Download 40.58 Kb.
Multivariate analyses reveal common and drug-specific genetic influences on responses to four drugs of abuse
John K. Belknap, Pamela Metten, Ethan H. Beckley and John C. Crabbe
Research Service, Veterans Affairs Medical Center, Department of Behavioral Neuroscience and the Portland Alcohol Research Center, Oregon Health & Science University, Portland, OR 97239, USA
Corresponding author: Crabbe, J.C. (firstname.lastname@example.org).
Introduction: The testing methods, choice of doses, and experimental design for the ethanol (E), morphine (M), pentobarbital (P) and diazepam (D) studies have been described in detail in prior papers 1-10 and are summarized in main text, Table 1. Therefore, only a brief summary is presented here. Because our plan from the outset was to make comparisons among drugs once all four drugs had been tested, we designed the testing regimen to include a number of traits appropriate for all four drugs, and doses were chosen to induce a similar range of responses whenever possible across the same 14-15 inbred strains. Some traits appropriate to only a subset of these four drugs were also studied in these earlier reports; these are not addressed here (see main text, Table 1).
Animals. Male mice from the following inbred strains were purchased from the Jackson Laboratory, Bar Harbor, ME at 5 to 6 weeks of age: A/HeJ, AKR/J*, BALB/cJ, CBA/J, CE/J, C3H/HeJ, DBA/1J, DBA/2J, C57BL/6J, C57BR/cdJ, C57L/J, PL/J*, SJL/J*, SWR/J*, and 129P3/J. Four of these strains are believed to have been derived from Mus musculus domesticus (shown by an asterisk); the others from a different subspecies, Mus musculus musculus 11,12. Brief descriptions of their origin and characteristics are available 11,12. Housing was in Thoren mouse shoebox caging with corn cob bedding; ambient temperatures were 21-23oC throughout. All procedures followed USDA and NIH guidelines for the care and use of laboratory animals and were approved by our local IACUC.
Activity and thermal response testing. Previously-published procedures were employed for testing of the four drugs 3-6. All mice were males of 6-8 weeks of age at time of testing. Briefly, one hr prior to testing, all animals were moved into the room where testing would occur. Each animal was restrained briefly in a Plexiglas tube and a lubricated 0.5 mm diameter probe was inserted 2.5 cm into the rectum. A baseline temperature was recorded 5-10 sec later using a rapid response sensor (Sensortek Thermalert Th 8 Digital Thermometer). Each animal was injected with one of three or four doses of each drug (i.p., in 0.9% NaCl) or the saline vehicle, and placed immediately into the center of an illuminated, sound-attenuated Omnitech Activity monitor. Infrared beam interruptions in the horizontal plane were recorded automatically each 5 min for 30 min, yielding a measure of horizontal distance traversed for each animal (A). When removed from the activity chamber, a single post-injection temperature was taken 30 min after injection. For each dose (trait), the saline control group values were subtracted from the drug administered groups (between groups correction). Separate groups of animals of each strain were administered doses of 1, 2 or 3 g/kg of E, 10, 20, 30 or 40 mg/kg of P, 2, 4, 8 or 16 mg/kg of D and 4, 8, 16 or 32 mg/kg of M. Mice tested with any one drug, as well as the saline groups, were run concurrently whenever possible, but testing for different drugs was spaced over a several-year period.
Two-bottle choice consumption. Detailed methods have been published 8,9. Preference drinking testing began at 9-11 weeks of age using mostly mice previously given saline as part of the activity/thermal response studies after at least one week of rest. Generally eight mice per strain (except DBA/2J, N=12; C57BL/6J, N=12) were housed singly in standard shoebox cages and provided with two inverted 25 ml graduated cylinders as drinking bottles in each home cage, one containing water and the other a drug solution. Daily fluid consumption was recorded at 2-3 hrs after light onset on a 12:12 L:D cycle. The position of the two bottles alternated at each concentration change, and fresh fluids provided. Drug concentrations in the drug bottle were 3, 6 and 10% of E; 25, 50 and 100 mg/100 ml of P; 5, 10 and 20 mg/100 ml of D; and 5 mg/10 ml of M. For each drug, these concentrations were presented serially for four days per concentration to each mouse; however, each mouse was exposed to only one of the four drugs. Body weights were monitored each week and food was available ad libitum.
We also included in our multivariate analyses published data from two-bottle choice studies 14-18 of a number of tastants (vs water), including sweet (sucrose, 1.7%; saccharin, .033%; glycine, 1.5%) and bitter (phenylthiocarbamide or PTC, .01%; cycloheximide, 1µM; quinine, .03%) on mice of 12-13 the same strains we used in our studies. We deemed this of interest because two of the four drugs we studied (M, P) are reported to have a bitter taste, at least to humans. In contrast, ethanol may have a mildly sweet taste 19-22. For comparison with these data, we also included our saccharin preference data at a 6-fold higher concentration (.2%, SAC2) pooled across two different studies 8,9 for these same strains.
Withdrawal severity. Acute withdrawal was induced by a single large dose of each of the four drugs (EWD, PWD, DWD, MWD) in additional groups of inbred mice. One group of naive mice was tested for both ethanol and pentobarbital withdrawal severity one week apart 23. For both drugs, withdrawal signs (handling-induced convulsions, or HICs) were monitored hourly as the drug was eliminated from the body. Separate groups of naive mice of these same 14-15 strains were tested for either diazepam or morphine withdrawal. For diazepam, a single large dose (20 mg/kg, i.p.) was given, HICs were assessed to establish that they were depressed, then one hour after diazepam injection, withdrawal was precipitated by an i.p. injection of 10 mg/kg flumazenil, a benzodiazepine receptor antagonist. HICs were assessed over the next 12 minutes; the short interval was due to the very short-acting antagonism induced by flumazenil 2. For morphine, a single large dose of 200 mg/kg s.c. was given in a sustained release preparation. Five hrs later, withdrawal was precipitated by an i.p. injection of 12 mg/kg of naloxone, and opioid receptor antagonist 10. Because HICs are not observed during morphine withdrawal, we used the number of jump responses as the index of withdrawal severity, the most common index reported in the literature for mice 24. In contrast, the handling-induced convulsion (HIC) was used to quantify withdrawal severity for the other three drugs studied. In addition, a chronic model based on 3 days of ethanol vapor inhalation was used for E (EWDC) in a separate group of naive mice per strain. HICs were monitored hourly 0-10 and 24, 25 hrs after withdrawal from the vapor chambers 7.
Genetic Correlational Analyses. Since all members of an inbred strain are genetically identical for all practical purposes, all variation within a strain is environmental in origin, while variation among strains can be presumed to be predominantly genetic in origin. Because of this, correlations (rG) between pairs of traits based on strain means for each trait are estimates of genetic correlations (within strain variability of mostly environmental origin is ignored); these index the degree of genetic similarity or codetermination between trait pairs 25. These are presented in the genetic correlation matrices shown in Supplementary Tables 1, 2 and 3. Each dose or concentration of each drug was considered to be a separate trait for analysis, which resulted in a total of 45 traits in this study, comprised of 15 activity, 15 thermal response, 10 voluntary consumption and 5 withdrawal traits. In addition, data from two-bottle choice studies of sweet or bitter tastants were added to the analysis to determine whether these strain differences resembled two-bottle choice data for E, M, D or P. This brought the total number of variables to 52 and the total number of possible genetic correlations to 1,326.
The underlying principles for these analyses are discussed in more detail elsewhere 25. The goals of the genetic correlational analyses were essentially exploratory, that is, to generate new hypotheses about the relationship among traits suitable for further testing rather than to test such hypotheses. The latter will require independent confirmation studies in other populations. In the past, we adopted the p < .05 level for initial exploration, which was r > |0.51| for n = 15 strains in order to maintain enough power to detect general relationships [e.g., 6]. The drawback is that an estimated 27% of correlations attaining p < .05 in this study are estimated to be false positives because of multiple testing as estimated by the false discovery rate (FDR) method (see below). However, this problem is largely (but not entirely) avoided in the present study because our focus is not on individual correlations, but on broad patterns involving several to many correlations throughout this paper. Thus, the units of statistical analysis are groups (clusters) of correlations, not single correlations, with very few exceptions as noted below. Another reason why this approach is important is that we have more variables (52) than we do strains (15); looking at groups of variables rather than individual variables greatly mitigates this problem because there are many-fold fewer groups than there are individual variables.
However, if the reader is interested in the significance of any one correlation, a correction for multiple testing using the false discovery rate (FDR) method is useful. The FDR is the probability of a false positive among all correlations declared to be significant, which is very different from the traditional (and overly conservative) method based on the probability of a false positive among all correlations in the matrix, e.g., a Bonferroni correction 26. With the significance level set at FDR <.05, this corresponds to p <.002 estimated by qvalue software (http://faculty.washington.edu/~jstorey/qvalue/ ) based on the entire matrix (all 1,326 correlations). This significance threshold is attained with correlations r > |0.73| with n =15 strains. Correlations exceeding this threshold have an estimated 5% chance of being a false positive.
Multivariate analyses. The genetic correlation matrices were subject to exploratory multivariate analysis using multidimensional scaling [MDS, 27] implemented using Systat, v. 11 software (SPSS, Inc., Evanston, IL) using the monotonic option. Single linkage cluster analysis was also performed on the same matrices for comparison purposes only for the Supplementary Table 1 data using the same software. These plots are shown in main text Figures 1A, 1B, 2 and 3. MDS constructs a map where each trait is plotted according its position on two linear coordinates (dimensions), here referred to as Dimensions 1 and 2 selected because they together account for the maximum amount of the covariance present in each correlation matrix compared to all other possible dimension pairs. Each dimension has been standardized, thus it has a mean of zero and an SD of 1. Dimension 1 was constructed to account for the largest amount of the total covariance in the correlation matrix. Dimension 2 was constructed to be independent (orthogonal) to Dimension 1 and to account for the largest amount of covariance when combined with Dimension 1 27,28. The relative linear (Euclidean) distances between plotted points, as they range from very small to very large, reflect genetic similarity ranging from strongly positive correlations at small distances to no correlation to strongly negative correlations when two traits map to opposite sides of the MDS plot. Note that the genetic distance (linear difference) between any pair of plotted traits depends not only on the single correlation value between that pair across the 14-15 strains (bivariate case), but also on the similarity of correlations between that pair and all of the other traits in the correlation matrix (multivariate case). The resulting MDS plots contain some error due to the effects of collapsing what is essentially a 30-dimensional space (when there are 30 traits as in Supplementary Table 1 and main text Figure 1A) down to only two dimensions for plotting, resulting in some loss of information contained in the original correlation matrix. However, the two-dimensional plots shown in main text Figures 1A, 2 and 3 captured 86%, 83% and 89% of the total covariation in the correlation matrices shown in Supplementary Tables 1, 2 and 3, respectively. This is a good outcome considering that all traits in a given category were included rather than only selected ones. For this reason, we did not resort to higher dimensionality to increase the percent of covariance accounted for.
MDS is conceptually similar to cluster analysis in that it seeks to depict which traits or variables are closely related and which are not, but the output is different in that MDS uses a two-dimensional plot of similarities represented by linear distance while cluster analysis uses a tree diagram (dendrogram) for this purpose. For comparison, we present the outcome of cluster analysis (single linkage, main text Fig. 1B) to compare to the MDS plot shown in main text Fig 1A, both based on the same genetic correlation matrix. Traits closely similar in their pattern of strain differences are connected by branches to the left (small distances; X axis), while relatively dissimilar traits are connected to the right (large distances) of the tree diagram. The distances shown are equal to 1 – r, where r is the correlation between the nearest members of a pair of clusters. MDS was chosen as the primary method because it typically accounts for more of the total covariation and the output is better suited to subsequent statistical analysis (see below).
The MDS map positions in two coordinates (x and y for Dimensions 1 and 2, respectively) of traits generated by MDS allow statistical tests of differences in location (reflecting genetic similarity) between groups (clusters) of traits. In most cases, a one-way ANOVA by drug was first performed to determine whether the four drugs differed significantly in map position for either dimension. For example, four doses per drug and four drugs would involve a total of 16 data points (expressed as either x or y for each trait), thus the F ratio for ANOVA would then have df =3 (between) and df =12 (within). Tukey HSD post hoc tests were then used to compare each drug with the other three 29. When a single trait compared to a cluster of traits was of interest, we used an outlier test [Grubb's test 29 ] to determine whether a single trait mapped significantly outside of a cluster. In all cases, only significant outcomes at p< 0.01 are mentioned in the text.
An alternative to the approach taken in this paper is to synthesize a new dimension based on Euclidean distances between groups or clusters of traits which should be somewhat more powerful than using only Dimension 1 or only Dimension 2 (the original MDS dimensions) for this purpose. In this case, the distances for a new linear dimension (z) are calculated for each trait derived from Dimensions 1 and 2 coordinates jointly as follows: z1 - z2 = √[ (x1 - x2)2 + (y1 - y2)2] 28, where x and y refer to Dimension 1 and Dimension 2 coordinates per trait, respectively, for group (or cluster) 1 and 2 (denoted by subscripts). The coordinates for each trait are then calculated in reference to this new linear dimension (z) that is no longer strictly horizontal (Dimension 1) or vertical (Dimension 2), but at an angle between the two. We do not report these results because the basic conclusions reported in the main text were essentially the same with either method, and using the original MDS dimensions is much simpler to present and interpret.
Supplementary Table S1. Genetic correlation coefficients (rG) among all activity (A) and thermal response (T) traits for the four drugs (ethanol, E; diazepam, D; pentobarbital, P; morphine, M) across 14-15 inbred strains. The 2nd character in each trait symbol refers to dose in g/kg (E only) or mg/kg (all others) given i.p. Correlations in bold are p <.05 and are shown only to highlight apparent clusters. The strain means upon which this matrix was derived are given in the original publications (see text).
Supplementary Table S2. Genetic correlation coefficients (rG) among all two-bottle choice voluntary consumption measures for the four drugs across 14-15 inbred strains (ethanol, E; diazepam, D; pentobarbital, P; morphine, M). The 2nd character in each trait symbol refers to drug concentration (see text). In addition, six tastants representing sweet (sucrose, SUC; saccharin, SAC; glycine, GLY) and bitter (phenylthiocarbamide, PTC; cycloheximide, CYC; quinine, QUI) were included based on published preference drinking data for these same strains (see text). Correlations in bold are p<.05 and are shown only to highlight apparent clusters. The strain means upon which this matrix was derived are given in the original publications from our group (see text).
Supplementary Table S3. Correlation coefficients (rG) among all two-bottle choice voluntary consumption measures (C) for four drugs (ethanol, E; diazepam, D; pentobarbital, P; morphine, M) across 14-15 inbred strains and all withdrawal severity (WD) traits for these same drugs. The 2nd character in each consumption trait symbol refers to drug concentration in percent (E only) or mg/10ml (all others). For ethanol, both acute withdrawal (EWD) and chronic withdrawal (EWDC) data were included from separate groups of mice. Correlations in bold are p<.05. The strain means upon which this matrix was derived are given in the original publications from our group (see text).
Supplementary Table S4. Overall means (±SEM) collapsed across strain for activity (counts) and thermal responses (oC) to morphine (M), ethanol (E), diazepam (D) and pentobarbital (P) are expressed as the difference from the corresponding saline group values (N =117-128 mice per dose). Scores for particular strains often differed greatly from these average values. Doses for each drug in mg/kg (except E, g/kg) are given within brackets.
Activity: M E D P
698 ± 116 [ 4] 1812 ± 95  896 ± 123  2428 ± 101 
504 ± 139  1920 ± 211  670 ± 208  5710 ± 295 
821 ± 255  1071 ± 123  1255 ± 309  1662 ± 356 
2832 ± 325  767 ± 248  1679 ± 123 
Temp: M E D P
0.30 ± 0.063  0.05 ± 0.034  0.60 ± 0.047  0.17 ± 0.030 
1.14 ± 0.104  1.26 ± 0.082  1.22 ± 0.078  0.01 ± 0.053 
2.32 ± 0.149  2.77 ± 0.075  1.40 ± 0.087  1.04 ± 0.102 
2.91 ± 0.214  1.79 ± 0.062  2.62 ± 0.142 
1 Metten,P. et al. (1998) High genetic susceptibility to ethanol withdrawal predicts low ethanol consumption. Mamm Genome 9, 983-990
2 Metten,P. and Crabbe,J.C. (1999) Genetic determinants of severity of acute withdrawal from diazepam in mice: commonality with ethanol and pentobarbital. Pharmacol Biochem Behav 63, 473-479
3 Crabbe,J.C. et al. (1998) Genetic determinants of sensitivity to diazepam in inbred mice. Behav Neurosci 112, 668-677
4 Belknap,J.K. et al. (1998) Genetic determinants of morphine activity and thermal responses in 15 inbred mouse strains. Pharmacol Biochem Behav 59, 353-360
5 Crabbe,J.C. et al. (1994) Genetic determinants of sensitivity to ethanol in inbred mice. Behav Neurosci 108, 186-195
6 Crabbe,J.C. et al. (2002) Genetic determinants of sensitivity to pentobarbital in inbred mice. Psychopharmacology 161, 408-416
7 Metten,P. and Crabbe,J.C. (2005) Alcohol withdrawal severity in inbred mouse (Mus musculus) strains. Behav Neurosci 119, 911-925
8 Belknap,J.K. et al. (1993) Voluntary consumption of morphine in 15 inbred mouse strains. Psychopharmacology 112, 352-358
9 Belknap,J.K. et al. (1993) Voluntary consumption of ethanol in 15 inbred mouse strains. Psychopharmacology 112, 503-510
10 Metten,P. et al. (2008) Genetic correlates of morphine withdrawal in 14 inbred mouse strains. Drug Alc Depen, DOI:10.1016/j.drugalcdep.2008.07.006
11 Festing,M.F.W. (1998) Inbred strains of mice and their characteristics. INTERNET http://www.informatics.jax.org/external/festing/mouse/INTRO.shtml
12 Wade,C.M. et al. (2002) The mosaic structure of variation in the laboratory mouse genome. Nature 420, 574-578
13 Beck,J.A. et al. (2000) Genealogies of mouse inbred strains. Nat Genet 24, 23-25
14 Lush,I.E. (1989) The genetics of tasting in mice. VI. Saccharin, acesulfame, dulcin and sucrose. Genet Res 53, 95-99
15 Lush,I.E. (1984) The genetics of tasting in mice III. Quinine. Gen Res Cambridge 44, 151-160
16 Lush,I.E. (1986) Differences between mouse strains in their consumption of phenylthiourea (PTC). Heredity 57 ( Pt 3), 319-323
17 Lush,I.E. and Holland,G. (1988) The genetics of tasting in mice. V. Glycine and cycloheximide. Genet Res 52, 207-212
18 Lush,I.E. et al. (1995) The genetics of tasting in mice VII. Glycine revisited, and the chromosomal location of Sac and Soa. Genet Res Cambridge 66, 167-174
19 Overstreet,D.H. et al. (1993) Saccharin intake predicts ethanol intake in genetically heterogeneous rats as well as different rat strains. Alcohol Clin Exp Res 17(2), 366-369
20 Blizard,D.A. and McClearn,G.E. (2000) Association between ethanol and sucrose intake in the laboratory mouse: exploration via congenic strains and conditioned taste aversion. Alcohol Clin Exp Res 24, 253-258
21 Lemon,C.H. et al. (2004) Alcohol activates a sucrose-responsive gustatory neural pathway. J Neurophysiol 92, 536-544
22 Dess,N.K. et al. (1998) Ethanol consumption in rats selectively bred for differential saccharin intake. Alcohol 16, 275-278
23 Metten,P. and Crabbe,J.C. (1994) Common genetic determinants of severity of acute withdrawal from ethanol, pentobarbital and diazepam in inbred mice. Behav Pharmacol 5, 533-547
24 Kest,B. et al. (2002) Naloxone-precipitated withdrawal jumping in 11 inbred mouse strains: evidence for common genetic mechanisms in acute and chronic morphine physical dependence. Neuroscience 115, 463-469
25 Crabbe,J.C. (1999) Animal models in neurobehavioral genetics: Methods for estimating genetic correlation. In Neurobehavioral Genetics: Methods and Applications (Mormede,P. and Jones,B.C., eds), pp. 121-138, CRC Press
26 Storey,J.D. and Tibshirani,R. (2003) Statistical significance for genomewide studies. Proc Natl Acad Sci USA 100, 9440-9445
27 Kruskal,J.B. and Wish,M. (1978) Multidimensional Scaling, Sage Publications
28 Manly,B. (1986) Multivariate Statistical Methods, Chapman and Hall
29 Sokal,R.R. and Rohlf,F.J. (1995) Biometry, Freeman