Candidate Quantitative Trait Loci and Naturally Occurring Phenotypic Variation for Bristle Number in Drosophila melanogaster: The Delta-Hairless Gene Region

Corresponding author: Trudy F. C. Mackay, Department of Genetics, Box 7614, North Carolina State University, Raleigh, NC 27695, trudy_mackay{at}ncsu.edu (E-mail).

Communicating editor: P. D. KEIGHTLEY

	ABSTRACT

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

Delta (Dl) and Hairless (H) are two chromosome 3 candidate neurogenic loci that might contribute to naturally occurring quantitative variation for sensory bristle number. To evaluate this hypothesis, we assessed quantitative genetic variation in abdominal and sternopleural bristle numbers among homozygous isogenic third chromosomes sampled from nature and substituted into the Samarkand (Sam) inbred chromosome 1 and 2 background; among homozygous lines in which the wild-derived Dl-H gene region was introgressed into the Sam chromosome 3 background; and among Dl-H region introgression lines as heterozygotes against the Sam wild-type strain and derivatives of Sam into which mutant Dl and H alleles had been introgressed. Variation among the Dl-H region introgression lines accounted for 36% (8.3%) of the total chromosome 3 among line variance in abdominal (sternopleural) bristle number and for 53% of the chromosome 3 sex x line variance in abdominal bristle number. Naturally occurring alleles in the Dl-H region failed to complement a Dl mutant allele for female abdominal bristle number and sternopleural bristle number in both sexes, and an H mutant allele for both bristle traits in males and females. These results are consistent with the hypothesis that naturally occurring alleles at Dl and H contribute to quantitative genetic variation in sensory bristle number.

MOST phenotypic differences between individuals, such as variation in anatomical dimensions, behavior, and susceptibility to disease, are quantitative in nature, of degree rather than kind (FALCONER and MACKAY 1996 ). Phenotypic variation of quantitative traits is characterized by the simultaneous segregation of alleles at multiple loci, the effects of which are sensitive to the environment. A major challenge in medicine, plant and animal breeding and evolution is to determine at what genetic loci allelic variation contributes to phenotypic variation in quantitative traits, and to estimate the effects of the segregating alleles. This task is not simple, because the allelic effects at each locus are, by definition, too small to be perceived over and above the background noise of segregation of other alleles affecting the trait of interest and environmental variation. Quantitative trait loci (QTLs) can now be mapped in many species by linkage to polymorphic, neutral molecular markers (reviewed by TANKSLEY 1993 ). However, even with dense molecular marker maps, it is difficult to localize QTLs to regions much smaller than a few cM with initial whole genome screens, as informative recombination events between closely linked loci become increasingly rare and require prohibitive sample sizes to recover. Although sufficiently precise for utilizing the QTL in a selective breeding program, for example, this level of resolution is still several orders of magnitude away from identifying allelic differences at a single locus responsible for variation in a quantitative trait phenotype.

The most promising approach for proceeding from QTL to genetic locus is to search for candidate genes in the region to which the QTL maps. Candidate loci can be proposed a priori, based on functional relationships of known genes to the trait phenotype (i.e., loci in the biochemical or developmental pathways leading to the phenotype of interest), or, with considerably more difficulty, from positional cloning subsequent to high-resolution meiotic mapping. Underlying the candidate gene approach is the assumption that variation in quantitative traits is caused by the segregation of "isoalleles" with small effects at loci previously identified by alleles with major effects (THOMPSON 1975 ; MACKAY 1985 ; ROBERTSON 1985 ). Obvious limitations of the candidate locus approach are that all of the loci involved in the relevant biochemical or developmental pathways leading to the phenotype of interest are usually not known, and loci that are not obvious candidates may have undescribed and unexpected pleiotropic effects on the trait.

Although roughly concordant map positions of QTLs and candidate loci from an initial mapping population suggest a testable hypothesis, several criteria need to be fulfilled before it can be accepted that the candidate gene and the QTL are one and the same. (1) If "near-isoallelic" or congenic strains, in which alternative alleles of the QTL are introgressed into a common background, can be established, variation among the near-isoallelic lines confirms the presence of loci affecting the trait, including the candidate locus, in the introgressed gene region (DOEBLEY et al. 1995 ; ESHED and ZAMIR 1996 ). In the absence of homologous recombination, the precision of this analysis is limited by the length of the introgressed fragment containing the QTL. (2) Quantitative complementation of QTL alleles with a major mutation and a standard wild-type allele at the candidate locus tests for genetic interaction between the QTL alleles and known alleles at the candidate locus (LONG et al. 1996 ; MACKAY and FRY 1996 ). This test is not definitive, because observed interactions could be allelic or epistatic, and failure to observe interaction may be due to interallelic complementation. (3) If the candidate gene has been cloned and characterized at the molecular level, population-level association studies of molecular variation in the candidate gene and phenotypic variation in the quantitative trait can implicate the candidate locus as a QTL (e.g., SING et al. 1988 ; MACKAY and LANGLEY 1990 ; CORDER et al. 1993 ; LAI et al. 1994 ). A strong association of a molecular polymorphism in a candidate gene with phenotypic variation could arise if the polymorphism causes the phenotypic variation, in which case the association will be found in all populations with that particular molecular variant. However, a strong association could also be found if the molecular variants are in linkage disequilibrium with the true causal polymorphism. In the latter case, conflicting results may be obtained from independent studies in different populations due to differences in population history, and misleading linkage disequilibria can arise from recent admixture of populations with different frequencies of molecular variants in the candidate gene region and different average phenotypes for the trait of interest. Further, sorting out which of the many nucleotide and length polymorphisms throughout the regulatory and coding sequences of the candidate gene are functional and which are phenotypically neutral is a nontrivial task. (4) In genetically tractable organisms, it should be possible to demonstrate functional complementation of a null mutation at the candidate locus with the QTL allele in transgenic individuals. Whether transgenic rescue of a null allele can be used to identify the molecular polymorphism(s) that differentiate alternative QTL alleles probably depends on the availability of targeted gene replacement, because position effects of randomly integrated transgenes on the trait might be as large as the QTL effects themselves (MACKAY et al. 1992A ; LYMAN et al. 1996 ). (5) Finally, if naturally occurring variation at a candidate gene causes phenotypic variation in a quantitative trait, it follows that spontaneous mutations at the candidate locus should have phenotypic effects on the trait. The converse is not necessarily true; loci at which spontaneous mutations affect the quantitative trait phenotype may not contribute to naturally segregating variation if functional variation in the gene decreases fitness.

The above map position, genetic complementation, allelic association and mutation detection criteria are not individually sufficient proof that a candidate gene is the genetic locus corresponding to a QTL. The case becomes stronger if multiple independent pieces of evidence point to the verity of this hypothesis. For this reason, comprehensive evaluations of the extent to which allelic variation at candidate loci contributes to quantitative genetic variation may be restricted to quantitative traits in model organisms for which the entire repertoire of genetic tests is feasible. One such model system is the number of sensory bristles of Drosophila melanogaster. Drosophila abdominal and sternopleural bristle numbers are highly genetically variable, with heritabilities of ~0.5 (FALCONER and MACKAY 1996 ). Bristle numbers have historically been used to elucidate basic quantitative genetic principles such as short- (CLAYTON et al. 1957 ; FRANKHAM et al. 1968 ) and long-term (CLAYTON and ROBERTSON 1957 ; JONES et al. 1968 ) response to selection; selection limits (YOO 1980A ); spontaneous mutation rates (reviewed by KEIGHTLEY et al. 1993 ; HOULE et al. 1996 ); the effects of recombination (MCPHEE and ROBERTSON 1970 ), population structure (MADALENA and ROBERTSON 1975 ) and environmental heterogeneity (MACKAY 1981 ) on genetic variation and response to selection; and the chromosomal locations, numbers and effects of QTLs contributing to selection response (BREESE and MATHER 1957 ; THODAY 1979 ; SHRIMPTON and ROBERTSON 1988A , SHRIMPTON and ROBERTSON 1988B ; LONG et al. 1995 ). As Drosophila bristles are sensory organs of the peripheral nervous system, appropriate candidate loci for bristle number QTLs are the many genes that are involved in the development of the nervous system and sensory organs (CAMPOS-ORTEGA 1993 ; JAN and JAN 1993 ). Other candidate bristle number QTLs are loci with major mutant effects on bristle number that have not (yet) been implicated in the bristle development pathway.

Several lines of evidence are consistent with the hypothesis that quantitative variation in bristle number is at least partly attributable to variation in candidate bristle loci. (1) QTLs responsible for the divergence between high and low selection lines map to approximately the same positions as several candidate bristle development loci (SHRIMPTON and ROBERTSON 1988B ; LONG et al. 1995 ). (2) Naturally occurring (LONG et al. 1996 ) and spontaneous mutant (MACKAY and FRY 1996 ) high and low alleles at bristle number QTLs genetically interact with mutations at candidate bristle number loci. (3) Molecular variation at candidate neurogenic loci is associated with naturally occurring phenotypic variation in bristle number (MACKAY and LANGLEY 1990 ; LAI et al. 1994 ). (4) Occasionally, alleles at candidate bristle loci with large effects are found segregating at low frequency in natural populations (MCBRIDE and ROBERTSON 1963 ; FRANKHAM and NURTHEN 1981 ). (5) Spontaneous mutations at candidate bristle number genes have contributed to long-term artificial selection response (HOLLINGDALE 1971 ; FRANKHAM et al. 1978 ; FRANKHAM 1980 ; YOO 1980B ). (6) P-element insertional alleles at neurogenic loci have quantitative phenotypic effects on bristle number (LYMAN et al. 1996 ).

Although these observations collectively support the hypothesis that genetic variation at known bristle loci contributes to quantitative variation in bristle number, they are nevertheless a somewhat eclectic set of data in that the various steps of the proof are illustrated by different candidate loci. To systematically evaluate the extent to which variation at candidate genes contributes to quantitative genetic variation for bristle number in general, it is necessary to collect information on QTL map position, genetic and functional complementation, allelic association and mutational analysis for a sample of several candidate loci. In this way, we can begin to disperse the statistical "fog" surrounding at least one model quantitative trait and begin to describe quantitative variation in terms of defined molecular variants, their frequencies, and allelic effects (ROBERTSON 1967 ). We have begun such a comprehensive investigation with a quantitative and molecular genetic analysis of naturally occurring variation in the Delta (Dl) gene region.

The Drosophila peripheral nervous system (PNS) is progressively determined by a number of interacting genes. Initially, the proneural genes endow small clusters of cells with the potential to become neuronal precursors, after which the neurogenic loci single out a subset of cells in the proneural cluster to become neuronal precursors (CAMPOS-ORTEGA 1993 ; JAN and JAN 1993 ). Subsequent steps in the development of the PNS are the commitment of neuronal precursors, specification of identity of neuronal precursors, and sensory organ cell fate specification (JAN and JAN 1993 ). Of the many loci known that control each step in the pathway (JAN and JAN 1993 ), interactions of the neurogenic loci during the process of lateral specification of neuronal precursors is particularly well established (ARTAVANIS-TSAKONAS et al. 1995 ). Neuronal precursor cell fate is specified by interactions between proteins of the Notch signaling pathway. The eponymous Notch (N) gene encodes a transmembrane receptor protein, and Dl encodes a transmembrane protein that is an extracellular ligand for Notch; Delta is thought to activate the Notch receptor by binding of the epidermal growth factor-like repeats of both molecules (ARTAVANIS-TSAKONAS et al. 1995 ). The Hairless (H) gene is closely linked to Dl and interacts genetically with both N and Dl (LINDSLEY and ZIMM 1992 ). It encodes a nuclear protein (MAIER et al. 1992 ) that is thought to act at one of the later stages in PNS development (JAN and JAN 1993 ; ARTAVANIS-TSAKONAS et al. 1995 ).

Previously, we mapped one of five chromosome 3 QTLs responsible for the divergence of abdominal bristle number between artificial selection lines, derived from a natural population, to cytogenetic interval 89D-92E, which includes both Dl (92A1-2) and H (92E12-14; MAIER et al. 1992 ; LONG et al. 1995 ). The effect of this interval was 3.0 abdominal and 1.4 sternopleural bristles, averaged over sexes, and it accounted for 22% (30%) of the third chromosome divergence in abdominal (sternopleural) bristle number. Mutations at Dl and H showed strong genetic interactions with high and low selection line chromosomes (LONG et al. 1996 ); further, mutations at Dl and H interact with spontaneous mutations affecting abdominal and sternopleural bristle number (MACKAY and FRY 1996 ). Although implicating segregating and mutational variation at Dl and H as contributing to quantitative genetic variation in bristle number, these studies are subject to the caveats that the QTL map position needs to be confirmed independently and that the genetic interactions observed between high and low selection lines (or whole chromosomes from selection lines) and the Dl and H mutant alleles might not be allelic. Here, we have constructed near-isoallelic lines of the Dl-H gene region for a sample of unselected chromosomes from the same natural base population used to establish the selection lines of LONG et al. 1995 , and we demonstrate that QTLs in this region both contribute to quantitative variation in bristle number and fail to complement Dl and H mutant alleles. In the companion article (LONG et al. 1998 ), we provide further evidence that Dl is a genetic locus at which segregating variation contributes to quantitative variation in bristle number from associations between molecular variation at Dl and phenotypic variation in bristle number.

	MATERIALS AND METHODS

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

All Drosophila stocks were reared in shell vials with 10 ml cornmeal-agar-molasses medium, at 25°. For details of the loci and balancer chromosomes used, see LINDSLEY and ZIMM 1992 .

Chromosome 3 substitution lines:
Gravid D. melanogaster females were obtained from the Raleigh, NC, Farmer's Market in 1988 and used to establish isofemale lines. This sample of isofemale lines was used to establish the base population from which the artificial selection lines for abdominal bristle number used in QTL mapping were derived (LONG et al. 1995 ). In addition, a single third chromosome was extracted from each of 63 of these isofemale lines and substituted into the genetic background of the highly inbred Sam; ry⁵⁰⁶ strain (see LYMAN et al. 1996 ) by standard techniques using balancer chromosomes (MACKAY et al. 1996 ). The genotype of these lines was thus Sam1; Sam2; C3i, where i = 1, 2, ... , 63 wild-derived third chromosomes.

Abdominal bristle number (the number of hairs on the most-posterior sternite; segment six of females and segment five of males) and sternopleural bristle number (the sum of the number of bristles on the right and left sternopleural plates) were recorded on 10 C3i males and females in each of two replicate vials per line (i.e., 63 lines x two vials/line x two sexes/vial x 10 individuals/sex = 2520 individuals).

Construction of Dl-H region introgression lines:
An allele of the Delta (Dl³, 3-66.2) locus was substituted into the Sam; ry⁵⁰⁶ background by 20 generations of backcrossing Dl³ heterozygous females to Sam; ry⁵⁰⁶ males. This backcross chromosome (Dl³ BC20) was subsequently maintained balanced against TM3, Sb ry^RK, in a stock of genotype Sam1; Sam2; Dl³ BC20/TM3, Sb ry^RK.

Let C3i refer to the wild isogenic third chromosomes in the Sam background, and Dl³ BC20/Sb to the strain with the balanced Dl allele in the Sam background. At G0, C3i females were crossed to Dl³ BC20/Sb males in two replicate vials per line. At G1, C3i/Dl³ BC20 females were backcrossed to Dl³ BC20/Sb males, keeping the two replicate backcrosses per line distinct. This cross was repeated for nine further generations. After 10 backcross generations, the introgressed C3i regions should contain the wild alleles at the Dl locus plus a linked segment of wild-derived chromosome ~10 cM on either side of Dl (CROW and KIMURA 1970 , pp. 94–95). Keeping two independent backcross lines of each wild chromosome controls for variation in the size of the introgressed fragment, as variation in bristle number among introgression lines can be partitioned into that common to the independent backcross lines and that between replicate backcross lines, which is attributable to variation in flanking chromosome fragments (see below). Because the Hairless (H) locus at 3-69.5 is only 3.3 map units away from Dl, it is likely that the introgressed regions also contain wild-type H alleles. Therefore, we refer to the genotype of the introgression lines as Dl-Hi. At G11, Dl³ BC20/Dl-Hi males from each independent backcross line were crossed to Dl³ BC20/Sb females. Sb/Dl-Hi females and males were mated inter se at G12, and at G13 Dl-Hi homozygous females and males were crossed to establish the Dl-H region introgression lines. There were a total of i = 62 pairs of introgression lines (one of the original isogenic third chromosomes did not survive the entire backcross procedure), of which 59 were homozygous viable.

Homozygous effects, Dl-H region:
Abdominal and sternopleural bristle numbers were recorded for 10 Dl-Hi males and females in each of two replicate vials per backcross replicate, that is, 59 lines x two backcrosses/line x two vials/backcross x two sexes/vial x 10 individuals/sex = 4720 individuals.

Heterozygous effects, Dl-H region:
Heterozygous effects of the 62 Dl-Hi introgression lines were determined against three strains: (1) the Sam; ry⁵⁰⁶ strain into which background the wild third chromosomes had been introgressed and which contain a wild-type allele at the Dl and H loci (simply abbreviated Sam below); (2) the Sam; Dl³ BC20 strain with an extreme Dl mutation and a wild-type allele of H (abbreviated Dl³ below); and (3) Sam; H BC20, constructed by 20 generations of backcrossing H to Sam; ry⁵⁰⁶ (MACKAY and FRY 1996 ), which contains the mutant H allele and a wild-type Dl allele (abbreviated H below). Abdominal and sternopleural bristle numbers of the three heterozygous genotypes were recorded for 10 males and females in each of two replicate vials per backcross replicate, that is, 62 lines x two backcross replicates/line x two replicate vials/backcross x two sexes x 10 individuals/sex = 4960 individuals per heterozygous genotype.

Quantitative complementation effects:
Complementation tests typically involve recessive mutations with large phenotypic effects, so the heterozygote between the strain containing a putative mutation (m*) interacting with the locus of interest and the strain containing a recessive mutation at the tested locus (m) can be unambiguously classified as mutant or wild type. The logic of complementation testing can be extended to alleles with small, quantitative and additive (i.e., not completely recessive) effects by recognizing that the traditional test is for a difference in heterozygous effect between m*/+ and m*/m genotypes. That is, failure to complement is inferred if the degree of dominance of m* changes according to whether the allele at the locus tested is mutant or wild type. Therefore, the analogous test for genetic interaction between a wild-derived allele affecting bristle number in a Dl-H introgression line and Dl³ or H is the difference in bristle number between either Dl-Hi/Dl³ or Dl-Hi/H (the Tester crosses) and Dl-Hi/Sam (the Control cross) genotypes. However, because the Dl³ and H mutations themselves have heterozygous effects on bristle number (MACKAY and FRY 1996 ) and the chromosomes containing these mutations have linked non-Sam genomic fragments that might also have heterozygous effects on bristle number, a difference in bristle number between Tester and Control genotypes for any one introgression line is uninformative. However, variation in the Tester-Control difference among lines is indicative of variation in degree of dominance of the Dl-H introgressed regions as heterozygotes against wild-type and mutant alleles at candidate loci and is interpreted as quantitative failure to complement. Formally, variation in Tester-Control bristle number effects is detected as a significant cross by line interaction term by analysis of variance.

Statistical analyses:
The whole chromosome 3 homozygote bristle number data were analyzed by two-way factorial analysis of variance (ANOVA), with sex (fixed) and line (random) cross-classified main effects and replicate vial a random effect nested within lines. Sums of squares were partitioned into sources (degrees of freedom) attributable to Sex, S (1), Line, L (62), S x L (62), Replicate, R(L) (63), S x R(L) (63) and Error, E (2268). The homozygous and heterozygous Dl-H region bristle number data were analyzed similarly by two-way factorial ANOVA, with sex and line cross-classified main effects, backcross replicate a random effect nested within lines and replicate vials nested within (line x backcross). Sums of squares (degrees of freedom) were partitioned into sources attributable to S (1), L (58), S x L (58), Backcross, BC(L) (59), S x BC(L) (59), R(L x BC) (118), S x R(L x BC) (118) and E (4248) for the homozygous introgression lines and to S (1), L (61), S x L (61), Backcross, BC(L) (62), S x BC(L) (62), R(L x BC) (124), S x R(L x BC) (124) and E (4464) for the heterozygous introgression lines. Analyses of the quantitative complementation test data included the effect of Cross (C) as an additional fixed, cross-classified main effect. Thus, sums of squares were partitioned by three-way factorial ANOVA into sources (degrees of freedom) attributable to C (1), S (1), C x S (1), L (61), C x L (61), S x L (61), C x S x L (61), BC(L) (62), C x BC(L) (62), S x BC(L) (62), C x S x BC(L) (62), R(C x L x BC) (248), S x R(C x L x BC) (248) and E (8928). All ANOVAs and tests of significance were calculated using the Proc GLM procedure in SAS (SAS Institute, Inc., Cary, NC). Variance component estimation was done using SAS Proc VARCOMP. Linear regressions and correlations of line means were computed using the SAS procedures Proc REG and Proc CORR, respectively.

	RESULTS

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

Distribution statistics:
Quantitative variation in abdominal and sternopleural bristle numbers was assessed among (1) 63 isogenic third chromosomes, sampled from the Raleigh natural population and substituted into the common inbred Sam strain background; (2) homozygous near-isoallelic lines, in which the Dl-H gene region from the Raleigh third chromosomes was introgressed into the Sam third chromosome background; and (3) near-isoallelic Dl-H gene region lines as heterozygotes against the wild-type Sam strain and congenic derivatives of Sam into which mutant Dl³ and H alleles had been introgressed. The distribution statistics of line means are given in Table 1. The mean bristle number is much greater (less) in the backcross lines heterozygous for Dl (H) mutant alleles than in the backcross homozygotes, consistent with the dominant loss-of-function effects of these alleles. As is commonly observed, females have more bristles than males, across all genotypes. The variance among homozygous Dl-H region line means is considerably reduced relative to the whole chromosome 3 homozygotes for both bristle traits (Figure 1), as expected if there are multiple QTLs affecting bristle number on chromosome 3 and if backcrossing to a common inbred strain has been successful. The variance among heterozygous introgression line means is generally less than that among homozygous Dl-H region lines, with the interesting exceptions of female abdominal bristle number in crosses to Dl³ and sternopleural bristle number for both sexes in crosses to H. These observations are discussed in detail below. The line means are for the most part normally distributed.

View larger version (34K): In this window In a new window Download PPT slide   Figure 1. —Distributions of mean abdominal and sternopleural bristle number among homozygous whole chromosome 3 substitution lines (Raleigh C3) and among homozygous Dl-H region introgression lines derived from them. Solid bars are for male, and open bars for female line means.

Chromosome 3 substitution lines:
The variance components (²) and results of tests of significance from analyses of variance of bristle number among the chromosome 3 substitution lines are given in Table 2. The main effect of Sex was highly significant for abdominal bristle number but not for sternopleural bristle number in this set of lines, indicating sexual dimorphism for the former trait (Table 1) but not the latter. There was highly significant variation among lines for both bristle traits, indicating substantial naturally occurring genetic variation for bristle number. The S x L interaction term was highly significant for abdominal bristle number and marginally significant for sternopleural bristle number. The interpretation of the significant S x L interaction is that the female-male difference in bristle score varied among the lines, that is, there was significant genetic variation for sexual dimorphism for bristle traits, particularly for abdominal bristle number.

View this table: In this window In a new window   Table 2. Variance components (2) and significance of mean squares (P) from analyses of variance of bristle number for isogenic chromosome 3 substitution lines

Estimates of the genetic variance (V_G) and heritability (h²) for abdominal and sternopleural bristle number attributable to chromosome 3 can be obtained from the estimates of variance components given in Table 2. Because these lines are inbred, the variance component among lines (²_L ) estimates 2FV_G, and the S x L variance component (²_SL ) estimates 2F(1/2V_G), assuming a strictly additive model (FALCONER and MACKAY 1996 ). For completely homozygous lines, F = 1; thus, the genetic variance of bristle number was estimated by ²_{L +} ²_SL . The estimates of V_G attributable to variation among isogenic third chromosomes extracted from nature were 2.38 for abdominal bristle number and 2.72 for sternopleural bristle number. Taking the error variance, ²_E, as the estimate of the environmental variance, V_E, the heritability of bristle number was calculated from h² = V_G/(V_G + V_E). The estimates were h² = 0.39 for abdominal bristle number and h² = 0.48 for sternopleural bristle number. HOULE 1992 has argued that it is appropriate for many purposes to scale genetic and environmental variances by the mean and recommends computing the coefficients of genetic (CV_G) and environmental (CV_E) variance as and , respectively, where is the mean value of the trait. For these data, = 21.4 for both abdominal and sternopleural bristle numbers. Therefore, CV_G = 7.2 for abdominal bristle number and 7.7 for sternopleural bristle number, and CV_E = 9.1 for abdominal bristle number and 8.0 for sternopleural bristle number. Whether expressed as h² or CV_G and CV_E, these estimated quantitative genetic parameters for the two bristle traits are typical of previous whole genome estimates that have been obtained for these traits in other populations using different experimental designs (e.g., ROFF and MOUSSEAU 1987 ; HOULE 1992 ; FALCONER and MACKAY 1996 ).

Homozygous effects, Dl-H region:
The Dl gene region from the isogenic third chromosome lines were introgressed into a common inbred chromosome 3 background by 10 generations of backcrossing, with two independent backcrosses per line. These lines are expected to contain, on average, 20 cM of wild-derived genome flanking the Dl locus and are thus expected to be co-isogenic for 90 cM/110 cM = 82% of the third chromosome, or 276 cM/296 cM = 93% of the entire genome, where 110 cM and 296 cM are the approximate map lengths of the third chromosome and the entire genome, respectively (LINDSLEY and ZIMM 1992 ).

The results of analysis of variance of abdominal and sternopleural bristle numbers among the Dl-H region homozygous introgression lines are given in Table 3. The main effect of Sex was significant in these and all other analyses of these lines and indicates sexual dimorphism for both bristle traits (Table 1). This common observation will not be considered further. There was significant variation for abdominal and sternopleural bristle numbers among introgression lines, indicating segregating variation at loci affecting these traits in this region of the genome. For abdominal bristle number, 1.090/3.018 = 36.1% of the total chromosome 3 variance among lines was attributable to variation in the Dl-H region, whereas only 8.3% (0.435/5.233) of the variation among whole chromosome 3 lines for sternopleural bristle number was attributable to genetic variation in the Dl-H region. These observations are consistent with those from mapping studies that showed multiple factors on the third chromosome affected response to artificial selection for abdominal (LONG et al. 1995 ) and sternopleural (M. C. GURGANUS, S. V. NUZHDIN and T. F. C. MACKAY, unpublished results) bristle numbers from the Raleigh base population. There was also significant S x L variation for abdominal bristle number among the Dl-H region introgression lines, accounting for 0.460/0.869 = 52.9% of the S x L variation observed for the entire chromosome 3. There was significant variation between backcross replicates for both bristle traits, indicating that additional loci affecting bristle number, linked to Dl, were present in one backcross replicate but not the other. These linked loci had significant effects on variation in sexual dimorphism of abdominal bristle number. Terms involving variation between replicate vials were often significant in these and other analyses of the introgression lines, but these nongenetic terms will not be discussed further.

View this table: In this window In a new window   Table 3. Variance components (2) and significance of mean squares (P) from analyses of variance of bristle number for Dl-H region introgression lines

Heterozygous effects, Dl-H region:
Each of the Sam Dl-H introgression lines was crossed to a strain containing wild-type Dl and H alleles (Sam), a strain containing a mutant Dl allele (Dl³) and a strain containing a mutant H allele; the variation in bristle number was determined among the three sets of heterozygous genotypes.

The results of analysis of variance of heterozygous bristle number are given in Table 3. There was significant variation in abdominal bristle number among lines for Dl-H/Sam and Dl-H/H heterozygotes, but no other significant source of genetic variation. However, for the Dl-H/Dl³ heterozygotes, all of the significant variation in abdominal bristle number among lines was in the S x L component, and additionally there was variation in heterozygous effects and sex dimorphism in heterozygous effects between backcross lines. Given the large S x L interaction for the Dl-H/Dl³ heterozygotes, analyses of variance of abdominal bristle number of these lines were done separately for males and females (data not shown). There was no significant variation among lines for males (² = 0.217, P > 0.05), but highly significant among-line variation for females (² = 1.390, P < 0.001). There was significant variation in sternopleural bristle number among lines for Dl-H/Dl³ and Dl-H/H heterozygotes, but not for Dl-H/Sam heterozygotes. For all three heterozygous genotypes, there was significant variation in heterozygous sternopleural bristle number effects between backcross replicates.

The average degree of dominance, k, of alleles affecting bristle number in the Dl-H region can be estimated from the regression, b, of line means of heterozygous bristle number on homozygous bristle number, as k = 2(b - 0.5) (MACKAY 1987 ). The range of k is from -1 (naturally occurring alleles completely recessive) through 0 (additive gene action) to +1 (naturally occurring alleles completely dominant). The regressions (b ± SE) of heterozygous abdominal bristle number on homozygous abdominal bristle number were 0.348 ± 0.038, 0.522 ± 0.075 and 0.191 ± 0.046 for the Dl-H/Sam, Dl-H/Dl³ and Dl-H/H heterozygous genotypes, respectively, giving respective estimates for k of -0.30 (partly recessive), 0.04 (additive) and -0.62 (partly recessive). The regressions of heterozygous sternopleural bristle number on homozygous sternopleural bristle number were 0.334 ± 0.046, 0.681 ± 0.061 and 0.258 ± 0.121, yielding estimates for k of -0.34 (partly recessive), 0.36 (partly dominant) and -0.48 (partly recessive) for the Dl-H/Sam, Dl-H/Dl³ and Dl-H/H genotypes, respectively.

These estimates of average degree of dominance are unbiased only if there is no error in estimating homozygous effects and the degrees of dominance of alleles are uncorrelated with their homozygous effects (CABALLERO et al. 1997 ). The "reliability ratio," V_L/(V_L + V_U), where V_L is the variance among lines and V_U is the error in estimating homozygous effects, is the amount by which an estimate of b from regression of heterozygous on homozygous effects is biased toward zero (CABALLERO et al. 1997 ). For our data, the reliability ratios, estimated from variance components in Table 3, are quite high (0.95 for abdominal bristle number and 0.93 for sternopleural bristle number) because homozygous line means are each estimated for 80 individuals. Thus, there is little bias in estimates of degrees of dominance from error in estimating homozygous effects. Further, nonlinear (quadratic) regression terms were not significant (data not shown), suggesting that degrees of dominance of naturally occurring alleles affecting bristle number in the Dl-H gene region are not significantly correlated with their homozygous effects.

Quantitative complementation effects:
To detect whether naturally occurring alleles in the Dl-H gene region quantitatively fail to complement major morphological mutant alleles at Dl and H, we determined if there was significant variation in the difference of line means of the Dl-H/Dl³ or Dl-H/H heterozygotes (the Tester cross) and the Dl-H/Sam heterozygotes (the Control cross). Variation in Tester-Control effects (failure to complement) will occur if there is a difference in the degree of dominance of the naturally occurring alleles against the wild-type and mutant standards.

The results of analyses of variance of heterozygous bristle number of the Dl-H region introgression lines, which include the additional cross-classified fixed effect of Cross (C), are given in Table 4. The important new terms in these analyses are the main effect of C and the C x S, C x L and C x S x L interaction terms. The main effect of C tests for a mean difference in bristle number between Dl or H and Sam heterozygotes, averaged over all lines and both sexes. The term was highly significant for both bristle number traits in crosses of the introgression lines to Dl and H, indicating the Dl and H alleles had large heterozygous effects on bristle number. The C x S interaction term tests for an effect of Dl or H on the sex dimorphism of bristle number, averaged over all lines. Both candidate locus mutations strongly affected sex dimorphism of abdominal and sternopleural bristle numbers.

View this table: In this window In a new window   Table 4. Variance components (2) and significance of mean squares (P) from analyses of variance of abdominal (AB) and sternopleural (ST) bristle number from quantitative complementation tests of Dl-H region introgression lines

The C x L and C x S x L interaction terms test for quantitative failure to complement. The C x L term tests whether there is variation in the difference in bristle number between the Tester and Control heterozygous genotypes of the different introgression lines, averaged over both sexes. The C x S x L interaction tests for variation in sex dimorphism of the failure to complement, that is, whether the C x L effects are different in males and females. We observed significant quantitative failure of the introgression lines to complement abdominal and sternopleural bristle effects of the Dl and H alleles. The failure of naturally occurring alleles in the Dl-H gene region to complement the abdominal bristle effect of Dl is complicated. There is complementation when Tester-Control scores are averaged over males and females (the C x L term is not significant), but highly significant variation in failure to complement between males and females (the C x S x L interaction is highly significant). Analyses of variance of each sex separately (data not shown) reveal that failure to complement is observed only in females. The variance component (²) for the C x L term in females was ² = 0.352 (0.001 < P < 0.01), compared to ² = 0.003 (P > 0.05) in males. This is consistent with the observation above that there is variation among Dl-H/Dl³ heterozygous lines for females only. The failure of naturally occurring alleles in the Dl-H region to complement the sternopleural bristle effect of Dl, and the abdominal and sternopleural bristle effects of H, do not show this sex specificity. The magnitude of the failure to complement can be inferred from the C x L and C x S x L variance components. For abdominal bristle number, the sum of the C x L and C x S x L effects is much larger for the Dl than the H tester allele, whereas the opposite is true for sternopleural bristle number; the C x L effect is much larger for the H than for the Dl tester allele, and C x S x L effects are not significantly different from zero. This is consistent with the effects of the Dl and H mutants on bristle number; abdominal bristle number is increased in Dl mutants, and sternopleural bristle number is decreased in H mutants. Variation among line means in Tester - Control (T - C) effects in crosses to Dl³ and H is shown in Figure 2.

View larger version (28K): In this window In a new window Download PPT slide   Figure 2. —Distributions of mean abdominal and sternopleural bristle number complementation effects (T-C) between heterozygous Dl-H/Dl3 or Dl-H/H(T) and heterozygous Dl-H/Sam (C) introgression lines. Solid bars are for male, and open bars for female complementation effects.

Correlations of line means between genotypes:
The correlations of line means for abdominal and sternopleural bristle numbers across the different genotypes are given in Table 5. The whole chromosome 3 line means are not highly correlated with those of homozygous backcross lines for either bristle trait, as expected if other chromosome 3 loci than those in the Dl-H gene region contribute to variation in bristle number. Neither are the chromosome 3 sternopleural and abdominal bristle number line means highly correlated with the Dl-H backcross heterozygotes, for the above reason plus the expected reduction in heterozygous effect compared to homozygous effect for loci that are not dominant or overdominant. The correlations between homozygous backcross bristle numbers and the Dl-H/Sam, Dl-H/Dl³ and Dl-H/H heterozygotes are as expected from the regressions of heterozygous on homozygous bristle numbers (see above). The correlations of bristle numbers between the various heterozygous genotypes are not high, reflecting the variable degrees of dominance of the wild alleles in the mutant allele and Sam background. The Dl³ complementation effect was highly correlated with Dl-H/Dl³ heterozygous effects, but not Dl-H/Sam heterozygous effects, for both bristle traits, presumably because the bristle number alleles in the Dl-H region were more additive/dominant in the Dl³ than in the Sam background. The H complementation effect was positively correlated with Dl-H/H abdominal and sternopleural bristle numbers, but negatively or not at all with the other genotypes. Thus, positive correlations of the T - C H effects with the Dl-H/H bristle numbers mean that the complementation effect is smaller (less negative) in lines with high bristle numbers and larger (more negative) in lines with lower bristle numbers. Conversely, the negative association of the T - C H effects in the other genotypes implies a reversal such that large negative T - C effects are associated with high bristle numbers, and vice versa, which could be a scale effect. Finally, correlations of mean T - C Dl³ abdominal bristle numbers with other genotypes were larger for females than males (Dl-H/Dl³) or only significant for females (C3, Dl-H, Dl-H/Sam and Dl-H/H) (data not shown), consistent with the observed female-specific complementation effect.

Correlations between males and females:
The departure of the genetic correlation (r_G) in bristle number between the sexes from unity indicates whether naturally occurring alleles affecting bristle number have sex-specific effects, that is, whether the genes are expressed and/or have the same effects in males and females. Across-sex genetic correlations are significantly different from one when the S x L interaction terms in the analyses of variance (Table 2 Table 3 Table 4) are significantly different from zero. The S x L variance component is significant for abdominal bristle number for the whole chromosome 3, Dl-H homozygote and Dl-H/Dl³ heterozygote data, where it is, respectively, 29, 42, and 193% times the among-line variance components. The genetic correlation of abdominal bristle number between the sexes was computed for these data from variance components as (ROBERTSON 1959 ), where the subscripts P, M and F refer to the among-line variance components for the sexes pooled and to male and female analyses, respectively (the separate sex analyses are not shown). The estimates were 0.811, 0.763 and 0.499 for the whole chromosome 3, Dl-H homozygote and Dl-H/Dl³ heterozygote genotypes, respectively. Further, the quantitative complementation test to Dl³ revealed highly significant C x S x L interaction variance, indicating sex-specific complementation effects. The genetic correlation between the sexes for the difference in mean abdominal bristle number between the Tester and Control cross (the complementation effect) was computed from variance components as ^1/2 ; the point estimate is 0.304. The genetic correlation between male and female abdominal bristle number in different genetic backgrounds is depicted graphically in Figure 3 as reaction norms; increasing departures of the genetic correlation from unity is manifested as more crossing of reaction norms.

View larger version (62K): In this window In a new window Download PPT slide   Figure 3. —Norms of reaction for abdominal bristle number between males and females in genotypes with significant sex x line interaction variance.

For sternopleural bristle number, the S x L interaction term was significant only for the whole chromosome 3 data, and the magnitude of the variance component was small, only 2% as large as the among-line variance. Consequently, the genetic correlation of sternopleural bristle number between males and females for this genotype, although significantly different from one, was very high (r_G = 0.984). Genetic correlations of sternopleural bristle number between the sexes were not significantly different from one in the other genetic backgrounds.

Correlations between bristle traits:
The correlations between line means for abdominal and sternopleural bristle numbers are significantly different for the whole chromosome 3 and the Dl-H region homozygotes. The correlation between bristle traits is not significantly different from zero for the aggregate of chromosome 3 loci (r = 0.180, averaged over males and females; P > 0.05). However, the traits are moderately, but significantly, positively correlated when only loci in the Dl-H region are considered (r = 0.431, averaged over sexes; P < 0.001). This suggests that pleiotropic effects of naturally occurring alleles affecting bristle number are variable across bristle number QTLs, and that alleles at other chromosome 3 bristle number QTLs have negatively correlated effects on the two bristle traits to compensate for the positive correlation in the Dl-H gene region. The correlations between bristle traits for the heterozygous genotypes are not given because they are confounded with degree of dominance, in the sense that they are increasingly similar to the homozygous Dl-H region correlations as the degree of dominance of the wild-type alleles approaches k = 1 (completely dominant) and are, conversely, increasingly similar to the correlations between these traits in the Sam, Dl³ and H background as the degree of dominance of the wild-type alleles approaches k = -1(completely recessive).

	DISCUSSION

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

Loci in the Drosophila PNS development pathway are, a priori, candidate QTLs at which naturally occurring allelic variation could cause quantitative genetic variation in numbers of adult sensory bristles (MACKAY 1985 ). Various disparate pieces of experimental evidence have specifically implicated the neurogenic loci Dl and H as Drosophila bristle number QTLs. In ascending order of credulity, the evidence is as follows. (1) The earliest efforts to analyze factors affecting long-term selection response for bristle number found, surprisingly, that the long-term selection lines contained high frequencies of third chromosome recessive lethals with large heterozygous effects on bristle number, in the direction of selection (e.g., CLAYTON and ROBERTSON 1957 ; FRANKHAM et al. 1968 ; YOO 1980B ). Mutations at Dl and H are generally recessive lethal with large heterozygous effects on bristle number (LINDSLEY and ZIMM 1992 ), but the lethal mutations in the long-term selection lines were never mapped below the level of chromosome [although YOO 1980B noted that two allelic lethals had a Dl-like wing vein phenotype]. (2) Fine-scale mapping of factors affecting response to divergent artificial selection for abdominal bristle number, from a base population recently collected from nature, revealed a QTL with a large effect on abdominal and sternopleural bristle numbers in a 9-cM interval containing Dl and H (LONG et al. 1995 ). (3) Mutations at Dl and H interact genetically with high and low bristle number selection lines containing naturally segregating (LONG et al. 1996 ) or spontaneous mutation (MACKAY and FRY 1996 ) alleles at bristle number QTLs. Here, we provide further evidence supporting Dl and H as genetic loci at which naturally segregating alleles affect quantitative variation in bristle number: QTLs in the Dl-H gene region both contribute to quantitative variation in bristle number and fail to complement Dl and H mutant alleles.

To test whether naturally occurring variation affecting bristle number is attributable in part to QTLs in the Dl-H gene region, we sampled isogenic third chromosomes from the same natural population used as the base for the selection-based QTL mapping experiment of LONG et al. 1995 and substituted each into the standard Sam inbred strain chromosome 1 and 2 background. Then, from each whole chromosome 3 line, we derived two lines in which the wild-type Dl-H region chromosomal segment was independently introgressed into the Sam chromosome 3 background by 10 generations of backcrossing. After 10 backcross generations, the Dl locus, and a linked chromosomal fragment originating from the wild-derived third chromosome of ~10 cM on either side of Dl, are expected to remain in the otherwise Sam background (CROW and KIMURA 1970 ). The null hypothesis that variation among third chromosomes for bristle number is not attributable in part to variation at loci in the 20-cM interval containing Dl and H is tested by the significance of the among-line component of variation from analysis of variance of bristle number of the introgression lines. For both bristle traits, we reject this null hypothesis and conclude there is significant genetic variation for bristle number in the Dl-H region.

The Drosophila third chromosome is 110 map units long (LINDSLEY and ZIMM 1992 ); therefore, on average the introgressed wild-derived fragments represent 18% of the chromosome. One extreme alternative hypothesis is the infinitesimal model, which assumes the underlying genetic basis for quantitative traits is a large number of QTLs with small allelic effects, evenly distributed over the chromosomes. Under this hypothesis, the prediction is that 18% of the whole chromosome 3 genetic variance for bristle number should be contained in the Dl-H introgressions. For abdominal bristle number, the among-line and sex x line variance of the introgressions are, respectively, 36 and 53% that of the whole chromosome 3 variances—much greater than the infinitesimal model prediction. For sternopleural bristle number, the among-line variance of the introgressions is 8.3% of the whole chromosome 3 variance—less than the infinitesimal expectation. These observations are consistent with previous studies that showed that segregating (BREESE and MATHER 1957 ; SHRIMPTON and ROBERTSON 1988A , SHRIMPTON and ROBERTSON 1988B ) and mutational (MACKAY et al. 1992A ; LYMAN et al. 1996 ) variation for bristle number follows a leptokurtic distribution, with most of the variance attributable to a few loci with large allelic effects and the remainder due to a larger number of loci with smaller effects (ROBERTSON 1967 ).

The extent to which genetic variation for bristle number in the Dl-H introgression lines could be attributable to genetic variation at Dl and H was tested by crossing each introgression line to Sam, the standard inbred line with wild-type Dl and H alleles (the Control cross) and to lines derived from Sam, into which mutations at Dl and H had been introgressed by 20 generations of backcrossing (the Tester cross). The test for genetic interaction of naturally occurring alleles affecting bristle number in the Dl-H region with the mutant Dl and H alleles is whether there is variation among the introgression lines in the difference of means between Tester and Control heterozygotes. The null hypothesis of no interaction was rejected for both bristle traits at both candidate loci. However, the magnitude of the complementation effects varied. For abdominal bristle number, the Dl complementation effect variance component was 2.9x greater than the H complementation effect variance; whereas for sternopleural bristle number, the H complementation effect variance was 3.3x greater than the Dl complementation effect variance. Further, the Dl complementation effect variance was significant only for females.

There are at least four possible genetic interpretations of the quantitative complementation test results. (1) Naturally occurring alleles at Dl have relatively large effects on female abdominal bristle number and smaller pleiotropic effects on sternopleural bristle number, in both sexes. Naturally occurring alleles at H have relatively large effects on sternopleural bristle number and smaller pleiotropic effects on abdominal bristle number, in both sexes. (2) Naturally occurring alleles at Dl affect female abdominal bristle number and naturally occurring alleles at H affect sternopleural bristle number in both sexes. The abdominal bristle complementation effects at H and the sternopleural bristle complementation effects at Dl are due, respectively, to epistatic interactions of the wild Dl abdominal bristle alleles with the H mutation and the wild H sternopleural bristle alleles with the Dl mutation. (3) There is naturally occurring allelic variation affecting both abdominal and sternopleural bristle numbers at either Dl or H, and the significant complementation effect variance at the other locus is caused by epistatic interactions of the mutation at this locus with the wild alleles. (4) Naturally occurring alleles affecting bristle number in the Dl-H region, but not at Dl nor H, interact epistatically with the mutant Dl and H alleles. It is not possible to discriminate among the various allelism and epistasis hypotheses based on complementation test results alone; all four interpretations are plausible. However, it becomes increasingly likely that one of hypotheses (1)–(3) is correct, as the size of the region containing the candidate loci and putative additional interacting loci shrinks. The combined evidence from mapping factors causing response to divergent artificial selection to the Dl-H interval (LONG et al. 1995 ), demonstrating appreciable segregating genetic variation in the region (this article), and failure of selected third chromosomes (LONG et al. 1996 ) and Dl-H introgression lines (this article) to complement mutant Dl and H alleles, strongly motivates a more fine-scale evaluation of the association between molecular polymorphism at the candidate loci and phenotypic variation in bristle number (LONG et al. 1998 ).

Our long-term goal is to understand the genetics and evolution of quantitative traits in terms of complex genetics rather than complex statistics. To do this, it is necessary to know the genetic loci contributing to mutational and segregating variation of the trait, and the additive, dominance, pleiotropic and epistatic effects of spontaneous mutant and naturally segregating alleles, gene frequencies and per locus mutation rates (ROBERTSON 1967 ; BARTON and TURELLI 1989 ). This is a tall order, even for the model system of Drosophila bristle number. The task would be considerably simplified if the leading loci causing most of the variation for the trait had similar genetic properties; one could then concentrate on evaluating the necessary parameters for a representative few. How likely is this to be the case? Or, what is the variation among loci in additive, dominance, pleiotropic and epistatic effects? Unfortunately, the analyses of homozygous, heterozygous and pleiotropic effects of bristle numbers in whole third chromosomes and Dl-H introgression lines hint at an underlying complexity and variation of allelic effects between loci.

Consider first the average degree of dominance of segregating alleles. This is an important parameter because the contribution to genetic variance of segregating alleles depends on their degree of dominance, as well as additive effects and gene frequency (FALCONER and MACKAY 1996 ). Naturally occurring variation for bristle number has been shown to be largely additive, with no significant directional dominance (CLAYTON et al. 1957 ; KIDWELL and KIDWELL 1966 ). However, we have observed that the degree of dominance for bristle number of naturally occurring alleles in the Dl-H gene region varies significantly depending on the allele used as a standard. Average degrees of dominance for both bristle traits range from additive or partly dominant relative to Dl³ and partly recessive relative to H and Sam wild-type alleles. That the degree of dominance is not a fixed property of an allele, but must be considered relative to other alleles, is well known for mutant alleles of large effect, and will need to be taken explicitly into account when predicting the magnitude of variation contributed by segregating alleles at genetic loci affecting quantitative traits. The strong possibility that some fraction of naturally occurring variation for bristle number is due to variation at candidate neurogenic loci raises the issue of variation in degrees of dominance across loci. Mutations of large effect at known bristle loci are rarely strictly additive and tend to be partly recessive or dominant. Furthermore, the average degree of dominance of mutations at bristle loci varies among loci, with, for example, most achaete-scute mutations recessive and most Dl mutations dominant (LINDSLEY and ZIMM 1992 ). It is possible that the observation of additive bristle number effects is an artefact of averaging over loci whose individual dominance deviations are predominantly dominant or recessive.

Phenotypic and genetic correlations between traits estimated from response to selection and resemblance among relatives are also averaged over all loci. The change in the correlation of line means for homozygous abdominal and sternopleural bristle numbers between the whole third chromosome, where the bristle traits were not significantly correlated, and the Dl-H region homozygotes, where the two traits were significantly positively correlated, illustrates the important principle that an overall lack of correlation does not mean lack of pleiotropy. Rather, there appears to be variation in pleiotropic effects for abdominal and sternopleural bristle numbers among the third chromosome loci contributing to variation for these traits, as the positive correlation observed for the Dl-H gene region must be balanced by negative (or no) pleiotropic effects at other loci to reduce the net correlation of traits on the whole chromosome 3. This is also true for the genetic correlation of abdominal bristle number between males and females. Over 50% of the whole chromosome 3 sex x line variance is accounted for by loci in the Dl-H gene region, which implies that, for at least some chromosome 3 loci outside this region, the genetic correlation between the sexes for abdominal bristle number must be higher. Variation in pleiotropic effects between loci is, however, a common feature of mutations at candidate bristle loci: mutations at Dl tend to increase and mutations at H tend to decrease both abdominal and sternopleural bristle numbers, whereas mutations at scabrous, a chromosome 2 candidate bristle locus (LAI et al. 1994 ), decrease sternopleural bristle number in both sexes and increase abdominal bristle number in females only (T. F. C. MACKAY, unpublished results).

An unusual and recently observed feature of segregating (MACKAY and LANGLEY 1990 ; LONG et al. 1995 , LONG et al. 1998 ; this report) and mutational (MACKAY et al. 1992A , MACKAY et al. 1992B , MACKAY et al. 1994 , MACKAY et al. 1995 ; LOPEZ and LOPEZ-FANJUL 1993 ; LYMAN et al. 1996 ; MACKAY and FRY 1996 ) quantitative genetic variation for abdominal, and to a lesser extent, sternopleural, bristle numbers is that homozygous, heterozygous and epistatic effects are often highly sex specific. Although large sex-specific abdominal bristle effects had been reported previously in long-term selection lines (CLAYTON and ROBERTSON 1957 ; FRANKHAM 1968 ; FRANKHAM et al. 1978 ; FRANKHAM 1980 ), they were shown to be caused by mutations at the bobbed locus, which is duplicated on the X and Y chromosomes, and thus the sex-specific effects were not thought to be general. However, sex-specific QTL effects have been observed for Drosophila olfactory behavior (ANHOLT et al. 1996 ; MACKAY et al. 1996 ) and longevity (MAYNARD SMITH 1959 ; NUZHDIN et al. 1997 ) and may be a common property of quantitative variation in this and other species. Past inferences from experimental results for Drosophila bristle number have been accurate predictors of the properties of quantitative genetic variation for equivalent traits in other species (FRANKHAM 1989 ). The general pattern of variation in sex dimorphism observed is that variation among the same set of genotypes is much greater for females than males, perhaps reflecting stronger genetic canalization in males than females. Sex-specific QTL effects are a special case of genotype x environment interaction, where the sexes are the different environments, and could in some circumstances contribute to the maintenance of quantitative genetic variation (GILLESPIE and TURELLI 1989 ). A low genetic correlation between the sexes also facilitates the evolution of further sexual dimorphism (LANDE 1980 , LANDE 1981 ). Further fine-scale mapping of the large, female-specific, Dl³ abdominal bristle number complementation effect should prove particularly interesting.

	ACKNOWLEDGMENTS

We thank FAYE LAWRENCE and LUCY REID for excellent and dedicated technical assistance, and TONY LONG for comments on the manuscript. This work was supported by grants GM-45344 and GM-45146 from the National Institutes of Health.

Manuscript received August 29, 1997; Accepted for publication February 27, 1998.

	LITERATURE CITED

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

ANHOLT, R. R. H., R. F. LYMAN, and T. F. C. MACKAY, 1996  Effects of single P element insertions on olfactory behavior in Drosophila melanogaster. Genetics 143:293-301[Abstract].

ARTAVANIS-TSAKONAS, S., K. MATSUNO, and M. E. FORTINI, 1995  Notch signaling. Science 268:225-232[Abstract/Free Full Text].

BARTON, N. and M. TURELLI, 1989  Evolutionary quantitative genetics: how little do we know? Annu. Rev. Genet. 23:337-370