Adult male Nestin-CFPnuc mice (Mignone et al., 2004; Encinas et al., 2006; Enikolopov et al., 2015) maintained on C57BL/6J background, were used for all experiments. Prior and during the experiment, mice were housed in groups of 2–4 animals per cage under standard conditions with 12/12 h light–dark cycle, in cages 36х21х13.5 cm, with food and water available ad libitum. All experiments were conducted in compliance with the requirements, regulations, and guidelines issued by the National Institutes of Health and Stony Brook University.

We examined mice of different ages, dividing them into two groups. The first group consisted of mice that were 6 months old at the start of the behavioral training sessions (referred to as the 6MO group, n = 13). The second group consisted of mice that were 14 months old at the start of the behavioral training sessions (referred to as the 14MO group, n = 15). The same animals from the 6MO and 14MO groups were used for all experiments.

To label dividing cells in the brain, experimental animals were injected with 5-ethynyl-2′-deoxyuridine (EdU, 123 mg/kg, Invitrogen, USA) intraperitoneally (Podgorny et al., 2018; Ivanova et al., 2022) 6 weeks before behavioral training. After the injection, all mice were returned to their home cages and housed under conventional conditions before the beginning of behavioral experiments.

Mice were transferred from the animal facility to the experimental room 1 h prior to the beginning of the experimental procedures. The mice were subjected to three consecutive Morris water maze- (MWM)-based tasks: spatial MWM (sMWM), followed by reversal MWM (rMWM), which was followed by context discrimination MWM (cdMWM). For the training in the sMWM (Figure 1), animals were presented with a circular pool (120 cm in diameter) made of blue plastic (Noldus, the Netherlands). The water was made opaque with non-toxic white paint and maintained at 23–24°C. Several distal cues were affixed to the walls surrounding the pool. Each mouse performed five 60 s trials per day, with 30 min inter-trial intervals, for five consecutive days, with the objective to locate a submerged platform positioned 0.5 cm below the water surface in one of the virtual quadrants (the target quadrant, T). At the beginning of each trial, a mouse was released into the pool from one of three virtual quadrants that did not contain the platform. If a mouse failed to find the platform, the experimenter gently guided it to the platform. The mouse was allowed to remain on the platform for 30 s. The platform position remained constant throughout the five training days. On day 6, the mice underwent a spatial memory test involving swimming for 60 s in the pool without the platform.

Training in rMWM (Figure 1) started 2 h after the sMWM memory test. The hidden platform was reintroduced to the pool but was placed in the quadrant opposite to that used during sMWM training. The distal cues on the walls remained the same and were located at the same positions relative to the pool. The mice were given three additional days to learn the new platform location. The overall procedure was identical to that described for sMWM. On day 9, the mice were subjected to a 60 s memory test without the platform.

cdMWM training (Figure 1) began 2 h after the rMWM memory test (day 9). The same blue plastic pool and the same distal cues on the walls were used. The platform was relocated to a new quadrant and two local cues (beacons) – a yellow rubber ball and a multicolor pyramid – were suspended 20 cm above the water surface. One beacon was placed above the platform (the goal cue), while the other was positioned above the opposite quadrant (the false cue). On the following day (day 10), each mouse was placed in a pool identical to the blue pool except it was constructed from white plastic. This pool was located in the adjacent room with different distal cues on the walls. Two beacons identical to those used in the blue pool (a ball and a pyramid) were suspended above the pool, but the goal cue and the false cue were switched, i.e., if the ball indicated the platform location in the blue pool, it served as the false cue in the white pool, while the pyramid became the new goal cue. The assignment of beacons indicating the platform location was counterbalanced between the pools: for half of the animals in each group in a particular pool the goal cue was the ball, whereas for the other half in the same pool the goal cue was the pyramid. Therefore, to choose the correct local cue indicating the platform location, each mouse needed to discriminate among the contextual factors of the room, pool color, and distal cues. Each mouse underwent five daily sessions lasting 60 s each, and after reaching the platform was allowed an additional 30 s on the platform. The pools were alternated daily for six days (day 9 through day 14). On day 15, the mice underwent 60 s memory tests in the blue and white pools, separated by a 2 h interval. The timeline of the tasks is depicted in Figure 1.

Animals’ behavior was recorded with the video camera mounted above the pool. For behavioral tracking, EthoVision XT, versions 8 and 17 (Noldus, the Netherlands) was used, and the following behavioral parameters were extracted from the video: escape latency – time to reach the platform in training trial; fraction of time spent in each quadrant (in %) in the memory test. For the classification analysis of the search strategies, we used the Pathfinder software (Cooke et al., 2019). Raw tracking files with xy coordinates over time were extracted from the EthoVision XT and uploaded to the Pathfinder. The following parameters, empirically determined for our combination of the pool and platform size and location were used: Goal Position (x,y): varied between sMWM, rMWM, cdMWM; Goal Diameter (cm): 10; Maze Diameter (cm): 120; Maze Centre (x,y): 0,0; Angular Corridor Width (degrees): 20; Chaining Angular Width (cm): 15; Thigmotaxis Zone Size (cm): 15. A trial was truncated when the animal reached goal location. As a result, each track was assigned to one of the seven search strategies: direct path, directed search, focal search, indirect search, scanning, random search, thigmotaxis (from the most spatially precise and efficient to the least one). In wild-type healthy rodents, learning leads to a decrease of the fraction of inefficient, spatially imprecise search strategies (presumably, hippocampal-independent) and is accompanied by an increase of the fraction of more efficient, spatially specific, and, presumably, hippocampal-dependent, strategies (indirect search, focal search, directed search, direct path) (Garthe et al., 2009, 2014, 2016; Cooke et al., 2019). The repertoire of strategies used in every trial was presented as a percentage of each strategy used by animals for each group. The ideal path error (IPE) extracted from the Pathfinder output file was calculated as follows: the summed error of the search path (cm) = the cumulative actual path distance (cm) – the cumulative ideal path (cm). This parameter served as another metric to characterize effectiveness of search strategies: the IPE is expected to be lower if mostly spatially precise strategies were used, and higher if mostly spatially imprecise strategies were used. The preference score was calculated as the time spent in the target (T) quadrant divided by the cumulative time spent in both the T quadrant and the opposite (O) quadrant: (time in T) × 100/(time in T + time in O) was used to directly compare the quadrant discrimination between the experimental groups. The correct first choice (CFC) score was calculated by subtracting the latency to reach the goal cue from the latency to reach the false cue.

After the completion of behavioral testing, mice were deeply anesthetized with a mixture of Zoletil 100 (40 mg/kg) and xylazine (5 mg/kg) and transcardially perfused with 30 mL of ice-cold PBS followed by 30 mL of 4% paraformaldehyde (PFA) in phosphate buffer (PBS), pH 7.4. Brains were removed from the skull and postfixed in 4% PFA overnight at 4°C. The following day, the brains were transferred into PBS and stored until sectioning. Free-floating 50 μm-thick sagittal sections were obtained using a Leica VT1000S vibratome (Leica, Germany). The sections were collected in PBS and kept in PBS at 4°C or in cryoprotectant (1 volume of ethylene glycol, 1 volume of glycerin, and 2 volumes of PBS) at −20°C until staining. For permeabilization and blocking, sections were incubated in a solution containing 2% Triton-X100 in PBS (2% TBS) and 5% normal goat serum (Abcam, USA, ab7481) for 1 h at room temperature on a rocking platform. Next, the sections were incubated with primary antibodies in 0.2% TBS and 3% normal goat serum overnight at room temperature on a rocking platform. After washing three times in 0.2% TBS, sections were incubated with secondary antibodies in 0.2% TBS and 3% normal goat serum for 2 h at room temperature in darkness on a rocking platform. The following antibodies were used: guinea pig anti-DCX (1:2000, Millipore, USA, AB2253) and goat anti-guinea pig AlexaFluor 647 (1:500, Molecular Probes, USA, A21450); mouse anti-NeuN (1:1000, Millipore AB377) and goat anti-mouse AlexaFluor 488 (1:400, Molecular Probes, A32723). After three washings in 0.2% TBS, the click reaction was performed with AlexaFluor 555 Azide, triethylammonium salt (Invitrogen, USA, A20012) according to Salic and Mitchison (2008) and Ivanova et al. (2022). After three washes in 0.2% TBS and three washes in PBS, the sections were glass-mounted using Fluorescent Mounting Medium (DAKO, USA, S3023). The glass slides were dried horizontally overnight at room temperature in darkness, then stored at 4°C until imaging.

Cell counting was performed by means of design-based stereology (Encinas and Enikolopov, 2008). One brain hemisphere was randomly selected for each animal. The hemisphere was sagittally sectioned in the lateral-to-medial direction, from the beginning of the lateral ventricle to the midline, thus covering the entire DG region. The sections were 50 μm thick and were collected in six parallel sets; thus, each set was comprised of sections that were 300 μm apart from each other in the brain. One set of 8–9 sections on average, covering the DG, was used for cell counting. The sections were imaged using a spinning-disc confocal microscope (Andor Revolution WD, Oxford Instruments, UK) with the iQ 3.1 software (Oxford Instruments, UK) and a 20x NA 0.75 objective (Nikon, Japan). All images were imported into Imaris software (v.7.6.4, Bitplane, UK) and cells were counted manually by an experimenter who was blinded to the group assignment. The cell counts for the section set were averaged, normalized to the average number of sections from all animals, and then multiplied by 6 and by 2 (the number of wells and hemispheres, respectively) to represent the total number of cells per two hippocampi.

To control the validity of cell counts, we compared the impact of between-sample variability of cell counts (i.e., the variance of cell counts within a set of brain sections from the same animal) with the impact of between-animals variability. We calculated the mean coefficient of error (CE) for the mice and compared it to the group variance (CV) (Slomianka and West, 2005; Basler et al., 2017; Slomianka, 2021). A mean C E s a m p l i n g 2 C V g r o u p 2 ratio less than 0.5 would indicate that between-sample variability in cell counts contributed less than 50% to the overall variability between the animals, a value considered to be acceptable as a measure of the results’ validity in most cases (Slomianka, 2021).

Statistical analysis was performed using Prism (version 6.04, GraphPad Software, USA) and SPSS (version 28.0.0, IBM, USA). We used unpaired (independent samples, repeated measures) design for the experiments. For the analysis of escape latencies we used two-way repeated measures ANOVA followed by multiple comparisons with Sidak’s correction, with a family-wise significance level set to 0.05 (α = 0.05). For the analysis of the effect of age on the search strategies, we used Generalized Linear Mixed Model (GLMM). The response variable was the strategy used by an animal to reach platform in each trial. If a spatially precise strategy (direct path, directed search, focal search, indirect search) was used, it was scored “1”; if a spatially imprecise strategy (scanning, random search, thigmotaxis) was used, it was scored “0.” Age and the day of training were the predictor variables; subjects (individual animals) were added to model a random effect. A separate GLMM analysis was conducted for each of the MWM tasks. The results were presented as odds ratio (OR) of using precise search strategies over imprecise search strategies in the aged (14MO) group compared to the mature adult (6MO) group; a significant difference for the between-group pairwise comparison was determined with Bonferroni correction. The ideal path error (IPE) data were analyzed with two-way repeated measures ANOVA followed by multiple comparisons with Sidak’s correction.

To assess the distribution of time spent in quadrants in the test trials for uniformity we applied the Dirichlet distribution, using the “Dirichlet package” (Maugard et al., 2019). If the distribution was found to diverge from a uniform distribution, we performed post-hoc single sample t-tests with Bonferroni correction to compare the percentage of time spent in a particular quadrant with a theoretical value of 25%. The average distributions of the track points (mouse locations) in the test trials were represented as heatmaps generated in Ethovision XT. For the preference score (transformed with probit function from non-normally distributed data to normally distributed) and the CFC score analysis, we used Mann–Whitney t-test.

For the cell counts, we used Mann–Whitney t-test, with a significance level set to 0.05 (α = 0.05). A total of 10 mice from the 6MO group and 14 mice from the 14MO group were examined. In the 6MO group, one animal had pronounced hydrocephaly and was excluded from the analysis. The brain sections from two other animals showed unexpectedly high number of DCX⁺ cells [three-fold higher than in the other 24 animals; values identified as outliers using the ROUT method (Motulsky and Brown, 2006)] and were excluded from the further cell count analysis. In the 14MO group, sections from one animal were damaged during staining and processing and could not be included in the cell number analysis. Of note, these four mice did not differ significantly from the remaining mice in each of the behavioral tests.

To assess potential joint variation of individual behavioral parameters and neurogenesis, we performed correlation analysis. We used (a) escape latencies (averaged across five training trials) or cumulative number of precise search strategies used by individual mice on specific days of training (ranging from 0 – if a mouse did not use any of precise search strategies in each of five training trials, to 5 – if a mouse used only precise search strategies in each of five training trials), and (b) cell counts (DCX⁺ and EdU⁺NeuN⁺DCX^- cells per DG) for the same animals. We used Spearman rank correlation (following D’Agostino & Pearson omnibus normality test), with a significance level set to 0.05 (α = 0.05).

Graphs were plotted in Prism, the “Dirichlet package,” and Ethovision XT. Images were obtained in Imaris. Final figures were prepared in Inkscape 1.1.

To investigate the effects of aging on learning, memory, behavioral flexibility, and neurogenesis, we focused on two age groups: mice that are considered mature adult (6 months old, the 6MO group) and those considered to be of late middle age (14 months old, the 14MO group), i.e., on the ages before the overt manifestation of impairments characteristic of old age (Flurkey et al., 2007). We assessed the animals’ performance in a series of consecutive MWM tasks designed to evaluate various behavioral domains and analyzed several parameters to quantify the age-related changes (scheme in Figure 1). Additionally, we tagged dividing cells using the thymidine analog EdU, administered 6 weeks prior to the start of the MWM series.

Both the 6MO and 14MO groups were first trained for 5 days to locate a hidden escape platform in the standard spatial MWM (sMWM) task (Figure 1). On day 6, the platform was removed, and a memory test was conducted. Next, the same animals underwent 3 days of training (days 6–8) in the reversal MWM (rMWM) task where the platform was moved to the quadrant opposite to that used in the sMWM task. Following another memory test without the platform on day 9, the mice underwent training for 6 days (days 9–14) in the context discrimination MWM (cdMWM), with the color of the pool (blue and white) and the position of the beacons (a ball and a pyramid) alternating daily. The training in cdMWM was followed by memory tests in both pool configurations, with the platform removed. To evaluate the overall fitness of the animals for the task series, we measured an average swimming speed on days 1, 6, and 10 of training and found no significant differences between the 6MO and 14MO groups, indicating comparable locomotor activity of the mice of both ages (Figure 2).

Certain aspects of learning and memory, particularly in complex settings, have been shown to rely on ongoing hippocampal neurogenesis. As the division of neural stem cells and production of new neurons in the DG continuously decline with age (Kuhn et al., 1996; Ben Abdallah et al., 2010; Encinas et al., 2011; Kirschen and Ge, 2019), it is plausible that age-related changes in animals’ performance in the learning and memory tests are associated with the changes in hippocampal neurogenesis. Therefore, we investigated neurogenesis in 6MO and 14MO mice and explored its potential connection to their test performance.

Six weeks prior to the start of learning and memory training (i.e., 8 weeks at the time of the final cdMWM test, Figure 1), mice from both age groups were administered a synthetic thymidine analog EdU. Following completion of the test, the mice hippocampi were examined for the EdU labeling and expression of NeuN and DCX markers. The EdU signal identifies cells that were undergoing DNA duplication at the time of EdU administration, NeuN expression marks differentiated neurons in the DG, and DCX expression marks advanced neuronal precursors and immature neurons (Kronenberg et al., 2003; Encinas and Enikolopov, 2008; Podgorny et al., 2018). Thus, EdU⁺NeuN⁺DCX⁻ cells correspond to newly generated fully differentiated mature neurons derived from progenitors that were dividing at the time of the EdU administration; DCX⁺ cells encompass a broad range of neuronal precursors; and EdU⁺DCX⁺ cells correspond to neuronal precursors that were in the cell cycle at the time when the label was injected (representative images are shown in Figures 6A,B).

We found a significantly lower number of EdU⁺, EdU⁺NeuN⁺DCX⁻, and DCX⁺ cells in the DG of 14MO mice compared to 6MO mice (p < 0.0001, p = 0.0018, and p < 0.0001, respectively) (Figures 6C–E), in line with previous reports (Bondolfi et al., 2004; van Praag et al., 2005; Gil-Mohapel et al., 2013). As expected, we did not detect any EdU⁺DCX⁺ cells in either group.

Notably, when we assessed the impact of between-sample variability on the total variability, we found that the mean C E s a m p l i n g 2 C V g r o u p 2 ratio was lower than 0.5 for each cell count in either group: 0.282 and 0.25 for the EdU⁺ cells in the 6MO and 14MO groups, respectively; 0.20 and 0.0133 for the EdU⁺NeuN⁺DCX^- cells in the 6MO and 14MO, respectively; and 0.130 and 0.0803 for the DCX⁺ cells in the 6MO and 14MO groups, respectively. This indicates that the variability introduced by the sampling procedure was less than half of the total variability, thus, supporting the validity of our cell counts (particularly relevant for the older animals).

Given the involvement of newly generated hippocampal neurons in a range of behavioral responses, we further investigated whether the number of newly produced cells in individual animals correlated with their performance in the learning and memory tasks. Specifically, we correlated the numbers of DCX⁺ and EdU⁺NeuN⁺DCX⁻ cells (representing immature and mature neurons, respectively) in each mouse with their escape latency and with their use of spatially precise strategies during the initial days of training in the sMWM, rMWM, and cdMWM tasks (day 1, day 6, and days 9–10 of the training sequence, respectively).

On the first day of training in sMWM we did not find a significant correlation (Spearman’s r here and later) between the number of DCX⁺ immature neurons in individual animals and their task performance (Figure 7A). Of note, the evident bimodal distribution of the cell counts reflects the difference in the number of these cells between the 6MO and 14MO groups.

During the initial training in rMWM, the correlation between the number of DCX⁺ cells and the animal’s escape latency or use of spatially precise search strategies did not reach statistical significance (Figure 7B, r = −0.3643, p = 0.0801, r = 0.3867, p = 0.0619, respectively). However, on the first and second days of training in cdMWM (in the blue and white pools, respectively), there were significant correlations between the number of DCX⁺ cells and the escape latencies (Figure 7C (left graph), day 9: r = −0.4292; p = 0.0364; day 10: r = −0.5576; p = 0.0046). Additionally, there was a trend for a correlation between the number of DCX⁺ cells and the use of spatially precise search strategies on the first day of cdMWM (Figure 7C (right graph), day 9: r = 0.4023, p = 0.0513), and a significant correlation on the second day of cdMWM (Figure 7C (right graph), day 10: r = 0.4512, p = 0.0269).

On the first day of training in the sMWM, the number of adult-born differentiated neurons (EdU⁺NeuN⁺DCX⁻) was correlated with the escape latency, but not with the use of spatially precise search strategies (r = −0.4774, p = 0.0183 and r = −0.1152, p = 0.5920, respectively) (Figure 7D). The number of adult-born differentiated neurons (EdU⁺NeuN⁺DCX⁻) showed correlations with the escape latency (Figure 7E, r = −0.6649, p = 0.0004) and the use of spatially precise search strategies (r = 0.4567, p = 0.0249) on the first day of training in the rMWM task. Furthermore, the number of mature neurons correlated with the average escape latency (day 9: r = −0.6224, p = 0.0012; day 10: r = −0.4749, p = 0.0190) and the use of spatially precise search strategies (day 9: r = 0.6699, p = 0.0003; day 10: r = 0.4554, p = 0.0253) during the initial days of training in cdMWM (Figure 7F). Moreover, when comparing the data for the mice in the two age groups, we found a significant correlation (p = 0.0205) and a highly significant correlation (p < 0.0001) between the number of mature neurons and their escape latencies on day 6 of rMWM, and on days 9 and 10 of cdMWM, respectively, in individual 14MO mice but not in 6MO mice. Importantly, this correlation in 14MO mice disappeared by the end of training in each of these tasks (day 8 of rMWM, days 13 and 14 for cdMWM, all p > 0.05; not shown). Taken together, these findings suggest that individual animals with higher levels of adult-born immature and mature neurons perform better at the beginning of training in tasks that require flexible adaptation to the changed demands of the task.