Unbiased analysis of spatial learning strategies in a modified Barnes maze using convolutional neural networks

Unbiased analysis of spatial learning strategies in a modified Barnes maze using convolutional neural networks | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Unbiased analysis of spatial learning strategies in a modified Barnes maze using convolutional neural networks Tomer Illouz, Lyn Alice Becker Ascher, Ravit Madar, Eitan Okun This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3075861/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 10 Jul, 2024 Read the published version in Scientific Reports → Version 1 posted 10 You are reading this latest preprint version Abstract Assessment of spatial learning abilities is central to behavioral neuroscience and a pillar of animal model validation and drug development. However, biases introduced by the apparatus, environment, or experimentalist represent a critical challenge to the test validity. We have recently developed the Modified Barnes Maze (MBM) task, a spatial learning paradigm that overcomes inherent behavioral biases of animals in the classical Barnes maze. The specific combination of spatial strategies employed by mice is often considered representative of the level of cognitive resources used. Herein, we have developed a convolutional neural network-based classifier of exploration strategies in the MBM that can effectively provide researchers with enhanced insights into cognitive traits in mice. Following validation, we compared the learning performance of female and male C57BL/6 mice, as well as that of Ts65Dn mice, a model of Down syndrome, and 5xFAD mice, a model of Alzheimer’s disease. Male mice exhibited more effective navigation abilities than female mice, reflected in higher utilization of effective spatial search strategies. Compared to wildtype controls, Ts65Dn mice exhibited reduced usage of spatial strategies despite similar success rates in completing this spatial task. These data exemplify the need for deeper strategy classification tools in dissecting complex cognitive traits. In sum, we provide a machine-learning-based strategy classifier that extends our understanding of mice’s spatial learning capabilities while enabling a more accurate cognitive assessment. Biological sciences/Neuroscience/Learning and memory/Spatial memory Biological sciences/Neuroscience/Cognitive ageing Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Spatial learning is an essential cognitive function that enables organisms to navigate and learn about their surroundings 1 . Indeed, the ability to acquire, store, and use spatial information is crucial for survival in many species, including mice and humans. Further, studying spatial learning in mice is vital for understanding the mechanisms that underlie neurodegenerative diseases such as Alzheimer's disease (AD) and Down syndrome (DS)-related AD 2 , 3 , as well as potential treatments 4 . To fully assess the spatial cognitive abilities of mice in these pathologies, it is important to consider the complexities of their behavior. Mice utilize different spatial strategies to navigate their environment 5 . It is crucial to understand the specific spatial strategies employed by mice under different physiological and pathological conditions, as different strategies can indicate cognitive abilities or deficits 6 , 7 . Various highly effective tasks that assess spatial learning and memory in rodents have been described, including the Morris water maze (MWM) 8 , radial arm maze (RAM) 9 , radial arm water maze (RAWM) 10 , and the Barnes maze (BM) 11 . For two of the most widely-used tasks, the MWM and the BM, we have previously developed online tools for classifying behavioral spatial strategies in the MWM 7 and the BM 6 using supervised machine-learning algorithms. These classifiers are superior to human manual classification, which tends to be biased, labor intensive, and depends on the degree of expertise of the human classifier. However, since each spatial learning tasks exhibits inherent specific disadvantages, we previously developed a modified variant of the classical BM (MBM) 12 . Specifically, the MBM combines the continuous nature of the MWM while avoiding water-related stress. The MBM enables high flexibility in task difficulty, along with overcoming inherent biases towards non-spatial strategies that are typical of the traditional BM task. In the present study, we describe the development of an unsupervised machine learning algorithm used to classify behavioral strategies in the MBM. We demonstrate the efficacy of this algorithm in classifying spatial strategies in four experimental settings: changing task difficulty, comparing male and female mice, and comparing two neurodegenerative mouse models, namely, AD and DS to wildtype (WT) controls. These four experimental settings represent physiological conditions in which subtle differences are expected, as well as pathological conditions in which significant cognitive impairments are observed, showcasing the dynamic range of strategy classification described herein. Methods Animals. Female and male C57BL/6 WT mice were purchased from Jackson Laboratories (stock #000664). Ts(17 16 )65Dn (Ts65Dn), a widely used mouse model for DS that encompasses a partial trisomy of Mmu16 and Mmu17, thus containing 92 genes orthologous to Hsa21, including mouse amyloid precursor protein (APP) and dual-specificity tyrosine phosphorylation-regulated kinase (DYRK1A) 13 , 14 , and their background strain (B6EiC3Sn.BLiAF1/J) were purchased from the Jackson Laboratories (stocks #005252, #003647). 5xFAD mice on a C57BL/6 genetic background, expressing mutant human APP and presenilin-1 (PSEN1) genes (B6.Cg-Tg;APPSwFILon,PSEN1*M146L*L286V), were purchased from Jackson Laboratories (stock #034848). Animals were housed in a reversed 12:12hr cycle. Animal care and experimental procedures followed Bar Ilan University’s guidelines and were approved by the Bar Ilan University Animal Care and Use Committee. All experiments were done in accordance with the recommendations of the ARRIVE guidelines. Modified Barnes maze. The MBM consisted of a circular, 110 cm-high, 122 cm-wide white Perspex table with 40 randomly placed holes, each with a diameter of 5 cm, located at least 7 cm from each other and at least 5 cm from the perimeter. Six holes were fabricated to function as an optional escape chamber. Lighting was measured at the center of the table and maintained at > 900 lux to motivate the animals to search a target hole that leads to a hidden escape chamber. During a 1-day habituation, animals were placed in a cylinder at the center of the maze. Five seconds later, the cylinder was removed, and the mice were allowed to explore the environment for 2 minutes. Mice that found the target hole could enter the escape chamber, while mice that did not find it within this period were placed back in the cylinder, now located above the target hole. Visual cues were presented on the walls surrounding the apparatus. In the spatial acquisition phase, mice were given 2 minutes per trial to find the target hole. Mice that did not find it were confined to the target hole area until they located it. In this task, each animal was given 3 trials with a 30-sec inter-trial interval. This procedure was repeated daily until no significant improvement in performance was identified. Performance parameters in this task were automatically calculated by the ANY-maze video tracking system (Stoelting Co.). Image processing. X, Y coordinates of the animals’ location throughout each trial were extracted from ANY-maze. All further processing was done in MATLAB (Mathworks). Trajectories were plotted as a black line on a white background, and the target location was indicated by a red dot. MBM table boundaries and holes were not plotted. MBM trials were randomly divided into train and test sets such that the test set contained 20% of the samples from each label. For train set samples, images were augmented ten times by five 72° rotations and an additional horizontal flip. Convolutional neural networks. Convolutional neural networks (CNN) were trained to classify exploration strategies using the MATLAB Deep Learning Toolbox (Mathworks). To implement a hierarchical classification architecture, a CNN was trained to classify strategies into three pan categories. Next, samples in each pan category were classified into final exploration strategies. The architecture of each CNN consists of an input layer and multiple repetitions of convolution, batch normalization, ReLU, and max pooling layers followed by fully connected, soft-max and classification output layers. At testing, accuracy rate per strategy was calculated as the fraction of correctly classified strategy out of the total number of samples in that class. Statistical analysis. The data presented as mean ± SEM were tested for significance in repeated measures (RM) two-way ANOVA or one-way ANOVA using Tukey’s test for multiple comparisons. All error bars presented are SEM calculated as for all numerical variables, and as for all binomial variables. Significant results were marked according to conventional critical P values: * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001. Results Identification of distinct spatial learning strategies in the MBM. The MBM combines the spatial continuity of the MWM with the advantages of a dry test environment, a key virtue of the BM (Fig. 1 a) 12 . Since the MWM and the BM share some exploration strategies 6 , 7 , 12 , we hypothesized that the strategies that characterize mice’s performance in the MBM would reflect the combined nature of this apparatus. To investigate this, we tested whether distinct exploration patterns can be identified using unsupervised learning methods. First, principal component analysis (PCA) was conducted on ten commonly used variables that were extracted from a data set of 1,508 MBM trials (Fig. S1 a). PCA revealed that linear combinations of variables could reflect different exploration patterns (Fig. S1 b-c). For example, principal component (PC)1 negatively correlated with mice’s path efficiency and positively correlated with all other variables, indicating that the largest fraction of variance in this dataset originates from differences between long- and short-duration searches, which represent efficient and inefficient searches, respectively (Fig. S1 b-c). PC3 positively correlated with the average distance of the mice from the maze center and negatively correlated with the standard deviation of that distance, indicating that PC3 is a good indicator for distinguishing between focal/random and circular searches. Next, we assessed the number of potential exploration strategies using the elbow method: The average distance of each datapoint from its nearest k-means cluster centroid and the variance explained by clustering were calculated on the same ten-dimensional dataset, with the increasing number of clusters ( k ). Both the distance from the nearest centroid (Fig. 1 b) and the variance explained (Fig. 1 c) were reduced when the data was clustered using \(k=2 to 6\) clusters, while only minor changes in these metrics were measured using \(k> 6\) . Two-dimensional tSNE projection following k-means clustering confirmed that the data is indeed under-classified when using \(k 6\) (Fig. 1 d, S1d), suggesting that six clusters are an accurate number of exploration strategies found in the MBM. Next, trajectory plots of 211 randomly sampled MBM trials were presented to seven human classifiers experienced in conducting spatial learning tasks for human labeling. Individuals were allowed to classify each trial to one of the previously defined sets of strategies characteristic of the MWM 7 (Fig. S2 a) and the BM 12 (Fig. S2 b). The final label of each trial was determined as the mode of human classifications (Fig. S3 a). Interestingly, the classification of some strategies was more consistent between human classifiers (e.g., Direct , 0.78 agreement level, Fig. S3 a), while others were less decisive (e.g., Long correction , 0.68 agreement level, Fig. S3 a). Since no trial was classified as a Serial search by any human classifiers, this strategy was removed from downstream analysis. Next, to meet the optimal number of exploration strategies (Fig. 1 b-d), the six most prevalent exploration strategies were selected for full dataset labeling (Fig. 1 e-f, Fig. S3 b). As hypothesized, some MBM exploration strategies overlapped with MWM strategies ( Circling, Accidental circling ). Long correction was shared between the BM and the MBM, and three MBM strategies overlapped with both MWM and BM ( Direct, Corrected, Random , Fig. S3 c). Using the same methodology, a set of 2035 MBM trials were classified by 7 individuals to one of the six predetermined exploration strategies (Fig. 2 a-b). A two-dimensional tSNE projection of this dataset reveals that strategies are ordered between two poles: Direct at the upper-left corner and Random search at the lower-right corner (Fig. S3 d). Indeed, the lower-right pole populates trials obtained at early stages of animal training, in which Random search is more prevalent, and the upper-left pole populates trials obtained at later stages of animal training, in which Direct and Corrected searches are more common (Fig. S3 e). Classification of exploration strategies in the MBM using convolutional neural networks. To obtain a generalized classifier independent of feature selection, variable calculation, and apparatus size, we chose to use CNNs. CNNs are a deep learning neural network commonly used in computer vision tasks. CNNs are designed to automatically detect and learn spatial hierarchies of features from input images or data. They consist of multiple layers of filters that convolve with the input image to extract relevant features such as edges, textures, and shapes. With their ability to automatically learn and extract features from images, CNNs are particularly effective in object recognition and classification tasks. In multi-category classifications, it is often preferable to use hierarchical rather than flat architectures, in which highly similar categories are first treated as pan-categories ( https://ieeexplore.ieee.org/document/7410671 ). In the next level in the classification dendrogram, such pan-categories can be treated separately to deal with highly similar categories. Indeed, we observed strong similarities between some of the MBM strategies. Using pairwise Jaccard similarity indices (JSI), we found strong similarities between Circling and Random (JSI = 0.8, Fig. 2 c), Long correction and Accidental circling (JSI = 0.74, Fig. 2 c), and Direct and Corrected strategies (JSI = 0.5, Fig. 2 c). These pairs were pooled into pan-categories that corresponded to trajectory length: Direct and Corrected (short), Long correction and Accidental circling (intermediate), and Circling and Random (long, Fig. 2 d, S4a). Based on these similarities, we devised a two-level hierarchical architecture in which classification into pan-groups is followed by classification into individual categories (Fig. 2 e). Next, a dataset of 2035 MBM trials was randomly divided into train and test sets (80% and 20%, respectively). For the training set, data augmentation was performed by 72° image rotations and a horizontal flip, yielding 10 images per each original sample. Next, we trained a neural network for each classification junctions in the dendrogram (Fig. 2 e). Classification accuracy, measured by comparing the model results with a human-labeled test set, reached 91.86% (Fig. 2 f-g, S4b-c), with no significant difference in classification-explained variance between human and machine classification (Fig. S4 d). As a reference, we trained a Random Forest classifier on the same datasets and obtained a classification accuracy of 76.38% (Fig. S4 e), indicating that the CNN was superior to a random forest classifier when tested against human observers. Target hole location affects task difficulty and alters the usage of spatial strategies. We previously observed that the difficulty of the MBM task could be manipulated by using central targets for a more difficult task and distal/peripheral targets for an easier task, allowing the experimenter to adjust task difficulty according to experimental needs 12 . To validate this finding using the strategy classifier, we trained eight-week-old WT male mice (n = 10 per group) in the MBM using a central and an off-center (distal) target (Fig. 3 a). As expected, mice that were trained to find the distal target used the Direct and Corrected strategies at a higher prevalence on the sixth day of training compared with mice trained to find the central, more difficult target (56.67%, 27.77%, respectively, Fig. 3 b). Interestingly, mice that were trained to find the central target used mostly long correction by the last day of training, suggesting that conversion to the higher Correction strategy was beyond the cognitive capacity of mice under this task difficulty level. To further quantify these differences, we established a scoring system for spatial cognition that localizes the animals’ performance on a scale relative to the averaged direct performance (ADP) in the MBM. To define a non-arbitrary scale, we calculated the similarity index of each item in our pre-labeled dataset to the ADP using the two-dimensional tSNE score (Fig. S3 e). Next, the averaged strategy similarities to the ADP were rescaled to the 0–1 range to yield the following cognitive scores: Circling = 0, Random = 0.1, Accidental circling = 0.32, Long correction = 0.61, Corrected = 0.77, and Direct = 1. The cognitive score did not significantly differ between groups (P = 0.058, Fig. 3 c) due to similar scores at the early stages of training. These results reflect a slower learning curve of mice when using the central target. Consistently, latency to target entry, exploration distance, and path efficiency were higher in mice trained to find the central target than in mice trained to find the distal target (P < 0.01, P < 0.05, P < 0.001, respectively, Fig. 3 d-f), whereas exploration speed was mildly lower for mice trained to find the central target (P < 0.05, Fig. 3 g). Additionally, mice trained to find the central target exhibited elevated time in non-target holes (P < 0.01, Fig. 3 h) and an increase in reference and working memory errors (P < 0.001, P < 0.05, Fig. 3 i-j). Accordingly, the trajectories of mice trained to find the central location covered a higher percentage of the surface of the MBM table (P < 0.05, Fig. S5 a-d). However, success rate did not differ between groups (P = 0.08, Fig. 3 k). In sum, we provided evidence that manipulation of target location in the MBM can be used to control task difficulty while affecting the combination of the spatial strategies utilized by mice. This feature of the MBM enables the experimenter to adjust task difficulty to comply with the experiment’s requirements. Male C57BL/6 mice exhibit a more effective navigation ability than females in the MBM. With respect to spatial abilities, males outperform females in both murine and in humans, with the underlying mechanisms not entirely clear 15 – 17 . To assess whether sex-related changes in mice performance in the MBM can be identified using our classifier, female and male C57BL/6 mice (n = 10 per group) were trained for nine days in the MBM. Since no significant deficit in spatial learning abilities was expected in these experimental groups and to enable the identification of subtle differences, we used the most difficult MBM setting in which the target is located at the central hole of the MBM table 12 (Fig. 4 a). On days 1–4 of the training, Circling was the most prevalent strategy used by female mice (40.74% at day 4, Fig. 4 b), while male mice mostly used long correction at this timepoint (29.16% at day 4, Fig. 4 b). Interestingly, the usage of spatially higher strategies (i.e., Direct, Corrected, Long correction ) gradually increased from the fifth day in females, while males exhibited such improvement from the second day of training (Fig. 4 b). This improvement reached optimum by the seventh day, indicated by 91.67% and 51.85% usage of spatially higher strategies by male and female mice, respectively (Fig. 4 b). Random search represented 33.33% of the trials at this timepoint among females and only 4.16% among males. These changes were also reflected in higher spatial cognitive scores in males than in females (P < 0.0001, Fig. 4 c). Earlier and faster conversion from non-spatial to highly spatial strategy in males compared with females was associated with reduced latency to target entry (P < 0.01, Fig. 4 d), reduced exploration distance (P < 0.0001, Fig. 4 e), and higher path efficiency (P < 0.0001, Fig. 4 f). Intriguingly, females exhibited elevated exploration speed (P < 0.001, Fig. 4 g). Exploration accuracy, indicated by the number of entries to non-target holes, reference memory errors, and working memory errors, was also reduced in females compared with males (P < 0.0001, P < 0.0001, P < 0.001, respectively, Fig. 4 h-j). Consistently, the area covered by exploration trajectories was higher in females than in males (P < 0.0001, Fig. S6 a-d), indicating more scattered searches by females. Success rate, however, did not differ between male and female mice (P = 0.51, Fig. 4 k), implying that the usage of less efficient strategies is compensated by increased speed in female mice (Fig. 4 g). To further compare exploration patterns between male and female mice, we segmented the MBM environment into 3×3mm bins and calculated the fold-change (and P values) in occupancy of male versus female mice in a bin-wise manner. These data may be represented as statistical occupancy maps and volcano plots (Fig. 4 l, S6e). Stronger spatial learning performance in males was reflected by a shorter and more focused exploration pattern, while females exhibited a more scattered exploration pattern (Fig. 4 l, left panel, S5e left panel). While no significant difference was observed on the first day of training (Fig. 4 l, middle panel, S6e middle panel), females continued to explore the periphery of the MBM surface by the last day of training (Fig. 4 l, right panel, S6e right panel). Altogether, we identified higher spatial learning accuracy in male compared with female mice tested in the MBM. Characterization of spatial learning deficits in the Ts65Dn mouse model of DS in the MBM. DS, caused by a trisomy in human chromosome 21 (Hsa-21), is the most common chromosomal abnormality in humans and the most prevalent genetic cause of intellectual disability 18 . Hsa-21 contains approximately 233 protein-coding genes, 423 non-protein coding genes, and numerous other functional genomic elements 18 . The Amyloid precursor protein (APP) gene, located within Hsa-21, is triplicated in DS, such that APP is over-expressed in affected individuals with DS compared with euploid individuals 19 . This heightened expression results in APP-dependent Alzheimer-like neuropathology in ~ 88% of all individuals with DS by the age of 65 19 . The Ts65Dn mouse model of DS encompasses a partial trisomy of mouse chromosome 16, which includes 92 genes orthologous to Hsa21. As a result, this model recapitulates many of the cognitive, behavioral, structural, and physiological abnormalities of DS 13 . To characterize the cognitive deficits of Ts65Dn mice, we investigated their performance in the MBM using our trained classifier. Eight-month-old male Ts65Dn and their respective genetic background control strain (n = 14 per group) were trained in the MBM for 10 days using a medium difficulty level achieved by placing the escape hole between the periphery and the center of the apparatus (Fig. 5 a). Although learning was observed throughout training in both groups, Ts65Dn mice exhibited a profound spatial learning deficit compared to WT controls, reflected in reduced usage of highly spatial strategies (Fig. 5 b) starting from the first day. Random search was more prevalent in Ts65Dn mice (66.67%) than in controls (57.14%) on the first day of training (Fig. 5 b). By the last day, 9.52% of Ts65Dn trials were classified as Random , compared with 2.38% Random searches in WT controls (Fig. 5 b). Overall, WT controls used more effective spatial strategies in 97.6% of the trials, compared with 85.71% in the Ts65Dn group. Accordingly, the cognitive scores of Ts65Dn mice were lower than those of WT mice throughout training (P < 0.0001, Fig. 5 c). Importantly, latency to reach the target hole entry did not differ between groups (P = 0.76, Fig. 5 d), which corresponded with higher distance (P < 0.0001, Fig. 5 e), higher speed (P < 0.0001, Fig. 5 f), and lower path efficiency (P < 0.0001, Fig. 5 g) in Ts65Dn mice compared with WT controls. These findings provide an example of the need for comprehensive analysis of mice performance in spatial learning tasks beyond comparison of latencies. Ts65Dn mice also exhibited reduced spatial accuracy, indicated by a higher number of entries to non-target holes (P < 0.0001, Fig. 5 h), a profound reference memory deficit (P < 0.0001, Fig. 5 i), and a milder working memory impairment (P < 0.01, Fig. 5 j). Additionally, the exploration trajectories of Ts65Dn mice covered a higher percentage of the MBM table (P < 0.0001, Fig. S7 a-d). However, success rate did not differ between groups (P = 0.77, Fig. 5 k), indicating compensation of less efficient exploration strategies by higher exploration speed. Using statistical occupancy maps, we found that Ts65Dn mice spent significantly less time in the vicinity of the target location throughout training (Fig. 5 l, left panel, Fig. S7 e, left panel). Accordingly, they spent more time near the periphery of the table on the first day (Fig. 5 l, middle panel, Fig. S7 e, middle panel), and exhibited less target-oriented exploration on the last day of training (Fig. 5 l, right panel, Fig. S7 e, right panel). In sum, our findings indicate a clear spatial learning impairment in the Ts65Dn mouse model of DS due to a deficit in reference memory capacity associated with using less-efficient exploration strategies and increased exploration speed. Association of spatial working memory impairment and circular explorations in the 5xFAD mouse model of Alzheimer disease. Spatial learning ability heavily relies on the integrity of hippocampal and para-hippocampal brain regions 20 . The hippocampus is also specifically vulnerable to Alzheimer disease (AD) pathology 21 . Therefore, we investigated the impact of Amyloid-β (Aβ) pathology on the spatial strategy utilization of transgenic AD mice in the MBM. The 5xFAD mouse strain, which models early-onset AD, encompasses five early-onset AD-related mutations: the Swedish (K670N, M671L), London (V717I), and Florida (I716V) mutations in APP and the M146L and L286V mutations in Presenilin 1 (PS1) 22 . As a result, 5xFAD mice exhibit early and profound Aβ pathology in the brain. Eight-month-old 5xFAD (n = 9) and their respective WT control male mice (n = 10) were trained to find the central (most difficult) target of the MBM for 6 days (Fig. 6 a). 5xFAD mice exhibited reduced usage of highly spatial strategies (i.e., Direct, Corrected , and long correction ) by the third day of training compared with WT controls (40.74%, 83.33%, respectively, Fig. 6 b). The most prevalent strategies in 5xFAD mice were Circling and Accidental circling , represent together 55.55% of strategies used. The performance of WT mice reached an optimum on day 5, with 80% of WT trials classified as highly spatial. In comparison, 5xFAD mice used these strategies at a prevalence of 44.44%. Accordingly, the overall cognitive score of 5xFAD mice was lower compared to WT controls (P < 0.05, Fig. 6 c). Unlike the performance of Ts65Dn mice, 5xFAD mice exhibited increased latency to target entry (P < 0.05, Fig. 5 d and Fig. 6 d, respectively), while only mild difference in exploration distance (P < 0.05, Fig. 6 e) and no difference in speed was observed (P = 0.19, Fig. 6 f). Path efficiency was also reduced in 5xFAD mice compared with controls (P < 0.01, Fig. 6 g), but was only associated with early stages of training. Importantly, lower accuracy was observed in 5xFAD mice, indicated by elevated time in non-target holes (P < 0.01, Fig. 6 h). Interestingly, reference memory capacity of 5xFAD mice did not differ from WT controls (P = 0.53, Fig. 6 i), but working memory capacity was significantly reduced in these mice (P < 0.01, Fig. 6 j), which resulted in lower success rate compared with WT mice (P < 0.0001, Fig. 6 k). Accordingly, statistical occupancy maps analysis revealed that exploration trajectories of 5xFAD mice covered a higher percentage of the MBM table surface (P < 0.05, Fig. S8 a-d), with a clear tendency to explore the periphery of the surface (Fig. 6 l, S8e). Overall, we report a working memory impairment in 5xFAD mice trained in the MBM, which is associated with a higher prevalence of the Circling and Accidental circling strategies. Discussion The MBM is a novel modified variant of the traditional BM task for spatial learning 12 . It combines the advantages of the MWM and the BM while avoiding their disadvantages 12 . As a result, the MBM avoids water stress and its related technical complications, including lengthy operation times characteristic of the MWM. It also avoids non-spatial strategies, such as circling, that are characteristic of the BM. As with more traditional spatial learning tasks such as the MWM and the BM, in which spatial strategy classifiers provide additional layers of information to be extracted 6 , 7 , we set out to generate an unbiased classifier that effectively classifies cognitive strategies in the MBM, as a tool for the research community. The algorithm presented herein can effectively analyze MBM data obtained from different transgenic mice with and without cognitive impairments in an unbiased manner while providing a cognitive score scale that assesses memory acquisition. Traditionally, performance in spatial learning tasks is analyzed according to one-dimensional parameters such as path efficiency, working and reference errors, and latency to reach the target. However, focusing solely on these parameters fails to fully capture the animal’s spatial cognitive capacity. We argue that utilizing a spatial learning paradigm superior to traditionally used paradigms (e.g., the MWM or the BM), combined with an added layer of information on the spatial strategies utilized by the rodents, is advantageous to optimizing experimental efficacy and promoting research output. In addition, the approach can sensitively identify behavioral nuances in pathologically relevant conditions, such as spatial learning deficits found in DS and AD mouse models. Therefore, in summary, the algorithm presented herein offers the research community an improved, powerful, and precise tool for assessing spatial learning in rodents. Declarations Data Availability statement All data including a user interface will be available upon request. Requests should be addressed to Prof. Eitan Okun, PH.D., ( [email protected] ) or Tomer Illouz, Ph.D., [email protected] ). Author contribution statement TI conducted the research and wrote the manuscript, LABA conducted behavioral experiments, RM assisted in all experiments, EO conceived the project and participated in writing the manuscript. Additional information The Authors declare no competing financial and/or non-financial interests. References Nyberg, N., Duvelle, E., Barry, C. & Spiers, H. J. Spatial goal coding in the hippocampal formation. Neuron 110, 394–422, doi: 10.1016/j.neuron.2021.12.012 (2022). Laczo, M. et al. Different Profiles of Spatial Navigation Deficits In Alzheimer's Disease Biomarker-Positive Versus Biomarker-Negative Older Adults With Amnestic Mild Cognitive Impairment. Front Aging Neurosci 14, 886778, doi: 10.3389/fnagi.2022.886778 (2022). Lavenex, P. B. et al. Allocentric spatial learning and memory deficits in Down syndrome. Front Psychol 6, 62, doi: 10.3389/fpsyg.2015.00062 (2015). Placido, J. et al. Spatial navigation in older adults with mild cognitive impairment and dementia: A systematic review and meta-analysis. Exp Gerontol 165, 111852, doi: 10.1016/j.exger.2022.111852 (2022). Johnsen, S. H. W. & Rytter, H. M. Dissociating spatial strategies in animal research: Critical methodological review with focus on egocentric navigation and the hippocampus. Neurosci Biobehav Rev 126, 57–78, doi: 10.1016/j.neubiorev.2021.03.022 (2021). Illouz, T. et al. Unbiased classification of spatial strategies in the Barnes maze. Bioinformatics 32, 3314–3320, doi: 10.1093/bioinformatics/btw376 (2016). Illouz, T., Madar, R., Louzoun, Y., Griffioen, K. J. & Okun, E. Unraveling cognitive traits using the Morris water maze unbiased strategy classification (MUST-C) algorithm. Brain Behav Immun 52, 132–144, doi: 10.1016/j.bbi.2015.10.013 (2016). Morris, R. Developments of a water-maze procedure for studying spatial learning in the rat. J Neurosci Methods 11, 47–60, doi: 10.1016/0165-0270(84)90007-4 (1984). Olton, D. S. & Samuelson, R. J. Remembrance of places passed: Spatial memory in rats. Journal of Experimental Psychology: Animal Behavior Processes 2, 97–116, doi: 10.1037/0097-7403.2.2.97 (1976). Alamed, J., Wilcock, D. M., Diamond, D. M., Gordon, M. N. & Morgan, D. Two-day radial-arm water maze learning and memory task; robust resolution of amyloid-related memory deficits in transgenic mice. Nat Protoc 1, 1671–1679, doi: 10.1038/nprot.2006.275 (2006). Barnes, C. A. Memory deficits associated with senescence: a neurophysiological and behavioral study in the rat. J Comp Physiol Psychol 93, 74–104, doi: 10.1037/h0077579 (1979). Illouz, T., Madar, R. & Okun, E. A modified Barnes maze for an accurate assessment of spatial learning in mice. J Neurosci Methods 334, 108579, doi: 10.1016/j.jneumeth.2020.108579 (2020). Reeves, R. H. et al. A mouse model for Down syndrome exhibits learning and behaviour deficits. Nat Genet 11, 177–184, doi: 10.1038/ng1095-177 (1995). Rueda, N., Florez, J. & Martinez-Cue, C. Mouse models of Down syndrome as a tool to unravel the causes of mental disabilities. Neural plasticity 2012, 584071, doi: 10.1155/2012/584071 (2012). Monfort, P., Gomez-Gimenez, B., Llansola, M. & Felipo, V. Gender differences in spatial learning, synaptic activity, and long-term potentiation in the hippocampus in rats: molecular mechanisms. ACS Chem Neurosci 6, 1420–1427, doi: 10.1021/acschemneuro.5b00096 (2015). Piber, D., Nowacki, J., Mueller, S. C., Wingenfeld, K. & Otte, C. Sex effects on spatial learning but not on spatial memory retrieval in healthy young adults. Behav Brain Res 336, 44–50, doi: 10.1016/j.bbr.2017.08.034 (2018). Yuan, L. et al. Gender Differences in Large-Scale and Small-Scale Spatial Ability: A Systematic Review Based on Behavioral and Neuroimaging Research. Front Behav Neurosci 13, 128, doi: 10.3389/fnbeh.2019.00128 (2019). Rafii, M. S., Kleschevnikov, A. M., Sawa, M. & Mobley, W. C. Down syndrome. Handb Clin Neurol 167, 321–336, doi: 10.1016/B978-0-12-804766-8.00017-0 (2019). Doran, E. et al. Down Syndrome, Partial Trisomy 21, and Absence of Alzheimer's Disease: The Role of APP. J Alzheimers Dis 56, 459–470, doi: 10.3233/JAD-160836 (2017). Tukker, J. J. et al. Microcircuits for spatial coding in the medial entorhinal cortex. Physiol Rev 102, 653–688, doi: 10.1152/physrev.00042.2020 (2022). Shipton, O. A., Tang, C. S., Paulsen, O. & Vargas-Caballero, M. Differential vulnerability of hippocampal CA3-CA1 synapses to Abeta. Acta Neuropathol Commun 10, 45, doi: 10.1186/s40478-022-01350-7 (2022). Oakley, H. et al. Intraneuronal beta-amyloid aggregates, neurodegeneration, and neuron loss in transgenic mice with five familial Alzheimer's disease mutations: potential factors in amyloid plaque formation. J Neurosci 26, 10129–10140, doi: 10.1523/JNEUROSCI.1202-06.2006 (2006). Additional Declarations No competing interests reported. Supplementary Files FigS1.tiff Figure S1. Principal component analysis reveals different exploration patterns in the MBM. 1508 MBM trials were subjected to principal component analysis (PCA) to identify meaningful linear combinations of commonly used metrics. (A) Two-dimensional PC projection. (B) fraction of variance explained by PCs. (C) Pearson’s correlation of each variable with the different PCs indicates different exploration patterns in the MBM. (D) Averaged Z-score of the different variables following k-means clustering using an increasing number of clusters ( k ). Abbreviations: principal component (PC); standard deviation (STD). FigS2.tiff Figure S2. Spatial learning strategies in the MWM and the BM. Pseudo-trajectory plots representative of previously defined learning strategies in the (A) MWM and (B) the BM. These strategies were used as optional categories for trial labeling by human classifiers. Abbreviations: Morris water maze (MWM); Barnes maze (BM). FigS3.tiff Figure S3. Defining spatial strategies in the MWM by human classifiers. 211 randomly selected MWM trials were presented to seven individuals with prior experience in behavioral spatial learning testing. Individuals were given the option to classify each trial to previously defined strategies in the MWM and the BM. The final label was determined using the winner-takes-all approach. (A) Confusion matrix of the Mode versus all labels. (B) Prevalence of different learning strategies obtained from human labeling, averaged between-judge agreement levels, are indicated in purple. (C) The MBM strategies were defined as the six most prevalent strategies in the defining set, which overlap with some strategies previously defined for the MWM and BM. (D) tSNE projection of 2035 MBM trials with color coding for the six learning strategies characteristic of the MBM and (E) day of acquisition. FigS4.tiff Figure S4. Classification of exploration strategies in the MBM using convolutional neural networks. Hierarchical neural-network classifier was trained to distinguish (A) pan categories, defined as short, intermediate, and long trajectories. (B) Accuracy level of the model versus human labeling at each classification node on the dendrogram. (C) human and model tSNE projection following the first (short, intermediate, and long), second (direct and corrected), third (long correction and accidental circling), and fourth (circling and random) classification nodes. The overall classification is presented in the bottom panels. (D) Percentage of variance explained by human and model classification do not show a difference. (E) Lower classification accuracy was obtained using a Random Forest classifier. FigS5.tiff Figure S5. Target hole location affects task difficulty and alters the usage of spatial strategies. (A) Overlayed trajectory plots and (B) occupancy plots at the first (upper panels) and last day (lower panels) of the acquisition phase, with the hidden escape box location indicated (upper and lower right panels, respectively). (C) Percentage of the MBM table covered by trajectories and (D) fraction of time mice occupied an increasing radius around the target at the first and last days. Repeated-measures two-way ANOVA, Abbreviations: Group effect (GE). FigS6.tiff Figure S6. Male C57BL/6 mice exhibit more effective navigation compared with females in the MBM. (A) Overlayed trajectory plots and (B) occupancy plots on the first and last day of the acquisition phase, with the location of the hidden escape box located at the center of the arena (right panel). (C) Percentage of the MBM table covered by trajectories and (D) fraction of time mice occupied an increasing radius around the target on the first and last days. (E) Volcano plots of the bin-wise occupancy of male mice compared with female mice used to generate the statistical occupancy maps; statistically significant bins are marked in color. Repeated-measures two-way ANOVA, *P<0.05, **P<0.01, Abbreviations: Group effect (GE). FigS7.tiff Figure S7. Characterization of spatial learning deficits in the Ts65Dn mouse model of DS in the MBM. (A) Overlayed trajectory plots and (B) occupancy plots on the first and last day of the acquisition phase, with the hidden escape box located midway between the center and the periphery of the apparatus (right panel). (C) Percentage of the MBM table covered by trajectories and (D) fraction of time mice occupied an increasing radius around the target on the first and last days. (E) Volcano plots of the bin-wise occupancy of Ts65Dn compared with C57BL/6 mice used to generate the statistical occupancy maps; statistically significant bins are marked in color. Repeated-measures two-way ANOVA, *P<0.05, **P<0.01, ***P<0.001 Abbreviations: Group effect (GE). FigS8.tiff Figure S8. Association of spatial working memory impairment and circular explorations in the 5xFAD mouse model of Alzheimer disease. (A) Overlayed trajectory plots and (B) occupancy plots on the first and last day of the acquisition phase, with the location of the hidden escape box located at the center of the arena (right panel). (C) Percentage of the MBM table covered by trajectories and (D) fraction of time mice occupied an increasing radius around the target on the first and last days. (E) Volcano plots of the bin-wise occupancy of 5xFAD compared with C57BL/6 mice used to generate the statistical occupancy maps; statistically significant bins are marked in color. Repeated-measures two-way ANOVA, *P<0.05, Abbreviations: Group effect (GE). Cite Share Download PDF Status: Published Journal Publication published 10 Jul, 2024 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Major revision 23 Oct, 2023 Reviews received at journal 23 Oct, 2023 Reviews received at journal 07 Sep, 2023 Reviewers agreed at journal 22 Aug, 2023 Reviewers agreed at journal 22 Aug, 2023 Reviewers invited by journal 22 Aug, 2023 Editor assigned by journal 22 Aug, 2023 Editor invited by journal 07 Jul, 2023 Submission checks completed at journal 07 Jul, 2023 First submitted to journal 17 Jun, 2023 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-3075861","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":216593973,"identity":"3d45a52e-e22c-4b28-90a2-e1dcd6d07bca","order_by":0,"name":"Tomer Illouz","email":"","orcid":"","institution":"Bar-Ilan University","correspondingAuthor":false,"prefix":"","firstName":"Tomer","middleName":"","lastName":"Illouz","suffix":""},{"id":216593974,"identity":"bf1b5df2-625e-40a1-acc9-a7930032f9d6","order_by":1,"name":"Lyn Alice Becker Ascher","email":"","orcid":"","institution":"Bar-Ilan University","correspondingAuthor":false,"prefix":"","firstName":"Lyn","middleName":"Alice Becker","lastName":"Ascher","suffix":""},{"id":216593975,"identity":"c0daada5-c500-4abb-8c23-7fe028547168","order_by":2,"name":"Ravit Madar","email":"","orcid":"","institution":"Bar-Ilan University","correspondingAuthor":false,"prefix":"","firstName":"Ravit","middleName":"","lastName":"Madar","suffix":""},{"id":216593976,"identity":"6a5a2d3f-0e8d-4e7f-9eb8-96e8144a5fdd","order_by":3,"name":"Eitan Okun","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA10lEQVRIiWNgGAWjYDACZghlAGQdY0gAsxkbD+DXwgzTwpYG09KAXwsDXAuPGVwMrxZ+dv6DnwsYbIwNbvd8e/CwjUGev4EZvy2SzczM0jMY0swM7pzdbpDYxmA44wABhxkcZmaQ5mE4bGNwI3ebBFAL4wZCfrE/zMz8G6Il5xlIiz1BLQbMzGwgW8yAWthAWhIJapE4zGxmzWOQZix5I83cIOGcRPKMwwS08PcffHybp8LGsO9G8rOHP8psbPvb2x8+wKcF6jyErfD0MApGwSgYBaOAAgAABChDQfpStdoAAAAASUVORK5CYII=","orcid":"","institution":"Bar-Ilan University","correspondingAuthor":true,"prefix":"","firstName":"Eitan","middleName":"","lastName":"Okun","suffix":""}],"badges":[],"createdAt":"2023-06-17 12:29:17","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3075861/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3075861/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-024-66855-8","type":"published","date":"2024-07-10T15:05:52+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":39921788,"identity":"521b9d30-d863-4b10-bc6a-024eeda9e11c","added_by":"auto","created_at":"2023-07-12 14:32:52","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":725856,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eIdentification of distinct spatial learning strategies in the MBM. \u003c/strong\u003eThe number of learning strategies used by mice in the MBM was estimated using unbiased techniques. 1508 MBM trials were subjected to \u003cem\u003ek-me\u003c/em\u003eans clustering (\u003cem\u003ek\u003c/em\u003e=2:18). (\u003cstrong\u003eA)\u003c/strong\u003e Scheme of the MBM table. The elbow method was utilized by evaluating the elbow point of (\u003cstrong\u003eB\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003ethe distance to the centroid and (\u003cstrong\u003eC\u003c/strong\u003e) the percentage of variance explained by clustering. (\u003cstrong\u003eD\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003etSNE projection of 1508 MBM trials clustered increasing \u003cem\u003ek.\u003c/em\u003e (\u003cstrong\u003eE\u003c/strong\u003e)\u003cem\u003e\u003cstrong\u003e \u003c/strong\u003e\u003c/em\u003ePseudo-trajectories typical of the six identified learning strategies\u003cem\u003e. \u003c/em\u003e(\u003cstrong\u003eF\u003c/strong\u003e)\u003cem\u003e\u003cstrong\u003e \u003c/strong\u003e\u003c/em\u003eAveraged Z-score of the variables used to differentiate learning strategies. Abbreviations: Standard deviation (STD).\u003c/p\u003e","description":"","filename":"Fig1.png","url":"https://assets-eu.researchsquare.com/files/rs-3075861/v1/d11c73ccad0946cfe504832f.png"},{"id":39921791,"identity":"1f956ffe-80f1-4eac-898e-463d8fd4bb04","added_by":"auto","created_at":"2023-07-12 14:32:52","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":553638,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eClassification of exploration strategies in the MBM using convolutional neural networks.\u003c/strong\u003e 2035 MWM trials were subjected to manual labeling by seven individuals to train and test the performance of a neural-network classifier. (\u003cstrong\u003eA\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eConfusion matrix of all manual labeling versus their mode reveals that \u003cem\u003eDirect\u003c/em\u003e and \u003cem\u003eRandom\u003c/em\u003eare the most coherently identified strategies. (\u003cstrong\u003eB\u003c/strong\u003e) Prevalence of different learning strategies obtained from human labeling, averaged between-judge agreement levels, are indicated in purple. (\u003cstrong\u003eC\u003c/strong\u003e) Pairwise Jaccard similarity coefficients were used to identify pan-categories. \u003cem\u003eDirect\u003c/em\u003eand \u003cem\u003eCorrected\u003c/em\u003e, \u003cem\u003eLong correction\u003c/em\u003e and \u003cem\u003eaccidental circling\u003c/em\u003e, and \u003cem\u003eCircling\u003c/em\u003e and \u003cem\u003eRandom\u003c/em\u003e were paired into three pan-categories based on a high Jaccard similarity coefficient (indicated within the confusion matrix). These pan categories were defined as (\u003cstrong\u003eD\u003c/strong\u003e) \u003cem\u003eshort, intermediate, \u003c/em\u003eand\u003cem\u003e long\u003c/em\u003e trajectories and were then divided into individual strategies on the (\u003cstrong\u003eE)\u003c/strong\u003eclassification dendrogram. (\u003cstrong\u003eF\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eAccuracy level of the hierarchical neural-network classifier. (\u003cstrong\u003eG)\u003c/strong\u003e tSNE projection of the test set; human (left) and neural-network model (right) show high similarity. One-way ANOVA, ***P\u0026lt;0.001, Abbreviations: Direct (Dir), Corrected (Cor), Long correction (LC), Accidental circling (AcC), Random (Rnd).\u003c/p\u003e","description":"","filename":"Fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-3075861/v1/8796b21b7693af37f483c24c.png"},{"id":39921789,"identity":"f1aa3007-afdd-45dc-b85b-cdff75683700","added_by":"auto","created_at":"2023-07-12 14:32:52","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":451815,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eTarget hole location affects task difficulty and alters the usage of spatial strategies.\u003c/strong\u003e (\u003cstrong\u003eA\u003c/strong\u003e) Male C57BL/6 mice (aged 8 weeks, n=10) were trained in the MBM using a central and a more distal target location. (\u003cstrong\u003eB\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eStrategy usage throughout the training was assessed using a neural-network classifier and (\u003cstrong\u003eC\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003ewas quantified by the cognitive score scaling. Between-group differences were measured for (\u003cstrong\u003eD\u003c/strong\u003e) latency to reach the target,\u003cstrong\u003e (E)\u003c/strong\u003e exploration distance,\u003cstrong\u003e (F)\u003c/strong\u003e path efficiency,\u003cstrong\u003e (G) \u003c/strong\u003ewalking speed, (\u003cstrong\u003eH\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003etime in non-target holes, (\u003cstrong\u003eI\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003ereference and (\u003cstrong\u003eJ)\u003c/strong\u003e working memory errors, and (\u003cstrong\u003eK\u003c/strong\u003e) success rate. Repeated-measures two-way ANOVA, *P\u0026lt;0.05, **P\u0026lt;0.01, ****P\u0026lt;0.0001 Abbreviations: Group effect (GE).\u003c/p\u003e","description":"","filename":"Fig3.png","url":"https://assets-eu.researchsquare.com/files/rs-3075861/v1/8216e36eaae5ab4e9683d98d.png"},{"id":39923368,"identity":"9109dba8-acab-4630-8b17-3f426a1f88d1","added_by":"auto","created_at":"2023-07-12 14:40:52","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":664666,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eMale C57BL/6 mice exhibit a more effective navigation ability compared with females in the MBM. (A) \u003c/strong\u003eFemale and male C57BL/6 mice (aged 8 weeks, n=10 per group) were trained in the MBM with the hidden escape box placed at the center of the arena. (\u003cstrong\u003eB\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eStrategy usage throughout the training was assessed using a neural-network classifier and (\u003cstrong\u003eC\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003ewas quantified by the cognitive score scaling. Inter-sex differences were observed for (\u003cstrong\u003eD\u003c/strong\u003e) latency to reach the target,\u003cstrong\u003e (E)\u003c/strong\u003eexploration distance,\u003cstrong\u003e (F)\u003c/strong\u003e path efficiency,\u003cstrong\u003e (G) \u003c/strong\u003ewalking speed, (\u003cstrong\u003eH\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003enumber of entries to non-target holes, \u003cstrong\u003e(I) \u003c/strong\u003ereference memory errors, and (\u003cstrong\u003eJ\u003c/strong\u003e) working memory errors but not for (\u003cstrong\u003eK\u003c/strong\u003e) success rate. (\u003cstrong\u003eL\u003c/strong\u003e) Statistical occupancy map for all (left), first (middle), and last (right) days of training. Bin-wise change in occupancy of male compared with female mice is indicated in blue (positive fold-change) or red (negative fold-change). P value is coded by the colors’ darkness. Non-significant differences are not shown. Repeated-measures two-way ANOVA, *P\u0026lt;0.05, **P\u0026lt;0.01, ***P\u0026lt;0.001, Abbreviations: Group effect (GE).\u003c/p\u003e","description":"","filename":"Fig4.png","url":"https://assets-eu.researchsquare.com/files/rs-3075861/v1/ba9266808da7ee004f52a5ad.png"},{"id":39923367,"identity":"e7300fea-165c-4da6-b347-aa6af8cabd83","added_by":"auto","created_at":"2023-07-12 14:40:52","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":689398,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCharacterization of spatial learning deficits in the Ts65Dn mouse model of DS in the MBM.\u003c/strong\u003e (\u003cstrong\u003eA\u003c/strong\u003e) Male Ts65Dn and WT mice (aged 8 months, n=14 per group) were trained in the MBM with the hidden escape box placed mid-way between the center and the periphery of the arena. (\u003cstrong\u003eB\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eStrategy usage throughout the training was assessed using a neural-network classifier and (\u003cstrong\u003eC\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003ewas quantified by the cognitive score scaling. Inter-strain differences were measured for (\u003cstrong\u003eD\u003c/strong\u003e) latency to reach the target,\u003cstrong\u003e \u003c/strong\u003e(\u003cstrong\u003eE\u003c/strong\u003e) exploration distance,\u003cstrong\u003e (F)\u003c/strong\u003e walking speed, (\u003cstrong\u003eG\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003epath efficiency,\u003cstrong\u003e \u003c/strong\u003e(\u003cstrong\u003eH\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003enumber of entries to non-target holes, (\u003cstrong\u003eI\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003ereference and (\u003cstrong\u003eJ\u003c/strong\u003e) working memory errors, and (\u003cstrong\u003eK\u003c/strong\u003e) success rate. (\u003cstrong\u003eL\u003c/strong\u003e) Statistical occupancy map for all (left), first (middle), and last (right) days of training. Bin-wise change in occupancy of Ts65dn compared with WT mice is indicated in blue (positive fold-change) or red (negative fold-change). P value is coded by the colors’ darkness. Non-significant differences are not shown. Repeated-measures two-way ANOVA, *P\u0026lt;0.05, **P\u0026lt;0.01, ***P\u0026lt;0.001, ****P\u0026lt;0.0001 Abbreviations: Group effect (GE).\u003c/p\u003e","description":"","filename":"Fig5.png","url":"https://assets-eu.researchsquare.com/files/rs-3075861/v1/ca652c9208d4fd091172ac9c.png"},{"id":39921793,"identity":"16aff9e7-5409-47db-891f-80a108734500","added_by":"auto","created_at":"2023-07-12 14:32:52","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":617331,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAssociation of spatial working memory impairment and circular explorations in the 5xFAD mouse model of Alzheimer disease.\u003c/strong\u003e (\u003cstrong\u003eA\u003c/strong\u003e) Male 5xFAD and WT mice (aged 8 months, n=9 per group) were trained in the MBM with the hidden escape box placed at the center of the arena. (\u003cstrong\u003eB\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eStrategy usage throughout the training was assessed using a neural-network classifier and (\u003cstrong\u003eC\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003ewas quantified by the cognitive score scaling. Inter-strain differences were measured for (\u003cstrong\u003eD\u003c/strong\u003e) latency to reach the target,\u003cstrong\u003e (E)\u003c/strong\u003e exploration distance,\u003cstrong\u003e (G)\u003c/strong\u003ewalking speed,\u003cstrong\u003e (G\u003c/strong\u003e) path efficiency\u003cstrong\u003e, (H) \u003c/strong\u003etime in non-target holes, \u003cstrong\u003e(I) \u003c/strong\u003ereference and \u003cstrong\u003e(J)\u003c/strong\u003e working memory errors, and \u003cstrong\u003e(K)\u003c/strong\u003esuccess rate. (\u003cstrong\u003eL\u003c/strong\u003e) Statistical occupancy map for all (left), first (middle) and last (right) days of training. Bin-wise change in occupancy of 5xFAD compared with WT mice is indicated in blue (positive fold-change) or red (negative fold-change). P value is coded by the colors’ darkness. Non-significant differences are not shown. Repeated-measures two-way ANOVA, *P\u0026lt;0.05, **P\u0026lt;0.01, ***P\u0026lt;0.001, ****P\u0026lt;0.0001 Abbreviations: Group effect (GE).\u003c/p\u003e","description":"","filename":"Fig6.png","url":"https://assets-eu.researchsquare.com/files/rs-3075861/v1/1425a62c82b16ee3c31fbf4d.png"},{"id":60225654,"identity":"591e5f9f-ad25-4725-88c5-a5ef6a5ba47f","added_by":"auto","created_at":"2024-07-13 15:06:00","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":4060011,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3075861/v1/88f52591-d2d7-4d93-a781-26d3a5a9fa51.pdf"},{"id":39921796,"identity":"98474081-5dd1-4798-a752-58613e85e47b","added_by":"auto","created_at":"2023-07-12 14:32:52","extension":"tiff","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":9700986,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure S1. Principal component analysis reveals different exploration patterns in the MBM. \u003c/strong\u003e1508 MBM trials were subjected to principal component analysis (PCA) to identify meaningful linear combinations of commonly used metrics. (\u003cstrong\u003eA\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eTwo-dimensional PC projection. (\u003cstrong\u003eB\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003efraction of variance explained by PCs. (\u003cstrong\u003eC\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003ePearson’s correlation of each variable with the different PCs indicates different exploration patterns in the MBM. (\u003cstrong\u003eD\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eAveraged Z-score of the different variables following \u003cem\u003ek-means \u003c/em\u003eclustering using an increasing number of clusters (\u003cem\u003ek\u003c/em\u003e). Abbreviations: principal component (PC); standard deviation (STD).\u003c/p\u003e","description":"","filename":"FigS1.tiff","url":"https://assets-eu.researchsquare.com/files/rs-3075861/v1/8fb829cd812b549f62aa38b3.tiff"},{"id":39921795,"identity":"bcd40dff-8f0e-4149-a343-ff31dea2ba30","added_by":"auto","created_at":"2023-07-12 14:32:52","extension":"tiff","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":9700986,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure S2.\u003c/strong\u003e \u003cstrong\u003eSpatial learning strategies in the MWM and the BM. \u003c/strong\u003ePseudo-trajectory plots representative of previously defined learning strategies in the (\u003cstrong\u003eA\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eMWM and (\u003cstrong\u003eB\u003c/strong\u003e) the BM. These strategies were used as optional categories for trial labeling by human classifiers. Abbreviations: Morris water maze (MWM); Barnes maze (BM).\u003c/p\u003e","description":"","filename":"FigS2.tiff","url":"https://assets-eu.researchsquare.com/files/rs-3075861/v1/06360d4c381f293794f32d23.tiff"},{"id":39923369,"identity":"706a8c00-76e9-4314-aa39-6a7b9e78f239","added_by":"auto","created_at":"2023-07-12 14:40:52","extension":"tiff","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":9700986,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure S3. Defining spatial strategies in the MWM by human classifiers. \u003c/strong\u003e211 randomly selected MWM trials were presented to seven individuals with prior experience in behavioral spatial learning testing. Individuals were given the option to classify each trial to previously defined strategies in the MWM and the BM. The final label was determined using the winner-takes-all approach. (\u003cstrong\u003eA\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eConfusion matrix of the Mode versus all labels. (\u003cstrong\u003eB\u003c/strong\u003e) Prevalence of different learning strategies obtained from human labeling, averaged between-judge agreement levels, are indicated in purple. (\u003cstrong\u003eC\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eThe MBM strategies were defined as the six most prevalent strategies in the defining set, which overlap with some strategies previously defined for the MWM and BM. (\u003cstrong\u003eD\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003etSNE projection of 2035 MBM trials with color coding for the six learning strategies characteristic of the MBM and (\u003cstrong\u003eE\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eday of acquisition.\u003c/p\u003e","description":"","filename":"FigS3.tiff","url":"https://assets-eu.researchsquare.com/files/rs-3075861/v1/1035ab565c69932a68f9815b.tiff"},{"id":39921794,"identity":"70f010d0-fd75-4b97-99a9-d55e8a9e3657","added_by":"auto","created_at":"2023-07-12 14:32:52","extension":"tiff","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":9700986,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure S4. Classification of exploration strategies in the MBM using convolutional neural networks. \u003c/strong\u003eHierarchical neural-network classifier was trained to distinguish (\u003cstrong\u003eA\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003epan categories, defined as \u003cem\u003eshort, intermediate, \u003c/em\u003eand\u003cem\u003e long\u003c/em\u003etrajectories. (\u003cstrong\u003eB\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eAccuracy level of the model versus human labeling at each classification node on the dendrogram. (\u003cstrong\u003eC\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003ehuman and model tSNE projection following the first (short, intermediate, and long), second (direct and corrected), third (long correction and accidental circling), and fourth (circling and random) classification nodes. The overall classification is presented in the bottom panels. (\u003cstrong\u003eD\u003c/strong\u003e) Percentage of variance explained by human and model classification do not show a difference. (\u003cstrong\u003eE\u003c/strong\u003e) Lower classification accuracy was obtained using a Random Forest classifier.\u003c/p\u003e","description":"","filename":"FigS4.tiff","url":"https://assets-eu.researchsquare.com/files/rs-3075861/v1/4c8671a34ebb9530a1995644.tiff"},{"id":39921801,"identity":"e5dbc657-99d8-4a6a-921d-8c905ea29862","added_by":"auto","created_at":"2023-07-12 14:32:52","extension":"tiff","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":9700986,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure S5. Target hole location affects task difficulty and alters the usage of spatial strategies.\u003c/strong\u003e (\u003cstrong\u003eA\u003c/strong\u003e) Overlayed trajectory plots and (\u003cstrong\u003eB\u003c/strong\u003e) occupancy plots at the first (upper panels) and last day (lower panels) of the acquisition phase, with the hidden escape box location indicated (upper and lower right panels, respectively). (\u003cstrong\u003eC\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003ePercentage of the MBM table covered by trajectories and (\u003cstrong\u003eD\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003efraction of time mice occupied an increasing radius around the target at the first and last days. Repeated-measures two-way ANOVA, Abbreviations: Group effect (GE).\u003c/p\u003e","description":"","filename":"FigS5.tiff","url":"https://assets-eu.researchsquare.com/files/rs-3075861/v1/d96331622e1f237c155b197a.tiff"},{"id":39921800,"identity":"b8569539-56df-40df-929d-9b8710416f17","added_by":"auto","created_at":"2023-07-12 14:32:52","extension":"tiff","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":9700986,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure S6. Male C57BL/6 mice exhibit more effective navigation compared with females in the MBM.\u003c/strong\u003e \u003cstrong\u003e(A) \u003c/strong\u003eOverlayed trajectory plots and (\u003cstrong\u003eB\u003c/strong\u003e) occupancy plots on the first and last day of the acquisition phase, with the location of the hidden escape box located at the center of the arena (right panel). (\u003cstrong\u003eC\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003ePercentage of the MBM table covered by trajectories and (\u003cstrong\u003eD\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003efraction of time mice occupied an increasing radius around the target on the first and last days. (\u003cstrong\u003eE\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eVolcano plots of the bin-wise occupancy of male mice compared with female mice used to generate the statistical occupancy maps; statistically significant bins are marked in color. Repeated-measures two-way ANOVA, *P\u0026lt;0.05, **P\u0026lt;0.01, Abbreviations: Group effect (GE).\u003c/p\u003e","description":"","filename":"FigS6.tiff","url":"https://assets-eu.researchsquare.com/files/rs-3075861/v1/7e22782ff4bfa262ecb27711.tiff"},{"id":39921797,"identity":"48657f45-ea1d-44c4-84a9-57bef71c6c8a","added_by":"auto","created_at":"2023-07-12 14:32:52","extension":"tiff","order_by":7,"title":"","display":"","copyAsset":false,"role":"supplement","size":9700986,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure S7. Characterization of spatial learning deficits in the Ts65Dn mouse model of DS in the MBM. (A) \u003c/strong\u003eOverlayed trajectory plots and (\u003cstrong\u003eB\u003c/strong\u003e) occupancy plots on the first and last day of the acquisition phase, with the hidden escape box located midway between the center and the periphery of the apparatus (right panel). (\u003cstrong\u003eC\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003ePercentage of the MBM table covered by trajectories and (\u003cstrong\u003eD\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003efraction of time mice occupied an increasing radius around the target on the first and last days. (\u003cstrong\u003eE\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eVolcano plots of the bin-wise occupancy of Ts65Dn compared with C57BL/6 mice used to generate the statistical occupancy maps; statistically significant bins are marked in color. Repeated-measures two-way ANOVA, *P\u0026lt;0.05, **P\u0026lt;0.01, ***P\u0026lt;0.001 Abbreviations: Group effect (GE).\u003c/p\u003e","description":"","filename":"FigS7.tiff","url":"https://assets-eu.researchsquare.com/files/rs-3075861/v1/970847c63f317bf2e7ec50e1.tiff"},{"id":39921799,"identity":"cec00eb0-dfaa-4eb0-ae7d-86a53c6445aa","added_by":"auto","created_at":"2023-07-12 14:32:52","extension":"tiff","order_by":8,"title":"","display":"","copyAsset":false,"role":"supplement","size":9700986,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure S8. Association of spatial working memory impairment and circular explorations in the 5xFAD mouse model of Alzheimer disease.\u003c/strong\u003e (\u003cstrong\u003eA\u003c/strong\u003e) Overlayed trajectory plots and (\u003cstrong\u003eB\u003c/strong\u003e) occupancy plots on the first and last day of the acquisition phase, with the location of the hidden escape box located at the center of the arena (right panel). \u003cstrong\u003e(C) \u003c/strong\u003ePercentage of the MBM table covered by trajectories and (\u003cstrong\u003eD\u003c/strong\u003e) fraction of time mice occupied an increasing radius around the target on the first and last days. (\u003cstrong\u003eE\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eVolcano plots of the bin-wise occupancy of 5xFAD compared with C57BL/6 mice used to generate the statistical occupancy maps; statistically significant bins are marked in color. Repeated-measures two-way ANOVA, *P\u0026lt;0.05, Abbreviations: Group effect (GE).\u003c/p\u003e","description":"","filename":"FigS8.tiff","url":"https://assets-eu.researchsquare.com/files/rs-3075861/v1/280755daf4e3525b8133da37.tiff"}],"financialInterests":"No competing interests reported.","formattedTitle":"Unbiased analysis of spatial learning strategies in a modified Barnes maze using convolutional neural networks","fulltext":[{"header":"Introduction","content":"\u003cp\u003eSpatial learning is an essential cognitive function that enables organisms to navigate and learn about their surroundings\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. Indeed, the ability to acquire, store, and use spatial information is crucial for survival in many species, including mice and humans. Further, studying spatial learning in mice is vital for understanding the mechanisms that underlie neurodegenerative diseases such as Alzheimer's disease (AD) and Down syndrome (DS)-related AD\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e,\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e, as well as potential treatments\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. To fully assess the spatial cognitive abilities of mice in these pathologies, it is important to consider the complexities of their behavior. Mice utilize different spatial strategies to navigate their environment\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. It is crucial to understand the specific spatial strategies employed by mice under different physiological and pathological conditions, as different strategies can indicate cognitive abilities or deficits\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e,\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. Various highly effective tasks that assess spatial learning and memory in rodents have been described, including the Morris water maze (MWM)\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e, radial arm maze (RAM)\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e, radial arm water maze (RAWM)\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e, and the Barnes maze (BM)\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e. For two of the most widely-used tasks, the MWM and the BM, we have previously developed online tools for classifying behavioral spatial strategies in the MWM\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e and the BM\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e using supervised machine-learning algorithms. These classifiers are superior to human manual classification, which tends to be biased, labor intensive, and depends on the degree of expertise of the human classifier. However, since each spatial learning tasks exhibits inherent specific disadvantages, we previously developed a modified variant of the classical BM (MBM)\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. Specifically, the MBM combines the continuous nature of the MWM while avoiding water-related stress. The MBM enables high flexibility in task difficulty, along with overcoming inherent biases towards non-spatial strategies that are typical of the traditional BM task. In the present study, we describe the development of an unsupervised machine learning algorithm used to classify behavioral strategies in the MBM. We demonstrate the efficacy of this algorithm in classifying spatial strategies in four experimental settings: changing task difficulty, comparing male and female mice, and comparing two neurodegenerative mouse models, namely, AD and DS to wildtype (WT) controls. These four experimental settings represent physiological conditions in which subtle differences are expected, as well as pathological conditions in which significant cognitive impairments are observed, showcasing the dynamic range of strategy classification described herein.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cstrong\u003eAnimals.\u003c/strong\u003e Female and male C57BL/6 WT mice were purchased from Jackson Laboratories (stock #000664). Ts(17\u003csup\u003e16\u003c/sup\u003e)65Dn (Ts65Dn), a widely used mouse model for DS that encompasses a partial trisomy of Mmu16 and Mmu17, thus containing 92 genes orthologous to Hsa21, including mouse \u003cem\u003eamyloid precursor protein (APP)\u003c/em\u003e and \u003cem\u003edual-specificity tyrosine phosphorylation-regulated kinase (DYRK1A)\u003c/em\u003e\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e, and their background strain (B6EiC3Sn.BLiAF1/J) were purchased from the Jackson Laboratories (stocks #005252, #003647). 5xFAD mice on a C57BL/6 genetic background, expressing mutant human \u003cem\u003eAPP\u003c/em\u003e and \u003cem\u003epresenilin-1 (PSEN1)\u003c/em\u003e genes (B6.Cg-Tg;APPSwFILon,PSEN1*M146L*L286V), were purchased from Jackson Laboratories (stock #034848). Animals were housed in a reversed 12:12hr cycle.\u003c/p\u003e\n\u003cp\u003eAnimal care and experimental procedures followed Bar Ilan University’s guidelines and were approved by the Bar Ilan University Animal Care and Use Committee. All experiments were done in accordance with the recommendations of the ARRIVE guidelines.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModified Barnes maze.\u003c/strong\u003e The MBM consisted of a circular, 110 cm-high, 122 cm-wide white Perspex table with 40 randomly placed holes, each with a diameter of 5 cm, located at least 7 cm from each other and at least 5 cm from the perimeter. Six holes were fabricated to function as an optional escape chamber. Lighting was measured at the center of the table and maintained at \u0026gt; 900 lux to motivate the animals to search a target hole that leads to a hidden escape chamber. During a 1-day habituation, animals were placed in a cylinder at the center of the maze. Five seconds later, the cylinder was removed, and the mice were allowed to explore the environment for 2 minutes. Mice that found the target hole could enter the escape chamber, while mice that did not find it within this period were placed back in the cylinder, now located above the target hole. Visual cues were presented on the walls surrounding the apparatus. In the spatial acquisition phase, mice were given 2 minutes per trial to find the target hole. Mice that did not find it were confined to the target hole area until they located it. In this task, each animal was given 3 trials with a 30-sec inter-trial interval. This procedure was repeated daily until no significant improvement in performance was identified. Performance parameters in this task were automatically calculated by the ANY-maze video tracking system (Stoelting Co.).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eImage processing.\u003c/strong\u003e X, Y coordinates of the animals’ location throughout each trial were extracted from ANY-maze. All further processing was done in MATLAB (Mathworks). Trajectories were plotted as a black line on a white background, and the target location was indicated by a red dot. MBM table boundaries and holes were not plotted. MBM trials were randomly divided into train and test sets such that the test set contained 20% of the samples from each label. For train set samples, images were augmented ten times by five 72° rotations and an additional horizontal flip.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConvolutional neural networks.\u003c/strong\u003e Convolutional neural networks (CNN) were trained to classify exploration strategies using the MATLAB Deep Learning Toolbox (Mathworks). To implement a hierarchical classification architecture, a CNN was trained to classify strategies into three pan categories. Next, samples in each pan category were classified into final exploration strategies. The architecture of each CNN consists of an input layer and multiple repetitions of convolution, batch normalization, ReLU, and max pooling layers followed by fully connected, soft-max and classification output layers. At testing, accuracy rate per strategy was calculated as the fraction of correctly classified strategy out of the total number of samples in that class.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStatistical analysis.\u003c/strong\u003e The data presented as\u0026nbsp; mean ± SEM were tested for significance in repeated measures (RM) two-way ANOVA or one-way ANOVA using Tukey’s test for multiple comparisons. All error bars presented are SEM calculated as\u0026nbsp;\u003cimg src=\"data:image/png;base64,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\" width=\"68\" height=\"58\"\u003e\u0026nbsp;for all numerical variables, and as \u003cimg src=\"data:image/png;base64,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\" width=\"100\" height=\"81\"\u003efor all binomial variables. Significant results were marked according to conventional critical P values: * P \u0026lt; 0.05, ** P \u0026lt; 0.01, *** P \u0026lt; 0.001, **** P \u0026lt; 0.0001.\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e \u003cb\u003eIdentification of distinct spatial learning strategies in the MBM.\u003c/b\u003e The MBM combines the spatial continuity of the MWM with the advantages of a dry test environment, a key virtue of the BM (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ea)\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. Since the MWM and the BM share some exploration strategies\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e,\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e,\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e, we hypothesized that the strategies that characterize mice\u0026rsquo;s performance in the MBM would reflect the combined nature of this apparatus. To investigate this, we tested whether distinct exploration patterns can be identified using unsupervised learning methods. First, principal component analysis (PCA) was conducted on ten commonly used variables that were extracted from a data set of 1,508 MBM trials (Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003ea). PCA revealed that linear combinations of variables could reflect different exploration patterns (Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003eb-c). For example, principal component (PC)1 negatively correlated with mice\u0026rsquo;s path efficiency and positively correlated with all other variables, indicating that the largest fraction of variance in this dataset originates from differences between long- and short-duration searches, which represent efficient and inefficient searches, respectively (Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003eb-c). PC3 positively correlated with the average distance of the mice from the maze center and negatively correlated with the standard deviation of that distance, indicating that PC3 is a good indicator for distinguishing between focal/random and circular searches. Next, we assessed the number of potential exploration strategies using the elbow method: The average distance of each datapoint from its nearest \u003cem\u003ek-means\u003c/em\u003e cluster centroid and the variance explained by clustering were calculated on the same ten-dimensional dataset, with the increasing number of clusters (\u003cem\u003ek\u003c/em\u003e). Both the distance from the nearest centroid (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eb) and the variance explained (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ec) were reduced when the data was clustered using \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(k=2 to 6\\)\u003c/span\u003e\u003c/span\u003e clusters, while only minor changes in these metrics were measured using\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(k\u0026gt; 6\\)\u003c/span\u003e\u003c/span\u003e. Two-dimensional tSNE projection following \u003cem\u003ek-means\u003c/em\u003e clustering confirmed that the data is indeed under-classified when using \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(k\u0026lt; 6\\)\u003c/span\u003e\u003c/span\u003e and over-classified when using \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(k\u0026gt; 6\\)\u003c/span\u003e\u003c/span\u003e (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ed, S1d), suggesting that six clusters are an accurate number of exploration strategies found in the MBM.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eNext, trajectory plots of 211 randomly sampled MBM trials were presented to seven human classifiers experienced in conducting spatial learning tasks for human labeling. Individuals were allowed to classify each trial to one of the previously defined sets of strategies characteristic of the MWM\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e (Fig. \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003ea) and the BM\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e (Fig. \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003eb). The final label of each trial was determined as the mode of human classifications (Fig. \u003cspan refid=\"MOESM3\" class=\"InternalRef\"\u003eS3\u003c/span\u003ea). Interestingly, the classification of some strategies was more consistent between human classifiers (e.g., \u003cem\u003eDirect\u003c/em\u003e, 0.78 agreement level, Fig. \u003cspan refid=\"MOESM3\" class=\"InternalRef\"\u003eS3\u003c/span\u003ea), while others were less decisive (e.g., \u003cem\u003eLong correction\u003c/em\u003e, 0.68 agreement level, Fig. \u003cspan refid=\"MOESM3\" class=\"InternalRef\"\u003eS3\u003c/span\u003ea). Since no trial was classified as a \u003cem\u003eSerial search\u003c/em\u003e by any human classifiers, this strategy was removed from downstream analysis. Next, to meet the optimal number of exploration strategies (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eb-d), the six most prevalent exploration strategies were selected for full dataset labeling (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ee-f, Fig. \u003cspan refid=\"MOESM3\" class=\"InternalRef\"\u003eS3\u003c/span\u003eb). As hypothesized, some MBM exploration strategies overlapped with MWM strategies (\u003cem\u003eCircling, Accidental circling\u003c/em\u003e). \u003cem\u003eLong correction\u003c/em\u003e was shared between the BM and the MBM, and three MBM strategies overlapped with both MWM and BM (\u003cem\u003eDirect, Corrected, Random\u003c/em\u003e, Fig. \u003cspan refid=\"MOESM3\" class=\"InternalRef\"\u003eS3\u003c/span\u003ec). Using the same methodology, a set of 2035 MBM trials were classified by 7 individuals to one of the six predetermined exploration strategies (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea-b). A two-dimensional tSNE projection of this dataset reveals that strategies are ordered between two poles: \u003cem\u003eDirect\u003c/em\u003e at the upper-left corner and \u003cem\u003eRandom search\u003c/em\u003e at the lower-right corner (Fig. \u003cspan refid=\"MOESM3\" class=\"InternalRef\"\u003eS3\u003c/span\u003ed). Indeed, the lower-right pole populates trials obtained at early stages of animal training, in which \u003cem\u003eRandom search\u003c/em\u003e is more prevalent, and the upper-left pole populates trials obtained at later stages of animal training, in which \u003cem\u003eDirect\u003c/em\u003e and \u003cem\u003eCorrected\u003c/em\u003e searches are more common (Fig. \u003cspan refid=\"MOESM3\" class=\"InternalRef\"\u003eS3\u003c/span\u003ee).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eClassification of exploration strategies in the MBM using convolutional neural networks.\u003c/b\u003e To obtain a generalized classifier independent of feature selection, variable calculation, and apparatus size, we chose to use CNNs. CNNs are a deep learning neural network commonly used in computer vision tasks. CNNs are designed to automatically detect and learn spatial hierarchies of features from input images or data. They consist of multiple layers of filters that convolve with the input image to extract relevant features such as edges, textures, and shapes. With their ability to automatically learn and extract features from images, CNNs are particularly effective in object recognition and classification tasks.\u003c/p\u003e \u003cp\u003eIn multi-category classifications, it is often preferable to use hierarchical rather than flat architectures, in which highly similar categories are first treated as pan-categories (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://ieeexplore.ieee.org/document/7410671\u003c/span\u003e\u003cspan address=\"https://ieeexplore.ieee.org/document/7410671\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). In the next level in the classification dendrogram, such pan-categories can be treated separately to deal with highly similar categories. Indeed, we observed strong similarities between some of the MBM strategies. Using pairwise Jaccard similarity indices (JSI), we found strong similarities between \u003cem\u003eCircling\u003c/em\u003e and \u003cem\u003eRandom\u003c/em\u003e (JSI\u0026thinsp;=\u0026thinsp;0.8, Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ec), \u003cem\u003eLong correction\u003c/em\u003e and \u003cem\u003eAccidental circling\u003c/em\u003e (JSI\u0026thinsp;=\u0026thinsp;0.74, Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ec), and \u003cem\u003eDirect\u003c/em\u003e and \u003cem\u003eCorrected\u003c/em\u003e strategies (JSI\u0026thinsp;=\u0026thinsp;0.5, Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ec). These pairs were pooled into pan-categories that corresponded to trajectory length: \u003cem\u003eDirect\u003c/em\u003e and \u003cem\u003eCorrected\u003c/em\u003e (short), \u003cem\u003eLong correction\u003c/em\u003e and \u003cem\u003eAccidental circling\u003c/em\u003e (intermediate), and \u003cem\u003eCircling\u003c/em\u003e and \u003cem\u003eRandom\u003c/em\u003e (long, Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ed, S4a). Based on these similarities, we devised a two-level hierarchical architecture in which classification into pan-groups is followed by classification into individual categories (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ee).\u003c/p\u003e \u003cp\u003eNext, a dataset of 2035 MBM trials was randomly divided into train and test sets (80% and 20%, respectively). For the training set, data augmentation was performed by 72\u0026deg; image rotations and a horizontal flip, yielding 10 images per each original sample. Next, we trained a neural network for each classification junctions in the dendrogram (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ee). Classification accuracy, measured by comparing the model results with a human-labeled test set, reached 91.86% (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ef-g, S4b-c), with no significant difference in classification-explained variance between human and machine classification (Fig. \u003cspan refid=\"MOESM4\" class=\"InternalRef\"\u003eS4\u003c/span\u003ed). As a reference, we trained a Random Forest classifier on the same datasets and obtained a classification accuracy of 76.38% (Fig. \u003cspan refid=\"MOESM4\" class=\"InternalRef\"\u003eS4\u003c/span\u003ee), indicating that the CNN was superior to a random forest classifier when tested against human observers.\u003c/p\u003e \u003cp\u003e \u003cb\u003eTarget hole location affects task difficulty and alters the usage of spatial strategies.\u003c/b\u003e We previously observed that the difficulty of the MBM task could be manipulated by using central targets for a more difficult task and distal/peripheral targets for an easier task, allowing the experimenter to adjust task difficulty according to experimental needs\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. To validate this finding using the strategy classifier, we trained eight-week-old WT male mice (n\u0026thinsp;=\u0026thinsp;10 per group) in the MBM using a central and an off-center (distal) target (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea). As expected, mice that were trained to find the distal target used the \u003cem\u003eDirect\u003c/em\u003e and \u003cem\u003eCorrected\u003c/em\u003e strategies at a higher prevalence on the sixth day of training compared with mice trained to find the central, more difficult target (56.67%, 27.77%, respectively, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb). Interestingly, mice that were trained to find the central target used mostly \u003cem\u003elong correction\u003c/em\u003e by the last day of training, suggesting that conversion to the higher \u003cem\u003eCorrection\u003c/em\u003e strategy was beyond the cognitive capacity of mice under this task difficulty level. To further quantify these differences, we established a scoring system for spatial cognition that localizes the animals\u0026rsquo; performance on a scale relative to the averaged \u003cem\u003edirect\u003c/em\u003e performance (ADP) in the MBM. To define a non-arbitrary scale, we calculated the similarity index of each item in our pre-labeled dataset to the ADP using the two-dimensional tSNE score (Fig. \u003cspan refid=\"MOESM3\" class=\"InternalRef\"\u003eS3\u003c/span\u003ee). Next, the averaged strategy similarities to the ADP were rescaled to the 0\u0026ndash;1 range to yield the following cognitive scores: \u003cem\u003eCircling\u0026thinsp;=\u0026thinsp;0, Random\u0026thinsp;=\u0026thinsp;0.1, Accidental circling\u0026thinsp;=\u0026thinsp;0.32, Long correction\u0026thinsp;=\u0026thinsp;0.61, Corrected\u0026thinsp;=\u0026thinsp;0.77, and Direct\u0026thinsp;=\u0026thinsp;1.\u003c/em\u003e The cognitive score did not significantly differ between groups (P\u0026thinsp;=\u0026thinsp;0.058, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ec) due to similar scores at the early stages of training. These results reflect a slower learning curve of mice when using the central target. Consistently, latency to target entry, exploration distance, and path efficiency were higher in mice trained to find the central target than in mice trained to find the distal target (P\u0026thinsp;\u0026lt;\u0026thinsp;0.01, P\u0026thinsp;\u0026lt;\u0026thinsp;0.05, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001, respectively, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ed-f), whereas exploration speed was mildly lower for mice trained to find the central target (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eg). Additionally, mice trained to find the central target exhibited elevated time in non-target holes (P\u0026thinsp;\u0026lt;\u0026thinsp;0.01, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eh) and an increase in reference and working memory errors (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001, P\u0026thinsp;\u0026lt;\u0026thinsp;0.05, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ei-j). Accordingly, the trajectories of mice trained to find the central location covered a higher percentage of the surface of the MBM table (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05, Fig. \u003cspan refid=\"MOESM5\" class=\"InternalRef\"\u003eS5\u003c/span\u003ea-d). However, success rate did not differ between groups (P\u0026thinsp;=\u0026thinsp;0.08, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ek). In sum, we provided evidence that manipulation of target location in the MBM can be used to control task difficulty while affecting the combination of the spatial strategies utilized by mice. This feature of the MBM enables the experimenter to adjust task difficulty to comply with the experiment\u0026rsquo;s requirements.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eMale C57BL/6 mice exhibit a more effective navigation ability than females in the MBM.\u003c/b\u003e With respect to spatial abilities, males outperform females in both murine and in humans, with the underlying mechanisms not entirely clear\u003csup\u003e\u003cspan additionalcitationids=\"CR16\" citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. To assess whether sex-related changes in mice performance in the MBM can be identified using our classifier, female and male C57BL/6 mice (n\u0026thinsp;=\u0026thinsp;10 per group) were trained for nine days in the MBM. Since no significant deficit in spatial learning abilities was expected in these experimental groups and to enable the identification of subtle differences, we used the most difficult MBM setting in which the target is located at the central hole of the MBM table\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea). On days 1\u0026ndash;4 of the training, \u003cem\u003eCircling\u003c/em\u003e was the most prevalent strategy used by female mice (40.74% at day 4, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb), while male mice mostly used \u003cem\u003elong correction\u003c/em\u003e at this timepoint (29.16% at day 4, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb). Interestingly, the usage of spatially higher strategies (i.e., \u003cem\u003eDirect, Corrected, Long correction\u003c/em\u003e) gradually increased from the fifth day in females, while males exhibited such improvement from the second day of training (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb). This improvement reached optimum by the seventh day, indicated by 91.67% and 51.85% usage of spatially higher strategies by male and female mice, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb). \u003cem\u003eRandom search\u003c/em\u003e represented 33.33% of the trials at this timepoint among females and only 4.16% among males.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThese changes were also reflected in higher spatial cognitive scores in males than in females (P\u0026thinsp;\u0026lt;\u0026thinsp;0.0001, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ec).\u003c/p\u003e \u003cp\u003eEarlier and faster conversion from non-spatial to highly spatial strategy in males compared with females was associated with reduced latency to target entry (P\u0026thinsp;\u0026lt;\u0026thinsp;0.01, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ed), reduced exploration distance (P\u0026thinsp;\u0026lt;\u0026thinsp;0.0001, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ee), and higher path efficiency (P\u0026thinsp;\u0026lt;\u0026thinsp;0.0001, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ef). Intriguingly, females exhibited elevated exploration speed (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eg). Exploration accuracy, indicated by the number of entries to non-target holes, reference memory errors, and working memory errors, was also reduced in females compared with males (P\u0026thinsp;\u0026lt;\u0026thinsp;0.0001, P\u0026thinsp;\u0026lt;\u0026thinsp;0.0001, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001, respectively, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eh-j). Consistently, the area covered by exploration trajectories was higher in females than in males (P\u0026thinsp;\u0026lt;\u0026thinsp;0.0001, Fig. \u003cspan refid=\"MOESM6\" class=\"InternalRef\"\u003eS6\u003c/span\u003ea-d), indicating more scattered searches by females. Success rate, however, did not differ between male and female mice (P\u0026thinsp;=\u0026thinsp;0.51, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ek), implying that the usage of less efficient strategies is compensated by increased speed in female mice (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eg).\u003c/p\u003e \u003cp\u003eTo further compare exploration patterns between male and female mice, we segmented the MBM environment into 3\u0026times;3mm bins and calculated the fold-change (and P values) in occupancy of male versus female mice in a bin-wise manner. These data may be represented as statistical occupancy maps and volcano plots (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003el, S6e). Stronger spatial learning performance in males was reflected by a shorter and more focused exploration pattern, while females exhibited a more scattered exploration pattern (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003el, left panel, S5e left panel). While no significant difference was observed on the first day of training (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003el, middle panel, S6e middle panel), females continued to explore the periphery of the MBM surface by the last day of training (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003el, right panel, S6e right panel). Altogether, we identified higher spatial learning accuracy in male compared with female mice tested in the MBM.\u003c/p\u003e \u003cp\u003e \u003cb\u003eCharacterization of spatial learning deficits in the Ts65Dn mouse model of DS in the MBM.\u003c/b\u003e DS, caused by a trisomy in human chromosome 21 (Hsa-21), is the most common chromosomal abnormality in humans and the most prevalent genetic cause of intellectual disability\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. Hsa-21 contains approximately 233 protein-coding genes, 423 non-protein coding genes, and numerous other functional genomic elements\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. The Amyloid precursor protein (APP) gene, located within Hsa-21, is triplicated in DS, such that APP is over-expressed in affected individuals with DS compared with euploid individuals\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. This heightened expression results in APP-dependent Alzheimer-like neuropathology in ~\u0026thinsp;88% of all individuals with DS by the age of 65\u003csup\u003e19\u003c/sup\u003e. The Ts65Dn mouse model of DS encompasses a partial trisomy of mouse chromosome 16, which includes 92 genes orthologous to Hsa21. As a result, this model recapitulates many of the cognitive, behavioral, structural, and physiological abnormalities of DS\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eTo characterize the cognitive deficits of Ts65Dn mice, we investigated their performance in the MBM using our trained classifier. Eight-month-old male Ts65Dn and their respective genetic background control strain (n\u0026thinsp;=\u0026thinsp;14 per group) were trained in the MBM for 10 days using a medium difficulty level achieved by placing the escape hole between the periphery and the center of the apparatus (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ea). Although learning was observed throughout training in both groups, Ts65Dn mice exhibited a profound spatial learning deficit compared to WT controls, reflected in reduced usage of highly spatial strategies (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eb) starting from the first day. \u003cem\u003eRandom\u003c/em\u003e search was more prevalent in Ts65Dn mice (66.67%) than in controls (57.14%) on the first day of training (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eb). By the last day, 9.52% of Ts65Dn trials were classified as \u003cem\u003eRandom\u003c/em\u003e, compared with 2.38% \u003cem\u003eRandom\u003c/em\u003e searches in WT controls (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eb). Overall, WT controls used more effective spatial strategies in 97.6% of the trials, compared with 85.71% in the Ts65Dn group. Accordingly, the cognitive scores of Ts65Dn mice were lower than those of WT mice throughout training (P\u0026thinsp;\u0026lt;\u0026thinsp;0.0001, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ec). Importantly, latency to reach the target hole entry did not differ between groups (P\u0026thinsp;=\u0026thinsp;0.76, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ed), which corresponded with higher distance (P\u0026thinsp;\u0026lt;\u0026thinsp;0.0001, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ee), higher speed (P\u0026thinsp;\u0026lt;\u0026thinsp;0.0001, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ef), and lower path efficiency (P\u0026thinsp;\u0026lt;\u0026thinsp;0.0001, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eg) in Ts65Dn mice compared with WT controls. These findings provide an example of the need for comprehensive analysis of mice performance in spatial learning tasks beyond comparison of latencies. Ts65Dn mice also exhibited reduced spatial accuracy, indicated by a higher number of entries to non-target holes (P\u0026thinsp;\u0026lt;\u0026thinsp;0.0001, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eh), a profound reference memory deficit (P\u0026thinsp;\u0026lt;\u0026thinsp;0.0001, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ei), and a milder working memory impairment (P\u0026thinsp;\u0026lt;\u0026thinsp;0.01, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ej). Additionally, the exploration trajectories of Ts65Dn mice covered a higher percentage of the MBM table (P\u0026thinsp;\u0026lt;\u0026thinsp;0.0001, Fig. \u003cspan refid=\"MOESM7\" class=\"InternalRef\"\u003eS7\u003c/span\u003ea-d). However, success rate did not differ between groups (P\u0026thinsp;=\u0026thinsp;0.77, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ek), indicating compensation of less efficient exploration strategies by higher exploration speed. Using statistical occupancy maps, we found that Ts65Dn mice spent significantly less time in the vicinity of the target location throughout training (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003el, left panel, Fig. \u003cspan refid=\"MOESM7\" class=\"InternalRef\"\u003eS7\u003c/span\u003ee, left panel). Accordingly, they spent more time near the periphery of the table on the first day (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003el, middle panel, Fig. \u003cspan refid=\"MOESM7\" class=\"InternalRef\"\u003eS7\u003c/span\u003ee, middle panel), and exhibited less target-oriented exploration on the last day of training (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003el, right panel, Fig. \u003cspan refid=\"MOESM7\" class=\"InternalRef\"\u003eS7\u003c/span\u003ee, right panel). In sum, our findings indicate a clear spatial learning impairment in the Ts65Dn mouse model of DS due to a deficit in reference memory capacity associated with using less-efficient exploration strategies and increased exploration speed.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eAssociation of spatial working memory impairment and circular explorations in the 5xFAD mouse model of Alzheimer disease.\u003c/b\u003e Spatial learning ability heavily relies on the integrity of hippocampal and para-hippocampal brain regions\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e. The hippocampus is also specifically vulnerable to Alzheimer disease (AD) pathology\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. Therefore, we investigated the impact of Amyloid-β (Aβ) pathology on the spatial strategy utilization of transgenic AD mice in the MBM. The 5xFAD mouse strain, which models early-onset AD, encompasses five early-onset AD-related mutations: the Swedish (K670N, M671L), London (V717I), and Florida (I716V) mutations in APP and the M146L and L286V mutations in Presenilin 1 (PS1)\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. As a result, 5xFAD mice exhibit early and profound Aβ pathology in the brain. Eight-month-old 5xFAD (n\u0026thinsp;=\u0026thinsp;9) and their respective WT control male mice (n\u0026thinsp;=\u0026thinsp;10) were trained to find the central (most difficult) target of the MBM for 6 days (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ea). 5xFAD mice exhibited reduced usage of highly spatial strategies (i.e., \u003cem\u003eDirect, Corrected\u003c/em\u003e, and \u003cem\u003elong correction\u003c/em\u003e) by the third day of training compared with WT controls (40.74%, 83.33%, respectively, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eb). The most prevalent strategies in 5xFAD mice were \u003cem\u003eCircling\u003c/em\u003e and \u003cem\u003eAccidental circling\u003c/em\u003e, represent together 55.55% of strategies used. The performance of WT mice reached an optimum on day 5, with 80% of WT trials classified as highly spatial. In comparison, 5xFAD mice used these strategies at a prevalence of 44.44%. Accordingly, the overall cognitive score of 5xFAD mice was lower compared to WT controls (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ec). Unlike the performance of Ts65Dn mice, 5xFAD mice exhibited increased latency to target entry (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ed and Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ed, respectively), while only mild difference in exploration distance (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ee) and no difference in speed was observed (P\u0026thinsp;=\u0026thinsp;0.19, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ef). Path efficiency was also reduced in 5xFAD mice compared with controls (P\u0026thinsp;\u0026lt;\u0026thinsp;0.01, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eg), but was only associated with early stages of training. Importantly, lower accuracy was observed in 5xFAD mice, indicated by elevated time in non-target holes (P\u0026thinsp;\u0026lt;\u0026thinsp;0.01, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eh). Interestingly, reference memory capacity of 5xFAD mice did not differ from WT controls (P\u0026thinsp;=\u0026thinsp;0.53, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ei), but working memory capacity was significantly reduced in these mice (P\u0026thinsp;\u0026lt;\u0026thinsp;0.01, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ej), which resulted in lower success rate compared with WT mice (P\u0026thinsp;\u0026lt;\u0026thinsp;0.0001, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ek). Accordingly, statistical occupancy maps analysis revealed that exploration trajectories of 5xFAD mice covered a higher percentage of the MBM table surface (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05, Fig. \u003cspan refid=\"MOESM8\" class=\"InternalRef\"\u003eS8\u003c/span\u003ea-d), with a clear tendency to explore the periphery of the surface (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003el, S8e). Overall, we report a working memory impairment in 5xFAD mice trained in the MBM, which is associated with a higher prevalence of the \u003cem\u003eCircling\u003c/em\u003e and \u003cem\u003eAccidental circling\u003c/em\u003e strategies.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe MBM is a novel modified variant of the traditional BM task for spatial learning\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. It combines the advantages of the MWM and the BM while avoiding their disadvantages\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. As a result, the MBM avoids water stress and its related technical complications, including lengthy operation times characteristic of the MWM. It also avoids non-spatial strategies, such as circling, that are characteristic of the BM. As with more traditional spatial learning tasks such as the MWM and the BM, in which spatial strategy classifiers provide additional layers of information to be extracted\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e,\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e, we set out to generate an unbiased classifier that effectively classifies cognitive strategies in the MBM, as a tool for the research community.\u003c/p\u003e \u003cp\u003eThe algorithm presented herein can effectively analyze MBM data obtained from different transgenic mice with and without cognitive impairments in an unbiased manner while providing a cognitive score scale that assesses memory acquisition.\u003c/p\u003e \u003cp\u003eTraditionally, performance in spatial learning tasks is analyzed according to one-dimensional parameters such as path efficiency, working and reference errors, and latency to reach the target. However, focusing solely on these parameters fails to fully capture the animal\u0026rsquo;s spatial cognitive capacity. We argue that utilizing a spatial learning paradigm superior to traditionally used paradigms (e.g., the MWM or the BM), combined with an added layer of information on the spatial strategies utilized by the rodents, is advantageous to optimizing experimental efficacy and promoting research output. In addition, the approach can sensitively identify behavioral nuances in pathologically relevant conditions, such as spatial learning deficits found in DS and AD mouse models. Therefore, in summary, the algorithm presented herein offers the research community an improved, powerful, and precise tool for assessing spatial learning in rodents.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData Availability statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll data including a user interface will be available upon request. Requests should be addressed to Prof. Eitan Okun, PH.D., ([email protected]) or Tomer Illouz, Ph.D., [email protected]).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contribution statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTI conducted the research and wrote the manuscript, LABA conducted behavioral experiments, RM assisted in all experiments, EO conceived the project and participated in writing the manuscript. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAdditional information\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe Authors declare no competing financial and/or non-financial interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eNyberg, N., Duvelle, E., Barry, C. \u0026amp; Spiers, H. J. Spatial goal coding in the hippocampal formation. Neuron 110, 394\u0026ndash;422, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.neuron.2021.12.012\u003c/span\u003e\u003cspan address=\"10.1016/j.neuron.2021.12.012\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLaczo, M. \u003cem\u003eet al.\u003c/em\u003e Different Profiles of Spatial Navigation Deficits In Alzheimer's Disease Biomarker-Positive Versus Biomarker-Negative Older Adults With Amnestic Mild Cognitive Impairment. Front Aging Neurosci 14, 886778, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fnagi.2022.886778\u003c/span\u003e\u003cspan address=\"10.3389/fnagi.2022.886778\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLavenex, P. B. \u003cem\u003eet al.\u003c/em\u003e Allocentric spatial learning and memory deficits in Down syndrome. Front Psychol 6, 62, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fpsyg.2015.00062\u003c/span\u003e\u003cspan address=\"10.3389/fpsyg.2015.00062\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePlacido, J. \u003cem\u003eet al.\u003c/em\u003e Spatial navigation in older adults with mild cognitive impairment and dementia: A systematic review and meta-analysis. Exp Gerontol 165, 111852, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.exger.2022.111852\u003c/span\u003e\u003cspan address=\"10.1016/j.exger.2022.111852\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJohnsen, S. H. W. \u0026amp; Rytter, H. M. Dissociating spatial strategies in animal research: Critical methodological review with focus on egocentric navigation and the hippocampus. Neurosci Biobehav Rev 126, 57\u0026ndash;78, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.neubiorev.2021.03.022\u003c/span\u003e\u003cspan address=\"10.1016/j.neubiorev.2021.03.022\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIllouz, T. \u003cem\u003eet al.\u003c/em\u003e Unbiased classification of spatial strategies in the Barnes maze. Bioinformatics 32, 3314\u0026ndash;3320, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/bioinformatics/btw376\u003c/span\u003e\u003cspan address=\"10.1093/bioinformatics/btw376\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIllouz, T., Madar, R., Louzoun, Y., Griffioen, K. J. \u0026amp; Okun, E. Unraveling cognitive traits using the Morris water maze unbiased strategy classification (MUST-C) algorithm. Brain Behav Immun 52, 132\u0026ndash;144, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.bbi.2015.10.013\u003c/span\u003e\u003cspan address=\"10.1016/j.bbi.2015.10.013\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMorris, R. Developments of a water-maze procedure for studying spatial learning in the rat. J Neurosci Methods 11, 47\u0026ndash;60, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/0165-0270(84)90007-4\u003c/span\u003e\u003cspan address=\"10.1016/0165-0270(84)90007-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (1984).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOlton, D. S. \u0026amp; Samuelson, R. J. Remembrance of places passed: Spatial memory in rats. Journal of Experimental Psychology: Animal Behavior Processes 2, 97\u0026ndash;116, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1037/0097-7403.2.2.97\u003c/span\u003e\u003cspan address=\"10.1037/0097-7403.2.2.97\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (1976).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlamed, J., Wilcock, D. M., Diamond, D. M., Gordon, M. N. \u0026amp; Morgan, D. Two-day radial-arm water maze learning and memory task; robust resolution of amyloid-related memory deficits in transgenic mice. Nat Protoc 1, 1671\u0026ndash;1679, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/nprot.2006.275\u003c/span\u003e\u003cspan address=\"10.1038/nprot.2006.275\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2006).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBarnes, C. A. Memory deficits associated with senescence: a neurophysiological and behavioral study in the rat. J Comp Physiol Psychol 93, 74\u0026ndash;104, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1037/h0077579\u003c/span\u003e\u003cspan address=\"10.1037/h0077579\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (1979).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIllouz, T., Madar, R. \u0026amp; Okun, E. A modified Barnes maze for an accurate assessment of spatial learning in mice. J Neurosci Methods 334, 108579, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jneumeth.2020.108579\u003c/span\u003e\u003cspan address=\"10.1016/j.jneumeth.2020.108579\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eReeves, R. H. \u003cem\u003eet al.\u003c/em\u003e A mouse model for Down syndrome exhibits learning and behaviour deficits. Nat Genet 11, 177\u0026ndash;184, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/ng1095-177\u003c/span\u003e\u003cspan address=\"10.1038/ng1095-177\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (1995).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRueda, N., Florez, J. \u0026amp; Martinez-Cue, C. Mouse models of Down syndrome as a tool to unravel the causes of mental disabilities. \u003cem\u003eNeural plasticity\u003c/em\u003e 2012, 584071, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1155/2012/584071\u003c/span\u003e\u003cspan address=\"10.1155/2012/584071\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMonfort, P., Gomez-Gimenez, B., Llansola, M. \u0026amp; Felipo, V. Gender differences in spatial learning, synaptic activity, and long-term potentiation in the hippocampus in rats: molecular mechanisms. ACS Chem Neurosci 6, 1420\u0026ndash;1427, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1021/acschemneuro.5b00096\u003c/span\u003e\u003cspan address=\"10.1021/acschemneuro.5b00096\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePiber, D., Nowacki, J., Mueller, S. C., Wingenfeld, K. \u0026amp; Otte, C. Sex effects on spatial learning but not on spatial memory retrieval in healthy young adults. Behav Brain Res 336, 44\u0026ndash;50, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.bbr.2017.08.034\u003c/span\u003e\u003cspan address=\"10.1016/j.bbr.2017.08.034\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYuan, L. \u003cem\u003eet al.\u003c/em\u003e Gender Differences in Large-Scale and Small-Scale Spatial Ability: A Systematic Review Based on Behavioral and Neuroimaging Research. Front Behav Neurosci 13, 128, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fnbeh.2019.00128\u003c/span\u003e\u003cspan address=\"10.3389/fnbeh.2019.00128\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRafii, M. S., Kleschevnikov, A. M., Sawa, M. \u0026amp; Mobley, W. C. Down syndrome. Handb Clin Neurol 167, 321\u0026ndash;336, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/B978-0-12-804766-8.00017-0\u003c/span\u003e\u003cspan address=\"10.1016/B978-0-12-804766-8.00017-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDoran, E. \u003cem\u003eet al.\u003c/em\u003e Down Syndrome, Partial Trisomy 21, and Absence of Alzheimer's Disease: The Role of APP. J Alzheimers Dis 56, 459\u0026ndash;470, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3233/JAD-160836\u003c/span\u003e\u003cspan address=\"10.3233/JAD-160836\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTukker, J. J. \u003cem\u003eet al.\u003c/em\u003e Microcircuits for spatial coding in the medial entorhinal cortex. Physiol Rev 102, 653\u0026ndash;688, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1152/physrev.00042.2020\u003c/span\u003e\u003cspan address=\"10.1152/physrev.00042.2020\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShipton, O. A., Tang, C. S., Paulsen, O. \u0026amp; Vargas-Caballero, M. Differential vulnerability of hippocampal CA3-CA1 synapses to Abeta. Acta Neuropathol Commun 10, 45, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s40478-022-01350-7\u003c/span\u003e\u003cspan address=\"10.1186/s40478-022-01350-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOakley, H. \u003cem\u003eet al.\u003c/em\u003e Intraneuronal beta-amyloid aggregates, neurodegeneration, and neuron loss in transgenic mice with five familial Alzheimer's disease mutations: potential factors in amyloid plaque formation. J Neurosci 26, 10129\u0026ndash;10140, doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1523/JNEUROSCI.1202-06.2006\u003c/span\u003e\u003cspan address=\"10.1523/JNEUROSCI.1202-06.2006\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2006).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-3075861/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3075861/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAssessment of spatial learning abilities is central to behavioral neuroscience and a pillar of animal model validation and drug development. However, biases introduced by the apparatus, environment, or experimentalist represent a critical challenge to the test validity. We have recently developed the Modified Barnes Maze (MBM) task, a spatial learning paradigm that overcomes inherent behavioral biases of animals in the classical Barnes maze. The specific combination of spatial strategies employed by mice is often considered representative of the level of cognitive resources used. Herein, we have developed a convolutional neural network-based classifier of exploration strategies in the MBM that can effectively provide researchers with enhanced insights into cognitive traits in mice.\u003c/p\u003e \u003cp\u003eFollowing validation, we compared the learning performance of female and male C57BL/6 mice, as well as that of Ts65Dn mice, a model of Down syndrome, and 5xFAD mice, a model of Alzheimer\u0026rsquo;s disease. Male mice exhibited more effective navigation abilities than female mice, reflected in higher utilization of effective spatial search strategies. Compared to wildtype controls, Ts65Dn mice exhibited reduced usage of spatial strategies despite similar success rates in completing this spatial task. These data exemplify the need for deeper strategy classification tools in dissecting complex cognitive traits. In sum, we provide a machine-learning-based strategy classifier that extends our understanding of mice\u0026rsquo;s spatial learning capabilities while enabling a more accurate cognitive assessment.\u003c/p\u003e","manuscriptTitle":"Unbiased analysis of spatial learning strategies in a modified Barnes maze using convolutional neural networks","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2023-07-12 14:32:47","doi":"10.21203/rs.3.rs-3075861/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Major revision","date":"2023-10-24T02:19:13+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2023-10-23T09:10:59+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2023-09-07T23:54:29+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"7ec2fad4-c1e2-4b1e-82fe-5f4e729a49af","date":"2023-08-22T18:52:48+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"f4765d79-e6bb-42f0-8d90-1333cd3d0c0c","date":"2023-08-22T12:52:46+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2023-08-22T12:44:54+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2023-08-22T12:44:07+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2023-07-07T10:57:15+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2023-07-07T10:52:18+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2023-06-17T12:16:13+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"3df21a44-dd6d-44ee-9f58-a9e265ed96ca","owner":[],"postedDate":"July 12th, 2023","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":23063999,"name":"Biological sciences/Neuroscience/Learning and memory/Spatial memory"},{"id":23064000,"name":"Biological sciences/Neuroscience/Cognitive ageing"}],"tags":[],"updatedAt":"2024-07-13T15:05:52+00:00","versionOfRecord":{"articleIdentity":"rs-3075861","link":"https://doi.org/10.1038/s41598-024-66855-8","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2024-07-10 15:05:52","publishedOnDateReadable":"July 10th, 2024"},"versionCreatedAt":"2023-07-12 14:32:47","video":"","vorDoi":"10.1038/s41598-024-66855-8","vorDoiUrl":"https://doi.org/10.1038/s41598-024-66855-8","workflowStages":[]},"version":"v1","identity":"rs-3075861","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3075861","identity":"rs-3075861","version":["v1"]},"buildId":"_2-kVJe1T_tPrBINL-cwx","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}