Chromosomal loci influencing chronic alcohol ... - Springer Link

2 downloads 0 Views 209KB Size Report
... Section of Neurobiology, University of Texas, Austin, Texas 78712, USA .... mapmgr.roswellpark.org) by using the interval ..... Kenneth Manly (Roswell Park Institute) for making ... Belknap JK, Danielson PW, Lame M, Crabbe JC (1988).
Chromosomal loci influencing chronic alcohol withdrawal severity Susan E. Bergeson,1 R. Kyle Warren,1 John C. Crabbe,2 Pamela Metten,2 V. Gene Erwin,3 John K. Belknap2 1

Waggoner Center for Alcohol and Addiction Research, Section of Neurobiology, University of Texas, Austin, Texas 78712, USA Portland Alcohol Research Center, Research Service (R&D5), Veterans Affairs Medical Center, and Department of Behavioral Neuroscience, Oregon Health & Science University, Portland, Oregon 97201, USA 3 School of Pharmacy (C238), University of Colorado Health Sciences Center, Denver, Colorado 80262, USA 2

Received: 22 November 2002 / Accepted: 3 March 2003

Introduction

Abstract

Ethanol (alcohol) withdrawal-induced convulsions are a key index of physical dependence on ethanol and a clinically important consequence of alcohol abuse in humans. In rodent models, severity of withdrawal is strongly influenced by genotype. For example, many studies have reported marked differences in withdrawal severity between the WSR (Withdrawal Seizure Resistant) and WSP (Withdrawal Seizure Prone) mouse strains selectively bred for over 25 generations to differ in chronic withdrawal severity. Therefore, we used an F2 intercross between the inbred WSP and WSR strains for a genome-wide search for quantitative trait loci (QTLs), which are chromosomal sites containing genes influencing the magnitude of withdrawal. We also used the recently developed HW, RHW (high withdrawal) and LW, RLW (low withdrawal) lines selectively bred for the same trait and in the same manner as the WSP, WSR lines. QTL analysis was then used to dissect the continuous trait distribution of withdrawal severity into component loci, and to map them to broad chromosomal regions by using the Pseudomarker 0.9 and Map Manager QT29b programs. This genome-wide search identified five significant QTLs influencing chronic withdrawal severity on Chromosomes (Chrs) 1 (proximal), 4 (mid), 8 (mid), 11 (proximal), and 14 (mid), plus significant interactions (epistasis) between loci on Chr 11 with 13, 4 with 8, and 8 with 14.

3 Correspondence to: J.K. Belknap, Research Service (R&D5), VA Medical Center, Portland, OR 97239, USA; E-mail: belnajo@ ohsu.edu

454

Convulsions are a well-known consequence of withdrawal from alcohol in physically dependent individuals and are one of the most feared manifestations of the alcohol withdrawal syndrome in alcoholics. Seizure or convulsive activity (tonic and/or tonic-clonic) are hallmark signs of central nervous system (CNS) hyperexcitability characteristic of withdrawal from alcohol in physically dependent organisms. These typically become manifest after alcohol is withdrawn and often reach their peak intensity shortly after blood (and brain) alcohol concentrations are at or approaching zero. The withdrawal syndrome is often considered to be a ‘‘rebound’’ phenomenon. During extended alcohol exposure, compensatory (homeostatic) processes develop in the CNS that lead to the development of tolerance in the presence of alcohol. When the alcohol is withdrawn, these previously compensatory neuroadaptive processes (tolerance) now become overcompensatory (rebound), leading to the CNS hyperexitability signs characteristic of the withdrawal syndrome (Metten and Crabbe 1996). Withdrawal convulsions are the most quantifiable and reliable index of physical dependence in the mouse. Because other signs of withdrawal are highly genetically correlated with convulsive activity (Kosobud and Crabbe 1986; Belknap et al. 1987, 1988), a common practice is to use convulsive activity as the primary index of withdrawal syndrome severity. The most often used index of convulsive activity in the mouse is the handling-induced convulsion (HIC), so called because it is elicited by picking up a mouse by the tail (Metten and Crabbe 1996). A rating scale is used to rate HIC severity, from mild clonic to more severe tonic-clonic convulsions (Crabbe 1998). These are seldom lethal or debilitating, so multiple measurements are routinely

DOI: 10.1007/s00335-002-2254-4 • Volume 14, 454–463 (2003) •  Springer-Verlag New York, Inc. 2003

S.E. B ERGESON

ET AL :

QTL S

FOR

455

A LCOHOL W ITHDRAWAL

made on each mouse at hourly intervals (Terdal and Crabbe 1994). The last decade has seen the development of several methods designed to search the genome for individual genes influencing a particular genetically complex (polygenic) trait. The best known of these are gene expression microarrays, random mutagenesis, and quantitative trait locus (QTL) analyses. The present paper employed QTL detection and mapping to search for chromosomal regions containing genes that have an effect on a trait of interest. With this approach, several markers are genotyped on each mouse chromosome, and any marker whose genetic variation significantly covaries with trait variation detects the presence of a closely linked QTL. This project made use of presently existing, selectively bred lines for ethanol withdrawal sensitivity because the alleles (genes) predisposing toward very high and very low ethanol sensitivity have essentially been isolated in these oppositely selected lines, making them and their crosses a valuable resource for gene mapping studies. These were the WSP (Withdrawal Seizure Prone) and WSR (Resistant) lines bred respectively for high and low ethanol withdrawal intensity (Crabbe et al. 1985). The WSP and WSR lines were selectively bred starting with the HS (heterogeneous) stock, which in turn was derived from an eight-way cross of inbred strains (McClearn et al. 1970). Thus, the pool of alleles from which selection can fix those with the largest phenotypic effects is much larger than a typical F2 cross between just two inbred strains. The WSP and WSR lines are at their maximum selection response (the selection limit), indicating that high withdrawal-disposing alleles have been largely fixed in the WSP lines, and low disposing alleles in the WSR lines. This was shown by relaxation of selection (random mating) after the 26th selected generation, which did not result in changes in trait scores toward intermediate values in any of the four lines (Crabbe and Phillips, 1993). Lander and Botstein 1 (1989) have argued that selection lines near the selection limit are especially efficient and effective gene mapping tools when the alleles at the QTLs affecting the trait under selection have been fixed for alternate alleles in the oppositely selected lines. This selection project was a replicated experiment; thus, there are two WSP (replicates 1 and 2) and two WSR (1 and 2) lines, which we shall refer to as P1, P2, R1, and R2 hereafter. In the foundation HS/Ibg population from which the lines were developed, the assignment of mice to each replicate was largely at random; thus, there is no difference between P1 and P2 lines at their inception other than sampling effects. The selected lines were systemi-

cally inbred beginning at the 26th generation of selection. The fully inbred IWSP2 and IWSR2 strains were intercrossed to produce the IP2xIR1 F2 population used in the present work. The Colorado HW (high withdrawal), RHW (replicated high withdrawal), LW (low withdrawal), and RLW (replicated low withdrawal) selection lines were recently developed by Erwin et al. (unpublished). This selection experiment started from the same HS/Ibg founding population and used the same ethanol treatment regimen and scoring procedure as was used to develop the WSP and WSR selection lines (Erwin et al. unpublished). This allowed us to determine whether the QTLs found in the IP2·IR1 F2 population could be independently supported by QTL data from these new selection lines. Materials and methods Animals. We used the inbred strains derived from the outbred WSP and WSR selection lines after selection was complete so that only two alleles would be present at each marker and for each QTL in equal frequencies. The inbred strains were developed by over 20 generations of brother · sister mating. Because two of the four inbred strains were unavailable at the time of these studies, we used the replicate 1 WSR inbred strain (IWSR-1) and replicate 2 WSP inbred (IWSP-2) as progenitors to generate the F2 (IP2·IR1 F2) mapping population (N = 440) used in the present study. These two inbred strains differ over tenfold in their severity of withdrawal convulsions following equivalent exposure to ethanol (Crabbe, unpublished). Inhalation treatment. This regimen was essen1 tially that described by Buck et al. (2002). All mice were injected with 1.5 g/kg ethanol to elevate blood ethanol concentrations (BEC) and were also given 68 mg/kg pyrazole HCl, an alcohol dehydrogenase inhibitor, to stabilize BECs. They were exposed to ethanol vapor (9.5 mg/liter of air) for 3 days. After 24 and 48 h of vapor exposure, mice were removed from the chamber and given pyrazole injections. After 72 h, mice were removed from the chamber, and a 20-lL blood sample was promptly drawn from the tail tip to determine BEC. The index of withdrawal was the handling-induced convulsion (HIC) elicited by lifting each animal by the tail. HIC was scored for intensity by using a rating scale ranging from 0 (no seizure) to 7 (spontaneous lethal tonic-clonic seizure). HIC scoring was carried out just prior to ethanol treatment (baseline HIC), immediately upon removal from vapor, and at hourly intervals postwithdrawal up to 10 h, and again at 24 and 25 h with methods

456

described previously (Terdal and Crabbe 1994). The area under the withdrawal curve was used as the index of withdrawal severity after correcting for baseline HIC and BEC differences among individual 1 mice by using multiple regression residuals (Buck et al. 2002). This procedure eliminates the effects of individual differences in baseline HIC and BEC during chronic exposure on withdrawal HIC scores, thus insuring that the correlation between corrected HIC withdrawal scores and either baseline HIC or BEC was zero. Only the corrected scores were used in all subsequent analyses. The progenitor inbred strains and the F2 mice were bred and tested in Crabbe’s mouse colony at the Portland VA Medical Center pursuant to AAALAC and NIH guidelines for animal care and use. All animals were maintained on a 12 h light/dark cycle and were given food and water ad libitum. After being weaned at age 21 ± 1 days, mice were housed with their same-sex littermates in groups of 2–4 and were tested at age 7–11 weeks. The Colorado selection line mice were maintained under very similar conditions at the University of Colorado Health Sciences Center. Genotyping. DNA was isolated from spleen and genotyped by using standard methods (Buck et al. 2001; Bergeson et al. 2001). Very briefly, genomic DNA was extracted from spleen by a salting out method on proteinase K digests. Genotyping with 1 PCR was carried out for SSLP loci (Dietrich et al. 1992, 1994) with primer pairs obtained from Research Genetics, lnc., (Huntsville, Ala.). PCR genotyping was a modification of that described in Dietrich et al. (1992) by using ethidium bromide staining and high-resolution agarose (MetaPhor, FMC) in place of 32P radiolabeling and polyacrylamide. Because the alleles for the progenitors were unknown, a large number of microsatellite markers were screened in the search for polymorphic markers distributed at about 15- to 25-cM intervals throughout the genome. We found 82 polymorphic markers covering about 90% of the genome to within 20 cM of the nearest marker in our intercross. Mapping approach. Selective genotyping was used in which only the extreme high and low scoring tails of the trait distribution were genotyped out of the total sample of 440 F2 mice. A two-stage sequential search was used to reduce genotyping costs. For the first stage, only 88 mice (44 in each tail) were genotyped genome-wide for all 82 markers. Those markers attaining p < .05 (two-tailed) in their association with the trait were then genotyped for an additional 88 mice (stage 2), bringing the total gen-

S.E. B ERGESON

ET AL :

QTL S

FOR

A LCOHOL W ITHDRAWAL

otyped to 176 mice (40% of the total sample) equally distributed across both tails and both genders. The data from both stages were analyzed by using Map Manager QT software for the Macintosh computer (v29b; Manly and Olson 1999; obtained from http:// mapmgr.roswellpark.org) by using the interval mapping option at 2-cM spacing. The second stage data were also analyzed for QTLs and for interactions (epistasis) by using the Pseudomarker 0.9 program written for the MATLAB (Mathworks Inc., Natick, Mass.) programming environment (Sen and Churchill 2001; obtained from http://www.jax.org/ research/churchill) by using default settings except where noted otherwise. The first program uses linear regression methods for interval mapping (Haley and Knott 1992), while the second uses Monte Carlo computer sampling to estimate the most likely values of unknown parameters (QTL genotypes, locations, missing data) by Bayesian methods. Sixty-four imputations (samplings) of unknown genotypes based on the observed marker data were made at 5cM intervals, which approximates interval mapping at this resolution. This program analyzes the effect of each marker (pseudomarker) or marker interval on the trait (MAINSCAN program), but also the phenotypic effects of marker pairs or intervals taken together (PAIRSCAN program). The latter allows a genome-wide search for epistasis (locus-locus interactions) across the 5-cM grid as described below. Statistical significance criteria. The criterion for statistical significance was a genome-wide p < .05 estimated by permutation tests on our F2 data (Churchill and Doerge 1994). This method takes the trait data for individual mice and reassigns them at random to the genotypes 5000 times. This simulates the null hypothesis that there are no QTLs anywhere in the genome; thus, any apparent QTLs are by definition false positives. QTL analysis with 2-cM interval mapping was carried out on each permutation with QT software. The 5% highest LOD scores were identified, and the threshold which separates them from the other 95% was taken as the genome-wide empirical significance threshold for this data set. This value was LOD 3.5 (p = .0003 for single markers or intervals). The threshold for a suggestive QTL was LOD 2.2 (p = .006), which allows an average of one false positive QTL genomewide. We also carried out permutation tests for each gender. We added LOD 0.3 to these gender-specific thresholds to correct for carrying out two genomewide searches, resulting in LOD 3.9 for males and LOD 3.8 for females. For tests of epistasis (PAIRSCAN), we first required that each marker of a pair be located on dif-

S.E. B ERGESON

ET AL :

QTL S

FOR

A LCOHOL W ITHDRAWAL

ferent chromosomes, or if on the same chromosome, that they be spaced at least 40 cM apart. This was to avoid closely linked pairs which often have missing or under-represented heterogametic genotypes owing to linkage. Second, we required that the combined (or full model) effects of a marker pair, which reflects the main effects of both markers plus their interaction, exceed the threshold for significance (5% genome-wide error rate) estimated by a permutation test carried out with the Pseudomarker PAIRSCAN program on F2 data. In this study, the threshold was set at LOD 8.0 (df = 8, p = 1 · 10)5) based on 500 permutations. Only when the combined effects of a marker pair were significant (15 marker pairs met this criterion) did we look to see whether their interaction attained the significance threshold of p < .01 (Sen and Churchill 2001). With this approach, the main protection against false positive interactions rests on the stringent criterion set for the combined or joint effects of marker pairs (which includes their interaction), which markedly reduces the number of locus pairs directly examined for interactions alone (only 15 pairs in this study). This approach represents a conditional search for epistasis (Chase et al. 1997; Hood et al. 2001) because only marker pairs attaining significance for their joint effects on a trait (the condition) are tested for interactions. This condition was preset into the software settings so that only significant results were reported. Those meeting the above significance criterion for interactions were then subjected to the epistatic interaction analysis of Cheverud and Routman (1995) and Routman and Cheverud (1997), which examines epistasis in an allele frequency-independent manner and partials out the interaction effect (df = 4) into the additive · additive, additive · dominance, dominance · additive, and dominance · dominance components (each df = 1). [A MathCad worksheet implementing this analysis is available from [email protected].] The statistical power of an experiment is the probability of correctly detecting a QTL of a given effect size, here expressed as the proportion of the trait (phenotypic) variance due to a QTL, or %VP. Stage 1, when a was set at .05, has a statistical power of about 0.8 to reliably detect a QTL with %VP of 4% based on equations given by Belknap and Atkins (2001). For stage 2, when a was set at .0003 (the significance threshold), the corresponding %VP value at the same power was about 5%. These calculations took into account the effects of selective genotyping as presented by Darvasi and Soller (1992). Selective genotyping causes an upward bias in estimates of the QTL effect size expressed as the

457

proportion of the trait variance due to a QTL (%VP) when compared with full genotyping when all 440 mice are genotyped. In stage 2, 40% of the mice from the extreme tails of the distribution were genotyped (20% in each tail), so the upward bias is about 2.1fold (tau is 2.1), as shown by Darvasi and Soller (1992). Thus, the observed QTL effect sizes reported by the QTL analysis software were divided by 2.1 to remove this bias, which provides estimates of what we would expect to observe with full genotyping. In contrast, selective genotyping causes little bias in LOD score estimates (usually a downward bias) because the effect of increased QTL effect size is offset by the effect of reduced N compared with full genotyping. Estimates of map location are also little affected by selective genotyping (Darvasi and Soller 1997). For the eight provisional QTLs showing p < .05 in the first stage, we also looked at each gender separately in stage 2. The LOD scores for each gender were subtracted, and this gender difference LOD score converted back to a p value, which is an estimate of p for the gender difference. Interconverting LOD (df = 2) and p were done by using the expressions: LOD = )log10(p) or p = 10)LOD. Since eight provisional QTLs were searched in the F2 for gender differences, we used a Bonferroni correction of eight fold; thus the significance threshold was set at p = .05/8, or p = .006 (a LOD score difference of 2.2) for the gender difference. The HW, LW, RHW and RLW selection lines. As a follow-up to the QTLs detected in the IP2·IR1 F2, we tested markers in the known QTL regions from our F2 study in four new selection lines developed by V. Gene Erwin and coworkers, using the same procedure and beginning with the same heterogeneous stock (HS/Ibg) as was used by J.C. Crabbe more than 20 years earlier. Because these Colorado selection line mice were in the early generations of selection when random drift is small (S5 for HW, LW lines, N = 63; S2 for RHW, RLW lines, N = 57), we were able to use the selection lines directly rather than generating an F2 intercross 1 by using the QTL analysis method of Belknap et al. (1997) explicitly designed for short-term selection line data. Evidence for the presence of a QTL was gained from the difference in relative allele frequencies between the high and low lines at a nearby marker (d = qH)qL) exceeding that expected from random drift and sampling error. Most markers showed only two alleles, but when three alleles were encountered, the allele showing the largest value of d was used in the analysis (its frequency was designated as q), and the frequencies of the

458

S.E. B ERGESON

ET AL :

QTL S

FOR

A LCOHOL W ITHDRAWAL

Table 1. QTLs emerging as significant (LOD>3.5) or suggestive (LOD>2.2) in a genome-wide scana based on stage 2 data (176 IP2·IR1 F2 mice) analyzed by Map Manager QT. The first column shows the chromosome and the 1.0 LOD support intervals in cM units, which approximates the 90% confidence interval. Column 2 gives the marker(s) and cM location in parentheses. Also given are the p values with LOD scores in parentheses (column 3), corrected percentage of the trait variance accounted for (%VP) by each QTL for both genders combined (column 4) or separately (column 2), and the degree and direction of dominance (column 5) Chrom. (±1 LOD) 1:28–47 cM 4:33–51 cM 8:16–43 cM 11:15–36 11:63-ter 14:24–48 cM

Nearest marker or flanking markers D1Mit46 (43) D4Mit186 (44) D8Mit4.25 (14,32) D11Mit340,4 (11,37) (Females, LOD 2.5, 5%; Males, LOD 5.7, 12%) D11Mit203 (75) (Females, LOD 3.5, 7%; Males, LOD 0.2, 0.5%) D14Mit160 (40)

Dominance

p (LOD)

%VP

)7

(6.5) (2.3) (3.5) (6.7)

7 3 4 7

Partial, R allele Overdom, P allele Partial, P allele Partial, R allele

3 · 10)3 (2.5)

3

None

6 · 10)5 (4.2)

5

None

3 5 3 2

· · · ·

10 10)3 10)4 10)7

a

Mit markers used for the genome-wide scan were Chr 1: 122, 46, 30, 16, 56, 511; Chr 2: 80, 296, 61, 484, 102, 493; Chr 3: 46, 43, 86; Chr 4: 263, 272, 186, 312, 33; Chr 5: 1, 297, 134, 312, 371; Chr 6: 307, 93, 25; Chr 7: 91, 30, 66, 334; Chr 8: 4, 25, 312, 320; Chr 9: 91, 31, 274, 18; Chr 10: 51, 194, 61, 161; Chr 11: 340, 4, 160, 203, 104; Chr 12: 38, 118, 262; Chr 13: 91, 224, 147, 130; Chr 14: 54, 160, 265; Chr 15: 12, 63, 149; Chr 16: 15, 5, 152; Chr 17: 198, 53; Chr 18: 68, 35, 9, 79, 4; Chr 19: 95, 80, 18, 55; Chr X: 68.

other two alleles were pooled with frequency, p. The value of Z, the normal deviate, was calculated as follows for each marker and was used to test for QTL significance: Equation 1: Z ¼ d=½2p0 q0 F þ pH qH =2nH þ pL qL =2nL 0:5 ;

where the 1st term in the denominator is the expected random drift variance (Falconer and Mackay 1996); the 2nd and 3rd terms are the variances due to sampling error in the high and low selection lines, respectively; nL and nH are the sample sizes in each line; pH, qH, pL, qL are the allele frequencies in each line; F is the inbreeding coefficient at a given selected generation (Falconer and Mackay, 1996); and p0 and q0 are the initial allele frequencies in the S0 founding population. [A MathCad worksheet implementing this analysis is available from [email protected].] Results A full genome search in the IP2·IR1 F2, based initially on genotyping 88 extreme-scoring mice (first stage), identified eight provisional QTLs at p < .05 for individual markers or intervals. These were on Chrs 1 (proximal), 1 (distal), 2 (proximal), 4 (mid), 8 (proximal), 11 (proximal), 13 (mid), and 14 (mid). These eight were followed up in the second stage by expanding the genotyped sample for markers in these chromosomal regions to 176 extreme-scoring mice out of 440. The more promising of these results are summarized in Table 1 and are shown in Figures 1 through 4.

Evidence for gender differences in QTL effect size. Of the eight provisional QTLs emerging from the stage 1 full genome search, only one was significantly different by gender after stage 2 analysis. The Chr 11 QTL at 15–36 cM (LOD 5.7 in males, LOD 2.5 in females) was strongly gender influenced and accounted for about 12% of the trait variance in males vs 5% in females, and about 7% for both genders combined (Fig. 2). Another gender-influenced QTL may exist at the distal end of Chr 11 (Fig. 2). At LOD 3.5 in females (vs LOD 0.2 in males), it failed to reach significance when compared with the permutation test threshold for females only, which was LOD 3.8. Results of genome scan. Four QTLs emerged as significant and two as suggestive. Fig. 1 shows the genome-wide results of the Pseudomarker MAINSCAN analysis, while the Map Manager QT results are given in Table 1. In general, the two computer programs gave closely similar results, as expected, with the possible exception of the Chr 8 QTL, which showed LOD 3.5 with QT interval mapping and LOD 4.5 with Pseudomarker 0.9 MAINSCAN analysis (Fig. 1). Table 1 also shows the 1.0 LOD support intervals from Map Manager QT interval mapping. These values were similar to the 95% confidence intervals in cM estimated by the method of Darvasi and Soller (1997) based on the equation: 95%Cl = 530/[N · (%VP/100)] for an F2 showing complete dominance. Because we observed only partial dominance, we used the constant 600 instead of 530. The four significant QTLs (Chrs 1, 8, 11, 14) together accounted for about 23% of the trait variance

S.E. B ERGESON

ET AL :

QTL S

FOR

A LCOHOL W ITHDRAWAL

459

Fig. 1. LOD plot (df = 2) for a genome scan for QTLs using MAINSCAN in Pseudomarker 0.9 by using a 5-cM grid with 64 imputations for stage 2 IP2·IR1 F2 data. The horizontal fine line shows the significance threshold of LOD 3.5 estimated from 5000 permutations. Four QTLs emerged as significant and one as suggestive, influencing corrected withdrawal scores following 3 days of ethanol vapor inhalation.

for both genders combined. This was calculated by adding the %VP from each QTL, which in this case was justified because we found no association between the markers best representing each QTL by using simple chi square tests of genotype frequencies. However, genome-wide scans tend to overestimate the QTL effect size as a function of the statistical power of the experiment to detect QTLs (Beavis 1998), which is in addition to the upward bias due to selective genotyping noted above. When an equation is fitted to the data presented by Beavis, the

fold increase of this upward bias is approximately [1/ Power]0.6. Since the estimated power of this experiment to detect each of the four ranged from .70 to .90, this upward bias can be estimated to range from roughly 5% (1.05-fold) for the two largest QTLs (Chrs 1 and 11) to about 25% (1.25-fold) for the two smallest (Chrs 8 and 14). Therefore, to be conservative, we conclude that about 20% of the trait variance has been accounted for by the four QTLs detected in this study.

Fig. 3 The corrected HIC withdrawal scores by genotype

Fig. 2. LOD (df = 2) plot along the length of Chr 11 for both males (N = 88) and females (N = 88) based on stage 2 IP2·IR1 F2 data. A highly significant QTL emerged at 20 cM for males, but not for females. In contrast, a suggestive QTL emerged at 75 cM for females, but not for males. The markers used and their cM locations (in parentheses) are 11-340 (11), 11-4 (37), 11-160 (58), and 11-203 (79).

for the D11Mit4 marker on Chr 11 as a function of genotype at a modifier locus (D13Mit147) on Chr 13 in the F2. An interaction is demonstrated by the fact that the QTL effect of the Chr 11 QTL is quite evident when the modifier locus genotype is RR, but is absent when the modifier genotype is PP. The Chr 13 modifier had no significant main effects in the MAINSCAN analysis, but emerged as significant only in the PAIRSCAN analysis owing mainly to the interaction with the Chr 11 QTL.

460

S.E. B ERGESON

ET AL :

QTL S

FOR

A LCOHOL W ITHDRAWAL

11 (25 cM) and a modifier locus on Chr 13 (50 cM) as shown in Fig. 3. [A locus without significant main effects but with a significant interaction with another locus is a modifier locus.] The second and third were interactions between two QTLs with significant main effects taken singly (detected by MAINSCAN), one on Chr 8 (55 cM) with Chr 14 (40 cM), and the other on Chr 4 (40 cM) with Chr 8 (35 cM). These three interactions were then subjected to the interaction analysis of Cheverud and Routman (1995) and Routman and Cheverud (1997), as shown in Table 3. The first two interactions were primarily additive · additive, while the third was primarily additive · dominant. Fig. 4. Two-dimensional plot of PAIRSCAN results for the six chromosomes of greatest interest with a 5-cM grid. The lower half shows the LOD scores for the full model (df = 8) for marker pairs, which includes the main effects of each marker plus their interaction, while the upper half shows LOD scores for the interaction alone (df = 4). Areas in white failed to reach p < .05 (