Dopaminergic inputs in the dentate gyrus direct the choice of memory encoding

Significance

Reward boosts forms of learning and memory through dopamine-mediated neuromodulation in the brain. However, the influence of dopamine has been an underappreciated component of episodic information in the hippocampus. Using a cross-disciplinary approach, we demonstrate that dopaminergic input in the dentate gyrus, a hippocampal subregion critical for the formation of high-resolution memories, impairs subsequent learning by suppressing cortical inputs and ensemble neuronal activity in this area. This work reveals a mechanism by which dopamine signal biases memory storage of events leading to rather than subsequent to the reward.

Abstract

Rewarding experiences are often well remembered, and such memory formation is known to be dependent on dopamine modulation of the neural substrates engaged in learning and memory; however, it is unknown how and where in the brain dopamine signals bias episodic memory toward preceding rather than subsequent events. Here we found that photostimulation of channelrhodopsin-2–expressing dopaminergic fibers in the dentate gyrus induced a long-term depression of cortical inputs, diminished theta oscillations, and impaired subsequent contextual learning. Computational modeling based on this dopamine modulation indicated an asymmetric association of events occurring before and after reward in memory tasks. In subsequent behavioral experiments, preexposure to a natural reward suppressed hippocampus-dependent memory formation, with an effective time window consistent with the duration of dopamine-induced changes of dentate activity. Overall, our results suggest a mechanism by which dopamine enables the hippocampus to encode memory with reduced interference from subsequent experience.

The brain structures crucial for memory formation are presumably under the control of the midbrain dopamine (DA) system, which selectively marks experiences that lead to reward (1, 2). In the striatum and the cortex, repetitive pairing of DA input after, but not before, sensorimotor stimulus within a narrow time window promotes structural and functional connectivity (3, 4), which may provide a cellular basis for reward to reinforce specifically an immediate past action. In the hippocampus, a structure that is instrumental in forming memories of contexts and objects making up the experiences (5, 6), DA must be present at the induction of long-term potentiation (LTP) to increase the magnitude of early- and late-phase LTP (7–10). When released during learning, DA also has been found to enhance the reactivation of newly formed neural ensembles (11). The requisite coincidence between the DA signal and the conditioning stimulation may serve to ensure that only inputs concurrent with or occurring shortly before reward are encoded in long-term memory. However, rewarding outcomes often may be delayed, and the involvement of the hippocampus is necessary when an event and its outcome are temporally discontinuous (12). This type of “memory” can be formed rapidly after even a single experience (5, 6), and behavioral studies demonstrate that application of DA agonists in the hippocampus hours after training promotes memory maintenance (13, 14), indicating that DA released from midbrain projections (Fig. S1) exerts distinct influences on the hippocampus to reinforce memory of earlier events selectively.

Fig. S1.

Photomicrographs of FG retrogradely traced neurons double-labeled by TH immunohistochemistry. (A, Left) FG signals at the injection site in the hippocampus. (Right) Image was taken from a section slightly ventral to the injection site. (Scale bar: 200 μm.) (B, Top Left) Sample image of FG-traced neurons. (Middle Left) The same horizontal section of TH⁺ neurons in the midbrain. (Bottom Left) Merged image showing colocalization of FG label with TH staining. (Right) Higher magnification of the boxed region in each left panel. Arrows show double-labeled cells. (Scale bar: 500 μm.)

Here we surveyed the dentate gyrus (DG) of the hippocampus to explore the possible sites and actions of DA. As the first stage of the intrahippocampal trisynaptic loop, the DG receives multiple processed sensory inputs from the entorhinal cortex (EC) and uses conjunctive encoding to integrate them for a memory representation (15). This region, together with area CA3, has been shown to be active during new memory formation in humans (16–18). The DG is thought to transform noisy cortical signals into sparse activation of distinct ensembles of granule cells (GCs) (19, 20), and this “pattern separation” function is indispensable for the discrimination and storage of similar experiences (15, 21–23). The DG also displays numerous types of plastic processes in the hippocampus that may underlie learning and memory, including life-long generation of new neurons (24, 25). The DG thus possesses a special place in hippocampal anatomy and function. Because in slice preparations exogenously applied DA has been demonstrated to down-regulate GC excitation by cortical stimuli (26), we hypothesize that the DA system may impair postreward learning, thereby allowing the preferential entry of information received before reward into the hippocampus.

Results

Electrophysiological and Behavioral Effects of Optogenetic DA Release in the DG.

To investigate how DA release regulates DG activity in vivo, we adopted two strategies to produce animals with restricted expression of the light-gated ion channel channelrhodopsin-2 (ChR2) in dopaminergic neurons. First, we used offspring of tyrosine hydroxylase (TH)-Cre mice crossed with Ai27 mice bearing a Cre-dependent ChR2-tdTomato fusion gene. Their WT littermates were used as control. Second, TH-Cre mice were transfected by injection of adeno-associated virus (AAV)-double-floxed inverted open reading frame (DIO)–ChR2–mCherry or control AAV-DIO–mCherry viral vectors into the ventral tegmental area (VTA). The specificity and efficacy of Cre-dependent expression of ChR2 in TH-Cre;Ai27 and AAV–ChR2 mice were determined separately by colocalization of TH and tdTomato or mCherry (Fig. 1 A and B). Using electrophysiological recordings from anesthetized animals, we confirmed that light stimulation delivered through an optic fiber in the VTA triggered a time-locked increase of phasic firing in most recorded neurons expressing ChR2 (Fig. 1 C and D), although a small portion (8 of 51 neurons) displayed a decrease in spiking rate. Because ChR2 was expressed not only in the soma but also in the axon (Fig. S2A), the optic fiber with electrodes affixed was moved to the DG to induce local DA release and to record GC activity (Fig. S2B). The electrophysiological readout revealed that field excitatory postsynaptic potentials (fEPSPs) evoked by electrical stimulation of the perforant path (PP) in TH-Cre;Ai27 mice were decreased significantly following exposure to light (5-ms pulses at 20 Hz for 5 min), with a kinetic profile similar to the long-term depression (LTD) caused by DA infusion into the DG of WT animals (Fig. 1 E and F). The same light stimulation induced an fEPSP reduction in the AAV–ChR2 group as well (Fig. 1G), but with a smaller amplitude and shorter duration than in the TH-Cre;Ai27 group (ANOVA: no significant group × time interaction, F_1,13 = 0.2815, P = 0.9933; no significant time effect, F_13,111 = 0.429, P = 0.9562; main group effect, F_1,111 = 11.65, P = 0.0009), possibly because of differences in the number of ChR2-labeled dopaminergic cells and/or ChR2 expression levels.

Fig. 1.

Optical stimulation of dopaminergic terminals in the DG induces LTD-like changes at PP–DG synapses. (A) Representative images showing coexpression of ChR2–mCherry and TH in the VTA of TH-Cre mice infected with AAV-DIO–ChR2–mCherry. (B) Quantification showing the overlap of ChR2⁺TH⁺ cells with TH⁺ or ChR2⁺ cells in the VTA (from seven TH-cre;Ai27 mice and seven AAV–ChR2 mice). Error bars indicate SEM. (C) Typical superimposed spike waveforms of one ChR2-expressing neuron in the VTA activated by blue light. (Scale bars: 20 μV and 200 μS.) (D) Single-unit recordings in the VTA. (Upper) The raster plot shows the spike times for five optical stimulation trials (473 nm, 5 ms, 20 Hz). (Lower) The peristimulus time histogram shows the averaged response across all repetitions (50-ms bins). (E) Summary data of the effects of light stimulation (5-ms pulses at 20 Hz for 5 min; black horizontal bar) in the DG of TH-cre;Ai27 mice and their control littermates (ANOVA: P < 0.0001; TH-cre;Ai27: n = 7, WT: n = 4). Data represent the fEPSP amplitudes normalized by the mean values before optical stimulation and averaged over 2-min bins for each experiment. Error bars indicate SEM. (F) Summary plot of changes in fEPSPs evoked by PP stimulation after DA or vehicle (Veh) infusion (black horizontal bar) into the DG of WT animals (ANOVA: P < 0.0001; DA: n = 11, vehicle: n = 8). Data represent the fEPSP amplitudes normalized by the mean values before DA or vehicle infusion and averaged over 2-min bins for each experiment. Error bars indicate SEM. (G) Summary data of the effects of light stimulation (5-ms pulses at 20 Hz for 5 min; black horizontal bar) in the DG of TH-cre mice infused with AAV–ChR2 or control viral vectors (ANOVA: P < 0.0001; AAV–ChR2: n = 5, AAV–mCherry: n = 6). Data are presented as in E. (H) Photostimulation (black horizontal bar) following infusion of DA receptor antagonists (red horizontal bar) into the DG does not induce significant changes in the fEPSP amplitude (n = 7). Data are presented as in E.

Fig. S2.

Stimulation and recording positions in the DG of anesthetized mice. (A) Confocal images showing the presence of ChR2-tdTomato–expressing terminals (arrowheads) in the DG. (Scale bars: 40 μm.) (B) Sample images showing the tracks of stimulating electrode (Left) and tetrode-optic fiber assembly in the DG (Right). Arrowheads indicate the electrode tips. (Scale bars: 400 μm.)

The TH-Cre line has been reported to drive apparent nondopaminergic expression patterns of Cre within VTA nuclei and a number of other brain structures, such as the locus coeruleus (LC) (27). It is thus possible that some ChR2-expressing fibers in the DG originated from noradrenergic nuclei in a TH-Cre;Ai27 mouse model. However, in line with most studies (28), norepinephrine (NE) potentiated GC responses elicited by PP stimulation, an effect opposite that of DA (Fig. S3A). We also quantified the colocalization between ChR2 and dopamine β-hydroxylase (DβH), a marker for noradrenergic neurons, and observed quite a low efficiency of Cre-dependent expression of ChR2 in the LC (Fig. S3 B and C). Taken together with similar LTD-like changes of fEPSPs following photostimulation in the TH-Cre;Ai27 and AAV–ChR2 groups (Fig. 1 E and G), these data suggested that activation of dopaminergic neurons in the VTA accounted for the synaptic change, although the involvement of NE could not be completely excluded. To validate the role of DA further, we preinfused antagonists specifically targeting DA receptors, a mixture of the D1-like receptor blocker SCH 23390 and the D2-like receptor blocker sulpiride, into the DG of TH-Cre;Ai27 mice and found that light-induced LTD of fEPSPs was fully abolished (Fig. 1H). This piece of evidence rules out the possible action of GABA, if DA and GABA were coreleased from dopaminergic neurons as previously documented (29, 30), and confirms that the effect of optogenetic manipulation was mediated by DA. It is also noted that the administration of DA antagonists itself elicited a small and steady, although not statistically significant (ANOVA: P = 0.1) increase in the fEPSP amplitude, indicating a possible impact of tonic DA release on basal synaptic transmission in the DG.

Fig. S3.

NE effect and ChR2 expression in noradrenergic neurons. (A) In comparison with time-matched saline controls, bath application of NE (black horizontal bar) induces a small potentiation of PP-evoked fEPSPs in the DG (ANOVA: P < 0.0001; NE group, n = 5; vehicle group, n = 4). Data represent the fEPSP slopes normalized against the average of baseline traces (dotted line) and binned over 2-min spans. Error bars indicate ± SEM. (B) Representative images showing coexpression of ChR2-tdTomato and DβH in the LC of TH-cre;Ai27 mice. (Scale bar: 20 μm.) (C) Quantification showing the overlap of ChR2⁺DβH⁺ cells with DβH⁺ or ChR2⁺ cells in the LC. Error bars indicate SEM.

Next, we performed bilateral DG photostimulation with unilateral DG recording in freely moving animals (Fig. S4). When the mice were allowed to behave freely in the home-cage environment, theta frequency (4–12 Hz) oscillation of local field potentials (LFPs) could be observed frequently. Control WT littermates and AAV–mCherry mice did not show any significant light-induced LFP alterations at DG sites throughout the course of the experiment (Fig. 2 A and B). In contrast, LFP oscillatory powers in TH-Cre;Ai27 and AAV–ChR2 mice appeared to be reduced at most frequencies following the light-stimulation epoch (Fig. 2 C and D). Spectral analysis of the last 10-min recording period (∼20–30 min after delivery of light pulses) relative to the 10-min control period before stimulation showed that, in comparison with the controls, the LFP power change in the theta frequency range was significant in AAV–ChR2 mice. Similarly, although it lacked statistical significance, a trend in LFP power reduction was observed in the TH-cre;Ai27 group (Fig. 2E). Theta power and frequency may vary with the locomotion speed. To corroborate that the observed decrease in theta power was a result of optical stimulation rather than behavioral variability, we analyzed the dataset of anesthetized animals included in Fig. 1. As shown in Fig. S5, light exposure resulted in a significant drop of theta power in AAV–ChR2 mice as compared with AAV–mCherry controls. TH-Cre;Ai27 mice also showed a trend toward theta power decrease compared with their WT littermates, and this reduction of theta power could be fully prevented by DA receptor blockers. Thus, optical stimulation appeared to have similar effects on theta oscillations in anesthetized and freely behaving animals.

Fig. S4.

Sites for optical stimulation and recording in freely moving mice. (A) Schematic drawing of electrode and optic fiber placement in the DG. (B) Sample images showing the tracks of an implanted microdrive consisting of an optetrode (Left) and an optic fiber (Right) in the DG. Arrowheads indicate the tips of the optetrode and optic fiber. (Scale bars: 200 μm.)

Fig. 2.

Prelearning optical stimulation causes changes in DG network oscillations and impairs memory formation. (A) Averaged power spectrogram of LFPs in WT littermates exposed to light stimulation (n = 9). The white curve indicates the total power of the theta band LFP signal (4–12 Hz). Dashed lines mark the time of blue light delivery (5-ms pulses at 20 Hz for 5 min). The horizontal white bar marks the time periods used for quantitative comparison of LFP powers between 4 and 12 Hz. (B) Averaged power spectrogram of LFPs in TH-Cre mice transduced with AAV-DIO–mCherry exposed to light stimulation (n = 11). Data are presented as in A. (C) Averaged power spectrogram of LFPs in TH-cre;Ai27 mice exposed to light stimulation (n = 8). Data are presented as in A. (D) Averaged power spectrogram of LFPs in TH-Cre mice transduced with AAV-DIO–ChR2–mCherry exposed to light stimulation (n = 11). Data are presented as in A. (E) Histogram showing LFP power (4–12 Hz) change measured during the interval indicated in A relative to a 10-min baseline segment before stimulation (t test: TH-cre;Ai27 vs. WT, P = 0.12; AAV–ChR2 vs. AAV–mCherry, P = 0.02). *P < 0.05. (F) Control but not TH-cre;Ai27 mice undergoing photostimulation before context preexposure freeze significantly longer in context A than in context B (ANOVA: group × context interaction, F_1,17 = 7.069, P = 0.0165; post hoc for WT vs. TH-cre;Ai27, context A, P > 0.05; context B, P > 0.05; post hoc for context A vs. context B: WT, P < 0.001, n = 10; TH-cre;Ai27, P > 0.05, n = 9). Mice in the AAV–mCherry group and AAV–ChR2 mice freeze similarly in context A and B (ANOVA: group × context interaction, F_1,21 = 6.160, P = 0.0216; post hoc for AAV–mCherry vs. AAV–ChR2, context A, P > 0.05; context B, P > 0.05; post hoc for context A vs. context B: AAV–mCherry, P < 0.001, n = 12; AAV–ChR2, P > 0.05, n = 11). ***P < 0.001. (G) The discrimination index is significantly lower in the TH-cre;Ai27 and AAV–ChR2 groups than in their respective control groups (t test: WT vs. TH-cre;Ai27, t₁₇ = 2.609, P = 0.018; AAV–mCherry vs. AAV–ChR2, t₂₁ = 2.482, P = 0.022). *P < 0.05.

Fig. S5.

Optical stimulation induces a reduction of theta power in anesthetized animals. (A) Averaged power spectrogram of LFPs in TH-cre;Ai27 mice exposed to light stimulation. Data are presented as in Fig. 2A. The horizontal white bar marks the time period used for quantitative comparison of LFP powers between 4 and 12 Hz. (B) Pretreatment with DA antagonists prevents light-induced changes in total power of LFPs in TH-cre;Ai27 mice. (C) Averaged power spectrogram of LFPs in WT littermates exposed to light stimulation. (D) Averaged power spectrogram of LFPs in TH-Cre mice transduced with AAV-DIO–ChR2–mCherry exposed to light stimulation. (E) Averaged power spectrogram of LFPs in TH-Cre mice transduced with AAV-DIO–mCherry exposed to light stimulation. (F) Histogram showing change in LFP power (4–12 Hz) measured during the interval indicated in A relative to the 10-min baseline segment before stimulation (t test: TH-cre;Ai27 vs. WT, P = 0.077; DA antagonist vs. WT, P = 0.908; AAV–ChR2 vs. AAV–mCherry, P = 0.016). *P < 0.05.

To evaluate whether the impact of DA was behaviorally relevant, we examined the performance of mice subject to optogenetic activation in a contextual fear-conditioning paradigm that combined context preexposure with an immediate foot shock (31). Taking the DG’s role in behavioral pattern separation into account, we not only measured the freezing behavior in the conditioned context (context A) but also assessed context discrimination by examining postconditioning fear behaviors of mice in another context (context B) that was similar but not identical to context A (32). The TH-Cre;Ai27 mice that received photostimulation ∼0.5 h before context preexposure exhibited a tendency to freeze less in context A than WT animals, whereas AAV–ChR2 mice subjected to optical stimulation appeared to show a slight but not statistically significant increase in the fear in context B as compared with AAV–mCherry mice (Fig. 2F). Despite this difference, both TH-Cre;Ai27 and AAV–ChR2 mice failed to distinguish context A from B, whereas the controls that underwent the same treatment paradigms froze significantly more in context A than in context B (Fig. 2F). To quantify the difference between fear responses to the two environments, we computed a discrimination index, defined as freezing in context A minus freezing in context B. As shown in Fig. 2G, the discrimination index was significantly lower in the TH-Cre;Ai27 group than in the WT group, and AAV–ChR2 mice had a lower discrimination index than AAV–mCherry controls. Because photostimulation did not alter exploratory behavior during context preexposure (Fig. S6), the prelearning DA signal appeared to impair new memory formation, which was most obviously reflected by the compromised context-discrimination ability.

Fig. S6.

Photostimulation before the context preexposure does not change the exploratory behavior. (A) The average motion of TH-cre;Ai27 mice is not significantly different from that of their WT littermates (ANOVA: no significant light × time interaction, F_1,9 = 0.7993, P = 0.6174; no significant light effect, F_1,170 = 0.6478, P = 0.422; main time effect, F_9,170 = 5.413, P < 0.0001; TH-cre;Ai27 group, n = 9; control group, n = 10). (B) The average motion of AAV–ChR2 mice is not significantly different from that of AAV–mCherry mice (ANOVA: no significant light × time interaction, F_1,9 = 0.4, P = 0.934; no significant light effect, F_1,200 = 0.0032, P = 0.9552; main time effect, F_9,200 = 12.8, P < 0.0001; AAV–ChR2 group, n = 12; AAV–mCherry group, n = 11).

To test the modulatory effect of DA directly, we infused DA into the DG before contextual preexposure. Consistent with the results of the optogenetic experiment, the DA-infused animals froze considerably less in context A than did the saline-infused controls. Furthermore, unlike the control mice that could discriminate context A from context B, DA-infused mice did not display significantly more freezing in context A than in context B, and their discrimination index was dramatically lower (Fig. 3A). However, their motility or explorative drive during context preexposure was not altered (Fig. 3B). Thus, hippocampal DA infusion and optical stimulation of ChR2-expressing dopaminergic axons in the DG exerted similar effects on contextual learning, demonstrating that the memory of events occurring subsequent to the dopamine signal was stored less efficiently.

Fig. 3.

Pretraining elevation of the hippocampal DA level is sufficient to cause a learning deficit. (A, Upper) Experimental scheme. On the first day, mice were infused with DA and ∼1 h later were allowed to explore context A. The next day, mice received an immediate foot shock in context A′. The fear behavior of mice then was tested in context A and B in a counterbalanced order. (Lower) Infusion of DA into the hippocampus before training causes a deficit in learning. (Left) The mice infused with control solution but not with DA freeze more in context A than in context B (ANOVA: group × context interaction, F_1,21 = 5.134, P = 0.035; post hoc for control vs. DA, context A, P < 0.05; context B, P > 0.05; post hoc for context A vs. context B: control group, P < 0.001, n = 12; DA group, P > 0.05, n = 11). (Right) The discrimination index is significantly lower in the DA group than in the control group (t test, t₂₁ = 2.266, P = 0.035). Data are presented as mean ± SEM. *P < 0.05. (B) DA infusion in the hippocampus does not change exploration during preexposure. The average motion of the mice that were infused with DA 1 h before preexposure is not significantly different from that of the control mice (ANOVA: no significant DA × time interaction, F_1,9 = 0.5464, P = 0.83; no significant DA effect, F_1,189 = 1.390, P = 0.25; main time effect, F_9,189 = 4.461, P < 0.0001; DA group, n = 11; vehicle group, n = 12).

DA Can Impact Temporal Associations in Dentate Function.

Although the observation that optogenetically induced DA release induces a strong LTD effect is intriguing, it is unclear what the functional impact of this depression would be in natural behaviors because the role of DA in DG function has not been well explored. It is well accepted that the impact of DA in other regions can be considered as temporal in nature, providing capabilities in reinforcement learning and consolidation. Likewise, we have previously hypothesized that the DG’s population activity would exhibit strong temporal structure because of its mixed population of young and old GCs (33, 34), a prediction that has been observed in a series of behavioral and physiology studies (35, 36). We accordingly predicted that the suppression of DG activity after DA may interact nontrivially with these temporal associations produced by the DG.

To explore whether DA-induced LTD could impact temporal coding in the DG after a reward, we implemented DA modulation in a computational model consisting of a highly abstracted, two-layer feed-forward network in which the output layer gradually added neurons (Fig. 4A). At a set time, DA was modeled by reducing the synaptic strengths of excitatory inputs as demonstrated in Fig. 1. Not surprisingly, DA-induced LTD sharply suppressed the activity of the outputs (Fig. 4 B and C), and this suppression had a marked effect on the correlations between the model’s outputs. Without DA, events presented close to one another had increased correlations, an effect we described previously as “pattern integration” (34). These associations were typically symmetrical in time; a given event was as likely to be associated with a preceding event as with a following event (Fig. 4D). However, the presence of DA disrupted this symmetry: If DA was released after an event, that event’s associations were biased toward the events that preceded it (Fig. 4 E and F). Similarly, events that occurred after DA release were preferentially associated with other postreward events (Fig. 4G). The results from this simulation thus suggest a role for DA in the DG similar to temporal difference learning (TDL) at the long time scales relevant for declarative memory encoding and that experience-dependent DA release could significantly impact hippocampal-dependent learning in a temporally asymmetric manner.

Fig. 4.

Implementation of DA in a simple neurogenic two-layer neural network. (A) Schematic of model. The neural network consists of an input layer of excitatory and inhibitory neurons and a neurogenic second layer that is initialized with no neurons. At each time step, the network is trained with a new event, and two immature neurons are added to the network. At time steps 65 through 70 (shaded green), DA is added to the network. Each plot below represents an average of 5,000 model runs. (B) DA (presented during shaded events) suppresses model output considerably. Even a relatively weak effect on excitatory synapses can greatly suppress network activity. (C) Strong DA suppression limits the population of responding neurons to only a subset of the youngest neurons in the network. Inactivity above the diagonal is indicative of neurons not yet having been born. (D) Cross-similarity plot of network encoding of events without DA; events are associated symmetrically in time because of neurogenesis. (E) Cross-similarity plot of network encoding of events with strong DA during events 65–70. Pre-DA and post-DA events are encoded using essentially distinct populations of neurons. (F) Events preceding DA are temporally associated with one another but show greatly reduced associations with events during (shaded area) and after DA. (G) Events following DA (shaded area) are temporally associated with other post-DA events. In F and G, note the symmetry of associations in the non-DA condition (blue).

We then examined whether a positive deflection of PP strength, which is a potential effect of NE signaling (Fig. S3), would produce a similar asymmetry in this simple model (Fig. S7 A–D). Because our model was highly abstract and thus lacks hilar inhibitory feedback that is known to normalize substantial positive deflections of DG activity, we considered only moderate increases in PP strength as a model of NE. Although an increase in PP strength indeed impacted the associations in our modeled DG, the effect was predominately isolated to the temporal window of NE potentiation, with minimal impact on associations between pre- and postperturbed activity (Fig. S7 E and F).

Fig. S7.

Implementation of NE in a simple neurogenic two-layer neural network. (A) Weak NE-induced LTP (+5% strength) broadly activates model DG output. (B) Events with 5% NE-induced LTP are weakly correlated with a broad range of temporally distant events. (C) Moderate NE-induced LTP (+12.5% strength) greatly activates model DG output. (D) Events with 12.5% NE-induced LTP are strongly correlated with a broad range of temporally distant events. Pre-NE and post-NE events are not differentially encoded using essentially distinct populations of neurons. (E) NE-induced LTP leads to strong correlation of events preceding NE with events during NE. (F) Events following NE (shaded area) are more strongly temporally associated with pre-NE events than events encoded without NE. In E and F, note the symmetry of associations in the non-NE condition (blue).

Impairment of Spatial Memory by Prelearning Reward.

To investigate if optogenetic control mimicked the presence of a natural reward, we gave WT mice sweetened condensed milk as reward, instead of exposing them to light pulses, before context preexposure during the fear-conditioning procedure. In contrast to the controls that were given water, the mice that received the milk reward ∼1 h before preexposure to the conditioning context showed remarkably less fear response in context A and were compromised in discriminating context A from context B, as indicated by a significant drop in the discrimination index (Fig. 5A). This finding is consistent with the result of the photostimulation experiment that indicated a DA-induced decrement in learning (Fig. 2) and with the observation of impaired correlations in the computational model (Fig. 4). However, a milk reward provided to a different cohort of mice 6 h before the contextual preexposure had no effect on contextual memory (Fig. 5B), suggesting that the behavioral effect of the reward was transient. Moreover, the impaired memory formation was not detected in mice given the reward at other time points during the behavioral procedure, including immediately after context preexposure, immediately before shock, and before memory test (Fig. S8), suggesting that the reward signal influenced information acquisition during preexposure rather than altering shock perception or the later association of context information with shock. Importantly, this experiment also ruled out the possibility that the reward led to a positive association with context A on the day of preexposure and thus to less freezing behavior when a shock was presented in the same environment the next day. Finally, we did not detect any significant effect of milk reward on the motility and exploration of mice in either the open field test or contextual preexposure (Figs. S9 and S10), indicating that the behavioral phenotype was not caused by an indirect effect of reward on motor activity and/or explorative drive. Hence, the prelearning DA increase resulted in a transient deficit in memory formation.

Fig. 5.

A natural reward transiently impairs spatial memory formation. (A) Appetitive reward treatment 1 h before training causes compromised learning. Control mice display significantly more freezing behavior in context A than do the milk-treated mice (ANOVA: group × context interaction, F_1,39 = 6.966, P = 0.012; post hoc: context A, P < 0.05, context B, P > 0.05; water group, n = 20; milk group, n = 21). Milk-treated mice have a significant lower discrimination index than the control mice (t test, t₃₉ = 2.069, P = 0.045). (B) Reward treatment 6 h before training has no effect on learning. Both groups of mice show significantly more freezing behavior in context A than in context B (ANOVA: group × context interaction, F_1,40 = 2.641, P = 0.11; n = 21 in each group). There is no significant difference in the discrimination index between milk-treated mice and the control mice (t test, t₄₀ = 1.625, P = 0.11). (C) Experimental scheme of Barnes maze experiment. (D) Milk-treated mice are impaired in reversal learning compared with control mice. (Upper Left) Errors made before reaching the target location (two-way ANOVA: group effect, F_1,45 = 6.077, P = 0.0176; training effect, F_2,45 = 3.159, P = 0.052, training × group interaction, F_1,45 = 0.8962, P = 0.41). (Lower Left) Average number of errors during reversal training (Mann–Whitney test, P = 0.0301). (Upper Center) Latency to reach the target location (two-way ANOVA: group effect, F_1,45 = 5.672, P = 0.0215; training effect, F_2,45 = 3.409, P = 0.0418, training × group interaction, F_1,45 = 0.5787, P = 0.56). (Lower Center) Average latency during reversal training (Mann–Whitney test, P = 0.0208). (Upper Right) Path length (two-way ANOVA: group effect, F_1,45 = 3.734, P = 0.0596; training effect, F_2,45 = 3.075, P = 0.056, training × group interaction, F_1,45 = 1.03, P = 0.36). (Lower Right) Average path length during reversal training (Mann-Whitney test, P = 0.036). (E) Milk reward has no effect on navigation speed (two-way ANOVA: group effect, F_1,45 = 0.2218, P = 0.64; training effect, F_2,45 = 1.379, P = 0.26, training × group interaction, F_1,45 = 0.6077, P = 0.54). Data are presented as mean ± SEM. In A, B, and D, *P < 0.05.

Fig. S8.

Reward has no effect on contextual conditioning if the reward is received immediately after preexposure, before immediate foot shock, or before memory test. (A) A posttraining reward does not affect contextual fear conditioning. In the fear test mice treated with sweetened condensed milk perform similarly to control mice (ANOVA: no significant treatment × context interaction, F_1,1 = 0.002353, P = 0.96; no significant treatment effect, F_1,44 = 0.1221, P = 0.72; main context effect, F_1,44 = 23.85, P < 0.0001; Bonferroni post hoc test, water group, P < 0.01, n = 24; milk group, P < 0.01, n = 22). (B) Sweetened milk reward received before shock or before testing does not alter fear behavior of mice (ANOVA: no significant treatment × context interaction, F_2,1 = 0.03321, P = 0.96; no significant reward effect, F_2,46 = 0.3762, P = 0.68; main context effect, F_1,46 = 21.34, P < 0.0001; n = 17 for the water group, n = 15 for the milk shock group, n = 17 for the milk test group). *P < 0.05.

Fig. S9.

Pretest reward does not change the exploratory or anxiety-related behaviors in either the open field test or the light–dark choice test. (A) The open field test. The exploratory behavior, as indicated by ambulatory path length (unpaired t test, t₂₂ = 1.692, P = 0.10, n = 12 mice in each group) and vertical counts (unpaired t test, t₂₂ = 0.3711, P = 0.71), is not affected by reward received about 1 h before the test. Receiving the reward does not affect the anxiety-like behavior, as indicated by the percentage of ambulatory length in the center (unpaired t test, t₂₂ = 0.0, P = 1.000) and the percentage of time spent in the center (unpaired t test, t₂₂ = 0.05182, P = 0.95). (B) The light–dark test. The anxiety-like behavior, as indicated by the percentage of ambulatory length in the light compartment (unpaired t test, t₂₂ = 0.1361, P = 0.89, n = 12 in each group), the time spent in the light compartment (unpaired t test, t₂₂ = 0.3029, P = 0.76), and the number of entries to the light compartment (unpaired t test, t₂₂ = 0.5222, P = 0.60), is not affected by a reward received about 1 h before the test. The exploratory behavior is not affected by reward, as indicated by similar total ambulatory path lengths (unpaired t test, t₂₂ = 0.5389, P = 0.59).

Fig. S10.

Prelearning reward does not change the exploratory behavior of mice during the context preexposure. (A) Changes in satiety affect the exploratory behavior during preexposure. The average motion of starved mice is significantly different from that of satiated ones (ANOVA: significant treatment × time interaction, F_1,9 = 2.743, P = 0.0071; no significant treatment effect, F_1,90 = 0.3803, P = 0.55; main time effect, F_9,90 = 3.972, P = 0.0003; starved group, n = 5; satiated group, n = 7). *P < 0.05. (B) Reward received ∼1 h before learning does not affect exploratory behavior during preexposure. The average motion of milk-treated mice is not significantly different from that of water-treated ones (ANOVA: no significant treatment × time interaction, F_1,9 = 1.605, P = 0.11; no significant treatment effect, F_1,207 = 0.2018, P = 0.65; main time effect, F_9,207 = 1.993, P = 0.042; milk group, n = 12; water group, n = 13).

We next examined how a reward received shortly before training would affect learning a new spatial location in a Barnes maze. We first trained mice for one target location in the maze and then divided the animals into two groups, with one group receiving a sweetened milk reward and the other group getting water as control. About 0.5–1 h after the reward, both groups were trained to learn a new target position that was located 90° from the original position (Fig. 5C). Mice in the reward group made significantly more errors than the control group, took significantly more time, and traveled a significantly longer distance to find the new target location (Fig. 5D). In contrast, no significant difference was detected in the navigational speed between the reward and control groups (Fig. 5E), suggesting that the reward had no effect on the general motility of the animals. These data corroborated the finding that a reward received shortly before learning had a detrimental effect on memory.

Natural Reward Occludes the in Vitro Effect of DA.

To verify the correlation between a prelearning reward-induced impairment in memory tasks and the electrophysiological effect of DA on the DG, we examined the action of DA perfusion in brain slices prepared from mice that were fed sweetened condensed milk immediately before they were killed (Fig. 6A), based on the observation that DA-induced LTD at PP–GC synapses was saturated by a single episode of drug administration (Fig. 6B). In slices obtained from milk-treated animals when the recordings were made 2–4 h but not >4 h after milk delivery, an ensuing DA perfusion failed to generate any further depression in GCs, including in GCs born during adulthood (Fig. 6 C–F). Consistent with our prior finding that activation of D2- but not D1-like receptors mediated DA-induced LTD upon neuronal maturation (26), preinjection of sulpiride completely overcame the occlusion effect of reward on mature GCs, whereas milk reward preceded by injection of SCH 23390 still significantly diminished subsequent LTD triggered by exogenous DA (Fig. 6G). Therefore, the activation of the natural DA system triggered PP–GC synaptic changes that closely resembled LTD induced by in vitro DA application, with a duration largely fitting with behavioral assessments. It is very likely that DA-induced LTD in new and old neurons was dependent on signaling mediated by D1- and D2-like receptors, respectively (26).

Fig. 6.

Appetitive reward occludes ensuing ex vivo DA-induced LTD within a certain time window. (A) A slice recording is made to examine the effect of exogenous DA on synaptic transmission from the PP to the DG immediately after the mouse receives a reward of sweetened condensed milk. (B) Summary of results obtained from GCs treated with two identical DA perfusions with a 50-min interval (black horizontal bars). LTD of PP-driven excitatory postsynaptic currents (EPSCs) induced by the first drug administration is not increased further by the second drug administration. Data represent the EPSC amplitudes normalized by the mean values before drug perfusion (dashed line) and averaged over 2-min bins for each experiment. Error bars indicate ± SEM. (C) DA fails to induce LTD in vitro within a critical time window after a milk reward. Summary data of the effects of DA perfusion on GCs in the outer GC layer in slices from control (gray) and milk-treated (black) mice. All recordings were made between 2 and 4 h after delivery of milk reward. Data are presented as in B. (D) Exogenous DA induces LTD again >4 h after milk reward. As in C except that all recordings were made >4 h after milk delivery. (E) Summary results of the effects of DA perfusion on GFP-labeled adult-generated GCs at 4–6 wpi in slices from control (gray in green) and milk-treated (green) mice. All recordings were made between 2 and 4 h after delivery of milk reward. Data are presented as in B. (F) As in E except that all recordings were made >4 h after milk delivery. (G) Comparison of DA actions on PP-elicited responses of mature GCs in brain slices from mice treated with milk (black) and animals i.p. preinjected with either sulpiride (open box) or SCH 23390 (gray) before milk reward. Data are presented as in B.

DA Regulation in Distinct Hippocampal Subfields.

To confirm further the involvement of the DG in the prereward-induced learning deficit, we analyzed DA regulation in the other hippocampal subareas. We first made extracellular field recordings from Schaffer collaterals (SC) and temporoammonic (TA) pathways to field CA1 (Fig. 7A). Consistent with previous reports (37, 38), DA suppressed fEPSPs evoked by TA stimulation for a short period but did not significantly alter the transmission at SC–CA1 synapses (Fig. 7B). Similarly, in the CA3 region, DA administration led to a transient reduction in fEPSPs evoked by PP projections, but no significant changes were detected in recurrent excitation via the extensive network of associational/commissural (A/C) connections among CA3 pyramidal cells (Fig. 7C). In contrast, field responses at PP–DG synapses were decreased to a level similar to that demonstrated in individual GCs by whole-cell patch-clamp recordings (Fig. 7D). As summarized in Fig. 7E, only the inhibition occurring in the DG was long lasting, although the levels of depression in distinct hippocampal subregions were not significantly different. To determine whether DA release produced a short-term depression (STD) in CA1 in vivo as in vitro, we recorded TA-evoked fEPSPs from brain slices of mice prerewarded with milk as in Fig. 6A. As predicted, DA applied to the bathing solution 2–4 h after milk treatment resulted in an STD almost identical to that in slices from control animals (Fig. 7F), indicating that food reward did not occlude ex vivo LTD in CA1 as it did in the DG. Taken together, our data indicate that the long duration of the DA effect appeared to be a unique feature of the DG and that, in this regard, DA modulation of DG activity was more likely to underlie the behavioral changes shown in Figs. 3 and 5. Because DA receptors are more abundant and diverse in the DG than in the other hippocampal subareas (26), we speculate that the observed differential effects of DA in distinct hippocampal subfields may be caused by different levels and/or combinations of DA receptors.

Fig. 7.

Region-specific dopaminergic regulation in the hippocampus. (A, Left) Location of stimulating and recording electrodes for extracellular field recordings from the SC (red) and TA (blue) pathways in area CA1. A cut was placed between the DG and the CA3 (dotted line) to avoid polysynaptic activation. (Right) Average traces of 30 consecutive sweeps of fEPSPs in response to SC or TA stimulation recorded before and within 10 min after bath application of DA. (B) In comparison with time-matched saline controls (Veh), DA administration (black horizontal bar) induces a significant STD in fEPSPs at TA–CA1 synapses but does not affect the strength of SC–CA1 synapses. Data represent the fEPSP amplitudes normalized against the average of baseline traces (dashed line) and binned over 2-min spans. Error bars indicate ± SEM. (C) DA attenuates PP- but not A/C-evoked fEPSPs in area CA3. Data are presented as in B. (D) DA causes an LTD of PP-elicited fEPSPs in the DG. (E) Comparison of DA effects on responses to cortical inputs in distinct hippocampal subfields. (F) No difference in DA-induced depression of TA–CA1 synaptic responses is observed in slices from control and milk-treated mice.

Discussion

To date, most studies characterizing the role of DA in hippocampal learning have focused on the CA1 region. For instance, when present at the time of induction, DA has been demonstrated to enhance LTP, especially late LTP at SC–CA1 synapses (10), indicating a mechanism by which information arriving coincident with or right before reward could be recorded into persistent memory by the hippocampus. However, several important issues remain unknown. First, how can reward facilitate memory of events occurring hours beforehand? Second, are memories formed more readily or simply prolonged in the presence of reward? Third, are memories influenced by DA’s actions on hippocampal subregions other than the CA1? In this study, we reveal an effect of DA on DG activity within a long time window after reward. By globally reducing excitation of GCs (Fig. 1), DA may drive the DG to filter out weak stimuli and to incorporate fewer inputs during “conjunctive encoding” (15), thereby probably leading to memory containing less information or possessing low “resolution” (33). Furthermore, we found that the prestimulus theta rhythm was weakened after DA release (Fig. 2 and Fig. S5). Although the function of theta oscillation is not yet clearly understood, prior studies have suggested its importance in mnemonic information encoding, because it reflects the “on-line” state or a state of readiness to process incoming signals in the hippocampus (39). It has been demonstrated that the amplitude of hippocampal oscillations before stimulus onset, especially in the theta range, is positively correlated with later episodic memory formation (40, 41). Here we suggest that the reduction of theta power may reflect a disruption of hippocampal ensemble activity and therefore altered neuronal encoding. Indeed, our behavioral tests demonstrate that prelearning optical activation of VTA impairs memory of a specific context in which animals receive one-trial contextual fear conditioning (Fig. 2). We noted that TH-Cre;Ai27 mice exhibit a tendency to freeze less in context A, whereas AAV–ChR2 mice display a tendency to freeze more in context B (Fig. 2F). A possible explanation for the difference between these two groups is that photostimulation seems to generate a weaker impact on both fEPSP and theta oscillation in AAV–ChR2 mice than in TH-Cre;Ai27 mice (Figs. 1 and 2 and Fig. S5). As a consequence, AAV–ChR2 mice can form the memory of context A but with some details missing, thereby failing to differentiate context B from A. In contrast, TH-Cre;Ai27 mice tend to ignore more information and therefore fail to remember context A at all. These results suggest that the prelearning DA signal primarily impairs the ability to discriminate between the training context and a similar environment. We further found that natural reward has an effect similar to that of optogenetic manipulation (Fig. 5) and that the reward can affect animal behavior within a long time window; this result in accordance with the key feature of long-lasting variation of neural activity associated with DA modulation in the DG (Figs. 6 and 7). Thus, our findings indicate an important role for the DG in the mediation of brain responsiveness to rewarding stimuli. It is possible that incoming information can be retained by default, unless prevented by a preceding DA signal. In this way, DA reduces interference from inputs arriving after a reward and increases the contrast between events taking place before and after reward, thereby virtually favoring the entrance of information temporally before reward into the hippocampus, be it a long time or shortly before reward. Complementing previous studies revealing that the midbrain DA system serves to stabilize the memory of events coincident with reward by promoting CA1 late LTP and network dynamics (10, 11), the present study provides a mechanism by which information acquired before reward is encoded into memory more readily than information acquired after reward, particularly when the input and DA signal are temporally discontinuous.

The DG has been identified as a major neurogenic region in the adult brain, and increasing evidence suggests that newborn neurons contribute to hippocampus-related learning and memory (33). Although the specific function of newly generated GCs in these processes remains elusive, they might work through two mechanisms that are not mutually exclusive. First, the plastic properties of immature GCs could allow them to participate in information storage per se, given prior findings that adult-generated GCs form a critical and enduring component of hippocampal memory traces (42, 43). Second, newborn GCs could introduce a degree of similarity to memories learned at the same time, a process referred to as “pattern integration” (34). They could be particularly important in combining inputs from various sources to form a complete representation of a context or an event. Thus, DA-induced suppression of neural transmission to these neurons could hinder the “conjunctive encoding” of the context, resulting in contextual discrimination deficits. Although further study is required to determine whether the role of young GCs is identical to or different from the role of their older counterparts, our results provide in vivo evidence that adult-generated neurons are indeed integrated into the broad hippocampal circuitry composed of subcortical modulatory systems (Fig. 6 E and F). The computational modeling of the DG network with the continuous addition of new neurons suggests that events occurring before DA release show stronger correlations with temporally proximal events without DA (Fig. 4), causing a retroactive bias in the temporal relationships if there is more DA for later events than for earlier ones. Such a function for DA has been suggested in a number of other regions for more immediate reward-based learning (44). At neurogenesis time scales, one perspective of this TDL function would be that, for very large rewards, DA in the DG could serve to separate episodic parts of memory into predictive components (Fig. 4E). Events in life would, in a sense, be binned together to represent episodes that lead to significant rewards; what happens postreward is important only with regard to the next rewarding event. Even smaller rewards, with smaller relative effects of DA, would be valuable in encoding causation into memory associations. One of the observations from our computational model is that DA in the DG interacts with the unique memory functions of adult-born neurons, a result consistent with their differential expression of DA receptors (26). As techniques for targeting specific ages of young neurons are developed further, it will be interesting to dissociate their contribution to our observed behavioral results.

Materials and Methods

Detailed methods are provided in SI Materials and Methods. Briefly, TH-Cre transgenic mice were crossed with Ai27 mice or were injected with AAV to express ChR2 in their dopaminergic neurons. Photostimulation (5-ms pulses at 20/50 Hz) was applied through an optical fiber connected to a blue laser (λ = 473 nm). The laser power intensity was kept constant at ∼13 mW for reliable activation of ChR2 in DA neurons. A bundle of four or eight tetrodes was used for recording LFPs in the DG. Time–frequency decomposition was performed for LFP signals obtained before and after optical stimulation from both freely moving and anesthetized animals. Mice subjected to photostimulation were used for fear-conditioning tests. C57BL/6 mice pretreated with milk were used for fear-conditioning or Barnes maze tests. Replication-incompetent retroviruses expressing GFP were prepared and injected into the DG of 6- to 7-wk-old female C57BL/6 mice. At 4–6 wk postinfection (wpi), acute brain slices were obtained from animals pretreated with water or milk, and whole-cell patch-clamp recordings were made from GFP-expressing GCs and their nonlabeled neighbors. Age-matched animals without virus injection were used for extracellular field recordings. Computational simulations were performed on an abstract two-layer neural network model in which DA modulation was simulated by decreasing activity of EC inputs by a variable fraction. All animal experiments were approved by the Animal Care and Use Committees of Huazhong University of Science and Technology and the Salk Institute. All values are reported as mean ± SEM, unless indicated otherwise. Statistical significance was determined by Student’s t test and two-way repeated-measures ANOVA for electrophysiology and behavioral data, respectively.

SI Materials and Methods

Subjects.

TH-Cre transgenic mice (45) and Ai27 mice (46) were both maintained as heterozygous in a specific pathogen-free animal facility and were exposed to a 12-h light/dark cycle with food and water provided ad libitum. They were crossed to produce a colony of TH-Cre;Ai27 mice in which ChR2 expression was restricted in dopaminergic neurons. WT littermates were used as control. Another cohort of TH-Cre mice was injected with viral vectors encoding Cre-dependent ChR2–mCherry or mCherry only. All animals (20–35 g, 2–4 mo old at the time of surgery) were single housed after surgery and throughout the rest of the experiments. For the milk reward experiments, C57BL/6 female mice were purchased from Harlan and were group housed until 1 wk before the start of the behavioral experiment. After being individually housed for 1–3 d, the animals (3–4 mo old) were trained to drink diluted condensed sweetened milk (1:3 milk/water). The sweetened milk was available for 10 min each day for 5 d, and the mice were allowed to drink ad libitum during the training. Typically, the mice drank about 0.7 mL milk in about 5 min and spent the rest of the time grooming and resting. All procedures were conducted in strict accordance with institutional guidelines for the care and use of laboratory animals.

Viral Labeling.

Replication-incompetent retroviruses expressing GFP were used to identify and birthdate adult-generated neurons. The virus was prepared by transfecting plasmids into HEK 293T cells. Concentrated virus-containing supernatant (10⁷–10⁸ cfu/mL) was stereotaxically infused into the DG of 6- to 7-wk-old female C57BL/6 mice with the following spatial coordinates: anteroposterior (A/P) = −d/2 mm from bregma (d represents the distance between bregma and lambda); lateral = −1.6 (if d ≤1.6 mm) or −1.7 mm; ventral = −1.9 mm from dura. Animals were killed for electrophysiological analysis at the indicated times. For optogenetics experiments, mice were anesthetized with urethane and placed into a stereotaxic frame. A volume of 0.5 μL AAV-DIO–ChR2–mCherry or control AAV-DIO–mCherry viral vectors (10¹³ cfu/mL) (Obio Technology) was bilaterally infused into the VTA [A/P −3.1; mediolateral (M/L) 0.5; and dorsoventral (D/V) −4.4–4.6] using a 5-μL Hamilton syringe at the rate of 0.1 μL/min. At the end of infusion, the syringe needle remained in place for 15 min before retraction.

Histology and Imaging.

Mice were killed and perfused with 4% paraformaldehyde (PFA) in PBS. After decapitation, the mouse brains were postfixed overnight and equilibrated in 30% sucrose. Brain tissue was then embedded in optimum cutting temperature (O.C.T.) compound and were either sectioned into 10-μm slices with an upright Leica cryostat or cut into 40-μm slices with a cooled-stage microtome. For immunostaining, slices were incubated with primary antibodies (rabbit anti-TH 1:500, Abcam; mouse anti-DβH 1:300, Millipore) followed by secondary antibodies (Alexa Fluor 488 anti-rabbit IgG 1:1,000, Abcam; Alexa Fluor 488 anti-mouse IgG 1:1,000, Molecular Probes) and finally DAPI. Immunostained tissue samples were mounted on glass slides and imaged using a Leica TCS SP5-II confocal microscope or Leica SD AF spinning-disk confocal system. Image stacks of the DG area were compressed into a single plane using a maximum intensity projection.

Extracellular Recording from Anesthetized Mice.

Mice were anesthetized with urethane and mounted on a stereotaxic frame; body temperature was maintained by a 37 °C recirculating-water heating pad. For simultaneously recording electrophysiological signals and infusing chemical reagents in close proximity to the electrodes, a cannula (100-μm silica capillary tubing; Polymicro Technologies) and four tetrodes made of four twisted Formvar-coated platinum-iridium wires (17 µm; California Fine Wire) were glued to form a bundle with the cannula located in the center and connected to a Hamilton syringe that was filled with drug and mounted in a syringe pump. For optogenetic experiments, an optical fiber (200-µm core diameter; Thorlabs) was affixed to four tetrodes, with the tip of the electrode extending 500 µm beyond that of the fiber. The assembly was first targeted at the VTA (A/P −3.2 mm, M/L ±0.3 mm, D/V −4.0 mm) to validate the expression of opsins before being placed in the DG. Light stimulation (5-ms pulses at 20/50 Hz with a power intensity of 13 mW) was applied through the optical fiber connected to a blue laser (λ = 473 nm; Shanghai Fiblaser Technology Co.) that was controlled by a function generator (AFG3200B; Tektronix). For simultaneous optical stimulation and extracellular recordings in the presence of DA receptor blockers, a cannula and an optic fiber were attached together to four tetrodes. Bipolar tungsten stimulating electrodes (A-M Systems) were placed at the angular bundle of the PP (A/P −3.7 to 3.8 mm, M/L ±2.1 mm, D/V −1.8 mm). The cannula and/or optic fiber assembly with multielectrode array was then lowered into the DG (A/P −1.7 mm, M/L ±0.9 mm, D/V −1.8 to 2.3 mm) to record PP-evoked fEPSPs. All electrode positions were verified histologically following recordings. DA was dissolved in PBS containing 25 μM ascorbic acid to a final concentration of 20 μM. The same volume of vehicle was used as a control in the experiment. The DA antagonist mixture was made in saline containing 10 μM SCH 23390 and 10 μM sulpiride. All drugs were purchased from Sigma.

Extracellular Recording from Freely Moving Mice.

The mice used for freely moving recordings and subsequent behavior tests were chronically implanted with a custom microdrive that performed simultaneous bilateral light delivery and unilateral electrophysiology measurements. We first made an optetrode by attaching a bundle of eight tetrodes to the optical fiber shaft, with the tetrode tip extending 500 μm beyond the fiber end. The optetrode, together with an additional optic fiber, was then attached to the drive unit with Epoxy (Precision Fiber Products). Surgical procedures were performed under deep anesthesia maintained with 0.5% pentobarbital. The assembled microdrive was secured to the skull using four anchor screws and dental cement, with the optic fiber targeted at the appropriate stereotaxic coordinates (A/P −1.7 mm, M/L ±0.9 mm, D/V −1.4 to 1.6 mm). After the animals recovered from the surgery (∼1 wk after implantation), the microdrive was advanced gradually to lower the optetrode to the desired anatomical location. Recordings then were performed in the animals’ home-cage environment. Following in vivo recordings and behavioral experiments, mice were perfused with 4% PFA. Their brains were sectioned to verify the electrode and optic fiber placement as well as opsin expression within axon terminals in the DG. Mice were excluded if the implantation site was incorrectly positioned.

Data Acquisition and Analysis.

All electrophysiological recordings in intact in vivo preparations were performed using the OmniPlex D Neural Data Acquisition System (Plexon Inc.). The electrical signal was filtered at 0.05–8,000 Hz, amplified at a gain of 250–5,000, and digitized at 40 kHz. Spike sorting was performed with Off-Line Sorter software (Plexon Inc.) using automatic sorting methods and manual checking of single-unit isolation. For LFP analysis, signals were down-sampled to a rate of 1,000 Hz and low-pass filtered at 250 Hz. Time-frequency decomposition of the LFP signal was performed using custom code in Matlab and Chronux, an open-source software package for the analysis of neural data.

Contextual Fear Conditioning.

The fear-conditioning paradigm was designed based on immediate shock deficit, a phenomenon referring to the inability of animals to form an association between the conditioning context and an aversive stimulus, such as foot shock, if the stimulus is delivered immediately after the animal is introduced into the context. The immediate shock deficit can be rescued by preexposing the animal to the conditioning context (47). Contextual fear conditioning started at least 5 d after optic fiber implantation or milk training and was performed in a conditioning apparatus from Taimeng or Med Associates. The conditioning chambers were located inside sound-attenuation boxes. On the first day, mice were allowed to explore the conditioning chamber (context A) with a plastic floor for 10 min. The next day, mice received an immediate foot shock (0.72 mA, 2 s) 5 s after being placed into context A′, where the plastic floor was removed to expose the wired floor for shock delivery. Thirty to sixty minutes after the shock, the fear behavior of mice was tested by placing them into context A and into another context (context B) in a counterbalanced order. Context B was modified from context A by changing the olfactory cues (vanilla extract-scented), the shape of the chamber (a curved plastic board was inserted), and distal visual cues (posted on the walls of sound attenuation boxes). Mice were allowed to drink either sweetened milk or water at various points during the behavioral procedure: 40–60 min before preexposure on the first day, immediately after preexposure, 30 min before immediate shock, or 30 min before testing. All behaviors of the mice were recorded and analyzed by video-freeze software (Stoelting ANY-maze or Med Associates).

Barnes Maze.

The Barnes maze was 91.4 cm (3 ft) in diameter with 20 circular holes around the periphery. The diameter of each hole was 5.1 cm (2 in). Each hole was connected to a shallow removable plastic box about 2 cm below the maze surface except for one randomly selected hole, which was connected to a hidden escape tunnel. The apparatus was brightly lit. In the initial training phase, mice were trained for three trials/d for 5 d. For each trial, an individual mouse was placed into an opaque cylinder in the center of the maze to promote spatial disorientation. Thirty seconds later, the cylinder was removed, and the mouse was allowed to explore the maze for 3 min to locate the hidden escape tunnel. The trial ended when the mouse found the escape tunnel or when 3 min had elapsed. Once the mouse reached the escape tunnel, it was left there for about 20 s. If a mouse failed to locate the escape tunnel, it was gently guided to the escape location and was allowed to stay there for 20 s. The position of the escape tunnel remained the same throughout the initial training phase for each individual animal. On the sixth day of training, the escape tunnel was moved to another hole that was 90° from the original position. Two positions were used for the original and the new target locations, which were counterbalanced across mice. Thirty to sixty minutes before training, one group was provided with sweetened milk, and the other group was provided with water. Mice were allowed to drink for 10 min. Mice were subsequently trained for three trials to locate the new escape position. All trials were recorded with EthoVision software from Noldus. The number of errors made by entering an incorrect hole location was scored by a blinded observer. Errors, the distance traveled before reaching the escape tunnel, and the latency to reach the escape tunnel were used as measurements of spatial learning.

Infusion of DA for Behavioral Assays.

Mice were deeply anesthetized with a mixture of ketamine (100 mg/kg) and xylazine (10 mg/kg), and bilateral 26-gauge cannulas (Plastics One Inc.) were implanted stereotaxically aimed at the DG of the dorsal hippocampus (A/P −1.9, M/L ±1.5, D/V −2.1). A jewelry screw was screwed into the skull next to the cannulas to facilitate the fixation of the cannulas. The cannulas were glued to the skull with a layer of cyanoacrylate gel followed by application of dental cement to secure the implants. Animals were allowed to recover from surgery for at least 1 wk before submitting to any procedures. Infusions were performed using 33-gauge infusion needles that were fitted into the cannulas and were connected to an infusion pump. Infusions (1 μL per side) lasted for 112 s, and the needles were left in place for an additional 60 s to minimize backflow. DA (Sigma) solution (7.5 mg/mL in PBS containing 0.2% ascorbic acid) was made fresh on the day of infusion. PBS containing 0.2% ascorbic acid was used for control infusions.

Computational Model.

An abstract two-layer neural network model was simulated with individual neurons represented as binary threshold neurons (i.e., perceptrons or McCullough–Pitts neurons). The input layer had 25,000 excitatory neurons (representing the EC) and 6,250 inhibitory neurons (representing local inhibition). The output layer (representing GCs) modeled neurogenesis by gradually adding neurons to the network; initially, the network had zero output neurons, and two neurons were added per time step up to a total of 200 output neurons. The model did not include feedback or recurrent dynamics, making each output neuron independent of all others. Therefore it was sufficient to model the activity and correlations of output neurons without requiring the highly expansive anatomical structure of the biological EC–DG projection. Neurons were initialized with zero synapses, and they were gradually matured by adding 100 excitatory (+) and 25 inhibitory (−) synapses per step, up to a total of 5,000 (+) and 1,250 (−) synapses per mature neuron. At each time step, a novel combination of input neurons was activated, driving activity in the output neurons. Neurons fired if excitation surpassed inhibition, and input synapses for active neurons were trained using a simple Hebbian-like learning rule. DA modulation of EC inputs to GCs was simulated by decreasing the activity of EC inputs by a variable fraction (5, 12.5, or 25%) between time steps 65 and 70. Likewise, NE modulation of EC inputs to GCs was simulated by increasing the activity of EC inputs by a variable fraction (5 or 12.5%) between time steps 65 and 70. Outputs from each time step were compared with outputs of other time steps using normalized dot product (cosine angle) similarity. A total of 5,000 replicates were performed for each level of DA effect; averages are plotted in Fig. 4.

Retrograde Tracing.

FluoroGold (FG) (Fluorochrome Inc.) was dissolved in 0.9% saline to make a 4% (wt/vol) solution. Seven-week-old female C57BL/6 mice were anesthetized and prepared for surgery in compliance with the guidelines of the Salk Institute. The FG solution was then stereotaxically injected into the DG of the animals with the following spatial coordinates: A/P = −d/2 mm from bregma (d represents the distance between bregma and lambda), lateral = −1.6 (if d ≤ 1.6 mm) or −1.7 mm, ventral = −1.9 mm from dura. Two weeks later, the animals were perfused transcardially with 0.9% saline followed by 4% PFA and then were decapitated. The mouse brains were fixed with 4% PFA and equilibrated in 30% sucrose. Coronal slices of 40-μm thickness were cut with a sliding microtome. Polyclonal anti-TH (1:250 dilution; Chemicon) and anti-FG (1:2,000; Chemicon) antibodies were used for immunostaining. AF488- and Cy3-conjugated secondary antibodies (1:250 dilution) were used against the primary anti-TH and anti-FG antibodies, respectively. Immunostained tissue samples were mounted on glass slides, and images were obtained using a Zeiss LSM 710 laser-scanning confocal microscope.

Slice Electrophysiology.

Mice housed in standard cages were anesthetized by isoflurane inhalation. For natural reward experiments, mice were trained to drink sweetened condensed milk in their home cage for three or four consecutive days and were presented with milk solution immediately before anesthetization. The brains were removed quickly and sliced horizontally using a Leica VT1000S vibratome (200- or 400-μm thickness) in chilled artificial cerebrospinal fluid (ACSF) containing (in mM): choline chloride 110, KCl 2.5, NaH₂PO₄ 1.3, NaHCO₃ 25.0, CaCl₂ 0.5, MgCl₂ 7, glucose 20, Na-ascorbate 1.3, and Na-pyruvate 0.6. Slices were recovered at 32 °C for >30 min in standard ACSF (in mM: NaCl 125, KCl 2.5, NaH₂PO₄ 1.3, NaHCO₃ 25, CaCl₂ 2, MgCl₂ 1.3, Na-ascorbate 1.3, Na-pyruvate 0.6, and glucose 10) bubbled with 95% O₂ and 5% CO₂ and subsequently were transferred to a recording chamber constantly perfused with ACSF at room temperature. Whole-cell perforated patch recordings were obtained from GCs visually identified by infrared differential interference contrast optics and fluorescence, in the presence of 50 μM picrotoxin. The microelectrodes (∼3 MΩ) were tip-filled with an internal solution composed of (in mM) K-gluconate 128, KCl 17.5, NaCl 9, MgCl₂ 1, EGTA 0.2, Hepes 10 (pH 7.3) and then were back-filled with the same internal solution containing 200 μg/mL amphotericin B. GCs were generally held at a potential of −70 mV in voltage-clamp mode. Input and series resistances were monitored continuously, and data were discarded if either changed more than 20%. Extracellular field potential recordings were made using microelectrodes (∼1 MΩ) filled with external ACSF that did not contain GABA receptor blockers. For both recording configurations, a concentric bipolar tungsten electrode was used for stimulation. All data were obtained using Axopatch 200B patch-clamp amplifiers, sampled at 5 kHz, and filtered at 2 kHz using a Digidata 1322A analog–digital interface (Molecular Devices).