Optogenetic perturbation of projections from thalamic nucleus reuniens to hippocampus disrupts spatial working memory retrieval more than encoding

Abstract

Background:

Working memory deficits are key cognitive symptoms of schizophrenia. Elevated delta oscillations, which are uniquely associated with the presence of the illness, may be the proximal cause of these deficits. Spatial working memory (SWM) is impaired by elevated delta oscillations projecting from thalamic nucleus reuniens (RE) to the hippocampus (HPC); these findings imply a role of the RE-HPC circuit in working memory deficits in schizophrenia, but questions remain as to whether the affected process is the encoding of working memory, recall, or both. Here, we answered this question by optogenetically inducing delta oscillations in the HPC terminals of RE axons in mice during either the encoding or retrieval phase (or both) of an SWM task.

Methods:

We transduced cells in RE to express channelrhodopsin-2 through bilateral injection of adeno-associated virus, and bilaterally implanted optical fibers dorsal to the hippocampus (HPC). While mice performed a spatial memory task on a Y-maze, the RE-HPC projections were optogenetically stimulated at delta frequency during distinct phases of the task.

Results:

Full-trial stimulation successfully impaired SWM performance, replicating the results of the previous study in a mouse model. Task-phase-specific stimulation significantly impaired performance during retrieval but not encoding.

Conclusions:

Our results indicate that perturbations in the RE-HPC circuit specifically impair the retrieval phase of working memory. This finding supports the hypothesis that abnormal delta frequency bursting in the thalamus could have a causal role in producing the WM deficits seen in schizophrenia.

Introduction

Working memory requires coordinated action between the hippocampus (HPC) and prefrontal cortex (PFC) [1]. While there is a direct pathway for information flow from HPC to PFC, bidirectional connectivity is facilitated through the thalamic nucleus reuniens (RE), which has dense reciprocal projections to both mPFC and the CA1 region of HPC [2–4]. During performance of spatial working memory (SWM) tasks, RE responses contain future trajectory information, and silencing RE reduces the magnitude of trajectory-specific information in HPC [5]. Chemical or optogenetic silencing of RE during SWM tasks also greatly reduces performance [6–8], suggesting a key role for RE in working memory.

Patients diagnosed with schizophrenia (SZ) commonly present with a loss of functional connectivity between HPC and PFC as well as impairments in working memory and executive function – facts which imply that dysfunction in the PFC-RE-HPC circuit might be a specific substrate for SZ pathogenesis[9–11]. Human studies of SZ have focused on cortical oscillatory activity via EEG and MEG. While many studies note disruptions in high-frequency gamma band (30–80 Hz) activity in SZ patients [12–14], meta-analyses across studies showed these abnormalities are variable in their direction (i.e. increases vs decreases in spectral power) and are not unique to the diseased state – also being present in clinically high-risk patients, similar psychiatric disorders and first-degree relatives of SZ patients. However, SZ patients also have an abnormal elevation of low-frequency delta (1–5Hz) oscillations in temporal and frontal areas, and this abnormality is unique to the diseased state of SZ and seen even in first psychotic episode patients [15–17].

Further research into the HPC-PFC dysfunction in SZ has come from studies using the NMDA-R hypofunction model of SZ, which is based on the observation that healthy subjects develop both positive and negative symptoms of SZ when exposed to an NMDA-R antagonist such as ketamine [18]. In rats, these studies have found that ketamine administration induces delta frequency bursting in RE and delta frequency oscillations in HPC [19, 20]. However, ketamine has many effects other than just NMDA-R antagonism and affects many systems in the body and brain. To address this, a prior study from our lab used optogenetic experiments to confirm a causal role of thalamic delta oscillations in working memory disruption [21]. They found that delta frequency stimulation of RE terminals in CA1 was sufficient to significantly impair performance of rats on a delayed non-match to sample (DNMS) task.

Exactly how delta oscillations impair working memory performance is still not well understood. The DNMS task involves distinct phases (Figure 1D), each of which require distinct processes for successful completion of the task. First there is the ‘sample phase’ in which an animal is given access to a single reward arm and must encode the memory of which arm was sampled. After a brief delay where the animal is held at the home port, the ‘choice phase’ begins in which the animal must recall the sample arm and enter the opposite arm in order to receive a reward. To investigate which memory process is impaired by our manipulation, we trained mice to complete the DNMS task on a Y-maze and selectively induced delta oscillations in projections from RE to CA1 during distinct phases of the task.

Figure 1. Overview of experimental setup.

(A) Optogenetic construct hCh2R-eYFP under the control of a CamKII promoter was encapsulated in a pAAV5 viral vector. Depiction of (B) virus injection site in RE and (C) optical fiber implant target above dorsal CA1. (D) Representative histological slice (~−2.4 mm AP) showing virus expression with an expanded window showing expression in hippocampal CA1 and the optical fiber tip (red arrow). (E) Schema of a single trial of the DNMS Y-maze task; green, reward locations; red, error location. Optogenetic stimulation occurred either over the full trial or specifically during the encoding or retrieval phases indicated above.

Materials and Methods

Animals.

Adult C57Bl/6J mice (5 female and 1 male), age 4–11 months, were communally housed in a temperature- and humidity-controlled environment on a 12-hour light/dark cycle. During training and testing, animals were water restricted to maintain 85% ad libitum body weight and given ad libitum access to food. All procedures were performed according to the guidelines of the Institutional Animal Care and Use Committee at Brandeis University.

Surgery.

Mice were anesthetized with an intraperitoneal injection of a ketamine/medetomidine cocktail (K/M, 50/0.5, mg/kg, IP) and mounted on a stereotaxic frame. After the skull was exposed and cleaned, craniotomies were made to allow for virus injection into RE (0.9 mm posterior to bregma, ± 1.25 mm lateral to midline, 4.28 & 4.08 mm below the brain surface @ 15° from vertical) and for optical fiber implants over CA1 (2.3 mm posterior to bregma, ± 2.0 mm lateral to midline, 1 mm below the brain surface). Using a calibrated glass pipette (VWR, Radnor, PA) attached to a stereotactic injector (Stoelting Co., Kiel, WI), adeno-associated virus (AAV) serotype 5 containing the gene hChR2(H134R)-EYFP (Figure 1A) mediated by a CaMKIIα promoter (pAAV5-CaMKIIα-hChR2(H134R)-EYFP) (Addgene, Watertown, MA) was bilaterally injected into the RE (Figure 1B), at 200 nL per site. Two 200-micron optical fibers (Thorlabs, Newton, NJ) were then implanted above CA1 and secured using Vitrebond (3M, St. Paul, MN), C&B-Metabond (Parkell, Edgewood, NY) and dental acrylic. Following surgery, the mice were injected with the antibiotic penicillin, the analgesic meloxicam, and the anti-sedative atimapezole.

Histology.

After experiments, the subjects were perfused and their brains stored in 4% paraformaldehyde solution for 24 hrs at 4°C. 120-micron sections were sliced on a model VT1000 S Leica microtome. Slices were mounted with DAPI mounting solution and imaged under a fluorescent microscope to confirm expression of ChR2/EYFP in RE and CA1 and the location of the optical fiber lesions above CA1.

Behavior.

After a minimum two-week recovery period following surgery, the animals were water-deprived to a minimum of 85% of their original weight. The custom-built Y-maze, used to execute the delayed alternation SWM task, had three arms with sensor-activated water dispensers, which were gated using motorized, automatically-operated doors. The Y-maze was connected to a Bpod state machine (SanWorks, Stony Brook, NY) and trial events were triggered through Matlab (Mathworks, Natick, MA). 10 μL of water were dispensed per reward. The mice were habituated to the maze over 4 days, after which training and then testing occurred. On day 1 of habituation, two doors were permanently down, and mice were allowed to freely navigate between two water ports—home and either the left or right arm—until 200 μL of water were collected. On days 2–4, mice were given a forced alternation task, where the left and right arms were opened and closed in alternation until 200 μL of water were collected. Training consisted of 10 or 15 trials per day, with each trial consisting of a sample phase, a 10-second delay at the home port and a choice phase. In the sample phase, only a single path from home to reward port was left open. After obtaining the reward and returning home, animals were locked in the home location for the delay. In the choice phase, all doors were opened and the animals needed to navigate to the arm opposite the sample to obtain a reward. A 40-second wait separated all trials. The side of the correct choice was randomized, and 3 consecutive days of ≥80% accuracy were needed to move on to testing.

Testing spanned 4 days and followed the same Y-maze protocol as training with 10 or 15 trials per day; the number of trials per day was constant throughout each testing period, but as observed in Duan et al. [21], animals learned the task a varying speed and some needed an increase to 15 trials per day during training in order to learn within a reasonable time frame. On days 1 and 3 of testing, laser stimulation (473 nm wavelength) was applied to RE terminals in CA1, according to one of three protocols: full-trial, encoding, or retrieval. Days 2 and 4 were used as control days where either no stimulation was given or stimulation was done using a 554 nm MINTF4 LED so as not to activate ChR2. Each 4-day test run consisted of only a single experimental stimulation condition (Full Trial, Encoding or Retrieval) and a single control condition (No Stimulation or Control stimulation).

RE-HPC projections were illuminated with either a 473 nm laser (experimental) or a 554 nm MINTF4 LED (control) via two patch cords attached to the implanted optical fibers. Optical fiber irradiances were 27.84 mW/mm² and 25.45 mW/mm² at the optical fiber tip for the laser and LED respectively. This gave estimated irradiances of 5.45 mW/mm² and 6.21 mW/mm², respectively, at 200 microns from tip through mammalian tissue. All stimulation occurred at 3 Hz flashing frequency (100 ms on, 233 ms off) roughly matching the frequency of oscillations seen in the RE local field potential generated by systemic ketamine injection [20]. Full-trial stimulation involved light stimulation persisting from trial start to trial end. Encoding stimulation was restricted to just the sample phase. Since we used a relatively short delay and could not determine exactly when a choice was being made, retrieval stimulation began at the start of the delay and persisted throughout the choice phase. No stimulation was ever given during the inter-trial interval (40 s) in order to protect the brain from potential photo-damage. Single animals were often tested under multiple testing protocols, in which case a rest period of ≥5 days was provided before the mouse was returned to the training stage in preparation for the next protocol. To account for order effects, we counterbalanced the order in which each mouse went through the different stimulation protocols; though not all mice completed all protocols. Training and re-training time varied greatly between animals.

Statistical Analysis

Data were analyzed using the SciPy, statsmodels and Pingouin python libraries. All comparisons were done with the non-parametric Kruskal-Wallis one-way analysis of variance (KW test) and the Games-Howell (GH) post-hoc test (if necessary), except where otherwise noted. Control groups were first compared, and finding no significant differences between control conditions, we then aggregated control data and compared experimental groups and the aggregated control. In order to mitigate animal differences in task performance, we normalized accuracy relative to mean control accuracy during each test run. We also tested for differences in animals’ performance during control sessions, and animal time latencies from start-to-sample (during the Sample phase) and start-to-goal (during the Choice phase). The differences between consecutive days of testing (Figure 3) under each stimulation condition were compared using the Mann-Whitney U-test and Bonferroni correction.

Figure 3.

(left) Accuracy relative to mean control accuracy over experiment days depicting training (Days −3 to −1), experimental stimulation days (Days 1 and 3) and control days (Days 2 and 4). Control data are averaged over no stimulation and control stimulation sessions. Consecutive days were compared with the Mann-Whitney U-Test and Bonferroni corrected. Only the Retrieval Stimulation condition showed significant differences from day-to-day [*p<0.05]. (right) Trial schema showing stimulation period on experimental and control stimulation sessions.

Results

Six mice (5 female and 1 male) were used in these experiments. Mice were bilaterally injected with rAAV5-CaMKIIa-hChR2(H134R)-eYFP targeting RE (Figure 1A&B). The virus primarily transduced RE with some overflow into nearby thalamic nuclei. However, this overflow was of no consequence (see below). Optical fibers were then implanted bilaterally, with tips targeting the region just dorsal to the hippocampal CA1 field. Histology confirmed that, by the time of experiments, virus had visibly transfected RE cells and axons, causing ChR2 and eYFP expression in the stratum lacunosum moleculare of CA1. Since RE is the only thalamic nucleus that innervates this hippocampal region[2–4], we can be assured that our fibers were selectively exciting RE axons terminating in HPC.

Once animals achieved ≥80% performance on the DNMS task for 3 consecutive days, testing began. Each testing period consisted of 4 days of alternating experimental and control sessions. On experimental days, animals had to complete the DNMS task while RE terminals were optogenetically activated at delta frequency (3Hz) either throughout the entire trial, just during encoding, or just during retrieval. During control days, animals either had no light stimulation or the same stimulation as during testing except with a yellow-green LED – which would not activate ChR2 – in order to see if the light alone had any effect on performance. Each testing period only consisted of one experimental and one control condition, and between testing periods animals were given a break and then re-trained.

We first assessed whether the light had any effect on performance by comparing training (Days −3 to −1) and all of the control conditions (Days 2 & 4): no stimulation, full-trial control stimulation, encoding control stimulation, and retrieval control stimulation. A Kruskal-Wallis test (p=0.0046, H=15.0432) followed by Games-Howell post-hoc test revealed that when compared to the last three data of training, the no stimulus (p=0.0020) condition showed significantly lower performance. However, none of the control conditions were significantly different from each other (all p>0.05; Figure 2A). For this reason and because of the low number of control samples compared to experimental, we combined all control conditions for further analysis. Comparison of control sessions between animals (KW Test: p=0.0030, H=17.92) revealed that one animal had significantly lower performance during control sessions than several other animals, and this animal’s performance was responsible for the difference in performance between the Training and No Stimulation conditions. However, all animals were found to have consistent performance during control sessions over time (KW Test: all p > 0.39).

Figure 2.

(A) Percent correct trials in various control conditions. Training (gray) indicates Days −3 to −1 and the rest represent Days 2 & 4. Compared with Kruskal-Wallis H-test (p = 0.0046; H = 15.0432) and Games-Howell post-hoc test (**p < 0.01). (B) Percent correct relative to mean control accuracy for each test run. Comparison is between all control conditions (excluding training; orange) and the various experimental stimulation conditions (blues). Kruskal-Wallis (p = 1.097e-5; H = 25.7089) with Games-Howell post-hoc (*p<0.05; **p<0.01). N: number of sessions under each condition.

Next, we assessed if and when 3Hz stimulation of RE terminals in CA1 impaired performance (Figure 2B; KW test: p=5.58e-6, H=27.113). To account for animal differences, experimental conditions were compared using accuracy relative to the mean control accuracy during each 4-day testing run. Full-trial stimulation produced results similar to those seen in rats (Duan et al 2015), with performance significantly (p = 0.0085) dropping on experimental days and recovering in the subsequent control sessions (Figure 3, top). Alternatively, encoding-specific stimulation did not significantly alter performance (p = 0.90) with testing days showing little to no change between experimental and control sessions (Figure 3, middle). Retrieval-specific stimulation did significantly decrease (p = 0.0010) performance, entirely re-capitulating the effect of full-trial stimulation (Figure 3, bottom). Furthermore, retrieval stimulation also caused significantly lower performance compared to encoding stimulation (p=0.0010). Together these data show that perturbation of RE inputs to CA1 specifically impairs the memory retrieval phase of SWM task performance.

Lastly, in order to evaluate whether our perturbations impacted other behavioral variables, we looked at run times on correct trials from start to reward during both the Sample phase and the Choice phase. No significant differences were found in run times during both the Sample phase (KW test: p=0.094, H=6.39) and the Choice phase (KW test: p=0.054, H=2.18) for all conditions (Control, Full Trial, Encoding and Retrieval). As such, our perturbations impair task performance without significantly altering animal behavior during the task.

Discussion

Our findings translate the results of Duan et al 2015 from rats to mice, showing that the RE projections to HPC play a critical role in SWM [21]. Further, by optogenetically interfering with the communication between RE and HPC at different phases of SWM processing, we demonstrated that these imposed oscillations specifically disrupt the memory retrieval phase and not the encoding phase during a SWM task.

While RE was previously identified as essential to proper SWM function [5, 7], its role in the specific phases of the memory process are not fully understood [6, 22]. Optogenetic inactivation of RE specifically impairs encoding during a SWM task [6], whereas, in a long-term spatial memory task, muscimol inactivation of RE generated a specific impairment of memory retrieval but not memory consolidation [22]. These findings suggest RE participates in both encoding and retrieval processes, possibly dependent of task setup or timescale. Distinct from direct inactivation of RE, our manipulation is spatially confined to RE inputs in the dorsal HPC and our optogenetic stimulation at delta frequency had a more significant impact on memory retrieval than encoding. Considering that RE projects to both HPC and PFC, the RE projections to each brain region could dynamically coordinate different aspects of working memory. The interrogation of other specific RE projections in the same task will provide a better picture of RE’s role in WM.

It is important to note that our retrieval stimulation protocol consisted of stimulation starting during the delay and continuing throughout the choice phase of the task. This is because our task used a short delay (10 s), and we could not assume exactly when the choice was being made and thus when the memory of the sample was retrieved. As such, it is possible that what is being disrupted is not memory retrieval, but memory maintenance. Some studies have noted that during the delay in WM tasks, decision-relevant memories can be maintained in active firing sequences in PFC [23]. Furthermore, studies in rats have found that HPC and PFC can work independently to process spatial memory at short delays (10 s), but interaction is required for long delays (5 min) [24]. Therefore, our paradigm, with its short delay, should be relatively unperturbed by inactivation of just HPC or RE, but what we observe is that imposition of delta oscillations impairs retrieval despite the short delay. This suggests that somehow our imposed oscillation is in some way disrupting decision-relevant firing in both HPC and PFC, possibly by forcing interaction, erroneous interaction, between HPC and PFC.

This effect may also be explained by a subtler phenomenon. With an intact circuit, HPC-mPFC interaction during SWM is known to be supported by coherent theta (5–10Hz) and gamma (40–80Hz) oscillations with the spiking of units phase-locking to these synchronous rhythms [25]. The synchronous theta rhythm specifically has been found to also exist in RE, and CA1-RE-mPFC theta coherence was found to increase when making route-related decisions [26]. Furthermore, this synchrony was found to be greatly reduced in SWM studies with mice genetically modified to model SZ [27, 28]. From this viewpoint, our findings suggest that the induction of delta oscillations in HPC could be sufficient to impair proper phase-locking to the theta and gamma bands and ruin proper cooperation between PFC and HPC, though further studies are needed to confirm whether HPC-mPFC synchrony is actually being impaired by our stimulation. Also, it still needs to be determined whether this impairment in memory retrieval is specific to the imposition of delta oscillations or if any imposed oscillation would produce the same impairment. These studies could greatly inform how the HPC-PFC-RE circuit works together to support proper working memory.

Highlights.

• The reuniens-hippocampal circuit plays a crucial role in a spatial memory task in mice.

• Delta-frequency stimulation of reuniens inputs to CA1 impaired spatial working memory.

• Perturbations during retrieval, but not encoding, reduced task performance.