Abstract
Exposure to elevated doses of ionizing radiation, such as those in therapeutic procedures, catastrophic accidents, or space exploration, increases the risk of cognitive dysfunction. The full range of radiation-induced cognitive deficits is unknown, partly because commonly used tests may be insufficiently sensitive or may not be adequately tuned for assessing the fine behavioral features affected by radiation. Here, we asked whether γ-radiation might affect learning, memory, and the overall ability to adapt behavior to cope with a challenging environment (cognitive/behavioral flexibility). We developed a new behavioral assay, the context discrimination Morris water maze (cdMWM) task, which is hippocampus-dependent and requires the integration of various contextual cues and the adjustment of search strategies. We exposed male mice to 1 or 5 Gy of γ rays and, at different time points after irradiation, trained them consecutively in spatial MWM, reversal MWM, and cdMWM tasks, and assessed their learning, navigational search strategies, and memory. Mice exposed to 5 Gy performed successfully in the spatial and reversal MWM tasks; however, in the cdMWM task 6 or 8 weeks (but not 3 weeks) after irradiation, they demonstrated transient learning deficit, decreased use of efficient spatially precise search strategies during learning, and, 6 weeks after irradiation, memory deficit. We also observed impaired neurogenesis after irradiation and selective activation of 12-week-old newborn neurons by specific components of cdMWM training paradigm. Thus, our new behavioral paradigm reveals the effects of γ-radiation on cognitive flexibility and indicates an extended timeframe for the functional maturation of new hippocampal neurons.
SIGNIFICANCE STATEMENT Exposure to radiation can affect cognitive performance and cognitive flexibility — the ability to adapt to changed circumstances and demands. The full range of consequences of irradiation on cognitive flexibility is unknown, partly because of a lack of suitable models. Here, we developed a new behavioral task requiring mice to combine various types of cues and strategies to find a correct solution. We show that animals exposed to γ-radiation, despite being able to successfully solve standard problems, show delayed learning, deficient memory, and diminished use of efficient navigation patterns in circumstances requiring adjustments of previously used search strategies. This new task could be applied in other settings for assessing the cognitive changes induced by aging, trauma, or disease.
- cognitive flexibility
- context discrimination
- gamma-radiation
- hippocampal neurogenesis
- search strategies
- spatial learning
Introduction
Therapeutic or environmental exposure to ionizing radiation may lead to long-lasting deficits in verbal and spatial memory, attention, problem-solving, and emotional status in humans (Kadan-Lottick et al., 2010; Krull et al., 2012; Schuitema et al., 2015; Williams et al., 2021; Kramkowski and Hebert, 2022; Lehrer et al., 2022). Radiation-induced deficits are particularly evident when new or changing events or environments require rapid adaptation and adjustment of behavior or problem-solving strategies — an ability broadly understood as cognitive/behavioral flexibility. Similarly, in animal models, brain-focused and whole-body irradiation can impair learning, memory, and relearning (Snyder et al., 2005; Saxe et al., 2006; Clelland et al., 2009; Kiffer et al., 2019).
Memory formation and processing critically depend on the integrity of the hippocampus and the hippocampal neural circuits (Frankland et al., 2019; de Sousa et al., 2021). Hippocampal neurons are particularly vulnerable to neurogenerative diseases, seizures, or trauma. Associated damage to the hippocampus produces learning and memory deficits in humans and animals resembling those observed after therapeutic, environmental, or experimental irradiation.
An additional potential link among radiation, induced cognitive deficits, and hippocampal function is the adverse effects of radiation on dividing hippocampal cells. Hippocampal circuits continue to incorporate locally produced new neurons after birth, in both animals (Aimone et al., 2014; Cameron and Glover, 2015; Goncalves et al., 2016) and humans (Spalding et al., 2013; Boldrini et al., 2018; Kempermann et al., 2018; Sorrells et al., 2018; Moreno-Jimenez et al., 2021). Newborn neurons originate from dormant adult neural stem cells residing in the dentate gyrus (DG) of the hippocampus, after a protracted cascade of events, including the proliferation, differentiation, and maturation of neural progenitors and their progeny, as well as integration of the new neurons into existing networks (Encinas et al., 2011; Vivar and van Praag, 2013; Enikolopov et al., 2015; Lazutkin et al., 2019; Urban et al., 2019; Denoth-Lippuner and Jessberger, 2021). Integrated adult-born neurons participate in spatial memory formation (Aimone et al., 2014), memory reconsolidation (Suarez-Pereira and Carrion, 2015; Lods et al., 2021), forgetting (Akers et al., 2014; Epp et al., 2016), and distinguishing between similar contexts (Clelland et al., 2009; Aimone et al., 2011; Sahay et al., 2011; Niibori et al., 2012; Anacker and Hen, 2017), among other processes. Radiation eliminates actively dividing cells and suppresses the proliferation and differentiation of surviving neural stem and progenitor cells, thus resulting in a prolonged deficit in new neurons (Tada et al., 2000; Monje et al., 2002; Mizumatsu et al., 2003; Encinas et al., 2008; Mineyeva et al., 2018). Therefore, some cognitive impairments observed after irradiation may be caused or potentiated by compromised neurogenesis-mediated hippocampal function.
Despite accumulating evidence of the detrimental effects of radiation on hippocampal function and hippocampal neurogenesis, the full range of radiation-induced cognitive deficits remains elusive. This lack of knowledge may reflect the multifactorial effects of radiation on the brain, including neuronal damage, vascular abnormalities, gliosis, demyelination, and white matter necrosis (Monje et al., 2007; Gibson and Monje, 2012; Pereira Dias et al., 2014; Son et al., 2015). However, current behavioral tests in animal models might conceivably be insufficiently targeted or insufficiently sensitive to reveal fine details of the behavioral features affected by compromised hippocampal functioning, particularly those dependent on ongoing production of new neurons. Notably, in a panel of the most widely used learning and memory tests, irradiation often does not lead to overt behavioral phenotypes.
Here, we investigated the effects of radiation on animals' learning and memory in a series of consecutively presented tasks entailing various types of cues and contexts and their combinations. We found that exposure of mice to a moderate dose of γ rays affected learning and memory and the choice of search strategies. Importantly, the radiation-induced deficits manifested fully only in a complex hippocampus-dependent context discrimination task requiring integration of information on the spatial and nonspatial cues and the changing contexts and readjustment of search strategies, thus demonstrating that radiation exposure impairs cognitive flexibility.
Materials and Methods
Mice
We used male heterozygous Nestin-GFP mice at 4-4.5 months of age (Mignone et al., 2004), which were maintained on C57BL/6J background. Before the experiment, the animals were housed in groups of 2-4 animals per cage under standard conditions with 12/12 h light-dark cycle, in cages 36 × 21 × 13.5 cm, with food and water available ad libitum. All experiments were performed in compliance with the requirements and regulations for studies using experimental animals and the animal care guidelines issued by the National Institutes of Health and Stony Brook University.
Morris water maze (MWM) training
Mice were transferred from the animal facility to the experimental room 1 h before the beginning of the experimental procedures. Mice were subjected to three MWM-based tasks in succession: spatial MWM (sMWM), followed by reversal MWM (rMWM), followed by context discrimination MWM (cdMWM). For training in the sMWM (see Fig. 1a), animals were introduced to a circular pool (120 cm in diameter) made of blue plastic (Noldus). The water was made opaque with nontoxic white paint and kept at 23° C-24° C. Several distal cues were mounted on the walls around the pool. Each mouse performed five 60 s trials per day (with 30 min intertrial intervals) for 5 consecutive days to find the platform submerged 0.5 cm under 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 (not containing the platform). If the mouse was unable to locate the platform, the animal was gently guided to the platform by the experimenter. The mouse was allowed to remain on the platform for 30 s. The platform position remained unchanged for 5 training days. On day 6, the mice were given a spatial memory test involving swimming in the pool without the platform for 60 s.
Training in rMWM (see Fig. 1b) began 2 h after the sMWM memory test. The hidden platform was returned to the pool but was now placed in the quadrant opposite to that used for 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 3 additional days to learn the new platform location. The overall procedure was the same as that described above for sMWM. On day 9, the mice were subjected to a 60 s memory test without the platform.
cdMWM training (see Fig. 1c) began 2 h after the rMWM memory test. 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 above the platform (the goal cue) and the other above the opposite quadrant (the false cue). On the next day (day 10), each mouse was placed in a pool that was identical to the blue pool, except that it was made of white plastic. This pool was located in the adjacent room with different distal cues on the walls. Two beacons identical to those used with the blue pool (a ball and a pyramid) were suspended above the pool, except that the goal cue and the false cue were switched; that is, if the ball signaled the platform location in the blue pool, it became the false cue in the white pool, whereas the pyramid became a new goal cue. Beacons indicating the platform location were counterbalanced between the pools (i.e., in each group for half the animals, in a particular pool the goal cue was a ball, whereas for the other half in the same pool the goal cue was a pyramid). Thus, to choose the correct local cue signaling the platform location, each mouse was required to discriminate among the contexts comprising the room, pool color, and distal cues. Each mouse performed five daily sessions of 60 s each, and after reaching the platform was allowed another 30 s to stay on the platform. The pools were alternated daily for 6 d (days 9-14). On day 15, the mice were given 60 s memory tests in the blue and white pools, separated by a 2 h interval. The timeline of the tasks is shown in Figure 1d.
Animals' behavior was recorded with the ceiling-mounted video cameras connected to the computers, using EthoVision XT, versions 8 and 17 (Noldus). We determined the latency to reach the platform (escape latency) during learning trials and the fraction of time spent in each quadrant (in %) during the memory test. To assess and directly compare the magnitude of quadrant discrimination between the experimental groups, we determined the quadrant preference score, calculated as the time spent in the target (T) quadrant as a fraction of the time spent in the T quadrant and the opposite (O) quadrant (time in T) × 100/(time in T + time in O). We also calculated the correct first choice (CFC) score by subtracting the latency to reach the goal cue from the latency to reach the false cue. This metric is expected to be higher if the animal's first choice was correct, that is, if the animal approached the goal cue first, thus, (goal cue latency) < (false cue latency); it is expected to be lower if the animal's first choice was incorrect (i.e., an animal approached the false cue first, hence (the goal cue latency) > (the false cue latency).
As yet another way to assess animals' performance in the MWM tasks, we analyzed search strategies used in every learning trial. For the classification analysis of the search strategies (here used in a more narrow sense to describe distinct navigation patterns), we used the Pathfinder software, as described by Cooke et al. (2019). Briefly, raw tracking files containing xy coordinates over time extracted from the EthoVision XT were uploaded to the Pathfinder. According to the variables defined in the Pathfinder, pool and platform geometry, each track was assigned to one of the seven distinct search strategies: direct path, directed search, focal search, indirect search, scanning, random search, and thigmotaxis. The search strategies, such as thigmotaxis, random search, and scanning, are low-precision and inefficient, and are hippocampal-independent. With learning progression, the fraction of these strategies decreases, whereas the fraction of more efficient, spatially specific, and, presumably, hippocampal-dependent, strategies, such as indirect search, focal search, directed search, and direct path, increases (Garthe et al., 2009, 2014, 2016; Cooke et al., 2019). We plotted the percent of each strategy class to demonstrate their contribution to an overall learning performance. This qualitative analysis was followed by the generalized linear mixed model (GLMM) analysis to assess the odds ratio (OR) of using spatially precise strategies by the experimental groups (see Statistical analysis). We determined the fraction of each strategy relative to all strategies for the group's performance in each trial for the sMWM, rMWM, and cdMWM tasks. To demonstrate the dynamics of the employed strategies, we analyzed strategies used during the first and last days of training in the standard tasks (for sMWM, days 1 and 5; for rMWM, days 6 and 8), and at each day of training in the cdMWM task (days 9-14). To reveal the difference between the groups' performance, we subtracted the fraction of each strategy (averaged for each day) in the irradiated group from its fraction in the Sham group. We also analyzed ideal path error (IPE), defined by the Pathfinder program as follows: the summed error of the search path (cm) = the cumulative actual path distance (cm) – the cumulative ideal path (cm). For example, if the direct path strategy would be used, the cumulative actual path distance would be close to the cumulative ideal path resulting in a small IPE; in contrast, if a spatially nonspecific strategy was used (e.g., scanning or random search), the IPE is expected to be high.
Intrahippocampal anisomycin administration
Anisomycin, a protein synthesis inhibitor, was used to disrupt hippocampal function (Sharma et al., 2012). Mice were anesthetized with a mixture of Zoletil 100 (40 mg/kg, Virbac) and xylazine (5 mg/kg, Bioveta). Guide cannulas (1.25 mm depth, Plastics One) were stereotaxically implanted bilaterally into the hippocampus (bregma −2.5 mm, ML 2 mm) and closed by a cap (2.25 mm depth, Plastics One) until the experiment. Anisomycin was administered using an injection cannula (2.25 mm depth, Plastics One) at a volume of 1 μl (concentration 16 mg/ml) per hemisphere at a rate of 0.7 μl/min, each day, 10 min before training, for the entire 6 d of training in the cdMWM. Control animals received saline under the same conditions. For morphologic control of the injection site, on completion of the behavioral experiment, mice were deeply anesthetized with a mixture of Zoletil 100 and xylazine, and the same guide cannulas were used to inject methylene blue (4% solution, 1 μl/hemisphere at a rate of 0.5 μl/min) into the hippocampus.
Irradiation and cell labeling
Whole-body γ irradiation was delivered by the 60Co source. Two doses of γ-radiation were applied: the dose of 1 Gy was delivered for 57 s (1 Gy group) and the dose of 5 Gy was delivered for 4 min 53 s (5 Gy group). Mice from the sham-irradiated group (Sham) were transported to the source facility but were not exposed to irradiation. All animals in each group were irradiated simultaneously. To label dividing cells in the brain, experimental animals were injected with 5-ethynyl-2′-deoxyuridine (EdU, 123 mg/kg, Lumiprobe) intraperitoneally 2 h before the irradiation procedure. After exposure to the γ ray source, all mice were returned to their home cages and housed in conventional conditions for 1, 4, or 6 weeks before the beginning of behavioral experiments (1, 4, and 6 week experiments, respectively); since the learning protocol encompassed 2 weeks, these experimental labels correspond to the final memory tests being performed 3, 6, or 8 weeks after irradiation, respectively. In 1 and 4 week experiments, mice were exposed to 5 Gy only, in 6 week experiment, mice were exposed to either 1 or 5 Gy of γ-radiation.
c-Fos expression in various populations of adult-born neurons
Newborn neurons of different ages were identified by injection of two thymidine analogs, BrdU (Sigma-Aldrich) and EdU, 12 and 7 weeks before death, respectively. To label the maximum number of dividing cells, injections were administered intraperitoneally for 5 consecutive days twice a day with an interval of 12 h at doses of 50 mg/kg for BrdU and 40 mg/kg for EdU. On the 11th week, all animals were trained in the sMWM for 5 d as described above. Starting from day 6, mice were divided into three groups. Control animals continued to be trained under the same conditions for 2 more days. The animals of the New Paradigm and New Context groups were trained in a blue pool on day 6 with beacons suspended above the pool. On day 7, only the animals from the New Context group were trained in a white maze with the same beacons, but with the platform located in the opposite quadrant. The mice were deeply anesthetized and perfused 2 h after the end of training on a respective day (Control and New Context groups on day 7, New Paradigm on day 6). The brains were extracted, and DG was analyzed for c-Fos expression, BrdU and EdU labeling, and colocalization of each of the thymidine analogs with c-Fos to assess the number of activated adult-born neurons of different ages.
Immunohistochemistry
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% PFA in PBS, pH 7.4. Brains were removed from the skull and postfixed in 4% PFA overnight at 4° C. The next day, the brains were transferred into PBS and stored until sectioning. Free-floating sagittal 50 μm sections were obtained using Leica VT1000S vibratome (Leica). 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. Staining started with the permeabilization and blocking procedure: sections were incubated in a solution containing 2% Triton X-100 in PBS (2% TBS) and 5% normal goat serum (Abcam, ab7481) for 1 h at room temperature on the rocking platform. Incubation with primary antibodies was performed in 0.2% TBS and 3% NGS overnight at room temperature on a rocking platform. After washing 3 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, AB2253) and goat anti-guinea pig Alexa Fluor 647 (1:500, Invitrogen, A21450); mouse anti-NeuN (1:1000, Millipore AB377) and goat anti-mouse Alexa Fluor 405 (1:400, Invitrogen, A31553); rabbit anti-c-Fos (1:2000, Synaptic Systems, 226003) and goat anti-rabbit Alexa Fluor 488 (1:500, Invitrogen, A48282); and mouse anti-BrdU (1:500, Biolegend, Mobu-1, 317902) and goat anti-mouse Alexa568 (1:500, Invitrogen, A-11031). After three washings in 0.2% TBS, the click reaction was performed with Alexa Fluor 555 Azide, triethylammonium salt (Invitrogen, A20012) according to Salic and Mitchison (2008). After three washings in 0.2% TBS and three washings in PBS, the sections were glass-mounted using Fluorescent Mounting Medium (DAKO, S3023). Glass slides were dried horizontally overnight at room temperature in darkness, then transferred to 4° C and kept until imaging.
Imaging and cell counting
Cell counting was performed by means of design-based (assumption free, unbiased) stereology, as detailed in Encinas and Enikolopov (2008). One brain hemisphere was randomly selected per animal. The hemisphere was sliced sagittal, in the lateral-to-medial direction, from the beginning of the lateral ventricle to the midline, thus including the entire DG. The 50 μm sections were collected in six parallel sets, sections in each set were 300 μm apart from each other. One set of eight to nine sections on average, covering the extent of the DG, was used for cell counting. Sections were imaged using a spinning-disk confocal microscope Andor Revolution WD (Oxford Instruments) equipped with the iQ 3.1 software (Oxford Instruments) and 20× (NA 0.75) objective (Nikon). All images were imported to Imaris software (version 7.6.4, Bitplane) and counted manually by an experimenter blind to the groups. Cell counts from all sections were averaged, then normalized to the average number of sections from all animals, then multiplied by 6 and by 2 (the number of wells and hemispheres, respectively), thus representing the total number of cells per two hippocampi.
Statistical analysis
Statistical analysis and graph plotting were performed in Prism (version 6.04, GraphPad Software) and SPSS (version 28.0.0, IBM) for Windows. For latencies to reach the platform during learning, 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 statistical analysis of the effect of anisomycin or irradiation on the search strategy, we applied GLMM. GLMM allows to analyze the effect of predictor variables, considering the potential nonindependence of response variables. The response variable was the strategy used by an animal to reach the platform in every trial. For this analysis, the strategies were scored as follows: if an animal used spatially precise strategy in a trial (direct path, directed search, focal search, indirect search), the strategy was scored 1; otherwise (i.e., if scanning, or random search, or thigmotaxis strategy was used), the strategy was scored 0. Predictor variables were treatment (anisomycin or dose of whole-body irradiation, in the respective experiments), and the day of training; subjects (individual animals) were added to model random effect. Separate GLMM analyses were performed for the pulled data of the sMWM and rMWM tasks, and for the data of the cdMWM task. The results were presented as OR of using precise search strategies over imprecise search strategies in a treated group compared with the respective control group, and a significant difference for the between group pairwise comparison with Bonferroni correction was determined. OR = 1 indicates no effect of a treatment, OR < 1 indicates that the event is less likely to occur in a treated group, and OR > 1 indicates that the event is more likely to occur in a treated group. For the IPE analysis, we used two-way repeated-measures ANOVA followed by multiple comparisons with Sidak's correction. For the time spent in quadrants in memory tests, we used the Dirichlet distribution to determine whether the fraction of time spent in the quadrants is uniformly distributed (Maugard et al., 2019). This allowed to overcome the problem of the constant-sum constraints of the MWM assay, in which the data points (i.e., the time that is spent in each quadrant of the maze as a fraction of the overall time of the assay) are not independent of each other (a requirement of the tests, e.g., ANOVA). For this analysis, we used the Dirichlet package developed by Maugard et al. (2019) and available for download from https://github.com/xuod/dirichlet/tree/master/dirichlet. This analysis was followed by single-sample t tests with Bonferroni correction for multiple comparisons to characterize the preference for a particular quadrant in the groups showing a divergence from uniformity. Using Ethovision XT (version 17), we plotted the heatmaps for the test trials of 1, 4, and 6 week experiments to qualitatively compare the average distributions of the track points (mouse locations) in these trials. For the preference score analysis, we used Prism built-in probit function to transform non-normally distributed data to normally distributed followed by Kruskal–Wallis test with Dunn's multiple comparisons correction (6 week experiment) or Mann–Whitney t test (anisomycin, 1 and 4 week experiments). For the CFC score analysis, we used Kruskal–Wallis test with Dunn's multiple comparisons correction (6 week experiment), or Mann–Whitney t test (anisomycin, 1 and 4 week experiments). One mouse from the 5 Gy group was excluded from analysis following the memory test in the white pool in the 6 week experiment because of software failure.
For cell counts, we used one-way ANOVA followed by Sidak's correction for multiple comparisons, with a family-wise significance level set to 0.05 (α = 0.05). One mouse was excluded from the analysis of recruitment of 12-week-old cells because of poor EdU staining.
Figures were prepared in Inkscape 1.1.1 software, using graphs plotted in Prism and Ethovision XT, images were obtained in Imaris, and pictures were created in www.BioRender.com and Freepik.
Results
Context discrimination task
To assess the effect of radiation on the animals' cognitive flexibility, we first sought to develop a challenging task that would require the mice to use various strategies to discriminate between similar but not identical contexts, akin to pattern separation tasks in humans and animals (Yassa and Stark, 2011; Leal and Yassa, 2018). In this task, which we called cdMWM, the mice were required, in addition to remembering the position of a platform in relation to surrounding spatial cues (as in conventional MWM tasks), also to consider the overall context, including the color of the pool and the presence of local nonspatial cues, and to integrate these distinct types of information to find the platform.
For the cdMWM task, the mice were first trained in conventional spatial and reversal versions of the MWM task (sMWM and rMWM, respectively, experimental scheme on Fig. 1a,b and details in Materials and Methods). After the training and testing in the sMWM and rMWM tasks, the mice were trained in the cdMWM task (Fig. 1c,d). In this task, the mice were required to rely on the distal cues, local cues (a ball and a pyramid), and the color of the pool (blue and white) to navigate to the location of the platform. Thus, in addition to testing the animals' navigation directed by spatial and nonspatial (distal and local) cues, the chosen succession of tasks required the trained animals to adapt and relearn the previously used strategies, thereby revealing their potential for cognitive flexibility. Relearning was first required in the rMWM task to learn the new platform location by using familiar distal cues. Subsequently, in the blue pool (cdMWM), the mice were required to learn that both distal and local cues were necessary for navigation. Finally, in the white pool (cdMWM), the mice were required to differentiate, in a context-dependent manner, which of the two local cues indicated the platform location. An additional feature of the cdMWM paradigm, beyond its greater complexity, was the timeline of the task sequence, wherein the last test performed was within the period of the maturation of new hippocampal neurons.
The context discrimination paradigm. a, sMWM task: schematic representation of the pool used for learning (left) and memory testing (right) of the platform location. b, rMWM task: schematic representation of the pool used for learning (left) and for memory testing (right) of the new platform location in the quadrant opposite to that used for sMWM. c, cdMWM task: schematic representations of the pools used for learning in daily alternating pools (two left panels) and memory testing (two right panels). Mice need to discriminate between the blue (B) and the white (W) pool with different distal cues to choose the correct local cue (a ball or a pyramid) suspended above the platform. d, Experimental timeline: mice were trained in succession of three tasks: sMWM learning, days 1-5, sMWM memory test, day 6; rMWM learning, days 6-8, rMWM memory test, day 9; cdMWM learning in daily alternating blue and white pools, days 9-14, cdMWM memory tests in the blue and white pools, day 15.
Context discrimination MWM is a hippocampus-dependent task
We first examined whether the cdMWM is a hippocampus-dependent task by blocking hippocampal function through local injection of the protein synthesis inhibitor anisomycin (Sharma et al., 2012). The mice were implanted bilaterally with cannulas, which were later used for injection of either anisomycin (the ANI group) or saline (the SAL group). The mice were then subjected to an abridged paradigm, with the rMWM omitted: they were trained and tested in the sMWM, and then anisomycin was administered intrahippocampally 10 min before each of the daily training in the cdMWM (Fig. 2a,b).
Context discrimination depends on protein synthesis in the hippocampus. a, Experimental timeline. Saline (SAL) or anisomycin (ANI) was injected through cannulas (shown with the syringe sign) to SAL (n = 10) and ANI (n = 9) mice. B, Blue pool; W, white pool. b, Escape latencies (time to reach a hidden platform) during training. ***p < 0.001 for day 5 versus day 1 for each of two groups (two-way ANOVA, multiple comparisons with Sidak's correction). c, Time spent in quadrants (in %) during test following sMWM training; here and below, on the graphs for the test trials (h,j): Each column represents an animal. Each color represents the percent of time spent in each quadrant. Black dashed line indicates mean value for the fraction of time spent in each quadrant. Error bar on the mean is approximated with the inverse Fisher information. Black symbols represent the results of Dirichlet distribution analysis. White symbols represent the results of post hoc single-sample t test comparison (with Bonferroni correction) with the theoretical value 25%. **p < 0.01. ***p < 0.001. ****p < 0.0001. d, Preference score, relative time spent in T quadrant in test. Dots represent individual values. Horizontal dashed line indicates chance level. e, Escape latencies (time to reach a hidden platform) during learning in the blue (B) and white (W) pools. Syringe sign indicates ANI (or SAL) injection before the training in the respective days. Black asterisks indicate statistically significant difference between the groups. **p < 0.01 (group effect, two-way ANOVA, multiple comparisons with Sidak's correction). Magenta asterisks indicate statistically significant differences between the ANI and the SAL groups on the respective days. *p < 0.05; **p < 0.01; two-way ANOVA, multiple comparisons with Sidak's correction. f, Search strategies analysis during training in cdMWM. Top, Search strategies for SAL mice. Bottom, Search strategies for ANI mice. Each color represents a strategy (see legend under the graph). Bars represent a fraction of animals used each strategy. Stacked bars totaling 100% represent the repertoire of strategies used in each trial. g, Top, OR of spatially precise strategies use in ANI compared with the SAL. *p < 0.05; **p < 0.01; ****p < 0.0001; values determined by fitting a generalized linear mixed effect model with binomial distribution. Bottom, Ideal path error on days 6-11 of cdMWM training. **p < 0.01; ****p < 0.0001; two-way ANOVA, multiple comparisons with Sidak's correction. h, Time spent in quadrants (in %) during test following cdMWM training in the blue pool. **p < 0.01. ****p < 0.000. i, Preference score, relative time spent in T quadrant in test following cdMWM training in the blue pool. *p < 0.05 for ANI versus SAL (Mann–Whitney t test). Dots represent individual values. Horizontal dashed line indicates chance level. Inset, CFC score (latency to reach the false cue – latency to reach the goal cue). j, Time spent in quadrants (in %) during test following cdMWM training in the white pool. **p < 0.01. ***p < 0.001. k, Preference score, relative time spent in T quadrant in test following cdMWM training in the white pool. Dots represent individual values. Horizontal dashed line indicates chance level. Inset, CFC score (latency to reach the false cue – latency to reach the goal cue). *p < 0.05 for ANI versus SAL (Mann–Whitney t test). b, d, e, g (bottom), i, k, Data are mean ± SEM. T, Target quadrant; O, opposite quadrant. For statistical analysis, see Extended Data Table 2-1 (lines 1-16).
Table 2-1
Statistical analysis data for behavioral experiments and cell counts. Download Table 2-1, XLSX file.
sMWM
Mice in both groups successfully learned the platform location, as indicated by a significant decrease in escape latencies by the end of training (Fig. 2b; significant effect of day: p < 0.0001, no interaction between group and day effects, two-way ANOVA, Extended Data Table 2-1, line 1).
In the spatial memory test, the distribution of the time spent by the ANI and the SAL groups in the four quadrants, as expected, significantly differed from a uniform (25%) distribution according to Dirichlet distribution analysis (Fig. 2c, ANI: p = 0.00028, SAL: p < 0.0001; here and below, to account for the constant-sum constraints of the MWM task, we used Dirichlet distribution to compare the distribution of time spent in quadrants with a uniform distribution in within-group analysis). Post hoc single-sample t tests revealed a longer time spent in the T quadrant than the theoretical value of 25% (T quadrant ANI: p = 0.0002, SAL: p = 0.002, here and below, single-sample t test with Bonferroni correction for multiple comparisons was used to compare the fraction of time spent in a particular quadrant with the theoretical value 25%).
To directly compare the extent of quadrant discrimination between the experimental groups, we calculated a quadrant preference score for each animal as the time spent in T as a fraction of the sum of time in T and O. We observed no difference in preference scores between the ANI and SAL groups (Fig. 2d, unpaired t test; Extended Data Table 2-1, line 4). Together, these results suggest that both groups showed normal performance in the sMWM task.
cdMWM
In the context discrimination task, either anisomycin or saline was administered before each learning session in the pools that were alternated daily. Two-way ANOVA of escape latencies revealed a significant interaction between the group and day effects (p = 0.0297), a significant effect of group (Fig. 2e, p = 0.0029), but no effect of day (Extended Data Table 2-1, line 5). ANI mice had longer escape latencies than SAL mice in the blue pool on days 8 and 10, and in the white pool on day 11 (Fig. 2e; here and below, pair-wise comparisons with Sidak's correction were used to compare escape latencies between groups; blue pool, day 8 ANI–SAL: p = 0.0303, day 10 ANI–SAL: p = 0.0474, white pool, day 11 ANI–SAL: p = 0.0035; Extended Data Table 2-1, line 6). These results indicate that, in the cdMWM task, learning was impaired by anisomycin.
The memory test in the blue pool indicated that the distribution of time spent in four quadrants significantly differed from a uniform distribution in SAL mice, but not ANI mice (Fig. 2h, SAL: p < 0.0001; ANI: p = 0.54). In the SAL group, post hoc single-sample t tests revealed a longer time spent in the T quadrant (p = 0.0015), whereas the time spent in the O quadrant did not differ from the theoretical value of 25% (Extended Data Table 2-1, line 10). The memory test in the white pool similarly showed that the distribution of time spent in the quadrants significantly differed from a uniform distribution in SAL mice, but not in ANI mice (Fig. 2j, SAL: p = 0.00028; ANI: p = 0.25). In the SAL group, post hoc analysis revealed a longer time spent in the T quadrant than the theoretical value of 25% (p = 0.0046). Single-sample t tests performed for the O quadrant revealed that the fraction of time spent in this quadrant did not differ from 25% (Extended Data, Table 2-1, line 12). The preference score was significantly lower in the ANI group than the SAL group in the blue pool (p = 0.034), but not the white pool (p = 0.22) (Fig. 2i,k; Extended Data Table 2-1, lines 13 and 14). We considered a possibility that, during the test session with the local cues presented, mice initially visited the goal cue, but in the absence of the platform, they started checking the other locations in the pool, which may have blurred the difference in the preference scores. Thus, we compared the preference scores obtained for the first 10 s of the 60 s test trial. We found that the preference score was lower in the ANI group than in the SAL group in either the blue or white pool (p = 0.0326 and p = 0.0141, respectively, Extended Data Table 2-1, lines 13 and 14). We also analyzed the CFC score as another approach to assess animals' memory for the cues. The score was calculated by subtracting the latency to reach the goal cue from the latency to reach the false cue. We found a lower score in ANI mice compared with SAL mice for the white pool but not for the blue pool (Fig. 2k, inset, p = 0.0472, Fig. 2i, inset, p = 0.325, respectively, Extended Data Table 2-1, lines 15 and 16). Together, the observed learning and memory deficits after ANI treatment indicate that the cdMWM is a hippocampus-dependent task.
Search strategies
In the conventional tests, even when the differences in parameters such as escape latency are subtle, the specific spatial navigation strategies that the animal uses to find the platform may conceivably differ. Therefore, we analyzed the navigation patterns of the animals in each group by categorizing the overall search path in each trial into several possible search strategies using the Pathfinder suite (Cooke et al., 2019). These strategies ranged from inefficient spatially imprecise strategies (e.g., thigmotaxis and random search) to highly efficient spatially precise strategies (e.g., direct path, directed search, and focal search). We used GLMMs to model the relation between the outcome variable (search strategy) and predictor variables (i.e., factors: treatment and day of training). Search strategies were scored as 1 for the spatially precise strategies (indirect search, directed search, focal search, direct path) or as 0 for spatially imprecise strategies (thigmotaxis, random search, scanning). Fixed effects of factors and their interactions, ORs for experimental groups to fall into target category (i.e., to use spatially precise strategies), and pairwise comparisons between the groups were calculated.
Anisomycin injection resulted in altered learning in the cdMWM task; we thus analyzed behavioral strategies used in this task (Fig. 2f, top, SAL; bottom graph, ANI). While the overall pattern of strategies used on days 6 and 7 did not differ between the groups, on days 8-11, we observed lower contribution of the precise search strategies (shown in shades of blue) in ANI mice than in SAL mice. GLMM revealed a significant effect of group × day interaction (p = 0.002) and a significant effect of group alone (p < 0.001, Extended Data Table 2-1, line 7) on the search strategies used. The estimated OR (Fig. 2g, top, Extended Data Table 2-1, line 7) for the ANI group to use spatially precise search strategies compared with the SAL group was OR = 0.681 (p = 0.465, here and below, in parentheses following the OR value, the significance level for pairwise comparison between the respective experimental groups with Bonferroni adjustment is shown) on day 6, OR = 0.75 (p = 0.576) on day 7, OR = 0.12 (p < 0.0001) on day 8, OR = 0.29 (p = 0.018) on day 9, OR = 0.17 (p = 0.001) on day 10, and OR = 0.005 (p < 0.0001) on day 11. These data indicate that, starting from day 8 through day 11, animals injected with the protein synthesis inhibitor are less likely to use spatially precise search strategies than the vehicle-treated controls.
We also determined the IPE for each group, defined as the difference between the actual path of the animal in the search of the platform and the ideal path (i.e., a straight line between the release site in the pool and the platform). The use of a more spatially precise search strategy would result in an actual path closer to the ideal path and a smaller IPE. For the IPE, we found a significant effect of treatment × day interaction (p = 0.0007) and a significant effect of treatment alone (p < 0.0001), but no effect of day alone (p = 0.673). Post hoc analysis (with Sidak's correction) revealed higher IPE in ANI mice compared with SAL mice on days 8 (p = 0.0021), 10 (p = 0.003), and 11 (p < 0.0001), confirming that anisomycin injection resulted in a marked decrease in precision of navigation in cdMWM (Fig. 2g, bottom; Extended Data Table 2-1, line 8).
Thus, protein synthesis inhibitor injection into the DG during training in the cdMWM task significantly decreased the chance of the animal to use spatially precise, hippocampus-dependent search strategies, thereby further confirming the hippocampus-dependency of this task.
Learning, memory, and search strategies in the sMWM, rMWM, and cdMWM tasks 6 weeks after irradiation
We next addressed the effects of γ-radiation on learning and memory in a full panel of MWM tasks. The mice were irradiated with 1 or 5 Gy of γ rays or were sham treated; 6 weeks later, they were trained for 5 d to locate a hidden platform in the sMWM task. After a memory test on day 6 with the platform removed, the mice were trained in the rMWM task for an additional 3 d (days 6-8). For that, the platform was returned to the pool, but in the quadrant opposite to that used for sMWM. After a memory test without the platform on day 9, the mice were trained in cdMWM with the blue and white pools and beacons alternating daily for the next 6 d (days 9-14). Memory was tested in the cdMWM on day 15 in both pools with the platform removed. We measured the total distance traveled and average swimming speed in two test trials (after the sMWM task and after the cdMWM task) and found no between-group differences (Fig. 3; Extended Data Table 2-1, lines 112-117), thus indicating normal physical activity of the irradiated mice. The full experimental scheme is shown in Figure 4a.
General activity of irradiated mice in test trials in 6 week experiment. a, Total distance traveled (left) and average velocity (right) in memory test after training in the sMWM task. No significnt between-group differences were found (two-way ANOVA, multiple comparisons with Sidak's correction). b, Total distance traveled (left) and average velocity (right) in memory test after training in the cdMWM task. No significnt between-group differences were found (two-way ANOVA, multiple comparisons with Sidak's correction). Data are mean ± SEM. For statistical analysis, see Extended Data Table 2-1 (lines 112-117).
Behavior in sMWM, rMWM, and cdMWM tasks in mice 6 weeks after irradiation. a, Experimental scheme. B, Blue pool; W, white pool. b–g, Learning and memory in spatial and reversal MWM. b, Escape latencies (time to reach a hidden platform) during training. ****p < 0.0001 for day 5 versus day 1 for each of three groups (two-way ANOVA, multiple comparisons with Sidak's correction). c, Time spent in quadrants (in %) during test following sMWM training; here and below, on the graphs for the test trials (f,i,k): Each column represents an animal. Each color represents the percent of time spent in each quadrant. Black dashed line indicates mean value for the fraction of time spent in each quadrant. Error bar on the mean is approximated with the inverse Fisher information. Black symbols represent the results of Dirichlet distribution analysis. White symbols represent the results of post hoc single-sample t test comparison (with Bonferroni correction) with the theoretical value 25%. Heatmaps represent average distribution of the track points (mouse locations) in the pool during test trial. **p < 0.01. ****p < 0.0001. d, Preference score, relative time spent in T quadrant in test. Dots represent individual values. Horizontal dashed line indicates chance level. e, Escape latencies (time to reach a hidden platform) during reversal learning. ***p < 0.001 for day 8 versus day 6 for each of three groups (two-way ANOVA, multiple comparisons with Sidak's correction). f, Time spent in quadrants (in %) during test following rMWM training. **p < 0.01. ****p < 0.0001. g, Preference score, relative time spent in T quadrant in test. Dots represent individual values. Horizontal dashed line indicates chance level. h–l, Learning and memory in context discrimination MWM task. h, Escape latencies (time to reach a hidden platform) during learning in the blue (B) and white (W) pools. Black asterisks indicate statistically significant difference between the groups. **p < 0.01 (two-way ANOVA, post hoc multiple comparisons with Sidak's correction). Red asterisks indicate statistically significant differences between the 5 Gy and the Sham groups on the respective days. *p < 0.05 for escape latencies 5 Gy versus Sham (two-way ANOVA, multiple comparisons with Sidak's correction). i, Time spent in quadrants (in %) during test following cdMWM training in the blue pool. **p < 0.01. ****p < 0.0001. j, Preference score, relative time spent in T quadrant in test in the blue pool. Dots represent individual values. Horizontal dashed line indicates chance level. Inset, CFC score (latency to reach the false cue – latency to reach the goal cue). k, Time spent in quadrants (in %) during test following cdMWM training in the white pool. **p < 0.01. ****p < 0.0001. l, Preference score, relative time spent in T quadrant in test in the white pool. Dots represent individual values. Horizontal dashed line indicates chance level. Inset, CFC score (latency to reach the false cue – latency to reach the goal cue). *p < 0.05 for 5 Gy versus Sham (Kruskal–Wallis test followed by multiple comparisons with Dunn's correction). N (Sham) = 10, N (1 Gy irradiated) = 14, N (5 Gy irradiated) = 15. b, d, e, g, h, j, l, Data are mean ± SEM. T, Target quadrant; O, opposite quadrant. For statistical analysis, see Extended Data Table 2-1 (lines 17-19, 21-26, 29-34, 37-43).
sMWM
Mice in all groups successfully learned the platform location, as revealed by a significant decrease in escape latencies by the end of training (Fig. 4b; significant effect of day: p < 0.0001, Extended Data Table 2-1, line 17; Sham day 5–day 1: p < 0.0001; 1 Gy day 5–day 1: p < 0.0001; 5 Gy day 5–day 1: p < 0.0001). The escape latencies did not differ between the Sham group and each irradiated group (Extended Data Table 2-1, line 19).
In the memory test with the platform removed, the distribution of time spent in the four quadrants significantly differed from a uniform distribution in all three groups (Fig. 4c; Sham: p < 0.0001; 1 Gy p < 0.0001; 5 Gy p < 0.0001). Further post hoc single-sample t tests revealed a longer time spent in the T quadrant than the theoretical value of 25% (Sham: p < 0.0001, 1 Gy: p = 0.0026, 5 Gy: p < 0.0001). The heatmaps, reflecting the average distributions of the track points (i.e., mouse locations in the pool during the sessions) demonstrate that mice from all experimental groups localized their search to the area of the platform (Fig. 4c). To compare the ability to discriminate quadrants among the groups, we analyzed quadrant preference scores. No overall difference was found between the Sham and irradiated groups (Kruskal–Wallis test followed by between-group comparisons with Dunn's correction, Extended Data Table 2-1, line 23). Thus, in the sMWM task, spatial learning and memory remained intact in the 1 and 5 Gy irradiated mice.
rMWM
We next assessed the relearning ability of the irradiated mice by training them in rMWM (Fig. 4e–g). Animals from all experimental groups learned the new platform location, as revealed by a significant decrease in escape latencies (Fig. 4e; significant effect of day: p < 0.0001, two-way ANOVA, Extended Data Table 2-1, lines 24 and 25; Sham day 8–day 6: p = 0.0006; 1 Gy day 8–day 6: p < 0.0001; 5 Gy day 8–day 6: p < 0.0001). As for sMWM, there was no difference in the escape latencies between the Sham group and each irradiated group (Extended Data Table 2-1, line 27).
In the rMWM memory test, in all experimental groups, the distribution of time spent in four quadrants significantly differed from a uniform distribution (Fig. 4f; Sham: p < 0.0001, 1 Gy: p < 0.0001, 5 Gy: p < 0.0001). Post hoc single-sample t tests indicated a longer time spent in the T quadrant than the theoretical value of 25% in the Sham and 1 Gy groups, with a trend in the 5 Gy group (Sham: p = 0.0012, 1 Gy: p = 0.0021, 5 Gy: p = 0.0628). Post hoc analysis showed that the percent of time spent in the O quadrant did not differ from 25% in all the groups (Extended Data Table 2-1, line 30). The heatmaps demonstrate that the distribution of track points of mice from the irradiated groups was less localized to the area of platform location, whereas the tracks in the Sham group were restricted to that area (Fig. 4f). No overall difference was found between the Sham and irradiated groups in preference scores (Fig. 4g, Kruskal–Wallis test followed by between-group comparisons with Dunn's correction, Extended Data Table 2-1, line 30). Thus, learning and memory in the rMWM task remained intact in the 1 Gy group. In the 5 Gy group, learning was also intact; given that the time spent in the T quadrant and the preference score did not significantly differ from those in the Sham, these results suggest that memory in the 5 Gy group was not affected.
cdMWM
Next, we addressed the ability of the mice to differentiate between daily alternating contexts in the cdMWM task (Fig. 4h–l). Two-way ANOVA of escape latencies indicated no significant interaction between group and day effects; however, there were significant effects of group (p = 0.0158) and day alone (p < 0.0001, Extended Data Table 2-1, line 32). When we assessed the effect of irradiation on escape latency, the post hoc comparison revealed a higher overall escape latency in the 5 Gy group than in the Sham group (p = 0.0086, Fig. 4h). When the daily performance was compared, a longer escape latency was observed in the 5 Gy group than in the Sham group on day 9 in the blue pool (Fig. 4h; 5 Gy–Sham: p = 0.036) and on day 10 in the white pool (Fig. 4h; 5 Gy–Sham: p = 0.03).
In the memory tests, in all groups, the distribution of time spent in four quadrants significantly differed from a uniform distribution (Fig. 4i,k; blue pool-Sham: p < 0.0001, 1 Gy: p < 0.0001, 5 Gy: p < 0.0001; white pool-Sham: p < 0.0001, 1 Gy: p < 0.0001, 5 Gy: p < 0.0001). Post hoc single-sample t tests showed that all groups spent a longer time in the T quadrant than the theoretical value of 25% in both pools (blue pool Sham: p = 0.0011, 1 Gy: p < 0.0001, 5 Gy: p < 0.0001; white pool Sham: p < 0.0001, 1 Gy: p < 0.0001, 5 Gy: p < 0.0001, Extended Data Table 2-1, line 36). The time spent in the O quadrant did not differ from 25% in the blue pool in all the groups but was higher than 25% in the white pool in all the groups (Extended Data Table 2-1, lines 36 and 37). These findings were further illustrated by the heatmaps: in the blue pool, the density of track points was high in the area of platform, but a second focus of higher density of the track points was observed in the opposite quadrant (Fig. 4i); in the white pool, the density of the track points was higher in the area of the platform in all three groups, with a second focus observed in the opposite quadrant in the irradiated groups (Fig. 4k). To further investigate a potential preference for the O quadrant, we performed Wilcoxon matched-pairs test to compare the time spent in the T and in the O quadrants for each group. Mice in all experimental groups spent longer time in the T than in the O quadrant (Extended Data Table 2-1, line 39). No between-group differences in the preference score were observed in either pool (Fig. 4j,l; Extended Data Table 2-1, lines 40 and 41). We also compared the preference scores obtained for the first 10 s of the 60 s test trial and found no between-group differences either (Extended Data Table 2-1, lines 40 and 41). Furthermore, as another approach to assess animals' memory for the cues, we determined the CFC score by subtracting the latency to reach the goal cue from the latency to reach the false cue. We found no difference in the CFC scores between the experimental groups for the blue pool (Fig. 4j, inset, p > 0.05), and a lower score in 5 Gy mice compared with Sham mice observed for the white pool (Fig. 4l, inset, p = 0.0452, Extended Data Table 2-1, lines 42 and 43). Thus, 5 Gy γ-radiation induced a learning deficit in the cdMWM task; however, this deficit was transient and did not preclude subsequent memory formation in this task.
In summary, in the 6 week experiment, mice exposed to 1 and 5 Gy of γ rays showed intact learning in the spatial and reversal MWM tasks, but mice exposed to 5 Gy showed transiently delayed learning in the cdMWM. In both irradiated groups, memory was not affected after the sMWM, rMWM, and cdMWM training.
Search strategies
We next performed a more detailed analysis of the distinct navigation strategies used by the irradiated mice, similar to that applied to the ANI and SAL groups. In the first trial of training in the sMWM, all groups were almost exclusively thigmotaxic. The contribution of this inefficient strategy decreased by the last trial on day 1 and was virtually absent by day 5 in each group (Fig. 5a–c, sMWM). This was accompanied by an increased contribution of the efficient spatially precise strategies in each group by day 5. However, the contribution of the distinct strategies to the group's performance in sMWM differed between the Sham and the irradiated groups. The proportion of efficient strategies, such as directed search and focal search, was higher in the Sham than in the irradiated groups. This difference was more evident when the results were presented as the relative difference in each strategy with respect to that in the control Sham group (Fig. 5d,e, sMWM).
Qualitative and quantitative analysis of strategies used in sMWM, rMWM, and cdMWM in mice 6 weeks after irradiation. For sMWM and rMWM, the data are shown for the first and the last day of training in each task. For cdMWM, the data are shown for all days of training. Each block of stacked bars represents strategies used for the 5 trials for each day. B, Blue pool; W, white pool. a, Search strategies for Sham mice. b, Search strategies for 1 Gy mice. c, Search strategies for 5 Gy mice. d, Difference in contribution of search strategies between the Sham and the 1 Gy groups. e, Difference in contribution of search strategies between the Sham and the 5 Gy groups. f, Ideal path error on days 1 and 5 of sMWM training. Data are mean ± SEM. g, Ideal path error on days 6 and 8 of rMWM training. Data are mean ± SEM. h, Left, Ideal path error on days 9-14 of cdMWM training. *p < 0.05; **p < 0.01; two-way ANOVA, multiple comparisons with Sidak's correction. Data are mean ± SEM. Right, OR of spatially precise strategies use in 5 Gy compared with the Sham in cdMWM. **p < 0.01 (determined by fitting a generalized linear mixed effect model with binomial distribution). For statistical analysis, see Extended Data Table 2-1 (lines 20, 27, 28, 35, 36).
Training in the rMWM task, when the platform location was changed, initially resulted in a higher percentage of spatially imprecise strategies in all groups (Fig. 5a–c, rMWM, day 6, Trial 1), suggesting errors in navigation. In Sham mice, the fraction of spatially precise strategies was higher than in 5 Gy irradiated mice. However, with further training, the proportion of spatially precise strategies increased in all experimental groups (from day 6 to day 8, Fig. 5a–e, rMWM).
GLMM analysis applied to the pulled data for strategies used in the sMWM and rMWM tasks did not reveal the effect of interaction between the irradiation dose and the day (p = 0.559) or of irradiation dose alone (p = 0.324) but showed a significant effect of day (p < 0.001, Extended Data Table 2-1, line 28). Thus, irradiation did not affect the chance of using spatially precise search strategies compared with spatially imprecise strategies in the sMWM and rMWM tasks.
A marked decrease in the IPE was observed from day 1 to day 5 of the sMWM task (Fig. 5f), and from day 6 to day 8 of the rMWM task (Fig. 5g), reflecting improved performance during learning; however, the IPE did not differ between the groups (Extended Data Table 2-1, lines 20 and 27).
Finally, in the cdMWM task, introduction of the local cues significantly increased the contribution of the highly precise search strategies, such as direct path, from the first trial on day 9 (Fig. 5a–c, cdMWM). However, compared with the Sham group, a notable delay in the use of spatially precise strategies was observed in the 1 and 5 Gy groups; the difference in the strategies used gradually decreased by the end of training (Fig. 5d,e, cdMWM). GLMM revealed a significant effect of irradiation dose or day alone (p = 0.049 and p < 0.001, respectively), but no interaction between the treatment and the day (p = 0.591, Extended Data Table 2-1, line 35). The estimated ORs for the 5 Gy group compared with the Sham group (Fig. 5h, right) were as follows: OR = 0.26 (p = 0.004) on day 9, OR = 0.56 (p > 0.05) on day 10, OR = 0.67 (p > 0.05) on day 11, OR = 0.64 (p > 0.05) on day 12, OR = 1.08 (p > 0.05) on day 13, and OR = 0.87 (p > 0.05) on day 14. The estimated ORs for the 1 Gy group did not differ significantly compared with the Sham group (for all days p > 0.05, Extended Data Table 2-1, line 35). This observation of the lower odds of using spatially precise search strategies by the 5 Gy group on day 9 of training in the cdMWM task was corroborated by a significantly higher IPE in the 5 Gy group than in the Sham group at the beginning of the training period (days 9 and 10), but also on day 14 (Fig. 5h, left; Extended Data Table 2-1, line 36).
Notably, the magnitude of differences between the 5 Gy and the Sham group in using the efficient spatially precise strategies was greatest in the cdMWM task, thus suggesting that this task is more sensitive to the effects of radiation than sMWM or rMWM.
In summary, our analysis of search strategies revealed a transition from inefficient to more efficient spatially precise strategies over the course of training in all three tasks, in both the control and the irradiated groups. However, the cdMWM task was able to reveal a delay in learning in the irradiated groups. This delay is supported by the observation of increased escape latencies, lower odds of using spatially precise search strategies early in the task, and increased IPE induced by irradiation.
Learning, memory, and search strategies in the sMWM, rMWM, and cdMWM tasks 1 week after irradiation
The period 4-8 weeks after birth is critical for newborn hippocampal neurons to integrate into preexisting neuronal circuits in the adult brain (Denoth-Lippuner and Jessberger, 2021). Therefore, the results obtained in the MWM 6-8 weeks after 5 Gy γ radiation are compatible with the potential involvement of hippocampal neurogenesis in the functions tested in the cdMWM task. To further examine whether the observed behavioral deficits might be confined to a certain time window, we irradiated two cohorts of mice at 5 Gy and trained them in the same paradigms as described above, at 1 week (Fig. 6a) or 4 weeks (see Fig. 8a) after irradiation.
Behavior in sMWM, rMWM, and cdMWM tasks in mice 1 week after irradiation. a, Experimental scheme. B, Blue pool; W, white pool. b–g, Learning and memory in spatial and reversal MWM. b, Escape latencies (time to reach a hidden platform) during training. ****p < 0.0001 for day 5 versus day 1 for each of two groups (two-way ANOVA, multiple comparisons with Sidak's correction). c, Time spent in quadrants (in %) during test following sMWM training; here and below, on the graphs for the test trials (f,i,k): Each column represents an animal. Each color represents the percent of time spent in each quadrant. Black dashed line indicates mean value for the fraction of time spent in each quadrant. Error bar on the mean is approximated with the inverse Fisher information. Black symbols represent the results of Dirichlet distribution. White symbols represent the results of post hoc single-sample t test comparison (with Bonferroni correction) with the theoretical value 25%. Heatmaps represent average distribution of the track points (mouse locations) in the pool during test trial. ****p < 0.0001. d, Preference score, relative time spent in T quadrant in test. Dots represent individual values. Horizontal dashed line indicates chance level. e, Escape latencies (time to reach a hidden platform) during reversal learning. ****p < 0.0001 for day 8 versus day 6 for each of two groups (two-way ANOVA, multiple comparisons with Sidak's correction). f, Time spent in quadrants (in %) during test following rMWM training. **p < 0.01. ****p < 0.0001. g, Preference score, relative time spent in T quadrant in test. Dots represent individual values. Horizontal dashed line indicates chance level. h–l, Learning and memory in context discrimination MWM task. h, Escape latencies (time to reach a hidden platform) during learning in the blue (B) and white (W) pools. i, Time spent in quadrants (in %) during test following cdMWM training in the blue pool. ***p < 0.001. ****p < 0.0001. j, Preference score, relative time spent in T quadrant in test in the blue pool. Dots represent individual values. Horizontal dashed line indicates chance level. Inset, CFC score (latency to reach the false cue – latency to reach the goal cue). k, Time spent in quadrants (in %) during test following cdMWM training in the white pool. ****p < 0.0001. l, Preference score, relative time spent in T quadrant in test in the white pool. Dots represent individual values. Horizontal dashed line indicates chance level. Inset, CFC score (latency to reach the false cue – latency to reach the goal cue). N (Sham) = 11, N (5 Gy irradiated) = 14. b, d, e, g, h, j, l, Data are mean ± SEM. T, Target quadrant; O, opposite quadrant. For statistical analysis, see Extended Data Table 2-1 (lines 47-49, 51-56, 59-65, 68-74).
sMWM
Mice in both the 5 Gy and Sham groups learned to locate the platform in the sMWM task 1 week after irradiation, as revealed by the decrease in escape latencies over 5 training days (Fig. 6b; significant effect of day: p < 0.0001, Extended Data Table 2-1, line 47; Sham day 5–day 1: p < 0.0001; 5 Gy day 5–day 1: p < 0.0001). The escape latencies did not differ between the 5 Gy and Sham groups (Extended Data Table 2-1, line 49).
In the memory test without the platform, in both groups, the distribution of time spent in each of the four quadrants significantly differed from a uniform distribution (Fig. 6c; Sham: p < 0.0001; 5 Gy: p < 0.0001, Extended Data Table 2-1, line 51), with the post hoc single-sample t tests indicating longer time spent in the T quadrant, than the theoretical value of 25% (T quadrant Sham: p < 0.0001, 5 Gy: p < 0.0001, Extended Data Table 2-1, line 52). The heatmaps demonstrate that the mice in both experimental groups localized their search to the area of the platform (Fig. 6c). The preference scores did not differ between the groups (Fig. 6d, Extended Data Table 2-1, line 53). Thus, exposure to 5 Gy 1 week before the sMWM training did not affect learning and memory in the irradiated mice.
rMWM
After the sMWM, the mice were trained in the rMWM task. Mice from both groups learned to locate the hidden platform in the new quadrant, as revealed by the decrease in escape latencies over the course of 3 training days (Fig. 6e; a significant day effect: p < 0.0001, two-way ANOVA, Extended Data Table 2-1, line 54; Sham day 8–day 6: p < 0.0001; 5 Gy day 8–day 6: p < 0.0001); no difference was observed between the groups on either day (Extended Data Table 2-1, line 56).
When memory was tested without the platform, in both groups, the distribution of the time spent in each quadrant significantly differed from a uniform distribution (Fig. 6f; Sham: p = 0.0011; 5 Gy: p < 0.0001). In both groups, the post hoc single-sample t tests showed a longer time spent in the T quadrant than the theoretical value of 25% (Sham: p = 0.0018, 5 Gy: p = 0.0043) and no difference in the time spent in the O quadrant and the 25% (Extended Data Table 2-1, line 60). The heatmaps demonstrate that mice from both experimental groups localized their search to the area of the platform (Fig. 6f). No difference was observed in the preference scores between the groups (Fig. 6g; Extended Data Table 2-1, line 61). Thus, mice exposed to 5 Gy demonstrated normal reversal learning and memory.
cdMWM
The mice were next trained in the cdMWM task. Two-way ANOVA of escape latencies indicated no significant interaction between group and day effects or group effect alone (Extended Data Table 2-1, line 62), but demonstrated a significant day effect (p = 0.0013). The escape latency decreased in the Sham group between days 9 and 11, and days 9 and 13 in the blue pool (Sham day 9–day 11: p = 0.0284, day 9–day 13: p = 0.006). The between-group comparison did not indicate differences between the 5 Gy and Sham groups in either pool (Extended Data Table 2-1, lines 64 and 65).
In the memory test without the platform (Fig. 6i–l), the distribution of time spent in each quadrant significantly differed from a uniform distribution in both groups in either pool (Fig. 6i,k; blue pool Sham: p < 0.0001, 5 Gy: p < 0.0001; white pool Sham: p < 0.0001, 5 Gy: p < 0.0001). Post hoc single-sample t tests demonstrated that both groups spent longer than 25% time in the T quadrant in either pool (blue pool Sham: p < 0.0001, 5 Gy: p = 0.0002; white pool Sham: p < 0.0001, 5 Gy: p < 0.0001). The time spent in the O quadrant did not differ from 25% (Extended Data Table 2-1, lines 69 and 70). The heatmaps demonstrate that the mice in all experimental groups localized their search to the area of the platform (Fig. 6i,k). No between-group differences were observed in preference score in either pool (Fig. 6j,l; Extended Data Table 2-1, lines 71 and 72). When we compared preference scores obtained for the first 10 s of the test trial, in the blue pool (but not in the white pool) a lower preference score was found in the 5 Gy group than in the Sham group (Extended Data Table 2-1, line 71). We found no difference for the CFC score between the experimental groups in either pool (Fig. 6j,l, insets, Extended Data Table 2-1, lines 73 and 74). In summary, our data on the sMWM, rMWM, and cdMWM tasks indicate normal learning and memory in the 1 week experiment following 5 Gy γ-irradiation.
Search strategies
Analysis of search strategies demonstrated that, similarly to the findings of the 6 week experiment, over the first day of training in the sMWM, both the Sham and the 5 Gy groups were mainly thigmotaxic; the contribution of this strategy decreased by the last trial on day 1, and further training increased the contribution of the spatially precise strategies (Fig. 7a–c, sMWM). Training in the rMWM task initially resulted in a higher percentage of spatially imprecise strategies in both groups (Fig. 7a,b, rMWM, day 6, Trials 1 and 2), suggesting errors in navigation. With training continued, however, the fraction of spatially precise strategies in both experimental groups increased by day 8. GLMM analysis of pulled strategies data for sMWM and rMWM did not reveal the effect of interaction between the treatment and the day (p = 0.96) or of treatment alone (p = 0.338), but there was a significant effect of day (p < 0.001, Extended Data Table 2-1, line 58). Thus, irradiation did not affect the chance of using spatially precise search strategies over imprecise search strategies in the sMWM and rMWM tasks. No between-group differences in the IPE were found in either task (Fig. 7d,e, Extended Data Table 2-1, lines 50 and 57).
Qualitative and quantitative analysis of strategies used in sMWM, rMWM, and cdMWM in mice 1 week after irradiation. For sMWM and rMWM, the data are shown for the first and the last day of training in each task. For cdMWM, the data are shown for all days of training. Each block of stacked bars indicates strategies used for the 5 trials for each day. B, Blue pool; W, white pool. a, Search strategies for Sham mice. b, Search strategies for 5 Gy mice. c, Difference in contribution of search strategies between the Sham and the 5 Gy groups. d, Ideal path error on days 1 and 5 of sMWM training. e, Ideal path error on days 6 and 8 of rMWM training. f, Ideal path error on days 9-14 of cdMWM training. d–f, Data are mean ± SEM. For statistical analysis, see Extended Data Table 2-1 (lines 50, 57, 58, 66, 67).
Behavior in sMWM, rMWM, and cdMWM tasks in mice 4 weeks after irradiation. a, Experimental scheme. B, Blue pool; W, white pool. b–g, Learning and memory in spatial and reversal MWM. b, Escape latencies (time to reach a hidden platform) during training. ****p < 0.0001 for day 5 versus day 1 for each of two groups (two-way ANOVA, multiple comparisons with Sidak's correction). c, Time spent in quadrants (in %) during test following sMWM training; here and below, on the graphs for the test trials (f,i,k). Each column represents an animal. Each color represents the percent of time spent in each quadrant. Black dashed line indicates mean value for the fraction of time spent in each quadrant. Error bar on the mean is approximated with the inverse Fisher information. Black symbols represent the results of Dirichlet distribution analysis. White symbols represent the results of post hoc single-sample t test comparison (with Bonferroni correction) with the theoretical value 25%. Heatmaps represent average distribution of the track points (mouse locations) in the pool during test trial. **p < 0.01. ****p < 0.0001. d, Preference score, relative time spent in T quadrant in test. Dots represent individual values. Horizontal dashed line indicates chance level. e, Escape latencies (time to reach a hidden platform) during reversal learning. **p < 0.01 for day 8 versus day 6 for each of two groups (two-way ANOVA, multiple comparisons with Sidak's correction). f, Time spent in quadrants (in %) during test following rMWM training. *p < 0.05. **p < 0.01. g, Preference score, relative time spent in T quadrant in test. Dots represent individual values. Horizontal dashed line indicates chance level. h–l, Learning and memory in context discrimination MWM task. h, Escape latencies (time to reach a hidden platform) during learning in the blue (B) and white (W) pools. Black asterisks indicate statistically significant difference between the groups. **p < 0.01 (two-way ANOVA, post hoc multiple comparisons with Sidak's correction). Red represents p values for the escape latencies comparison between the 5 Gy and the Sham groups on the respective days (two-way ANOVA, multiple comparisons with Sidak's correction). i, Time spent in quadrants (in %) during test following cdMWM training in the blue pool. ***p < 0.001. ****p < 0.0001. j, Preference score, relative time spent in T quadrant in test in the blue pool. Dots represent individual values. Horizontal dashed line indicates chance level. Inset, CFC score (latency to reach the false cue – latency to reach goal cue). k, Time spent in quadrants (in %) during test following cdMWM training in the white pool. *p < 0.05. **p < 0.01. ****p < 0.0001. l, Preference score, relative time spent in T quadrant in test in the white pool. Dots represent individual values. Horizontal dashed line indicates chance level. *p < 0.05 (Mann–Whitney t test). Inset, CFC score (latency to reach the false cue – latency to reach goal cue). **p < 0.01 for 5 Gy versus Sham. N (Sham) = 8, N (5 Gy irradiated) = 10. b, d, e, g, h, j, l, Data are mean ± SEM. T, Target quadrant; O, opposite quadrant. For statistical analysis, see Extended Data Table 2-1 (lines 75-77, 79-84, 87-92, 95-101).
Finally, in the cdMWM task, we observed a moderate increase in the fraction of spatially precise strategies and a concomitant decrease in the fraction of imprecise strategies over the training days in both groups (Fig. 7a,b, cdMWM). Although some differences were observed in strategies contributing to groups' performance between Sham and irradiated mice, they were inconsistent and small in magnitude (Fig. 7c, cdMWM). GLMM analysis of search strategies did not reveal the effect of interaction between the treatment and day (p = 0.3) or of treatment alone (p = 0.889); there was a significant effect of day (p < 0.001, Extended Data Table 2-1, line 66). Thus, irradiation did not affect the chance of using spatially precise search strategies compared with spatially imprecise strategies in the cdMWM task. The IPE predictably decreased over the course of training, with no difference observed between the groups (Fig. 7f; Extended Data Table 2-1, line 67).
In summary, at 1 week after irradiation, the repertoire and dynamics of change of search strategies used to navigate the pools were highly similar between the irradiated and control mice.
Learning, memory, and search strategies in the sMWM, rMWM, and cdMWM tasks 4 weeks after irradiation
sMWM
When trained in the sMWM task 4 weeks after irradiation, both 5 Gy and Sham irradiated mice successfully learned the platform location, showing a gradual decrease in escape latency (Fig. 8b; significant day effect: p < 0.0001, Extended Data Table 2-1, line 75; Sham day 5–day 1: p < 0.0001; 5 Gy day 5–day 1: p < 0.0001), which did not differ between the groups (Extended Data Table 2-1, line 77).
In the memory test without the platform, in both groups, the distribution of time spent in each quadrant significantly differed from a uniform distribution (Fig. 8c; Sham: p < 0.0001; 5 Gy: p < 0.0001). Post hoc single-sample t tests revealed a longer time spent by both groups in the T quadrant than the theoretical value of 25% (T quadrant Sham: p = 0.001, 5 Gy: p = 0.0044). The heatmaps demonstrate that mice in both experimental groups localized their search to the area of the platform (Fig. 8c). We observed no difference in preference scores between the 5 Gy and Sham groups (Fig. 8d, Extended Data Table 2-1, line 81). Thus, 4 weeks after irradiation, the irradiated and control mice were equally successful in the learning and memory tests of the sMWM task.
rMWM
In the rMWM task, animals from the Sham and the 5 Gy groups successfully learned the new platform location as revealed by the decrease in escape latencies over the course of three training days (Fig. 8e; significant effect of day: p = 0.0001, two-way ANOVA, Extended Data Table 2-1, line 82; Sham day 8–day 6: p = 0.0051; 5 Gy day 8–day 6: p = 0.0033), and no difference was observed in the escape latencies between the groups (Extended Data Table 2-1, line 84).
In the memory test without the platform, in both groups, the distribution of the time spent in each quadrant was significantly different from uniform (Fig. 8f; Sham: p = 0.0197, 5 Gy: p = 0.00202). For both groups, the post hoc single-sample t tests showed that the time spent in the T quadrant significantly exceeded 25% (Sham: p = 0.017, 5 Gy: p = 0.014), whereas no difference was observed in the time spent in the O quadrant (Extended Data Table 2-1, line 88). The heatmaps demonstrate that mice in both experimental groups localized their search to the area of the platform (Fig. 8f). The preference scores did not differ between the Sham and the 5 Gy groups (Fig. 8g; Extended Data Table 2-1, line 89). Thus, 4 weeks after exposure, mice exposed to 5 Gy demonstrate normal reversal learning and memory and did not differ from the Sham group in the rMWM task.
cdMWM
In the cdMWM task, two-way ANOVA of escape latencies indicated no significant interaction between the group and day effects in either pool (Extended Data Table 2-1, line 90) but demonstrate a significant effects of group (p = 0.0014) and day alone (p = 0.0029) (Fig. 8h). In the 5 Gy group, a decrease in escape latencies over the training days was observed in the blue pool (day 9–day 13: p = 0.001). In the Sham group, no difference was observed in escape latencies during training in both pools (Extended Data Table 2-1, line 91). The between-group difference in escape latencies on days 9 (blue pool) and 10 (white pool) showed a tendency to increase in the 5 Gy group but did not reach the significance level (p = 0.0735, p = 0.0619, respectively, Extended Data Table 2-1, line 92).
In the memory tests, the distribution of time spent in the four quadrants significantly differed from a uniform distribution in both groups in either pool (Fig. 8i,k; blue pool Sham: p < 0.0001, 5 Gy p < 0.0001; white pool Sham: p < 0.0001, 5 Gy: p < 0.0001). Post hoc single-sample t tests indicated that both groups spent >25% of the time in the T quadrant in either pool (blue pool Sham: p = 0.0001, 5 Gy: p = 0.0007; white pool Sham: p = 0.0012, 5 Gy: p < 0.0001). In the blue pool, the time spent in the O quadrant did not differ significantly from 25% (Extended Data Table 2-1, line 96), whereas in the white pool, it was lower than 25% in the Sham group, but not in the 5 Gy group (Sham: p = 0.014, 5 Gy: p = 0.49, Extended Data Table 2-1, line 97). The heatmaps demonstrate that mice from the Sham group localized their search to the area of the platform in both pools (Fig. 8i,k), whereas in the 5 Gy group, two foci of distributions of the track points were observed in both pools (Fig. 8i,k). In addition, the preference score in the blue pool did not differ between the groups (Fig. 8j, Extended Data Table 2-1, line 98); however, in the white pool, it was lower in the 5 Gy group than in the Sham group (Fig. 8l; p = 0.041, Mann–Whitney unpaired t test). When we compared the preference scores of the first 10 s of the test trial, we did not find differences between the groups in either pool (Extended Data Table 2-1, lines 98 and 99). However, the CFC score was lower in the 5 Gy group than in the Sham group in the white (but not blue) pool (p = 0.0062, p = 0.272, respectively Fig. 8j,l, insets; Extended Data Table 2-1, lines 100 and 101). Thus, 5 Gy irradiated mice demonstrated delayed learning, and memory deficit in the white pool in the cdMWM task 6 weeks after irradiation exposure.
Overall, in the 4 week experiment, mice exposed to 5 Gy irradiation demonstrated intact learning in the spatial and reversal MWM tasks, but transiently delayed learning in the cdMWM task. Memory in the 5 Gy mice did not differ from that in the control mice in the sMWM and rMWM tests but was deficient in the cdMWM test.
Search strategies
Similarly to the results of the 6 and 1 week experiments, both groups in the 4 week experiment were thigmotaxic over the first day of training in the sMWM. The fraction of this strategy decreased by the last trial on day 1, was absent on day 5, and was accompanied by an increased contribution of the spatially precise strategies (Fig. 9a,b, sMWM).
Qualitative and quantitative analysis of strategies used in sMWM, rMWM, and cdMWM in mice 4 weeks after irradiation. For sMWM and rMWM, the data are shown for the first and the last day of training in each task. For cdMWM, the data are shown for all days of training. Each block of stacked bars indicates strategies used for the 5 trials for each day. B, Blue pool; W, white pool. a, Search strategies for Sham mice. b, Search strategies for 5 Gy mice. c, Difference in contribution of search strategies between the Sham and the 5 Gy groups. d, Ideal path error on days 1 and 5 of sMWM training. e, Ideal path error on days 6 and 8 of rMWM training. f, Left, Ideal path error on days 9-14 of cdMWM training. *p < 0.05 (two-way ANOVA, multiple comparisons with Sidak's correction). Right, OR of spatially precise strategies use in 5 Gy compared with the Sham in cdMWM. *p < 0.05; **p < 0.01; determined by fitting a generalized linear mixed effect model with binomial distribution. d–f, Left, Data are mean ± SEM. For statistical analysis, see Extended Data Table 2-1 (lines 78, 85, 86, 93, 94).
Training in the rMWM task resulted in a higher initial percentage of spatially imprecise strategies in both groups (Fig. 9a,b, rMWM, day 6), suggesting errors in navigation. By the end of training in rMWM, the fraction of the spatially precise strategies in both experimental groups increased (Fig. 9a,b, rMWM, day 8). In both groups, no clear preference for the use of spatially precise strategies (Fig. 9c, sMWM and rMWM), and no between-group differences in the IPE (Fig. 9d,e, Extended Data Table 2-1, lines 78 and 85) were observed.
The GLMM analysis of the pulled data for strategies used in the sMWM and rMWM tasks did not reveal the effect of interaction between the treatment and day (p = 0.663) or of the treatment alone (p = 0.915) but revealed a significant effect of day (p < 0.001, Extended Data Table 2-1, line 86). Thus, irradiation did not affect the chance of using spatially precise search strategies compared with spatially imprecise strategies in the sMWM and rMWM tasks.
Finally, in the cdMWM task, in both groups, the fraction of spatially imprecise strategies was higher at the beginning of the training (days 9 and 10), but decreased by days 13 and 14 (Fig. 9a,b, cdMWM). A consistent and highly pronounced difference was observed between the 5 Gy and Sham groups, with the irradiated mice using fewer efficient search strategies during the entire training in the cdMWM task (Fig. 9c). GLMM analysis revealed a significant effect of either the treatment or day alone (p < 0.001 and p = 0.016, respectively), but no treatment × day interaction (p = 0.959). The OR for the 5 Gy group compared with the Sham group (Fig. 9f, right) was OR = 0.443 (p = 0.061) on day 9, OR = 0.28 (p = 0.004) on day 10, OR = 0.34 (p = 0.027) on day 11, OR = 0.38 (p = 0.044) on day 12, OR = 0.56 (p = 0.28) on day 13, and OR = 0.41 (p = 0.058) on day 14. This finding was in agreement with the significantly higher IPE in the 5 Gy group than the Sham group on days 9 and 10 of training (Fig. 9f, left; Extended Data Table 2-1, line 94).
Thus, similarly to the results of the 6 week experiment (Fig. 5h, right), irradiated mice in the 4 week experiment demonstrated decreased odds of using spatially precise strategies in cdMWM, indicating that this task is particularly prone to disruption by irradiation compared with the sMWM and rMWM tasks.
In summary, our analysis of search strategies at 1, 4, and 6 weeks after exposure to γ rays revealed a transition from nonspecific spatially imprecise strategies to the spatially precise ones with progressive training in each experimental group in each training paradigm (sMWM, rMWM, and cdMWM). However, the cdMWM task was particularly advantageous in revealing the differences in strategy use between the groups. The higher sensitivity of cdMWM further supports and expands the conclusions based on the escape latencies, that learning is deficient at 6 and 8, but not 3 weeks after irradiation.
γ irradiation results in long-term impairment of hippocampal neurogenesis
To assess the effect of our irradiation procedure on hippocampal neurogenesis, we examined the cohorts of newborn cells accumulated throughout 8 weeks after irradiation in the DG in mice subjected to behavior testing at 6-8 weeks (i.e., 6 weeks between irradiation and beginning of training, followed by 2 weeks of training; Fig. 4). To differentiate between mature and immature neuronal cells, we injected a synthetic thymidine analog EdU 2 h before the irradiation and inspected hippocampal sections for labeling with EdU and the expression of NeuN and DCX, markers of differentiated and immature neurons, respectively. Older neuronal cell cohorts in the DG (i.e., the progeny of cells dividing at the time of irradiation) were identified based on the presence of EdU and NeuN, but not DCX, signals; newer cell cohorts corresponding to immature neurons were identified by the presence of DCX, but not NeuN or EdU, signals.
Irradiation with either 1 or 5 Gy dose resulted in lower numbers of total EdU-labeled cells in the DG than were observed in the Sham control group (Fig. 10a,b; main effect revealed by one-way ANOVA: p < 0.0001; between-group pair-wise comparison with Sidak's correction: 1 Gy–Sham: p = 0.0007; 5 Gy–Sham: p < 0.0001, Extended Data Table 2-1, line 44). The number of EdU+ cells that became mature neurons (EdU+NeuN+DCX– cells) was diminished in the 1 Gy group, and those cells were not detected in the 5 Gy group (Fig. 10c; main effect revealed by one-way ANOVA: p = 0.0002; between-group pair-wise comparison with Sidak's correction: 1 Gy–Sham: p = 0.0016; 5 Gy–Sham: p = 0.0004). Of note, 4 of 7 mice in the 1 Gy group and all 7 mice in the 5 Gy group display no EdU+NeuN+DCX– cells (Fig. 10c). No EdU+ cells in the 5 Gy group carried neuronal markers DCX or NeuN. Furthermore, no EdU+DCX+ cells were detected in all three groups. Finally, the number of DCX+ cells was significantly lower in the 5 Gy-irradiated mice than in the Sham and 1 Gy-irradiated mice (Fig. 10d; p < 0.0001, one-way ANOVA; 5 Gy–Sham: p < 0.0001; 1 Gy–5 Gy: p = 0.0051, between-group pair-wise comparisons with Sidak's correction, Extended Data Table 2-1, line 45).
Neurogenesis at 8 weeks after irradiation. a, Representative images of DG of Sham, 1 Gy- and 5 Gy-irradiated mice. Arrowheads indicate EdU+ cells. Scale bar, 100 µm. b, Number of EdU+ cells. EdU was injected 2 h before irradiation. ***p < 0.001; ****p < 0.0001; one-way ANOVA, multiple comparisons with Sidak's correction. c, Number of EdU+NeuN+DCX– cells (mature new neurons) per DG. **p < 0.01; ***p < 0.001; one-way ANOVA, multiple comparisons with Sidak's correction. Although we have 0 cells in the 5 Gy group, we used one-way ANOVA taking into account that our cell count is a result of sampling and may not correspond to an actual zero. d, Number of DCX+ cells (immature new neurons) per DG. **p < 0.01; ****p < 0.0001; one-way ANOVA, multiple comparisons with Sidak's correction. Data are mean ± SEM. For statistical analysis, see Extended Data Table 2-1 (lines 44-46).
Together, these results indicate that the number of adult-born neurons generated and differentiated within an 8 week interval after irradiation markedly decreased after exposure to 1 or 5 Gy of γ rays. In addition, the absence of EdU+DCX+ cells in all three groups suggests that stem and progenitor cells, with or without irradiation, did not reenter the cell cycle and engage in neuronal differentiation 6-8 weeks after they were labeled with EdU. Furthermore, a decrease in the number of DCX-expressing immature neurons in the 5 Gy group at 8 weeks after irradiation suggests that exposure to γ rays induced a long-lasting impairment in neuronal production by the neurogenic niche.
cdMWM-induced c-Fos activation in different populations of young neurons
Next, we investigated the functional involvement of new neurons in the cdMWM task. Because the integration of new neurons into the preexisting circuitry requires a certain degree of maturity and may be confined to a critical time window (Denoth-Lippuner and Jessberger, 2021), we evaluated the expression of the neuronal activation marker c-Fos in young neurons of two ages: 7 and 12 weeks. To identify these neuronal populations, we labeled young neurons with two synthetic thymidine analogs: BrdU at 11 weeks before the start of the behavioral experiments and EdU at 6 weeks before the start of the experiments (Fig. 11a).
Expression of c-Fos in adult-born neurons of the DG. a, Experimental scheme. Groups: Control, c-Fos+ cells were examined after 2 additional days of sMWM training; New Paradigm, c-Fos+ cells were examined after a single day of cdMWM training in the blue (B) pool with the beacons available; New Context, c-Fos+ cells were examined after 2 d of cdMWM training in both blue (B) and white (W) pools. b, c, Representative images of c-Fos expression (green) colocalized with thymidine analogs (red) EdU (b) and BrdU (c). Scale bar, 5 µm. d, Escape latencies (time to reach a hidden platform) during training. ****p < 0.0001, day 5 versus day 1 for each of three groups (two-way ANOVA, multiple comparisons with Sidak's correction). e, Total number of c-Fos+ cells in the DG. **p < 0.01 (one-way ANOVA, multiple comparisons with Sidak's correction). f, Number of EdU+-cells in the DG. g, Number of BrdU+ cells in the DG. h, Number of EdU+c-Fos+ (7-week-old) cells in the DG. i, Number of BrdU+c-Fos+ (12-week-old) cells in the DG. *p < 0.05 (one-way ANOVA, multiple comparisons with Sidak's correction). Data are mean ± SEM. For statistical analysis, see Extended Data Table 2-1 (lines 102-111).
All mice were trained in the sMWM task for 5 consecutive days. After the sMWM training, the animals were divided into three groups (Fig. 11a), and the expression of c-Fos was examined after the last trial in the following behavioral setting: (1) Control: c-Fos+ cells were examined after 2 additional days of sMWM training when the expression of c-Fos was expected to decrease because of the decreased novelty; (2) New Paradigm: c-Fos+ cells were examined after a single day of cdMWM training in the blue pool with the beacons available; and (3) New Context: c-Fos+ cells were examined after 2 d of cdMWM training in both pools (blue and white).
Behavioral testing indicated that the animals successfully learned in the sMWM and cdMWM tasks (Fig. 11d; Extended Data Table 2-1, lines 102-106). We next examined the total number of c-Fos-expressing cells (Fig. 11e), the number of cells labeled with each of the thymidine analogs (Fig. 11f,g), and the number of c-Fos-expressing cells colabeled with each thymidine analog (Fig. 11b,c,h,i). Among the total neuronal population of the DG, an increase in c-Fos+ cells was observed only in the New Context group (Fig. 11e; p = 0.0013, one-way ANOVA; New Context–Control: p = 0.0024, New Context–New Paradigm: p = 0.0048, between-group pair-wise comparisons with Sidak's correction, Extended Data Table 2-1, line 107). Among the younger (7-week-old) adult-born neurons (i.e., EdU+ cells), training in neither setting increased the number of c-Fos-expressing cells (Fig. 11h; Extended Data Table 2-1, line 108). Among the older (12-week-old) adult-born neurons (i.e., BrdU+ cells), training in both the New Paradigm and the New Context settings increased the number of c-Fos-expressing cells (Fig. 11i; p = 0.022, one-way ANOVA; New Context–Control: p = 0.023, New Paradigm–Control: p = 0.042, between-group pair-wise comparisons with Sidak's correction, Extended Data Table 2-1, line 109). Analysis of the number of EdU+ and BrdU+ cells did not reveal intergroup differences for 7- and 12-week-old labeled cell cohorts in the DG (Fig. 11f and Fig. 11g, respectively; Extended Data Table 2-1, lines 110 and 111). Thus, our results suggest that neuronal subpopulations in the DG are differentially recruited by new experience: the total population of neuronal cells in the DG (of which the predominant fraction corresponds to preexisting neurons) responded to the new context with a significant increase in c-Fos expression; the cohort of older adult-born neurons was activated by both the New Context and New Paradigm settings, and the cohort of younger adult-born neurons was not preferentially activated by either of the settings.
Discussion
Here we show that exposure to a moderate dose of γ-radiation affects animals' cognitive flexibility: irradiated mice, while being able to perform conceivably simpler tasks by using distal spatial cues, showed deficiencies when they were required to combine spatial and nonspatial cues and contexts and readapt their repertoire of search strategies.
Exposure to 1 or 5 Gy of γ rays did not affect learning or memory when mice were tested in conventional sMWM and rMWM tasks. We then asked whether a more challenging task might be more sensitive in revealing irradiation-induced changes. We designed the cdMWM task as a new hippocampus-dependent task in which animals rely on allocentric and egocentric reference frames and learn new combinations of spatial and nonspatial cues in modified contexts to find the escape platform. cdMWM conceivably poses greater demands on animals' ability to reassess and adjust their search strategies; therefore, performance in this task may be a better indicator of animals' cognitive flexibility than conventional sMWM and rMWM.
Indeed, in irradiated mice, the cdMWM task allowed us to detect deficits in learning, including higher escape latency, a higher reliance on inefficient spatially imprecise non–hippocampal-dependent strategies versus spatially precise hippocampal- and neurogenesis-dependent strategies (Garthe et al., 2009), and a higher IPE, as well as impaired memory. These deficits were evident 6 and 8 weeks after radiation exposure; the lack of observed effects 3 weeks after irradiation supports the specificity of these observations.
Our findings that exposure to γ rays spared the acquisition and retention of memory in the sMWM and rMWM tasks (Fig. 12) are consistent with those in numerous reports of normal performance (assessed on the basis of escape latency) in spatial MWM tasks by animals exposed to different radiation doses (from 5 cGy to tens of Gy), radiation types (x rays, γ rays, or proton particles), post-irradiation time intervals (ranging from 4 to 52 weeks), body parts subjected to radiation (hippocampus, whole brain, or whole body), and species (mice or rats) (summarized references in Table 1).
List of references confirming normal learning and memory in spatial and reversal MWM task following irradiation
Schematic representation of the experimental design and main results. Blue and white circles represent training in the respective pools in the MWM tasks. Black square represents hidden platform. Yellow asterisk, and green triangle represent local cues. Dashed line indicates animal's navigation trajectory to the platform (black represents normal trajectory; magenta represents altered trajectory). Lower horizontal arrows indicate the key stages of adult-born neurons' development.
Notably, the radiation-induced deficits observed in cdMWM were transient and, in the case of escape latencies and search strategies, were overcome after additional training. These findings parallel the performance improvement reported after additional training in spatial navigation tasks (Shukitt-Hale et al., 2000; Garthe et al., 2009; Bellone et al., 2015) or tasks with aversive reinforcement (Burghardt et al., 2012). The eventual improvement in cdMWM performance might have been associated with familiarity with the experimental context because, by the beginning of cdMWM training (day 9), the mice had undergone >1 week of training in that context. Other factors enabling the performance improvement in the irradiated mice in the cdMWM task might have included intact nonspatial components of navigation (Hamilton et al., 2004) and the inability of irradiation-induced damage to completely abolish the hippocampus-dependent component of learning.
We used several metrics to characterize the radiation-induced effects revealed by cdMWM. To describe the changes in learning, in addition to the analysis of escape latencies, we used automated classification of search strategies in Pathfinder. This application, combined with GLMM analysis and determination of IPE, allowed us to quantitatively characterize the repertoire of navigation strategies used by the mice during learning. We also extended the array of metrics used to assess memory in cdMWM: in addition to the time spent in the different quadrants, we calculated the preference score and CFC score.
Although some of these metrics are interassociated to some degree (e.g., shorter escape latency suggests shorter IPE, whereas longer time spent in the T quadrant suggests higher preference score), when considered in combination, they strengthened our conclusions regarding the effects of irradiation on cognitive flexibility.
Together, our results demonstrate that γ-radiation induces dose-, task-, and time interval-specific deficits which become apparent under the higher demand for cognitive flexibility in the cdMWM task. cdMWM and similarly designed complex tasks may be more broadly applied to analysis of animal behavior requiring continual strategy adjustment in response to altered circumstances. We suggest that cdMWM, compared with sMWM and rMWM, better reflects the multistep, multicued learning situations that animals encounter in their natural environment and is eventually more ethological. Notably, both the vulnerability of the skills required for solving complex tasks and the transient nature of the deficits observed in animals after irradiation parallel observations in humans exposed to radiation during oncotherapy (Schuitema et al., 2015; Kramkowski and Hebert, 2022; Lehrer et al., 2022).
As expected, exposure to 1 or 5 Gy of γ-radiation eliminated most dividing cells in the DG and, consequently, their immediate neuronal progeny. Remarkably, irradiation also had long-lasting effects on neurogenesis: suppressed production of young DCX-expressing neurons was evident 8 weeks after irradiation when the production of new neurons by the surviving stem cells was expected to be reestablished. This finding is compatible with observations that adult-born neurons of several ontogenic ages are underproduced after exposure to radiation (Andres-Mach et al., 2008; Park et al., 2012; Mineyeva et al., 2018).
Our results suggest that the radiation-induced decrease in the production of new hippocampal neurons is a key contributor to the observed behavioral deficits in cdMWM. This possibility is supported by the importance of adult neurogenesis for local cues discrimination in the MWM task (Arruda-Carvalho et al., 2011) or for the choice of efficient strategies in spatial alternation in MWM (Yu et al., 2019), contextual discrimination in contextual fear conditioning (Tronel et al., 2012) and in the radial arm maze (Clelland et al., 2009).
Although the decrease in new neurons was likely to contribute to the observed learning and memory deficits in irradiated mice, the role of hippocampal neurogenesis in behavioral responses may be defined by not only the number, but also the intrinsic properties of newborn DG neurons. Our results on c-Fos expression in the DG as a proxy for neuronal activation suggest differential recruitment of adult-born and preexisting neurons by different aspects of the context discrimination task.
Whereas neither the New Paradigm nor the New Context settings affected the activation of 7-week-old DG neurons, the activation of 12-week-old neurons increased in both settings, and the activation of preexisting neurons (which constitute most neurons in the DG) increased in the New Context setting, but not the New Paradigm setting; these results are in line with earlier observations of distinct activation of adult-born and developmentally generated neurons by exposure of animals to single or multiple contexts (Snyder et al., 2012) or to tasks requiring discrimination of dissimilar contexts and spatial problem solving (Tronel et al., 2015). Of note, the New Context setting presented a particular challenge to the animals: they were trained in an unfamiliar environment, with a new pool, new distal cues, and a new platform location; moreover, the local cues, although familiar, had switched roles (with the former false cue becoming the goal cue and vice versa). Thus, the mice were required to learn new spatial cues, incorporate familiar local cues into the new context, and update the cues' predictive value. The increased complexity of the New Context might have selectively activated the preexisting DG neurons.
Our results on neuronal activation, together with our behavioral data, suggest a substantially higher upper bound (6-12 weeks) for the age of newborn neurons that are competent in the context discrimination task than has usually been ascribed according to the initial steps of maturation of new neurons (4-8 weeks). These initial maturation steps might conceivably be necessary for the circuit integration of new neurons but insufficient for efficient performance in the complex tasks that new neurons must support; thus, increasingly challenging tasks may rely on continual development of new neurons and associated circuitry. Remarkably, recent reports have greatly extended the timeframe of the morphologic and functional maturation of new hippocampal neurons in rodents and humans (Masachs et al., 2021; Zhou et al., 2022); for example, from 4 to 6 weeks to >24 weeks in mice, with a gradual expansion of dendritic branches, mushroom spines, and presynaptic terminals of the mossy fibers (Cole et al., 2020). Together, our results suggest that precise context discrimination may be achieved through the interaction of DG neuronal populations of different ontogenic ages.
We posit that adult-born cells may not be necessary for the initial acquisition of spatial task but be important when modification of the acquired experience is needed. Our results highlight a subpopulation of adult-born neurons that are recruited by new demands of the cdMWM task and may contribute to circuit plasticity in the DG. In cdMWM, a delayed behavioral response (6-8 weeks) to the ablation of neurogenesis by irradiation, and selective activation of the 12-week-old neuronal cohort may be relevant to these subpopulations' apparent function in memory retrieval and updating, as well as in pattern separation — the ability to discriminate among distinct contexts incorporating a range of familiar features. These findings may also underlie animals' potential for adjusting and integrating various search strategies, and, ultimately, their cognitive flexibility.
Footnotes
This work was supported by Russian Science Foundation Grants 17-15-01426, 19-15-00247, and 20-15-00283 to K.V.A. and National Institute on Aging Grants R01 AG057705-01 and R01 AG076937-01 to G.E. This work was partially accomplished using the resource facilities at the National Research Center Kurchatov Institute. We thank Dr. Nikolai S. Lobanov (Kurchatov Institute National Research Center, Moscow) for help with irradiation procedures; Lidiya Zamyatina for technical assistance; and Dr. Jie Yang (Renaissance School of Medicine at Stony Brook University Biostatistical Consulting Core) for consultations on statistical analysis.
The authors declare no competing financial interests.
- Correspondence should be addressed to Grigori Enikolopov at grigori.enikolopov{at}stonybrookmedicine.edu or Alexander A. Lazutkin at lazutkin.a.a{at}gmail.com
