Physiology& Behavior,Vol. 50, pp. 403-408. ©PergamonPress plc, 1991.Printedin the U.S.A.
0031-9384/91 $3.00 + .00
A System for Automated Recording and Analysis of Feeding Behavior M A R T I N G. HULSEY AND ROY J. MARTIN
Department of Foods and Nutrition, College of Family and Consumer Sciences The University of Georgia, Athens, GA 30602 Received 21 February 1991 HUI_~EY, M. G. AND R. J. MARTIN. A systemfor automated recording and analysis of feeding behavior. PHYSIOL BEHAV 50(2) 403--408, 1991.--We have designed and implemented a system that utilizes a network of top-loading balances digitally interfaced to a Macintosh computer. The system simultaneouslycollects two forms of data which allow the evaluation of the animal's biting and/or licking behavior in addition to cumulativefood intake and meal patterns. The system is capable of resuming data acquisition following a power failure without user intervention.Plexiglas cages utilized with the system features adjustable tunnel feeders and are appropriate for use with small rodents. Givenappropriate caging, the system may be utilized to evaluate the feeding behavior of other species. Food intake
Ingestion
Computer
Microcomputer
THE assessment of feeding behavior is prerequisite for studying physiological controls of feeding or evaluating the efficacy and specificity of agents that stimulate or inhibit feeding, Due to the prohibitive cost of apparatus for automated acquisition and processing of such data, many studies are performed without its benefit. Short-term measurements of food intake with deprived animals are common, but this paradigm has been criticized (9,14). Intermittent manual weighing of feed containers with freely feeding rats is another common method, but it has been noted that manipulation of feed containers per se can affect feeding behavior (7). Numerous approaches have been employed to capture feeding data. Operant paradigms (1,19) impose constraints which alter feeding as compared to the free-feeding state (14), but these constraints can be used to advantage when investigating feeding in an ecological context (5). An automated recording device was developed whereby the rat's removal of a pellet triggers the release of another (12). This approach was later refined by adding automated analysis of pellet consumption (11,24). Lickometer circuitry (21) has been employed to examine the microstructure of consumption of liquid diets (6), and has been adapted to a modified food container or "eatometer" to record a rat's contact with the feed (8,22). It is noteworthy that drinkometers and eatometers provide superior temporal resolution, but only indirect measurements of the amounts of food consumed. Another indirect approach is to measure the animal's entry into a feeding tunnel (19). This method has also been utilized to measure diet selection behavior (16). Continuously monitoring the weight of conventional feed containers with strain gauges permitted utilization of diets of a consistency that precludes its extrusion into pellets (10,17). A recently reported system (15) utilized a similar approach and featured photocells to simultaneously record a rodent's noningestive approaches to the feed container. Systems that utilized a
Meal patterns
network of electronic digital balances with a serial communications interface provided automated analysis of meal pattern data on a computer (3,4). In developing this system, our engineering goals were: 1) temporal resolution equivalent to a lickometer or pellet dispenser; 2) mass resolution equivalent to a load cell; 3) automated processing of meal pattern and cumulative food intake data; and 4) utilization of diets of widely varying composition. We have designed and implemented a system that fulfills these criteria. METHOD
Cage Design Plexiglas cages utilized with the system are appropriate for use with small rodents. The cage modules consist of twin Plexiglas enclosures measuring 28 cm in width, 27 cm in depth, and 38 cm in height. Each enclosure has a removable floor of 1-cm mesh stainless steel construction, and removable lids fabricated of the same material are secured with stainless steel pins. Each enclosure has an adjustable tunnel feeder. The entrance to the tunnel is elevated 5 cm above the floor of the enclosure so that rodents may enter from the forward direction only. The bottom of the tunnel is fabricated of 1-cm stainless steel mesh. An orifice is provided in the bottom of the tunnels through which the animals can feed. Stainless steel rings may be substituted to provide an opening of 2.5 to 5.0 cm in 0.6-cm increments. The sides of the tunnels are fabricated of 6 mm Plexiglas and are adjustable in width over a range of 5 to 10 cm at the rear and 2.3 to 6.5 cm at the front. The top of the tunnel is fabricated of 24-gauge stainless steel with a height adjustable from 3 to 16.5 cm. These dimensions were selected to accommodate rodents of various sizes, from weanling rats to mature obese fedfa Zucker rats. At its extreme height adjustment, the top of the tunnel is intended to clear chronic infusion leads.
403
404
HULSEY AND MARTIN
Sampleallbalancestwice
Hardware Glass feed containers (Allentown Caging, Inc.) rest upon Mettler PM4000 top-loading digital balances. Polystyrene containers 15 cm in diameter contain spillage within the balance pan. Balances rest atop scissor-type laboratory jacks (Fisher Scientific) so that the height of the feeder can be brought in proximity to but not contacting the orifice in the bottom of the feeding tunnel. The balances are networked to an eight-channel code-operated switch (COS-8, Black Box Corp.) using a threewire (pins 2, 3 and 7) RS-232 protocol at 9600 baud. The master port of the switch is connected through a shielded threeconductor cable to the RS-422 serial port of a Macintosh SE microcomputer located in an adjacent room. The system currently consists of eight channels, but is expandable to 64 or more channels by chaining multiple switches.
7/
J
yes
ISelo~ burstchantm, I
I
I recordbursttodisk J
L. yes
~
no
Software Programs for data acquisition and analysis were written inhouse using compiled BASIC (QuickBASIC, Microsoft Corp.) and conform to Macintosh user-interface quidelines. In the Macintosh FinderTM environment, the data acquisition program is selected as a start-up application. Thus the system is capable of resuming data acquisition following a power failure without user intervention. Figure 1 depicts a flow diagram for the data acquisition routine. The acquisition program collects two types of data. Lowfrequency data are collected from all animals for the determination of cumulative food intake and meal patterns. The program polls all balances twice and records the first feeder weight value to a random-access disk file. In the interval between low-frequency samples, the program determines if any animal is eating based upon the absolute value of the difference between the two sampies collected. If this difference exceeds a user-selectable criterion (nominally 0.2 grams), high-frequency data are collected from that animal for evaluation of the fine structure of bites and/or licks. If more than one animal is eating simultaneously, the program will sample the animal for which the least highfrequency data has been previously recorded. The time interval between low-frequency samples is user-selectable within a range of 2-60 seconds, and is nominally set at 20 seconds. The highfrequency data are collected at a fixed maximal rate of 6 Hz, which represents a limitation of the balance rather than the computer or network. Bursts of high-frequency data are recorded to a separate random-access disk file. Storage requirements for data acquisition are variable, based upon the sampling interval and feeding behavior, but average less than 1 megabyte per day for an 8-channel system. When a trial is completed or at desired intervals, data acquisition is suspended, and data files for high- and low-frequency data are backed up to floppy disks. These data are restored to the fixed disk of another Macintosh computer for analysis. Figure 2 depicts the flow diagram for determination of feeding events (i.e., meals and/or bouts). The criterion for initiation of a feeding event is deflection of the feeder weight beyond a user-selectable threshold. This threshold is nominally set to 0.1 grams. The criterion for termination of a feeding event is a lack of deflection of the feeder weights above the event initiation threshold for a user-selectable number of samples. This number is adjustable within a range of 1 to 200 samples, and is nominally set to a number of low-frequency samples corresponding to 30 seconds for determination of bouts (2) or 10 minutes for
FIG. 1. Flow diagram for data acquisitionroutine. The data acquisition program samples all selected balances continuouslyuntil interruptedby the user.
determination of meals (4). Meal or bout data are exported in tab-delimited format for further analysis by a spreadsheet program (Excel 2.2, Microsoft Corp.). In the algorithm for calculating cumulative food intake, the ending feeder weight for the last event is subtracted from the beginning feeder weight of the first event. In addition to cumulative food intake, the postevent interval and overall rate of feeding (i.e., size/duration) are determined. Cumulative food intake data for selected periods are plotted using Cricket Graph 1.3.2 (CA-Cricket Software, Inc.). High-frequency data are exported in tab-delimited format to a data file utilized by HyperCard 2.0 (Clads Software, Inc.) for automated charting and analysis. Numerical and graphic data for each high-frequency burst is represented upon a single "card" within the HyperCard "stack." Routines written in HyperTalk, the scripting language of HyperCard, plot the feeder weight data for each burst in Cartesian coordinates. The raw feeder weight data for each burst is displayed in a scrolling text field, and the time interval encompassing the burst is also recorded. RESULTS Data obtained with lean (fa/?) and obese (fa/fa) Zucker rats exemplify use of the system for evaluation of cumulative food intake and meal patterns. Figure 3 portrays a 65-hour segment of cumulative food intake data for 4 markedly obese male Zucker rats and 4 lean male litterrnates. Figures 4 and 5 illustrate the diurnal distribution of postmeal interval and meal size for both phenotypes obtained during the same sampling period. Figure 6 depicts the acute effect of intracerebroventricularly administered ~-hydroxybutyric acid on the duration of meals. Figures 7 and 8 are screen captures from the HyperCard stack which was developed to view and evaluate high-frequency burst data collected with the system. These example cards depict the different waveforms obtained from a domestic cat consuming commercial dry cat food and premium canned cat food, respectively.
RECORDING AND ANALYSIS OF FEEDING BEHAVIOR
405
DISCUSSION
m e a l initiation
I •-b i R,~ ~xt.,~a. I
~m d~, I ~
I compare w~h previo~-,,rr,~
I
I [
I n c r e m e n t t e r m i n a t i o n interval
I
no
I ~o~
fin-'*,~h,
I
FIG. 2. Flow diagram for data evaluation routine. By selection of appropriate user-adjustable parameters, feeding bouts rather than meals are evaluated.
Representative data obtained with the system illustrate the merits of the dual-frequency approach to acquisition of feeding behavior data. Three models of altered feeding behavior, which have genetic, pharmacological and sensory etiology, demonstrate the utility of the system• The hyperphagia of genetically obese Zucker rats is evident within minutes after placing rats in the tunnel feeders. Analysis of the diurnal distribution of meal size and postmeal interval reveals that the hyperphagia of the obese phenotype is primarily attributable to increased meal size. In the obese phenotype only, an increasing motivation to eat as the dark phase progresses is evidenced by a trend toward increasing meal size. Despite differences in feeder configuration (tunnel vs. upright hopper) and feed consistency (powdered vs. pelleted), these data are in agreement with previously reported data (18). Transient, pharmacologically induced increases of meal duration can be captured by the system. An example of this is shown in rats administered ICV with "y-hydroxybutyric acid. The acute increase in meal duration would have eluded detection using conventional approaches of weighing food cups 2-3 hours postinjection. The sensory qualifies of the diet have been shown to affect the microstructure of feeding behavior as well as the quantity of food consumed (6). Feeder waveforms such as those presented in Figs. 7 and 8 suggest that spectral components of biting and/or licking may provide an index of palatability or preference. Temporal resolutions of less than one second are the hallmark of lickometer- and eatometer-based systems, but have been previously unavailable in systems that directly measure the amount of food consumed. The system is primarily used to measure the feeding behavior of rats, but data obtained from the domestic cat confirm the utility of the system for evaluating the feeding behavior of other species. Mettler " P M " series electronic digital balances with capacities up to 30 kg may be used without software modification, so the system is applicable to studies with larger animals. High-frequency burst data are voluminous, so it is not feasible to manually graph the data associated with every burst. Depending upon the length of an experiment and the number of channels sampled, a single experiment will contain thousands of
Cumulative Food Intake off
100.
off
off
8070-. E •
~
'
~ '
50-
30
0 ~. -409
~
~
-
-1031
•
~. 2471
_ 39~ 1
elspeed minutes
FIG. 3. Cumulative food intake for 4 lean (fa/?, dashed line) and 4 obese (fa/fa, solid line) Zucker rats. Lean and obese rats weighed 209.0-+8.3 and 269.9__.6.0 grams (mean-+SEM), respectively, when data acquisition was initiated, F(1,3)=35.4, p
0031-9384/91 $3.00 + .00
A System for Automated Recording and Analysis of Feeding Behavior M A R T I N G. HULSEY AND ROY J. MARTIN
Department of Foods and Nutrition, College of Family and Consumer Sciences The University of Georgia, Athens, GA 30602 Received 21 February 1991 HUI_~EY, M. G. AND R. J. MARTIN. A systemfor automated recording and analysis of feeding behavior. PHYSIOL BEHAV 50(2) 403--408, 1991.--We have designed and implemented a system that utilizes a network of top-loading balances digitally interfaced to a Macintosh computer. The system simultaneouslycollects two forms of data which allow the evaluation of the animal's biting and/or licking behavior in addition to cumulativefood intake and meal patterns. The system is capable of resuming data acquisition following a power failure without user intervention.Plexiglas cages utilized with the system features adjustable tunnel feeders and are appropriate for use with small rodents. Givenappropriate caging, the system may be utilized to evaluate the feeding behavior of other species. Food intake
Ingestion
Computer
Microcomputer
THE assessment of feeding behavior is prerequisite for studying physiological controls of feeding or evaluating the efficacy and specificity of agents that stimulate or inhibit feeding, Due to the prohibitive cost of apparatus for automated acquisition and processing of such data, many studies are performed without its benefit. Short-term measurements of food intake with deprived animals are common, but this paradigm has been criticized (9,14). Intermittent manual weighing of feed containers with freely feeding rats is another common method, but it has been noted that manipulation of feed containers per se can affect feeding behavior (7). Numerous approaches have been employed to capture feeding data. Operant paradigms (1,19) impose constraints which alter feeding as compared to the free-feeding state (14), but these constraints can be used to advantage when investigating feeding in an ecological context (5). An automated recording device was developed whereby the rat's removal of a pellet triggers the release of another (12). This approach was later refined by adding automated analysis of pellet consumption (11,24). Lickometer circuitry (21) has been employed to examine the microstructure of consumption of liquid diets (6), and has been adapted to a modified food container or "eatometer" to record a rat's contact with the feed (8,22). It is noteworthy that drinkometers and eatometers provide superior temporal resolution, but only indirect measurements of the amounts of food consumed. Another indirect approach is to measure the animal's entry into a feeding tunnel (19). This method has also been utilized to measure diet selection behavior (16). Continuously monitoring the weight of conventional feed containers with strain gauges permitted utilization of diets of a consistency that precludes its extrusion into pellets (10,17). A recently reported system (15) utilized a similar approach and featured photocells to simultaneously record a rodent's noningestive approaches to the feed container. Systems that utilized a
Meal patterns
network of electronic digital balances with a serial communications interface provided automated analysis of meal pattern data on a computer (3,4). In developing this system, our engineering goals were: 1) temporal resolution equivalent to a lickometer or pellet dispenser; 2) mass resolution equivalent to a load cell; 3) automated processing of meal pattern and cumulative food intake data; and 4) utilization of diets of widely varying composition. We have designed and implemented a system that fulfills these criteria. METHOD
Cage Design Plexiglas cages utilized with the system are appropriate for use with small rodents. The cage modules consist of twin Plexiglas enclosures measuring 28 cm in width, 27 cm in depth, and 38 cm in height. Each enclosure has a removable floor of 1-cm mesh stainless steel construction, and removable lids fabricated of the same material are secured with stainless steel pins. Each enclosure has an adjustable tunnel feeder. The entrance to the tunnel is elevated 5 cm above the floor of the enclosure so that rodents may enter from the forward direction only. The bottom of the tunnel is fabricated of 1-cm stainless steel mesh. An orifice is provided in the bottom of the tunnels through which the animals can feed. Stainless steel rings may be substituted to provide an opening of 2.5 to 5.0 cm in 0.6-cm increments. The sides of the tunnels are fabricated of 6 mm Plexiglas and are adjustable in width over a range of 5 to 10 cm at the rear and 2.3 to 6.5 cm at the front. The top of the tunnel is fabricated of 24-gauge stainless steel with a height adjustable from 3 to 16.5 cm. These dimensions were selected to accommodate rodents of various sizes, from weanling rats to mature obese fedfa Zucker rats. At its extreme height adjustment, the top of the tunnel is intended to clear chronic infusion leads.
403
404
HULSEY AND MARTIN
Sampleallbalancestwice
Hardware Glass feed containers (Allentown Caging, Inc.) rest upon Mettler PM4000 top-loading digital balances. Polystyrene containers 15 cm in diameter contain spillage within the balance pan. Balances rest atop scissor-type laboratory jacks (Fisher Scientific) so that the height of the feeder can be brought in proximity to but not contacting the orifice in the bottom of the feeding tunnel. The balances are networked to an eight-channel code-operated switch (COS-8, Black Box Corp.) using a threewire (pins 2, 3 and 7) RS-232 protocol at 9600 baud. The master port of the switch is connected through a shielded threeconductor cable to the RS-422 serial port of a Macintosh SE microcomputer located in an adjacent room. The system currently consists of eight channels, but is expandable to 64 or more channels by chaining multiple switches.
7/
J
yes
ISelo~ burstchantm, I
I
I recordbursttodisk J
L. yes
~
no
Software Programs for data acquisition and analysis were written inhouse using compiled BASIC (QuickBASIC, Microsoft Corp.) and conform to Macintosh user-interface quidelines. In the Macintosh FinderTM environment, the data acquisition program is selected as a start-up application. Thus the system is capable of resuming data acquisition following a power failure without user intervention. Figure 1 depicts a flow diagram for the data acquisition routine. The acquisition program collects two types of data. Lowfrequency data are collected from all animals for the determination of cumulative food intake and meal patterns. The program polls all balances twice and records the first feeder weight value to a random-access disk file. In the interval between low-frequency samples, the program determines if any animal is eating based upon the absolute value of the difference between the two sampies collected. If this difference exceeds a user-selectable criterion (nominally 0.2 grams), high-frequency data are collected from that animal for evaluation of the fine structure of bites and/or licks. If more than one animal is eating simultaneously, the program will sample the animal for which the least highfrequency data has been previously recorded. The time interval between low-frequency samples is user-selectable within a range of 2-60 seconds, and is nominally set at 20 seconds. The highfrequency data are collected at a fixed maximal rate of 6 Hz, which represents a limitation of the balance rather than the computer or network. Bursts of high-frequency data are recorded to a separate random-access disk file. Storage requirements for data acquisition are variable, based upon the sampling interval and feeding behavior, but average less than 1 megabyte per day for an 8-channel system. When a trial is completed or at desired intervals, data acquisition is suspended, and data files for high- and low-frequency data are backed up to floppy disks. These data are restored to the fixed disk of another Macintosh computer for analysis. Figure 2 depicts the flow diagram for determination of feeding events (i.e., meals and/or bouts). The criterion for initiation of a feeding event is deflection of the feeder weight beyond a user-selectable threshold. This threshold is nominally set to 0.1 grams. The criterion for termination of a feeding event is a lack of deflection of the feeder weights above the event initiation threshold for a user-selectable number of samples. This number is adjustable within a range of 1 to 200 samples, and is nominally set to a number of low-frequency samples corresponding to 30 seconds for determination of bouts (2) or 10 minutes for
FIG. 1. Flow diagram for data acquisitionroutine. The data acquisition program samples all selected balances continuouslyuntil interruptedby the user.
determination of meals (4). Meal or bout data are exported in tab-delimited format for further analysis by a spreadsheet program (Excel 2.2, Microsoft Corp.). In the algorithm for calculating cumulative food intake, the ending feeder weight for the last event is subtracted from the beginning feeder weight of the first event. In addition to cumulative food intake, the postevent interval and overall rate of feeding (i.e., size/duration) are determined. Cumulative food intake data for selected periods are plotted using Cricket Graph 1.3.2 (CA-Cricket Software, Inc.). High-frequency data are exported in tab-delimited format to a data file utilized by HyperCard 2.0 (Clads Software, Inc.) for automated charting and analysis. Numerical and graphic data for each high-frequency burst is represented upon a single "card" within the HyperCard "stack." Routines written in HyperTalk, the scripting language of HyperCard, plot the feeder weight data for each burst in Cartesian coordinates. The raw feeder weight data for each burst is displayed in a scrolling text field, and the time interval encompassing the burst is also recorded. RESULTS Data obtained with lean (fa/?) and obese (fa/fa) Zucker rats exemplify use of the system for evaluation of cumulative food intake and meal patterns. Figure 3 portrays a 65-hour segment of cumulative food intake data for 4 markedly obese male Zucker rats and 4 lean male litterrnates. Figures 4 and 5 illustrate the diurnal distribution of postmeal interval and meal size for both phenotypes obtained during the same sampling period. Figure 6 depicts the acute effect of intracerebroventricularly administered ~-hydroxybutyric acid on the duration of meals. Figures 7 and 8 are screen captures from the HyperCard stack which was developed to view and evaluate high-frequency burst data collected with the system. These example cards depict the different waveforms obtained from a domestic cat consuming commercial dry cat food and premium canned cat food, respectively.
RECORDING AND ANALYSIS OF FEEDING BEHAVIOR
405
DISCUSSION
m e a l initiation
I •-b i R,~ ~xt.,~a. I
~m d~, I ~
I compare w~h previo~-,,rr,~
I
I [
I n c r e m e n t t e r m i n a t i o n interval
I
no
I ~o~
fin-'*,~h,
I
FIG. 2. Flow diagram for data evaluation routine. By selection of appropriate user-adjustable parameters, feeding bouts rather than meals are evaluated.
Representative data obtained with the system illustrate the merits of the dual-frequency approach to acquisition of feeding behavior data. Three models of altered feeding behavior, which have genetic, pharmacological and sensory etiology, demonstrate the utility of the system• The hyperphagia of genetically obese Zucker rats is evident within minutes after placing rats in the tunnel feeders. Analysis of the diurnal distribution of meal size and postmeal interval reveals that the hyperphagia of the obese phenotype is primarily attributable to increased meal size. In the obese phenotype only, an increasing motivation to eat as the dark phase progresses is evidenced by a trend toward increasing meal size. Despite differences in feeder configuration (tunnel vs. upright hopper) and feed consistency (powdered vs. pelleted), these data are in agreement with previously reported data (18). Transient, pharmacologically induced increases of meal duration can be captured by the system. An example of this is shown in rats administered ICV with "y-hydroxybutyric acid. The acute increase in meal duration would have eluded detection using conventional approaches of weighing food cups 2-3 hours postinjection. The sensory qualifies of the diet have been shown to affect the microstructure of feeding behavior as well as the quantity of food consumed (6). Feeder waveforms such as those presented in Figs. 7 and 8 suggest that spectral components of biting and/or licking may provide an index of palatability or preference. Temporal resolutions of less than one second are the hallmark of lickometer- and eatometer-based systems, but have been previously unavailable in systems that directly measure the amount of food consumed. The system is primarily used to measure the feeding behavior of rats, but data obtained from the domestic cat confirm the utility of the system for evaluating the feeding behavior of other species. Mettler " P M " series electronic digital balances with capacities up to 30 kg may be used without software modification, so the system is applicable to studies with larger animals. High-frequency burst data are voluminous, so it is not feasible to manually graph the data associated with every burst. Depending upon the length of an experiment and the number of channels sampled, a single experiment will contain thousands of
Cumulative Food Intake off
100.
off
off
8070-. E •
~
'
~ '
50-
30
0 ~. -409
~
~
-
-1031
•
~. 2471
_ 39~ 1
elspeed minutes
FIG. 3. Cumulative food intake for 4 lean (fa/?, dashed line) and 4 obese (fa/fa, solid line) Zucker rats. Lean and obese rats weighed 209.0-+8.3 and 269.9__.6.0 grams (mean-+SEM), respectively, when data acquisition was initiated, F(1,3)=35.4, p
