Accuracy and reliability of total body electrical conductivity (TOBEC) for determining body composition of rats in experimental studies

Physiology & Behavior,Vol. 56, No. 4, pp. 767-773, 1994 Copyright © 1994 ElsevierScienceLtd Printed in the USA. All rights reserved 0031-9384/94 $6.00 + .00

Pergamon 0031-9384(94)E0123-L

Accuracy and Reliability of Total Body Electrical Conductivity (TOBEC) for Determining Body Composition of Rats in Experimental Studies 1 RHONDA

C. B E L L , .2 A M Y J. L A N O U , t E D W A R D A. F R O N G I L L O , D A V I D A. L E V I T S K Y t A N D T. C O L I N C A M P B E L L t

JR.,t

*Department of Medicine, Division of Endocrinology and Metabolism, University of Alberta, Edmonton Alberta, Canada, T6G 2S2 and "/'Division of Nutritional Sciences, Cornell University, Ithaca, NY 14853 R e c e i v e d 19 O c t o b e r 1993 BELL, R. C., A. J. LANOU, E. A. FRONGILLO, JR., D. A. LEVITSKY AND T. C. CAMPBELL. Accuracyand reliabilityof total body electricalconductivity(TOBEC)for determiningbody compositionof rats in experimentalstudies. PHYSIOL BEHAV 56(4) 767-773, 1994.--Total body electrical conductivity (TOBEC) has been promoted as a noninvasive method to estimate body composition in small mammals. Validation of this method has primarily been under normative conditions and has generally been inadequate. This article reports on the reliability and accuracy of TOBEC methodology to assess gradual, physiologically induced changes in body composition in rats under different experimental conditions. Reliability of the index of electrical conductivity (EM number) was assessed by analyzing components of variance. Accuracy was assessed by comparing EM number to actual lean body mass (LBM, from carcass analysis), across different experimental conditions, within a particular experimental condition, and over time for a given set of animals. Reliable measurements were obtained by strictly adhering to a standard protocol. TOBEC was inaccurate across experimental conditions, within experimental conditions, and within a single experimental condition during the course of an experiment. This inaccuracy apparently stemmed from the lack of a direct relationship between EM number and LBM; EM number was more strongly correlated with body weight than with LBM. At the present time, TOBEC cannot be used in place of carcass analysis to accurately predict the body composition of rats during or following the administration of a variety of experimental conditions. Total body electrical conductivity (TOBEC)

Body composition

T O T A L body electrical conductivity (TOBEC) has recently been promoted as a rapid, noninvasive method to estimate body composition in small animals (26). An electrical conductivity index, known as the EM number, is calculated from the output of the instrument and, ideally, the EM number can be related to the lean body mass (LBM) of animals (obtained by direct carcass analysis) by a regression equation, which can then be used to predict the LBM of other animals. This eliminates the necessity of sacrificing representative groups of animals to determine body composition over the course of an experiment. This technique, if accurate and reliable, would make the assessment of body composition of small animals possible in a variety of experimental and field situations where termination of the animal's life is undesirable. If this instrumentation is to be useful for prediction of body composition, then the observations obtained from it must first be reliable. For a method to be reliable, the observations from the

Rats

instrument must be both precise and dependable, under a wide variety of experimental conditions. Second, the EM number must accurately reflect differences or changes in LBM. For the TOBEC instrument to be of use, the relationship between EM number and body composition must accurately predict the body composition of animals during or following different experimental manipulations, such as dietary changes or the administration of a drug. In several studies where TOBEC measures were made only at a single time point, in normal animals, strong correlations between the lean mass of rodents and small birds and the EM number have been reported (5,9,18,22,26), suggesting that it would be possible to predict LBM of animals from EM number. Three recent reports (3,4,22) have questioned the usefulness of TOBEC measures. Rumpler et al. (22) reported that the inclusion of EM number in a regression model that already included weight and sex did not improve the prediction of LBM in rodents. Baer et

Presented in part at the 75th Annual Meeting of the Federation of the American Societies of Experimental Biology, Atlanta, GA, 1991. Mention of a trademark or proprietary product does not constitute a guarantee or warranty of the product and does not imply its approval to the exclusion of other like products. 2 Requests for reprints should be addressed to Rhonda C. Bell, Ph.D., at her present address: Department of Health Studies and Gerontology, University of Waterloo, Waterloo, Ontario, Canada N2L3Gl. 767

768

t3El,t. I~l AI_

al. (3) compared TOBEC to direct carcass analysis and concluded that TOBEC may not be useful for predicting LBM for individuals, but may be useful for predicting mean body mass for a group of individuals. Bellinger and Williams (4) evaluated the ability of TOBEC to accurately assess acute changes in body composition induced by injecting known amounts of saline or oil into rats. They concluded that results from TOBEC may not be valid across size, gender, or breed of rat. Most of these studies used correlation methods, which are not appropriate for determining whether a new technique, such as TOBEC, sufficiently agrees with an established one to be a valid substitute for it (1,2). None of these studies, except Rumpler et al. (22), evaluated whether TOBEC was a better predictor of LBM than was body weight. None of these studies except Baer et al. (3) and Bellinger and Williams (4) evaluated TOBEC under conditions of altered body composition. No comprehensive validation study has been reported on animals under experimental conditions that were expected to result in gradual, physiologically induced differences or changes over time in body composition. The purpose of this paper is to report on the reliability and accuracy of the TOBEC methodology to assess gradual, physiologically induced changes in body composition in groups of rats under several experimental conditions and over time, using results from a series of five studies. This validation is comprehensive because it includes: 1) the assessment of reliability by replication, 2) the assessment of accuracy by comparison to a gold standard, 3) comparison of the accuracy of TOBEC to the accuracy of competing indicators (body weight), and 4) the use of applied conditions that challenge the theoretical basis for the method. The limitations and usefulness of the TOBEC instrument are discussed. METHOD

Determination of Body Composition TOBEC method. TOBEC instrumentation was originally developed to determine fat or lean content of ground meat (Dj-me 100, Dickey John, Auburn, IL) and is now produced specifically for use with live animals (EM-Scan, Springfield, IL). The method is based on the principle that lean tissue is more electrically conductive than fat tissue. An immobilized animal is placed inside the measurement chamber of the apparatus, which is surrounded by a coil. The animal's presence alters the electromagnetic conductivity of the coil; this change is a function of the total electrical conductivity of the animal's body, which in turn is related to the mass of lean and fat tissue, body geometry, and fluid balance of the animal (20,26). In theory, the output generated by the TOBEC instrument is related to the animal's LBM, and fat mass is calculated by difference from the rat's body weight. The TOBEC of each animal was measured in a Dickey John Meat Fat Tester (now equivalent to the EM-Scan SA-1 Small Animal Body Composition Analyzer, EM-Scan, personal communication) using procedures outlined by Walsberg (26) and following the recommendations given in discussions with EM-Scan and Dickey John. The carrier tube for the TOBEC instrument was marked to clearly identify the center of the tube and 1.5 in. from each end of the tube. A carrier platform, placed inside of the horizontal carrier tube, was selected so that the animal was centered vertically inside the tube. Animals were anesthetized until they were completely limp using an inhalant type of anesthetic (methoxyflurane, Pitman-Moore, Inc., Washington Crossing, N J) and were then positioned prone, along the long axis of the carrier platform, with the tail tucked under the body, so that the entire

conductive mass of the animals was inside the measurement chamber (26). The platform was placed inside the carrier tube with the animal's nose toward the proximal end of the tube; this allowed a nose cone containing additional anesthetic to remain over the animal's nose at all times. This type of short-term an esthesia was chosen to minimize the time that an animal's eating behavior and metabolism would be disrupted. The nose cone with the anesthetic did not affect the TOBEC readings (unpublished data). The carrier platform supporting the animal was moved inside the carrier tube so that the animal's sternum was brought in line with the center marking of the carrier tube. TOBEC readings are known to be affected both by the geometry of the object and by the position of the object within the instrument because the conductivity of only one "'slice'" of the animal is measured. In Studies I and II (see below), animals were wrapped tightly in a piece of cloth to produce a more cylindrical shape and to reduce small movements (similar to birds being placed in a stocking) before being placed in the TOBEC instrument. In Studies IV and V, animals were measured without the wrap. Both the wrapped and no-wrap conditions were used in Study Ill. Two or three independent readings were taken at each position in each of the wrapped and/or no-wrap conditions, as specified in each study. The means of multiple readings were used in the analyses of accuracy described. The phantom tube (provided with the TOBEC instrument) was measured at regular intervals each day that TOBEC measurements were done, and did not vary. Carcass analysis. Whole-carcass analysis was performed according to standard protocols as follows and was comparable to the methods outlined by Baer et al. (3). On specified days of a study and following the final TOBEC measurement, animals were weighed, euthanized (by CO~_ inhalation), and the carcasses were frozen. Carcasses were later autoclaved for 3 h and then ground in a blender. The homogenate was aliquoted into approximately 10-ml samples (15), weighed, freeze-dried, and reweighed. The freeze-dried sample was further ground to a fine powder. In a minimum of two subsamples, the fat content was determined by ether extraction in a Soxlet apparatus ( 13,14); lean mass was determined by difference as LBM (g) = body weight (g) -- body fat (g). For the extraction procedure, the coefficient of variation among replicate samples was less than 3%.

Data Sources and Study Descriptions Data for this report came from five studies conducted between 1989 and 1990, all of which used TOBEC to assess changes in the body composition of rats. Generally, the purpose of each study was to assess changes in body composition under different dietary treatments or following treatment with anorectic drugs. Study l--preliminary study. This study familiarized us with the equipment and the protocol outlined by Walsberg (26) and by the EM-Scan Operation Manual (10). Six F344 male rats were maintained on one of five diets containing different amounts of protein (4%, 8%, 12%, 16%, or 20%, % casein w/w) from approximately 8 weeks of age until the conclusion of the study. Three TOBEC measurements were taken on 2 consecutive days, three times during a 6-week period. Animals were measured using the wrapped condition each time. Study H--weight loss study. The purpose of this study was to assess weekly changes in body composition using TOBEC in groups of adult rats that were losing weight due to dietary manipulation or due to the incorporation of the anorectic drug dfenfluramine into the diet. Female Long Evans hooded rats (Blue Spruce, NY) weighing approximately 350 g were housed individually in wire mesh cages. Animals were maintained on lab

TOBEC IN ASSESSMENT OF BODY COMPOSITION

chow (Purina) for 2 weeks, and then were assigned to one of the following six dietary treatment groups: high fat (n = 10), high fat + fenfluramine (n = 9), high fat diluted with cellulose (n = 10), high carbohydrate (n = 9), high carbohydrate + fenfluramine (n = 9), or high carbohydrate diluted with cellulose (n = 10); groups were matched for initial body weight. The composition of the high-fat diets was 45% carbohydrate, 25% fat, and 20% protein (w/w) whereas the high-carbohydrate diets were 65% carbohydrate, 5% fat, and 20% protein. Because the caloric densities of the diets were different, fenfluramine was added to the diets on a per kcal basis (0.04 g/kcal). Fenfluramine has been shown to have anorectic effects and is thought to work by depressing food intake, increasing peripheral glucose uptake, mobilizing fat stores, and reducing fat synthesis (6). Body weight was measured daily; body composition was assessed using TOBEC (three readings, wrapped) at the start of the experiment and weekly thereafter for the 6 weeks of the study. A group of animals (n = 9) was sacrificed before the introduction of the treatments to assess initial body composition. The remaining animals were sacrificed at the conclusion of the study for carcass analysis. Study Ill--single time point study. The purpose of this study was to develop a regression equation for prediction the lean body mass (LBM) of rats for use in subsequent studies as suggested by the EM-Scan Operation Manual (10). Twenty rats (five male Sprague-Dawley, 15 female Long Evans hooded) weighing 130-420 g were measured in the TOBEC instrument in the wrapped and no-wrap condition. Two independent readings were taken in each of the wrapped and no-wrap conditions. Immediately after measuring in the TOBEC instrument, animals were euthanized and the carcasses frozen for direct determination of fat tissue by extraction (13). Study IV--growth study. Thirty male Fisher 344 rats weighing approximately 205 g were individually housed in hanging stainless steel cages. All animals were fed an A1N-76A-based diet containing 22% casein as the protein source (control diet, Dyets, Bethlehem, PA). At the beginning of the study, 15 animals were randomly assigned to receive fenfluramine mixed into their food (0.15 g/kg food or 0.04 g/kcal). The 30 animals reported here were control groups for a larger concurrent study. Body weight was measured twice weekly. Body composition was assessed using the TOBEC instrument before the introduction of the drug, and after 1, 3, 6, and 9 weeks of control diet or drug treatment to assess differences in the accumulation of LBM in growing animals. Three independent readings were taken and only the no-wrap condition was used in this study. Study V--fenfluramine/fluoxetine study. Study V was designed to assess changes in body composition using TOBEC and concomitant carcass analysis among animals given d-fenfluramine or fluoxetine to promote weight loss. Female, Long Evans hooded rats (Blue Spruce, NY) weighing about 250 g were housed individually in hanging stainless steel cages equipped with water feeders. At the start of the experiment, 10 animals were measured in the TOBEC instrument and were sacrificed immediately to determine initial carcass composition. The remaining 100 animals were measured using TOBEC, and were then assigned to one of the following groups (numbers of animals in each experimental group are shown in Table 2): fenfluramine, fluoxetine, weight matched/restricted intake, or no drug/ad lib control. Within each drug-treated group, animals were assigned to receive a high, medium, or low dose of the drug. Each drug was mixed with powdered, AIN-76A-based diet containing 20% (w/w) casein, 5% corn oil, and 65% corn starch. The doses of fenfluramine used were 0.30 mg/kg diet, 0.15 mg/kg diet, and 0.075 mg/kg diet, whereas the doses of fluoxetine used were 400

769

mg/kg diet, 200 mg/kg diet, and 100 mg/kg food. A weightmatched/restricted intake group was fed to induce the weight loss observed in the medium dose fenfluramine group and the high dose fluoxetine group. Food was available ad lib to all animals except those in the weight-matched groups. Body weight and food intake were measured daily during this study. After 3 weeks of treatment, a subset of animals from the control, high-dose fenfluramine, medium-dose fenfluramine, high-dose fluoxetine, and weight-matclaed fenfluramine groups were measured using TOBEC and were sacrificed for determination of carcass composition. At the conclusion of the study (6 weeks after the initiation of treatments), all remaining animals were measured in the TOBEC instrument and sacrificed for subsequent carcass analysis.

Statistical Analysis Reliability. Reliability refers to the total variability among observations when the same animal is measured repeatedly. Although conceptually the components of reliability are precision and dependability, in practice they are quantified as sources of error known as imprecision and undependability. The degree to which imprecision and undependability contribute to the variation among repeated observations is quantified by analyzing the components of variance. Imprecision is the degree to which replicate measurements are spread out or scattered about their target value (16,17). It is random error in a given measurement and, for the TOBEC measurements, may include such things as slight variation in the position of the animal inside the chamber or minor movements of the animal while under anesthetic. Undependability is the error in a measurement that results from changes in conductivity due to physiological changes within the animal (other than body composition), as well as changes in environmental conditions that affect replicate measurements. For example, fluid shifts within the animal due to anesthetic may affect the conductivity and may thereby affect the observed EM number, although the LBM of the animal would not have changed. In addition, day-to-day differences in humidity or temperature may impact on the instrument, thus affecting the undependability of the EM number, though again, the actual LBM of the animal remained constant. Variance component techniques were used to partition the total variability in repeated measures in the TOBEC measurement process. The variation among animals and the random sources of variation within animals (imprecision and undependability) were assumed to be random effects. Dietary treatment was considered to be fixed effect for the purpose of estimating the variance components. The particular sources of variability were identified for each study, and a maximum likelihood procedure was used to quantify the variance components (SAS Institute, Cary, NC, proc varcomp) (23). The variance components were also estimated using a method of moments (Type I ) and a norm quadratic method (Mivque0), and the answers were similar to those obtained using the maximum likelihood procedure. Accuracy. Accuracy refers to the degree to which the predicted outcome reflects the true value of that outcome. If the TOBEC instrument is to be useful, accurate prediction of LBM is critical, and at least one of three criteria would have to be met. The order of these criteria is hierarchical, and has implications for the general applicability of this technique as a method for the determination of body composition. To achieve the broadest applicability, it is necessary to satisfy the first criterion of a uniform relationship between the EM number and the actual LBM (as measured by carcass analysis) across all experimental conditions. This implies that the ordering of the

77(}

BI']A

"FABLE 1 COMPONI:!NTSOF VARIANCE* FOR FIVE SUCCESSIVESTt DIES THAT USED TOBEC TO ASSESS BODY COMPOSITION IN RATS

Study 1 Study 11 Study 111 No wrap Wrapped Study IV Study V 3 Weeks 6 Weeks

lions were compared among the various experimental groups u~ing regression analysis [proc reg (231]. RESIq I"S

Unreliability eWithin-.Animal Variation) Study

I,"1 , \ I .

Reliability

Among-Animal VaIiation

Imp.

Total Undep. Variance

Variability¢ (%)

2702 4720

921 892

4.55 82/)§

3620 5694

8.4 5.2

5999 6138 968

231 347 239

NA¶ NA NA

623(I 6485 1911

1,2 1.8 6.6

4918 6228

292 181

NA NA

5210 6409

1.9 0.9

Imp. = Imprecision; Undep. = Undependability. * Based on maximum likelihood estimates. [ Percentage of total variability contributed by imprecision when three replicate measurements were averaged. $ Day-to-day variation. § Week-to-week variation. ¶ NA = not applicable.

The variance components from seven sets of data observed in the five studies outlined above are summarized in Table 1. The variation among animals accounted for the largest portion of the total variance in each study. Differences in this quantity among the seven sets of data reflects the range of variation in body weight or body size in a study at the time of the TOBEC measurements. The variation contributed by imprecision differed among the studies. In the initial studies (Studies I and II), the imprecision was relatively high, accounting for approximately 10% of the total variance. In Studies III, IV, and V, which were performed later, imprecision accounted for 1 - 7 % or less of the total variance. The imprecision was slightly higher in the wrapped condition relative to the no-wrap condition. This is supported by the observation that the imprecision in Study IV was similar to Studies III and V, though the total variance was smaller in Study IV than in either of the other two studies. Undependability, as evidenced by day-to-day and week-toweek variation, contributed little to the overall variation.

Relationship in Normal Animals mean LBM of experimental groups would follow the same pattern as would the ordering of the mean EM numbers. If this criterion is met, then a single regression equation could be developed to predict the LBM of animals in a variety of experimental conditions. In this situation, the instrument could be used under a wide variety of experimental situations to predict LBM. The second criterion is a uniform relationship between the individual EM numbers and the LBM within any given experimental condition. If this second criterion was satisfied (even if the first were not), then the TOBEC instrument could be used if the investigator determined separate regression equations for each experimental condition. The applicability of this technique would be diminished, but it could still be useful in situations where an investigator used a series of similar experimental conditions in different studies. The third criterion is that a regression equation for a particular group of animals is stable over time. In this case, prediction of the LBM of animals at several times during an experiment would be possible by assessing the relationships between the EM number and LBM at the conclusion of the experiment and applying this equation to TOBEC measures made during the study. If only the third criterion was met, then the usefulness of the TOBEC methodology would be limited to those protocols where comparison between direct and T O B E C measures of body composition could be done routinely. Regression equations relating LBM (as determined by carcass analysis), and the EM numbers from individual animals observed during TOBEC measurements were generated by regressing EM numbers on the LBM. If the regression equations were to be used for prediction of LBM from EM numbers, the equation would be rearranged, as suggested by other authors (7,19). Data from Studies III and V were used to assess accuracy since the imprecision of the TOBEC measurements in these two studies had been determined to be relatively low, the body composition of the animals bad been determined by carcass analysis, and, in the case of Study V, measures had been taken over the duration of the experiment. The slopes and intercepts of these regression equa-

To confirm a strong linear relationship at a single time measurement, Fig. 1 shows the relationship between individual EM numbers and actual LBM for the 20 animals used in Study III, along with the regression equation that best fits these data points (R 2 = 0.97). The addition of a quadratic term changes the regression equation to EM = 17.2 + 0.712(LBM) - 4.3 * 10 4(LBM)Z, although the R 2 is not increased significantly (R 2 = 0.975, quadratic model). The relationship is influenced by the five points with relatively low LBM. Excluding these points alters the regression equation such that EM = 3.0761(LBM) 334.30, R 2 = 0.78; the addition of a quadratic term does not change the R 2 in this case, although the equation changes to EM = -6.96 + 0.635(LBM) + 4.52(LBM) z. This confirms a strong linear relationship between LBM and EM numbers from normal animals over a broad weight range at a single measurement time.

Accuracy The first criterion of a uniform relationship between EM number and LBM across all experimental groups is not met. Figure 2 shows the relationship between mean EM number and mean 8OO

==

700

•

600 E

500

Z

40O

•

x - 225.73 2OO tOO

•

•

1 O0

, 3O0

20O Lean

Body

Mass

•

, 4OO

(g)

FIG. I. Relationship between Lean body mass (LBM) from carcass analysis and EM number from TOBEC instrument for Study III. Scatterplot of the individual values is superimposed on the regression line.

T O B E C IN A S S E S S M E N T O F B O D Y C O M P O S I T I O N

480

'

460

"

440

'

771

TABLE 2 REGRESSION EQUATIONS RELATING EM NUMBER TO LBM FOR GROUPS FROM STUDY V

J~

E ,-,l z

=Z

420 -

Body Weight (g)

Experimental Condition

Regression Equation

400-

ILl

38036O 220

230

240

250

260

Lean Body Mass (g) FIG. 2. Scatterplot of the group mean LBM vs. mean EM number for the 10 experimental treatment groups of Study V. There is no pattern that is readily discernible among these points; the correlation coefficient among these points is 0.19.

(actual) L B M for the 10 experimental groups of Study V. There is no uniform pattern that is readily observable among these points. This suggests that no single regression equation could be adequate for describing the relationship between the E M n u m b e r and the actual L B M of animals under different experimental conditions. The actual L B M of animals in Study V ranged from approximately 190 to 290 g, whereas the mean (__+SD) E M number was 418 (_+ 84), which suggests that there is sufficient variability in these data set to support the regression analysis. Table 2 gives the regression equations for each experimental group in Study V. The slopes and the intercepts of the regression equations from several experimental conditions were significantly different from the initial kill condition (normal), again supporting the observation that the relationship between E M n u m b e r and L B M is not uniform among the different experimental conditions• Two situations were used to assess whether the second criteflon, uniformity of the relationship between EM number and L B M under similar conditions, was met. First, regression equations for similar experimental groups in Study V were compared (see Table 2) and were observed to be different from each other. That is, regression equations for the three fenfluramine conditions were different from each other, and those for the three fluoxetine conditions were different from each other. Second, regression equations from the initial kill and control groups in Study V (see Table 2) and the 15 animals from Study III (see text above and Fig. l) were compared. These equations were also found to differ• Therefore, the second criterion was not satisfied. The third criterion of a stable relationship between E M number and L B M for a particular experimental group over time was evaluated by determining regression equations for several experimental conditions after 3 and 6 weeks of treatment in Study V. These are presented in Table 3. Equations for several of the conditions were not stable over time; generating a regression equation for a treatment group at the end of the study and applying this to E M numbers recorded earlier may not accurately predict L B M of the animals during the course of a study, depending on the experimental condition. Therefore, the third criterion was not satisfied. DISCUSSION

Reliability Data from this compilation of studies were used to assess the reliability and accuracy of T O B E C instrumentation as a method for assessing body composition in rats under a variety of experimefital treatments and over time. In our later studies (III, IV,

Initial kill (n = 10) Control (n = 9) High fenfluramine (n = 9) Medium fenfluramine (n = 15) Low fenfluramine (n = 7) High fluoxetine (n = 13) Medium fluoxetine (n = 10) Low fluoxetine (n = 10) Weight-Matched fen (n = 10) Weight-Matched fluox (n = 8)

314.0 327.9 300.5 317.5 319.8 289.6 286.6 320.1 315.5 272.3

± 11.9 ± 14.9 ± 11.8 ± 11.4 +_ 19.6 ± 9•5 ± 13.0 ± 11.5 ± 9•7 ± 9.5

3.46(LBM) 1.95(LBM) 3.04(LBM) 3.17(LBM) 2.08(LBM) 2.96(LBM) 2.30(LBM) 2.28(LBM)2.19(LBM) 3.90(LBM) -

385.1 15.0*t 300.7 314.8 60.5"t 332.2 185.0" 148.3 91.5"t 546.7

* Slopes are significantly different from initial kill group, p < 0.10. t Intercepts are significantly different from initial kill group, p < 0. I 0.

and V), we were able to generate reliable E M numbers from the T O B E C instrument using the protocol outlined by Walsberg (26) and EM-Scan/Dickey John (see Table 1). By averaging multiple repetitions per measurement, the imprecision of a given T O B E C observation represented 1 - 7 % of the total variability observed. The imprecision observed in these studies probably resulted from several factors, including: slight variation in the placement of the animals inside the chamber, the duration and amount of anesthetic administered to the animal, and small movements of the a n i m a l ' s limbs or tail when the animal was inside the chamber.

TABLE 3 REGRESSION EQUATIONS RELATING LBM AND EM NUMBER FOR DIFFERENT EXPERIMENTAL CONDITIONS AT TWO MEASUREMENT TIMES OVER THE COURSE OF THE STUDY* Regression Equation After 3 Weeks of Treatment

RegressionEquation After 6 Weeks of Treatment

Control (n = 13)

12.2(LBM) - 239.0 (n = 4) 308.1 ± 29.75

1.9(LBM) - 15.3t (n = 9) 337.6 ± 12.5

High fenfluramine (n = 9)

1.70(LBM) - 5.8 (n = 4 ) 296.6 _+ 14.8

3.46(LBM) - 397.8t (n = 5 ) 303.7 ± 19.0

Medium fenfiuramine (n = 15)

3.30(LBM) - 363.7 (n = 5) 317.2 ± 20.2

3.20(LBM) - 318.5 (n = 10) 317.7 ± 14.5

High fluoxetine (n = 13)

3.46(LBM) - 414.8 (n = 4) 324.7 ± 19.3

2.21(LBM) - 163.8 (n = 9) 279.1 + 8.8

Weight-matched fenfluramine (n = 10)

1.81(LBM) + 1 9 . 1 (n = 5) 315.2 _ 15.8

1.05(LBM) - 204.0 (n = 5) 315.7 ± 12.8

Experimental Condition

* Regression equations are from only those experimental conditions where a group was sacrificed at both the 3- and 6-week time points. t Significantly different from regression equation after 3 weeks of treatment, p < 0.05. :l: Group mean body weight at 3- or 6-week time point.

772

Recently, the timing of the measurement after the administration of an injected type of anesthetic has been shown to have a dramatic effect on the predicted LBM of rats (24). It is unknown whether similar effects would be observed under conditions where an inhalant type of anesthetic was used. All animals in the studies reported here were limp and motionless when positioned on the carrier platform and measured in the TOBEC instrument. EM Scan now suggests ( 11 ) that rodents be positioned with their nose pointing into the instrument: therefore, an injected anesthetic is implicitly recommended. Subject positioning is critical because this determines the precise location of the measurement on the animal's body, thus affecting the "'extent of interaction between the sensing magnetic field and the conductive volume of the subject'" (11). Careful attention to a standard protocol, as was done in our studies, will help to control some of these factors, although further reductions in imprecision may require additional technical advancement. Both day-to day and week-to-week variability were small relative to the total variability observed among the individual rats. Thus, undependability did not contribute significantly to the observed total variance.

Accuracy At a single time point, a strong linear relationship was observed between the LBM and the EM numbers of normal animals over a wide range of LBM (see Fig. ! ). This correlation decreased when the range of LBM was restricted. Correlations such as this have been reported previously (5,9,18,22,26); however, the accuracy of regression equations during or following an experimental treatment that alters body composition has not been discussed previously. To assess accuracy requires evaluating the relationship between LBM and EM number under conditions where body composition would be expected to change. The relationship between mean LBM and EM number did not satisfy any of the three criteria stipulated of a uniform relationship across experimental conditions, within experimental conditions, or over time during a single study. These results suggest that the use of any one regression equation to predict LBM would yield inaccurate results, and would either underestimate or overestimate the LBM of animals, depending on the experimental condition. Furthermore, variation in the relationship between EM number and LBM among similar experimental conditions, either in the same or different studies, prohibits the use of a condition-specific equation to accurately predict LBM of other animals treated under this same condition. Finally, variability in the regression equation from a single experimental condition at different times during a study implies that it would be inappropriate for an investigator to generate an equation at the conclusion of a study and apply it to earlier TOBEC observations to predict the LBM of a group of animals over time. Failure to meet these criteria severely limits the usefulness of the TOBEC method for many types of experimental studies. Direct carcass analysis on at least a subset of animals is necessary for each experimental condition and at each time of interest. Other investigators have recently questioned the accuracy of TOBEC for estimating body composition in experimental animals. Baer et al. (3) noted that although the mean amount of body fat of a population of animals was similar to that predicted by regression equations, the range included several physiologically impossible values. Thus, the prediction of body composition of an individual animal is highly questionable. Bellinger and Williams (4) purposefully adjusted fat mass and total body water of animals and made sequential TOBEC measurements. They too

BI-II~I b.'l AI~.

questioned the accuracy of TOBEC and suggest that diflercnt regression equations may be necessary for rats of different sizes, strains, or genders. Factors that affect the relationship between LBM and EM number will in turn affect the accuracy of the regression equations. Such factors must be understood if the TOBEC methodology is to be useful in an experimental setting. Rumpler el al. (22) noted that the EM number was highly correlated with both body weight and LBM in normal male and female rats, over a wide weight range. In that study, body weight and LBM were also highly correlated. An examination of other previously published papers on TOBEC indicates that, regardless of the TOBEC instrument used, the correlation of EM number with body weight is consistently at least as high as the correlation of EM number with LBM ( 12,21,25). In Study V, the correlation coefficient between EM number and LBM across conditions was 0.19, and the correlation coefficient between body weight and LBM was -0.14. However, there was a strong correlation (0.76) between EM number and body weight. This is because the administration of the drugs in this study changed LBM, but these changes were independent of changes in body weight. Thus, it appears that EM number was related to body weight, but not to LBM. Because EM number is affected by the geometry of the object being measured, changes in the geometry of an animal's body due to an experimental treatment or because the animal has grown may lead to alterations in the conductive volume of the space measured, thus affecting the accuracy of prediction. To generalize a regression equation to different experimental conditions or over time, the conductivity of the section measured is assumed to be representative of the conductivity of the whole animal. In addition, the EM number is assumed to represent a constant proportion of the total conductivity across animals and over time. Under experimental conditions that would cause differences in growth or would produce differential changes in body composition, these assumptions are probably violated. This suggestion has been supported by Bellinger and Williams (4). With the advent of the EM-Scan model SA-2 instrument, EM number can be validly obtained, according to the manufacturer's documentation (I 1), either by standard anatomical positioning (fixed mode) or at the conductive center of the animal (peak mode). Fixed mode has the advantage of being faster, whereas peak mode "is slightly less sensitive to positioning" (I II. All EM numbers used in our five studies were obtained using fixed mode, because that was the only mode available in our instrument and because obtaining fast measurements was important. Because we used careful, standard anatomical positioning, and because the use of peak mode does not resolve the problems of differential conductivity within an animal due to experimental conditions, our results would not have been different had we used peak mode in the SA-2 instrument. A recent study that did use peak mode questioned the validity of TOBEC ~4t. It has been suggested that the incorporation of indices of geometry, such as animal length (10) or anthropometric measures in humans (8), into multiple regression equations may improve the accuracy of these equations. Body length was not measured in our studies. However, because adult animals were used in Study V, the key study for assessing accuracy, body length did not vary over time. Furthermore, in each study, animals were assigned randomly to the experimental conditions. Consequently, we would not expect large differences in body size among the experimental groups initially. Therefore, the addition of body length into the regression equations would not have improved their accuracy. The inclusion of body length did not improve the accuracy of regression equations in other studies (4) and the der-

T O B E C IN A S S E S S M E N T OF B O D Y C O M P O S I T I O N

ivation of the manufacturer's equations, including body length, has been questioned (3).

Summary Reliable observations can be obtained from the T O B E C instrument by careful attention to a standard protocol. The appeal of this methodology is that it is a relatively noninvasive method, and theoretically, the same animals can be assessed over time for changes in body composition. However, the TOBEC method in its current state is inaccurate. The inaccuracy stems from the lack of a direct relationship between an E M number and LBM, which results in nonuniform relationships between L B M and EM number, among

773

and within experimental treatments. Before a new method based on T O B E C can be applied, first a direct relationship between the index of conductivity and L B M must be demonstrated. Second, an understanding of the relationship between the index of conductivity and L B M in live animals must be demonstrated. Third, a complete and thorough evaluation of both the reliability and accuracy of the improved method must be completed before it can be used with confidence in experimental settings. ACKNOWLEDGEMENTS Supported in part by the Cancer Research Foundation of America, NIH Training Grant No. DK07158, and NCI Grant R01 CA34205.

REFERENCES I. Altman, D. G.; Bland, J. M. Measurement in medicine: The analysis of method comparison studies. Statistician 32:307- 317; 1983. 2. Altman, D. G.; Bland, J. M. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1:301 310; 1986. 3. Baer, D. J.; Rumpler, W. V.; Barnes, R. E.; Kressler, L. L.; Howe, J. C.; Haines T. E. Measurement of body composition of live rats by electromagnetic conductance. Physiol. Behav. 53:1195-1199; 1993. 4. Bellinger, L. L.; Williams, F. E. Validation study of a total body electrical conductive TOBEC) instrument that measures fat-free body mass. Physiol. Behav. 53:1189-1194; 1993. 5. Bracco, E. F.; Yang, M.-U.; Segal, K.; Hashim, S. A.; Van Itallie, T. B. A new method for estimation of body composition in the live rat. Proc. Soc. Exp. Biol. Med. 174:143-146; 1983. 6. Burland, W. L. A review of experience with fenfluramine. In: Bray, G., ed. Obesity in perspective. Washington, DC: US Government Printing Office; 1975:429-439. 7. Castro, G.; Wunder, B. A.; Knopf, F. L. Total body conductivity (TOBEC) to estimate total body fat of free-living birds. The Condor 92:496-499; 1990. 8. Cochran, W. J.; Won, W. W.; Fiorotto, M. L.; Sheng, H.-P.; Klein, P. D.; Klish, W. J. Total body water estimated by measuring totalbody electrical conductivity. Am. J. Clin. Nutr. 48:946-950; 1988. 9. Cunningham, J. J.; Molnar, J. A.; Meara, P. A.; Bode, H. H. In vivo total body electrical conductivity following perturbations of body fluid compartments in rats. Metabolism 35:572-575; 1986. 10. EM-Scan SA-1 small research animal body composition analyzer, operation manual; 1989. 1 !. EM-Scan Model SA-2 question and answer bulletin, October; 1990. 12. Harrison, G. G.; Van Itallie, T. B. Estimation of body composition: A new approach based on electromagnetic principles. Am. J. Clin. Nutr. 35:1176-1179; 1982. 13. Horwitz, W., ed. Official methods of analysis of the association of official analytical chemists. 12th ed. Washington, DC: Association of Official Analytical Chemists; 1975.

14. McCann, J. P.; Bergman, E. N.; Beermann, D. H. Dynamic and static phases of severe dietary obesity in sheep: Food intakes, endocrinology and carcass and organ chemical composition, J. Nutr. 122:496505; 1992. 15. McCargar, L. J.; Baracos, V. E.; Clandinin, M. T. Influence of dietary carbohydrate-to-fat ratio on whole body nitrogen retention and body composition in adult rats. J. Nutr. 119:1240-1245; 1989. 16. McShane, L. M.; Clark, L. C.; Combs, G. F., Jr.; Tumbull, B. W. Reporting the accuracy of biochemical measurements for epidemiologic and nutrition studies. Am. J. Clin. Nutr. 53:354-360; 1991. 17. Moore, D. S. Statistics: Concepts and controversies. San Francisco: W. H. Freeman; 1979. 18. Morbach, C. A.; Brans, Y. W. Determination of body composition in growing rats by total body electrical conductivity. J. Pediatr. Gastroenterol. Nutr. 14:283-292; 1992. 19. Morton, J. M.; Kirkpatrick, R. L. Comments on estimating total body lipids form measures of lean mass. The Condor 93:463-465; 1991. 20. Petbig, R. Dielectric and electronic properties of biological materials. Chichester, England: Willey; 1979. 21. Presta, E.; Segal, K. R.; Gutin, B.; Harrison, G. G.; Van Itallie, T. B. Comparison in man of total body electrical conductivity and lean body mass derived from body density: Validation of a new body composition method. Metabolism 32(5):524-527; 1983. 22. Rumpler, W. V.; Kressler, L. L.; Baer, D. L; Howe, J. C. Determination of body composition of live rats by electromagnetic conductance. Working papers from 12th Symposium on Energy Metabolism of Farm Animals. Switzerland; 1991. 23. SAS user's guide. Statistics. version 5 edition. Cary, NC: SAS Institute Inc.; 1985. 24. Tobin, B. W.; Finegood, D. T. Duration of anesthesia alters TOBEC measurement of lean body mass in rodents. FASEB J. 15(5):A960, #3426; 1991 (abstract). 25. Van Loan, M. D. Assessment of fat-free mass in teen-agers: Use of TOBEC methodology. Am. J. Clin. Nutr. 52:131-137; 1990. 26. Walsberg, G. E. Evaluation of a nondestructive method for determining fat stores in small birds and mammals. Physiol. Zool. 61(2):153-159; 1988.