High-Throughput Task to Study Memory Recall During Spatial Navigation in Rodents

Abstract

Spatial navigation is one of the most frequently used behavioral paradigms to study memory formation in rodents. Commonly used tasks to study memory are labor-intensive, preventing the simultaneous testing of multiple animals with the tendency to yield a low number of trials, curtailing the statistical power. Moreover, they are not tailored to be combined with neurophysiology recordings because they are not based on overt stereotyped behavioral responses that can be precisely timed. Here we present a novel task to study long-term memory formation and recall during spatial navigation. The task consists of learning sessions during which mice need to find the rewarding port that changes from day to day. Hours after learning, there is a recall session during which mice search for the location of the memorized rewarding port. During the recall sessions, the animals repeatedly poke the remembered port over many trials (up to ∼20) without receiving a reward (i.e., no positive feedback) as a readout of memory. In this task, mice show memory of port locations learned on up to three previous days. This eight-port maze task requires minimal human intervention, allowing for simultaneous and unsupervised testing of several mice in parallel, yielding a high number of recall trials per session over many days, and compatible with recordings of neural activity.

Keywords: correlation between neuronal activity and behavior; data output for machine-learning algorithms analysis tools; freely-moving calcium imaging recordings; high-throughput experimentation; single-session memory test; spatial navigation and memory.

FIGURE 1

Trial structure of the spatial memory within a learning session. (A) Schematic of the open field built-in transparent acrylic with eight water spouts (ports) controlled by Python-based software coupled to Arduino boards. The arena is placed inside a 1 cubic meter sound isolation chamber with cards placed on the walls serving as distal visual cues for the animals. (B) Top: Schematic of the behavioral protocol, showing handling, pre-training, and actual-experiment phases. Learning and recall sessions are performed on a daily basis until the animal’s motivation persists over days. Bottom: Schematics of two consecutive trials during a learning session of the task. Top traces show the temporal structure of the sound, trigger zone activation, and the reward, along with the pokes in the different ports (colored dots). Bottom diagrams show the trajectory of the animals and the location of the ports in the maze. Ports are colored according to the distance to the correct port (black port at E). In the first part of each trial, animals walk around the box seeking to start the reward time window during which water is available (black trajectory in bottom panels). The reward time window lasting for a maximum of 6 s, starts when the animal steps into an invisible trigger zone, randomly placed in a different location on every trial (small violet circle in bottom diagrams). During this time window, a sound is played cueing the mouse about the availability of reward (green trajectory). If the animal pokes the correct port, the sound cue stops. If the animal does not reach the correct port during this window, perhaps because it poked in a non-rewarding port, the trial is considered incorrect (trial i−1 in B). In contrast, if the animal reaches the correct port, independently of whether it poked in incorrect ports before (typically close to the correct port), it receives the reward (10 μL of water), the sound stops, and the trial is considered correct (trial i in B). The correct port is fixed for each day (learning and recall sessions) but changes randomly from day to day. (C) Performance (i.e., correct over total trials) versus the trial number in learning sessions. All measures in (C–F) represent averages across all animals and sessions (n = 23 mice). Shaded areas represent a 95% confidence interval of the SEM. The performance in the first trial was higher than 1/8 because animals poked in multiple ports during the tone (E), setting the probability to hit the correct port on the first trial above 40%. (D) Response time during learning sessions. Response time was defined as the interval from reward time window onset to nose poke in the correct port in three different conditions: for correct trials only (poke correct port before 6 s, blue), for correct and incorrect trials (magenta) (Note: incorrect trials response time was set to 6 s), and total time to reach the correct port (orange) even after the 6 s sound cue window. (E) The number of errors during the sound cue is the number of ports poked before poking the correct port in learning sessions. Errors decrease with the trial number. (F) Time to find trigger-zone in learning sessions. After the reward time window is finished, we computed the time from this moment until the animal hits the randomly placed trigger zone. We show here that animals improve navigation toward the correct port within each session.

FIGURE 2

Mouse behavior during the learning sessions. (A) Top: raster plot of pokes during an example learning session shows the timing of pokes at different ports (colored dots matching the port color code) ordered in time by the sound cue onset on each trial (vertical dashed line). Dark gray bars represent the sound cue duration (as in Figure 1D magenta line). Light gray bars indicate the duration of the trigger zone seeking phase of the trial (as in Figure 1F). Small dots indicate persistent licking in a given port. In this example session, the animal sought the correct port during the first five trials after which the animal started to accurately find it in almost all consecutive trials. Bottom: pokes cue-triggered time histograms for the correct port (black line) and the average overall incorrect ports (gray line). The time bin for the poke rate is 0.5 s. (B) Trajectories during four example trials. The color code of pokes and trajectory parts are the same as in Figure 1B. Trials at the beginning of the session were more explorative (Example 1) until the animal unequivocally identified the correct port location (e.g., in trials 6–9 in this example). From that point on, trajectories were either directed toward the correct port (Example 2), or with only one error port during the sound cue (Examples 3 and 4). (C) All trajectories from all trials accumulated over the entire learning session. Violet dots represent the position of the animal at sound cue onset. (D) Left: cumulative poke count vs. trial number for each individual port during sound cue. The correct port (black line) shows a higher poke count than any other port during the tone. Pokes on different ports are plotted with color code as in (A–C). Dashed lines indicate the counterpart port at equal distance from the correct port. Right: poke histogram during sound cue, for each port index ordered by the distance to the correct port. Magenta dashed line represents the significance level (P < 0.01) over which poking probability was significantly larger than that expected from a uniform distribution. Gray dashed line is the mean value of the uniformly shuffled data.

FIGURE 3

Mouse behavior during the memory recall sessions. (A) Pokes raster plot during an example memory recall session performed by mouse 4032, displayed as explained in Figure 2A. During recall sessions, correct pokes during the first n minutes of the session (n = 0, 1, 3, or 5) were not rewarded, to avoid re-learning of the reward location. Sound cues were never interrupted during this period. After the n minutes, correct pokes were rewarded as during the learning session (blue rectangle). In this example session, during the first n = 5 min, corresponding to trials 1–16, water was not available, whereas in trials 17–31 water was available. (B,C) All trajectories from all trials accumulated over the recall session for the period without water (F) and water (G). Representation is equivalent to that shown in Figure 2C. Cumulative poke count versus trial number for each port during sound cue (D) and sound cue off (E). Correct port (black) and one neighbor port (dashed brown) show higher poke count in both cue on and cue off conditions during the period without water. Once the water was available and harvested (blue area), the rewarded port was mostly poked. Poke histograms in polar coordinates for the example session (black dots), showing the number of pokes in each port as the radial distance of each point to the center, for the cue on (F) and cue off (G) conditions. Colored squares mark the angle of each port (as in Figure 1B). Small gray numbers show the count number of the radius of the inner and outer circles. The dashed magenta line shows the pointwise significance bound obtained from shuffled data (P < 0.01, one-tailed; see section “Methods”). Black arrows show the vector summation of the counts over all the ports. Red lines show the MI for the session relative to the 2 h correct port. (H) Histogram of MIs obtained from individual recall sessions of mouse 4032 (black bars). Gray histogram shows the MI values from the surrogate data set drawn from a uniform poking distribution. The magenta vertical line shows the significance bound of individual session MI (P < 0.01). The yellow arrow marks the session shown in (A–G). (I) Accumulated poke histogram for all sessions (n = 127) of mouse 4032 considering both on and off cue conditions. Magenta dashed line indicates significance bound (P < 0.01, one-tailed). The red line shows the MI over all sessions. (J) 2 h average MI versus trial index during the recall session. Despite the absence of water during these trials, MI was significantly larger than zero over the first 16 trials, illustrating the persistence of the animal to recall the rewarding port.

FIGURE 4

Memory recall from previous sessions. Normalized poke histograms aligned at the port learned 2 h (A), 24 h (B), 48 h (C), and 72 h (D) before (black line + dots) show a significant preference to poke in the ports memorized in the three previous days (P < 0.01, permutation test). Histograms were the average across animals (n = 23 mice), each with a different number of sessions (mean no. sessions 55; range: 11–127). Lower insets show the same poke histograms unfolded for finer visualization. The red dashed line is the top part of the confidence interval of the surrogate data generated with the same 2 h memory strength contained in the real data. (E) Corrected memory index MI (MI – mean surrogate shuffles) versus the session lag (black dots, n = 23 mice) shows that there was significant memory recall of the correct ports from the four previous sessions (i.e., up to 72 h). Significance was assessed generating a surrogate data set with only 2 h memory that followed the same sequence of rewarded ports across days (mean across surrogates is shown in gray). Results from control (no injection) and saline injections sessions are combined here as they show no difference. (F) Averaged corrected MI for 2 and 24 h memories. Consecutive sessions were grouped in three conditions: (i) sessions with no CPP (black line), (ii) sessions with CPP injected only on the testing day (dark green line), and (iii) sessions with CPP injected only on the previous day (light green line). CPP decreased the MI only on the day that it was injected: MI decreased in lag 0 when CPP was injected on the test day (P < 0.05, one tail t-test) and in lag 1 when it was injected on the previous day (P = 0.08, one tail t-test). Lag and CPP conditions showed a significant interaction (P < 0.022, two-way ANOVA). Error bars are the 95% confidence interval.

FIGURE 5

Calcium imaging recordings while animals perform the task. (A) Example frame calcium imaging registration of the cells recorded simultaneously while the animal is performing the task (approximately 180 neurons recorded). (B) Calcium activity traces of relative fluorescence (ΔF/F_o) recorded from selected neurons (colored in B). Traces were processed off-line and used to decode different behavioral tasks (e.g., animal’s position). (C) Performance measure in interleaved sessions with and without mini-microscope. The performance versus trial number does not show a significant difference if we compare the conditions with the animal carrying the mini-microscope (P = 0.5, one-tailed t-test). Shaded areas are the mean’s 99% conf. interv. (D) Example place-fields obtained from two different animals while performing the task. Place-fields were chosen from the distribution of spatial information measured by the entropy of the place-fields neuronal tuning (Markus et al., 1994). All recorded place-cells have entropies higher than 2.5 bits/sec.