Identifying the genes underlying genetically complex traits is of fundamental importance for medicine, agriculture, and evolutionary biology. However, the level of resolution offered by traditional quantitative trait locus (QTL) mapping is usually coarse. We analyze here a trait closely related to fitness, ovariole number. Our initial interspecific mapping between Drosophila sechellia (8 ovarioles/ovary) and D. simulans (15 ovarioles/ovary) identified a major QTL on chromosome 3 and a minor QTL on chromosome 2. To refine the position of the major QTL, we selected 1038 additional recombinants in the region of interest using flanking morphological markers (selective phenotyping). This effort generated approximately one recombination event per gene and increased the mapping resolution by approximately seven times. Our study thus shows that using visible markers to select for recombinants can efficiently increase the resolution of QTL mapping. We resolved the major QTL into two epistatic QTL, QTL3a and QTL3b. QTL3a shows sign epistasis: it has opposite effects in two different genetic backgrounds, the presence vs. the absence of the QTL3b D. sechellia allele. This property of QTL3a allows us to reconstruct the probable order of fixation of the QTL alleles during evolution.
MOST phenotypic variation within and between species appears to be controlled by multiple genes. Such genes with quantitative effects are more difficult to identify than genes underlying Mendelian traits because of their low penetrance, their environmental sensitivity, and epistasis with other loci (Glazier et al. 2002). Interval mapping (Lander and Botstein 1989) and its variants [collectively named quantitative trait loci (QTL) mapping] (reviewed in Manly and Olson 1999; Broman 2001) have been used extensively to map QTL. In such experiments, inbred strains differing in traits of interest are crossed to generate a mapping population. QTL are identified and localized to chromosomal regions by correlating segregating genetic markers with trait values. This approach typically identifies large genomic regions that do not allow identification of the genes involved. The major problem is that an increasingly large number of recombination events is required to map the QTL to an increasingly small interval. Several strategies to increase the number of recombination events in the mapping population, either by using multiple generations of crossing or by selecting for recombination events at targeted loci, have been proposed (reviewed in Darvasi 1998, 2005). Here we exploit selective phenotyping, in which individuals that are recombinant in a region known to contain a QTL are selected for phenotyping (Darvasi 1998, 2005; Ronin et al. 2003; see also Jin et al. 2004; Xu et al. 2005). Such selective phenotyping requires only two generations of crossing to produce the mapping population, but a large number of progeny individuals must be screened. This strategy is commonly employed in mice to narrow QTL regions (see for example Purcell et al. 2001). Flies of the genus Drosophila are ideally suited to this approach because they can be reared en masse and morphological markers can be used to facilitate the screening for recombinants. This technique, also known as meiotic recombination mapping, is extensively used in D. melanogaster to identify genes underlying Mendelian traits. However, to our knowledge, selective phenotyping has not yet been used for QTL mapping in Drosophila. Our analysis highlights some of the advantages and pitfalls of this approach for mapping QTL.
Ovarioles represent the functional units of insect ovaries; they are the tubes in which eggs undergo maturation (Mahowald and Kambysellis 1980). Ovariole number is determined during the early pupal stage (King et al. 1968; Hodin and Riddiford 2000). Ovariole number is a fertility trait that is closely related to fitness; the more ovarioles, the more eggs females can potentially lay. Similar latitudinal clines in ovariole number have been found on four continents (David and Bocquet 1975a,b; Lemeunier et al. 1986; Capy et al. 1993; Azevedo et al. 1996; Gibert et al. 2004), suggesting that ovariole number is under contemporary natural selection. Variation in ovariole number is controlled by several loci, both within D. melanogaster (at least five QTL, Coffman et al. 2003; Wayne et al. 2001; Wayne and McIntyre 2002) and between species of the melanogaster species subgroup (at least three QTL, Coyne et al. 1991). Ovariole number is sensitive to environmental conditions such as temperature (David and Clavel 1967; Delpuech et al. 1995; Moreteau et al. 1997; Morin et al. 1997; Hodin and Riddiford 2000), rearing density (Robertson 1957), and larval nutrition (Hodin and Riddiford 2000), which complicates QTL analysis by introducing nongenetic variation.
D. sechellia has approximately half as many ovarioles as its closest relatives (Figure 1A), suggesting that ovariole number has decreased in the evolutionary lineage leading to D. sechellia. Such a dramatic decline in potential fecundity is possibly adaptive. D. sechellia is endemic to the Seychelles Islands where it feeds exclusively on the freshly dropped fruits of Morinda citrifolia. These fruits are highly toxic to other Drosophila species (Rkha et al. 1991), although the level of toxins declines as the fruits rot. Ecological specializations in Hawaiian drosophilids and Tephritid flies in the genus Dacus have previously been associated with reductions in ovariole number (Kambysellis and Heed 1971; Fitt 1990; Kambysellis et al. 1995). The possible advantages of such reductions in ovariole number are not known but may be related to larger egg sizes (Kambysellis and Heed 1971; Montague et al. 1981; Fitt 1990). Thus, D. sechellia specialization on a brief temporal niche during fruit rotting might have favored a dramatic reduction in its ovariole number.
Using four genetic markers, Coyne et al. showed that both chromosomes 2 and 3 harbored QTL controlling the ovariole number difference between D. sechellia and D. simulans (Coyne et al. 1991). The X chromosome had no effect. We performed an initial QTL mapping experiment to map more precisely the main regions responsible for the decrease in ovariole number in D. sechellia. We then used selective phenotyping to increase the resolution in the major QTL region.
MATERIALS AND METHODS
Strains and crosses:
Female f;nt,pm;st,e D. simulans flies (kindly provided by C. Jones) were crossed to male D. sechellia flies (Tucson Drosophila Species Stock Center strain 14021-0248.07) and the female progeny were backcrossed to D. simulans males (D. simulans backcross) or D. sechellia males (D. sechellia backcross). Flies were raised on standard media at 25°. Since ovariole number is sensitive to nutrient conditions (Hodin and Riddiford 2000), flies were reared in uncrowded conditions.
Ovaries were removed from 2- to 4-day-old females. Ovarioles were dissected with tungsten needles in phosphate-buffered saline + 0.1% Tween and counted under a Nikon SMZ1500 stereoscopic microscope. For each fly, the mean ovariole number of the left and right gonad was calculated. Only flies for which both ovaries were scored were included in the analysis.
Following dissection, flies were frozen at −80°. DNA was isolated from frozen individuals (Gloor and Engels 1992). Molecular markers (supplemental Table 1 at http://www.genetics.org/supplemental/) were PCR amplified and separated on 2% agarose or 4.5% agarose SFR (AMRESCO). We scored natural variation in sequence length or differences in restriction enzyme sites.
Genetic marker map:
We used the genetic map previously determined from D. simulans/D. sechellia hybrids (Macdonald and Goldstein 1999) (supplemental Table 1 at http://www.genetics.org/supplemental/). For markers not included in the original map, we estimated their genetic map position on the basis of their physical map position in D. melanogaster relative to flanking markers. The large inversion on chromosome 3 relative to D. melanogaster (84F1-93F6-7) that is present in both species (Mahowald and Kambysellis 1980) was taken into account.
The distribution of markers and our sample size allowed us to identify every QTL responsible for a minimum of a one-ovariole effect (and thus a two-ovariole difference between parents if the QTL is additive) (calculation not shown, adapted from Soller et al. 1976; Soller and Genizi 1978; Darvasi and Soller 1992; Lynch and Walsh 1998). This detection limit is in the worst case of a QTL located midway between two markers. In most genomic locations, QTL with lower effects were potentially detectable. To avoid any bias related to selective genotyping (Lander and Botstein 1989), data from the nonextreme progeny were included and their molecular marker genotypes were entered as missing. Composite interval mapping was performed using R/qtl (Broman et al. 2003) with the multiple imputation method of Sen and Churchill (2001). Background markers were chosen at the location of the maximum LOD score calculated by simple interval mapping. Background markers were included only if located on a different chromosome than the test position (see supplemental material at http://www.genetics.org/supplemental/). Statistical significance was determined by permutation tests (Churchill and Doerge 1994). Two- and three-dimensional scans for epistatic QTL were performed using the scantwo and scanqtl functions of R/qtl, respectively (Broman et al. 2003). To test for additional QTL, the maximum LOD score was compared between a model containing an additional epistatic QTL and a model without any additional QTL. An LOD difference <2 was considered as not significant. To test for epistasis between QTL, the model allowing for an interaction between the QTL was compared to the additive QTL model. With the present data, evidence for interaction between QTL3a and QTL3b was so clear that the formal calculation of a P-value was deemed unnecessary.
The mapping strain of D. simulans has a mean ovariole number of 14.9 ± 0.1 (SE) and D. sechellia has a value of 7.8 ± 0.1 ovarioles (Figure 1, A–C). F1 hybrids have a mean of 12.4 ± 0.1 ovarioles (Figure 1C). This value is not significantly different from the mean of the D. simulans and D. sechellia values, as previously observed with other strains of D. sechellia and D. simulans (Coyne et al. 1991). This suggests that the ovariole difference involves alleles with additive effects or several alleles with opposite dominant effects in each species.
To survey the genomic regions responsible for the difference in ovariole number between D. sechellia and D. simulans, we performed composite interval mapping on separate D. sechellia and D. simulans backcross populations. As we were interested primarily in identifying the major QTL, we chose to selectively genotype the progeny flies exhibiting extreme phenotypes, i.e., with lowest and highest ovariole numbers (selective genotyping, Lander and Botstein 1989). We genotyped ∼42% of 226 progeny for the D. sechellia backcross and 25% of 383 for the D. simulans backcross. Selective genotyping reduces cost and time and efficiently detects major QTL (Lander and Botstein 1989; Darvasi and Soller 1992). However, the estimates of QTL position and effect are usually less precise than total genotyping and the identification of epistasis can be more difficult. We estimate here the effect of a QTL as the effect of substituting one D. simulans allele with a D. sechellia allele. Since D. sechellia contains fewer ovarioles than D. simulans, QTL are expected to have negative effects on ovariole number if they follow the general direction of evolution.
In both backcrosses the marker on the fourth chromosome was not significantly associated with the trait and results from this chromosome, which comprises only ∼1% of the genome, are not shown.
D. sechellia backcross:
Results for the D. sechellia backcross are shown in Figure 2A. One region of chromosome 3 has the largest effect (a difference of −1.24 ovarioles between a D. simulans/D. sechellia heterozygote and a D. sechellia homozygote). Another QTL is detected on chromosome 2 with an effect of −0.58 ovarioles (LOD = 2.74, permutation-based LOD threshold = 1.15). No QTL is detected on chromosome 1. Tests for additional QTL on chromosome 2 or 3 are not significant (LOD = 0.66 for chromosome 2 and LOD = 0.87 for chromosome 3). Estimates of QTL positions and effects are summarized in Table 1. No significant epistatic interaction is detected between QTL (LOD = 0.67) and no further QTL exhibiting solely epistatic effects were identified (not shown).
D. simulans backcross:
Initial analysis of the D. simulans backcross using the five visible markers suggested that a major QTL resides between scarlet (st) and ebony (e) on chromosome 3 (not shown). Therefore, to increase the resolution of this QTL, 70 additional st–e recombinant flies from the original backcross were scored (see supplemental material at http://www.genetics.org/supplemental/). Composite interval mapping with all markers revealed a major QTL on chromosome 3 with an effect of −0.92 ovarioles (Figure 2B, LOD = 15.06, permutation-based LOD threshold = 1.49). Two minor peaks were also detected on chromosome 2 with effects of −0.48 and −0.44 ovarioles, respectively (Figure 2B). However, a test for two linked QTL on chromosome 2 is not significant (LOD = 0.67). We cannot, therefore, conclude that there are two QTL on chromosome 2. No significant epistatic interaction was detected between QTL (LOD = 0.34) and no further QTL exhibiting solely epistatic effects were identified (not shown).
High-resolution QTL mapping of the main region on chromosome 3:
To increase the resolution of the major QTL region on chromosome 3, we screened a large number of progeny flies for those that were recombinant between the morphological markers st and e in the D. simulans backcross population. The conspicuous phenotypes of the st (bright red eyes) and e (dark brown body) markers facilitated the screen. We estimated the recombination rate between st and e from our first backcross as 14.5%. We selected 1038 recombinants and thus screened ∼7158 flies. Because the resolution of the QTL position is theoretically proportional to the number of analyzed progeny flies (Darvasi et al. 1993; Dupuis and Siegmund 1999; Visscher and Goddard 2004) the resolution is expected to increase by a factor of seven within the st–e region. We then selectively genotyped the progeny with extreme phenotypes (48% of 1038 flies).
Composite interval mapping using only the recombinants from this second D. simulans backcross indicates that the QTL responsible for the decrease in ovariole number in D. sechellia is located near e (Figure 3A). In the second backcross, the LOD score peak over st with precisely the opposite effect is an artifact resulting from the fact that every fly that is heterozygous for st is homozygous for e, and vice versa. Composite interval mapping on the pooled data from the first and second backcrosses reduces the artifactual peak near st and improves the resolution of the peak (Figure 3A). The new estimated effect of the main QTL on chromosome 3 (−0.73 ovarioles) is lower than the estimate obtained from the first backcross experiment (−0.92), as might be expected with the use of a larger mapping population (Broman 2001).
The selected st–e recombinants were also genotyped for several markers on chromosome 2. Two peaks are observed on the right arm of chromosome 2 (Figure 3B) but a test for two QTL on chromosome 2 is not significant (LOD = 1.08). We cannot, therefore, conclude that there are two QTL on chromosome 2. The estimated position and effect of the QTL on chromosome 2 is given in Table 1.
The major QTL splits into two QTL with epistatic interactions:
From the pooled data of both D. simulans backcrosses, a model with two epistatic QTL on chromosome 3 (and one QTL on chromosome 2) gives a maximum LOD score of 40.9, whereas a model with a single QTL on chromosome 3 (and one QTL on chromosome 2) gives a maximum LOD score of 33.7. The LOD difference of 7.2 between these models provides strong evidence that there are actually two linked QTL on chromosome 3. To determine the positions and effects of the linked QTL, we performed a two-dimensional scan for QTL on chromosome 3, while simultaneously controlling for the effects of the QTL on chromosome 2 (Figure 4A). The highest LOD score is found at the intersection of 82C and 93E (see cross in Figure 4A) and the 2-LOD support interval delimits a region around 82C that we named QTL3a and a region around 93D named QTL3b (Figure 4A). A test for epistasis between the linked QTL is highly significant (LOD = 5.5, Figure 4B), but no epistasis is detected between the QTL on chromosome 2 and QTL on chromosome 3 (LOD = 0.16 with QTL3a and LOD = 0.41 with QTL3b). The D. sechellia allele of QTL3b decreases ovariole number in the absence of the QTL3a D. sechellia allele (Figure 4D), with an effect of −0.36 ovarioles at the highest LOD score position. Surprisingly, QTL3a acts in the opposite direction in the absence of the QTL3b D. sechellia allele, with an effect of +0.15 ovarioles at the highest LOD score position (Figure 4C). When both QTL are combined, they have a total effect of −0.88 ovarioles at the highest LOD score position (Figure 4E). The estimated effects and positions of QTL are summarized in Table 1. A test for a third QTL on chromosome 3 is not significant (LOD = 0.80).
QTL controlling ovariole number:
Our study illustrates how finely one can map interspecific QTL of large effect for a quantitative trait that is susceptible to considerable environmental variation, on the basis of a backcross mapping population. We identified a minimal number of three QTL (if the same QTL are detected with both backcrosses) that account for ∼42% of the total species difference (Table 1). The remaining 58% is due either to additional loci of smaller effect or to epistatic interactions that are not detectable in backcross conditions between the identified QTL.
Our results agree with previous lower-resolution mapping experiments between D. simulans and D. sechellia that identified a major effect on ovariole number near ebony (93C-D) and a smaller effect linked to cinnabar (43E) in a D. simulans backcross (Coyne et al. 1991). The previous study used different fly lines than our experiment, suggesting that the QTL that we have identified control the species difference in ovariole number and do not reflect simply strain differences. Our interspecific QTL for ovariole number are largely distinct from the intraspecific QTL for ovariole number that have been identified in D. melanogaster, with the possible exception of the 65A–87F region (Wayne et al. 2001; Wayne and McIntyre 2002; Coffman et al. 2003).
It has been suggested that evolutionary changes at the ovo/shaven-baby locus might have contributed to the evolutionary decrease in ovariole number in D. sechellia (Jones 2005). This hypothesis was motivated by the observation that evolution of the ovo/shaven-baby gene caused the loss of larval trichomes in the D. sechellia lineage (Sucena and Stern 2000). Since ovo/shaven-baby is required for oogenesis in D. melanogaster (Mevel-Ninio et al. 1995), Jones hypothesized that evolutionary changes at this locus might have pleiotropic effects on larval trichome patterning as well as on ovariole number. We can now reject this hypothesis for intermediate or large effects of the ovo/shaven-baby gene on ovariole number since we found no QTL for ovariole number on chromosome 1, the location of the ovo/shaven-baby locus.
Traditional QTL mapping, which utilizes every individual from a mating design, usually provides limited resolution of QTL locations. This is due largely to the limited number of recombination events that are scored. Selective phenotyping, in which only individuals carrying a recombination event in a region of interest are selected for phenotyping, has been predicted to greatly improve the resolution of QTL mapping (Darvasi 1998). We used selective phenotyping to better resolve the major QTL in the st–e region.
We phenotyped 1038 st–e recombinant flies. There are ∼1150 genes between st and e, so we generated approximately one recombination event per gene in this region. Theoretically (Darvasi et al. 1993; Dupuis and Siegmund 1999; Visscher and Goddard 2004) this should refine the confidence interval for the QTL position by a factor of seven. We indeed observed a significant increase in the QTL mapping resolution. The initial QTL on chromosome 3, estimated to be responsible for 13% of the parental difference, was first mapped to a 2-LOD support interval of 20 cM (not shown). Then, using selective phenotyping, the QTL3a region was mapped to a region four times smaller (5 cM) although it accounts for only 2% of the parental difference in the absence of the QTL3b D. sechellia allele (Table 1). Selective phenotyping is therefore an efficient method to increase resolution and power to detect QTL of small effect.
Unfortunately, our st–e recombinants do not help to resolve the QTL3b region that extends to the right of e. Correspondingly, the 2-LOD support interval for the position of QTL3b is large on the right of e and relatively small on the left of e (Figure 4A), with only 21 genes on the left of the highest LOD score peak. This suggests that a greater resolution would have been obtained if a visible marker located to the right of e had been used instead of e. This observation illustrates an inherent tradeoff in selective phenotyping. Markers must be close enough to allow the exclusion of many uninformative individuals, but they should also be far enough apart so that the flanked region covers the potential QTL. In our study, we mistakenly assumed that the QTL would lie within the st–e region because the initial highest LOD score peak was actually caused by two closely linked QTL. One solution to this problem is to use distant markers in a first step to refine the region and then closer markers in a second step.
Our selective phenotyping strategy can be applied to other genomic regions (there are ∼50 available visible markers across the D. simulans genome at the Tucson Drosophila Species Stock Center) and to other organisms that can be raised en masse in the lab. For species with a limited number of existing visible mutations, transgenes carrying visible reporter constructs could be used as an alternative.
Identifying the genes underlying QTL:
The identification of the genes and mutations underlying QTL, which appears to be extremely difficult, is nevertheless essential to improve our understanding of the genetic basis of complex traits. The reduced ovariole number in D. sechellia is associated with a slower rate of cell proliferation in the ovaries during the third instar larval stage (Hodin and Riddiford 2000). We observed no difference in the number of embryonic ovary cells between D. sechellia and D. melanogaster (V. Orgogozo and D. L. Stern, unpublished data). This suggests that the genes involved in the decrease in ovariole number in D. sechellia are likely to regulate cell proliferation or cell death. Each QTL region that we identified contains several such genes (Table 1). The insulin receptor gene, which regulates cell size and cell number in individual organs (Brogiolo et al. 2001), is an attractive candidate gene for the QTL3b region. It falls directly beneath the QTL3b peak and ovariole number is reduced in D. melanogaster mutants of the insulin pathway (Tu and Tatar 2003; Richard et al. 2005; V. Orgogozo and D. L. Stern, unpublished data).
Several approaches can be undertaken to identify the genes underlying the QTL for ovariole number (reviewed in Flint and Mott 2001). For example, candidate genes can be tested via interspecific transgenesis. Unfortunately Drosophila transgenesis is currently not feasible for genes that are longer than 40–50 kb (Venken and Bellen 2005), such as the insulin receptor gene. Alternatively, classical meiotic recombination mapping could be pushed further to identify the genes underlying QTL. First, a QTL should be isolated from the effects of other QTL by introgressing only the region of interest. Then, using markers that closely flank the introgression, a large number of recombination events could be identified. Measuring a large number of individuals for each recombination event (progeny testing) should allow identification of the causal genes and possibly nucleotides. An unresolved problem for genes with small effects and for traits that are sensitive to environmental variation such as ovariole number is that large sample sizes are required to estimate the effect of a genomic region. A second unresolved problem is that such an introgression approach is unlikely to identify epistatic QTL such as QTL3a, which causes a small increase in ovariole number in the absence of other QTL. Thus, a combination of approaches that search for QTL at multiple hierarchical levels of resolution is required to fully elucidate the genetic basis of quantitative traits.
Epistasis and evolution:
There is increasing evidence that epistatic interactions play an important role in the expression of complex traits (see, for example, Nagel 2005). Detection of epistasis generally requires high-resolution mapping studies (see for example Gadau et al. 2002; Brem et al. 2005; Brem and Kruglyak 2005). Selective phenotyping enabled us to detect two linked epistatic QTL. The QTL3a D. sechellia allele increases ovariole number in the absence of the QTL3b D. sechellia allele, but decreases ovariole number in the presence of the QTL3b D. sechellia allele. This type of epistasis, in which an allele has opposite effects on two different genotypic backgrounds, has been named sign epistasis (Weinreich et al. 2005). This is in contrast with synergistic epistasis (the effect of two loci is higher than the sum of their individual effects) and antagonistic epistasis (the effect of two loci is lower than the sum of their individual effects, but still in the same direction).
Because sign epistasis introduces limitations on the selectively accessible mutational trajectories across a fitness landscape (Weinreich et al. 2005), it is possible to order past evolutionary events. If we assume that the ancestor of D. simulans and D. sechellia was D. simulans-like regarding ovariole number, and that alleles that increase ovariole number decreased fitness during D. sechellia evolution, it is unlikely that the D. sechellia QTL3a allele (which slightly increases ovariole number in the absence of QTL3b) appeared before QTL3b. It is more likely that the D. sechellia QTL3b allele appeared and was fixed first and then the QTL3a allele arose (Figure 5). It is also possible that both D. sechellia alleles segregated in the ancestral population at the same time and that the generation of a QTL3a + QTL3b super-allele by recombination allowed rapid fixation of the pair of QTL. In any case, our results suggest that the D. sechellia QTL3a allele could not have been selectively fixed in the absence of the D. sechellia QTL3b.
Our results also have implications for use of the Orr Sign Test (Orr 1998) for detecting positive selection using QTL mapping data. This test determines whether the number of QTL with effects in one direction is greater than expected under a neutral model. An excess of QTL with effects in the direction expected, given the parental difference, suggests that directional selection operated on the trait. The possibility that sign epistasis may exist between QTL will complicate interpretation of the Orr Sign Test. In particular, sign epistasis may bias the test toward suggesting that directional selection did not act when in fact it did. This is because during evolution a QTL showing sign epistasis might have had an effect opposite that of its calculated effect if it happened to be in a different genetic background. It is not yet clear how often QTL underlying complex traits display sign epistasis (Kroymann and Mitchell-Olds 2005; Weinreich et al. 2005). Investigating sign epistasis between closely linked loci will require QTL mapping studies with higher resolution than is usually obtained.
We thank C. Jones and the Tucson Drosophila Species Stock Center for flies, T. Doak for the multichannel pipette, and D. Storton for the use of the high-throughput liquid handling Beckman Coulter Biomek FX and MJ Research Tetrad thermocyclers. We thank L. Kruglyak and M. Rockman for providing useful comments on this manuscript. V.O. is a Damon Runyon Fellow supported by the Damon Runyon Cancer Research Foundation (DRG-1789-03). K.W.B. acknowledges the National Institutes of Health (grant GM-074244) and D.L.S. acknowledges the National Institutes of Health (grant GM-6362201), the David and Lucile Packard Foundation, and Princeton University for financial support.
Communicating editor: G. Gibson
- Received November 30, 2005.
- Accepted February 11, 2006.
- Copyright © 2006 by the Genetics Society of America