72, 9 I- 10 1 ( 1984)
Application of Biplot Methods to the Multivariate Analysis of Toxicological and Pharmacokinetic Data
JANET S. SHY-M• DJESKA,’ J. EDMOND RIVIERE,“~ AND JOHN 0. RAWLINGS”
Laboratory Program, of Pharmacology and Department and Toxicology, School of Veterinary Medicine of Statistics, North Carolina State University, and Interdisciplinary Toxicology Raleigh, North Carolina 27606
Application of Biplot Methods to the Multivariate Analysis of Toxicological and Pharmacokinetic DJESKA, J. S., RMERE, J. E., AND RAWLINGS, J. 0. ( 1984). Toxicol. Appl. Pharmacol. 72, 91-101. The biplot technique was applied to aminoglycoside renal toxicological and pharmacokinetic data in beagles. The biplot obtains a two-dimensional approximation to a matrix and plots row effects and column effectsjointly, depicting relationships among different observed variables and simultaneously showing the relationship of experimental units as individuals and as treatment groups to those variables. This graphical representation of the matrix allows inspection of relationships, trends, clusters, approximate correlations, and variances existing in the data. Biplots were generated from gentamicin dosage regimen nephrotoxicity data. Six dogs classified as being intoxicated by established indicators of renal toxicity were a distinct cluster. A cluster of nonintoxicated dogs was separated into two groups approximating nephrectomized and normal dogs, thus revealing variables signihcant in separating toxic and nontoxic as well as nephrectomized and normal dogs. Biplots from pharmacokinetic data were able to separate different renal disease states on the basis of disease-induced changes in gentamicin pharmacokinetic parameters. In conclusion, the biplot technique proved to be a very useful tool in exploring this type of data by revealing clear relationships between nephrotoxicity and physiological and pharmacokinetic variables and by separating different disease states based on these data.
The biplot is a graphical multivariate statistical tool which displays a two-dimensional approximation to a data matrix of samples and variables and allows for a visual examination of the structure of the data. A single graph can depict relationships among different observed parameters, (i.e., clinical pathologic data, histopathologic scores, and pharmacokinetic constants) and simultaneously illustrate the relationships of experimental units as individuals and as treatment groups to each other and to these parameters. The technique, based on a singular value decomposition of
’ Laboratory of Pharmacology and Toxicology. * Address correspondence and reprint requests to Dr. J. E. Biviere. ’ Department of Statistics. 91
the data matrix, allows visual inspection of patterns existing in the data (e.g., relationships, trends, and clusters) as well as approximate correlations and variances (Gabriel, 197 1, 1972). Although this technique has been pub lished in some fields (Gabriel, 1972, 198 1; Strauss et al., 1979), it has not been extensively utilized in toxicology (Tepper et al., 1982) although other multivariate techniques have proved to be of value (Gad and Weil, 1982). It is the purpose of this paper to demonstrate the application of the biplot technique to toxicological and pharmacokinetic data through the analysis of the nephrotoxic potential and pharmacokinetics of the aminoglycoside antibiotic gentamicin in dogs with experimentally induced renal insufficiency. The biplot was particularly useful in analyzing a com0041-008X/84 $3.00
Copyright 0 1984 by Academic Press Inc. All rights of reproduction 1” any for”? rewn’ed
a clear relationship between nephrotoxicity and both physiological and pharmacokinetic parameters was seen. X..w.
posite set of data previously reported in independent studies (Riviere and Coppoc. an idealized biplot is presented in Fig. 1981a. Since all variables were standardized. for i = 1. 1972). respectively. the sum of squares for each is equal to its degrees of freedom. m and j = 1. dimensions.vs X2w2j.. “Total variabiity” is measured here as the sum of sum of squares for each variable. For illustrative purposes. 2.e.1. m . 3. 1982a). Row constants. ~ri). This twodimensional plot is the best two-dimensional representation of the original n-dimensional space defined by the data matrix. . The biplot is especially useful if the first two principal components account for most of the total variability:
where n = rank of matrix. in the least-squares sense. METHODS
Description of Biplot Analysis The technique of biplotting uses the singular value decomposition (SVD) of a matrix to give a graphical display of its two-dimensional approximation. If two dimensions are inadequate.
.w. etc. 2. a maximum proportion of the dispersion in the original space is retained and the rank two matrix represented by this plot is the best rank two approximation. of the original rank n matrix. 4.n.). the m X n matrix Y consisting of n variables observed on each of m experimental animals was standardized by correcting each variable (column) for its mean and dividing by its standard deviation to give variables of unit variance. Riviere. Each plotted vector represents the projection of
0. Column vectors were constructed by drawing lines from the origin (0.. i. Idealized biplot of data matrix.2. * . Rk). were then plotted jointly as tlri vs t)ri and as h. ylr. Riviere et al. 1981. X~W~. fourth.1). 1.92
SHY-MODJESKA. The approximation obtained by using the first and second singular value components of a matrix is the best possible two-dimensional approximation in the least-squares sense (Gabriel. and C xf = n(m . In spite of the complex influence of kidney disease on the detection of nephrotoxicosis and on the elimination of drugs from the kidney.b. and column constants.0
FIG. . In our analysis.
RIVIERE. This new matrix X was used asthe input matrix for Gabriel’s biplot analysis to compute the first four singular values (X3 and singular vectors
(yk.. 1. The biplot reveals the structure of the data insofar as it can be represented in two dimensions. Row vectors were left as points (Uri. k = 1. .O) to their end points (Xl Wrj. more biplots can be used to study the third. The reader is referred to Gabriel (1972) for the details of these computations.
Clusters of points would represent experimental units with similar parameter values.. 1978). Renal tissue samples were collected and prepared for light microscopy: histological sections were examined to quantib the degree of interstitial nephritis. The transformation of the data giving the largest separation among dosage regimens. A histopathologic score (0 to 20) was defined for each beagle as the sum of these individual scores. The daily elimination rate constant (Kel) was then calculated according to the formula: Kel (hr-‘) = In peak (&ml) . Pharmacokinetics. experimental units for that variable. the transformation with the highest F ratio for treatments from an analysis of variance of a randomized incomplete block design. Klz. and 14. . 1980. A second data set consisting of 8 pharmacokinetic parameters(ClB.
Sample Data Sets
culated by dividing the electrolyte clearance by the simultaneously determined CL. decreased CL. and 0 if post = pre). normalized to body weight... Kel ratio. Vd(-).g. Angles between vectors represent approximate correlations with small acute angles showing high positive correlations between variables (e.. right angles showing correlations of 0 (e.In trough (&ml) dosing interval .0 3 none
n Nephrectom&d 6 6 6 6 Normal 6 -
(hr) 8 8 16 or 24 none
a Time between successive dosesduring the 14day treatment period. tubular dilatation. Since all the vectors are of equal length. Established indicators of gentamicin toxicity were considered to be increased SCR. In addition the ratio of the first and last Kel’s was calculated: Kel( -z 14) Kel( I) Kel of Day 14 of treatment Kel of Day I of treatment
Nephrotoxicity. The second biplot (Fig. Freewater clearance was calculated by subtracting the osmolar clearance from the rate of urine production. dogs were euthanized and necropsied..5 hr
where peak and trough are defined as before.. @.. Had the variables not been standardized before the analysis. and expressed as a percentage. Both data sets were standardized before analysis. m . and a transformation on the magnitude of change of the 7 physiological parameters from pre.1 6 r C 1). i. Vs. urea and free water clearances (Cl. a and b).5 or 1. and fractional sodium and potassium clearances (CINa.1 if post-pre < 0. 24-hr endogenous creatinine clearance (CLJ. log (post &/pre CIK)... All four treatment groups contained approximately equal numbers of high and low serum creatinine. and clusters among the 29 dogs were computed by Proc CLUSTER in SAS to confirm relationships suggested in the biplots. (Riviere et al. 1979).g. Lufi ct
al. 1982. peak (30 min postdosing) and trough (immediately prior to next dose) gentamicin concentrations were determined in all dogs on Days 1.0. Perpendicular projections of the points onto any vector approximate the relative values of these. Correlations among the 9 variables were computed for both data setsby Proc CORR in Statistical Analysis System (SAS. tubular necrosis. Cronin et al.
where Ux = urinary concentration of compound X.. P. the relative lengths of the plotted vectors reflect the closeness with which they fall in this two-space.. log (post SCR/pre SCR). and large obtuse angles showing high negative correlations (e. the relative lengths would include differences in variances. and CIK.APPLICATION
a variable’s original vector onto this two-space and each poini X represents an experimental unit./pre C4A pre Cl.
The first biplot (Fig. 4. A fifth group of six dogs.-postcl. Brinker et al. was assigned to the standard dosage regimen.e. Fractional electrolyte clearances were calTABLE 1 EXPERIMENTAL DESIGN Dosageregimen DW
Treatment Standard Fixed interval h/kg) 3 1. 1982). and low KeI ratio (Schentag. and tubular regeneration by assigning values from 0 through 5 for severity of the condition.to postdrug administration. Twenty-four adult female beagles were subtotally nephrectomized (3/4 or 7/8) and assigned to one of four gentamicin dosage regimens so that mean pretreatment serum creatinine 2 weeks postsurgery was as balanced as possible among treatments (Table 1).
Clearance = i&V/P. All clearances used in the analyses were the average of two 24-hr determinations. 1981... SUN. 3) was based on the same 29 beagles and 9 variables but with the 7 physiological parameters recorded as direction of change from pre to post (+ 1 if post-pre > 0. serum urea nitrogen (SUN).-post Ck. with intact kidneys. During drug dosage. I 1. Renal clearances of all substances were calculated according to the formula:
Following the final assessment of renal function.g. preCI. CINa. e and b) (. b and c). was chosen for each variable: log (post Cb.. = serum concentration of compound X. 2) was based on the 29 dogs having complete data for all variables: histopathologic score. and log (post CI&pre Cl&.1 in n-space. Clx). post SUN/pre SUN. and V = urine flow (ml/min).. V. Renal function was assessedin all dogs by the following methods: serum creatinine (SCR).
2.40 1 1 0. = 69%. 0.40
pharmacokinetic FIG.60 +
.0 I 0. 0.0
-0. Riviere ef al.20
0. and histopathologic index.94
SHY-MODJESKA..60I I -0. Toxic dogs.-0. Biplot of transformed physiologic data.40 r I 0. A. transformations of variables. nontoxic nephrectomized dogs.
. and histopathologic index. 0.. nontoxic dogs with intact kidneys.40
0. K2. R& = 76%. samples were collected for 6 hr and analyzed for gentamicin by radioimmunoassay.
RIVJERE.60 I 1 -0. Biplot of direction of change of physiologic data.0
-0. nontoxic dogs with intact kidneys.
3. Kel ratio.
Kek/Kel 0. Kel ratio.0
0. for 23 beagles was also analyzed by the biplot technique.40 -0. Kel) and Cl. represents nontoxic nephrectomized dogs. 1981a. l . Six of the dogs which were nephrectomized (7/8.60
1 0. *. Toxic dogs. AND RAWLINGS
0.40 .20 0. The parameters obtained by nonlinear regression analysis were then fit to a two-compartment open pharmacokinetic model by methods previously described (Riviere and Cop pot.3/4) had received the standard
. 0.4a I
-0. These data were obtained from analysis of the serum gentamicin decay curve after a single iv dose of drug was administered. 1983).
primary pharmacokinetic. The subset of 18 dogs containing information on all 9 variables consisted of 8 normal. and SCR and histopathologic scores (p significantly different from 0 at p < 0. The correlation between Clk and Cln. and correlations among the four sets of variables were again computed with Proc CORR in SAS for verification of correlations shown by the biplots. SUN and SCR. 5). Because five beagles had no Cl. similarly.. (Fig. and Cl.I exist between SCR and CL. 2). Vss. and Cl. Biplot of pharmacokinetic data. Kel ratio.25 I
eters. and histopathologic score) are highly correlated among themselves but are negatively correlated to each other. and glomerulonephritis dogs with a subset of the 9 variables: C&. SCR.0prmcipal -0.50 1
component 0. Correlations approaching . normal and nephrectomized dogs.00 5
0. The group of six dogs had physiological and pharmacokinetic alterations consistent with those previously associated with aminoglycoside nephrotoxicosis and were therefore classified as “toxic” dogs.. X. and between Clk and CIN. Normal dogs. The four data sets used to generate these biplots were standardized before analysis. and histopathologic scores had a rank 2 goodness of fit of 76% (I?$) = 76%) (Fig. High positive correlations are seen between Kel and Q. Vc.50 I d 4.. Biplots were constructed for the normal dogs. 0.. 198 I). 1982a).) and those to the left of center (SUN. These pharmacokinetic variables were selected because they are the model independent. and normal. is not significantly different from 0. (Table 2). The
for 0. leaving just pharmacokinetic data to be analyzed for all 23 dogs (Fig. 4). and 4 glomerulonephritis beagles. glomerulonephritis dogs. 1981b). Vectors falling on the right-left axis (SUN. nephrectomized. A cluster analysis (Proc CORR) grouped the same six dogs into a distinct cluster and the remaining 23 dogs into another cluster. Seventeen additional dogs.o and between CIN. and 3 had catheters surgically implanted in the left lateral ventricle of the brain for determination of gentamicin concentrations in the cerebral spinal fluid (CSF) (Riviere and Coppoc.
RESULTS Nephrotoxicity The biplot analysis of transformations of magnitude of change of physiological paramFirst -0. Vectors to the right of center (Kel ratio and Cl. parameters.. nephrectomized * spinal fluid dogs. 6 nephrectomized. They appear to separate the dogs into two clusters. not part of the previous experiment. A.. Rc2) 75%.25 I animals 0.APPLICATION
dosage regimen and were part of the previously described experiment (see data above). 4 had glomerulonephritis (Riviere et al. kel ratio. 4. SCR. the 23 dogs were classified as “nontoxic” (Tables 3 and 4). this parameter was eliminated from the first biplot on this data set. histopathologic score and Kel. histopathologic score. were also utilized: 10 were considered to be normal (Riviere and Coppoc. and C&J define a primar-y axis and form a highly correlated set of variables. data. 198 1a. Riviere.
1. Normal dogs. (A) Biplot of pharmacokinetic data for normal dogs.25
I 0.40 C
I I -0.40 -0.0
-2. 0.c k I
-3.a 2 'C P 2 8 fz
C 0 0 T 0 0 O
-0. nephrectomized dogs.30 vs
V 0. and glomerulonephritis dogs. 0.10 0. 5.0 -4. 96
0.C 10.0 -1.0
2.0 I 0.0
-2 i? . nephrectomized. R.0 I -0.83%. nephrectomized dogs.0 I 0.c )-0.0 0.4 0.0
040 0.0 2.o
I I 0.: -1.f3l.00 -0.c
-1. 0.50 4.0
I I 0.5 E ‘u
1.0I1. Normal dogs. Normal dogs.0 for
0. glomerulonephritis dogs.20
3. R$) = 93%. (C) Biplot of pharmacokinetic data for normal.c
lki E 2 c
zl E x
0. R& = 93%.0 3.c -1.0
-3. A. (B) Biplot of * pharmacokinetic data for normal and nephrectomized dogs.f 0. X.30 -0. A.50 0.F~rsl prmcipal
I o 0.
the right-left axis (Kel ratio.63‘ -0.59’ -0. (When R& is low.
’ All variables transformed as defined under Methods. b NA. and normal dogs and an R$. 3). The top-bottom axis (C1n20. while most of the non-nephrectomized dogs fell in another cluster. the biplot procedure can be extended to include more dimensions.76’ 0.61’ -0. nephrectomized.51 (0. Not appropriate..82’ -0.. gave a wider separation among toxic.
. 3).07) 10.05.83’ 1.50 (0. Histo b Kel 1. SUN Cl”..67 (0.54’ -0.50’ 0..48 (0.98 (4. The nontoxic dogs are beginning to be separated into groups approximating those that were nephrectomized and those six that were not. CIHZO.83) 0. and SUN) is made up of two sets of variables highly correlated among themselves and uncorrelated to each other and separates normal from nephrectomized dogs (Fig.83 (7.29 -0. The same six dogs still grouped together in the biplot and the cluster analyses and again were classified as toxic.75’ 1.10) 27..36’ 0.16 1.53’ 0.00
top-bottom axis (Clx .52’ 1.77) 1.. Cl.1.56)
a Mean (SE).02 0. Clx and CIN.) High positive correlations exist between SUN and Clx..oo
Histo’ 0.47’ 0. histopathologic score..44’ 1.79’ -0.) is only partially correlated to the right-left axis. Ch.32 -0. ClK.44 (0.25) 40.31) 61.23) 72.42’ 1. Direction of change of physiological parameters. and CINa. and histopathologic score) is made up of two sets of variables highly correlated among themselves and negatively correlated to each other and apparently separate nephrotoxic and nontoxic dogs.00 -0. 0.APPLICATION
OF BIPLOT TABLE 2
CORRELATION COEFFICIENTS OF TRANSFORMED PHYSIOLOGICAL PHARMACOKINETIC DATA AND HISTOPATHOL~GIC INDEX
SCR” SCR CL.. when subjected to a biplot analysis.54’ -0. Clx. ’ r significantly different from 0 at p < 0.55’ -0. and Kel and C&.78’ 0. SUN and CIN. and the remaining nephrectomized (nontoxic) dogs fell in a third cluster. Correlations among these variables are near 0 and . Cl. and Cl.20 1.oo
Kel -0. This biplot was used to help generate a “nephrotoxicity index” by giving an indication of variables that significantly separate toxic and nontoxic dogs.76’ 1. -0. Cl.95 (0. and Kel ratio.51’ -0.1 with histopathologic score.45 i 0. -0.07 -0.70’ 0.54” 0. PATHOLOGIC PHARMACOKINETIC. Negative water clearance has a correlation not significantly different from 0 with SUN.86) NAb NA
Post 0.24 0.32 (22.. have correlations near . while both Kel and Cl..44’ 0.00
Cl..17 (19. Cl”. CIN. PARAMETERS IN TOXIC AND HIST~Dots (n = 6)
Parameter CL SCR SUN Cl?& ClK Kel ratio Histopathologic index
SUN 0.32 -0. ’ Histopathologic score.11)” 1.31 -0. Cl.00
Cl.. nontoxic.3 1 0.97) 0..14) 4. SCR.
= 69% (Fig.87 (0.00
0.69’ 0.27 (1.
The CSF dogs are within the cluster of normal dogs suggesting that the catheterization did not affect the kinetics of gentamicin.78” 0.46 (0. Not appropriate.13 -0.96) 0.67 (4.05.31 -0. and VC (Table 5).61’ 0.17) 1.70” 1.12 (4.83 (4.15 -0. and Kel and p.98
SHY-MODJESKA.09 (0.16 -0.08) 0.64 (0.66 (0.23) 0.60 (0.93 (0. between K. and CIB.14) NA NA Post 1. Inclusion of nephrectomized dogs gives an entirely different biplot.50 (0.22 -0.95 (0.04) 3..11) 2.91) NAb NA Post 1.26) 35.13 (8.14) 17.04) 0.91 (2.
Positive correlations exist among Vc.00
Kel 0. one sees normal dogs near the center of the biplot.76) NA NA Post 2.14) 32.13 0. Vss.66) 0.09 (0. reflecting random variation among the dogs (Fig.94’ 1. b NA.46 (0.28) 1.12) 29. The glomerulonephritis dogs tend to fall to the right of the normal dogs with slightly higher Vc.68 (0.11 (0.10) 12.19 0.32 -0.02 1.00 0.90” 1. Vd(area)y VSS.50 (0.08) 18.60)
Nephrectomized dogs receiving no gentamicin (n = 6) Pre 0.00
cLNa CL.28 (2.44) Normal dogs (intact kidneys) receiving gentamicin (n = 6) Pre 2. implying little change in their parameters.66) 0.57 (0.15 0.72 (0.93 (0.35) 0.
Kel ratio Histopathologic index
’ Mean (SE).02 0.oo
-0. Nephrectomized dogs fall to the left.00
1.82* 0.22 -0.67 (0.22) 34. Vc.30 0.81” 1.17 (5. The biplot of the eight normal dogs with respect to CL.ANDHISTOPATHOLOGICPARAMETERSFORNONTOX~CDOGS(~ Nephrectomized dogs receiving gentamicin (n = 11) Parameter Pre 0.48 (0.00
Kel Br significantly different from 0 at p < 0. Vss. have correlations not significantly different from 0 with both Kel and /3 and have negative correlations with Cle . and Cln values (R& was 75%).40” 0.35 -0.00 (2.74” 0.33)
CL.92) 0.42 (4.03) 13.11) 1.09) 5.44” 1.15) 3 1.oo
0.05 (0.10 (0.18 -0.17 (0.3 1 (0.PHARMACOKINETIC. Cln.86” 0.23) 0.
Considering pharmacokinetic data similarly (Fig.
. Normal dogs cluster together midway up the four vectors while nephrec-
TABLE 5 COWLATIONCOEFFKIENTSOFPHARMACOKINETICDATA
-0.z and & .36) NA 0.76“ 1. 4).17) 30. with low Kel and Cla values. TABLE 4
PHYSIOLOGIC. Vdc-).68 (0.32 0.85 (0.10) 20. 5A).07)” 1.70 (0. and VSS reveals no clusters or relationships. & and K2.10) 1.
This uncoupling can be appreciated from the clustering evident in the biplots (Figs. and 93%. and C1ii compared to nephrectomized dogs which have markedly reduced CIB and somewhat contracted Vssand Vc compared to normal. Although there is more variability among the glomerulonephritis dogs... in all cases presented in this study. Lufi et al. it also gives insight into relationships between toxicologic and pharmacokinetic variables as well as between individual treatment responses. A nephrotoxicity index consisting of variables separating toxic and nontoxic dogs was developed by employing the biplot in this manner. The biplot clearly suggests an uncoupling of this normal association between Cla and CL. Riviere. nephrectomy) and further. 1979.1) increased as more dogs were included in the analysis. Schentag. Pechere and Dugal. The biplot can also discriminate between the effects of different disease states on drug disposition in animals. they tend to have higher kinetic parameters and lower CL. CL. vectors when glomerulonephritis dogs are added to the input
tomized dogs cluster along the vectors in the opposite direction (Fig. It is of interest to note that the high correlation between Cl.. i. 1976. 1981.. In healthy animals aminoglycoside CIB is directly correlated to CL or to other measures of glomerular filtration rate (Chiu et al. such a table of clinical data does not also show the relationships among toxicologic variables.
however. SUN) may be confounded with variables indicative of the underlying renal insufficiency. This technique replaces the need for multiple twodimensional scatter plots or tables of correlation coefficients. Addition of the four glomerulonephritis dogs reveals another relationship. .e.g. Riviere. Rank 2 goodness of fit for those three biplots was 83. respectively. For example. DISCUSSION This study has shown that the biplot is a useful tool in exploratory analysis of the structure of toxicological and pharmacokinetic data. 1980. 1978).. Brinker et al. however. what variables separate the clusters. are similar to those reported by other investigators for aminoglycoside toxic nephropathy (Schentag. Kel ratio. 198 1.. and the Kel ratio implies that changes in drug elimination over time in this study are primarily due to the drug-induced decrease in glomerular filtration rate. 4 and 5B).. Although nephrectomized and glomerulonephritis dogs both have renal insufficiency characterized by a low glomerular filtration rate estimated by low CL. Not only does it provide graphic approximations to complex data sets. 1982a).. Glomerulonephritis dogs have higher Vss. the nephrectomized dogs have lower values for all four parameters when compared to both glomerulonephritis and normal dogs. an event discussed in the original studies (Riviere et al. The biplot was successful in separating these two states of renal insufficiency because additional variables specific for each of the conditions were utilized. 1982b)..e. The primary parameters separating toxic dogs. gentamicin. and histopathologic index. Cronin et al.. the relationships between variables demonstrated on the biplot were confirmed by independent correlation and cluster analyses. 1980. 5) because degrees of freedom (m . The biplot shows both kinds of information simultaneously..e. A biplot analysis can determine if clusters of samples exist as a result of specific disease states (e. 5B). serves as an excellent marker of glomerular f&ration. normal dogs and nephrectomized dogs differed in all pharmacokinetic variables measured and therefore appeared as two separate groups in the biplot (Figs. differences become apparent in the biplot. Vector lengths increase (Fig. In this situation the compound under study. Again. I’. This attribute is especially powerful in the situation above where variables indicative of nephrotoxicosis (i. readings than the normal dogs. SCR. 4 and 5C) or from the increased angle between the Cl* and Cl. 93. i. CL.. 1982.
Schentag. Pechere and Dugal. the parameters which make up the primary axis. However.
data matrix (Fig. 12-050300-595. only the correlation structure of homogeneous individuals within one treatment group is being displayed. note that if only relatively homogeneous data are utilized as input (Fig. as a graphical tool to discern the effects of different pathophysiological states on the parameters of drug disposition by studying the rotation of the parameter vectors induced by the addition of animals with specific disease states to the overall data matrix. techniques such as the cluster analysis can be used to verify such conditions or conclusions. Note that when vectors do not fall into the two-space being plotted. the biplot appears to be a useful tool in toxicologic and pharmacologic research as a graphical device to explore relationships between physiologic or pharmacologic variables and individual treatment responses to these variables. This information would be especially beneficial if performed in a real-time environment on a video monitor where vector displacement could be quantified and highlighted.. they are short vectors and the angles between them may not be good reflections of their correlations. Another application for the biplot is in determining whether individual animals can be classified as normal in terms of more than one parameter. William W. Purdue University in West Lafayette. Although it is usually evident in a biplot analysis when only random variation is present. can be selected. distribution rate constant vectors are perpendicular to the Kel vector indicating no correlation. Richard Rogers of the Laboratory of Pharmacology and Toxicology for assistance in graphical preparation of the biplots.
RIVIERE. relationships between parameters may not be very meaningful.
. Biologically significant variation must be present if meaningful interpretations are to be made by this technique. The relationships seen between pharmacokinetic parameters in Fig. for their support in the original studies from which these data originated. Dr. 5C vs Fig. 1977). This is implicit to their derivation. the dogs involved in the CSF experiment cluster with the normal
dogs in the biplot (Fig. 4 are consistent with theoretical assumptions (Gibaldi and Perrier. Its use is analogous to that of scattergrams and residual plots in regression analysis (Chattejee and Price. The biplot would be useful in toxicologic studies utilizing normal animals to determine relationships between independent toxicologic parameters. Coppoc. 1980. 1975. namely.e. In this case. In summary. 5A). Hinsman of the School of Veterinary Medicine. O’Flaherty. The use of these parameters in subsequent toxicologic studies should increase the efficiency and power of the experiments. Wagner. i. The biplot is particularly useful as an exploratory tool and is not intended for statistical test of hypotheses. We acknowledge Mr. 1982). 4). 5B). Gordon L. Carlton. 1982. In addition. Edward J. 1981. 1983). Portions of this work were presented in abstract form at the Annual Meeting of the American Society of Pharmacology and Experimental Therapeutics in Philadelphia. This illustration suggests a powerful use for the biplot in pharmacokinetic studies. the optimal subset which discriminates between toxic and nontoxic animals. by generating sequential biplots composed of subsets of all measured parameters. Pennsylvania on August I 1. This work was supported in part by the USDAES No. lending support to the assumption that the surgical procedure involved in implanting ventricular cannulas did not alter drug disposition as compared to historical controls. Indiana. The above mentioned disease states can be compared and contrasted easily from the biplot. the K12 and &. For instance.100
SHY-MODJESKA. Associations or hypotheses suggested by the biplot should be verified with independent sets of data to insure the results are not unique to that set of data. 232. Biologically significant variation in the data should be present before the biplot is applied. 1979. For example. and outliers readily noted. 1983 (Pharmacologist 25. ACKNOWLEDGMENTS
The authors thank Dr. conclusions drawn. and Dr.
S. Bamett. 64. (198 I). AND PERRIER. K.. E. Proc.. Chemotherapy 21. eds. glycoside body clearance in renal disease when volume of distribution increases. J. E. S. R. F. Wiley. Amer.. STAREY. R. 309-312. W. The biplot graphic display of matrices with application to principal component analysis. Natural history of aminoglycoside nephrotoxicity in the dog. New York. COPPOC. M. J. Drug Intelligence Publications. J. O’FLAHERTY.. Pharmacol. D. 147173. 174-209.. J. ed. J. R. M. 4. LUFT. A. Species dependent gentamicin phannacokinetics and nephrotoxicity in the young horse. 98. CRONIN. Comparative nephrotoxicity of fixed-dose versus fixedinterval reduction of gentamicin dosage in partially nephrectomized dogs. R. Appl. W. 95. (198 1). E. R.. C. 273-320.APPLICATION
BRINKER. AND MAXWELL.S. AND CARVER. B. Biometrika 58.. AND COPPOC. Wiley. W. HINSMAN.G.. pp. G.. J. CARLTO~. pp. (1983).G.. K. Neu.J.4pp/.R. D. R. 277-282. E. R. C. 1071-1077. E.. WAGNER.rJl Appl. W. AND TRAVER. PECHERE. R. Paradoxical increase in amino-. CARLTON. R. RIVI~RE.. E. A. W. (1977). Antimicrob. BLOCK. pp. Intersci. Res. Hayes. AND TARANA.). E. Med. E. TEPPER. E. Pharmacokinetics. E. Med. COSTELLO. (198 1). Microanalysis of ozone depression of motor activity. R. (1980). To. AND CRONIN. Inc ~ Hamilton. WEISS. Pharmacokin. D~BYAN. Tovi<. Analysis of meteorological data by means of canonical decomposition and biplots. Pharmacoi. 141167. (1978). Effect of potassium depletion on gentamicin nephrotoxicity. Dis. C. 167: 105-113. (1981). HENRICH. B. K. L. Dekker. L. 1979.. SOUTHERN.
. R. J. G.. 42. Antimicrob.463-474.D.. AND HENRICH.. 387. P. 10. GABRIEL. Applied Therapeutics. L. 2nd ed. Statistics for toxicologists. YUM. Clin. GABRIEL.. (1979). Inc. CHIU. Injtict... (1979). J. J. Pathol. In Principles and Methods of Toxicology (A. AND CARLTON.. Dekker. Agents Chemother. Determination G. Clinical pharmacokinetics of aminoglycoside antibiotics. Do psychiatric patients fit their diagnoses? Patterns of symptomatology as described with the biplot. RIVIERE. KOKES. Chem. L. In Applwd Pharmacokinetics (W. W. R. J. J.. S. New York. Statistical Analysis System Institute Inc. Meteor.).. Jusko. BULGER.ricants and Drugs: Kinetics and Dynamics. 3 1-42. Biplot display of multivariate matrices for inspection of data and diagnosis.. S. . B. L.. Gentamicin pharmacokinetic changes in induced acute canine nephrotoxic glomerulonephritis. Pharm. 720-72 I RIVIERE.. W. W. Res. Comparative nephrotoxicity of aminoglycoside antibiotics in rats.). Clin. (1975).. Evans. AND LONG..J. J. New York. New York. AND WEIL. J. CHATTERJEE.G. C. 138. R. L. Regression Analysis by Example.. (in press). SCHENTAG. AminogJycoside pharmacokinetics as a guide to therapy and toxicology.. LANTZ.C. M.r. 292-301. 541-545.
of cerebrospinal fluid gentamicin in the beagle using an indwelling cerebral ventricular cannula. R. In The Aminoglycosides (A. C. (1982). G. C/in. New York. GABRIEL. (1982). pp. Whelton and H. 11.
RIVIERE. J. STRAUSS. SOUTHERN. Agents Chemother. SAS User’s Guide. S. Wiley.. 170. Commun. 22. Toxicol. RIVIERE. C. BROWN. E. S. AND Cox. (1982). VANORD. Lab. Schentag. J. E.. Lab. SLOAN. ed. (198 I b).. HINSMAN.. Vet. F. 74. Renal extraction of gentamicin in anesthetized dogs. (198213). Conf Antimicroh Agents Chemother. A. ( 1982). N. J. ( 1976). K. COPFQC.. E. 20.. D. Fundamentals of Clinical Pharmacokinetics. (1982a).. M. W. GIBALDI. Limitations on the physiologic interpretation of aminogiycoside body clearance derived from pharmacokinetic studies.199. Pharmacokinetics of gentamicin in the juvenile dog. F. AND PRICE. J.
392.. Mental Di. C. R. E. P. J. K. eds. J. 71. (1971). Raven Press. J. GABRIEL.. (1980). AND COPPOC. San Francisco.. J. J. New York. S. 38. and W. MILLER. Aminoglycosides. J. Nerv. M..). AND DUGAL. RIVIERE. (198 la).. RIVIERE. R. COPPOC. C. J. J. J. 458-467. P. J. N.. Fundam. BULGER. E. Raleigh. E. III. P. GAD. SCHENTAG. Sci. 3 17-326.. In Interpreting Multivariate Data (V. J. 1621-1623.. J. (1972). L. RITZLF. L. (1982). HINSMAN.. T. E.