software package designed for animal pose estimation. 14 mice, some of which were part of an unrelated study, from 19 different behavioral sessions and 169 frames in total were manually labeled to mark 26 face points of mouse pupil, nose, whiskers, chin, tongue, and paw. 95% of the labeled frames were used to train the network and the remaining 5% were used as the test set. The model was trained until the loss converged at 1,000,000 iterations. The model performance, measured by root mean square difference between the manual and the model generated labels, was 1.83 pixels for the training set and 2.63 pixels for the test set. Chin and tongue position was tracked using the x and y of the chin pose point. Frames were filtered for likelihood values (numbers between 0 and 1 that reflect the confidence level of the estimate) greater than 0.99.

Acceleration was calculated using the diff of the diff of 0.5 second smoothed version of the positions. Lick vigor was then measured as the AUC from t=−0.15 to 0.1 seconds where t=0 is the lick contact.

Statistics

Means of data shown with SEM bars or shading. Either Wilcoxon rank sum test, paired-tests, or Wilcoxon signed rank were used. Data reported in text for non parametric tests are median [1st quartile, 3rd quartile], or mean±SEM for parametric tests. One way ANOVA tests were used for multiple comparisons with bonferroni-holm post-hoc correction. Spearman correlation was used for relationships over days. Otherwise, Pearson correlation was used. * = p < 0.05; ** = p < 0.01; *** = p < 0.005; **** = p < 0.001; ***** = p < 0.0005. For specifics, see Supplementary Table 1.

Results

Spatially-specific dopamine transients during Pavlovian conditioning

To precisely measure dopamine dynamics with high spatio-temporal resolution, we recorded with optical sensors, GRAB-DA1h, 2h, 2m, and the ligand insensitive controls, GRAB-DAmut (Sun et al., 2018, 2020) or EGFP using two-photon imaging to image a volume of tissue with multiple axial planes in head-fixed mice during behavior (Fig. 1A, B). Mice sensor composition varied by experiment throughout the paper (see Methods); for Pavlovian conditioning, we used GRAB-DA2h, 2m, and DAmut. Based on results, we split dopamine responses into two domains: the deep domain comprising stratum oriens, the layer corresponding to the basal dendrites of CA1 pyramidal neurons, and the superficial domain, comprised of stratum pyramidale, radiatum, and lacunosum moleculare, corresponding to cell somata and apical dendrites. Note that we use historical hippocampal terminology so that our defined “deep” domain is more superficial or dorsal in the animal, while the “superficial” domain is deeper or ventral. All recordings were done in CA1 of dorsal hippocampus.

Figure 1. Reward-related deep domain dopamine in late Pavlovian conditioning.

A, Schematic of imaging. Inset, approximate imaged area in dorsal CA1. Electric tunable lens (ETL) allows sequential imaging at different focal planes corresponding to anatomical layers. Icon to right shows color code of plane fluorescence traces to layers of the hippocampus. In some subsequent panels, results from SP, SR, and SLM are averaged together as the superficial domain while the SO is the deep domain. B, Schematic of virtual reality (VR) setup. Projection screen not used for Pavlovian conditioning. Not shown, lick spigot for water rewards. C, schematic of Pavlovian conditioning. D, raw data from early Pavlovian conditioning (Day 2) showing fluorescence in different cell layers, along with CS, licks and speed of treadmill. Dotted line = CS. Middle, CS-triggered average of dopamine signal in different layers. Right, pseudocolored images of CS-triggered dopamine overlaid on mean fluorescence image from different planes. Red for increased dopamine and blue for decreased. E, Mean CS-triggered dopamine in early Pavlovian conditioning across mice, along with speed. Individual planes (SP, SR, SLM) are averaged to “Superficial”. Control signal comes from dopamine insensitive version of the sensor, GRAB-DAmut. F, No significant CS-triggered dopamine signals when compared to controls. Icon identifies color code of deep and superficial layers. G, raw data from late Pavlovian conditioning (Day 11) showing dopamine transients at CS. Middle, CS-triggered average of dopamine signal in different layers. Right, pseudocolored images of CS-triggered dopamine overlaid on mean fluorescence image from different planes. H, Mean CS-triggered dopamine in early Pavlovian conditioning across mice, along with speed. F, Significant positive deep dopamine transient at CS and negative superficial domain transient, when compared to controls.

Deep domain dopamine transients emerge with extended Pavlovian conditioning

In a Pavlovian task with randomly timed rewards, we paired a conditioned stimulus (CS, sound cue, brief click) with water reward (unconditioned stimulus, UCS, at 0.5s delay from CS, Fig. 1C). Mice can freely move on the treadmill but locomotion has no effect on the conditioning paradigm. During early Pavlovian conditioning, mice lick often and randomly, indicating poor understanding of the CS-UCS association. Speed on the treadmill is also very low (Fig. 1D). Dopamine signals show little modulation in all recorded layers. To quantify dopamine changes near reward, we made CS-triggered averages and compared amplitudes between experimental and control animals (Fig. Fig. 1E). We averaged signals in the stratum pyramidale, radiatum, and lacunosum moleculare since they were similar into the “superficial” domain, corresponding to the apical dendrite and pyramidal cell layers. There were no significant differences in GRAB-DA mice in either domain from corresponding control mice (Fig. 1F, Note: for non-parametric tests we report median and 1st and 3rd quartiles in brackets, while for parametric tests mean±SEM; deep GRAB-DA, median: −0.03 dF/F, 1st and 3rd quartiles: [−0.17,0.01], N=7 mice, deep control, −0.04 [−0.24,0.03], N=5 mice, p=1, superficial GRAB-DA, 0.18 [0.03,0.36], N=7 mice, superficial control, −0.07 [−0.08,0.02], N=5 mice, p=0.1; Wilcoxon rank-sum).

In contrast, late in Pavlovian training (after 10–21 training sessions, one per day), we found clear layer-specific dopamine responses (Fig. 1G, H, S1A–C, Multimedia Movie 1). In the stratum oriens, positive dopamine transients coincided with CS presentation and reward, while dopamine in other layers decreased at CS. These changes were significant in comparison to control signals from dopamine insensitive GRAB-DAmut mice (Fig. 1I, deep GRAB-DA, 0.61 [0.48,1.26] dF/F, N=7 mice, deep control, −0.05 [−0.12,0.03], N=5 mice, p=0.0025, superficial GRAB-DA, −0.40 [−0.48,−0.24], N=7 mice, superficial control, 0.01 [−0.07,0.11], N=5 mice, p=0.0025; Wilcoxon rank-sum).

We looked for variability in the magnitude of dopamine transients in the x-y dimensions (corresponding to anterior/posterior and medial/lateral within a layer) but did not see clear evidence (Multimedia Movie 2), although imaging at a higher frame rate and a more sensitive reporter may reveal such hot spots. Thus dopamine in the hippocampus is functionally organized in restricted spatial domains defined by cell-layer anatomy, with peri-CS transients in the deep dopamine domain during late Pavlovian conditioning.

For these results, we combined data from GRAB-DA2h and 2m, high and medium affinity sensors, respectively. Waveforms were similar between the two sensors with some differences (S1A–D). The decay kinetics of the 2m sensor after CS were qualitatively faster, as expected from a lower affinity sensor. Importantly, expression of the high affinity sensor 2h did not distort dopamine peak amplitude, relative to the lower affinity 2m. Overall there was no significant difference between the amplitude of CS-triggered dF/F between 2h and 2m (S1E, deep 2h, 1.0 [0.64,1.91] dF/F, N=4, deep 2m, 0.48 [0.16,0.61], N=3 mice, p=0.15, superficial 2h −0.32 [−0.44,−0.21], N=4, superficial 2m, −0.45 [−0.50,−0.27], N=3, p=0.27; Wilcoxon rank-sum).

GRAB-DA sensors are selective for dopamine but can also be activated at higher concentrations by the structurally similar catecholamine norepinephrine (NE), with 10X or 15X higher affinity for dopamine over NE for GRAB-DA2h and 2m, respectively (Sun et al., 2020). In theory, our observed GRAB-DA signal could be triggered by high concentrations of NE release with little corresponding dopamine release in the hippocampus. To examine this possibility, we recorded the NE sensor GRAB-NE1h during Pavlovian conditioning in a separate cohort of mice. We did not observe a deep domain GRAB-NE transient and signals were stable over training (S1F–J). Similarly, there was no development of anticorrelation between deep and superficial GRAB-NE signals over days of Pavlovian conditioning (S1F). A direct comparison of deep domain GRAB-DA and NE signals showed both a larger amplitude GRAB-DA signal and distinct waveform differences (S1J, deep GRAB-DA, 0.61 [0.48,1.26] dF/F, N=7, deep GRAB-NE, −0.13 [−0.30,0.05], N=3 mice, p=0.0167; Wilcoxon rank-sum). These results demonstrate that deep domain GRAB-DA signal does not originate through NE cross talk.

We tracked behavioral changes and dopamine signals across conditioning in all mice. In an example mouse over days of training, licking gradually became more selective for the post CS period (licking in response to the CS appearing as a conditioned response) while locomotion speed in the inter-CS interval increased followed by rapid stopping at CS (Fig. 2A). Dopamine signals were weak in early days of conditioning, then positive transients appeared abruptly in the deep domain. Around the same time, negative going transients appeared in the superficial domain, consistent with anticorrelation of dopamine signals across the two domains.

Figure 2. Deep domain dopamine transients develop over experience.

A, licking, dopamine dF/F dynamics, and locomotion speed over days of training in an example mouse showing appearance of deep transients in Pavlovian conditioning. Negative going superficial dopamine transients appear around the same time as positive deep transients. Mice developed stereotyped behavior of running in between reward presentation and stopping to lick at CS. This behavior arose spontaneously in all mice. B, Time course of task behavior, CS to first lick latency, averaged across all mice, showing learning of CS to UCS association by decreasing lick latency. Traces are aligned by last day of conditioning. C, Time course of CS to locomotion stop latency. Mice stopped spontaneous locomotion more quickly as they learn the CS to UCS association. D, Time course of deep and superficial domain CS-triggered dopamine vs. days of training, showing appearance of transients when mice have low latency stopping and licking after CS. E, Correlation between deep and superficial CS-triggered dopamine signals show development of anticorrelation around the time that dopamine transients emerge.

To look for possible behavioral correlates of the emergence of deep dopamine transients, we plotted the latency between the CS and the first lick as a measure of learning of the CS-UCS association (Fig. 2B, S2A, top). Most mice learned the CS-UCS association within a few days, shown by sharply decreasing lick latency, followed by an intermediate period of low but variable latency, and finally low latency with little variability. We also examined the time course of latency to stop locomotion after CS (Fig. 2C, S2A, bottom). This measure showed a very similar dynamic to CS to lick latency, being initially high, then decreasing to a lower but variable level, before settling at a low level with little variability. This late period reflects automatic and stereotyped licking behavior in response to the CS where mice stop immediately and lick.

We compared these behavioral data to the peri-CS deep and superficial dopamine domain signal over days of conditioning, where deep dopamine transients appeared in GRAB-DA mice but not in controls (Fig. 2D, S2B). Since there was significant individual variability in when deep dopamine transients appeared, we aligned these comparisons to the end of training rather than the start, which would wash out the magnitude of transients in well-trained mice. Behaviorally, deep dopamine signals appeared and grew during the late period of CS to lick and stop latency where behavior was highly stereotyped with low variability. These results illustrate the dynamic behavior of hippocampal dopamine release over Pavlovian conditioning and constrain possible functional roles for this type of signaling. Since deep dopamine does not emerge until late in training, it is unlikely to play a role in reward signaling or reward prediction error (RPE) in initial Pavlovian conditioning. Instead this dopamine signal emerges only after extended training at a time when conditioned response behavior has become automatic. These findings also suggest a link between stop and lick action with dopamine transients, which we will explore in subsequent figures.

We also examined the time course of other behavioral parameters over training (S2C). Total time spent stopped at rewards decreased, total stopped time outside of rewards decreased, licks outside of rewards decreased, and average speed outside of reward increased, all consistent with the development of stereotyped behavior where mice run between rewards without licking, then stop and lick at CS.

In figure F1, we averaged the superficial planes together because in late Pavlovian conditioning, dopamine signal in these planes was correlated. We more closely examined correlation across the superficial planes across conditioning. In early conditioning, signals were small but occurrences of correlation across all superficial planes or lack of unified correlation both appeared (S2D). We tracked the correlations between pairwise superficial planes over conditioning, averaged across mice or individually (S2E). The data were noisy and only a few clear conclusions could be drawn. Correlation across superficial planes was initially variable before becoming strongly correlated towards the end of conditioning, when deep dopamine transients appeared. Before that, the most likely superficial plane to not correlate with others was the SP layer which is anatomically adjacent to the deep SO plane.

It was also notable that marked anticorrelation between deep and superficial dopamine signals was initially absent in early Pavlovian conditioning, then appeared around the time when deep dopamine transients increased in amplitude (F2A, E). The anticorrelation between deep and superficial planes in late Pavlovian conditioning was significant in comparison to controls (S2F, GRAB-DA deep-superficial dopamine correlation, −0.63 [−0.83,−0.44], N=7, control deep-superficial dopamine correlation, 0.17 [−0.07,0.41], N=5, p=0.0025; Wilcoxon rank-sum).

These results demonstrate the dynamic nature of domain-specific hippocampal dopamine signals over weeks of Pavlovian conditioning. The delayed appearance of dopamine transients over days highlights the slow learning or experience-dependent nature of hippocampal dopamine signals, in marked contrast to striatal dopamine transients that appear at rewards prior to learning.

We further examined the characteristics of deep dopamine transients in well-trained mice. We noted an inverse relationship between locomotion and dopamine after rewards where decreases in dopamine seemed time locked to resumption of running (S2G). To illustrate this effect, we plotted the trial-by-trial dopamine signal, licking, and locomotion of one mouse for one imaging session and ordered the trials by duration of stopped time after CS (S2H). This visualization clearly shows that dopamine drops rapidly at the resumption of movement after rewards. To quantify, we made a correlation plot of area under the curve (AUC) of deep dopamine signal after CS versus duration of locomotion stop in individual trials across mice. We found a strong correlation between the size of dopamine transients and duration of stopped periods (S2I, r=0.23, p=8 × 10⁻¹⁴, n=1001 stops from 7 mice; Pearson correlation). We conclude that locomotion strongly correlates with decreased deep domain dopamine release after CS-UCS events. While we focused on locomotion in this quantification, it was also striking that there was a tight link between licking and stopped periods, with licking ending when mice began to run again. We explore the relationship between licking, movement, and dopamine in the next analyses.

Deep dopamine transients at stop and lick events

Given the strong relationship between dopamine release and rewards in the ventral striatum and the peri-reward timing of dopamine transients in the hippocampus, it would be natural to assume that these events are triggered by CS or UCS. However dopamine in the dorsal striatum is strongly tied to actions and there are significant behavioral covariates that occur at CS, specifically stopping and licking. Since we identified an inverse relationship between dopamine and locomotion, we further examined the link between locomotion and dopamine, independent of CS and rewards. When we searched for spontaneous locomotion stops away from CS-UCS events (in well-trained mice), we were surprised to see deep dopamine transients that looked very similar to those at CS (Fig. 3A, B) and were significantly larger than signals in control mice (Fig. 3C, deep GRAB-DA, 0.70 [0.24,1.22] dF/F, N=7 mice, deep control, 0.03[−0.07,0.15] N=5 mice, S3A for superficial, superficial GRAB-DA, −0.44 [−0.50,−0.30] dF/F, N=7 mice, deep control, −0.09[−0.11,−.01] N=5 mice; Wilcoxon rank-sum). These spontaneous stops were often accompanied by licking, even though these events occur in the absence of CS and reward. To determine if licking contributed to dopamine events at spontaneous stops, we selected spontaneous stops that occurred without lick sensor contact (Fig. 3D). Here we found a significant decrease in dopamine at stops without lick contact indicating a combination of stopping and licking was sufficient to trigger dopamine release in the absence of CS and reward (Fig. 3E, Note for parametric tests, we report mean±SEM, mean difference 0.66±0.26 dF/F, N=7, p=0.005; paired t-test; S3B, superficial difference, 0.001±0.16, N=7, p=0.3; paired t-test). At the same time, there still seemed to be hints of a deep dopamine transient at stopping events in the absence of lick sensor contact, where we expected no signal (Fig. 3D, right panel). We investigated the origin of this small, unexplained blip.

Figure 3. Spontaneous stop and lick actions are sufficient for dopamine transients.

A, raw data showing spontaneous dopamine transients at behavioral events marked by licking and locomotion stopping, away from CS and reward. B, average dopamine at spontaneous stopping events, showing dopamine transient in experimental mice but not controls. Lick rate also increased at these stopping events even though no reward was present. C, significantly larger deep dopamine transients at spontaneous stop events in experimental mice vs. controls. D, splitting all spontaneous stopping events in experimental mice into events with lick sensor contact and those without, dopamine appears larger in stops with lick sensor contact, although there may be a small signal in events with no contact. E, dopamine at stops with lick sensor contact are significantly larger than those without. F, tongue tracking with DeepLabCut (DLC) combined with simultaneously recorded lick sensor voltage, locomotion speed, and dopamine signal. Three example events with lick spigot contact (left), tongue appearance without lick spigot contact (center), and spontaneous locomotion stop without tongue appearance (right). Lick-like tongue appearance that did not contact the lick spigot also had dopamine signal while those without tongue appearance did not. Red vertical line marks time of lowest tongue tip position. Red dots on tongue in picture mark DLC tongue markers. Other dots are tracked facial features not used in this analysis. G, stop triggered averages with tongue appearance and no lick sensor contact (left) show clear dopamine transients while those without tongue appearance did not (right). H, stops without tongue had smaller dopamine signals than either stops with lick sensor contact or tongue appearance without contact. Dopamine amplitude was larger for licks that contacted the sensor than those that did not. I, complementary analysis filtering for spontaneous licks rather than locomotion stops. Spontaneous licks with locomotion stops show dopamine transients while licking during movement did not. J, Licking while stopped had larger deep dopamine signals than licking while moving.

By observing mice face videos, we noted that mice sometimes made licking motions at stops without the tongue contacting the lick sensor. One possibility is that there were lick-related movements with associated dopamine elevations that were not detected by the lick circuit because no contact was made. We investigated this possibility using movies of the mouse face and synchronized to two-photon imaging and behavior. To quantify mouth movement, we trained DeepLabCut to track facial features including the chin and tongue. We split spontaneous stopping events without lick sensor contact into those with tongue appearance and without. In this dataset of 813 stops, 17.8% of stops had no tongue and 10.7% had tongue emergence with no lick sensor contact, with the remainder being licks with sensor contact. Indeed there were dopamine elevations that corresponded to licks where the tongue emerged but did not contact the lick spigot (Multimedia movie 3, Fig. 3F). In contrast, stops without tongue appearance had no clear dopamine signal. Also observable in movies of mouse behavior, occasional paw movements towards the lick spigot did not have corresponding dopamine transients, suggesting that these dopamine signals are specific to licking and not general movement or effort.

We quantified dopamine at unrewarded stops with no tongue appearance, tongue but no lick contact, and tongue with lick spigot contact. Average waveform of no tongue stops showed no deep domain transient and dopamine for these events was smaller than stops where the tongue appeared (Fig. 3G, H, deep dopamine for stops without tongue, 0.26±0.03 dF/F, n=145 events, tongue without contact, 0.62±0.11, n=87 events, and tongue with contact, 0.92±0.05, n=581 events from 7 mice, stops with no tongue vs. tongue but no contact, p=0.04, stops with no tongue vs. lick contact, p=1.1×10⁻⁹, tongue no contact vs lick contact, p=0.04, one-way ANOVA, post hoc t-test with bonferroni-holm). Furthermore, events with lick spigot contact had larger dopamine signals than those where the tongue appeared but did not touch the spigot, consistent with this dopamine signal scaling with the vigor of the action (Panigrahi et al., 2015; da Silva et al., 2018). An alternate possibility is that sensory feedback from touching the lick spigot can boost dopamine signal. Regardless, from this analysis, it is clear that stops with tongue movement are sufficient for dopamine transients while sensory inputs, either CS tone or lick spigot feedback, or reward are not required.

Next, rather than using stopping behavior to search for events, we used the complementary approach of detecting lick sensor contact events. First we identified spontaneous licks where mice come to a complete stop, again away from CS-UCS events. Making a lick-triggered average, there was a clear dopamine signal at these events (Fig. 3I, left). In contrast, when mice did not stop completely at licks, there was a no dopamine transient (Fig. 3I, right). The dopamine signal at licks with complete stops was significantly larger than licks without stopping (Fig. 3J, mean difference 0.62±0.28 dF/F, N=7, p=0.027; paired t-test, S3C for superficial domain dopamine, mean difference, −0.13±0.10, N=7, p=0.3; paired t-test). Note that mice did not stop completely at these events, although they did always slow down, probably reflecting an innate inverse relationship between licking and running.

To summarize, spontaneous stop and lick events had significant dopamine transients, indicating that neither CS or reward were necessary for dopamine transients. Furthermore stops with licking motions where the tongue did not even contact the lick spigot had larger dopamine signals than similar stopping events without licking actions. Results from searching for licking events with and without stops came to the same conclusion: only the specific conjunction of stop and licking behavior was accompanied by dopamine transients.

Before moving on to examine the relationship between dopamine and rewards, we first asked if there a general relationship between dopamine and locomotion, outside of stop and lick events and dopamine transients. Previous work found that VTA and SNc neurons, along with dopaminergic axons in the dorsal striatum showed positive activity correlation with locomotion (Wang and Tsien, 2011; Barter et al., 2015; Howe and Dombeck, 2016). First, we tested whether particular phases of locomotion correlated with dopamine signal by chunking behavior into different categories such as stopped, slowing down, starting, speeding up, and relatively constant speed (S3D). To screen categories, we looked for correlations that were significantly different from zero and found the largest correlations occurred during stopping or speeding up periods and were stronger for speed than acceleration (S3F, for statistical results see Supplemental Table 1). Thus we focused on the relationship between dopamine and locomotion at stopping and starting periods in the absence of lick sensor contact. Making a time course of start and stop events showed the superficial domain had higher correlation with speed than the deep domain (S3G, I). To quantify, we analyzed early and late points in conditioning, using DLC to remove events with tongue appearance but no lick sensor contact so all dopamine transients are removed.

Early Pavlovian dopamine signals did not show correlation with locomotion, while superficial domain dopamine during late conditioning was significantly correlated with locomotion at stops and starts, in comparison to control mice, while deep domain dopamine did not (S3H, early superficial start-triggered GRAB-DA correlation to locomotion, 0.38 [0.08,0.70], N=7, vs control correlation 0.06 [−0.00,0.28], N=5, p=0.2; S3H, early deep start-triggered GRAB-DA correlation to locomotion, −0.18 [−0.25,0.04], N=7, vs control correlation, 0.03 [−0.16,0.18], N=5, p=0.34; S3H, late superficial start-triggered GRAB-DA correlation to locomotion, 0.66 [0.55,0.80], N=7, vs control correlation 0.00 [−0.23,0.19], N=5, p=0.005; late deep start-triggered GRAB-DA correlation to locomotion, −0.33 [−0.43,0.01], N=7, vs control correlation 0.12 [−0.11,0.21], N=5, p=0.005; Wilcoxon rank-sum; S3J, early superficial stop-triggered GRAB-DA correlation to locomotion, 0.19 [0.02,0.70], N=7, vs control correlation 0.09 [−0.03,0.17], N=5, p=0.43; early deep stop-triggered GRAB-DA correlation to locomotion, −0.15 [−0.27,−0.05], N=7, vs control correlation −0.05 [−0.11,0.19], N=5, p=0.15; late superficial stop-triggered GRAB-DA correlation to locomotion, 0.67 [0.38,0.75], N=7, vs control correlation 0.03 [−0.09,0.17], N=5, p=0.005; late superficial stop-triggered GRAB-DA correlation to locomotion, −0.24 [−0.58,0.28], N=7, vs control correlation −0.01 [−0.18,0.14], N=5, p=0.43; Wilcoxon rank-sum). Thus removing dopamine transients associated with stop and lick periods, reveals an underlying superficial dopamine correlation with locomotion at stop and start events. This correlation was dependent on experience, appearing in well-trained mice but not early in training.

Having shown that neither CS or rewards were necessary for deep domain dopamine transients in well-trained mice, we tested if the magnitude of dopamine transients was modulated by CS or reward by comparing dopamine at cued reward events (CS and UCS) with spontaneous stop and lick events (no CS or UCS) in the same mice (Fig. 4A, S4A). To make this comparison, we needed to align the dopamine waveforms for the two types of events. Since spontaneous stop and lick events had neither a CS nor UCS to use for alignment, we made a stop-triggered average and then added a temporal offset that brings the stop-triggered waveform in alignment with the CS-triggered waveform. This offset was necessary because stopping occurs after CS in cued reward events. We could have also aligned by stop in both type of events but chose to maintain the consistency of CS-alignment with previous panels.

Figure 4. No detectable reward prediction error in deep dopamine transients.

A, left, schematic of spontaneous stop and lick events and cued reward events (CS-UCS pair). Bottom, timeline of experiment. Events are taken from the late Pavlovian block, after prior training. B, average waveform of deep dopamine signal at cued reward events (left), spontaneous stop and lick events (middle), and a zoomed in and overlaid version (right). Dotted line in the middle panel marks locomotion stopping with an additional temporal offset to align stop-triggered waveforms in stop and lick events with CS-triggered cued reward events. Two windows are used to compare dopamine signals. Window 1 is CS to delivery of UCS (0.5s) in cued reward and the equivalent time window for spontaneous stop and lick. Window 2 is the next 0.5s and corresponds to post UCS time or the equivalent in spontaneous stop and lick events. This end of this window is set by the earlier resumption of locomotion in spontaneous stop and lick events because movement itself decreases dopamine signal (see S2G–I). We did not analyze dopamine outside of these windows because of the confounding locomotion effects on dopamine. C, no difference in window 1 or 2, indicating no effect of CS or reward on dopamine. D, schematic of reward omission (~5% of events), which are cued with CS but no UCS. E, average waveform of deep dopamine signal at cued reward events (left), reward omission events (middle), and a zoomed in and overlaid version (right). Window 2 is 1s, again based on early resumption of locomotion during reward omit events. F, No effect of reward omission on dopamine in either window. Paired t-test, N=6 mice. G, schematic of uncued reward events (~5% of events) which are UCS with no preceding CS. H, average waveform of deep dopamine signal at cued reward (left), uncued reward (middle), and a zoomed in and overlaid version (right). Uncued reward trace is the average of rewarded events, where mice detect and collect reward from spontaneous licks, since these events are uncued. ~40% of uncued rewards are undetected and not shown here. Window 2 is 1s but unlike events with no reward, locomotion is similar in these two conditions because they both have longer stopped time for reward consumption. I, No effect of surprise, uncued reward on dopamine in either window.

We evaluated dopamine magnitude in two windows. The early window (0.5s duration) is equivalent to the time after CS and before reward delivery and represents dopamine triggered by the CS, or the initial part of the dopamine transient in the stop and lick events. Changes in this window will reflect dopamine differences for events with and without CS, although depending on latency, differences could appear in a later temporal window. The second window corresponds to post-UCS reward consumption or the equivalent time window in stop and lick events. We expect changes in this window to reflect dopamine differences for expected UCS vs. no reward. Note that the duration of this second window (0.5s) is timed to avoid dopamine comparisons when locomotion has resumed in stop and lick events, while mice are still stopped in cued reward events. We did not compare dopamine past the resumption of locomotion since locomotion itself decreases dopamine (S2G–I) and will confound possible reward effects. Locomotion resumes earlier in the stop and lick events since there is no reward to consume.

We found no significant difference in dopamine at spontaneous stop and lick events and cued reward events in either the first window corresponding to CS or the second window where reward was delivered and consumed, consistent with little effect of CS or reward on dopamine transients in comparison to spontaneous stopping and licking (F4A-C, spontaneous stop and lick deep domain vs. rewarded, window 1, mean difference −0.25±0.17 dF/F, p=0.21, window 2, mean difference, −0.18±0.33, N=6, p=0.46; paired t-test; S4B, C, spontaneous stop and lick superficial domain vs. rewarded, window 1, mean difference −0.06±0.06 dF/F, p=0.43, window 2, mean difference, 0.02±0.09, N=6, p=0.79; paired t-test).

The timing of dopamine transients relative to stopping and licking are also apparent in these panels. Transients were simultaneous with the initiation of deceleration for a stop, while licking was slightly delayed and was true in both cued reward and spontaneous stop and lick events (Fig. 4B). This is consistent with dopamine being released at the decision to stop and lick, with licking delayed relative to dopamine rise and deceleration. We reiterate that stop and lick events are a concerted action since neither stopping nor licking alone produce dopamine transients (F3G-J).

We also directly tested the effect of locomotion resumption on truncating dopamine signals during spontaneous stop and lick events. We split events into short and long duration stops and compared domain signals in an early and late window. As expected, dopamine signals were significantly smaller in the second window in shorter stopped mice, indicating that the resumption of locomotion decreased dopamine (S4D, E, spontaneous stop and lick superficial domain long stop vs. short stop, window 1, mean difference −0±0.09 dF/F, p=1, window 2, mean difference, −0.70±0.25, N=6, p=0.04, spontaneous stop and lick deep domain long stop vs. short stop, window 1, mean difference 0.23±0.26 dF/F, p=0.43, window 2, mean difference, 1.51±0.56, N=6, p=0.04; paired t-test).

While we cannot eliminate the possibility that reward might increase dopamine at later time points, where we could not analyze because confounding differences in locomotion, two factors argue against this. First, when we split spontaneous stop and lick events into short and long locomotion stops, long locomotion stop events showed longer lasting dopamine transients, directly demonstrating longer stops prolong dopamine signals in the absence of reward. Second, reward omission experiments (CS with no expected UCS) in striatum using GRAB-DA show clear dopamine decreases within 0.5–1s, the same window where we see no change in dopamine (Ishino et al., 2023; Costa et al., 2025; Golden et al., 2025). Our experiments also show that the CS tone itself has a limited role in directly driving dopamine signal. Spontaneous stop and lick events with no CS show similar dopamine levels when compared to CS-cued events in the initial time window. While we cannot completely eliminate the possibility of CS-driven dopamine responses, this experiment sets the upper bound for such a response, which will be low. Since the CS to UCS delay was relatively short in our experiments (0.5s) dopamine evoked by conditioned response actions (stopping and licking) might also contribute to the early dopamine signal. Future experiments with a longer delay will more cleanly resolve possible CS-specific dopamine signals. Thus we conclude that within the temporal windows we analyzed, the occurrence of CS or UCS had little detectable effect on dopamine levels in comparison to spontaneous stop and lick events.

No detectable reward prediction error in deep dopamine transients

To summarize, deep dopamine transients appear over extended Pavlovian conditioning and correspond to conjunctive stop and lick events. This behavior is normally triggered by CS and is used to consume rewards, however, when they occur spontaneously in the absence of CS and UCS, the dopamine signal is similar. These results strongly suggest deep dopamine transients do not reflect a value signal. To further explore possible dependence of dopamine signals on reward, we evaluated reward prediction error (RPE) properties using reward omission experiments (CS and omission of expected UCS) and surprise uncued reward experiments (uncued UCS). Reward omission should produce negative RPE where expected reward is less than expected, with a corresponding decrease in dopamine. We inserted a small percentage of reward omission trials (~5%) and compared dopamine to regular CS-UCS events in the same sessions (Fig. 4D, E). We used analogous time windows to the spontaneous stop and lick analysis with an early window corresponding to CS and a late corresponding to UCS or omission. The late window ended at the initiation of locomotion in reward omission trials (1s after reward omission). This second window was longer in reward omission experiments in comparison to spontaneous stop and lick events (1s vs 0.5s) because mice stopped longer in reward omission trials. Again, there was no difference in the dopamine amplitude in either window with reward omission (Fig. 4F, deep dopamine reward omit vs. rewarded, window 1, mean difference, −0.15±0.06 dF/F, p=0.053, window 2, mean difference, −0.18±0.33, p=0.39, N=6; paired t-test, S4F, G, superficial dopamine reward omit vs. rewarded, window 1, mean difference, 0.02±0.01, p=0.57, window 2, mean difference, 0.06±0.03, p=0.07, N=6; paired t-test). As before, we also found that resumption of locomotion decreased dopamine as dopamine signals were significantly smaller in the second window in shorter stopped mice. (S4H, I, spontaneous stop and lick superficial domain long stop vs. short stop, window 1, mean difference −0.02±0.08 dF/F, p=0.6, window 2, mean difference, 0.62±0.27, N=6, p=0.07, spontaneous stop and lick deep domain long stop vs. short stop, window 1, mean difference 0.0004±0.0031 dF/F, p=0.91, window 2, mean difference, −1.34±0.43, N=6, p=0.03; paired t-test). Thus we conclude that hippocampal dopamine transients do not show negative reward prediction error within 1s after reward omission.

Surprise reward should have positive RPE, where reward is greater than expected with a corresponding elevation in dopamine release. In surprise, uncued (no CS) rewards, we first examined behavioral responses to UCS alone. ~40% of these rewards went undetected, indicating mice could not detect the UCS alone. Mice did not show time locked licking to UCS alone events and uncued reward was only discovered when mice spontaneously licked before the next CS-UCS pair (S4J). When surprise rewards were found, there was no difference in dopamine signal in the early or late window in comparison to CS-UCS events (Fig. 4G–I, deep dopamine uncued UCS vs. CS-UCS, window 1, mean difference, −0.31±0.51, p=0.57, window 2, mean difference, 0.01±0.30, p=0.76, N=5; paired t-test; S4L, superficial dopamine uncued UCS vs. CS-UCS, window 1, mean difference, 0.03±0.22, p=0.9, window 2, mean difference, 0.03±0.16, p=0.84, N=5; paired t-test). Importantly, in this experiment, there was no difference in late locomotion since both conditions had reward consumption. As expected there was no difference in late dopamine when locomotion was matched. For quantification, we kept window 2 at 1s duration to match reward omission experiments, although dopamine past this window was clearly identical (Fig. 4H, right).

We also aligned by the data by first rewarded lick after the UCS. We did this to align directly on the consumption of reward, since variability in the time between stopping and licking could potentially wash out reward responses. Using this alternate alignment, there was still no difference in dopamine between cued reward and surprise, uncued reward (S4M, window 1, mean difference, −0.27±0.33, p=0.46, window 2, mean difference, 0.08±0.23, p=74; paired t-test). We conclude that hippocampal dopamine does not show positive RPE at surprising, unexpected rewards.

We also instituted another type of unexpected reward by inserting a small number of double rewards to see if this increased the size of dopamine transients. We did this by adding a second CS-UCS pair that partially overlapped but was temporally delayed from the first CS-UCS pair such that the second CS overlapped with the first UCS. These events have two CS tones and double the reward. To analyze, we split dopamine signal into the initial CS window corresponding to the time after the first CS and the second window where the second CS is delivered and both rewards. We found no difference between single and double rewards in either window (S4O, P, deep dopamine double CS-UCS vs. single CS-UCS, window 1, mean difference, 0.12±0.06, p=0.12, window 2, mean difference, −0.01±0.08, p=0.88, N=7; paired t-test). As expected there was no difference in dopamine in the CS window between single and double rewards since the first CS tone was identical. However, there was no change in dopamine in the second window, indicating that doubling CS and UCS did not change dopamine release.

Deep dopamine transients at initiation of stop and lick events

Our results show that stop and lick events are sufficient for dopamine transients in the absence of CS and reward. Next we explored the specific association of dopamine transients to behavior. We tested two broad possibilities for how dopamine signals could be generated. The first possibility is that there could be a small dopamine signal for every lick that summates over licks to produce the observed dopamine signal. The second possibility is that there is a dopamine signal at the initiation of stop and lick events and the number of individual licks is less important. To distinguish between these possibilities we made simple linear kernel convolution models. To test the possibility of an individual lick dopamine signal, we generated an event kernel from the mean lick dopamine signal for isolated individual unrewarded licks (no other licks within 0.5s). To model dopamine signals, we convolved this kernel with the binary lick signal, summed the kernel convolution, and quantified how well the modeled traces corresponded to the actual dopamine signal (Fig. 5A, left). For these convolution models, we used a GLM to regress locomotion-correlated signal (speed and acceleration) from the dopamine signal. To test the model fit, we looked at dopamine at rewarded events since these are completely independent of the data used to make the kernel which came from unrewarded lick events. Convolving the lick dopamine kernel with the lick signal at rewards produced modeled deep dopamine traces that showed poor similarity to actual dopamine traces, greatly overestimating actual dopamine. Although the focus for this modeling was on deep domain transients, we also tested model fit to the superficial domain. This was also poorly fit with control r² being higher than experimental (Fig. 5A–D, lick convolution model r² to deep dopamine signal for GRAB-DA mice, 0.10 [0.02,0.70], N=7, vs control mice, 0.74 [0.17,0.94], N=5, p=.94; Wilcoxon rank-sum; Fig. 5C, lick convolution model r² to superficial dopamine signal for GRAB-DA mice, 0.30 [0.07,0.53], N=7, vs control mice, 0.78[0.46,0.88], N=5, p=.02; Wilcoxon rank-sum).

Figure 5. Deep dopamine transients at initiation of stop and lick events.

A, left, schematic of lick model, assessing a per lick dopamine transient. Lick dopamine kernel is made from average of isolated unrewarded licks in same session (no other licks within 0.5s). Dopamine kernel is convolved with the recorded lick signal and waveform summated to model dopamine trace. Right, example dopamine trace and modeled dopamine signal showing overestimate of actual dopamine. B, top, mean peri-reward deep domain traces for model and actual dopamine. Dotted line is first lick. Bottom, no difference in r² of modeled dopamine for GRAB-DA in comparison to same approach applied to control traces (GRAB-DAmut) indicating poor fit of lick model. C, top, mean peri-reward superficial domain traces for model and actual dopamine. Bottom, greater r² of modeled dopamine for controls over GRAB-DA showing the lick model also poorly fits superficial domain signals. D, mean control traces and modeled dopamine for deep (top) and superficial (bottom) domains. E, left, schematic of initiation model, assessing dopamine transient at the initiation of stop and lick events. Initiation dopamine kernel is made from average of isolated unrewarded licks and the initiation of unrewarded lick bouts (not a per lick kernel convolution). Right, example dopamine trace and modeled dopamine signal. F, top, mean peri-reward traces for model and actual dopamine showing good fit of modeled dopamine. Dotted line is first lick. Bottom, larger r² of modeled dopamine for GRAB-DA than control traces indicating significant explanation of dopamine variance at rewards. Note that the events used to make the kernel are from unrewarded, spontaneous events and completely independent from modeled dopamine at rewarded events. G, top, mean peri-reward superficial domain traces for model and actual dopamine. Bottom, no difference in r² of modeled dopamine between GRAB-DA and controls indicating the initiation model is not well fit to superficial domain signals. H, mean control traces and modeled dopamine for deep (top) and superficial (bottom) domains.

To test a model with dopamine tied to the initiation of a stop and lick events rather than individual licks, we generated a dopamine kernel using both isolated unrewarded licks (single lick) and unrewarded lick bouts (multiple licks), aligned to first lick. We then convolved this kernel to rewarded bout initiations, but just once at the start, rather than to each individual lick (Fig. 5E, left). We found that modeled deep domain dopamine at lick bout initiation corresponded well to dopamine at rewarded lick bouts, as shown by higher r² values of this model in comparison to controls. Again we also tested model fits to superficial domain signals and found the r² not significantly different (Fig. 5F–H, initiation convolution model r² to deep dopamine signal for GRAB-DA mice, 0.44 [0.28,0.74], N=7, vs control mice, −0.10 [−0.11,−0.06], N=5, p=.005; Wilcoxon rank-sum; Fig. 5G, initiation convolution model r² to superficial dopamine signal for GRAB-DA mice, 0.24 [−0.00,0.50], N=7, vs control mice, 0.27 [0.02,0.49], N=5, p=.20; Wilcoxon rank-sum). Thus during late Pavlovian conditioning, deep dopamine transients correspond to the initiation of lick bouts rather than individual licks. This compression of multiple actions into an action initiation dopamine signal with extended training has also been observed for striatal dopamine (Collins et al., 2016; Coddington and Dudman, 2018, 2019; van Elzelingen et al., 2022). Finally this interpretation of dopamine as an action initiation signal is consistent with our finding that even “aborted” licks where the tongue appears but does not contact the sensor also have dopamine transients corresponding to the initiation of action.

Dopamine domains in goal-directed learning

To define dopamine dynamics during more complex, goal-directed learning, we developed an instrumental learning paradigm, the VR hidden reward zone task (HRZ). In the HRZ task, mice run on a treadmill to control their position on a virtual track (Fig. 6 A, B). On reaching the end of the track, they experience a random duration dark time, and then begin another trial at the beginning of the track. Mice are trained to lick the water reward spigot as an exploratory behavior. A lick when the mouse is in the unmarked reward zone results in CS and subsequent water reward. After discovering the location of the reward zone, mice refine their lick pattern closer towards the reward zone. At the end of each epoch, there are three unrewarded probe trials, and the reward zone moves to another hidden location. Well trained mice typically complete three such epochs in a 20 min session.

Figure 6. Dopamine transients and ramps in goal-directed task.

A, Hidden Reward Zone (HRZ) behavior. Running moves mice through a visual virtual reality corridor with proximal and distal cues. Licks in the uncued, hidden reward zone trigger CS and water reward. At the end of the track, there is random duration dark time, then a new trial begins. Blue line is mouse's position on virtual track. At the end of each epoch, there are three unrewarded, uncued probe trials, then the reward zone moves to a new location. B, licks, plane fluorescence, and speed expanded from marked period in A. C, CS-triggered average from example experimental and control mouse showing positive dopamine ramp prior to reward in superficial domain and negative ramp in deep domain. D, Dopamine transients at CS and ramping prior to goals are apparent in CS-triggered averages from experimental and control mice. E, positive deep domain and negative superficial domain dopamine signals at CS in comparison to controls, similar to those seen in Pavlovian conditioning. F, deep dopamine transient is apparent in spontaneous stop and lick events but not for licking without stopping, indicating that stopping and licking are sufficient for dopamine transients in HRZ while CS and rewards are not necessary. G, deep dopamine transient is significantly larger in stop and lick events vs. licking while moving. H, schematic of quantification of dopamine ramp slope during reward approach. I, significant negative going deep domain dopamine ramps and positive going superficial domain ramps, in comparison to control slopes. J, no dopamine ramps for equivalent windows of Pavlovian conditioning suggesting that ramps are specific to goal-oriented behavior.

Dopamine dynamics in this task showed some similarities and significant differences in comparison to Pavlovian conditioning (Fig. 6C, D, S5A). First, distinct dopamine signaling in deep and superficial domains was readily apparent. Second, deep and superficial dopamine signals were anticorrelated as seen in late Pavlovian conditioning. Third, again there was a deep domain dopamine transient at reward. Fourth, in marked contrast to Pavlovian conditioning, dopamine levels ramped prior to rewards, ramping down in the deep domain and up in the superficial.

We first asked if deep dopamine transients in HRZ showed similar functional correlates to those in Pavlovian conditioning. Quantitatively there was a positive going dopamine transient in the deep domain and a negative transient relative to baseline in the superficial in rewarded, CS-triggered events (Fig. 6E, S5A, deep GRAB-DA, 0.88 [0.22,2.56], N=8, deep control, −0.03 [−0.39,0.11], N=5 mice, p=0.002, superficial GRAB-DA, −0.54 [−1.00,−0.27], N=8 mice, superficial control, −0.11 [−0.13,−0.09], N=5 mice, p=0.02; Wilcoxon rank-sum). In Pavlovian conditioning, the specific conjunction of licking and stopping in the absence of CS and reward was sufficient to produce dopamine transients, while licking while moving was not. Similarly in HRZ, spontaneous stop and lick events showed a clear dopamine transient in comparison to lick events while moving (Fig. 6F, G, dopamine at licks while stopped vs. licks while moving, mean difference 0.99±0.42 dF/F, N=8, p=0.02; paired t-test). In conclusion, deep dopamine transients during goal-directed HRZ behavior shared critical properties with those seen in Pavlovian, most notably they mark stop and lick episodes and do not require CS or UCS.

The most striking difference in dopamine signaling in HRZ was the appearance of dopamine ramps prior to the reward zone. We quantified the slope of these ramps in a window between start of the track and the reward zone (Fig. 6H, S5B). The ramps were better fit by linear regression in experimental animals over controls (S5C, pre-reward linear fit r² of superficial dopamine in GRAB-DA mice, 0.92 [0.82,0.96], N=8, vs. control mice, 0.44 [0.08,0.61], N=5, p=0.002, pre-reward linear fit r² of deep dopamine in GRAB-DA mice, 0.90 [0.86,0.94], N=8, vs. control mice, 0.20 [0.05,0.20], N=5, p=0.002; Wilcoxon rank-sum). There were significant dopamine ramps in comparison to controls, with the slope increasing in the superficial domain and decreasing in the deep domain (Fig. 6I, pre-reward slope of superficial dopamine domain in GRAB-DA mice, 0.0008 [0.0006,0.0013], N=8, vs. control mice, 0.0002 [5.090 ×10⁻⁵,0.0006], N=5, p=0.002, pre-reward slope of deep dopamine domain in GRAB-DA mice, −0.0021 [−0.0034,−0.0012], N=8, vs. control mice, 0.0003 [5.565×10⁻⁵,0.0007], N=5, p=0.03; Wilcoxon rank-sum).

Previous work found characteristic ramping of dopamine in the striatum prior to reward, when the reward occurs at a predictable time or place, but not in Pavlovian conditioning (Howe et al., 2013; Howe and Dombeck, 2016; Engelhard et al., 2019; Mohebi et al., 2019; Guru et al., 2020; Kim et al., 2020; Hamilos et al., 2021). Consistent with findings in the striatum, there was no ramping in the hippocampal dopamine in the Pavlovian task, where intertrial timing is variable (Fig. 6J, pre-reward slope of superficial dopamine domain in GRAB-DA mice, 5.5×10⁻⁶[−0.0001,8.2 ×10⁻⁵], N=7, vs. control mice, −2.3 ×10⁻⁶ [−6.8 ×10⁻⁵,4.2 ×10⁻⁵], N=5, p=0.56, pre-reward slope of deep dopamine domain in GRAB-DA mice, −4.3 ×10⁻⁵ [−0.0003,3.3 ×10⁻⁵], N=7, vs. control mice, −7.2 ×10⁻⁵ [−0.0003,0], N=5, p=0.85; Wilcoxon rank-sum).

Finally we examined the appearance of the dopamine ramping in the HRZ task to determine if this is an innate or experience-dependent property of hippocampal dopamine signaling. All mice were initially trained in Pavlovian conditioning so the deep dopamine transients at CS were present from the beginning of HRZ training. We found that dopamine ramping developed over training in the HRZ task, with the slope gradually increasing over initial days of training before stabilizing (S6D). We conclude that hippocampal dopamine ramps develop with experience in the HRZ task. This process is somewhat similar to the experience-dependent appearance of deep dopamine transients in Pavlovian conditioning, although time scales and dynamics are very different, with dopamine ramps appearing rapidly after starting the HRZ task.

To summarize, the same distinction between superficial and deep dopamine domains seen in Pavlovian conditioning was apparent during a goal-directed spatial task. Similar to late Pavlovian conditioning, there was a prominent dopamine transient at rewards in the deep domain. This dopamine transient did not require CS or reward but were triggered by stop and lick events. Notably in the HRZ task, ramping dopamine signals appeared during reward approach, similar to those seen in the striatum.

Discussion

Our results reveal previously unsuspected complexity in spatio-temporal hippocampal dopamine release, identifying anatomically defined, functionally distinct domains. During Pavlovian conditioning, deep domain transients appeared after extensive training. These transients were most tightly correlated with the initiation of stop and lick behaviors, the actions used to collect water rewards. This signal was not strongly dependent on CS or rewards; accordingly, hippocampal dopamine showed limited evidence of reward prediction error. The same two domains showed distinct dopamine dynamics in a spatial, goal-directed task. Deep domain transients at stop and lick events were also present while dopamine ramping now appeared during reward approach, with positive ramping in the superficial domain and negative ramping in the deep.

We identify a restricted spatial scale for the deep dopamine domain, on the order of ~100μm (corresponding to the thickness of the basal dendritic layer), with functionally and spatially distinct domains in adjacent layers. Importantly domains at such spatial scales cannot be observed using fiber photometry or miniscopes. Since these domains are axially segregated, single photon approaches with no optical sectioning ability will not resolve them. Our findings raise the possibility of similar subcellular dopamine domains in other brain regions, when measured at sufficient resolution during appropriate behaviors.

VTA neuron activity is strongly associated with dopaminergic RPE in the ventral striatum. These neurons also project throughout the brain and are hypothesized to broadcast RPE signals to modulate reward-based reinforcement learning in local circuits. We found little evidence for dopaminergic RPE signaling in the hippocampus during late Pavlovian conditioning. Stop and lick events themselves, in the absence of CS or reward, were sufficient to generate dopamine transients. We tested negative reward prediction error with omission of expected UCS and initial dopamine was unchanged, although we could not assess late changes in dopamine due to differences in locomotion affecting dopamine levels. However, other recent publications show detection of negative RPE within 1s of reward omission, suggesting if there were such changes, we could detect them here (Ishino et al., 2023; Costa et al., 2025; Golden et al., 2025). Similarly, we saw no positive reward prediction error, tested with surprise, uncued rewards. While there could be RPE signaling below our level of detection, our finding of limited RPE properties in the hippocampus is consistent with the conclusions of another study finding no dopaminergic RPE signal in numerous brain areas (not including the hippocampus) with the exception of the ventral prelimbic cortex (Mohebi et al., 2019). These conclusions only apply to late Pavlovian conditioning, where there were reliable deep dopamine transients; no transients were detectable in early conditioning. This experience-dependent development indicates that hippocampal dopamine is unlikely to contribute to early learning in Pavlovian conditioning, consistent with dopamine release in the striatum driving initial association of CS and reward.

What are the possible functional roles of hippocampal dopamine transients and ramps? Since ramping and transients are dissociable, we believe they serve different purposes. We propose that hippocampal dopamine transients at stop and lick events act equivalently to a reward signal, which will reinforce temporally adjacent reward-related information. In the HRZ task, this translates to dopamine facilitating the learning and recall of reward zone location. That the proximal trigger for hippocampal dopamine transients is the action of reward collection rather than CS or reward is quite surprising. We speculate that this distinction reflects specialized computations in different brain areas operating on distinct timescales. The striatum slowly learns associations between stimuli and responses to maximize rewards. Over experience, midbrain dopamine sources develop a learned, abstracted version of reward, tied to reward collection. This action signal is broadcast to the hippocampus and is used as a cue for fast learning and remembering of reward-associated events. Recently, this concept that action-related dopamine signals can directly drive associative learning, rather than a VTA value signal, has received computational (Daw et al., 2005; Miller et al., 2019; Bogacz, 2020) and experimental support (Greenstreet et al., 2025). To clarify, we do not propose that hippocampal dopamine transients directly drive reward collection actions but instead they play a retrospective role reinforcing reward-related information.

In contrast to dopamine transients that are tied to reward collection actions, dopamine ramps precede the reward zone, suggesting they may play a more instructive role in guiding successful operant behavior. In the case of the HRZ task, gradually increasing superficial domain dopamine levels may increase the probability to lick, peaking near the reward zone. Thus dopamine ramps in the hippocampus might help gate spatially specific actions that are necessary for rewards. Supporting evidence for ramps playing a role in the timing of actions comes from a study showing that optogenetically generating steeper dopamine ramps biased mice to earlier action initiation while decreasing the slope delayed action initiation (Hamilos et al., 2021). While significant future experiments will be necessary to test these proposals, the critical contribution of this work is to reveal the complexity of spatiotemporal hippocampal dopamine release that enable data-driven hypotheses. Being able to observe these dynamics gives us the power to specifically manipulate dopamine to test the roles of spatially and temporally specific dopamine release.

The finding of distinct dopamine signals in close apposition was striking and naturally raises the question of their anatomical and functional origins. Both dopaminergic midbrain (VTA and SNc), and LC dopaminergic neurons project to the hippocampus. Studies of layer-specific targeting by VTA dopaminergic axons are contradictory with one finding primarily SO targeting (Adeniyi et al., 2020), and another finding preferential SP targeting (Tsetsenis et al., 2021), although both found axons in all layers. There are fewer studies examining SNc projections to the hippocampus (Gasbarri et al., 1994b) and little in the way of layer-specific anatomy. Based on functional signals associated with midbrain dopamine sources, we propose that SNc primarily innervates the deep domain and VTA the superficial. Functionally SNc signals are tied to action initiation, similar to the deep dopamine transients in the hippocampus (Jin and Costa, 2010; Collins et al., 2016; da Silva et al., 2018). Furthermore a class of macaque dopaminergic SNc neurons that project to the tail of the striatum show similar functional properties to hippocampal deep dopamine transients (Kim et al., 2015). These neurons very slowly develop activity to visual stimuli associated with rewards and become insensitive to reward omission, indicating this is not a rapid feedback signal like RPE. Future experiments will determine if this type of SNc neuron preferentially innervates the hippocampal deep dopamine domain.

VTA neurons often show positive ramping during reward approach, similar to the positive ramping we see in the superficial domain (Howe et al., 2013; Howe and Dombeck, 2016; Engelhard et al., 2019; Mohebi et al., 2019; Kim et al., 2020). Furthermore positive dopamine ramping was observed in calcium activity of VTA dopaminergic axons recorded in the hippocampus (Krishnan et al., 2022; Heer and Sheffield, 2024), although the layer-specific distribution of these axons has not been described. Surprisingly, LC inputs can also release dopamine in the hippocampus, in addition to norepinephrine. LC axons also seem to target all layers (Takeuchi et al., 2016), although with preference for the superficial domain (Kempadoo et al., 2016).

Our findings show that postsynaptic neurons can see differing levels of dopamine with subcellular resolution, rather than a uniform global signal. These restricted domains imply correspondingly fine scale postsynaptic mechanisms to process distinct dopaminergic signals with subcellular or microcircuit resolution. This view aligns with recent work in the striatum identifying spatially restricted dopamine release that can produce dendrite-level functional effects (Banerjee et al., 2022; Yee et al., 2025). D1 and D2 dopamine receptors are expressed exclusively on hippocampal interneurons (Gangarossa et al., 2012; Puighermanal et al., 2015), marking inhibition as a major target of dopaminergic action (Godino et al., 2025). A clear target for spatially restricted dopamine domains is the preferential activation of specific classes of interneurons. OLM cells, a subset of somatostatin-expressing neurons that target their inhibition to the distal apical tuft of pyramidal neurons express the inhibitory D2 receptor and have dendritic arbors completely within the deep domain. Dopamine transients could act to disinhibit the apical tuft of CA1 pyramidal neurons, facilitating dendritic plasticity mechanisms to encode reward-related information (Bittner et al., 2017). Conversely subsets of CCK- and CB-expressing interneurons have dendrites within the superficial domain (Gulyás and Freund, 1996; Mátyás et al., 2004). Indeed hippocampal dopamine signaling mediated through D1/5 receptors was necessary for sustained PV neuron plasticity and memory consolidation (Karunakaran et al., 2016). Furthermore a subset of CA1 pyramidal neurons express D5 receptor (Sariñana et al., 2014) where dopamine could facilitate or stabilize long-term plasticity (Frey and Morris, 1997; Li et al., 2003). While the domain specific influences on hippocampal circuitry remain to be elucidated, our discoveries point to multiple dopamine domains and anatomical origins that likely play distinct and dissociable roles in hippocampal-dependent learning and memory.

Supplementary Material

Multimedia Movie 2

Multimedia movie 2.

Dopamine fluorescence across layers at CS.

Pseudo-colored movie of dopamine fluorescence intensity in peri-CS window (−5 to +5s) with red showing increased intensity and blue decreased.

Multimedia Movie 3

Multimedia movie 3.

Dopamine transients at licks with no lick sensor contact.

Movies showing mouse faces with simultaneous deep domain dopamine signal, tongue position (tracked by DLC), lick sensor electrical contact, and mouse speed. Lick that does not touch the lick spigot has a similar size dopamine transient as one that does, indicating deep dopamine transients are an action initiation signal.

Multimedia Movie 1

Multimedia movie 1.

Dopamine fluorescence across layers at CS.

Fluorescence movie showing intensity changes at CS. Plane 4 = stratum oriens or the deep layer, Plane 3 = stratum pyramidale, Plane 2 = stratum radiatum, and Plane 1 = stratum lacunosum moleculare

Significance statement.

The hippocampus is necessary for forming long-term memories and the neuromodulator dopamine is critical in this process. Exactly how dopamine acts in the hippocampus remains opaque since there is no clear picture of when and where dopamine is released. Using newly developed optical dopamine sensors, we find that hippocampal dopamine release is split into two distinct spatial domains. In Pavlovian conditioning, the “deep” domain shows dopamine transients that largely correspond to actions used to consume rewards, while in a goal directed task, dopamine ramping appears in both domains during reward approach. Notably, distinct domains were initially absent and developed over experience. Our results reveal specific dopamine dynamics that are likely to play distinct and dissociable roles during behavior.