Dynamic changes in chloride homeostasis coordinate midbrain inhibitory network activity during reward learning

Woo, Joyce; Uprety, Ajay; Reid, Daniel J.; Chang, Irene; Ketema Samuel, Aelon; de Carvalho Schuch, Helena; Swain, Caroline C.; Ostroumov, Alexey

doi:10.1038/s41467-025-66838-x

Download PDF

Article
Open access
Published: 09 December 2025

Dynamic changes in chloride homeostasis coordinate midbrain inhibitory network activity during reward learning

Nature Communications volume 16, Article number: 10903 (2025) Cite this article

7545 Accesses
1 Citations
101 Altmetric
Metrics details

Subjects

Abstract

The ability to associate environmental stimuli with positive outcomes is a fundamental form of learning. While extensive research has focused on midbrain dopamine neurons during associative learning, less is known about learning-mediated changes in the afferents that shape dopamine neuron responses. We demonstrate in rats that during critical phases of learning, anion homeostasis in midbrain inhibitory GABA neurons – a primary source of input to dopamine neurons – is disrupted due to downregulation of the potassium chloride cotransporter KCC2. This alteration in GABA neurons preferentially impacted lateral mesoaccumbal dopamine pathways and was not observed after learning was established. At the network level, learning-mediated KCC2 downregulation was associated with enhanced synchronization between individual GABA neurons and increased dopamine responses to rewards and reward-related stimuli. Conversely, enhancing KCC2 function during learning reduced GABA synchronization, diminished relevant dopamine signaling, and prevented cue-reward associations. Thus, circuit-specific adaptations in midbrain GABA neurons are crucial for forming new reward-related behaviors.

Silencing KCC2 in mouse dorsal hippocampus compromises spatial and contextual memory

Article 27 October 2022

Acetylcholine demixes heterogeneous dopamine signals for learning and moving

Article Open access 25 March 2026

Ventral tegmental area GABA neurons integrate positive and negative valence

Article Open access 24 November 2025

Introduction

An organism’s survival and adaptation depend on its ability to learn associations between environmental stimuli and rewards. Associative reward learning links events to outcomes, and is a fundamental process essential to adaptive behaviors; its disruption is a core feature of many debilitating neuropsychiatric disorders, such as major depressive disorder¹, addiction², and schizophrenia³.

Seminal work has shown that the rate of associative reward learning relies on phasic dopamine (DA) signaling within mesolimbic pathways projecting from the ventral tegmental area (VTA) targeting the nucleus accumbens (NAc)^4,5. These learning-related DA responses exhibit remarkably focused circuit-specificity, in which reward and cue-related phasic activation dominate in lateral VTA DA neurons that project to the lateral subregions of the NAc^4,6,7. During initial phases of learning, these neurons exhibit transient firing primarily in response to reward presentations. However, following repeated cue-reward pairings, DA neurons begin to exhibit additional bursts of activity at cue onset, indicating the strength of association between rewards and environmental stimuli^5,8. Despite extensive research involving DA and reward encoding in the brain, the mechanisms shaping these representations are still unclear.

GABA-releasing neurons in the VTA are key regulators of VTA DA neurons^9,10. These VTA GABA neurons increase firing during cues that predict rewards^10,11, and exhibit drastic alterations in transmission after exposure to drugs of abuse^12,13,14. However, there has been surprisingly little focus on how enhanced GABAergic signaling contributes to the acquisition of reward-associated behaviors.

Recent studies have shown that highly salient experiences, such as stress, addictive drugs, and chronic pain, induce a shift toward excitatory GABA_A receptor signaling in VTA GABA neurons^14,15,16. This shift in GABA signaling stems from a downregulation in the neuron-specific Cl^- extrusion pump, potassium-chloride transporter 2 (KCC2), which primarily functions to maintain low intracellular Cl^- levels. While midbrain DA neurons exhibit little to no expression of KCC2¹⁷, VTA GABA neurons are significantly affected by KCC2 impairment. The resulting intracellular Cl^- accumulation leads to depolarizing shifts in the Cl^- reversal potential (E_GABA), and compromises GABA_A receptor-mediated inhibition¹⁵. Ultimately, functional KCC2 downregulation in VTA GABA neurons following stress and drug exposure enhances excitatory GABA_A receptor signaling and amplifies the acquisition of drug-taking behaviors^13,14,15. Changes in chloride homeostasis within the VTA circuitry may reflect a pathological usurpation of mechanisms that normally support reward-related learning, a framework that has previously been proposed for addiction². Despite its previous implication during pathological states, intracellular chloride regulation during naturalistic learning contexts has not yet been examined.

To address this gap, we show that dynamic changes in intracellular anion homeostasis in GABAergic signaling emerge during reward learning and thereby shape the activity of downstream VTA DA targets. Utilizing behavioral, electrophysiological, and molecular approaches, we demonstrate that VTA GABA neurons exhibit altered Cl^- homeostasis through functional KCC2 downregulation within fine temporal windows during reward learning. Importantly, we show that enhancing KCC2 function in the VTA halts the progression of cue-reward learning. Moreover, our findings indicate that functional KCC2 downregulation is pathway-specific and plays a crucial role in synchronizing the activity of midbrain GABAergic networks. Finally, we show that synchronized GABAergic activity can amplify phasic firing in DA neurons. These findings indicate that circuit-specific alterations in KCC2 are a fundamental mechanism that sculpts experience-induced circuit remodeling and reward learning.

Results

Learning-dependent downregulation of KCC2 in VTA GABA neurons

We first investigated whether reward learning altered anion homeostasis in VTA GABA neurons. Given previous findings indicating that depolarizing shifts in E_GABA are calcium (Ca²⁺)-dependent¹⁸, we focused selectively on GABAergic populations that were activated during cue-reward pairings. To achieve this, male and female GAD-Cre rats received intra-VTA injections of Cre-dependent blue fluorescent protein (BFP) and a fluorescent Ca²⁺ sensor CaMPARI2 that converts from green to red fluorescence in the presence of high calcium concentrations and 405 nm (UV) light¹⁹ (Fig. 1A). In freely moving rats undergoing Pavlovian conditioning (see Methods), the UV illumination was delivered through an implanted optic fiber during the 5-s auditory tone (conditioned stimulus, CS) and subsequent sugar pellet presentations (unconditioned stimulus, US, Fig. 1B, Paired Conditioning). In a separate unpaired conditioning group, the UV light was also delivered throughout the CS and US, but these two stimuli were never presented in immediate succession (Fig. 1B, Unpaired Conditioning). Compared to the unpaired group, reward-seeking behaviors during CS presentations significantly increased after CS-US pairing over the 13-day period (Fig. 1C), indicating the formation of cue-reward association. Sex differences were not detected in the paired group (Supplementary Fig. 1A).

Fig. 1: Learning-dependent downregulation of KCC2 shifts EGABA in VTA GABA neurons. — **Fig. 1: Learning-dependent downregulation of KCC2 shifts E_GABA in VTA GABA neurons.**

Upon examination of the paired conditioning time course, we delineated three distinct phases of learning: baseline – referring to the first day of learning, acquisition – characterized by a rapid ascent in rate difference (change in number of port entries during CS relative to 5-s preceding CS), and plateau – a period of stable responding with <10% variation in performance over 3 consecutive days (Fig. 1C). To assess learning-induced changes in anion homeostasis in VTA GABA neurons, we measured the reversal potential for GABA_A receptor mediated currents (E_GABA) in cells expressing BFP and photoconverted red CaMPARI (Fig. 1D). After days 1, 5–7, and 13 of Pavlovian conditioning, we performed gramicidin-perforated patch-clamp recordings to preserve the intracellular anion concentrations (Fig. 1E and Supplementary Fig 1B). E_GABA was determined as the membrane potential at which evoked inhibitory postsynaptic currents (eIPSCs) change their direction from inward to outward (Fig. 1F). During acquisition, VTA GABA neurons from both male and female rats showed significantly more depolarized E_GABA values compared to baseline and plateau phases (Fig. 1G and Supplementary Fig. 1E). Notably, this depolarizing shift occurred without changes in the resting membrane potential (Supplementary Fig. 1F). While total CaMPARI2 expression did not differ across learning stages, the percentage of photoconverted (red) GABA neurons, indicating elevated calcium levels, was significantly higher during acquisition (days 5–7) compared to both baseline and plateau phases (Supplementary Fig. 1C, D). Importantly, VTA GABA neurons that were not photoconverted to red CaMPARI during acquisition maintained hyperpolarized E_GABA (Supplementary Fig 1G). To determine whether similar learning-induced changes occurred in VTA DA neurons, we performed analogous experiment in TH-Cre rats (Supplementary Fig. 1H). In contrast to GABA neurons, VTA DA neurons showed no changes in E_GABA during acquisition compared to baseline (Supplementary Fig. 1I).

In the unpaired conditioning group, total CaMPARI2 expression was comparable to the paired group, but significantly fewer red CaMPARI-expressing GABA neurons were observed during acquisition (Supplementary Fig. 1J, K). Moreover, E_GABA in these photoconverted GABA neurons remained hyperpolarized across all timepoints, indicating that salient stimuli alone are insufficient to induce detectable changes in chloride homeostasis (Fig. 1H).

A depolarizing shift in E_GABA reflects reduced Cl^- extrusion capacity, which has been previously associated with dephosphorylation of KCC2 at serine 940 (S940)^15,18. To examine whether KCC2 dephosphorylation was altered during reward learning, we performed western blot analysis using antibodies against total KCC2 and phospho-S940 KCC2 at distinct phases of Pavlovian conditioning. Immunoblots revealed prominent bands at 140 and 270 kDa for both total and phospho-S940 KCC2, indicating the presence of monomeric and dimeric forms of KCC2 (Fig. 1I and Supplementary Fig. 1O). As anticipated, the ratio of phosphorylated S940 KCC2 to total KCC2 during acquisition of learning (days 5–7) was significantly lower compared to baseline (day 1, Fig. 1J). In contrast, no significant differences in the ratio of phosphorylated S940 KCC2 to total KCC2 were observed between baseline and plateau groups (day 13, Fig. 1K, L and Supplementary Fig. 1O). Additionally, phospho-S940 levels did not differ by sex (Supplementary Fig. 1L), and total KCC2 expression remained unchanged across learning phases (Supplementary Fig. 1M). Together with previous immunolabeling studies showing that KCC2 protein is selectively expressed in non-DA, GABAergic neurons within the VTA^14,17, our results suggest that dephosphorylation of KCC2 protein at S940 decreases KCC2 function in VTA GABA neurons only during the acquisition of cue-reward association.

Although the chloride importer NKCC1 can also influence E_GABA²⁰, western blot analysis revealed no significant difference in NKCC1 expression between the acquisition and baseline phases of learning (Supplementary Fig. 1N, O).

Learning-mediated KCC2 downregulation impacts NAc lateral shell projecting DA neurons

Next, we examined whether depolarizing shifts in E_GABA within VTA GABA neurons altered GABA release on distinct DA neuron projections to different subregions of the NAc. To label projection-specific DA neurons, we injected a retrogradely transported Cre-dependent AAV vector containing GFP in the NAc lateral shell, medial shell, and core of male and female TH-Cre rats (Fig. 2A)^4,21,22. Then, we assessed learning-related changes in GABA release, by measuring the frequency of spontaneous inhibitory postsynaptic currents (sIPSCs) in GFP-expressing VTA DA neurons immediately after days 1, 5–7, and 13 of Pavlovian conditioning (Fig. 2B).

Lateral shell-projecting DA neurons were located predominantly in the lateral portions of the VTA (Fig. 2C). While sIPSC frequency did not change across learning (Supplementary Fig. 2A), we observed an increase in spontaneous GABA release events occurring in close temporal proximity to each other during acquisition (Fig. 2D, top black and red traces with insets, see Methods). Quantitative analysis revealed that after days 5–7 of Pavlovian conditioning, lateral shell-projecting DA neurons exhibited a significantly higher percentage of sIPSCs with interevent intervals <10 ms, compared to days 1 and 13 (Fig. 2E, left). Notably, the emergence of significant changes in sIPSC event proximity during the acquisition phase coincided with a significant increase in conditioned responding relative to day 1 (Supplementary Fig. 2B). To examine the role of KCC2 downregulation in this effect, we recorded sIPSCs in slices incubated with pharmacological KCC2 activator CLP290 (~1 h, 10 µM)^13,14. CLP290 treatment after days 5–7 of learning decreased the percentage of sIPSCs with interevent intervals of <10 ms to the values observed at days 1 and 13 (Fig. 2E, middle). Furthermore, lateral shell-projecting DA neurons from naïve control rats treated with KCC2 inhibitor VU0463271 (10 µM) showed a potentiation of closely occurring sIPSCs that was similar to days 5–7 of learning (Fig. 2E, right). Importantly, neither CLP290 nor VU0463271 altered the total number of sIPSCs per second (Supplementary Fig. 2C, D). An increased number of sIPSCs occurring in close temporal proximity, without changes in the total number of events, suggests enhanced synchronized GABA release on a single DA neuron^23,24.

To confirm the impact of KCC2 modulation on GABA release onto DA neurons, we next used a pharmacological approach. Prior studies showed that KCC2 downregulation leads to excitation of VTA GABA neurons during intense GABA_A receptor stimulation¹⁵. Compared to DA neurons, GABA_A receptors in VTA GABA neurons are much more sensitive to benzodiazepines due to differences in subunit composition²⁵. Thus, benzodiazepines, such as diazepam, potentiate GABA_A receptor function primarily in VTA GABA, but not in DA neurons. Indeed, diazepam was previously shown to suppress VTA GABA neuron activity and to attenuate GABA release onto DA neurons¹³. Upon KCC2 downregulation, however, diazepam-mediated potentiation of GABA_A receptor signaling excited VTA GABA neurons and enhanced GABA release onto DA neurons¹³. Consistent with this model, bath-application of diazepam (5 µM) during the acquisition phase of learning significantly increased sIPSC frequency in DA neurons compared to baseline and plateau (Fig. 2D, F, left). Incubation with CLP290 prevented diazepam-mediated increase in sIPSC frequency (Fig. 2F, right). An increase in the frequency of sIPSCs following diazepam suggests enhanced presynaptic GABA release onto DA neurons. Changes in sIPSC amplitude, in contrast, would suggest adaptations in DA neurons, yet diazepam did not alter sIPSC amplitude across all phases of learning (Supplementary Fig. 2E, F).

In contrast to lateral shell projections, DA neurons projecting to medial shell and core (Fig. 2G, J) showed no significant learning mediated changes in the percentage of sIPSCs with interevent intervals <10 ms (Fig. 2H, K) or total number of spontaneous events (Supplementary Fig. 2G, I). Further, bath application of diazepam led to a consistent decrease in sIPSC frequency (Fig. 2I, L) without changes in sIPSC amplitude across all stages of learning (Supplementary Fig. 2H, J). Overall, these findings highlight a striking circuit specificity of learning-induced KCC2 downregulation, with its impact on VTA GABA neurons selectively innervating DA projections to the NAc lateral shell, a DA pathway that was previously shown to develop phasic cue-evoked responses over the course of learning^6,7.

VTA GABA neurons and KCC2 downregulation contribute to reward learning

Given that changes in VTA GABA neurons developed during the acquisition phase of Pavlovian conditioning, we determined whether these neurons mediate cue-reward association. First, we used optogenetics to suppress GABA neuron activity during cue and reward presentations throughout a 13-day Pavlovian training period. Because previous findings have shown that unilateral manipulation of VTA GABA neurons is sufficient to alter reward-related behaviors^26,27, GAD-Cre rats were unilaterally injected in the lateral VTA with Cre-dependent archaerhodopsin (Arch) and implanted with optic fibers (Fig. 3A and Supplementary Fig. 3A). When compared to control rats (GFP-expressing with light stimulation and Arch-expressing with no light stimulation, Supplementary Fig. 3B), photoinhibition of VTA GABA neurons in Arch-expressing rats significantly attenuated conditioned responding to the reward-predictive CS (Fig. 3B). No sex differences were observed in conditioned responding (Supplementary Fig. 3C). Decreased conditioned responding suggests that photoinhibition of VTA GABA neurons delays acquisition of cue-reward association. To test this possibility, we calculated the number of conditioning sessions until the animals reached the middle of acquisition phase, defined as half of the rate difference values observed during the plateau phase (dashed line in Fig. 3B). Compared to GFP controls, Arch-expressing rats had a significantly greater number of training sessions below the middle of acquisition phase (Fig. 3C), indicating that suppression of VTA GABA neuron activity delayed reward-related learning.

Next, we examined whether VTA GABA neuron suppression during acquisition of Pavlovian conditioning arrested the development of cue-reward association. VTA GABA neurons were photoinhibited during CS-US starting the day after animals reached the middle of acquisition phase of learning (Fig. 3D, shaded green bars, and Supplementary Fig. 3D). In marked contrast to GFP controls, which exhibited an increased rate difference, Arch-expressing rats failed to show an enhancement in conditioned responding (Fig. 3E).

We have previously shown that bilateral pharmacological KCC2 activation in the VTA attenuates drug-related behaviors^14,15. To examine whether KCC2 activation also affects the formation of cue-reward associations, we bilaterally administered CLP290 (1 µL, 60 µM) or vehicle into the lateral VTA (Supplementary Fig. 3E) prior to conditioning sessions on alternate days (Fig. 3F, G, shaded blue bars). Compared to rats that received intra-VTA infusion of vehicle (Fig. 3G, black data), intra-VTA infusion of CLP290 significantly decreased conditioned responding to the reward-predictive CS (Fig. 3G, red data). The number of conditioning sessions until the animals reached the middle of acquisition phase was significantly higher in CLP290 compared to vehicle-treated controls (Fig. 3H). Furthermore, when CLP290 was administered beginning the day after the middle of acquisition (Fig. 3I and Supplementary Fig. 3F), rats failed to exhibit increase in rate difference (Fig. 3J, red data), in contrast to vehicle-treated controls, which showed enhanced conditioned responding (Fig. 3J, black data). To further confirm that KCC2 activation specifically affects the acquisition phase of conditioned responding, we administered CLP290 during the plateau stage of learning (Supplementary Fig. 3G). CLP290 injections on days 14 and 16 of learning did not produce significant differences in behavioral performance compared to vehicle-treated controls (Supplementary Fig. 3H), suggesting that KCC2 activation selectively influences the acquisition, but not the expression, of reward learning.

Finally, we attenuated KCC2 expression selectively in VTA GABA neurons prior to learning by bilaterally microinfusing a viral vector carrying Cre-dependent KCC2 targeting shRNA into the lateral VTA of GAD-Cre rats (Fig. 3K)^28,29. This approach reliably downregulated KCC2 expression and function in VTA GABA neurons compared to control animals that received a scrambled (non-targeting) shRNA construct (Fig. 3L and Supplementary Fig. 3I, J). Notably, KCC2 knockdown in VTA GABA neurons significantly accelerated the acquisition of Pavlovian reward conditioning relative to scrambled shRNA controls (Fig. 3M). Taken together, these results indicate that VTA KCC2 downregulation is critical for Pavlovian reward-related learning.

VTA GABA neurons show enhanced synchronized activity during acquisition periods of learning

We next investigated the mechanisms by which KCC2 downregulation in the VTA contributed to learning. Patch clamp recordings indicated that KCC2 downregulation during acquisition increased the number of concurrent GABA release events with synchrony of less than 10 milliseconds (Fig. 2D, E). This phenomenon could arise from increased temporal coincidence of action potential firing between individual GABA neurons in the VTA. To determine whether VTA GABA neurons exhibit millisecond timescale synchrony during Pavlovian conditioning, we performed tetrode and silicon probe electrophysiological recordings in freely behaving rats (Fig. 4A). For cell-type specific recordings, we expressed Cre-dependent Arch and Channelrhodopsin (ChR2) in GAD-Cre and TH-Cre rats, respectively (Supplementary Fig. 4A). Animals were chronically implanted with microdrives carrying recording electrodes and optic fiber cannulas allowing for the classification of lateral VTA cells as putative GABA and DA neurons based on opto-tagging and spike waveform features (Supplementary Fig. 4B–G, see Methods for details).

**Fig. 4: VTA GABA neurons show enhanced coordinated activity during learning acquisition.**

In agreement with previous work, the proportion of lateral VTA DA and GABA neurons responding to CS increased with the progression of associative learning^10,11 (Supplementary Fig. 5A–D). At all stages of learning, we observed synchronous spiking between individual VTA GABA neurons (Fig. 4B). To quantify changes in synchrony across learning phases, we calculated cross-correlograms (CCGs) between pairs of simultaneously recorded putative lateral VTA GABA neurons³⁰ (Fig. 4C, see Methods). Our results reveal that during acquisition, a significantly higher percentage of putative GABA neuron pairs exhibited millisecond timescale synchrony compared to baseline and plateau phases of learning (Fig. 4D). During the CS, synchrony occurred on average between 100 to 200 ms after cue-evoked peaks in VTA GABA firing. Notably, significant increases in synchrony were also observed outside the CS and US periods, extending into the intertrial intervals (Fig. 4D, Pre-CS).

We then postulated that pharmacological manipulation of KCC2 in the VTA would alter synchronized activity of VTA GABA neurons in vivo. Rats were chronically implanted with a microdrive system designed for unilateral drug infusions, opto-tagging, and simultaneous electrophysiological recordings from the lateral VTA (Fig. 4E). First, ipsilateral to the microdrive implantation (Supplementary Fig. 5E), we microinfused KCC2 activator CLP290 (1 µL, 60 µM) or vehicle in the VTA of learning animals once they reached the middle of the acquisition period (defined based on animal’s performance in Supplementary Fig. 5A). Across all parts of a trial, lateral VTA GABA neurons exhibited a significant decrease in coordinated firing activity after intra-VTA CLP290 injections compared to vehicle treatment (Fig. 4F). Importantly, intra-VTA CLP290 injections did not impact locomotion (Supplementary Fig. 6A–C). Next, we microinfused KCC2 antagonist VU0463271 (1 µL, 100 µM) in the lateral VTA of naïve, freely moving animals in their home cage (Fig. 4G and Supplementary Fig. 5E). When compared to the pre-injection period, VU0463271 significantly increased the mean percentage of synchronized VTA GABA neuron pairs (Fig. 4H, I). Cumulatively, these findings indicate that during acquisition phases of reward learning, individual VTA GABA neurons synchronize their activity within a millisecond time frame, a phenomenon bidirectionally dependent on KCC2 function.

Changes in KCC2 function enhance DA signaling in the VTA and NAc

Since associative reward learning relies on phasic DA neuron firing in the lateral VTA^4,5, we hypothesized that KCC2 downregulation in VTA GABA neurons altered DA neuron bursting activity during acquisition of learning. To test this, we first assessed burst firing parameters of putative VTA DA neurons during CS and US across all learning stages in animals with tetrode and silicon probes (Fig. 5A). No significant differences were observed in total burst count during CS-US periods or in the mean spike frequency within bursts across baseline, acquisition, and plateau phases of learning (Supplementary Fig. 7A, B). However, during the CS and US, the mean number of spikes within a burst was significantly higher during acquisition (Fig. 5B). The onset of CS-induced bursting occurred approximately 100 ms after the initiation of VTA GABA synchrony.

**Fig. 5: Changes in KCC2 function enhance DA signaling in the NAc and VTA.**

Next, we reassessed burst firing parameters during CS-US in animals that received intra-VTA CLP290 or vehicle, administered ipsilateral to the microdrive implantation on the day after reaching the middle of acquisition (Fig. 5C). There were no significant differences in the total number of bursts or intra-burst spike frequency during CS-US in VTA DA neurons between the two groups (Supplementary Fig. 7C, D). However, CLP290 treatment significantly reduced the mean number of spikes per burst during both the CS and US compared to vehicle-treated controls (Fig. 5D).

Finally, we examined the impact of suppressing VTA GABA neuron activity on spikes per burst during the CS and US in the acquisition phase. When VTA GABA neurons were photoinhibited ipsilateral to the electrophysiological recordings on the day following the midpoint of acquisition (Fig. 5E), Arch-expressing rats showed a significant reduction in the mean number of spikes per burst during CS and US compared to GFP-expressing controls (Fig. 5F).

Since reward learning involves DA signaling in the NAc lateral shell⁶, we hypothesized that KCC2 downregulation in the VTA might also alter DA release in this region. To test this hypothesis, we injected a genetically encoded DA sensor GRAB_DA2m³¹ and implanted an optical fiber in the NAc lateral shell (Supplementary Fig. 7E and Fig. 5G). Rats were subjected to the Pavlovian conditioning task and DA transients were measured during baseline and acquisition of learning. Fiber photometry recordings in the NAc revealed increased DA responses to the reward-predictive cue, while DA responses to reward delivery remained unchanged (Fig. 5H and Supplementary Fig. 7F). Unilateral CLP290 microinfusions the day after animals reached the middle of acquisition (Fig. 5I) attenuated DA responses to CS in the lateral shell (Fig. 5J and K). In contrast to our bursting data, CLP290 administration did not alter DA responses to the US (Fig. 5J, K).

Optogenetic synchronization of VTA GABA neurons enhances stimulus-induced burst firing ex vivo and in vivo

Based on our findings in Figs. 4, 5, we hypothesized that VTA GABA network synchronization potentiates phasic DA firing. To test this, we aimed to mimic learning-mediated neuronal synchronization by using optogenetic stimulation of VTA GABA neurons. First, we determined the frequency at which spikes in one VTA GABA neuron exhibited millisecond synchrony with spikes in another VTA GABA neuron during the acquisition period. To this end, we generated joint peristimulus time histograms that enabled the quantification of time intervals between two consecutive instances of synchrony in spike trains from pairs of VTA GABA neurons (Supplementary Fig. 8A). This analysis revealed that during acquisition, pairs of VTA GABA neurons displayed millisecond time frame synchrony at a mean frequency of around 10 Hz (Supplementary Fig. 8B, C).

Next, we studied the effects of optogenetically-induced GABA network synchronization on DA bursting. GAD-Cre animals were injected with Cre-dependent ChR2 in the VTA to stimulate GABA neurons at 10-Hz frequencies. In the first set of experiments, we prepared midbrain horizontal slices and performed patch-clamp cell-attached recordings of DA neurons (Fig. 6A). Putative DA neurons in the lateral VTA were identified using established electrophysiological criteria and suppression of spontaneous action potential firing during ChR2 stimulation (Supplementary Fig. 9A–D, see Methods). Phasic bursts of action potentials were triggered by direct iontophoretic application of glutamate near recorded neurons. Consistent with previous reports, iontophoretic application of glutamate elicited high-frequency spike trains in lateral VTA DA neurons (Fig. 6B)³². Strikingly, the same DA neurons showed an increased number of spikes per burst when glutamate application was paired with 10-Hz light-induced synchronization of VTA GABA neurons (Fig. 6C, D). Optogenetic VTA GABA neuronal synchronization did not alter the mean spike frequency within glutamate-induced bursts in DA neurons (Supplementary Fig. 9E).

**Fig. 6: Optogenetic synchronization of VTA GABA neurons enhances stimulus-induced burst firing ex vivo and in vivo.**

Qualitatively similar effects of GABAergic synchronization on phasic DA firing were observed during in vivo single-unit recordings of VTA DA neurons in anesthetized rats. Two-to-three weeks after Cre-dependent ChR2 injections, a glass electrode coupled to an optic fiber was lowered to the lateral VTA of GAD-Cre rats (Fig. 6E). Putative lateral VTA DA neurons were identified based on their electrophysiological properties and light-induced suppression of spontaneous action potential firing (Supplementary Fig. 9F, G, see Methods). Stimulus-evoked burst activity in DA neurons was assessed via electric stimulation of the pedunculopontine tegmental nucleus (PPTg, Fig. 6E), which relays cue-related sensory information to these cells^33,34,35. In 8 of 15 recorded DA neurons, PPTg stimulation elicited a burst of action potentials (Fig. 6F). In those cells in which bursts were evoked, PPTg stimulation was repeated with concurrent optogenetic synchronization of VTA GABA neurons at 10 Hz. Notably, optogenetic VTA GABA synchronization potentiated PPTg-driven phasic bursting by increasing the number of spikes within a burst (Fig. 6G, H). In contrast, synchronizing GABA neuron activity did not alter the intraburst firing rates in lateral VTA DA neurons (Supplementary Fig. 9H).

Discussion

Associative reward learning is pivotal for an organism’s survival, but the neuronal adaptations underlying this fundamental process have not been well delineated. We found that circuit-specific changes in Cl^- homeostasis within GABA neurons of the VTA contribute to the formation of Pavlovian cue-reward association. Specifically, during the acquisition phase of learning, the neuron-specific Cl^- transporter KCC2 undergoes transient downregulation, resulting in depolarized E_GABA in VTA GABA neurons. These alterations were associated with increased firing synchrony within VTA GABA neuronal networks and enhanced phasic bursting in VTA DA neurons. Most importantly, enhancing KCC2 function or silencing GABA networks in the VTA during acquisition of learning attenuated the formation of cue-reward associations.

Despite considerable focus on VTA DA neurons in reward learning^5,8,36, learning-related changes in VTA GABA neurons remain poorly understood. To address this, we used a photoconvertible Ca²⁺ indicator to selectively examine VTA GABA neurons active during cue and reward presentations¹⁹. In these neurons, Cl^- homeostasis was altered only during the acquisition phase of learning, when conditioned responding was rapidly growing. We found that functional KCC2 was downregulated via dephosphorylation at S940, occurring during acquisition but not maintenance of the cue-reward association. This dephosphorylation, mediated by protein phosphatase 1, promotes KCC2 internalization and can be triggered within minutes by NMDA receptor-mediated Ca²⁺ influx, which is elevated in VTA GABA neurons during acquisition (Supplementary Fig. 1F). Re-phosphorylation of S940 is driven by PKC signaling, which can also be rapidly activated via metabotropic glutamate or serotonin receptors^18,37,38. While other regulatory mechanisms – such as alternative phosphorylation sites, small non-coding RNAs, and transcriptional control – can influence KCC2, they typically also alter NKCC1 activity or KCC2 protein levels^{39,40,41,42,43,44}. However, our data show no changes in total KCC2 or NKCC1 protein expression (Supplementary Fig. 1M, N), and pharmacological activation of KCC2 alone was sufficient to restore VTA GABA signaling (Figs. 2, 4). These findings strongly support S940 dephosphorylation as the primary driver of the transient Cl⁻ dysregulation observed during learning. Interestingly, similar transient dynamics have been observed in glutamatergic receptor changes on DA neurons during learning³⁶, suggesting that short-term changes in VTA neurons act to establish new associations but are not necessary for the long-term storage of cue-reward information. The preservation of reward-related memories may rely on other mechanisms outside the VTA, while restoring Cl^- homeostasis after acquisition of associations may facilitate future reward learning.

Prior studies have shown that KCC2 downregulation can lead to paradoxical excitation of VTA GABA neurons in response to GABA_A receptor stimulation^13,15. Perhaps unsurprisingly, when we enhanced GABA_A receptor function in VTA GABA neurons with diazepam, reduced KCC2 function led to increased GABA release on downstream targets (Fig. 2F). The unexpected finding, however, was that during acquisition phases of learning, KCC2 downregulation increased coincident spiking amongst VTA GABA neurons (Figs. 2E, 4). The link between KCC2 and synchrony, rather than firing rate, may stem from impaired inhibitory coupling among lateral VTA GABA neurons^9,26. This idea is supported by studies in the substantia nigra pars reticulata, where sparse, hyperpolarizing GABAergic collaterals have minimal impact on baseline firing but reduce synchronous responses to shared input^45,46. A similar architecture exists in the VTA, where KCC2 downregulation during learning may weaken collateral inhibition, allowing shared input to enhance synchrony without affecting firing rates. Additionally, reduced KCC2 and depolarizing GABA_A signaling may shift collateral inhibition toward excitation of neighboring neurons, further promoting millisecond-scale synchrony with little effect on average firing.

Previous in vivo multi-unit recordings have reported learning-related increases in synchronized GABA neuron activity in the VTA and substantia nigra, along with enhanced oscillations below 10 Hz^47,48, but the causal role of this synchrony in learning have remained unexplored. Our results show that conditioned learning is disrupted when network synchronization is suppressed via KCC2 activation. Although reduced KCC2 function and increased synchrony have been linked to pathological states in the adult brain⁴⁹, growing evidence also implicates Cl^- transport in physiological synchrony, such as during sleep and seasonal rhythms in the hypothalamus and cortex^50,51. Together, our findings suggest that KCC2 downregulation is a key driver of learning-related synchrony in VTA GABA neurons, and more broadly, that KCC2-mediated Cl^- dynamics may serve as a common mechanism for neuronal synchronization in both pathological and physiological contexts.

At first glance, the notion that KCC2 downregulation and resulting GABAergic synchronization enhance cue-induced phasic DA signaling might appear contradictory. Previous work has associated KCC2 downregulation with reduced DA neuron firing in the VTA and blunted DA release in the NAc^13,15,17; however, these studies primarily examined DA firing in acute brain slices or anesthetized animals. Additionally, measurements of DA release in the NAc were conducted using microdialysis over minutes, which may not capture millisecond-scale changes in phasic DA release. Despite reports of attenuated DA activity, previous studies have also identified correlations between KCC2 downregulation in VTA GABA neurons and increased consumption of addictive drugs^13,15. Hence, we propose that enhanced phasic DA signaling due to KCC2-induced changes in VTA GABA neuron firing patterns may constitute one mechanism that promotes addictive behaviors. In support of this hypothesis, mounting evidence demonstrates that VTA GABA neurons exhibit excitatory responses in response to reward-related stimuli (Supplementary Fig. 4)^11,30,52. There is also increasing recognition of the crucial role of concerted activity in both VTA DA and GABA neurons in enhancing phasic DA neuron firing^30,53. Most notably, mathematical simulations revealed that synchronized inhibitory input to DA neurons can evoke additional spikes during bursting³⁰. Moreover, electrophysiological experiments suggest that brief inhibitory pulses interleaved with depolarizing currents can extend burst firing in midbrain DA neurons³². These studies suggest two major mechanisms by which synchronized inhibition can augment DA burst responses: First, removal of inhibition between GABAergic pulses increases the probability of spike generation in DA neurons³⁰. Second, GABA-mediated hyperpolarization attenuates depolarization-induced inactivation of the sodium current in DA neurons, a putative mechanism for burst firing termination³². Our slice and in vivo experiments strongly support these mechanisms by showing that optogenetic synchronization of VTA GABA neurons enhances phasic bursting of DA neurons via increasing the number of spikes within a burst (Fig. 6).

A dominant model of reward learning posits that DA neurons signal a reward prediction error (RPE) – the difference between expected and received reward^5,8. VTA GABA neurons are thought to contribute to this process by inhibiting DA neurons during the delivery of expected rewards, thereby shaping reward expectation¹⁰. Notably, most previous studies supporting this model have examined neural activity after cue-reward associations have already been established. In contrast, our findings reveal a previously unknown role for VTA GABA neurons during the acquisition phase of learning. Specifically, we found that either KCC2 activation or suppression of VTA GABA neurons delayed learning (Fig. 3B–J) and led to reduced burst firing of DA neurons in response to both cues and rewards (Fig. 5D, F). This attenuation of DA activity during reward delivery suggests that disrupting VTA GABA synchrony hinders learning via impairing the encoding of RPEs⁵⁴. However, our results do not exclude alternative theoretical models beyond the RPE framework. For example, dopamine signals may also support learning through mechanisms such as incentive salience⁵⁵, behavioral activation or flexible approach strategies^56,57, or inferred causality between cue and reward⁵⁸. These alternatives are compatible with our fiber photometry data, which, unlike DA spiking, showed no significant change in reward-evoked DA release following KCC2 activation (Fig. 5K). This dissociation between DA neuron firing and DA release could reflect axon terminal modulation in the NAc⁵⁹. Alternatively, it may stem from technical limitations of the DA sensor used in our study, which can exhibit reduced sensitivity at high DA concentrations⁶⁰. Notably, US-evoked DA signals were substantially larger than cue-evoked signals in our experiments (Fig. 5H). Together, these findings highlight the complexity of reward circuitry and identify KCC2-dependent changes in VTA GABA neurons as a mechanism critical for reward-related learning. Future work should explore how KCC2 modulation aligns with distinct theoretical frameworks of reward-related learning.

A guiding principle for the development of neuropsychiatric disorders is that normal learning and memory processes are ‘usurped’ into a pathological state². Several studies have highlighted that KCC2 downregulation represents a common yet often overlooked form of VTA GABA-specific dysregulation in maladaptive states^15,16,17. Our results suggest that this GABA-specific switch in Cl^- gradient is a normal form of neuroadaptation during learning that is ‘hijacked’ in addiction and other neuropsychiatric disorders. In summary, given that impaired reward learning is central to many neuropsychiatric disorders and aberrant function of KCC2 is observed after exposure to drugs of abuse and stress, our findings provide a broader context for understanding neuropsychiatric disorders and offer a rich avenue for future investigations.

Methods

Subjects

The following rat lines (200–500 g, 2 months old, males and females) were used for the experiments: GAD-Cre Long-Evans rats (Harlan/Envigo/Inotiv), TH-Cre Long-Evans rats (Harlan/Envigo/Inotiv), and wild-type Long-Evans rats (Harlan/Envigo/Inotiv). Littermates of the same sex were randomly allocated to each experimental group. Rats were maintained on a 12-h light-dark cycle with food ad libitum and were handled at least 5 days prior to the onset of surgery/behavioral testing. During behavioral testing, animals were food restricted but maintained 90% of their original weight. Room temperatures were kept around 22–25° Celsius and at 55% humidity. Animals with silicon or tetrode probe implantations were single-housed for the protection of the implant. All procedures and animal care standards were approved by Georgetown University’s IACUC committee and Division of Comparative Medicine.

Stereotaxic surgeries

All animal surgeries were performed under isoflurane gas anesthesia (2–3% in 100% O2, flow rate 0.8–1 L/min) using a stereotaxic apparatus (Stoelting Co, Model 51900). All AAVs used in this study were from Addgene. The injections were performed using a 10 μL Hamilton syringe at a rate of 0.2 μL/min, resulting in a final volume of 0.7 μL, administered with a microinjection pump (KDS Legato). All the intra-VTA virus infusions were performed at the following coordinates: anterior-posterior (AP) = −5.40 ÷ −5.50, medial-lateral (ML) = ± 0.90 ÷ ± 1.00, dorsal-ventral (DV) = −7.9 ÷ −8.1 (with slightly smaller numbers for female and larger numbers for male rats (f/m)). For labeling VTA GABA populations that were active during learning, GAD-Cre rats were injected unilaterally with AAV1-hSyn-NES-CaMPARI2 and AAV8-EF1α-Con/Foff2.0-BFP. For in vivo VTA GABA silencing experiments, GAD-Cre rats received unilateral injections of AAV9-flex-Arch-GFP. For Cre-dependent KCC2 gene silencing, shRNA constructs targeting KCC2 sequences²⁸ were generated using the pSico-Red system by Vector Biosystems²⁹. AAV8(or DJ)-EF1-mCherry-SICO-GFP-KCC2-shRNA or AAV8(or DJ)-EF1-mCherry-SICO-GFP-scrmbl-shRNA was then bilaterally injected into the VTA of GAD-Cre animals. This vector constitutively expresses mCherry and contains a GFP cassette downstream of the U6 promoter, followed by a stop codon and the shRNA sequence. In the absence of Cre-recombinase, transcription from the U6 promoter produces GFP, but not shRNA. In Cre-expressing cells, the GFP cassette and stop codon are excised, enabling expression of KCC2 or scrambled shRNA along with mCherry.

For VTA GABA and DA neuron activation, AAV5-EF1a-DIO-hChR2(H134R)-eYFP was injected bilaterally into the VTA of GAD-Cre and TH-CRE animals, respectively. Following injections, craniotomy sites were covered with bone wax. For some experiments, intra-VTA virus infusions were followed by chronic implantation of an optic fiber (200 µm, 0.22 NA, Newdoon Inc) slightly above the VTA: DV = −7.80 ÷ −8.00 (f/m). For in vivo pharmacology experiments, a stainless-steel guide cannula (26 G, 1.8 mm spacing, 10 mm below pedestal, Protech International) was bilaterally implanted above the VTA at coordinates: AP = −5.40, ML = ± 0.90, DV = −7.25 mm (m). Multiple layers of cement (C&B Metabond; Parkell) were applied to secure the optic fiber or guide cannula to the skull. For labeling NAc-projecting DA neurons, AAV2/5-pCAG-Flex-eGFP was bilaterally injected into the NAc medial shell (AP = +1.5, ML = ± 0.7, DV = −7.3 ÷ −7.5 mm (f/m)), NAc core (AP = +1.7, ML = ± 1.6, DV = −6.5 ÷ −6.7 (f/m)), or NAc lateral shell (AP = +1.1, ML = ± 2.4, DV = −7.8 ÷ −8.00 mm (f/m)). Details for in vivo electrode, silicon probe, and fiber photometry implantations are provided under In Vivo Electrophysiology and In Vivo Fiber Photometry sections.

Behavioral procedures

Rats underwent mild food restriction before behavioral sessions, aiming to achieve 90% of their free-feeding weight (standard laboratory chow at ~9 g/d for females and ~15 g/d for males). Behavioral sessions took place in standard operant chambers (MedAssociates) equipped with grid floors, a house light, a food tray, and auditory stimulus generators emitting tones at 4.5 kHz frequencies. Rats were habituated to the chamber and food retrieval through a magazine training session, during which 15 sugar pellets (BioServ) were delivered non-contingently at variable intervals of 45 ± 15 s. Subsequently, rats underwent up to 13 Pavlovian conditioning sessions (1 session per day), each comprising 50 trials. During these trials, the termination of a 5-s audio cue (Conditioned Stimulus (CS); 4.5 kHz tone) prompted the delivery of a single sugar pellet (Unconditioned Stimulus (US)) and the illumination of the food port light for 5 s. The audio cue and sugar pellet were delivered at 45 ± 15 s variable intervals. Conditioned responding was quantified by assessing the rate difference, i.e., the change in the rate of head entries during the 5-s CS period compared to the 5 s preceding CS delivery⁶¹. During the control unpaired conditioning sessions, CS and US were presented in the same trial in random order but never close together in time (12–30 s between CS and US). During the collection of all behavioral assays, experimenters were blinded to all treatment conditions.

Drugs and experimental design

All drugs were dissolved in sterile saline; however, if specified otherwise, a stock solution was prepared with drugs dissolved in Dimethyl Sulfoxide (DMSO), which was then further diluted in saline on the day of treatment. The intra-VTA concentration of CLP290 was 60 μM delivered at 0.5 μL/min with a total volume of 1.0 μL delivered 1 h prior to the behavioral task. CLP290 is a carbamate prodrug of CLP257 and should be metabolized by carboxylesterase, which is mostly abundant in the liver and intestine⁶². Due to its short half-life (<15 min), CLP257 is not expected to be effective for behavioral procedures used in this study⁶³. However, previous findings have demonstrated carboxylesterase expression in the brain^64,65,66,67 and have shown that carbamate bonds can be broken even in brain extracts^68,69. Most importantly, our previous experiments have confirmed that CLP290 is effective in impacting KCC2 levels in VTA GABA neurons in slice and when applied locally in the VTA^14,15. VU0463271 was used at a concentration of 100 μM and was injected into the VTA at 1 μL/min with a total volume of 1 μL during the in vivo recording session⁷⁰. Intra-VTA microinfusions were delivered by pump and the microinfusion injector was left in place for 2 additional min and then removed. After the experiments, Pontamine Sky Blue was injected into the VTA to determine the location of the microinfusion.

Ex vivo electrophysiology

Horizontal slices (220 μm) containing the VTA were cut (VT1200s, Leica Microsystems) from adult Long–Evans rats in ice-cold, oxygenated (95% O₂, 5% CO₂), high-sucrose artificial cerebrospinal fluid (aCSF): 205.0 mM sucrose, 2.5 mM KCl, 21.4 mM NaHCO₃, 1.2 mM NaH₂PO₄, 0.5 mM CaCl₂, 7.5 mM MgCl₂, and 11.1 mM dextrose. Immediately after cutting, slices were transferred to normal aCSF buffer: 120.0 mM NaCl, 3.3 mM KCl, 25.0 mM NaHCO₃, 1.2 mM NaH₂PO₄, 2.0 mM CaCl₂, 1.0 mM MgCl₂, 10.0 mM dextrose, and 20.0 mM sucrose. The slices were constantly oxygenated (95% O₂, 5%CO₂) and maintained at 32 °C in aCSF for 40 min, then at room temperature for at least 60 min prior to slice electrophysiology. To perform electrophysiological recordings, slices were transferred to a holding chamber and perfused with normal aCSF at a constant rate of 2–3 mL/min at 32 °C. Patch electrodes made of thin-walled borosilicate glass (1.12 mm ID, 1.5 mm OD, World Precision Instruments) had resistances of 1.0–2.0 MΩ when filled with the internal solution: 135.0 mM KCl, 12.0 mM NaCl, 2.0 mM Mg-ATP, 0.5 mM EGTA, 10.0 mM HEPES, and 0.3 mM Tris-GTP (pH 7.2–7.3). The liquid junction potential between the bath and the pipette solutions was corrected. Recordings were made using Axon Instruments Multiclamp 700B amplifier, filtered at 10 kHz, digitized at 20 kHz using pClamp 11, and analyzed using Clampfit 11 (Molecular Devices).

For E_GABA perforated-patch recordings in GAD-Cre rats, putative lateral VTA GABA neurons were identified based on the expression of Cre-dependent BFP. Gramicidin was first dissolved in methanol to a concentration of 10 mg/mL and then diluted in a pipette solution to a final concentration of 150 μg/mL. Recordings were conducted in the presence of 6,7-dinitro quinoxaline-2,3-dione (DNQX, 20 μM, Sigma), DL-2-amino-5-phosphonopentanoic acid (AP5, 50 μM, HelloBio), CGP55845 (1 μM, Sigma), and tetrodotoxin (0.5 μM, Abcam) to isolate GABA_A receptor-activated currents. Once perforation was achieved, a bipolar tungsten-stimulating electrode (World Precision Instruments) was positioned 100–150 μm away from the recording electrode to evoke inhibitory postsynaptic currents (eIPSCs) measured under voltage clamp at various holding potentials. The amplitudes of eIPSCs were plotted against voltage to estimate the reversal potential of E_GABA (Illustrated in Fig. 1E). After each cell-attached experiment, recordings were converted to the whole-cell configuration and I_h current was measured.

For spontaneous IPSC (sIPSC) recordings in Th-Cre rats, projection-specific DA neurons were identified based on the expression of Cre-dependent eGFP. In whole-cell configuration, spontaneous IPSCs (sIPSCs) were recorded in voltage clamp while holding VTA DA neurons at −60mV. Before each experiment, the I_h current was measured. To isolate sIPSCs, AP5 and DNQX were added to the aCSF perfusing solution to inhibit ionotropic glutamatergic synaptic transmission. Additionally, we employed a high Cl^- intracellular solution, which helps to detect individual sIPSCs by improving the signal-to-noise ratio^13,15. Following a 10-min baseline period, diazepam (5 μM, Sigma) was added to the perfusion solution (aCSF) to assess diazepam-induced alterations in sIPSCs. In experiments using the KCC2 activator, CLP290, slices were incubated in CLP290 (10 μM, Aobious) dissolved in aCSF solution for an additional hour prior to recording. In experiments with the KCC2 antagonist, VU0463271 (10 μM, Tocris) was dissolved in aCSF and bath applied during recordings. During sIPSC analysis, we carefully controlled for noise by removing events with deflections smaller than 10 pA and excluding events smaller than 3 standard deviations of the baseline noise. These steps helped to minimize spurious detections that could arise from noise during the rise time and early decay phases of the IPSCs.

For experiments examining glutamate-evoked DA neuron firing in GAD-Cre rats, the firing rates of lateral VTA DA neurons were recorded in cell-attached configuration in passive voltage-follower mode. DA neurons were identified in the lateral VTA by their morphology (>20 μm soma size), low firing frequency (1–5 Hz), which was suppressed during ChR2 stimulation of VTA GABA neurons, and the presence of a large I_h current, which was measured in whole-cell configuration upon the termination of cell-attached experiment. Together, these parameters were previously shown to correlate (~98%) with tyrosine hydroxylase (TH)-positive cells¹⁵. For iontophoretic glutamate application, iontophoresis pipettes (30–90 MΩ) were filled with 1 M Na + -glutamate. A retaining current (+10 to +15 nA) was applied to control for glutamate leakage from the pipette. The ejection currents (−10 to −100 nA, 0.3 to 1 s) were used to apply glutamate 100–150 μm away from the recording electrode. Iontophoretic glutamate application was performed with and without optogenetic stimulation of neighboring ChR2-expressing GABA neurons. Here, ChR2 was stimulated by flashing 473 nm light through the light path of the microscope (10 Hz, 5 ms pulse, 10 s train) using pE-300 LED Illumination System (CoolLED) and Master-9 pulse stimulator (A.M.P. Instruments). The whole slice was illuminated, and the light intensity of the LED was not changed during the experiments. During all slice electrophysiological data collection, experimenters were blinded to all treatment conditions.

Histology and microscopy

For all immunohistochemistry (IHC) experiments, animals were transcardially perfused immediately following completion of behavioral testing. Animals were first deeply anesthetized with isoflurane followed by perfusion with 50 mL of phosphate buffered saline (PBS, EMD Millipore) and subsequently 50 mL of 4% paraformaldehyde (PFA, Chembiotec). Brains were harvested and post-fixed in 4% PFA for 2 h. Following post-fixation, brains were transferred into increasing concentrations of sucrose solutions dissolved in PBS accordingly (10, 20, 30%) until equilibration. Brains were then embedded in OCT, frozen and stored at −80 °C. 30 µm sections containing the VTA were collected, and free-floating sections were stored in PBS. For CaMPARI2 expression labeling, perfusion, and IHC experiments, all PFA/PBS solutions additionally contained 10 mM EGTA (PBS-E, Sigma).

To label red-shifted CaMPARI2-expressing GABA neurons, we followed the IHC protocol established by Moeyaert et al. 2018¹⁹. Tissues were washed 3 x 10min using wash buffer (PBS-E, 0.5% Tween 20, 10 µg/ml Heparin) and then placed in a permeabilization buffer (PBS-E, 0.5% TX, 20% DMSO, 23 mg/ml Glycine) for 3 h at 37 °C. Tissues were then blocked (PBS-E, 0.5% TX, 10% DMSO, 6% NGS) for 3 h at 37 °C. Following blocking, tissues were then incubated with primary anti-CaMPARI2 1:1000 (Absolute Antibody, PBS-E, 0.5% Tween 20, 5% DMSO, 3% NGS, 10 µg/ml Heparin) for 20 h at 37 °C. Tissues were removed and washed 3 times for 10 min each, then incubated with the secondary antibody goat-anti-mouse 594 at 1:500 dilution (0.5% Tween 20, 3% NGS, 10 µg/mL Heparin). After incubation, tissues were rinsed 3 times for 10 min each in PBS-E, mounted, and cover slipped using Vectashield (Vectorlabs). Images were acquired using a Leica Mica Microhub System (Leica Microsystems). Regions of interest (ROIs) were defined, and image montages were captured using a 10x/0.32 dry-immersion lens with a zoom factor of 1.24. All illumination parameters were automatically set using the OneTouch imaging settings within LAS X software (Leica Microsystems). Images were exported and further processed using Fiji (ImageJ) and analyzed using Qupath. Finally, sections were labeled relative to the bregma using anatomical landmarks and neuroanatomical terminology according to a rat brain atlas.

To label DA projections to the NAc subregions, tissues were washed 3x10min in PBS followed by a 2 h incubation in blocking buffer consisting of 3% normal goat serum (NGS), 0.3% Triton X-100 (TX, Sigma) in PBS. After blocking, tissues were incubated in a blocking buffer with a rabbit anti-TH primary antibody 1:1000 (Invitrogen) for 12 hrs at room temperature. The following day, tissue was again rinsed with PBS 3x10min and then placed into a blocking buffer containing a goat anti-rabbit 594 secondary antibody 1:1000 (Invitrogen) for 2 hrs at room temperature. Tissues were then rinsed in PBS 3×10 min, mounted, and cover slipped using Vectashield (Vectorlabs). After curing for 24 h at room temperature, slides were boxed and stored at −20 °C until imaging was performed, typically within 72 h of slides curing.

To label GRAB_DA2m and DA terminals for photometry experiments, tissues were incubated in blocking buffer with primary antibodies rabbit anti-TH 1:500 (Invitrogen) and a mouse anti-GFP 1:500 (Invitrogen) for 12 hrs at room temperature. The following day, tissue was rinsed with PBS 3x10min and then placed into a blocking buffer containing secondary antibodies goat anti-rabbit 594 1:1000 (Invitrogen) and a goat anti-mouse 488 1:1000 (Invitrogen) for 2 hrs at room temperature. Sections were then washed in PBS 3 x 10min, mounted and coverslipped using Vectashield (Vector Labs) and DAPI. To label KCC2 expression in KCC2 shRNA or scrambled shRNA expressing neurons, tissues were incubated in blocking buffer with primary rabbit anti-KCC2 antibody 1:500 (Sigma) overnight at 4 °C. The following day, tissue was rinsed with PBS 3 × 10 min and then placed into a blocking buffer containing secondary antibodies goat anti-rabbit 647 1:1000 (Invitrogen). Sections were then washed in PBS 3x10min, mounted and coverslipped using Vectashield (Vector Labs). Confocal image montages were captured using a 63x water-immersion lens. All illumination parameters were automatically set using the OneTouch imaging settings within LAS X software (Leica Microsystems). Images were exported and further analyzed using Fiji. ROIs, defined as individual cell borders, were used to quantify KCC2 fluorescence intensity.

Western blots

Tissue punches of the Ventral Tegmental Area (220 μm) were extracted from horizontal slices of adult Long-Evans rats at baseline, acquisition, and plateau phases of learning (VTA slices were prepared as described in the ex vivo electrophysiology section) and homogenized in 100 µL of ice-cold RIPA buffer (Thermo Fisher Scientific) containing phosphatase and protease inhibitors (Thermo Fisher Scientific). Protein concentrations were detected using a Qubit fluorometer (Thermo Fisher Scientific). Samples (30 µg of protein in 2.5% Beta Mercaptoethanol and Laemmli sample buffer solution) were separated through a 4–15% Tris-Glycine gel (Bio-Rad Laboratories). Following electrophoresis, proteins were transferred to a nitrocellulose membrane (Bio-Rad Laboratories) at 100 V for 30 min in a transfer buffer (25 mM Tris, 192 mM glycine, 0.1% SDS, 20% methanol) and blocked in milk (2.5%) for one hour. The primary antibodies applied were rabbit anti-KCC2 antibody 1:2000 (Sigma), rabbit anti-Phospho-Ser940 KCC2 antibody 1:500 (Phospho-Solutions), NKCC1 antibody at 1:1000 (DSHB), and mouse anti-GAPDH antibody at 1:300 (Sigma). Goat anti-rabbit IgG (H + L) and goat anti-mouse IgG, IgM (H + L) secondary antibodies (Invitrogen) were used at a dilution of 1:5000. All antibodies were diluted in a Signalboost solution (Sigma). Membranes were developed for imaging using Tropix CDP-Star solution, then scanned using an AI600 Chemiluminescence Imager. Densiometric analyses were performed using Fiji (ImageJ) open-source software. The optical densities of KCC2 and Phospho-Ser940 KCC2-specific bands were measured and normalized to GAPDH.

In vivo fiber photometry and DA signal analysis

For in vivo fiber photometry experiments, male rats were unilaterally injected with AAV9-hSyn-GRAB_DA2m and implanted with an optical fiber (400 μm, 0.66 NA, Doric Lenses) targeted to the NAc Lateral Shell. To combine fiber photometry with in vivo pharmacology, a stainless-steel guide cannula (26 G, 1.8 mm spacing, 10 mm below pedestal, Protech International) was bilaterally implanted above the VTA. DA transients were recorded using the Synapse Software and real-time LUX RZ10X processor (Tucker Davis Technologies). Fluorescent signals were obtained by stimulating cells expressing GRAB_DA2m with a 465 nm LED in freely behaving rats. GRAB_DA2m transients were processed using pMAT software, developed by Bruno et al. 2021⁷¹. Noise-related changes in fluorescence across the whole experimental session were removed by scaling the isosbestic control signal (405 nm) and regressing it onto the DA-sensitive signal (465 nm). This regression generated a predicted model of the noise based on the isosbestic control. DA-independent waveforms on the 405 nm model were then subtracted from the raw GRAB_DA2m signal to remove artifacts such as movement and photobleaching. Time stamps were acquired using transistor-transistor logic (TTL) pulses generated by Med Associates chambers. Peri-event time histograms were constructed using 100 ms bins surrounding the time points of interest. Each bin was normalized (z-scored) on original ΔF/F signal using the pre-CS time-window (−5 to 0 s on Fig. 5B). Custom MATLAB scripts were utilized to examine the area under the curve (AUC). The AUC was defined as the integral between the first peak since stimulus presentation and its eventual decay back to 0.

In vivo electrophysiology in awake rats and data analysis

For in vivo electrophysiology experiments, male rats underwent unilateral implantation of an optic fiber (200 μm, 0.22 NA, Newdoon Inc) coupled to either a silicon probe or a tetrode microdrive in the lateral VTA at coordinates: AP = −5.40; ML = +0.90; DV = −8.1. Tetrode microarrays comprising 16 tetrodes (64 channels) were assembled using tungsten wires (California Fine Wire), customized electronic interface boards (https://likhtiklab.com/tools), and 36-channel connectors (Omnetics). 128-channel silicon probes were obtained from Diagnostic Biochips. 3D-printed plastic microdrives and headcap systems for silicon probes were adapted from Vöröslakos et al. 2021⁷². In the experiments with VU0463271 and CLP290 intra-VTA microinfusions, a drug infusion cannula was coupled to the microdrives. Plastic shielding encasing the microdrives were lined with copper tape to function as a Faraday enclosure. A thin layer C&B Metabond was initially applied on the skull followed by dental cement (Stoelting), to secure the headgear and shielding in place. Two ground screws were affixed adjacent to the craniotomy and soldered to the ground wires on the microdrives.

Recording sessions were conducted in rats during the Pavlovian learning paradigm over approximately two weeks. Data were collected using an Intan headstage for tetrode recordings, Open Ephys Acquisition board, Open Ephys plugin, and processed using Kilosort 2.5 (https://github.com/cortex-lab/kilosort), followed by manual curation of units using Phy2 (https://github.com/cortex-lab/phy). Post-processing was conducted using Cell Explorer (https://cellexplorer.org) and custom MATLAB scripts. During data collection and analysis, investigators were blinded to treatment conditions.

Cell type classification

Spikes were curated using Phy2, and waveform metrics were processed through Cell Explorer. To distinguish between distinct neuron types in the VTA, we employed an opto-tagging approach to obtain ground-truth data for the classification of putative DA and GABA neurons. For DA neuron identification, TH-Cre animals were injected with AAV5-EF1a-DIO-hChR2-EYFP (Supplementary Fig. 4A) and subjected to 40 Hz, 450 nm light stimulation (~5–8 mW·mm⁻²) for 2 s (Supplementary Fig. 4B, top). Units were identified as putative DA neurons and included in the ground-truth repository if they passed a spike latency test, demonstrating an average spike latency of less than 10 ms and a spike probability greater than 0.5 following each light pulse⁷³. For GABA neuron identification, GAD-Cre animals were injected with AAV-FLEX-Arch-GFP (Supplementary Fig. 4A) and GABA neurons were inhibited with 520 nm light (~5–8 mW·mm⁻²) for 10 s (Supplementary Fig. 4B, bottom). Archaerhodopsin (Arch) was selected for this protocol because VTA GABA neurons are known to exhibit a broad range of high firing rates⁷⁴; hence, employing an ChR2-induced opto-tagging protocol would make it challenging to differentiate between light-induced discharges and those due to intrinsic firing rates. Instead, Arch-induced inhibition provided a clear indication of neuronal responsiveness to light, regardless of firing rate. If the firing rate decreased by 50% compared to the rate measured 5 s before light application, the neuron was classified as a putative GABA neuron and included in a repository of ground-truth data. These opto-tagging protocols were applied starting a week prior to behavioral testing and daily at the end of each behavioral session. Importantly, our experiments measuring coordinated neuronal activity required simultaneous recordings from many pairs of neurons (at least 10 pairs of VTA GABA neurons). However, one technical limitation of using opto-tagging for neuron type labeling is that opsin gene expression levels can vary significantly across the targeted cells⁷⁵, and often we observed an insufficient number of units exhibiting notable responsiveness to light. This issue was compounded by the fact that extracellular recording probes sample only a fraction of the total neuron population. To address this, we used the spike features of opto-tagged units and developed a comprehensive protocol that (A) removed outliers that are neither DAergic nor GABAergic and (B) classified VTA GABA or DA neurons to include units that were not responsive to opto-tagging into our analyses (Supplementary Fig. 4C).

To first remove outliers, we constructed two Gaussian Mixture Models (GMMs) based on the opto-tagged dataset, using autocorrelogram (ACG) tau rise and firing rate as the variables. These models clustered the data and created probability distributions around each cluster: one around the opto-tagged DA units and the other around the opto-tagged GABA repository (Supplementary Fig. 4C). The rationale for selecting these spike features was twofold. First, firing rate has been a critical criterion for differentiating GABA and DA, as described in a number of seminal papers^76,77. Secondly, after comparing firing rate to an array of different spike features, ACG tau rise yielded the highest Fisher’s Score (used to evaluate the effectiveness of a set of features in distinguishing between different clusters) without overfitting and with minimal overlap between units within each GMM. Units beyond 3 standard deviations of at least one GMM were labeled as “unclassified,” accounting for 6.2% of the total recorded units, and removed from further analyses. This is consistent with previous literature indicating that 5–10% of VTA neurons are neither DAergic nor GABAergic⁷⁸.

Due to the overlapping spectra of spike shapes between the two cell type populations, assignment of recorded units to particular cell types remained difficult. Therefore, all units that were not excluded underwent further classification into “GABA” or “DA” neuron types using a decision tree model implemented in MATLAB (Version 2021a). This model was trained on key features extracted from the opto-tagged neurons, including ACG tau rise, ACG decay, ACG tau burst, ACG asymptote, firing rate, coefficient of variation of firing rate, firing rate interspike-intervals, and spike width (trough to peak) of the opto-tagged VTA GABA or DA units. The model achieved an accuracy rate of 97.5% distinguishing between these two neuron types on a separate ground-truth test dataset constructed after opto-identifying cell types.

Following neuron type classification, we further characterized neuron diversity by examining their response to reward-related stimuli during Pavlovian conditioning. To classify these response profiles, we applied Uniform Manifold Approximation and Projection (UMAP) to z-scored peri-stimulus time histograms of classified DA and GABA neuron types across pre-CS, CS, and US periods to identify distinct clusters. Unsupervised agglomerative hierarchical clustering was performed on the UMAP outputs, derived from the squared Euclidean linkage distance between data objects⁷⁹. The optimal number of clusters was determined using the elbow method, which allowed us to identify the point at which adding more clusters no longer significantly reduced the within-cluster variance⁸⁰. For each learning phase, two major clusters emerged: ‘type 1’, showing a positive response to stimuli, and ‘type 2’, comprising neurons with either no response or a negative inflection to stimuli. The z-scored response profiles are shown in Supplementary Fig. 4F, G, with the distribution of each neuron type depicted across learning.

Spike synchrony

Among GABA-identified neurons, pairwise synchronization was examined across learning phases. We utilized cross-correlograms (CCGs) to analyze spike timing correlations between GABA neuron pairs during 5 s of intertrial interval (pre-CS), CS, and US epochs. To do this, we compared the spike trains of two neurons by shifting one spike train forward and backward by ±10 milliseconds and measuring how closely the spikes from the two neurons aligned at each shift. We averaged these measurements across all 50 trials per session. A peak in the CCGs at or near zero shift indicated higher synchronization of spike firing between the two neurons. Only recordings with at least 10 pairs of GABA neurons were included in the analysis.

To assess the statistical significance of the peaks, spike train pairs were shuffled using the jitter method⁸¹, which accommodated variations in the underlying firing rate. Specifically, we randomly shuffled the spike trains from one neuron 100 times per trial within a ± 100 millisecond range. This created 100 versions of the CCGs from the shuffled dataset for each pair of neurons. We then calculated the average and standard deviation of these shuffled CCGs for each session. Peaks in the original CCGs were considered significantly synchronous if they exceeded 2 standard deviations above the average of the CCGs from the shuffled datasets⁸¹. We then measured the percentage of neuron pairs that exceeded this threshold for each animal across different learning phases.

When examining spike synchronization before and after the administration of VU0463271, animals were first placed in their home cage for 20 min to record basal activity. Subsequently, they were injected with VU0463271 and returned to their home cage for an additional 20–30 min for post-injection recording. Spike synchronization was assessed for 50-s periods: approximately 2 min before the injection as the baseline, and 5–10 min post-injection. Each 50-s epoch was divided into 10 intervals of 5 s to measure the time course of synchronization.

To explore the temporal evolution of spike-spike correlations, we generated a joint peri-stimulus time histogram (JPSTH). We plotted the spike trains of a neuron pair in 10 ms bins to create peri-event matrices. A Cartesian matrix was constructed where the horizontal axis represents bins for spikes from neuron A, and the vertical axis represents bins for spikes from neuron B. These joint histograms were summed across trials to obtain an average or cumulative representation of the joint spike activity (Supplementary Fig. 8A). To compute the inter-synchronization interval, we analyzed entries in the JPSTH matrix that were off the 45-degree diagonal, also referred to as the correlation time histogram (Supplementary Fig. 8B). Afterwards, we analyzed the frequency at which spikes from the two neurons overlapped within each trial (Supplementary Fig. 8C).

In Vivo Electrophysiology in Anesthetized Rats

For in vivo experiments examining the effect of VTA GABA synchronization on DA neuron firing in anesthetized animals, AAV5-EF1a-DIO-hChR2-EYFP was injected into the VTA of GAD-Cre animals. Two to three weeks following the injection, animals were anesthetized under isoflurane gas anesthesia, with their body temperature maintained at 37° Celsius using an isothermal pad. They were then positioned on a stereotaxic frame and an incision was made to expose the skull. Craniotomies were performed for ground electrodes, lateral VTA recordings and pedunculopontine tegmentum (PPTg) stimulation. For single-unit recordings and light stimulation in the lateral VTA, a recording electrode coupled to an optic fiber (300 μm, 0.22 NA, Thorlabs) was lowered into the VTA for single-unit recordings and light stimulation. Recording electrodes were pulled on a horizontal puller (Sutter Instrument) from filament-containing borosilicate glass (0.68 mm inner diameter (ID), 1.2 mm outer diameter (OD), World Precision Instruments) and backfilled with 0.5 M Na⁺- acetate and 2% Chicago sky blue (5–15 MΩ). The electrode-optic fiber complex was slowly lowered with a micromanipulator into the lateral VTA. Dopamine neurons were identified based on low frequency firing (0.1–10 Hz), broad triphasic action potential waveform (≥3 ms) and suppressed firing during ChR2 stimulation of VTA GABA neuron¹⁵. To evoke burst firing in lateral VTA DA neurons, bipolar stimulation electrode (World Precision Instruments) was inserted to the PPTg (AP = − 7.8, ML = 1.9 ÷ 2.1, DV = −7.0 ÷ −7.2). The PPTg was stimulated in single-pulse mode (0.1–1 mA) and bursts were defined as a series of >2 spikes with an initial interspike interval of ≤80 ms, and terminated when the ISI exceeded 160 ms⁷⁶. Signals were recorded using an AM Systems Model 1700 amplifier, filtered at 0.3– 5 kHz, and monitored using pClamp 8.0 (Molecular Devices) and an audio monitor (Grass Instruments).

After the experiment, the recording location was labeled by administration of Chicago Sky Blue dye by positive pressure (1–2 min) through the suction port in microelectrode holder. Next, the brains were isolated and were kept in 10% formalin for at least a day. Finally, the brains were cut into 75-μm coronal sections to verify the recording site by light microscopy.

DA neuron burst firing analysis

For each pre-CS, CS, and post-CS period, we calculated the total number of bursts, the number of spikes within each burst train, and the intraburst frequency. Bursts were defined as a series of more than 2 spikes, with an initial interspike interval (ISI) of ≤80 ms and terminating when the ISI exceeded 160 ms⁷⁶. The intraburst frequency was determined by dividing the number of spikes within a single burst by the duration of that burst.

Recording site identification

At the end of the experiments, the final recording sites for tetrode recordings were marked with electrolytic lesions (0.5 mA, 1 s per channel). For probes paired with injectors, Pontamine blue dye was injected 1 h prior to dissection. Following euthanasia, the brains were then fixed in 4% PFA in a phosphate buffer, sectioned using a vibratome, and imaged on a light microscope for post hoc confirmation (Supplementary Fig. 4H).

DeepLabCut video analysis

To characterize movement during classical conditioning, videos were recorded each day using an overhead webcam (29 fps) during the conditioning task. We utilized DeepLabCut, (https://github.com/DeepLabCut/DeepLabCut), an open-source pose estimation software to accurately label body parts of the rat and track movement over time⁸². To train a model, we labeled 400 frames, identifying the nose, head stage, right ear, left ear, tail base, and the food magazine location in the chamber. We subsequently trained the model to 500,000 iterations with a p-cutoff of 0.8 and pixel error of 15.85 px and analyzed videos taken 1 day before drug injection and the day of drug injection. This generated outputs of the Cartesian coordinates and associated likelihoods for each body part in each frame. DLCAnalyzer (https://github.com/ETHZ-INS/DLCAnalyzer) calculated the distance moving, speed, and time spent moving for each video. Points with a likelihood of less than 0.9 were excluded from the dataset.

Quantification and statistical analysis

Two-sided student paired t-tests or unpaired t-tests, one- or two-way ANOVA, three-way ANOVA, one- or two-way repeated measures (RM) ANOVA were used to determine statistical differences for anatomical, behavioral, and electrophysiological data using GraphPad Prism 9 (GraphPad Software). A mixed effect model was applied for the behavioral datasets in Fig. 1C, to account for variations in Ns across different days. Tukey’s post hoc analysis was applied when ANOVA showed a significant main effect. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. All data are presented as means ± SEM.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

All processed data supporting the findings of this study are deposited in Dryad (https://doi.org/10.5061/dryad.44j0zpcv7). The raw electrophysiology, imaging and behavioral datasets are too large to be publicly shared, yet they are available for research purposes from the corresponding author upon request. Source data are provided with this paper.

Code availability

The codes used or generated during the current study are available in GitHub repository (https://github.com/aolabgeorgetown/Ostroumovlab/tree/main/KCC2_Learning).

References

Halahakoon, D. C. et al. Reward-processing behavior in depressed participants relative to healthy volunteers: a systematic review and meta-analysis. JAMA Psychiatry 77, 1286–1295 (2020).
Article PubMed PubMed Central Google Scholar
Hyman, S. E., Malenka, R. C. & Nestler, E. J. Neural mechanisms of addiction: the role of reward-related learning and memory. Annu Rev. Neurosci. 29, 565–598 (2006).
Article CAS PubMed Google Scholar
Gold, J. M., Waltz, J. A., Prentice, K. J., Morris, S. E. & Heerey, E. A. Reward processing in schizophrenia: a deficit in the representation of value. Schizophr. Bull. 34, 835–847 (2008).
Article PubMed PubMed Central Google Scholar
Saunders, B. T., Richard, J. M., Margolis, E. B. & Janak, P. H. Dopamine neurons create Pavlovian conditioned stimuli with circuit-defined motivational properties. Nat. Neurosci. 21, 1072–1083 (2018).
Article CAS PubMed PubMed Central Google Scholar
Steinberg, E. E. et al. A causal link between prediction errors, dopamine neurons and learning. Nat. Neurosci. 16, 966–973 (2013).
Article CAS PubMed PubMed Central Google Scholar
de Jong, J. W. et al. A neural circuit mechanism for encoding aversive stimuli in the mesolimbic dopamine system. Neuron 101, 133–151 e7 (2019).
Article PubMed Google Scholar
de Jong, J. W., Liang, Y., Verharen, J. P. H., Fraser, K. M. & Lammel, S. State and rate-of-change encoding in parallel mesoaccumbal dopamine pathways. Nat. Neurosci. 27, 309–318 (2024).
Article PubMed PubMed Central Google Scholar
Schultz, W., Dayan, P. & Montague, P. R. A neural substrate of prediction and reward. Science 275, 1593–1599 (1997).
Article CAS PubMed Google Scholar
Omelchenko, N. & Sesack, S. R. Ultrastructural analysis of local collaterals of rat ventral tegmental area neurons: GABA phenotype and synapses onto dopamine and GABA cells. Synapse 63, 895–906 (2009).
Article CAS PubMed PubMed Central Google Scholar
Eshel, N. et al. Arithmetic and local circuitry underlying dopamine prediction errors. Nature 525, 243–246 (2015).
Article ADS CAS PubMed PubMed Central Google Scholar
Kim, Y.-B., Matthews, M. & Moghaddam, B. Putative γ-aminobutyric acid neurons in the ventral tegmental area have a similar pattern of plasticity as dopamine neurons during appetitive and aversive learning. Eur. J. Neurosci. 32, 1564–1572 (2010).
Article PubMed PubMed Central Google Scholar
Nugent, F. S., Penick, E. C. & Kauer, J. A. Opioids block long-term potentiation of inhibitory synapses. Nature 446, 1086–1090 (2007).
Article ADS CAS PubMed Google Scholar
Ostroumov, A. et al. Acute nicotine exposure alters ventral tegmental area inhibitory transmission and promotes diazepam consumption. eNeuro 7, 10.1523/ENEURO.0348-19.2020 (2020).
Thomas, A. M. et al. Adolescent nicotine exposure alters gabaa receptor signaling in the ventral tegmental area and increases adult ethanol self-administration. Cell Rep. 23, 68–77 (2018).
Article CAS PubMed PubMed Central Google Scholar
Ostroumov, A. et al. Stress Increases Ethanol Self-Administration via a Shift toward Excitatory GABA Signaling in the Ventral Tegmental Area. Neuron 92, 493–504 (2016).
Article CAS PubMed PubMed Central Google Scholar
Taylor, A. M. et al. Microglia disrupt mesolimbic reward circuitry in chronic pain. J. Neurosci. 35, 8442–8450 (2015).
Article CAS PubMed PubMed Central Google Scholar
Taylor, A. M. et al. Neuroimmune regulation of gabaergic neurons within the ventral tegmental area during withdrawal from chronic morphine. Neuropsychopharmacology 41, 949–959 (2016).
Article CAS PubMed Google Scholar
Lee, H. H., Deeb, T. Z., Walker, J. A., Davies, P. A. & Moss, S. J. NMDA receptor activity downregulates KCC2 resulting in depolarizing GABAA receptor-mediated currents. Nat. Neurosci. 14, 736–743 (2011).
Article CAS PubMed PubMed Central Google Scholar
Moeyaert, B. et al. Improved methods for marking active neuron populations. Nat. Commun. 9, 4440 (2018).
Article ADS PubMed PubMed Central Google Scholar
Ben-Ari, Y. NKCC1 chloride importer antagonists attenuate many neurological and psychiatric disorders. Trends Neurosci. 40, 536–554 (2017).
Article CAS PubMed Google Scholar
Rothermel, M., Brunert, D., Zabawa, C., Diaz-Quesada, M. & Wachowiak, M. Transgene expression in target-defined neuron populations mediated by retrograde infection with adeno-associated viral vectors. J. Neurosci. 33, 15195–15206 (2013).
Article CAS PubMed PubMed Central Google Scholar
Witten, I. B. et al. Recombinase-driver rat lines: tools, techniques, and optogenetic application to dopamine-mediated reinforcement. Neuron 72, 721–733 (2011).
Article CAS PubMed PubMed Central Google Scholar
Sharma, G. & Vijayaraghavan, S. Modulation of presynaptic store calcium induces release of glutamate and postsynaptic firing. Neuron 38, 929–939 (2003).
Article CAS PubMed Google Scholar
Sharma, G., Grybko, M. & Vijayaraghavan, S. Action potential-independent and nicotinic receptor-mediated concerted release of multiple quanta at hippocampal CA3-mossy fiber synapses. J. Neurosci. 28, 2563–2575 (2008).
Article CAS PubMed PubMed Central Google Scholar
Tan, K. R. et al. Neural bases for addictive properties of benzodiazepines. Nature 463, 769–774 (2010).
Article ADS CAS PubMed PubMed Central Google Scholar
Tan, K. R. et al. GABA neurons of the VTA drive conditioned place aversion. Neuron 73, 1173–1183 (2012).
Article CAS PubMed PubMed Central Google Scholar
van Zessen, R., Phillips, J. L., Budygin, E. A. & Stuber, G. D. Activation of VTA GABA neurons disrupts reward consumption. Neuron 73, 1184–1194 (2012).
Article PubMed PubMed Central Google Scholar
Chen, L. et al. KCC2 downregulation facilitates epileptic seizures. Sci. Rep. 7, 156 (2017).
Article ADS PubMed PubMed Central Google Scholar
Jung, J. Y., Lee, S. E., Hwang, E. M. & Lee, C. J. Neuronal expression and cell-type-specific gene-silencing of best1 in thalamic reticular nucleus neurons using psico-red system. Exp. Neurobiol. 25, 120–129 (2016).
Article PubMed PubMed Central Google Scholar
Morozova, E. O. et al. Contribution of synchronized GABAergic neurons to dopaminergic neuron firing and bursting. J. Neurophysiol. 116, 1900–1923 (2016).
Article CAS PubMed PubMed Central Google Scholar
Sun, F. et al. Next-generation GRAB sensors for monitoring dopaminergic activity in vivo. Nat. Methods 17, 1156–1166 (2020).
Article CAS PubMed PubMed Central Google Scholar
Deister, C. A., Teagarden, M. A., Wilson, C. J. & Paladini, C. A. An intrinsic neuronal oscillator underlies dopaminergic neuron bursting. J. Neurosci. 29, 15888–15897 (2009).
Article CAS PubMed PubMed Central Google Scholar
Pan, W. X. & Hyland, B. I. Pedunculopontine tegmental nucleus controls conditioned responses of midbrain dopamine neurons in behaving rats. J. Neurosci. 25, 4725–4732 (2005).
Article CAS PubMed PubMed Central Google Scholar
Floresco, S. B., West, A. R., Ash, B., Moore, H. & Grace, A. A. Afferent modulation of dopamine neuron firing differentially regulates tonic and phasic dopamine transmission. Nat. Neurosci. 6, 968–973 (2003).
Article CAS PubMed Google Scholar
Lokwan, S. J., Overton, P. G., Berry, M. S. & Clark, D. Stimulation of the pedunculopontine tegmental nucleus in the rat produces burst firing in A9 dopaminergic neurons. Neuroscience 92, 245–254 (1999).
Article CAS PubMed Google Scholar
Stuber, G. D. et al. Reward-predictive cues enhance excitatory synaptic strength onto midbrain dopamine neurons. Science 321, 1690–1692 (2008).
Article ADS CAS PubMed PubMed Central Google Scholar
Kimmey, B. A., Ostroumov, A. & Dani, J. A. 5-HT2A receptor activation normalizes stress-induced dysregulation of GABAergic signaling in the ventral tegmental area. Proc. Natl. Acad. Sci. USA https://doi.org/10.1073/pnas.1911446116 (2019).
Banke, T. G. & Gegelashvili, G. Tonic activation of group I mGluRs modulates inhibitory synaptic strength by regulating KCC2 activity. J. Physiol. 586, 4925–4934 (2008).
Article CAS PubMed PubMed Central Google Scholar
Zhang, J. et al. MicroRNA-mediated downregulation of potassium-chloride-cotransporter and vesicular γ-aminobutyric acid transporter expression in spinal cord contributes to neonatal cystitis-induced visceral pain in rats. Pain 158, 2461–2474 (2017).
Article CAS PubMed PubMed Central Google Scholar
Barbato, C. et al. Silencing of Ago-2 Interacting Protein SERBP1 Relieves KCC2 Repression by miR-92 in Neurons. Cells 11, 1052 (2022).
Article CAS PubMed PubMed Central Google Scholar
Kelsch, W. et al. Insulin-like growth factor 1 and a cytosolic tyrosine kinase activate chloride outward transport during maturation of hippocampal neurons. J. Neurosci. 21, 8339–8347 (2001).
Article CAS PubMed PubMed Central Google Scholar
Lee, H. H. C., Jurd, R. & Moss, S. J. Tyrosine phosphorylation regulates the membrane trafficking of the potassium chloride co-transporter KCC2. Mol. Cell Neurosci. 45, 173–179 (2010).
Article CAS PubMed PubMed Central Google Scholar
Rivera, C. et al. BDNF-induced TrkB activation down-regulates the K+-Cl- cotransporter KCC2 and impairs neuronal Cl- extrusion. J. Cell Biol. 159, 747–752 (2002).
Article CAS PubMed PubMed Central Google Scholar
Kahle, K. T. et al. WNK3 modulates transport of Cl- in and out of cells: implications for control of cell volume and neuronal excitability. Proc. Natl Acad. Sci. USA 102, 16783–16788 (2005).
Article ADS CAS PubMed PubMed Central Google Scholar
Brown, J., Pan, W.-X. & Dudman, J. T. The inhibitory microcircuit of the substantia nigra provides feedback gain control of the basal ganglia output. eLife 3, e02397 (2014).
Article PubMed PubMed Central Google Scholar
Higgs, M. H. & Wilson, C. J. Unitary synaptic connections among substantia nigra pars reticulata neurons. J. Neurophysiol. 115, 2814–2829 (2016).
Article PubMed PubMed Central Google Scholar
Kim, Y., Wood, J. & Moghaddam, B. Coordinated activity of ventral tegmental neurons adapts to appetitive and aversive learning. PLoS One 7, e29766 (2012).
Article ADS CAS PubMed PubMed Central Google Scholar
Rossi, M. A., Fan, D., Barter, J. W. & Yin, H. H. Bidirectional modulation of substantia nigra activity by motivational state. PLoS One 8, e71598 (2013).
Article ADS CAS PubMed PubMed Central Google Scholar
Di Cristo, G., Awad, P. N., Hamidi, S. & Avoli, M. KCC2, epileptiform synchronization, and epileptic disorders. Prog. Neurobiol. 162, 1–16 (2018).
Article PubMed Google Scholar
Rohr, K. E. et al. Seasonal plasticity in GABAA signaling is necessary for restoring phase synchrony in the master circadian clock network. Elife 8, e49578 (2019).
Article PubMed PubMed Central Google Scholar
Alfonsa, H. et al. Intracellular chloride regulation mediates local sleep pressure in the cortex. Nat. Neurosci. 26, 64–78 (2023).
Article CAS PubMed Google Scholar
Pan, W.-X., Brown, J. & Dudman, J. T. Neural signals of extinction in the inhibitory microcircuit of the ventral midbrain. Nat. Neurosci. 16, 71–78 (2013).
Article CAS PubMed Google Scholar
Tolu, S. et al. Co-activation of VTA DA and GABA neurons mediates nicotine reinforcement. Mol. Psychiatry 18, 382–393 (2013).
Article CAS PubMed Google Scholar
Maes, E. J. P. et al. Causal evidence supporting the proposal that dopamine transients function as temporal difference prediction errors. Nat. Neurosci. 23, 176–178 (2020).
Article CAS PubMed PubMed Central Google Scholar
Berridge, K. C. The debate over dopamine’s role in reward: the case for incentive salience. Psychopharmacol. (Berl.) 191, 391–431 (2007).
Article CAS Google Scholar
Robbins, T. W. & Everitt, B. J. A role for mesencephalic dopamine in activation: commentary on Berridge (2006). Psychopharmacol. (Berl.) 191, 433–437 (2007).
Article CAS Google Scholar
Nicola, S. M. The flexible approach hypothesis: unification of effort and cue-responding hypotheses for the role of nucleus accumbens dopamine in the activation of reward-seeking behavior. J. Neurosci. 30, 16585–16600 (2010).
Article CAS PubMed PubMed Central Google Scholar
Jeong, H. et al. Mesolimbic dopamine release conveys causal associations. Science 378, eabq6740 (2022).
Article CAS PubMed PubMed Central Google Scholar
Mohebi, A. et al. Dissociable dopamine dynamics for learning and motivation. Nature 570, 65–70 (2019).
Article ADS CAS PubMed PubMed Central Google Scholar
Simpson, E. H. et al. Lights, fiber, action! A primer on in vivo fiber photometry. Neuron 112, 718–739 (2024).
Article CAS PubMed Google Scholar
Lefner, M. J., Dejeux, M. I. & Wanat, M. J. Sex differences in behavioral responding and dopamine release during pavlovian learning. eNeuro 9, ENEURO.0050-22.2022 (2022).
Article PubMed PubMed Central Google Scholar
Taketani, M., Shii, M., Ohura, K., Ninomiya, S. & Imai, T. Carboxylesterase in the liver and small intestine of experimental animals and human. Life Sci. 81, 924–932 (2007).
Article CAS PubMed Google Scholar
Gagnon, M. et al. Chloride extrusion enhancers as novel therapeutics for neurological diseases. Nat. Med 19, 1524–1528 (2013).
Article CAS PubMed PubMed Central Google Scholar
Yamada, T. et al. Immunohistochemistry with an antibody to human liver carboxylesterase in human brain tissues. Brain Res 658, 163–167 (1994).
Article CAS PubMed Google Scholar
Yamada, T., Kawaguchi, N., Hosokawa, M. & Satoh, T. Localization of an isoform of carboxylesterase in rat brain differs from that in human brain. Brain Res 674, 175–179 (1995).
Article CAS PubMed Google Scholar
Jones, R. D., Taylor, A. M., Tong, E. Y. & Repa, J. J. Carboxylesterases are uniquely expressed among tissues and regulated by nuclear hormone receptors in the mouse. Drug Metab. Dispos. 41, 40–49 (2013).
Article CAS PubMed PubMed Central Google Scholar
Holmes, R. S., Cox, L. A. & Vandeberg, J. L. A new class of mammalian carboxylesterase CES6. Comp. Biochem Physiol. Part D. Genomics Proteom. 4, 209–217 (2009).
Article Google Scholar
Fernández, C. et al. Synthesis and biological studies of glycosyl dopamine derivatives as potential antiparkinsonian agents. Carbohydr. Res 327, 353–365 (2000).
Article PubMed Google Scholar
Fernández, C. et al. Synthesis of glycosyl derivatives as dopamine prodrugs: interaction with glucose carrier GLUT-1. Org. Biomol. Chem. 1, 767–771 (2003).
Article PubMed Google Scholar
Sivakumaran, S. et al. Selective inhibition of KCC2 leads to hyperexcitability and epileptiform discharges in hippocampal slices and in vivo. J. Neurosci. 35, 8291–8296 (2015).
Article CAS PubMed PubMed Central Google Scholar
Bruno, C. A. et al. pMAT: An open-source software suite for the analysis of fiber photometry data. Pharm. Biochem Behav. 201, 173093 (2021).
Article CAS Google Scholar
Vöröslakos, M., Petersen, P. C., Vöröslakos, B. & Buzsáki, G. Metal microdrive and head cap system for silicon probe recovery in freely moving rodent. Elife 10, e65859 (2021).
Article PubMed PubMed Central Google Scholar
Heymann, G. et al. Synergy of distinct dopamine projection populations in behavioral reinforcement. Neuron 105, 909–920 e5 (2020).
Article CAS PubMed Google Scholar
Steffensen, S. C., Svingos, A. L., Pickel, V. M. & Henriksen, S. J. Electrophysiological characterization of GABAergic neurons in the ventral tegmental area. J. Neurosci. 18, 8003–8015 (1998).
Article CAS PubMed PubMed Central Google Scholar
Lin, J. Y. A user’s guide to channelrhodopsin variants: features, limitations and future developments. Exp. Physiol. 96, 19–25 (2011).
Article PubMed Google Scholar
Grace, A. A. & Bunney, B. S. The control of firing pattern in nigral dopamine neurons: burst firing. J. Neurosci. 4, 2877–2890 (1984).
Article CAS PubMed PubMed Central Google Scholar
Chieng, B., Azriel, Y., Mohammadi, S. & Christie, M. J. Distinct cellular properties of identified dopaminergic and GABAergic neurons in the mouse ventral tegmental area. J. Physiol. 589, 3775–3787 (2011).
Article CAS PubMed PubMed Central Google Scholar
Morales, M. & Margolis, E. B. Ventral tegmental area: cellular heterogeneity, connectivity and behaviour. Nat. Rev. Neurosci. 18, 73–85 (2017).
Article CAS PubMed Google Scholar
Li, W., Doyon, W. M. & Dani, J. A. Quantitative unit classification of ventral tegmental area neurons in vivo. J. Neurophysiol. 107, 2808–2820 (2012).
Article PubMed PubMed Central Google Scholar
Gentle, J. E., Kaufman, L. & Rousseuw, P. J. Finding Groups in Data: An Introduction to Cluster Analysis. Biometrics 47, 788 (1991).
Article Google Scholar
Amarasingham, A., Harrison, M. T., Hatsopoulos, N. G. & Geman, S. Conditional modeling and the jitter method of spike resampling. J. Neurophysiol. 107, 517–531 (2012).
Article PubMed Google Scholar
Mathis, A. et al. DeepLabCut: markerless pose estimation of user-defined body parts with deep learning. Nat. Neurosci. 21, 1281–1289 (2018).
Article CAS PubMed Google Scholar

Download references

Acknowledgements

We thank Dr. Patrick Forcelli, Dr. Stefano Vicini, Dr. Daniil Berezhnoi, and Anna Pearson for helpful discussions and feedback. This work was supported by grants from the National Institutes of Health [MH125996 (AO), DA048134(AO), NS139517 (JW), DA061493 (JW)], Brain & Behavior Research Foundation [NARSAD 28113 (AO)], the Whitehall Foundation [2020-12-35 (AO)], and the Brain Research Foundation [BRFSG-2022-06 (AO)].

Author information

Authors and Affiliations

Department of Pharmacology and Physiology, Georgetown University, Washington, DC, USA
Joyce Woo, Ajay Uprety, Daniel J. Reid, Irene Chang, Aelon Ketema Samuel, Caroline C. Swain & Alexey Ostroumov
Interdisciplinary Program in Neuroscience, Georgetown University, Washington, DC, USA
Helena de Carvalho Schuch & Alexey Ostroumov

Authors

Joyce Woo
View author publications
Search author on:PubMed Google Scholar
Ajay Uprety
View author publications
Search author on:PubMed Google Scholar
Daniel J. Reid
View author publications
Search author on:PubMed Google Scholar
Irene Chang
View author publications
Search author on:PubMed Google Scholar
Aelon Ketema Samuel
View author publications
Search author on:PubMed Google Scholar
Helena de Carvalho Schuch
View author publications
Search author on:PubMed Google Scholar
Caroline C. Swain
View author publications
Search author on:PubMed Google Scholar
Alexey Ostroumov
View author publications
Search author on:PubMed Google Scholar

Contributions

J.W. and A.U. designed and performed ex vivo and in vivo electrophysiological experiments, assisted by I.C. and A.K.S. D.J.R. and C.C.S. performed behavioral experiments. J.W., A.U., and H.C.S. performed western blot and immunohistochemistry experiments. A.O. originated, planned, and oversaw the experiments with J.W.’s and A.U.’s assistance. Led by A.O., all the authors contributed to writing the manuscript.

Corresponding author

Correspondence to Alexey Ostroumov.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplementary Information (download PDF )

Reporting Summary (download PDF )

Transparent Peer Review file (download PDF )

Source data

Source Data (download XLSX )

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Cite this article

Woo, J., Uprety, A., Reid, D.J. et al. Dynamic changes in chloride homeostasis coordinate midbrain inhibitory network activity during reward learning. Nat Commun 16, 10903 (2025). https://doi.org/10.1038/s41467-025-66838-x

Download citation

Received: 01 November 2024
Accepted: 14 November 2025
Published: 09 December 2025
Version of record: 09 December 2025
DOI: https://doi.org/10.1038/s41467-025-66838-x

Subjects

Abstract

Similar content being viewed by others

Introduction

Results

Learning-dependent downregulation of KCC2 in VTA GABA neurons

Learning-mediated KCC2 downregulation impacts NAc lateral shell projecting DA neurons

VTA GABA neurons and KCC2 downregulation contribute to reward learning

VTA GABA neurons show enhanced synchronized activity during acquisition periods of learning

Changes in KCC2 function enhance DA signaling in the VTA and NAc

Optogenetic synchronization of VTA GABA neurons enhances stimulus-induced burst firing ex vivo and in vivo

Discussion

Methods

Subjects

Stereotaxic surgeries

Behavioral procedures

Drugs and experimental design

Ex vivo electrophysiology

Histology and microscopy

Western blots

In vivo fiber photometry and DA signal analysis

In vivo electrophysiology in awake rats and data analysis

Cell type classification

Spike synchrony

In Vivo Electrophysiology in Anesthetized Rats

DA neuron burst firing analysis

Recording site identification

DeepLabCut video analysis

Quantification and statistical analysis

Reporting summary

Data availability

Code availability

References

Acknowledgements

Author information

Authors and Affiliations

Contributions

Corresponding author

Ethics declarations

Competing interests

Peer review

Peer review information

Additional information

Supplementary information

Source data

Rights and permissions

About this article

Cite this article

Share this article

Search

Quick links