Abstract
The hippocampal region has long been considered critical for memory of time, and recent evidence shows that network operations and single-unit activity in the hippocampus and medial entorhinal cortex (MEC) correlate with elapsed time. However, whether MEC activity is necessary for timing remains largely unknown. Here we expressed DREADDs (designer receptors exclusively activated by designer drugs) under the CaMKIIa promoter to preferentially target MEC excitatory neurons for chemogenetic silencing, while freely moving male rats reproduced a memorized time interval by waiting inside a region of interest. We found that such silencing impaired the reproduction of the memorized interval and led to an overestimation of elapsed time. Trial history analyses under this condition revealed a reduced influence of previous trials on current waiting times, suggesting an impairment in maintaining temporal memories across trials. Moreover, using GLM (logistic regression), we show that decoding behavioral performance from preceding waiting times was significantly compromised when MEC was silenced. In addition to revealing an important role of MEC excitatory neurons for timing behavior, our results raise the possibility that these neurons contribute to such behavior by holding temporal information across trials.
SIGNIFICANCE STATEMENT Medial temporal lobe (MTL) structures are implicated in processing temporal information. However, little is known about the role MTL structures, such as the hippocampus and the entorhinal cortex, play in perceiving or reproducing temporal intervals. By chemogenetically silencing medial entorhinal cortex (MEC) excitatory activity during a timing task, we show that this structure is necessary for the accurate reproduction of temporal intervals. Furthermore, trial history analyses suggest that silencing MEC disrupts memory mechanisms during timing. Together, these results suggest that MEC is necessary for timing behavior, possibly by representing the target interval in memory.
- DREADDs chemogenetics neural silencing
- medial entorhinal cortex excitatory neurons
- memory for time
- temporal goal-directed task
- timing behavior
- trial history analyses
Introduction
Perceiving and memorizing time intervals is integral to our ability of interacting with the changing world. For example, appropriately reproducing a temporal interval entails comparing elapsed time with a reference target interval, stored in memory, all the while continually updating this reference based on feedback from previous trials. What brain circuits are involved in maintaining such a memory for time is still unclear. The hippocampus (HIPP) has been considered critical for memory of elapsed time (Meck et al., 1984, 2013; Jacobs et al., 2013; Yin and Meck, 2014; Sabariego et al., 2019). Recent evidence shows that neural activity in the HIPP and its major input region, the medial entorhinal cortex (MEC), encodes elapsed time through the sequential firing of time-sensitive neurons, or time cells, at specific moments of delay periods (Gill et al., 2011; Kraus et al., 2013; MacDonald et al., 2013; Kraus et al., 2015; Salz et al., 2016; Heys and Dombeck, 2018; Umbach et al., 2020). This sequential organization of firing patterns in the HIPP is likely dependent on intact MEC inputs (Schlesiger et al., 2015; Robinson et al., 2017; but see Sabariego et al., 2019), which are critical to holding information across delays (Sauvage et al., 2010; Suh et al., 2011; Kitamura et al., 2014). A recent optogenetic silencing study further showed that MEC is necessary for memorizing information across delays, and that the effects of MEC silencing outlasted this manipulation, persisting in subsequent nonmanipulated trials (Robinson et al., 2017). Recent data show that transiently silencing nonspecific MEC neural activity using optogenetics disturbs learning of a timing task, but not postlearning timing behavior (Heys et al., 2020). What class of MEC neurons is involved in timing and what specific role they play in timing behavior remains unknown. Based on previous findings, we suggest that MEC excitatory neurons are involved in holding a reference memory of the target interval during timing behavior. Thus, disrupting these neurons would impair the ability of rats to access the reference memory of the target interval, impairing timing performance.
To address this question, we used the hM4Di inhibitory DREADD (designer receptor exclusively activated by designer drugs; Roth, 2016) under the control of the CaMKIIa promoter to preferentially silence MEC excitatory neurons (Nathanson et al., 2009; Watakabe et al., 2015; Dimidschstein et al., 2016; Xu et al., 2016; Meng et al., 2018), while rats reproduced a memorized interval by waiting inside a region of interest (ROI). First, we hypothesized that if MEC excitatory activity were necessary for accurate timing, this neuronal silencing would compromise task performance. Second, if MEC were involved in maintaining or updating a putative reference memory of the target interval, our manipulation would also modulate intertrial influences and dissociate recent trials from current behavior (Taatgen and van Rijn, 2011; Bausenhart et al., 2014).
We found that silencing MEC excitatory neurons critically impaired timing performance and dissociated immediate past trials from waiting behavior.
Materials and Methods
Subjects
Seventeen male Long Evans rats (Rattus norvegicus), 4–10 months old and weighing approximately between 450 and 700 g (Charles River Laboratories), were used for all experiments. Five animals were used for the MEC silencing protocol. During experiments, the animals were housed individually and kept under a 14/10 h light/dark cycle. Rats were gradually food deprived to ∼90% of their free-feeding body weight. All the procedures were approved by the Portuguese National Authority for Animal Health, as well as by the animal well being office at Instituto de Medicina Molecular João Lobo Antunes.
Apparatus
Behavior training and MEC silencing experiments were performed on a 200 × 15 cm black acrylic Plexiglas linear track. At one end of the track was a start position area (wider area with 25 × 29.5 cm), where animals were placed at the beginning of each session. From this start position, a ROI was placed ∼3 cm into the track where the animals were trained to wait to obtain reward. This ROI was outlined with high-contrast tape, and bordered by a 5 cm barrier on each side. Reward was delivered at the opposite end of the track through a semiautomatic, custom-built, peristaltic pump circuit controlled by an UNO board (Arduino) with Bonsai software (Lopes et al., 2015). A green LED and a Go speaker (JBL), both of which were affixed to the wall at the reward end of the track, were used to deliver sound and light cues. Before behavior training, animals were accustomed to wear a removable low-power LED on their heads to facilitate position tracking. Overhead video was recorded with a Flea 3 camera (Point Grey) at 30 frames/s. Custom Bonsai routines were used for online position tracking, task timing, stimulus and reward delivery, as well as recording data.
Behavior training—waiting for trajectory task
Pretraining.
At the start of behavior training, the food-deprived rats were briefly introduced to the track (one or two sessions) and allowed to freely explore. When needed, chocolate milk was sprinkled on the track to promote exploration. On the second day, rats received a reward every time they performed a complete trajectory—a full traversal of the linear track from the starting position to the reward port, and back to the initial position without back laps (Fig. 1A, Movie 1). This was repeated until the animals performed full traversals with minimal interruptions (one to four sessions).
Two example trials of the Waiting for Trajectory task. The first trial is an incorrect trial (the animal waited less than 2.5s) and the second trial is a correct one (the animal waited 2.5s or more). Note that the animal has no access to external time cues and performed the task in a self-paced way.
Waiting for trajectory task.
After the initial pretraining stage, rats were introduced to the waiting-for-trajectory (WfT) task. A trial started when the animals entered the ROI from the start position and traversed the full length of the track to the reward location. For another trial to begin, the animals need to traverse the whole extent of the track back to the start position, where they could move freely before starting a new trial (Movie 1, example trials). Whenever animals entered the ROI while traveling from the starting to the reward points on the track, a brief tone (50 ms beep) marked the beginning of the waiting interval. Simultaneously, a green light was turned on for the duration of the target time. Trials were rewarded only when animals waited inside the ROI until the light was turned off. The initially rewarded waiting period was 1 s. The interval was progressively increased to the final 2.5 s duration over the subsequent sessions (see below). To train animals to voluntarily wait for the predefined period inside the ROI, a transparent door was manually placed at its exit. For the first sessions of training, the door was used in all trials (i.e., “door sessions”). During door sessions, the experimenter was inside the behavior room and manually placed and removed the door as a function of light onset and offset. Two to five door sessions were usually necessary before animals started to wait inside the ROI without the door. A correct trial was defined as a trial in which the rat entered the ROI and waited for the specified time interval after the beep. Furthermore, one door trial was introduced after five consecutive incorrect trials, until animals performed waiting periods voluntarily and reliably. This ended once they reached 50% correct trials for the first interval (1 s). Throughout the training and experimental stages, the door was used for the first three trials of each session as a refresher of the waiting period inside the ROI. To increase the waiting interval, the median of correct trial times (per session) was used as the new interval duration for the subsequent session (i.e., light cue duration). When performance reached 50% of correct trials at 2.5 s waiting time, the light cue was removed. From this stage onward, the animals had to reproduce the waiting period inside the ROI with only the tone cue and without any end-of-interval cue (i.e., light duration/offset). At this stage, the task became fully self-paced, and the production of the target interval was dependent on the internal representation of time of the animals. After achieving 50% correct trials, the viral vector for DREADDs expression AAV8-CaMKIIa-hM4D(Gi)-mCherry was stereotaxically injected in MEC, for animals in the experimental group (hM4Di group). Control animals started the MEC silencing protocol immediately after reaching this performance criterion (two to three consecutive sessions >50%).
MEC silencing protocol
After recovery, hM4Di animals were food deprived and reintroduced to the WfT task following the protocol described above. Behavior training was maintained until ≥30 d postsurgery when needed to achieve 50% correct trials at 2.5 s without the light cue. Typically, postsurgery training took between 2 and 3 weeks. To silence MEC, the inhibitory DREADD was activated through administration of clozapine N-oxide (CNO; 3 mg/kg, i.p.). A saline (SAL) solution (same volume as CNO) was administered in these same animals during control sessions. In addition to this group of animals, a control group was administered CNO in the absence of the expression of DREADDs to rule out known off-target effects of CNO metabolites per se (Gomez et al., 2017). For 10 consecutive days, alternate injections of either CNO or saline were performed ∼60 min before behavioral testing in the WfT task. The same protocol was administered to animals in the control group.
Surgical procedures
Viral injection surgery.
To allow the manipulation of MEC neurons, a commercially prepared suspension of viral vector AAV8-CaMKIIa-hM4D(Gi)-mCherry (Addgene) was stereotaxically injected bilaterally. Before surgery, the animals were placed in an induction chamber saturated with 5% isoflurane until absence of the righting reflex. Once the animals were deeply anesthetized, they were weighed and transferred to a heating pad, where they received an intraperitoneal injection of a ketamine (25 mg/kg) and xylazine (3 mg/kg) cocktail and their heads were shaved. After that, the animals were installed in the stereotaxic apparatus, with lidocaine applied to the ears, and eyes covered with a protection gel (Lubrithal) to prevent drying of the cornea. Anesthesia was maintained with 1–2% isoflurane in oxygen until the end of the procedure. Once animals were placed on the frame, the surgical site was scrubbed with 10% povidone iodine alternated with 70% ethanol three times before an incision was made to expose the skull. Bregma and lambda were carefully identified and marked on the skull surface. These two landmarks were used to level the skull horizontally. Once the skull was leveled, the craniotomy coordinates were marked on the bone. Before drilling, the bone surface was treated with Baytril to minimize infection.
A high-power microdrill tool was used to perform a single craniotomy on each hemisphere over the microinjection spots. Loose bone pieces were carefully removed, and a small needle was used to remove the dura mater and expose the brain surface. The viral construct was injected using a microinjection control system attached to the stereotaxic frame and a glass micropipette. The micropipette was filled with mineral oil and placed in the microinjector, after which the viral suspension was aspirated. Using bregma as a reference, the pipette was carefully positioned over the microinjection coordinates and slowly lowered until reaching the targets [rat 1: anteroposterior (AP), −9.0; mediolateral (ML), ±5.2; dorsoventral (DV), −5.0 to 1 µl; and AP, −8.4; ML, ±4.5; DV, −6.0 to 0.5 µl; rat 2: AP, −8.8; ML, ±5.1; DV, −5.0 to 1 µl; and AP, −8.4; ML, ±4.5; DV, −6.0 to 0.5 µl; rat 3: AP, −9.0; ML, ±5.2; DV, −5.0 to 1 µl; and AP, −8.4; ML, ±4.5; DV, −6.0 to 0.5 µl; rat 4, left hemisphere: AP, −9.0; ML, −5.07; DV, −5.0 to 1 µl; and AP, −8.4; ML, −4.5; DV, −6.0 to 0.5 µl; rat 4, right hemisphere: AP, −9.12; ML, 5.0; DV, −5.0 to 1 µl; and AP, −8.6; ML, 4.5; DV, −6.0 to 0.5 µl]. The virus was delivered after a 5 min rest period at a rate of 100 nl/min. To avoid spillover onto overlaying structures, we waited 10 min before slowly retracting the micropipette. Finally, the wound was closed using absorbable sutures, and animals were administered Ringer's lactate and carprofen for hydration and analgesia, respectively. Animals were then placed in a heated cage during the first 24 h postsurgery for recovery. During this period, ad libitum food, nutritional gel, and water were provided. For the firsts 2 d, a postoperative analgesic (buprenorphine, 0.3 mg/kg) dissolved in chocolate milk was administered when necessary.
Implant surgery.
To confirm local and distal (HIPP) effects of MEC silencing, we recorded electrophysiological data from two of the animals that underwent viral injection surgery (hM4Di group). For this purpose, two implants (Liang et al., 2017) carrying 27 independently movable tetrodes were used in these animals to record local field potentials (LFPs) and single-unit spikes from MEC and HIPP (see Fig. 3B). Tetrode bundles targeted the following coordinates: MEC (9 tetrodes): AP, between −9.0 and −8.0; ML, between −4.0 and −5.0; HIPP (18 tetrodes): AP, between −3.2 and −4.2; ML, between −2.4 and −4.4. The implant surgery was performed using similar procedures for the viral injections. Once the surgical field was ready, nine anchor screws were fixed to the skull (one placed over the cerebellum to be used as a ground) before a craniotomy was made for the implant. The dura mater was removed, and mineral oil was applied to the surface of the brain. Finally, the hyperdrive was carefully lowered and firmly fixed to the skull and to the anchor screws with self-curing dental acrylic (TAB 2000, KERR). Once the dental acrylic hardened, the surgical wound was closed, and the animal was left to recover over a period of 1 week. The protocol was adapted from the studies by Remondes and Wilson (2013, 2015).
Reagents and solutions
Viral constructs.
The AAV8-CaMKIIa-hM4D(Gi)-mCherry used to silence neural activity was acquired from Addgene (catalog #50 477-AAV8).
Clozapine N-oxide solution preparation.
CNO (25 mg; Enzo Life Sciences) was dissolved in 0.125 ml dimethylsulfoxide (DMSO) and then further diluted in 24.875 ml of isotonic saline to a final concentration of 1 mg/ml. Care was taken to obtain a solution with the lowest percentage possible of DMSO; in this case, the final concentration of DMSO in the CNO solution was 0.5%.
Data acquisition: in vivo recordings
Electrophysiological data were recorded from two of the hM4Di animals used in the silencing protocol, to monitor the effects of the activation of silencer DREADDs on MEC and its CA1 target neurons. Three days after the surgery, tetrodes were lowered toward the targets. Recordings were performed under CNO and saline conditions, while animals performed the WfT task. Single-unit spikes and LFPs were recorded using an Open Ephys system based on Intan acquisition boards. Extracellular action potentials and continuous LFP were acquired at 30 kHz per channel, digitized and amplified using RHD2164 amplifier boards, and sent to the acquisition computer under the control of Bonsai (Lopes et al., 2015).
Histology
Expression of hM4Di-mCherry was confirmed postmortem. At the end of the experiments, animals were killed with an isoflurane overdose, and transcardially perfused with 300 ml of PBS followed by 500 ml of 10% formalin. Brains were dissected and kept in formalin for 24 h at room temperature, then placed in 15% sucrose followed by 30% sucrose in PBS. Once sunk, the brains were embedded in gelatin, frozen, and sectioned in 50 µm sagittal slices using a cryostat (model CM3050 S, Leica). To amplify the hM4Di-mCherry signal in labeled tissue, anti-mCherry immunohistochemical staining was performed. Fixed brain slices were degelatinized in PBS at 37°C for 10 min, followed by a 20 min incubation with 0.1 m glycine in PBS. Slices were permeabilized and blocked in a solution of Tris-buffered saline (TBS) with 0.5% Triton X-100 and 10% fetal bovine serum (FBS) for 1 h. Brain sections were then incubated in a rabbit anti-mCherry primary antibody (1:200; catalog #PA534974, Thermo Fisher Scientific) diluted in a solution of TBS with 0.2% Triton X-100 and FBS 4% at 4°C for 24 h. The sections were subsequently washed three times for 15 min with TBS with 0.1% Triton X-100 and then incubated overnight at 4°C in secondary Alexa Fluor-546 donkey anti-rabbit (1:400; catalog #A32754, Thermo Fisher Scientific) diluted in block solution. Following further washing, the sections were incubated for 10 min with Hoechst 33342 stain (final concentration, 12 μg/ml; Thermo Fisher Scientific) and washed for 15 min with PBS. Slices were mounted and coverslipped Fluoromount and left to dry in a dark place. The expression of the hM4Di viral construct was confirmed using an Axio Observer wide-field fluorescence microscope (Zeiss) equipped with an Axiocam 506 mono CCD (Zeiss), and an LSM 880 confocal point-scanning microscope with Airyscan (Zeiss). The histology protocol was also performed for control animals.
Data analysis: electrophysiological analysis of MEC activity disruption
All analyses were performed using MATLAB (MathWorks) custom-written code.
Raw data were bandpass filtered between 700 and 8000 Hz, and single-unit and multiunit spiking activity was extracted based on 4 SD threshold crossings. Single units were merged with multiunit clusters for the purpose of multiunit analysis. Putative clusters were visually inspected for noise contamination and only clean clusters were included in further analyses. UltraMega Sort 2000 (Hill et al., 2011), as well as custom-written MATLAB (MathWorks) code, were used offline for single-unit detection and sorting.
Experimental design and statistical analysis
Task performance was computed for individual sessions as the percentage of correct trials (waiting time inside the ROI, ≥2.5 s). To determine whether MEC inactivation led to a disruption of the ability of the animals to judge elapsed time, the performance levels of the CNO and saline sessions were compared using a two-way ANOVA (group × condition) followed by Tukey's post hoc tests. Waiting times were compared between conditions to better understand how MEC silencing affected behavior. In both conditions, the distribution of waiting times was bimodal with a right heavy tail. The Wilcoxon rank-sum test was used to determine whether the median of the waiting times differed significantly across conditions. To better characterize the differences between distributions of waiting times, the shift function method was used to compare the spread of the distributions of waiting times across conditions (saline vs CNO; Rousselet et al., 2017). In essence, a shift function shows quantile differences as a function of the distributions. All trials were used to measure performance and the distribution analyses of waiting times. To test trial history effects, we used partial autocorrelation functions (MATLAB function parcorr) to measure the correlation between each trial and its lag-n, after adjusting for effects of intermediate lags (e.g., lag-n-1). The 2D trial plots for lags n-1, n-2, and n-3 were binned in 200 ms bins, and the insets were smoothed using the MATLAB contourf function. Logistic regression (GLM) was used to model trial outcome as a function of the time of the previous trial (Trialn-1), or the difference between current trial (Trialn) and Trialn-1 times (Trialn – Trialn-1), as well as condition (CNO vs saline). Cook's distance was used to remove high influence data points (for Trialn-1: 44 of 1369 trials; for Trialn – Trialn-1: 74 of 1369 trials). We report values as the mean ± SD, and the significance level was established at p < 0.05. All tests were one tailed, unless otherwise noted in the respective sections of Results and in figure legends. The experimenter was not blind to group or condition.
Results
Rats learn to reproduce a memorized interval in a temporal goal-directed task—the waiting-for-trajectory task
Most behavioral protocols testing the encoding of time in MEC and HIPP do not include explicit timing requirements. In these protocols, time is circumstantial, but not integral, to the goal. Here we developed a goal-directed timing task—the WfT task—to test the involvement of the MEC in explicit timing. Unrestrained rats were taught to wait motionless in a fixed region of interest at the start of a 2 m linear track, for the duration of a previously memorized target interval (2.5 s), to obtain reward (drops of chocolate milk) at the end of the track (Fig. 1A, Movie 1). A 50 ms tone signaled trial onset, and the start of the waiting period. Rats needed to voluntarily withhold movement for the duration of the memorized interval without the guidance of external timing cues or physical barriers. If the waiting period corresponded to, or exceeded, the memorized target interval, reward was delivered at the far end of the track (i.e., correct trials; Fig. 1A, Movie 1).
Estimation of elapsed time using a self-paced and uncued waiting task. A, Schematic depiction of the WfT task. A trial started when the animals left the start position (wide area on the left of the maze) and entered the ROI. Rats were rewarded at the end of the track with chocolate milk for waiting inside the region of interest for ≥2.5 s. See Movie 1 for example trials. B, Example learning curve over interval steps (1 s L, 1.5 s L, 2 s L, 2.5 s L, and 2.5 s NL). Colored circles indicate sessions at target performance (≥50% correct trials) over learning and interval steps. Dashed line indicates 50% performance. C, Mean speed across interval steps for one rat. Vertical dashed lines indicate target time for each interval step. D, Proportion of rats that learned the uncued (NL) version of the waiting-to-trajectory task, rats that only learned the cued version of the task (≤2.5 s L), and rats that did not learn the task (nonlearners).
During pretraining, rats were rewarded for waiting inside the ROI with the help of an additional light cue that signaled the target waiting time. Rats waited inside the ROI for the duration of the light, which was progressively increased across sessions (Fig. 1B,C; see Materials and Methods for details). After achieving 50% correct trials in a session, with the light cue at 2.5 s (L), the self-paced and uncued “no-light” (2.5 s; NL) version of the task was introduced (Fig. 1B,C). At this stage, rats had no access to external cues guiding the waiting period and had to retrieve a previously memorized target time to successfully complete the task. Comparable to previous work with mice in a virtual reality head-fixed preparation (Heys and Dombeck, 2018), waiting behavior in the freely moving rat became highly stereotyped, consisting of a well defined period of immobility inside the ROI, followed by significant acceleration starting on completion of the target interval (Fig. 1C). This pattern suggested a similar strategy across distinct intervals, tuning to the target interval in the absence of external temporal cues, and showed that the interval was memorized. Although differing in the number of sessions to reach the NL stage (mean ± SD, 40.11 ± 7.93; minimum, 32; maximum, 52), the learning process was stereotyped across animals, with 52.94% reporting the target time without the help of external cues, 35.29% reporting only the cued version (pretraining), and 11.77% reporting none (Fig. 1D).
These results show that rats can indeed memorize and reproduce an interval, to reach a goal, and that we can use the WfT task to consistently train animals to memorize and report a temporal interval.
Rats retrieve a learned interval from memory during successful performance of the WfT
To accurately perform the NL version of the WfT task, rats must retrieve the previously memorized interval and reproduce its duration by waiting inside the ROI. This entails reproducing in NL the behavioral parameters manifest in the presence of a light cue (L sessions). To verify whether this was the case, we compared the speed profiles of individual trials in the two conditions. We found that in both conditions the instantaneous speed profile of rats was similar around the ROI, exhibiting consistent kinetics across trials (Fig. 2A,B), starting with an initial peak immediately before entry in the ROI, followed by a variable period of immobility, corresponding to the actual waiting epoch. In most trials, a second speed bout happened when rats entered the linear track on their way to reward. This pattern was consistently found in NL trials, when rats reported time in an uncued and self-paced manner (Fig. 2B; N = 6; mean speed: Mann–Whitney U test = 13,269, p = 0.2233; mean ± SD: NL, 6.18 ± 1.88; L, 7.13 ± 4.31). Furthermore, the accuracy of reported time was comparable between the light-cued condition and the uncued one, where rats needed to rely on the memorized reference of the target interval (Fig. 2C; N = 6; Mann–Whitney U test = 37,753; p = 0.612; mean ± SD: NL, 2.58 ± 1.12; L, 2.55 ± 1.16). In both L and NL conditions, speed and waiting time followed similar distributions, indicating that animals memorized the target interval and that they used such memory to reproduce the target interval in the absence of external timing cues. Thus, rats relied on the retrieval of an enduring representation of the target time, which allowed them to report time accurately and efficiently.
Rats wait for a learned target time in the absence of external temporal cues. A, Top, Speed across all trials of an example cued session at 2.5 s (2.5 s L). Bottom, Speed across all trials of an example uncued session at 2.5 s (2.5 s NL). Data from the same animal. Trials were sorted according to waiting time inside the ROI. The white line at 0 s represents trial onset tone, and white line at 2.5 s represents the target time. Black arrowheads represent the boundary between correct and incorrect trials for each session (≥2.5 s waiting inside the ROI). B, Mean speed of combined correct trials for 2.5 s L versus 2.5 s NL (solid line, mean; shading, SD; N = 6; Mann–Whitney U test = 13 269, p = 0.2233). C, The distributions of combined waiting times of uncued and cued sessions do not differ (N = 6; Mann–Whitney U test = 37 753, p = 0.612). For each rat, the first session at ≥50% correct trials for each condition was used (2.5 s NL vs WL). Colored arrowheads represent median waiting times of distributions (100 ms bins). WL, With light.
Activation of inhibitory DREADDs using CNO partially silences MEC and HIPP
To preferentially silence MEC excitatory activity, we bilaterally expressed the hM4Di inhibitory DREADD (Roth, 2016), under the control of the CaMKIIa promoter, a technique generally used to restrict the expression to MEC excitatory neurons (Nathanson et al., 2009; Watakabe et al., 2015; Dimidschstein et al., 2016; Xu et al., 2016; Meng et al., 2018; Fig. 3A, Extended Data Fig. 3-1). We first tested the effectiveness of this silencing approach by using tetrodes (Liang et al., 2017) to simultaneously record neural activity from MEC and dorsal HIPP (Fig. 3B) while activating the inhibitory DREADD using CNO. We compared neural activity under CNO activation with baseline activity recorded after vehicle (saline) injection from two rats while they performed the task on the track.
Local MEC and distal HIPP multiunit activity is reliably silenced by CNO administration in rats expressing hM4Di. A, Schematic (left) of double, bilateral, hM4Di DREADD injection in MEC and anti-mCherry immunostained sagittal section (right) showing extensive expression of hM4Di across the dorsoventral axis of MEC. Note abundant staining of canonical HIPP projections. See Extended Data Figure 3-1 for individual examples. B, Schematic of simultaneous MEC (black) and HIPP (gray) tetrode recordings (left) and 2 s of example LFP (right) recorded simultaneously in MEC and HIPP. C, MEC multiunit activity (20 s bins) from example pair of consecutive sessions. D, Average firing rate for MEC across recording session (N = 5) of two rats (one-tailed Wilcoxon signed-rank test, p = 0.031). E, Average firing rate for HIPP across recording session (N = 5) of two rats (one-tailed Wilcoxon signed-rank test, p = 0.031). F, Percentage change of multiunit activity in CNO sessions compared with consecutive saline sessions.
Figure 3-1
Histology of individual animals. A, Individual examples of left and right hemisphere slices for each animal in the hM4D group. Although variable, the mCherry-tagged hM4D DREADD expression is significant in all animals. Note abundant staining of canonical HIPP projections in all animals. Download Figure 3-1, TIF file.
We found that the administration of CNO significantly reduced the rate of detected multiunit spikes in the MEC when compared with vehicle injection (Fig. 3C,D,F; one-tailed Wilcoxon signed-rank test, p = 0.031; CNO, 50.99 ± 23.97; saline, 73.83 ± 31.05), with no other apparent differences in the LFP. The effects of our manipulation in MEC extended downstream to the HIPP, which is known to be involved in memory for time (Jacobs et al., 2013; Sabariego et al., 2021), with all CNO sessions showing a reduction in firing rate relative to paired saline sessions (Fig. 3E,F; one-tailed Wilcoxon signed-rank test, p = 0.031; after CNO, 69.00 ± 36.33; after saline, 87.76 ± 49.30).
MEC is critical for accurate estimation of elapsed time
If MEC circuitry were necessary to accurately reproduce a memorized interval, disturbing the activity of MEC would impair the overall performance of the rats in the WfT. To test this, we pretrained animals in the WfT task, and then bilaterally expressed the hM4Di inhibitory DREADD, under the control of the CaMKIIa promoter as before, thus expressing hM4Di preferentially in MEC excitatory neurons (Extended Data Fig. 3-1; n = 4). Non-virus-injected animals (n = 3) served as controls for the action of CNO per se. Following recovery and reacquisition training to target performance, we started a schedule of 10 WfT sessions. For this, we intraperitoneally injected either CNO (MEC silencing) or SAL (control condition) on 10 alternate days of one WfT session (Fig. 4A), for a total of five SAL and five CNO sessions per animal. We found that silencing MEC significantly lowered behavioral performance, by impairing the reproduction of the target interval in CNO sessions, but not SAL sessions (Fig. 4B, left; ANOVA, group × condition interaction: F(1,64) = 4.97, p = 0.029; multiple comparisons: hM4Di CNO × hM4Di SAL: p = 7.26 × 10−4; mean difference = −15.24; CI, −25.11, −5.38). Conversely, control animals were unaffected by CNO administration (Fig. 4B, right; ANOVA, multiple comparisons: control CNO × control SAL: p = 0.918; mean difference = −2.70; CI, −13.80, 8.40), with none showing performance decline after CNO administration (Fig. 2B), ruling out possible off-target effects of CNO on behavior.
Silencing MEC impairs the estimation of elapsed time. A, Inactivation protocol. CNO or saline was alternately injected intraperitoneally for a period of 10 d while behavior was recorded. B, Behavioral performance during silencing (CNO) of MEC and control (saline) conditions (horizontal black line, mean; vertical black line, SD). Silencing the excitatory activity of MEC significantly impairs estimation of elapsed time in the hM4Di rats (ANOVA, multiple comparisons: hM4Di CNO × hM4Di SAL: p = 7.26 × 10−4) but not in control rats (ANOVA, multiple comparisons: control CNO × control SAL: p = 0.918). Each color corresponds to one animal. Circles with black edge represent mean performance, and light-colored circles represent individual sessions. C, Shift function analysis shows a consistent overestimation (i.e., leftward shift) of waiting times for hM4Di rats. Colored circles represent individual waiting times across all sessions (CNO vs saline). Gray vertical bars represent deciles, with the widest bars representing the median of the distributions. D, Decile differences between conditions (vertical lines, 95% bootstrap confidence interval). CNO waiting times are consistently bellow the saline deciles. E, Distributions of waiting times of hM4Di rats (Mann–Whitney U test = 461,825.5, p = 1 × 10−7). All waiting times, in both conditions, were used for statistical analysis. For display purposes, distributions were truncated at 7.5 s. Colored arrowheads represent the median waiting times of each distribution. Data in 100 ms bins. F, Cumulative distribution of waiting times of hM4Di rats (solid line, mean; shading, confidence bounds). Differences between CNO and saline distributions are visible around the target time (2.5 s). *p < 0.05. n.s., Not significant.
To better understand the effects of our manipulation, beyond the observed impairment, we analyzed waiting times across all trials, between CNO and SAL treatments in the hM4Di group. Specifically, we asked how silencing MEC impacted interval reproduction at the individual trial level (Fig. 4C). Our results show that silencing MEC reduced waiting times by ∼ 15% (Fig. 4D,E; Mann–Whitney U test, U = 461825.5, p = 1 × 10−7; CNO median = 2116 ms; SAL median = 2500 ms). When MEC was silenced, rats apparently judged that more time had elapsed than it did as measured by a clock (i.e., veridical time), ending waiting periods prematurely, and possibly overestimating elapsed time. To compare the spread of the two distributions, we used a shift function analysis (Rousselet et al., 2017). In essence, we subtracted each decile value of one condition (saline) from the decile value of the other (CNO). We observed a consistent, significant reduction in waiting times under CNO compared with SAL (Fig. 4D,E), with all deciles, except for the first and last, of the CNO waiting times falling significantly below the ones under SAL (Fig. 4D; zero is outside confidence intervals; first decile (D1) = −145.10, p = 0.006 (n.s.); D2 = −307.03, p = 0; D3 = −449.85, p = 0; D4 = −456.08, p = 0; D5 = −380.36, p = 0; D6 = −300.54, p = 0; D7 = −305.10, p = 0; D8 = −313.67, p = 0.002; D9 = −251.32, p = 0.188 (n.s.); p values were corrected for multiple comparisons with Hochberg's method (Hochberg, 1988; Rousselet et al., 2017). This overestimation of elapsed time is consistent with previous results observed in rodents and humans after HIPP lesions or dysfunction (Meck et al., 2013).
Importantly, the overestimation of time was more pronounced around the target (Fig. 4F), with 34.01% of SAL trials exhibiting waiting times between 2.5 and 3.5 s, contrasting with only 23.56% of CNO trials found in this range. Consistent with this observation, 31.84% of CNO trials fell between 1.5 and 2.5 s, whereas only 23.63% of the SAL trials fell within this target (Fig. 4G). Furthermore, the proportion of very low and very high waiting times (D1 and D9) accounted for a comparably low percentage of trials in both conditions (D1: CNO = 9.82%; saline = 10.38%; D9: CNO = 9.54%; saline = 9.80%) and was not significantly different between CNO and SAL trials (Fig. 4C,D). In other words, the bulk of differences between conditions was found around the 2.5 s target, while extreme waiting times did not account for the observed impairment. Together, these results indicate that excitatory activity in MEC is necessary to accurately reproduce a previously memorized interval.
Silencing MEC excitation modulates contextual effects of previous trials
Human studies have long established that the perception and production of temporal intervals are influenced by previously experienced trials (i.e., the temporal context; Vierordt, 1868; Jazayeri and Shadlen, 2010; Taatgen and van Rijn, 2011; Shi et al., 2013; van Rijn, 2016; de Jong et al., 2021). In the WfT task, we define temporal context as the recently experienced or reproduced intervals (i.e., trial history). Having found that silencing the MEC critically affects the reproduction of a memorized interval, we asked whether it also affects the mnemonic maintenance of the temporal context and its influence on individual trial waiting times. Specifically, we first asked whether the recent trial history (Fig. 5A) had any quantifiable effect over current trial waiting times, and then asked whether silencing MEC could affect such putative effects. To test whether trial history influenced current trials, we first quantified the correlation of current waiting times with the ones from n preceding trials. Using partial autocorrelation function (PACF), we found that current trial waiting times were significantly, and independently, correlated with the waiting times up to the three preceding trials (Fig. 5B; PACF for SAL: Trialn-1 = 0.17, CI, −0.08, 0.08; Trialn-2 = 0.18, CI, −0.08, 0.08; Trialn-3 = 0.14, CI, −0.08, 0.08; PACF for CNO: Trialn-1 = 0.20, CI, −0.08, 0.08; Trialn-2 = 0.16, CI, −0.08, 0.08; Trialn-3 = 0.05, CI, −0.08, 0.08). Furthermore, using multiple linear regression models, we confirmed that contextual effects were present under both conditions (SAL: R2Adjusted = 0.075; F(3,633) = 18.1, p = 2.7 × 10−11; CNO: R2Adjusted = 0.057; F(3,652) = 14.2, p = 5.61 × 10−9), indicating that current timing behavior is not independent of previous waiting times (Table 1) and suggesting that trial history information is incorporated in later judgments of elapsed time.
Estimated coefficients from multiple linear regressions for CNO and saline trials for each lag
MEC modulates the contextual effect of previous trials. A, Schematic depiction of the trial history. The contextual effect of Trialn-i decay over time (gray shading), and more recent trials have a greater influence on Trialn. B, PACF of waiting times across all animals (dashed lines, confidence bounds). The y-axis represents correlation values, and the x-axis represents the trial lags. Note that for all time series, at lag zero the autocorrelation = 1. C, 2D histograms for mean-centered (i.e., residuals) waiting times of the Trialn compared with the mean-centered (i.e., residuals) waiting times three preceding trials (Trialn-1, Trialn-2, and Trialn-3). Left panels correspond to pooled saline data, and the right panels correspond to CNO. Insets, 2D histograms represent smoothed counts between −2.5 and 2.5 s. Color bars represent normalized counts. Data in 200 ms bins. D, Median waiting times for preceding and succeeding correct (Hits) or incorrect (Miss) trials (vertical lines, SD). The pre-Hit and post-Hit times are significantly lower in the CNO condition (pre-Hit: Mann–Whitney U test = 116,374, p = 1.70 × 10−5; Hit: Mann–Whitney U test = 105,019.5, p = 0.2897; post-Hit: Mann–Whitney U test = 115,003.5, p = 2.58 × 10−4). The pre-Miss and post-Miss times are significantly lower in the CNO condition (pre-Miss: Mann–Whitney U test = 115,726.5, p = 0.0327; Miss: Mann–Whitney U test = 112623.5, p = 0.3069; post-Miss: Mann–Whitney U test = 117,167.5, p = 0.008). ***p < 0.001, **p < 0.01, *p < 0.05.
To better visualize the effects of temporal context on timing behavior, we analyzed consecutive waiting times and compared waiting times in a 2D space where Trialn waiting times were compared with the ones in Trialn-i, with i=1,2,3 corresponding to Trial n-1, n-2 and n-3. Additionally, to account for differences between conditions we mean centered the waiting times. As expected from the observed overestimation of time, we found that silencing MEC increased the number of consecutive short waiting times (Fig. 5C, right; increased counts in the bottom left quadrant). This contrasts with the control condition (Fig. 5C, left), where trials clearly clustered at the intersection of target times (Trialn vs Trialn-i). Silencing MEC excitatory activity led to a predictable increase in the number of consecutive misses (i.e., incorrect trials) and a reduced number of consecutive hits (i.e., correct trials), as if the contingency between waiting time accuracy and reward were partially lost. This observation is strengthened by the relative increase in the number of short consecutive waiting times, observed independently of the correctness (hit or miss) of the current trial, with both correct and incorrect trials being preceded by shorter waiting times when MEC excitation was silenced (Fig. 5D: pre-Hit, Mann–Whitney U test = 116,374, p = 1.70 × 10−5; pre-Miss, Mann–Whitney U test = 115,726.5, p = 0.0327).
These results suggest that decreasing MEC excitatory activity might influence the effects of trial history (i.e., previously reproduced intervals) on timing behavior.
Silencing MEC excitation hinders waiting time-based prediction of trial outcome
To complete the WfT task, animals must compare the perceived elapsed time with the memorized reference of target time, but also continually update this reference based on trial-by-trial feedback. This sort of adaptive behavior is defined by the influence of the past on future behavior, and by the presence of predictive information in each individual trial about the subsequent ones. Previous work suggests that such an influence is limited in time, with more recent trials having a greater influence than earlier ones (Taatgen and van Rijn, 2011). Thus, having found that trial history is associated with current waiting behavior, we tested the hypothesis that partially silencing MEC would cancel the predictive power of the putatively most influential trial, the immediately preceding one (Trialn-1), over the next one (Trialn). To quantitatively assess this, we used logistic regression to decode trial correctness (correct vs incorrect) from Trialn-1 waiting times and treatments (CNO or SAL). We found that a model including both Trialn-1 and treatment (CNO or SAL) significantly predicted trial correctness (Fig. 6A; deviance = 1725.14; χ2(1321, N = 1325) = 101, p = 7.96 × 10−22). Furthermore, predictions of trial correctness, based on the waiting time of the preceding trial, were less accurate in the CNO than in the SAL condition (treatment × Trialn-1 interaction, p = 2.43 × 10−4), with the model performing essentially at chance level under the former [Fig. 6A, right; Table 2; SAL area under the curve (AUC) = 0.61, 95% CI, 0.56, 0.65; CNO AUC = 0.56, 95% CI, 0.51, 0.60]. Silencing the MEC decoupled the current trial outcome from the Trialn-1 waiting time, putatively the most informative one, suggesting a manipulation of the memorized reference of the target interval.
Confusion matrices from trial classification of Trialn-1
Trial outcome is predicted by the immediately preceding waiting times and by the difference between current and previous waiting times. A, The probability of Trialn (left) being correct given the waiting time of Trialn-1 (shading, 95% confidence interval). Colored ticks represent trial ground truth [true correct (top) and true incorrect (bottom)]. The receiver operating characteristic curve (right) of the previous trial time classification model as a function of condition (shading, 95% bootstrap confidence interval). B, Waiting time differences of Trialn – Trialn-1 versus the probability of the current trial being a correct trial (left; shading, 95% confidence interval). Colored ticks represent trial ground truth [true correct (top) or true incorrect (bottom)]. The receiver operating characteristic curve (right) of trial differences classification model as a function of condition (shading, 95% bootstrap confidence interval).
Next, to test whether updating the reference waiting time across adjacent trials could predict trial correctness, we quantified the difference between adjacent waiting times and used such difference as predictor of success. When using the difference Trialn – Trialn-1 to decode trial correctness, the model performed better (Fig. 6B: deviance = 969.72; χ2(1291, N = 1295) = 804, p = 4.68 × 10−174), something we anticipated because of the implicit inclusion of current trial information by differencing. Yet, surprisingly, the model performed best in differentiating correct from incorrect trials for CNO sessions (Fig. 6B, right; SAL AUC = 0.79, 95% CI, 0.76, 0.83; CNO AUC = 0.84, 95% CI, 0.81, 0.88; condition × Trialn-1 interaction, p = 7.04 × 10−05). In addition to decreasing the percentage of correct trials, silencing MEC excitatory neurons increased the contrast between correct and incorrect trials, with correct trials (p correct) following larger trial time differences (Fig. 6B, left: Trialn – Trialn-1 intercepting p correct at 0.5; the detection threshold moved from the origin, at ∼0 s, to ∼750 ms). Moreover, silencing MEC steepened the p correct versus Trialn – Trialn-1 curve, with a sharper transition between low and high probability of correct trials. Sensitivity (SAL = 0.74; CNO = 0.65; Table 3) and specificity (SAL = 0.70; CNO = 0.90; Table 3) values show an increase in true negatives at the cost of true positives because of such a detection threshold shift (SAL = 30 ms; CNO = 750 ms). Thus, the predictive power of subtle intertrial differences observed in control conditions, allowing for continuous adjustment of behavior, is lost when MEC is partially silenced, remaining only for large intertrial differences. This is consistent with a disconnection between current trial accuracy and subtle deviations from target occurring in the immediately preceding trial.
Confusion matrices from trial classification of trial difference (Trialn – Trialn-1)
Discussion
Through chemogenetic interrogation of the MEC–HIPP memory circuit in freely moving rats, we show that MEC excitatory neural activity is necessary for accurate reproduction of a memorized interval and is involved in the effects of temporal context in goal-directed timing, possibly by holding temporal information in memory across trials. Using rats trained in an explicit goal-directed timing task, we show that MEC is necessary for accurate online estimation of time. Interestingly, MEC time cells were previously found to correlate with elapsed time during immobility when animals needed to report an interval by waiting (Heys and Dombeck, 2018). This earlier study showed that sequences of putatively excitatory MEC time cells progressed slower or faster when mice reported longer or shorter intervals, respectively. The presence of sequences of variable duration suggests the existence of a code for consecutive moments of elapsed time, regardless of lapse duration. What role, if any, these neurons play in interval timing remains largely unknown. Our work clarifies this question, as we preferentially silence excitatory neurons, likely including MEC time cells, and find significantly reduced timing accuracy when freely moving rats needed to reproduce a memorized interval (i.e., WfT task). Our results expand recent findings on the involvement of MEC neurons in learning timing (Heys et al., 2020) and show that continuous silencing of excitatory MEC neurons impairs timing in a well learned task, presumably by disrupting the reference memory of the target (see below). The neural manipulation technique used in the study by Heys et al. (2020) acts briefly across a subpopulation of nonspecific MEC neurons, precisely during the few seconds the animal recalls a previously learned task, which shows it to be robust to such manipulation, provided its learning was left intact. During learning, however, optogenetics effectively disturbed progressive performance increases. In our study, we drive the expression of inhibitory DREADDS with the CaMKIIa promoter to preferentially silence excitatory MEC neurons across 10 d of repeated sessions of timing behavior. We observe a significant reduction in spiking activity in MEC and HIPP, sustained in time, likely resulting from widespread silencing mostly of excitatory neurons, and an overall imbalance favoring the suppression of MEC–HIPP excitatory inputs.
Experimental differences aside, head-fixed versus freely moving rats versus mice, our manipulation results in thorough, prolonged silencing of MEC activity, disturbing the main source of depolarizing drive to HIPP, and affects excitatory–inhibitory balance. Furthermore, Heys et al. (2020) silence 100% of all trials, partially and transiently, during learning, but only 20% of postlearning trials. We are convinced that our silencing procedure was less time specific, but was more effective in blocking MEC–HIPP transmission throughout the whole behavioral testing session, including intertrial periods. This sustained silencing likely explains why we observe decreased performance after learning.
Our observed timing behavior impairment consisted in overestimating time [e.g., perceiving more elapsed time than that which is measured by a clock (i.e., veridical time)]. This bias is consistent with well established findings across a range of timing tasks in rodents with lesioned HIPP (Bannerman et al., 1999) and epilepsy patients with medial-temporal lobe resections (for review, see Meck et al., 2013). The magnitude of the bias we observed, 10–15%, is in line with that reported in previous work relating hippocampal damage with prospective memory for time (MacDonald, 2014). In other words, these findings suggest that, like damage to the HIPP, silencing MEC impairs memory-dependent timing behavior.
One possible explanation for the overestimation we observed is the speeding up of a putative “clock mechanism” because of the partial silencing of MEC excitation. This hypothesis would be explained by pacemaker accumulator models (Treisman, 1963; Gibbon, 1977), a long-standing explanation for the instantiation of time perception, which proposes that a central clock emits pulses that are accumulated and compared with an internal reference of a given target interval. In this scenario, if the MEC were part of the clock mechanism, silencing its neurons would result in lower accumulation of pulses and thus longer waiting time. However, our data show the opposite: decreasing firing rates in MEC caused animals to produce shorter waiting times. These results suggest that, rather than taking part in a clock mechanism, MEC might instead hold a reference of the target in memory across trials (Meck et al., 1984). Interestingly, when animals are asked to hold a memory across a delay, silencing MEC impairs memory performance and CA1 time cells are disturbed (Robinson et al., 2017). However, spatial firing patterns, recorded on the return arm of the maze are preserved, suggesting that MEC inputs to the HIPP are necessary for bridging delays (Sauvage et al., 2010; Suh et al., 2011; Kitamura et al., 2014) and specifically drive time cell patterns of activity. Moreover, like previous results, this study shows that time cell activity is related to memory performance (Pastalkova et al., 2008; MacDonald et al., 2013; but see Salz et al., 2016).
Another possible explanation could be that silencing MEC led to an increase in activity levels or impulsivity. To test this hypothesis, we analyzed the speed profiles of small subthreshold movements during the waiting period (i.e., 2.5 s before exit detection), as well as the running speed during the track traversals toward reward, and found no significant differences between conditions during either period (Fig. 7A–C). Likewise, saline and CNO sessions were comparable in total duration and the number of performed trials (Fig. 7D,E). Together, these results strongly argue against possible increases in activity levels (Teitelbaum and Milner, 1963) or impulsivity (Rawlins et al., 1985; Mariano et al., 2009), which would imply a specific increment of very short waiting times, increased movements or jitter during waiting, higher speed during track traversals, or a higher number of trials, all of which are notably absent from our results.
Silencing MEC does not change behavioral measures of activity and impulsivity. A, Speed profile of all trials from all animals in the hM4D group measures from time of exit (0 s) to 2.5 s prior. Solid lines represent the mean speed, and colored areas represent SDs. B, Mean speed inside the ROI per session [same period as before (−2.5 to 0 s); Mann–Whitney, Z = −1.81, p = 0.07]. C, Mean speed along the linear track (outbound) per session (Mann–Whitney test, Z = 0, p = 1). D, Mean session duration (Wilcoxon signed-rank test: Z = −1.61, p = 0.11). E, Number of trials per session (Wilcoxon signed-rank test: Z = 1.26, p = 0.23). Note: for animal 1 of the hM4Di group, the position data and videos of the first three sessions were lost because of file corruption. As such, the data of two CNO sessions and one saline session were not included in ROI and track speed analyses, and an unpaired test was used. n.s., Not significant.
Rather, the results we observe suggest that MEC excitatory activity is critical for holding temporal information in memory, and thus maintaining a continually updated representation of target time across trials (Vierordt, 1868; Jazayeri and Shadlen, 2010; Taatgen and van Rijn, 2011; Shi et al., 2013; van Rijn, 2016; de Jong et al., 2021). Indeed, when we ran trial history analysis, we observed that preceding trials influence interval reproduction in the subsequent trial. Silencing MEC disrupts this effect, breaking the cross-trial influence we observed under control conditions, which suggests that MEC excitatory neurons are important for maintaining this information in memory over trials. Others have suggested that trial history, reflecting the influence of local context, is limited in time, with more recent trials having a greater influence than earlier ones over current behavior (Taatgen and van Rijn, 2011). Indeed, we found that the duration of the putatively most influential trial, the immediately preceding one (Trialn-1), accurately predicted behavioral adaptation and consequently the outcome of the subsequent one (Trialn). Importantly, silencing MEC suppressed this effect of local context by decoupling the duration of the preceding trial from current behavior, suggesting a manipulation of duration memory content. Accordingly, under silenced MEC, only large differences between consecutive trials carried significant performance decoding power, pointing toward an impairment in fine trial-to-trial behavioral adjustments (Fig. 6). It is possible that silencing MEC impaired the dynamic updating of the target reference memory based on the waiting time of the previous trial and corresponding feedback (i.e., rewarded vs not rewarded), thus compromising behavioral performance. This supports the existence of adaptive updating of target representation wherein, under control conditions, pressure to obtain reward would lower waiting times by suppressing above-target durations, whereas the presence of subtarget, nonrewarded, trials (errors) would drive waiting times in the supratarget direction to guarantee reward. Notably, MEC is known to receive important projections from brain areas related to reward processing, namely the amygdala (Stefanacci et al., 1996; Pikkarainen et al., 2002; Tomás Pereira et al., 2016) and orbitofrontal cortex (Kondo and Witter, 2014), which could support this reward-driven updating of the reference target interval. Consistent with this updating mechanism, we observed that, under control conditions, correct trials were surrounded (i.e., preceded and succeeded) by shorter waiting times, and incorrect trials by longer waiting times. Yet, when MEC was silenced, the preceding and succeeding times of correct trials were even shorter, resulting in an increased difference between correct trials and the surrounding ones. Likewise, for incorrect trials the preceding and succeeding waiting times were also lower, suggestive of a significant impairment of the dynamic goal-directed adjustment of timing behavior in response to reward.
At the population level, two putative mechanisms could potentially support the maintenance of temporal information from one trial to the next one. First, sequences of time cells (Kraus et al., 2015; Heys and Dombeck, 2018) are optimal to convey temporal information to downstream structures (Zhou et al., 2020) and, in principle, to hold temporal information in their ordered pattern of activity. Thus, silencing MEC time cells would not only compromise local sequential activity but also the readout by downstream structures such as HIPP (Kraus et al., 2013; but see Sabariego et al., 2019). Furthermore, a recent study showed that CA1 time cells are active during online estimation of elapsed time (Shimbo et al., 2021), supporting the idea of their involvement in timing.
Second, sustained neural activity over the waiting period could putatively keep temporal information online (Egorov et al., 2002; Frank and Brown, 2003; Zylberberg and Strowbridge, 2017). MEC layer III inputs to HIPP seem to be an ideal candidate for such a mechanism as they are known to be critical for holding information across delays (Suh et al., 2011). In vitro studies showed the existence of persistent spiking activity in MEC layer III (Yoshida et al., 2008), suggesting that these neurons could, in principle, maintain information over extended periods and be impaired in doing so by MEC silencing via disruption of spiking activity, impaired plasticity in output pathways (Remondes and Schuman, 2002), and defective consolidation of learned information (Remondes and Schuman, 2004). Moreover, silencing these neurons in MEC would also impact time cell activity in the HIPP (Schlesiger et al., 2015; Robinson et al., 2017; but see Sabariego et al., 2019) and the encoding of time. Recent theoretical work supports this idea by showing that time cells can be generated by exponentially decaying inputs (Liu et al., 2019).
Overall, through different processes the MEC can maintain a reference memory of the target interval across trials, and this memory can be continuously updated and used to adjust ongoing behavior (Bausenhart et al., 2014). Independently of the implementation, we propose that the MEC is involved in holding the target interval in memory, and that such information is conveyed to downstream structures, including the CA1, where it might be compared with the ongoing estimation of time.
Footnotes
This work was funded by national funds through the FCT-Fundação para a Ciência e a Tecnologia, I.P. under the following Grants UIDB/50005/2020, Research Grant PTDC/MED-NEU/29325/2017, a PhD fellowship to M.D. (PD/PB/114125/2015), and an Exploratory Grant (IF/00201/2013) to M.R., who held an Investigator FCT Position at iMM (IF/00201/2013). We thank Bioimaging and Rodent facilities at iMM for critical help. We also thank all the members of the Remondes laboratory for fruitful discussions and help. In addition, we thank Inês Marques-Morgado for assistance with histology and immunohistochemistry.
The authors declare no competing financial interests.
- Correspondence should be addressed to Miguel Remondes at amcpremondes{at}gmail.com or Marcelo Dias at marcelofvdias{at}gmail.com
