- Email: [email protected]
A system utilizing radio frequency identification (RFID) technology to monitor individual rodent behavior in complex social settings
Journal of Neuroscience Methods 209 (2012) 74–78
Contents lists available at SciVerse ScienceDirect
Journal of Neuroscience Methods journal homepage: www.elsevier.com/locate/jneumeth
Basic Neuroscience Short communication
A system utilizing radio frequency identification (RFID) technology to monitor individual rodent behavior in complex social settings Christopher L. Howerton a,∗ , Joseph P. Garner b , Joy A. Mench a a b
Animal Science Department, University of California, Davis, United States Department of Comparative Medicine, Stanford University, United States
a r t i c l e
i n f o
Article history: Received 22 November 2011 Received in revised form 30 May 2012 Accepted 1 June 2012 Keywords: RFID Social living Individual behavior Rodent behavior
a b s t r a c t Pre-clinical investigation of human CNS disorders relies heavily on mouse models. However these show low predictive validity for translational success to humans, partly due to the extensive use of rapid, high-throughput behavioral assays. Improved assays to monitor rodent behavior over longer time scales in a variety of contexts while still maintaining the efficiency of data collection associated with highthroughput assays are needed. We developed an apparatus that uses radio frequency identification device (RFID) technology to facilitate long-term automated monitoring of the behavior of mice in socially or structurally complex cage environments. Mice that were individually marked and implanted with transponders were placed in pairs in the apparatus, and their locations continuously tracked for 24 h. Video observation was used to validate the RFID readings. The apparatus and its associated software accurately tracked the locations of all mice, yielding information about each mouse’s location over time, its diel activity patterns, and the amount of time it was in the same location as the other mouse in the pair. The information that can be efficiently collected in this apparatus has a variety of applications for preclinical research on human CNS disorders, for example major depressive disorder and autism spectrum disorder, in that it can be used to quantify validated endophenotypes or biomarkers of these disorders using rodent models. While the specific configuration of the apparatus described here was designed to answer particular experimental questions, it can be modified in various ways to accommodate different experimental designs. © 2012 Elsevier B.V. All rights reserved.
1. Introduction Monitoring laboratory mouse behavior is increasingly important for biomedical research, especially for translational science addressing human developmental and central nervous system disorders. Established rodent models for these disorders typically show very low predictive validity – on average only 8% of results successfully translate to humans (Kola and Landis, 2004). Multiple factors play a role in poor predictive validity, including the use of high-throughput behavioral phenotyping. Several authors have criticized high-throughput procedures, citing the limited construct validity of interpreting complex behavioral data, such as social interactions, in the context of these assays (Markou et al., 2009). In particular, these assays primarily measure transient behavioral characteristics, which are greatly affected by environmental conditions (Nestler and Hyman, 2010). For example, environmental variables (e.g. season/humidity, time of day, cage density) have twice as much effect on tail flick measures of pain sensitivity as
∗ Corresponding author. E-mail address: [email protected] (C.L. Howerton). 0165-0270/$ – see front matter © 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jneumeth.2012.06.001
genotype (Chesler et al., 2002). Such influences both reduce power and produce false positive results, leading to poor construct and predictive validity (Richter et al., 2010). There is a growing consensus that the predictive validity of rodent models could be improved by assessing more permanent and stable behavioral characteristics and by monitoring behavioral responses during the normal active period (at night), over a longer time scale (i.e. hours or days), and in a wider range of contexts (Editorial, 2011; Nestler and Hyman, 2010). Ethologically based approaches to collecting data with high predictive validity would involve monitoring the behavior of the mice in their home cage in their familiar social group, or in another socially and/or environmentally complex setting where they are able to perform a wider range of behaviors. However, these types of detailed assessments of complex behavioral characteristics require collecting data (usually from video) for several hours per individual or social group, generating hundreds of hours of video that must be coded for even a small behavioral study. An additional complication is accurately identifying each individual in a social group in order to increase statistical power, which is a particular problem during the dark phase of the light cycle because mice can appear identical under the infrared lighting used for videotaping.
C.L. Howerton et al. / Journal of Neuroscience Methods 209 (2012) 74–78
We developed a method for using radio frequency identification (RFID) to improve efficiency in collecting behavioral data from rodent models. RFID technology is widely used commercially for asset tracking, but has rarely been used to evaluate rodent behavior (see Winter and Schaefers, 2011). It has two components: the tag, which has a unique character string that identifies the item to which it is fixed, and the reader, which recognizes the tag and its associated string. Passive RFID tags have no internal power source and thus require excitation from an external magnetic field to transmit their signal, but because of this tend to be small enough to be implanted in animals. There are technological constraints to the use of these tags for tracking rodent movements. Multiple passive tags cannot be read concurrently by a reader and the magnetic fields emitted by RFID readers can interfere with an adjacent reader’s ability to recognize tags. We developed an apparatus that controlled these factors to use passive RFID technology to track individual rodent movements by providing location information for each animal over time. As we illustrate, this location information can then be paired with other information, such as video data, to greatly improve the efficiency of recording behavior. 2. Materials and methods 2.1. Behavioral Testing Apparatus The Behavioral Testing Apparatus (BTA; Fig. 1a) was comprised of polycarbonate mouse cages (Dura Cage PN RC71D-PC, Alternative Design Manufacturing & Supply Inc., Siloam Springs, AR, USA) connected by acrylic tubing (3.81 cm outside diameter, 19 cm in length). There was a central cage connected to four satellite cages, and an RFID reader (PhidgetRFID PN 1023, Phidgets Inc., Calgary, Alberta, CA) under each connecting tube. The reader was encased in a modified steel box (10.16 cm Square Box PN 8189, RACO, South Bend, IN, USA) to dampen the magnetic field emitted by each reader (Fig. 1b). The arrangement of the readers underneath the tubing also ensured that only one tag (implanted subcutaneously in each mouse, see Section 2.2) was within the read range of each reader at a time. Scaffolding constructed of 1.27 cm PVC and fittings measuring approximately 37 cm × 107 cm × 78 cm (H, W, and D) was arranged above each cage. A digital camera (15-CJ31, COP Security, Doetinchem, NL) was suspended from the scaffolding above each cage to capture video specific to that cage. To maximize video quality during the dark phase of the light cycle, the on-board infrared (IR) illuminator for each camera was disabled, and 2 supplementary IR illuminators (IR-300, Speco Technologies, Amityville, NY, USA) were arranged above the scaffolding. To maximize visualization, each cage was topped by custommade wire mesh covers (0.635 cm 23 gauge hardware cloth). Each of the satellite cages was provisioned with ad libitum feed (Purina Mills Mouse Chow 5010, St. Louis, MO, USA) and water. Cages were bedded with a mixture of the recycled paper bedding products Carefresh® (International Absorbents, Ferndale, WA, USA) and Paperchip® (Shepherd Specialty Papers, Kalamazoo, MI, USA) in a 2:1 ratio. The entire apparatus was affixed to a sheet of white opaque polycarbonate approximately 120 cm × 90 cm × 0.64 cm.
75
C57/BL6NCrl; Crl:CD1) to validate the BTA. The mice were bred and maintained in a conventional laboratory animal facility, from breeding stock originally obtained from Charles River Laboratories (Wilmington, MA, USA). They were housed in same sex and strain dyads in the same type of polycarbonate mouse cages used in the BTA. In these home cages they were provided with water, feed and bedding as described above. Each mouse was individually identified with both a unique RFID tag and conspicuous visual markings. We individually identified the mice by anesthetizing them with a mixture of ketamine and xylazine at 60:10 mg/kg and then subcutaneously implanting an RFID tag (Small Glass Ampoule Tag, Trossen Robotics, Downers Grove, IL, USA) by making a small incision (3–4 mm) on the ventral surface. The orientation of the tag was lengthwise and parallel to the midline of the mouse, the orientation previously determined to optimize read range. Each mouse was marked on the shoulders or the rump with either a commercially available hair lightening product (Clairol Professional© Basic White Extra Strength, Clairol Professional, Jacksonville, FL, USA, for the C57/BL6NCrl mice) or a black hair dye (Clairol Professional© Premium Crème Haircolor, 1N Black Neutral, Clairol Professional, Jacksonville, FL, USA for the Balb/cAnNCrl and the Crl:CD1 mice). All procedures were performed according to protocols approved by the Institutional Animal Care and Use Committee of the University of California, Davis. 2.3. Data collection software Video data were collected continuously from each camera and organized into 1-min video segments by the software (Geovision Digital Surveillance System v 8.2.0.0, USA Vision Systems Inc., Irvine, CA, USA). Each segment was automatically coded by date, time (24 h format) and recording camera. These codes were then used to identify video segments of interest based upon RFID data. The Geovision software can also be used to detect motion, and this feature was used to generate a list of motion segments, coded in the same manner as the video segments. RFID data were collected using software we developed based upon open-source programming that supports products manufactured by Phidgets Inc. (http://www.phidgets.com/programming resources.php). The software organizes tag readings to give the location of each mouse over time as being either (1) within the central cage or within the connecting tubing on the central side of the RFID reader or (2) within a specific satellite cage or the connecting tubing on the distal side of the RFID reader corresponding to that specific satellite cage. Location data are calculated by evaluating the order in which a mouse’s tag is read. The initial reading of a tag by any reader indicates that the mouse has entered a satellite cage and the next reading of that tag indicates that the mouse has left that cage, allowing the location of each mouse over time to be determined. Location data for each mouse were then used to generate a list of video segments in which that mouse was recorded. 2.4. BTA validation
2.2. Subject animals
Ten days after individual identification, one pair of mice was placed within the BTA for 24 h, and both RFID and video data were continuously recorded. The BTA was thoroughly cleaned after the data were collected, and was then supplied with fresh food, water and bedding. This process was repeated for the remaining 5 pairs of mice.
In order to represent the range of variation in size and activity (variables that could affect the efficacy of RFID readings), as well as coat color (a variable that could affect identification during video monitoring), found among common strains of laboratory mice, we used 2 pairs of mice from each of 3 strains (Balb/cAnNCrl;
2.4.1. Validation of RFID location In order to verify that the locations determined by RFID data were accurate and consistent, these data were compared to location of each mouse determined by video observations. Location was recorded using both methods at the beginning of every minute
76
C.L. Howerton et al. / Journal of Neuroscience Methods 209 (2012) 74–78
Fig. 1. (a) Behavioral Testing Apparatuses (BTAs). Each BTA consisted of polycarbonate cages with RFID readers under the connecting tubing and digital video cameras above each satellite cage. (b) An RFID reader, which was encased in a steel box to dampen emitted magnetic fields to avoid interference. (c and d) RFID readings and minutes of activity. The number of RFID readings recorded for each mouse is plotted against the number of minutes of activity determined from video observations. Data are presented using real numbers (c) and a square-root transformation and back-transformed axes labels (d). The predicted outcome is plotted (solid line) ± the 95% CI for the estimate (dashed line).
(n = 1440) within the data collection period. For the video recordings, we recorded the location of each mouse in the first frame of each 1-min video segment. The video and RFID locations were then compared by calculating both the Pearson’s correlation coefficient between the two measures and the total number of mismatches between the location of a mouse according to the BTA software and the actual location as determined by video. The source of all discrepancies was the reader failing to read a tag when the mouse passed over it, meaning that the software placed that mouse in the center cage when it was actually in one of the satellite cages, or vice versa. 2.4.2. RFID readings and activity To determine the total minutes of activity for each mouse, a list of video segments in which a mouse was recorded was compiled and then coded for activity using one-zero sampling, with a mouse coded as active during that minute if it moved for more than 5 consecutive seconds. A sum of both the active minutes as determined by the Geovision motion detection software and RFID readings was then calculated for each hour for all mice. To confirm the accuracy of the Geovision motion detection software in estimating mouse activity, the data were compared to the minutes of activity determined by video observations. A generalized linear mixed model with varying intercepts for each mouse, using R 2.11.1 with both
the ‘arm’ and ‘bbmle’ packages, was used to test the hypothesis that the number of minutes of activity estimated with the Geovision motion detection software accurately predicted the number of RFID readings. Since RFID readings were unbounded counts, a Poisson distribution with a log-link function was used to fit the data. The data were modeled using an exhaustive list of the combinations possible with the explanatory variables pair, strain and minutes of activity. The best-fit model was then determined using a second order transformed Akaike information criterion (AICc), with the parameter estimates from this model then used for reporting and predictions. 95% Confidence intervals (95% CI) for each parameter estimate were then created by re-sampling the original dataset using functions in the ‘boot’ package. 2.4.3. Space use over time For each mouse, the number of minutes spent in each of the 5 sub-areas of the BTA (the center or satellite cages) as calculated from RFID data was summed, and then depicted visually (Fig. 2). 3. Results The BTA and its associated software reliably measured the location of each mouse over time, as indicated by the correlation between visual observations and location calculated from
C.L. Howerton et al. / Journal of Neuroscience Methods 209 (2012) 74–78
77
Fig. 2. Locations of each pair of mice over time. In each panel, the number of minutes each mouse in a particular pair (the white bar is one member of the pair and the gray bar the other) spent in each of the five sub-areas of the BTA (C = center and 1–4 = each satellite cage) is shown.
RFID readings (average Pearson’s correlation coefficient ± SD: 0.935 ± 0.02) and the number of discrepancies between the two measures (mean number of discrepancies ± SD: 15.3 ± 5.7). These location data can easily be plotted to determine preferences for a location, as well as intra-group variations in these preferences (Fig. 2). Analyses of activity data show that both the Geovision software and the BTA can be used as an automated measure of rodent activity over time. Video observations were perfectly predicted by the Geovision motion detection feature, with no discrepancies between the two measures. RFID readings were also highly predicted by activity; the best fit model (Fig. 1 c and d; estimated as: e(2.342+0.044×(Minutes of Activity)) included only minutes of activity as an explanatory variable, indicating that there was a strong exponential relationship between the minutes of activity and RFID
readings, and that the effects of strain and pair of mice on activity levels were minimal. 4. Discussion The BTA we developed allowed mouse behavior to be automatically monitored in a complex environment while maintaining the mice in social groups. We used four satellite cages to accommodate our particular experimental questions (CLH and JAM manuscripts in preparation), but the system can be customized in virtually limitless configurations. The BTA and its associated software translate the RFID data into location information, which can be used to: (1) generate a list of video segments in which each individual mouse appears; (2) calculate the amount of time that each individual spends within the different sub-areas of the BTA and
78
C.L. Howerton et al. / Journal of Neuroscience Methods 209 (2012) 74–78
variation between social groups in time spent in these sub-areas; and (3) monitor general activity. The BTA allows researchers to efficiently assess these responses in complex environmental and social settings, and over extended timeframes, in order to increase validity. The data automatically collected by the BTA have a variety of potential uses. Information about location over time could be used to determine conditioned place preferences associated with rewarding experiences such as the use of drugs of abuse in a social context (Thiel et al., 2009), environmental preferences of the mice (e.g. enrichments, thermal conditions) important to improve rodent welfare (CLH and JAM manuscript in preparation), and deviations from normal activity patterns. Individual rodents in social groups may show deviations in species-typical diel activity patterns that are similar to those seen in humans with major depressive disorder (Nestler and Hyman, 2010). These have been interpreted as a depression-like endophenotype (Blanchard et al., 1995) and thus could represent a robust measure of depression-like phenotypes in rodents. Finally, being able to generate a list of video segments in which particular mice recorded would allow manipulations, such as drug interventions, to be applied to individual animals rather than the cage of mice in order to increase statistical power and reduce the number of animals needed. The major advantage of our method is that it reduces the total amount of video that must be analyzed in order to quantify social interactions, which are important for the evaluation of models of many neurodevelopment diseases such as Autism spectrum and affective disorders (Laviola and Terranova, 1998). Using the BTA, only data from the video segments during which more than one mouse is present and motion is detected have to be coded, which in our studies was only about 25% of the time (CLH and JAM manuscript in preparation). RFID technology has previously been used with rodent models to facilitate the performance of operant tasks while maintaining social housing (e.g. Winter and Schaefers, 2011) in order to reduce known sources of variability related to handling and isolation (Crabbe et al., 1999), for example the development of abnormal behaviors like stereotypies in some individuals (Olsson and Westlund, 2007). In contrast, the method we describe allows individual behavioral characteristics to be measured in the context of a stable social group, allowing the evaluation of the effects of group characteristics (such as group size or dominance relationships) on individual behavioral traits (such as diel activity patterns or enrichment use). There are other methods available for automatic coding of rodent behavior. For example, photocell arrays have been developed to measure circadian activity patterns (van’t Land and Hendriksen, 1995). These provide extremely accurate information but do not allow data to be collected from socially housed mice.
Similarly, methods for using advanced video analysis to automatically code mouse behavior such as Ethovision® or Anymaze© excel at analyzing behavior of singly housed animals, but not socially housed animals. Conversely, the BTA can track individuals in a complex social environment, but cannot determine which behaviors the animals are performing. It is conceivable that these types of methods could be integrated by, for example, providing location over time information (from the BTA) to the video analysis software. Our study showed that RFID technology can be a useful tool for monitoring mouse behavior. Its utility ranges from increasing the efficiency of coding behavioral data from recorded video to determining mouse preferences for specific locations. Further integrating this technology with more complex video analysis software and other data collection devices would offer exciting possibilities for automation of collecting increasingly complex behavioral data for basic discoveries in translational science, and dramatically reducing the time required to analyze ethologically relevant behavioral data. References Blanchard DC, Spencer RL, Weiss SM, Blanchard RJ, McEwen B, Sakai RR. Visible burrow system as a model of chronic social stress—behavioral and neuroendocrine correlates. Psychoneuroendocrinology 1995;20:117–34. Chesler EJ, Wilson SG, Lariviere WR, Rodriguez-Zas SL, Mogil JS. Identification and ranking of genetic and laboratory environment factors influencing a behavioral trait, thermal nociception, via computational analysis of a large data archive. Neurosci Biobehav Rev 2002;26:907–23. Crabbe JC, Wahlsten D, Dudek BC. Genetics of mouse behavior: interactions with laboratory environment. Science 1999;284:1670–2. Editorial. Building a better mouse test. Nat Methods 2011;8:697–700. Kola I, Landis J. Can the pharmaceutical industry reduce attrition rates? Nat Rev Drug Discov 2004;3:711–5. Laviola G, Terranova ML. The developmental psychobiology of behavioural plasticity in mice: the role of social experiences in the family unit. Neurosci Biobehav Rev 1998;23:197–213. Markou A, Chiamulera C, Geyer MA, Tricklebank M, Steckler T. Removing obstacles in neuroscience drug discovery: the future path for animal models. Neuropsychopharmacology 2009;34:74–89. Nestler EJ, Hyman SE. Animal models of neuropsychiatric disorders. Nat Neurosci 2010;13:1161–9. Olsson IAS, Westlund K. More than numbers matter: the effect of social factors on behaviour and welfare of laboratory rodents and non-human primates. Appl Anim Behav Sci 2007;103:229–54. Richter SH, Garner JP, Auer C, Kunert J, Wurbel H. Systematic variation improves reproducibility of animal experiments. Nat Methods 2010;7:167–8. Thiel K, Sanabria F, Neisewander J. Synergistic interaction between nicotine and social rewards in adolescent male rats. Psychopharmacology (Berl) 2009;204:391–402. van’t Land CJ, Hendriksen CFM. Change in locomotor activity pattern in mice: a model for recognition of distress? Lab Anim 1995;29:286–93. Winter Y, Schaefers ATU. A sorting system with automated gates permits individual operant experiments with mice from a social home cage. J Neurosci Methods 2011;196:276–80.
Contents lists available at SciVerse ScienceDirect
Journal of Neuroscience Methods journal homepage: www.elsevier.com/locate/jneumeth
Basic Neuroscience Short communication
A system utilizing radio frequency identification (RFID) technology to monitor individual rodent behavior in complex social settings Christopher L. Howerton a,∗ , Joseph P. Garner b , Joy A. Mench a a b
Animal Science Department, University of California, Davis, United States Department of Comparative Medicine, Stanford University, United States
a r t i c l e
i n f o
Article history: Received 22 November 2011 Received in revised form 30 May 2012 Accepted 1 June 2012 Keywords: RFID Social living Individual behavior Rodent behavior
a b s t r a c t Pre-clinical investigation of human CNS disorders relies heavily on mouse models. However these show low predictive validity for translational success to humans, partly due to the extensive use of rapid, high-throughput behavioral assays. Improved assays to monitor rodent behavior over longer time scales in a variety of contexts while still maintaining the efficiency of data collection associated with highthroughput assays are needed. We developed an apparatus that uses radio frequency identification device (RFID) technology to facilitate long-term automated monitoring of the behavior of mice in socially or structurally complex cage environments. Mice that were individually marked and implanted with transponders were placed in pairs in the apparatus, and their locations continuously tracked for 24 h. Video observation was used to validate the RFID readings. The apparatus and its associated software accurately tracked the locations of all mice, yielding information about each mouse’s location over time, its diel activity patterns, and the amount of time it was in the same location as the other mouse in the pair. The information that can be efficiently collected in this apparatus has a variety of applications for preclinical research on human CNS disorders, for example major depressive disorder and autism spectrum disorder, in that it can be used to quantify validated endophenotypes or biomarkers of these disorders using rodent models. While the specific configuration of the apparatus described here was designed to answer particular experimental questions, it can be modified in various ways to accommodate different experimental designs. © 2012 Elsevier B.V. All rights reserved.
1. Introduction Monitoring laboratory mouse behavior is increasingly important for biomedical research, especially for translational science addressing human developmental and central nervous system disorders. Established rodent models for these disorders typically show very low predictive validity – on average only 8% of results successfully translate to humans (Kola and Landis, 2004). Multiple factors play a role in poor predictive validity, including the use of high-throughput behavioral phenotyping. Several authors have criticized high-throughput procedures, citing the limited construct validity of interpreting complex behavioral data, such as social interactions, in the context of these assays (Markou et al., 2009). In particular, these assays primarily measure transient behavioral characteristics, which are greatly affected by environmental conditions (Nestler and Hyman, 2010). For example, environmental variables (e.g. season/humidity, time of day, cage density) have twice as much effect on tail flick measures of pain sensitivity as
∗ Corresponding author. E-mail address: [email protected] (C.L. Howerton). 0165-0270/$ – see front matter © 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jneumeth.2012.06.001
genotype (Chesler et al., 2002). Such influences both reduce power and produce false positive results, leading to poor construct and predictive validity (Richter et al., 2010). There is a growing consensus that the predictive validity of rodent models could be improved by assessing more permanent and stable behavioral characteristics and by monitoring behavioral responses during the normal active period (at night), over a longer time scale (i.e. hours or days), and in a wider range of contexts (Editorial, 2011; Nestler and Hyman, 2010). Ethologically based approaches to collecting data with high predictive validity would involve monitoring the behavior of the mice in their home cage in their familiar social group, or in another socially and/or environmentally complex setting where they are able to perform a wider range of behaviors. However, these types of detailed assessments of complex behavioral characteristics require collecting data (usually from video) for several hours per individual or social group, generating hundreds of hours of video that must be coded for even a small behavioral study. An additional complication is accurately identifying each individual in a social group in order to increase statistical power, which is a particular problem during the dark phase of the light cycle because mice can appear identical under the infrared lighting used for videotaping.
C.L. Howerton et al. / Journal of Neuroscience Methods 209 (2012) 74–78
We developed a method for using radio frequency identification (RFID) to improve efficiency in collecting behavioral data from rodent models. RFID technology is widely used commercially for asset tracking, but has rarely been used to evaluate rodent behavior (see Winter and Schaefers, 2011). It has two components: the tag, which has a unique character string that identifies the item to which it is fixed, and the reader, which recognizes the tag and its associated string. Passive RFID tags have no internal power source and thus require excitation from an external magnetic field to transmit their signal, but because of this tend to be small enough to be implanted in animals. There are technological constraints to the use of these tags for tracking rodent movements. Multiple passive tags cannot be read concurrently by a reader and the magnetic fields emitted by RFID readers can interfere with an adjacent reader’s ability to recognize tags. We developed an apparatus that controlled these factors to use passive RFID technology to track individual rodent movements by providing location information for each animal over time. As we illustrate, this location information can then be paired with other information, such as video data, to greatly improve the efficiency of recording behavior. 2. Materials and methods 2.1. Behavioral Testing Apparatus The Behavioral Testing Apparatus (BTA; Fig. 1a) was comprised of polycarbonate mouse cages (Dura Cage PN RC71D-PC, Alternative Design Manufacturing & Supply Inc., Siloam Springs, AR, USA) connected by acrylic tubing (3.81 cm outside diameter, 19 cm in length). There was a central cage connected to four satellite cages, and an RFID reader (PhidgetRFID PN 1023, Phidgets Inc., Calgary, Alberta, CA) under each connecting tube. The reader was encased in a modified steel box (10.16 cm Square Box PN 8189, RACO, South Bend, IN, USA) to dampen the magnetic field emitted by each reader (Fig. 1b). The arrangement of the readers underneath the tubing also ensured that only one tag (implanted subcutaneously in each mouse, see Section 2.2) was within the read range of each reader at a time. Scaffolding constructed of 1.27 cm PVC and fittings measuring approximately 37 cm × 107 cm × 78 cm (H, W, and D) was arranged above each cage. A digital camera (15-CJ31, COP Security, Doetinchem, NL) was suspended from the scaffolding above each cage to capture video specific to that cage. To maximize video quality during the dark phase of the light cycle, the on-board infrared (IR) illuminator for each camera was disabled, and 2 supplementary IR illuminators (IR-300, Speco Technologies, Amityville, NY, USA) were arranged above the scaffolding. To maximize visualization, each cage was topped by custommade wire mesh covers (0.635 cm 23 gauge hardware cloth). Each of the satellite cages was provisioned with ad libitum feed (Purina Mills Mouse Chow 5010, St. Louis, MO, USA) and water. Cages were bedded with a mixture of the recycled paper bedding products Carefresh® (International Absorbents, Ferndale, WA, USA) and Paperchip® (Shepherd Specialty Papers, Kalamazoo, MI, USA) in a 2:1 ratio. The entire apparatus was affixed to a sheet of white opaque polycarbonate approximately 120 cm × 90 cm × 0.64 cm.
75
C57/BL6NCrl; Crl:CD1) to validate the BTA. The mice were bred and maintained in a conventional laboratory animal facility, from breeding stock originally obtained from Charles River Laboratories (Wilmington, MA, USA). They were housed in same sex and strain dyads in the same type of polycarbonate mouse cages used in the BTA. In these home cages they were provided with water, feed and bedding as described above. Each mouse was individually identified with both a unique RFID tag and conspicuous visual markings. We individually identified the mice by anesthetizing them with a mixture of ketamine and xylazine at 60:10 mg/kg and then subcutaneously implanting an RFID tag (Small Glass Ampoule Tag, Trossen Robotics, Downers Grove, IL, USA) by making a small incision (3–4 mm) on the ventral surface. The orientation of the tag was lengthwise and parallel to the midline of the mouse, the orientation previously determined to optimize read range. Each mouse was marked on the shoulders or the rump with either a commercially available hair lightening product (Clairol Professional© Basic White Extra Strength, Clairol Professional, Jacksonville, FL, USA, for the C57/BL6NCrl mice) or a black hair dye (Clairol Professional© Premium Crème Haircolor, 1N Black Neutral, Clairol Professional, Jacksonville, FL, USA for the Balb/cAnNCrl and the Crl:CD1 mice). All procedures were performed according to protocols approved by the Institutional Animal Care and Use Committee of the University of California, Davis. 2.3. Data collection software Video data were collected continuously from each camera and organized into 1-min video segments by the software (Geovision Digital Surveillance System v 8.2.0.0, USA Vision Systems Inc., Irvine, CA, USA). Each segment was automatically coded by date, time (24 h format) and recording camera. These codes were then used to identify video segments of interest based upon RFID data. The Geovision software can also be used to detect motion, and this feature was used to generate a list of motion segments, coded in the same manner as the video segments. RFID data were collected using software we developed based upon open-source programming that supports products manufactured by Phidgets Inc. (http://www.phidgets.com/programming resources.php). The software organizes tag readings to give the location of each mouse over time as being either (1) within the central cage or within the connecting tubing on the central side of the RFID reader or (2) within a specific satellite cage or the connecting tubing on the distal side of the RFID reader corresponding to that specific satellite cage. Location data are calculated by evaluating the order in which a mouse’s tag is read. The initial reading of a tag by any reader indicates that the mouse has entered a satellite cage and the next reading of that tag indicates that the mouse has left that cage, allowing the location of each mouse over time to be determined. Location data for each mouse were then used to generate a list of video segments in which that mouse was recorded. 2.4. BTA validation
2.2. Subject animals
Ten days after individual identification, one pair of mice was placed within the BTA for 24 h, and both RFID and video data were continuously recorded. The BTA was thoroughly cleaned after the data were collected, and was then supplied with fresh food, water and bedding. This process was repeated for the remaining 5 pairs of mice.
In order to represent the range of variation in size and activity (variables that could affect the efficacy of RFID readings), as well as coat color (a variable that could affect identification during video monitoring), found among common strains of laboratory mice, we used 2 pairs of mice from each of 3 strains (Balb/cAnNCrl;
2.4.1. Validation of RFID location In order to verify that the locations determined by RFID data were accurate and consistent, these data were compared to location of each mouse determined by video observations. Location was recorded using both methods at the beginning of every minute
76
C.L. Howerton et al. / Journal of Neuroscience Methods 209 (2012) 74–78
Fig. 1. (a) Behavioral Testing Apparatuses (BTAs). Each BTA consisted of polycarbonate cages with RFID readers under the connecting tubing and digital video cameras above each satellite cage. (b) An RFID reader, which was encased in a steel box to dampen emitted magnetic fields to avoid interference. (c and d) RFID readings and minutes of activity. The number of RFID readings recorded for each mouse is plotted against the number of minutes of activity determined from video observations. Data are presented using real numbers (c) and a square-root transformation and back-transformed axes labels (d). The predicted outcome is plotted (solid line) ± the 95% CI for the estimate (dashed line).
(n = 1440) within the data collection period. For the video recordings, we recorded the location of each mouse in the first frame of each 1-min video segment. The video and RFID locations were then compared by calculating both the Pearson’s correlation coefficient between the two measures and the total number of mismatches between the location of a mouse according to the BTA software and the actual location as determined by video. The source of all discrepancies was the reader failing to read a tag when the mouse passed over it, meaning that the software placed that mouse in the center cage when it was actually in one of the satellite cages, or vice versa. 2.4.2. RFID readings and activity To determine the total minutes of activity for each mouse, a list of video segments in which a mouse was recorded was compiled and then coded for activity using one-zero sampling, with a mouse coded as active during that minute if it moved for more than 5 consecutive seconds. A sum of both the active minutes as determined by the Geovision motion detection software and RFID readings was then calculated for each hour for all mice. To confirm the accuracy of the Geovision motion detection software in estimating mouse activity, the data were compared to the minutes of activity determined by video observations. A generalized linear mixed model with varying intercepts for each mouse, using R 2.11.1 with both
the ‘arm’ and ‘bbmle’ packages, was used to test the hypothesis that the number of minutes of activity estimated with the Geovision motion detection software accurately predicted the number of RFID readings. Since RFID readings were unbounded counts, a Poisson distribution with a log-link function was used to fit the data. The data were modeled using an exhaustive list of the combinations possible with the explanatory variables pair, strain and minutes of activity. The best-fit model was then determined using a second order transformed Akaike information criterion (AICc), with the parameter estimates from this model then used for reporting and predictions. 95% Confidence intervals (95% CI) for each parameter estimate were then created by re-sampling the original dataset using functions in the ‘boot’ package. 2.4.3. Space use over time For each mouse, the number of minutes spent in each of the 5 sub-areas of the BTA (the center or satellite cages) as calculated from RFID data was summed, and then depicted visually (Fig. 2). 3. Results The BTA and its associated software reliably measured the location of each mouse over time, as indicated by the correlation between visual observations and location calculated from
C.L. Howerton et al. / Journal of Neuroscience Methods 209 (2012) 74–78
77
Fig. 2. Locations of each pair of mice over time. In each panel, the number of minutes each mouse in a particular pair (the white bar is one member of the pair and the gray bar the other) spent in each of the five sub-areas of the BTA (C = center and 1–4 = each satellite cage) is shown.
RFID readings (average Pearson’s correlation coefficient ± SD: 0.935 ± 0.02) and the number of discrepancies between the two measures (mean number of discrepancies ± SD: 15.3 ± 5.7). These location data can easily be plotted to determine preferences for a location, as well as intra-group variations in these preferences (Fig. 2). Analyses of activity data show that both the Geovision software and the BTA can be used as an automated measure of rodent activity over time. Video observations were perfectly predicted by the Geovision motion detection feature, with no discrepancies between the two measures. RFID readings were also highly predicted by activity; the best fit model (Fig. 1 c and d; estimated as: e(2.342+0.044×(Minutes of Activity)) included only minutes of activity as an explanatory variable, indicating that there was a strong exponential relationship between the minutes of activity and RFID
readings, and that the effects of strain and pair of mice on activity levels were minimal. 4. Discussion The BTA we developed allowed mouse behavior to be automatically monitored in a complex environment while maintaining the mice in social groups. We used four satellite cages to accommodate our particular experimental questions (CLH and JAM manuscripts in preparation), but the system can be customized in virtually limitless configurations. The BTA and its associated software translate the RFID data into location information, which can be used to: (1) generate a list of video segments in which each individual mouse appears; (2) calculate the amount of time that each individual spends within the different sub-areas of the BTA and
78
C.L. Howerton et al. / Journal of Neuroscience Methods 209 (2012) 74–78
variation between social groups in time spent in these sub-areas; and (3) monitor general activity. The BTA allows researchers to efficiently assess these responses in complex environmental and social settings, and over extended timeframes, in order to increase validity. The data automatically collected by the BTA have a variety of potential uses. Information about location over time could be used to determine conditioned place preferences associated with rewarding experiences such as the use of drugs of abuse in a social context (Thiel et al., 2009), environmental preferences of the mice (e.g. enrichments, thermal conditions) important to improve rodent welfare (CLH and JAM manuscript in preparation), and deviations from normal activity patterns. Individual rodents in social groups may show deviations in species-typical diel activity patterns that are similar to those seen in humans with major depressive disorder (Nestler and Hyman, 2010). These have been interpreted as a depression-like endophenotype (Blanchard et al., 1995) and thus could represent a robust measure of depression-like phenotypes in rodents. Finally, being able to generate a list of video segments in which particular mice recorded would allow manipulations, such as drug interventions, to be applied to individual animals rather than the cage of mice in order to increase statistical power and reduce the number of animals needed. The major advantage of our method is that it reduces the total amount of video that must be analyzed in order to quantify social interactions, which are important for the evaluation of models of many neurodevelopment diseases such as Autism spectrum and affective disorders (Laviola and Terranova, 1998). Using the BTA, only data from the video segments during which more than one mouse is present and motion is detected have to be coded, which in our studies was only about 25% of the time (CLH and JAM manuscript in preparation). RFID technology has previously been used with rodent models to facilitate the performance of operant tasks while maintaining social housing (e.g. Winter and Schaefers, 2011) in order to reduce known sources of variability related to handling and isolation (Crabbe et al., 1999), for example the development of abnormal behaviors like stereotypies in some individuals (Olsson and Westlund, 2007). In contrast, the method we describe allows individual behavioral characteristics to be measured in the context of a stable social group, allowing the evaluation of the effects of group characteristics (such as group size or dominance relationships) on individual behavioral traits (such as diel activity patterns or enrichment use). There are other methods available for automatic coding of rodent behavior. For example, photocell arrays have been developed to measure circadian activity patterns (van’t Land and Hendriksen, 1995). These provide extremely accurate information but do not allow data to be collected from socially housed mice.
Similarly, methods for using advanced video analysis to automatically code mouse behavior such as Ethovision® or Anymaze© excel at analyzing behavior of singly housed animals, but not socially housed animals. Conversely, the BTA can track individuals in a complex social environment, but cannot determine which behaviors the animals are performing. It is conceivable that these types of methods could be integrated by, for example, providing location over time information (from the BTA) to the video analysis software. Our study showed that RFID technology can be a useful tool for monitoring mouse behavior. Its utility ranges from increasing the efficiency of coding behavioral data from recorded video to determining mouse preferences for specific locations. Further integrating this technology with more complex video analysis software and other data collection devices would offer exciting possibilities for automation of collecting increasingly complex behavioral data for basic discoveries in translational science, and dramatically reducing the time required to analyze ethologically relevant behavioral data. References Blanchard DC, Spencer RL, Weiss SM, Blanchard RJ, McEwen B, Sakai RR. Visible burrow system as a model of chronic social stress—behavioral and neuroendocrine correlates. Psychoneuroendocrinology 1995;20:117–34. Chesler EJ, Wilson SG, Lariviere WR, Rodriguez-Zas SL, Mogil JS. Identification and ranking of genetic and laboratory environment factors influencing a behavioral trait, thermal nociception, via computational analysis of a large data archive. Neurosci Biobehav Rev 2002;26:907–23. Crabbe JC, Wahlsten D, Dudek BC. Genetics of mouse behavior: interactions with laboratory environment. Science 1999;284:1670–2. Editorial. Building a better mouse test. Nat Methods 2011;8:697–700. Kola I, Landis J. Can the pharmaceutical industry reduce attrition rates? Nat Rev Drug Discov 2004;3:711–5. Laviola G, Terranova ML. The developmental psychobiology of behavioural plasticity in mice: the role of social experiences in the family unit. Neurosci Biobehav Rev 1998;23:197–213. Markou A, Chiamulera C, Geyer MA, Tricklebank M, Steckler T. Removing obstacles in neuroscience drug discovery: the future path for animal models. Neuropsychopharmacology 2009;34:74–89. Nestler EJ, Hyman SE. Animal models of neuropsychiatric disorders. Nat Neurosci 2010;13:1161–9. Olsson IAS, Westlund K. More than numbers matter: the effect of social factors on behaviour and welfare of laboratory rodents and non-human primates. Appl Anim Behav Sci 2007;103:229–54. Richter SH, Garner JP, Auer C, Kunert J, Wurbel H. Systematic variation improves reproducibility of animal experiments. Nat Methods 2010;7:167–8. Thiel K, Sanabria F, Neisewander J. Synergistic interaction between nicotine and social rewards in adolescent male rats. Psychopharmacology (Berl) 2009;204:391–402. van’t Land CJ, Hendriksen CFM. Change in locomotor activity pattern in mice: a model for recognition of distress? Lab Anim 1995;29:286–93. Winter Y, Schaefers ATU. A sorting system with automated gates permits individual operant experiments with mice from a social home cage. J Neurosci Methods 2011;196:276–80.