Abstract
There is a growing demand for sophisticated behavioral analysis systems that minimize bias in multi-animal and semi-natural environments. We present “IntelliProfiler,” an advanced system using radio frequency identification (RFID) technology to enable fully automated behavior analyses of multiple mice within a home cage. IntelliProfiler continuously monitors up to 16 mice, capturing both locomotor activity and social dynamics over extended periods. Our findings revealed that male mice maintain broader social distances than females, with group size playing a key role in shaping male social network dynamics. Furthermore, aging in males significantly impacts both locomotor activity and social interaction in a group size-dependent manner. With its ability to provide in-depth analysis of individual and group behaviors, IntelliProfiler offers a novel approach for exploring complex social interactions and group dynamics, advancing the field of behavioral science.
Introduction
Behavioral analyses using model animals are crucial for understanding neural functions and pathological states. Due to its ease of handling and suitability for genetic manipulations, the laboratory mouse is widely used as a model animal to study neural bases of various conditions, including psychiatric, neurodevelopmental, and neurodegenerative disorders. Traditional methods for analyzing neural functions in mice involve assessing behaviors, such as locomotion, exploration, social interaction, and learning/memory through specific tests including the open field test, the elevated plus maze, the social interaction test, and the Morris water maze1–4. However, these traditional approaches come with several limitations.
A major limitation of traditional behavioral analyses is the low throughput of data acquisition, as most analyses are manually conducted over limited periods under artificial conditions. Typically, behavior is recorded for a single mouse in a specific experimental setup. For instance, the open field test measures general locomotor activity and anxiety-related behavior by placing a single mouse in an open arena and tracking its movements and time spent in different zones1. Similarly, the three-chamber social interaction test assesses sociability by comparing the time a mouse interacts with a novel mouse versus an empty chamber, offering insights into social preferences5. A more significant concern, however, is that interactions between experimenters and animals often introduce biases that influence results6–8. Handling by experimenters can elevate heart rate, body temperature, spontaneous activity, and anxiety, potentially impacting memory and learning abilities7,9. Moreover, variations in behavioral data have been linked to the researcher’s gender and environmental conditions8,10. Therefore, there is a growing need for automated, unbiased systems to analyze behavior in natural (home-caged) conditions to ensure more robust and reliable data.
Recent trends in behavioral analyses have shifted towards studying group dynamics in semi-natural environments. One prominent approach is the use of video tracking systems with computer visions, such as DeepLabCut (DLC)11–13. DLC leverages deep learning to achieve markerless pose estimation, offering high precision tracking with minimal training data14. Despite its accuracy, video tracking systems face limitations when tracking larger groups. Identifying individual animals becomes technically challenging due to issues like visual cue swapping or obscuring during group behaviors, such as huddling or fighting.
Another approach involves tracking systems using contactless tags, such as radio frequency identification (RFID) chips. IntelliCage, a pioneering RFID system, utilizes RFID readers placed at the corners of a cage to automatically monitor the frequency and timing of animals entering chambers15. IntelliCage has been widely used to investigate spontaneous behavior and learning in group-housed mice16,17. Another RFID-based system, ECO-HAB, consists of four open fields connected by tubes, with RFID readers positioned at the junctions, allowing the study of free movement and social interactions among mice by tracking their movement between fields18,19. More recently, RFID floor arrays have been developed to track multiple animals with higher time resolution20. While these RFID systems provide valuable insights into group behavior, they often lack the spatial or temporal resolution required for more detailed analyses.
In this study, we developed an innovative RFID-based behavioral analysis system, IntelliProfiler, to automatically monitor multiple mice in a home-cage environment. IntelliProfiler excels in its ability, accommodating groups ranging from four to 16 mice, while offering superior spatial and temporal resolution compared to existing systems. It enables long-term, continuous monitoring of both activity and sociability with minimal researcher intervention. Using IntelliProfiler, we conducted 72-hour continuous behavioral analyses on large groups of male and female wild-type mice and compared behavioral patterns between sexes, as well as between young and aged male mice, to investigate the effects of aging on social dynamics. The insights gained through IntelliProfiler provide a deeper understanding of group behavior dynamics, advancing the precision and reliability of behavior research.
Results
Long-term activity measurement of group-housed mice
First, we set up our IntelliProfiler system using the antenna board comprising 96 antenna tiles arranged in a 12 x 8 grid (Fig. 1a). A large plastic cage (58 cm x 38 cm x 21.5 cm = width × depth × height) with a cage lid with food pellets, and three water bottles for long-term housing was placed on the antenna board (Fig. 1a, b). The IntelliProfiler software (see Methods) offers multiple functions: (1) data collection and preprocessing, (2) calculation of travel distance, relative distance between two mice, and spatial interaction proportion, (3) network analysis, (4) principal component analysis (PCA), and (5) figure visualization. Data including time, antenna ID, and transponder ID, were exported from the antenna board via Universal Serial Bus (USB) using TeraTerm software (Fig. 1c). In the data collection and preprocessing step, raw data were converted into a per-second format with time, antenna ID, transponder ID, and X-Y coordinates using initial function of IntelliProfiler software we developed (Fig. 1d, see Data collection and preprocessing in Method section). Further calculations, analyses, and figure visualization were conducted using the converted data generated in the data collection and preprocessing step (see Travel distance for each mouse, Relative distance between two mice, Spatial interaction analysis, Close contact: defining social proximity, Network Analysis, Principal component analysis, Figure visualization in Method section). We conducted continuous 72-hour behavioral analyses on 8-week old male and female mice with group sizes of four, eight, and 16 mice (Fig. 1e). Due to the dislodging of one RFID tag in the group of 16 male mice, data from 15 male mice were used for analysis, following the methodology of a previous article21.
a Schematic of the IntelliProfiler. The system tracks the location of mice implanted with RFID tags via detectors placed in a 96-grid array at the base, with each grid measuring 5 cm x 5 cm. Location information is converted into XY coordinates (X=12, Y=8), based on antenna IDs (1-96). b Photograph of the IntelliProfiler. c Original data, capturing information on Time, Antenna ID, and Transponder ID. d Processed data converted for the IntelliProfiler. The data is refined to includ Time, Antenna ID, Transponder ID, and XY coordinates. Decimal information is removed from the time data. Iif multiple individuals are detected on the same antenna ID simultaneously, their information is separated into different rows. e Behaior analysis was conducted over a continuous 72-hour period for male and female mice in groups of four-, eight– and 15/16. Additionally, young (8-week old) and aged (53-week old) male mice were analyzed in groups of four and eight.
In the males, cumulative travel distances for every 2 hours exhibited the highest travel distance values in the four-mouse group than the eight– and 15-mouse groups during most of the time (Fig. 2a). The male mice exhibited the highest travel distance immediately after entry, dropping by less than half within 2 hours (Fig. 2a). Their travel distance increased again during the transition from light to dark period (Fig. 2a). In contrast, females showed less active and stable travel distances across almost all time points (Fig. 2b).
a-j Cumulative travel distances measured in 2-hour intervals (a, b, e-g) and over 12-hour intervals (light-dark cycles) (c, d, h-j). a-d Analysis of locomotor activity based on group size, comparing groups of four, eight, and 15 males (a, c) and four, eight, and 16 female mice (b, d). e-j Analysis of sex differences in locomotor activity, comparing groups of four (e, h), eight (f, i) and 15-16 (g, j) male and female mice. k-n Heatmaps showing relative locomotor activity over 2-hour intervals (k, l) and 12-hour intervals (m, n) for groups of four, eight, and 15-16 mice. Values in the bar or line graphs are presented as mean ± SD, with box plots displaying min/max whiskers. Statistical analysis was conducted using Holm-Sidak method for multiple comparison. Significance levels are indicated as *p < 0.05, **p < 0.01, and ***p < 0.001. p < 0.001 (c), p = 0.011 (d), p < 0.001 (e), p = 0.894 (f), p = 0.009 (g), p < 0.001 (h), p = 0.920 (i), p = 0.139 (j).
To understand the general trends of locomotor activity in the male and female mice, we analyzed cumulative travel distances in 12-hour intervals (Fig. 2c, d). The males showed the highest travel distance just after their entry and significantly higher travel distance in the group of four mice than in the groups of eight and 15 mice (Fig. 2c). Small but significant differences were observed between the eight– and 15-mouse groups in the light period on day 1 and the dark period on day 3 (Fig. 2c). On the other hand, females showed similar travel distances through the period among groups except for day 1 (Fig. 2d).
To examine sex differences in locomotor activity within the same group size, we compared 2-hour intervals activities in the groups of four, eight, and 15/16 mice (Fig. 2e-g). In the four-mouse group, male mice showed higher travel distances throughout the period than females, with significantly higher travel distances observed at six time points in the dark periods (Fig. 2e). However, no periodic trend was observed in eight– and 15/16-mouse groups (Fig. 2f, g). In the cumulative travel distances in 12-hour intervals (Fig. 2h-j), males showed higher travel distances throughout the period in the four-mouse group than female mice except for the light period on day 3 (Fig. 2h). On the other hand, no noticeable sex difference was observed in the groups of eight and 15/16 mice (Fig. 2i, j).
The time course of relative locomotor activity for each mouse was visualized as a heatmap (Fig. 2k-n). Male mice showed a marked increase in activity immediately after the introduction and at the onset of the dark period (Fig. 2k). A pronounced contrast of locomotor activity was observed between light and dark periods over the course of 3 days (Fig. 2m). In contrast, female mice displayed relatively more variable activity patterns over time (Fig. 2l), with some showing elevated activity even during the light period (Fig. 2n). These findings suggest that male mice, particularly in smaller group sizes, exhibit higher locomotor activity, pointing to potential sex-specific behavioral responses to group size.
Social interactions between two individuals of group-housed mice
To assess social interactions among group-housed mice, we tracked changes in the distance between two individual mice. The spatial relationships were classified into four categories based on proximities: “Same” – the other mouse occupied the same grid as the focal mouse; “Close” – the other mouse was in the adjacent grids; “Intermediate” – the other mouse was located beyond the surrounding grids; and “Away” – the other mouse was farther than “Intermediate” (Fig. 3a). The transition in social distance ratios were recorded in 2-hour intervals for groups of four, eight, and 15 male mice (Fig. 3b-d) and for groups of four, eight, and 16 female mice (Fig. 3e-g).
a Definition of the spatial relationships between two individual mice. When another mouse is in the same grid as the focal mouse (both depicted in yellow), the relationship is defined as “Same.” If the other mouse is in the surrounding grids (depicted in green), it is defined as “Close.” If the other mouse is outside the surrounding grids (depicted in light blue), it is defined as “Intermediate.” If the other mouse is beyond the “Intermediate” zone (depicted in blue), the relationship is categorized as “Away.” b-g The 2-hour bin transition of proportions based on “Same,” “Close,” “Intermediate,” and “Away” for groups of four male mice (b), eight male mice (c), 15 male mice (d), four female mice (e), eight female mice (f), and 16 female mice (g). The average proportion of distance categories during the light period and the dark period is shown for each group. Chi-square p-values comparing the proportions between light and dark periods: p < 0.001 (b, four males), p = 0.256 (c, eight males), p = 0.114 (d, 15 males), p = 0.005 (e, four females), p = 0.049 (f, eight females), p = 0.007 (g, 16 females). The p-values of chi-square for the group size, p < 0.001 (b-d, males in the light phase), p = 0.002 (b-d, males in the dark phase), p < 0.001 (e-g, females in the light phase), p = 0.092 (e-g, females in the dark phase).
The proportion of “Same” increased during the light period and decreased during the dark period in the four-mouse group, but this pattern was not observed in the other groups (Fig. 3b-g). A similar trend was found for the “Close” category, where its proportion rose in the light period and declined in the dark period across groups of four, eight, and 15 males, as well as four, eight, and 16 females (Fig. 3b-g). The proportion of “Intermediate” was relatively low in the four-mouse group but increased with group size, without showing a consistent periodic trend observed (Fig. 3b-g). In contrast, the proportion of “Away” decreased during the light period and increased in the dark period across all groups (Fig. 3b-g).
The average proportion of each social distance category during the light and dark periods was compared across groups (Fig. 3b-g). The proportion of “Same” decreased as group size increased (Fig. 3b-g). During the light period, the percentage of “Close” was higher than during the dark period across all groups, though there was no clear trend concerning group size (Fig. 3b-g). The proportion of “Intermediate” increased with group size during the light period, and was higher in the groups of eight and 15/16 mice than the four-mouse group during the dark period, but similar between the eight-mouse and 15/16-mouse groups (Fig. 3b-g). In contrast, the proportion of “Away” was highest in the 15-male group during both periods, although no clear pattern emerged for the female groups. In summary, the “Same” category was more prevalent in the four-male and four-or eight-female groups during the light period, but no consistent periodic trend was found in the larger groups. Meanwhile, the proportion of “Away” was consistently lower across all groups during the light periods (Fig. 3b-g). These analyses revealed qualitatively distinct patterns of social distance in the eight– and 15/16-mouse groups compared to the four-mouse group.
Significant differences in the proportions of each category between the light and dark periods were observed in the groups of four males, as well as in the groups of four, eight, and 16 females (Fig. 3b, e-g). However, no significant differences were found in the groups of eight and 15 males (Fig. 3c, d). Regarding group size, significant differences were noted in the male groups during both the light and dark periods, and in the female groups during the light period (Fig. 3b-g), but no significant differences were observed in the female groups during the dark period (Fig. 3e-g).
We then introduced the Close Contact Ratio (CCR), a metric to quantify social proximity to analyze social interactions. This ratio reflects the proportion of time two mice are either in the “Same” or “Close” (Fig. 4a). When CCR data were divided into 2-hour intervals for both male and female groups (Fig. 4b, c), males in the four-mouse group exhibited the highest CCR compared to the eight– and 15-mouse groups throughout the entire period (Fig. 4b). In male mice, the CCR was initially low immediately after entry, gradually increasing until 18:00, followed by a gradual decline until 20:00, with a similar pattern recurring from day 2 onward (Fig. 4b). In contrast, the females in the 16-mouse group showed a significantly lower CCR compared to the four– and eight-mouse groups (Fig. 4c), although no consistent time-dependent pattern was observed. We also analyzed the average CCR over 12-hour intervals (Fig. 4d, e). In male mice, the CCR decreased as the group size increased from four to eight to 15 mice during the light period (Fig. 4d). The 15-mouse group had the lowest CCR, while no significant difference was found between the four– and eight-mouse groups during the dark period (Fig. 4d). Similarly, female mice in the 16-mouse group tended to have the lowest CCR during the light period compared to the four– and eight-mouse groups, though no consistent differences were observed between the latter two groups (Fig. 4e). No clear trend was observed during the dark period in female mice (Fig. 4e).
a Definition of close contact, where two mice either in the “Same” grid (depicted in yellow) or in adjacent grids, classified as “Close” (depicted in green). b-k Average Close Contact Ratio (CCR) measured in in 2-hour intervals (b, c, f-h) and 12-hour intervals (d, e, i-k). b-e Analysis of CCR based on group size, comparing groups of four, eight, and 15 male mice (b, d) and four, eight, and 16 female mice (c, e). f-k Analysis of sex differences in CCR, comparing groups of four (f, i), eight (g, j) and 15/16 (h, k) male and female mice. l-o Heatmaps showing the time course of CCR over 2-hour intervals (l, m) and 12-hour intervals (n, o) for groups of four, eight, and 15-16 mice. Values in the bar or line graphs are presented as mean ± SD, with box plots showing min/max whiskers. Statistical analysis was conducted using Holm-Sidak method for multiple comparisons. Significance levels are indicated as *p < 0.05, **p < 0.01, and ***p < 0.001. p < 0.001 (b), p < 0.001 (c), p = 0.882 (f), p < 0.001 (g), p < 0.001 (h), p = 0.873 (i), p < 0.001 (j), p < 0.001 (k).
We also examined the sex differences in CCR using 2-hour intervals in four-, eight-, and 15/16-mouse groups (Fig. 4f-h). In the four-mouse group, females showed a significantly higher CCR during the initial 4 hours (Fig. 4f). However, CCR fluctuated between males and females throughout the remaining time points, suggesting that consistent sex differences in social proximity might not be present in this group (Fig. 4f). In contrast, in the eight– and 15/16-mouse groups, females consistently exhibited higher CCR than males at most time points, indicating more pronounced sex differences in social proximity (Fig. 4g, h). To assess overall trend, we analyzed the CCR in 12-hour intervals (Fig. 4i-k). In the four-mouse group, no clear sex differences in CCR were observed, except during the light period on day 2, when females had a higher CCR (Fig. 4i). However, in the eight-mouse group, females had a higher CCR than males across most periods, except during the dark period on day 3 (Fig. 4j). Similarly, in the 16-mouse group, females consistently exhibited a higher CCR than males across all the periods (Fig. 4k).
The time course of CCR, visualized as a heatmap for individual mice (Fig. 4l-o), revealed a strong periodic trend in the males, with CCRs peaking during the light period and decreasing during the dark period (Fig. 4l). In contrast, the female groups generally displayed higher CCRs for longer durations compared to males (Fig. 4l, m). The periodic patterns of CCRs were less pronounced in the 15/16 mouse groups, though the trend remained more apparent in males than females (Fig. 4l-o). These findings suggest that sex-specific social dynamics become more prominent in larger groups.
Aging effect
Since aging has been reported to affect many behavioral aspects, including locomotor activity and social interaction22,23, we compared young (8-week-old) and aged (53-week-old) male mice in groups of four and eight (Fig. 1e). Activity levels of the mice were analyzed using methods as those in Fig. 2. Both young and aged males in the four-mouse group generally had longer travel distances compared to those in the eight-mouse group at most time points (Fig. 5a). Regarding the group effect, aged males in the four-mouse group showed a longer travel distance immediately after their introduction and after the next transition from the dark and light period than those in the eight-mouse group (Fig. 5b). Regarding the age effect, there was no significant difference in travel distance between the young and aged males (Fig. 5c). In contrast, the aged male mice in the eight-mouse group had consistently longer travel distances at almost all time points compared to young male mice (Fig. 5d). The differences between young and aged mice in the eight-mouse group was also significant than other comparison between aged mice in the four– and eight-mouse groups as well as between young and aged mice in the four-mouse group (Fig. 5e-g).
a-g Cumulative travel distances measured in 2-hour intervals (a-d) and over 12-hour intervals (e-g). b, e Analysis of locomotor activity based on group size, comparing groups of four and eight aged mice. c, d, f, g Analysis of aging effects in locomotor activity, comparing groups of four young and aged male mice (c, f), groups of eight young and aged male mice (d, g). h-k Heatmaps showing relative locomotor activity over 2-hour intervals (h, i) and 12-hour intervals (j, k) for groups of four and eight mice. Values in the bar or line graphs are presented as mean ± SD, with box plots displaying min/max whiskers. Statistical analysis was conducted using Holm-Sidak method for multiple comparisons. Significance levels are indicated as *p < 0.05, **p < 0.01, and ***p < 0.001. p = 0.338 (c), p < 0.001 (d), p = 0.585 (f), p < 0.001 (g).
In the heatmap analysis, activity was also prominent immediately after the introduction and at the onset of the dark period in all groups, although aged mice in the eight-mouse group showed more activity throughout the period (Fig. 5h, i). In the heatmap of cumulative activity over 12-hour intervals, aged males exhibited more diverse and desynchronized activity, particularly in the eight-mouse group (Fig. 5j, k). Therefore, the spontaneous activity was different between young and aged mice.
Next, we examined the changes in spatial relationships among young and aged male mice in the four– and eight-mouse groups (Fig. 6a), similar to those in Fig. 3. Regarding group effects, aged mice in the four-mouse group exhibited lower percentage of “Same” category during the light and dark periods compared to young mice in the four-mouse group, while aged mice in the eight-mouse group showed lower percentage of “Away” in both periods compared to younger counterparts (Fig. 6b-e). In contrast, aged mice in both the four– and eight-mouse groups consistently exhibited higher percentage of “Intermediate” during both periods compared to young mice (Fig. 6b-e). These analyses revealed distinct social proximity patterns in aged male mice compared to young male mice, especially in the increased proportion of “Intermediate” in the aged groups.
a Definition of the spatial relationships between two individual mice. When another mouse is in the same grid as the focal mouse (both depicted in yellow), the relationship is defined as “Same.” If the other mouse is in the surrounding grids (depicted in green), it is defined as “Close.” If the other mouse is outside the surrounding grids (depicted in light blue), it is defined as “Intermediate.” If the other mouse is beyond the “Intermediate” zone (depicted in blue), the relationship is categorized as “Away.” b-e The 2-hour bin transition of proportions based on “Same,” “Close,” “Intermediate,” and “Away” for groups of four young male mice (b), eight young male mice (c), four aged male mice (d), eight aged male mice (e). b-e The average proportion of distance categories during the light period and the dark period is shown for each group. Chi-square p-values comparing the proportions between light and dark periods, p < 0.001 (b, four young males), p = 0.256 (c, eight young males), p = 0.836 (d, four aged males), p = 0.439 (e, eight aged males). The p-values of chi-square for the group size, p < 0.001 (b, c, young males in the light phase), p = 0.020 (b, c, young males in the dark phase), p = 0.115 (d, e, aged males in the light phase), p = 0.477 (d, e, aged males in the dark phase).
We then analyzed CCRs in 2-hour intervals for both young and aged male mice in the four– and eight-mouse groups (Fig. 7a), using methods similar to those in Fig. 4. Young males in the four-mouse group exhibited a higher CCR than those in the eight-mouse group at most time points (Fig. 7b). In contrast, aged males in the four-mouse group displayed a higher CCR from 18:00-20:00 on day 2 compared to those in the eight-mouse group (Fig. 7c). Regarding the age effect, aged mice in the four-mouse group consistently showed a lower CCR during the light period at nearly all time points compared to young mice in the four-mouse group (Fig. 7d). On the other hand, aged mice in the eight-mouse group displayed non-consistent periodic trends, with lower CCR amplitude of the light-dark cycle, compared to young mice in the eight-mouse group (Fig. 7e). The CCR differences between young and aged mice in the four-mouse group during the light period was also consistently more significant than those in the comparisons between aged mice in the four– and eight-mouse groups, as well as between young and aged mice in the eight-mouse group (Fig. 7f-h). In the heatmap analysis, the young mice exhibited robust periodic trends, with elevated CCR values during the light period and decreased values during the dark period, particularly in the four-mouse group (Fig. 7i). In contrast, the aged mice in both the four– and eight-mouse groups showed more stable, intermediate CCR values compared to young mice (Fig. 7j). In the heatmap of cumulative activity over 12-hour intervals, aged males demonstrated more ambiguous cyclic CCR patterns, particularly in the eight-mouse group (Fig. 7k, l). These findings suggest that aging in male mice altered not only locomotor activity but also the temporal dynamics of social behaviors.
a Definition of close contact, where two mice either in either “Same” (depicted in yellow) or in adjacent grids, classified as “Close” (depicted in green). b-h Average Close Contact Ratio (CCR) measured in 2-hour intervals (b-e) and 12-hour intervals (f-h). c, f Analysis of CCR based on group size, comparing groups of four and eight young and aged males. d, e, g, h Analysis of CCR based on aging effects, comparing groups of four young and aged male mice (d, g), groups of eight young and aged male mice (e, h). i-l Heatmaps showing the time course of CCR over 2-hour intervals (i, j) and 12-hour intervals (k, l) for groups of four and eight mice. Values in the bar or line graphs are presented as mean ± SD, with box plot showing min/max whiskers. Statistical analysis was conducted using Holm-Sidak method for multiple comparisons. Significance levels are indicated as *p < 0.05, **p < 0.01, and ***p < 0.001. p < 0.001 (c), p = 0.338 (d), p = 0.072 (e), p < 0.001 (f), p = 0.007 (g), p = 0.109 (h).
Characterization of group behaviors
To investigate the behavioral characteristics of each group, we conducted principal component analysis (PCA) using four key parameters. First, we defined normalized activity as the activity at time t, divided by the maximum activity value recorded over Days 1-3, and multiplied by 100 to standardize the data. This allowed us to calculate the normalized activity in both the light (AL) and dark (AD) periods by averaging activity values at each time point over three days. We selected four parameters for PCA: sociability, measured as CCR values during the light (SL) and dark (SD) periods (parameters 1 and 2, respectively), and normalized activity during the light (AL) and dark (AD) periods (parameters 3 and 4, respectively). Correlations among these parameters are shown in Supplementary Fig. 1, 2, while the PCA results with estimated clusters are displayed in Fig. 8. The number of clusters was determined using the Elbow method24 and further assessed through Silhouette plots25 (Supplementary Fig. 3). The PCA revealed distinct sex-biased groupings. Notably, the male mice in the 15-mouse group formed a well-defined separated cluster, whereas the male and female mice in the four-mouse groups and female mice in the eight-mouse group clustered together (Fig. 8a). Interestingly, female mice in the 16-mouse group split into two clusters, with one cluster aligning with the male mice in the eight-mouse group (Fig. 8b). This suggests that while smaller groups tend to blend, significant sex differences emerge in larger groups. In the feature space, AL and AD vectors pointed in the same direction, indicating similar patterns of activity, while SL and SD vectors diverged, reflecting different sociability dynamics between light and dark periods (Fig. 8c).
a-f Principal component analysis (PCA) of group behavior in four, eight, and 15/16 male and female mice (a-c) and in young and aged mice in groups of four and eight (d-f). b, e K-means clustering in PCA plots. c, f Analysis of feature vectors, focusing on Close Contact Ratio (CCR) and relative activity during the light and dark periods.
In terms of aging effects, young and aged mice in the four-mouse groups were largely positioned in the positive X and Y directions, whereas those in the eight-mouse groups clustered near the origin (0,0) or in the negative X direction (Fig. 8d). K-means clustering effectively separated the four– and eight-mouse groups (Fig. 8e), and each feature vector had a distinct direction (Fig. 8f), indicating that aging influenced both activity and sociability in distinct ways, aligning with our earlier observations (Fig. 5-7).
Network analysis of social interactions
We explored how group size influences social interaction networks (Fig. 9). Qualitatively, there was no noticeable difference in network structure between male and female mice in the four– and eight-mouse groups (Fig. 9). However, in the 15/16-mouse groups, male mice exhibited increased social proximity and developed more expanded networks, influenced by light-dark circadian rhythms. In contrast, female mice maintained relatively uniform network structures (Fig. 9). When comparing young and aged male mice in the four-mouse groups, no significant differences in network structure were observed (Fig. 10). In the eight-mouse groups, however, young male mice displayed increased social proximity and formed diverse networks that fluctuated with light-dark cycles, while aged male mice maintained more stable, uniform networks with only slight variation corresponding to circadian rhythms (Fig. 10). These findings suggest that male and female mice create distinct social environments, and that aging may impact the complexity and heterogeneity of the networks in larger groups.
Network analysis of social interactions during the light period on day 1 and the dark period on day 3 in groups of four, eight, and 15/16 male and female mice. Nodes represents individual mice, and edges reflects the relative social contacts between two individuals. The network graph was generated using Cytoscape. Supplementary Fig. 4 provides additional network diagrams, where unique color code for each individual highlight the individual-specific social dynamics within the network.
Network analysis of social interactions during the light period on day 1 and the dark period on day 3 in groups of four and eight young and aged males. Each circle represents an individual mouse, and the proximity of the lines indicates the relative social distance between two individuals. The network graph was generated using Cytoscape. Supplementary Fig. 5 provides further details, offering a clear understanding of individual-specific social dynamics across the networks.
Analysis of individual social interactions
Lastly, we focused on individual differences in social interactions through a pairwise analysis of CCR. In the heatmap presentations (Supplementary Fig. 6), no obvious individual differences were observed between sexes in the four– and eight-mouse groups. However, in the 15/16-mouse groups, two males (mouse ID #1 and #2) exhibited lower CCR scores compared to the others, while all female mice showed similar CCR patterns. Additionally, male mice tended to display higher CCR during the light period, which decreased during the dark period, whereas female mice maintained relatively stable CCR levels with less fluctuation across the light and dark periods, indicating a sex-dependent response to light. Regarding the aging effect, aged males showed lower CCR scores and more ambiguous, less defined time-dependent patterns compared to young males (Supplementary Fig. 7), consistent with mean values presented in Fig. 7. These findings suggest that individuality may develop progressively with aging in males, potentially contributing to the observed social dynamics.
Discussion
In this study, we developed a novel RFID-based behavioral analysis system, IntelliProfiler, which allows for continuous, unbiased monitoring of up to 16 mice within the home-cage environment. This system has provided clear insights into the effects of group size, sex, and aging on both locomotor activity and social interactions, offering a more comprehensive understanding of the semi-natural behaviors of group-housed mice than using traditional behavioral methods can achieve.
One of our key findings was the impact of group size on behavior. IntelliProfiler revealed significant differences in activity levels and social proximity, particularly among males, where larger groups reduced activity. Traditional behavior paradigms, such as the open-field test, are typically designed for individual animals26,27, and therefore fail to capture the complexities of behavior in group settings. These conventional paradigms may overlook critical dynamics unique to multi-animal environments. A recent article in Nature Digest emphasized the importance of studying animals under ‘natural conditions’ to gain deeper insights into their behavior28.
IntelliProfiler enables us to monitor authentic locomotor activity under semi-natural conditions, offering a more comprehensive understanding of how group size shapes individual activity and social dynamics.
In this study, we introduced CCR as a novel social parameter to measure social proximity, representing a significant advancement in behavioral analysis (Fig. 4). Traditional social interaction tests, such as the three-chamber test, offer limited insight into short-term interactions, but the CCR provides a more detailed, long-term view of group dynamics. Notably, we found that females tended to form stable clusters, while males exhibited more fluid, group size-dependent social patterns (Fig. 4, 8, 9). This underscores IntelliProfiler’s ability to detect subtle, yet important sex-specific social behaviors, often overlooked by conventional methods.
Aging also influenced locomotor activity and social interactions, with aged males displaying less distinct and more variable CCR patterns compared to young males. This finding aligns with previous research on aging and social behavior,23,29 but also suggests that IntelliProfiler offers a more nuanced understanding of how aging affects social dynamics, especially in larger groups (Fig. 7).
IntelliProfiler further provides an inexpensive, high-throughput platform for social network analysis, surpassing the limitations of human observation. Although sexual interaction between males and females is beyond our current purpose, our system would be applicable to analyze sexual or courtship behaviors in large groups. While video tracking systems have been useful for analyzing small groups, IntelliProfiler extends this capability to larger groups, making it a crucial tool for studying population-level behaviors.
Despite its advantages, IntelliProfiler has a few limitations. First, its spatial resolution is lower than video-based tracking systems like DLC11,30, as the current grid size is 5 cm x 5 cm, allowing two mice located in parallel. Additionally, determining body orientation is challenging when the animal remains within the same grid. Second, the system has difficulty distinguishing detailed social contacts, such as sniffing, mounting, or huddling. Lastly, the current RFID system requires surgery for transponder implantation, and it is not yet possible to eliminate the invasive effects due to the size of commercially available transponders. These challenges should be addressed in future developments.
Conclusion
The IntelliProfiler makes a significant advancement in animal behavior research, enabling continuous, detailed, and unbiased monitoring of individual movements and interactions within dynamic and complex social environments. Our findings highlight the critical roles of group size, sex, and aging in shaping behavior. This system provides a powerful tool for studying behavior in more naturalistic settings, offering new avenues for large-scale behavioral research and deeper insights into social dynamics.
Methods
Animals
C57BL/6J mice at 7-or 52-week old were purchased from CLEA Japan, Inc., and maintained in the animal facility at Tohoku University Graduate School of Medicine. An RFID tag (7 mm x 1.25 mm, Phenovance LCC) was subcutaneously implanted into the abdomen of each mouse under anesthesia using 2.0% isoflurane (MSD Japan). The mice were recovered for 1 week before starting behavioral recording. The recordings started at 8-or 53-week old. All experimental procedures were approved by the Ethics Committee for Animal Experiments at the Tohoku University Graduate School of Medicine (approval number: 2021MdA-020-13).
Hardware
The antenaboard, comprising 96 antenna tiles arranged in 12×8 grids (5 cm by 5 cm each) (Fig. 1a), was purchased from Phenovance LCC, Japan. The home cage used for the study was 2000P cage (58 cm x 38 cm x 21.5 cm = width × depth × height, Tecniplast, Italy), equipped with a cage lid with food pellets, and three water bottles for long-term housing (Fig. 1a, b). Both the antenaboard and home cage were placed inside a soundproof box (Phenovance LCC). The light-dark cycle in the soundproof box and testing room was 12:12 hours (08:00-20:00, light; 20:00-08:00, dark). Mice were introduced at 8:00 in the light period on day 1.
Software
R language (Version 4.2.2) was mainly used for preprocessing, data analysis, and visualization31. Python (Version 3.9.7), Prism 9, and Microsoft Excel (Version 16.86) were also used for preprocessing, data analysis, and visualization. R and Python packages required for the scripts are listed in Supplementary Table 1.
Data collection and preprocessing
Data including time, antenna ID, and transponder ID, were exported from the antenabord via Universal Serial Bus (USB) using TeraTerm software (Fig. 1c). The raw data were converted into a format per second with time, antenna ID, transponder ID, and X-Y coordinates (Fig. 1d). Missing values were interpolated by the immediately preceding value.
Travel distance for each mouse
The Euclidean distances of mouse movement were calculated from consecutive RFID positions at time t and t+1. This was pivotal in understanding the spatial dynamics of the subjects. Graphical representations of tracked movements, including 2D positional plots and time series plots of X and Y coordinates, were generated to visualize trajectories and movement patterns over time.
Relative distance between two mice
The distances between different pairs of mice were calculated from the Euclidean distance between RFID positions. The total travel distance for each mouse was also calculated and plotted over the experimental period, providing a quantitative measure of activity levels.
Spatial interaction analysis
Spatial relationships of paired mice on the XY plane of the IntelliProfiler were classified, based on the distance L between the paired mice, into four categories (Fig. 3a). “Same” indicates that two mice were in the same grid (L=0 (cm)), suggesting immediate proximity. “Close” represents two mice were in adjacent grids (L=5 or 5√2 (cm)), demonstrating close but distinct positions. “Intermediate” (L=10, 5√5 or 10√2 (cm)) places one mouse on the outer perimeter of “Close,” while “Away” (L>10√2 (cm)) signifies that one mouse was outside of the range of “Intermediate,” indicating highly separation.
Close contact: defining social proximity
To examine social interactions within a group of mice, we defined the Close Contact Ratio (CCR) as the cumulative percentage of time a pair of mice spent in the same grid (defined as “Same”) or adjacent grids (defined as “Close”). This measure serves as a quantitative score of sociality between two mice, indicating their tendency to remain in immediate or proximity. Utilizing Excel and Prism 9 (GraphPad Software), we calculated the proportionate shifts of social distances (“Same,” “Close,” “Intermediate,” and “Away”) and CCR every 2 hours and in light and dark periods.
Network Analysis
Using Cytoscape 3.9.132, we conducted network analyses for four-, eight-, and 15/16-mouse groups. We imported values ten times the CCR as edge betweenness into Cytoscape 3.9.1. The network layout was generated using the Edge-weighted Spring Embedded Layout, facilitating the visual interpretation of social interactions among the mice based on their proximity metrics32,33.
Principal component analysis
Principal component analysis (PCA) was conducted using Python 3 (Version 3.9.7) on four datasets derived from long-term behavior mice in large groups under light and dark periods, focusing on relative activity levels and CCR. Three distinct types of PCA plots were generated: 1) PCA contribution vectors displayed on the PCA plot; 2) K-means clustering analysis, with the optimal number of clusters determined using the elbow method24 and silhouette method25 applied to the PCA plot; and 3) PCA plots color-coded by gender and genotype to distinguish different groups.
Figure visualization
Graphical representations were primarily created using Prism 9 (GraphPad Software). Time series line graphs and heat maps of travel distances or cumulative CCRs as well as transitions in the proportionate of social distances (“Same,” “Close,” “Intermediate,” and “Away”) were visualized. Pie charts illustrating the distribution across social distances were also developed. The PCA plots were designed in Python 3, and the social network graphs for the groups were constructed using Cytoscape version 3.9.1.
Statistical analysis
Statistical analyses were performed with Prism version 9 (GraphPad). Two-way ANOVA followed by Holm-Sidak’s multiple comparisons test was applied to assess differences between the two groups, which has been presented in part in a previous article34. Values in the bar or line graphs were presented as mean ± SD, and p* < 0.05, p**<0.01, p***<0.001 indicated statistically significant differences. The p-values, F values and degrees of freedom for all results, are provided in Supplementary Table 2. All experiments were performed in a single repetition.
Data availability
The behavioral data generated in this study are provided in Supplementary Information.
Code availability
All code and scripts used for data analysis in this study are available on GitHub. To access the public repository, please use the following access token along with the repository URL: – Repository URL: https://github.com/ShoheiOchi/IntelliProfiler
Contributions
Conceptualization, S.O., H.I. and N.O.; behavior analysis, S.O.; data process, S.O. and H.I.; code preparation, S.O. and H.I.; figure preparation, S.O.; writing—review and editing, S.O., H.I. and N.O.; supervision, N.O.; project administration, S.O. and N.O.; funding acquisition, S.O., and N.O. All authors have read and agreed to the published version of the manuscript.
Competing interests
The authors declare that this study was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Acknowledgements
The authors are grateful to K. Tsutsui for temporally providing the behavioral analysis environment, and K. Abe for advising on the setup of the behavioral analysis room, and T. Endo for technical advice on constructing the antenaboard hardware. This research was funded by the JSPS KAKENHI (Grant-in-Aid for Transformative Research Areas) under grant number 24H01419 (to S.O.), AMED under grant number JP21wm0425003 (to N.O.). Takeda Science Foundation (to S.O.) and FY2024 Tohoku University Graduate School of Medicine Grant-in-Aid for Joint Research by Young Researchers (to S.O.).
References
