An Introgression Analysis of Quantitative Trait Loci That Contribute to a Morphological Difference Between Drosophila Simulans and D

Copyight 0 1997 by the Genetics Societv of America

An Introgression Analysis of Quantitative Trait Loci That Contribute to a Morphological Difference Between Drosophila simulans and D. mauritiana

Cathy C. Laurie, John R. True, Jianjun Liu and John M. Mercer Department of Zoology, Duke University, Durham, North Carolina 27708 Manuscript received August 12, 1996 Accepted for publication October 29, 1996

ABSTRACT Drosophila simulans and D. maum'tiana differ markedly in morphology of the posterior lobe, a male- specific genitalic structure. Bothsize and shapeof the lobe can be quantifiedby a morphometric variable, PC1, derived from principal components and Fourier analyses. The genetic architectureof the species difference in PC1 was investigated previously by composite interval mapping, which revealed largely additive inheritance, with a minimum of eight quantitative trait loci (QTL) affecting the trait. This analysis was extended by introgression of marked segments of the maun'tiana third chromosome into a simulans background by repeated backcrossing. Thetwo types of experiment are consistentin suggesting that several QTL on the third chromosome may have effects in the range of 10-15% of the parental difference andthat all or nearly all QTL have effects in the same direction. Since the parental difference is large (30.4 environmental standard deviations), effects of this magnitude can produce alternative homozygotes with little overlap in phenotype. However, these estimates may not reflect the effects of individual loci, since each interval or introgressed segment may contain multiple QTL. The consistent direction of allelic effects suggests a history of directional selection on the posterior lobe.

HE genetic analysis of a quantitative trait usually ual genes that affect continuously variable traits. The T begins with a pair of strains that differ widely in new tools are of two types. One type consists of experi- the value of the trait. These strains are intercrossed to mental techniques for the detection and convenient produce segregating populations, which then are ana- assay of molecular marker variation, which allows local- lyzed to estimate the number,locations and distribution ization of QTL due to cosegregation in F2 or backcross of effects of the quantitative trait loci (QTL) that con- populations (reviewed by TANKSLEY1993). The second tribute to the parental difference. The results of this type of tool consists of statisticaltechniques forinterval analysis may provide inferences about the evolutionary mapping, which utilize molecular marker data to local- processes that caused the trait difference between the ize and estimate the effects of QTL (LANDER and pair of strains being analyzed and possibly about the BOTSTEIN1989; ZENC 1994 and others). genetic structure of their common ancestral popula- The new QTL techniques have been utilized exten- tion. For example, analysis of a cultivated plant and its sively to analyze cultivated plants and their wild relatives wild ancestor may reveal the types of genetic variants (TANKSLEY1993), with interesting implications for crop that were utilized during the historical process of do- evolution under domestication (e.g., DOEBLEYet al. mestication. In contrast,analysis of pairs of natural spe- 1995). However, so far therehave been only a few inter- cies may provide information about thetypes ofgenetic val mapping studies motivated by an interest in natural variants that become fixed under natural (rather than populations. For example, LONGet al. (1995) have ana- artificial) selection, operating over long time periods. lyzed high and low bristle number lines of Drosophila, Analysis of divergent lines produced by short-term arti- produced by short-term artificial selection, to make in- ficial selection may reveal information about standing ferences about standing variation in a base population. variation in the base population, whereas such informa- In addition, BRADSHAWet al. (1995) have analyzedfloral tion may be obscured in theanalysis of population pairs differences between Mimulus species to understand the that have been isolated for long periodsof time. Thus, genetic basis of isolating mechanisms. each situation potentially provides novel types of infor- We have initiated a detailed genetic analysis of a dif- mation about quantitative trait evolution. ference in male genital morphology between species of In recent years there has been a resurgence of inter- Drosophila (LIUet al. 1996).Two closelyrelated species, est in quantitative trait evolution because of the devel- D. simulans and D. mauritiana, differ markedly in the size and shape of the posterior lobe of the genital arch opment of new tools for detectingand mapping individ- (Figure 1). This morphological difference can be quan- tified by a morphometric descriptor (PC1) based on Corresponding author: Cathy C. Laurie, DCMB/Zoology, Box 91000, Duke University, Durham, NC 27708. elliptical Fourier and principal components analyses. E-mail: [email protected] Previously, the genetic architecture of this trait was in-

Genetics 145 999-348 (Frbrnay, 1997) 840 C. C. Laurie el (IC.

FIGURE1.-The ventral portion of the genital arch from D. simulnns and D. mauritiuna. The lateral plate is covered with bristles, while the posterior lobe has only a single bristle at its base.

D. simulans D. mauritiana vestigated by hybridizing inbred lines of each species to in flies to be used for morphological analysis. Control produce two backcross populations of -200 individuals introgressions were also performed by repeated back- each, which were analyzedjointly by composite interval crossing of the mauritiana host strain (without P ele- mapping (ZENG 1994) with the aid of 18 marker loci. ments) to assess the effects of any residual maun'tianu The parental lines showed a large difference in PC1 material that is not associated with the selected P ele- (30.4 environmental SD) and the markers accounted ment. Thedifference in posterior lobe morphology be- for >80% of the phenotypic variation. Among 15 inter- tween the P-marked introgression lines and the controls vals analyzed, eight showed convincing evidence of gives information about the magnitude of QTL effects. QTL, and the range of estimated QTL effects was 5.7- This information provides a useful comparison with es- 15.9% of the parental difference. Although there was timates from interval mapping, which is based on a very some evidence for partial dominance of maun'tiana al- different methodology. In addition, the introgression leles and for epistasis, the pattern of inheritance of PC1 lines allow a direct visualization of the effects of small is largely additive. chromosomal segments on posterior lobe morphology. Here we extend our previous study of posterior lobe morphology by supplementingthe interval mapping re- MATERIALSAND METHODS sults with an introgression analysis. The focus of this Drosophila stocks: The experiment was initiated with a ana1ysis is the third chromosome^ which appears to 'On- 7uhi/p mutant stock of each species, both provided by J. A. tain the greatest nw~~berof QTL, as well as those QTL COYNE.From the initially outbred simulnns 7u stock ( 18w), an with the largest effects. The basic approach is to intro- inbred line (13wI) was derived by 20 generations of full sib gress marked segments of the maun'tiana third chromo- mating and subsequentlymaintained by mass transfer. AIl crosses were carried out on standard cornmeal-molasses me- SOme into a simulans genetic background. A total of 48 dium at room temperature. positions throughoutthe chromosome were tagged Construction and cytogenetic analysis of mauritiana stocks with a P-dement insertion containing a selectable eye Gth Rw+]insem The mari'liana 7u stock was transformed color marker (Figure 2). Each maun'liann insert stock, by embryo injection with a Pelement containinga mini-white which contains a single Pelement, was hybridized and gene, 13Lac-70+1 (BIER d a[. 1989), to produce a set of single- then backcrossed for 15-24 generations to a simulans insert transformant stocks. These insertions are stable because of lack of endogenous Pelements.Each Pinsert was localized strain, producing a set Of marked introgression lines* by in silu hybriclization to the polytene chromosomes(Figure Then the Pekment (and associated maudianachromo- 2) and the recombination fraction between pairs of inserts at soma1 segment) in each line was made homozygous adjacent locations was estimated toproduce a genetic map.

:wv vvv VWT" v :I: v v vvvw w wwv v vv v vv vv v f v 61 162 163 lU k5 1% 167 168 169 170 b1172b3 b4h5 b6 b7 h8 b91.&183 1841 931 921 911 901 891 881 871 861 851{)94 195 1% 197 198 199 I I I\ Id0 80 81 3rd chromosome FIGURE2.-Cytogenetic locations of the p[ru'] insertions introduced into a w mutant strain of D. mnuritiana. A total of 48 third chromosome insertions at 42 distinct locations are represented. Thejagged lines in 3R represent a fixed inversion difference between D.mnuritiana and D. melanogastpr; otherwise, those two species have homosequential polytene chromosomes (LEMEUNIER and ASHBURNER 1976). Introgression Xn;dysis in DrosopI1ila 3-1 1

D.nmuririona mating schcmc~samples a single introgrcwerl segtnct1t from D. simslons H' - one male pel- srthline and makes it homozygotts in thc malrs P[loc-w+l insertion- V that \vu"' rrsctl for nlorphological analysis (csccpt for cross- overs that occur in fc~tnalcsduring thc 1;~cross). D..simulons D m,t,t,iwta A total of 21 I~;~ckcrosscontrol srrhlinrs ~vet-calso cotl- -/X =& strrtctcd by using the same introgrcssion proccrlurc. csccp that thrl-r was no I' clcnlcnt in thc vwuri/i/rj//tliw that \\':IS FI lhyhrd crossed to .xiwtt/urr.s in thr first generation and, thct-cfixc.110 R-/x =dY cyc color markrr to selcct each gcncration. ~\tnongtllcse RE!!% x BdY 21 suhlincs, 11 went tht-ough tllc s;ltnc mating schemr and I maintcnancc rcgimc ;IS the fisctl sul~lincs;Ind IO u~nt repealedbackcrossing I for 15 generations I through thr samc rcgimc ;IS thc nonfisctl sublines. I Sampling males for morphological analysis: Two s(*ts of I lines wcrc sntnplcdat tlifli~renttimcs to obtain males for Inor- J. I addilional I phologic;11 analysis. The first sct consists of the -18 origind R~Xl"-Ic? hackcrosses J v/rtrtri/iuurt insert lincs ;IS well ;IS thr tw7ttri/inno TI! host strain JI f'iom which those lines were dcrivcrl. Thr second s('t consists RdbXL'ydY 8- X e1 d of t11e IO^ fisetl ;1nt1 61 nonfiserl introgression sul>Iincs,the JI JI 21 backcross control suhlinrs and thc recurrent .simr/nnx 71' R*X"-dY Rex pi!!i+a p;lrcnt;ll stock ( IR~olltl)rrtl) usrtl in t11r Ixtckct-ossing. FOI- thr vmrtri/frtnrtrcrantl ,sinrrt/rtns rcrlines, five males wt~~sampled from rach of thrre replicate vials (1.5 males per line). For a11 homozygous othcr lines, three replicate ~ialswere set up and an elTot-t 1 sampled fordY 1 55s morphological analysis e was matlr to sample two males per vial. Actual sample sizc-s avcragc~l6.0 for thr nwwi/irtnn insert lines and the 1,ackcross Fixed suhlincs Nonfixed suhlincs control srthlincs, 5.8 Ihr the fisrd introgrcssion sublines and F~~;~~~~~ TI^^. prOccr~urefor introgressing ;1 pm;lrketl 4.6 for thc nonfixctl introgression sublines. scgment of the D. tn/rrtri/irtwt genomc into ;I .tincdrttrs Ix~ck- Thr mini-\vhitr gene usrtl in this rxperimcnt has wcak cx- grorlnr~anrl tllrn n,aking t~lalsegmellt ~lolnozygolls. TI^^ 1ri- pression that \arics due to chromosomal position cflccts. In ;Ingle synll,ol represents a p[lrtr-7t,j1 illscrtion. p,otll tllc ,s;,,,- most C;ISCS. cyc color is semidominant so thflies homozy- rt/,t,i,s ;llltl t,rctrtr;/;(t,l(t strains llsctl :lr(. ,11111~1111 ;11 tllr gous or hctrrozygorts fbr a particular insert can he tlistin- cwtlogcnolts 10 1oc11son the X c11romosome. pishctl. In c;ws ~vcrcthere \vas ambiguity in eye color classes in nonliscrl lincs, m;Iles with the darkest eve color were se- Further details about the I>-insertlinc construction and the lcctcd for pl1enotypic analysisant1 assumcd to he homozygous for tl~r/>insc*rt. cytogenetic map are given in TRIT I?/ rrl. (1996a). Introgression procedure: The introgression proccdurc. is Morphological data: From each sampled male. an outline shown in Fiqlrc 3. For each trtnwi/ioun insrrt stock, ndcs ol' 011cposterior lohe and the length of the tibia on onc wn" as homozygous for the I' insert ~verccrossctl to f+tn;des from 21 forclcg oI~t;1incrl;und analyzed described previously sivrrtlots rcrstock to produce F, femalcs. XI this point, suldincs (I.lt, 14 rrl. 1996). Thc area of e.;lch outline was estimated ;IS of cxh inscrt ~vcrccstahlishctl by sctting up sevcl-;11inclrpcn- the arca of the polygon formed 1,y.joining the points in the tlrnt CI-OSSCS of thr F, fc.tnalcs to sinrtr/rtr?.x 70 malrs. In sr~l~e- coorrlinatc list with linc segmrnts. I

(SAS Institute, Inc.). Confidence intervals for linear combina- simulans except for an intact segment of mauritiana tions of subline means were calculated according BOWERMAN chromosome surrounding the selected P insert. The and O'CONNELL(1990, p. 742), using the mean square of vials within sublines from an analysis of variance. intact segmentis expected to have an exponentialdistri- bution with an average length of 9.4 cM after 15 generations, which is of the genome or 18% of the third RESULTS -7% chromosome. The actual lengths are probably more The size of introgressed segments: In principle, the variable than the expected distribution because recom- set of introgression sublines could be used to localize bination rates may vary among chromosomal regions QTL by a modification of the interval mapping concept and because there is variation among lines in the num- in which a likelihood ratio is calculated based on the ber of backcrosses (15-24). probability that the mauritiana allele of a QTL is still Variation among mauritiana insert lines: Analysis of linked to a selected P marker after t backcrosses to variance was used to detect variation among the48 pure simulans. Implementation of this idea requires accurate mauritianu thirdchromosome P-insert lines andthe estimates of recombination rates during the backcross mauritiana w host strain. Lines are highly significant for procedure. Originally, we anticipated that recombina- PC1 (P < 0.007), ADJPCl (P< O.OOOl), posterior lobe tion rates in hybrids and introgression stocks would be area (P < 0.0025) and tibia length (P < 0.0001). The very similar to the rates in pure mauritiana flies, so that correlations over line means between tibia length and the necessary information could be obtained by measur- each of thethree posterior lobe variables are small ing recombination between each pair of adjacent P- (0.03-0.16) and not significantly different from zero, element inserts in the mauritiana stocks. However,three suggesting that posterior lobe variation is not due to typesof evidence indicatethat the effectivelevel of general body size variation. recombination during thebackcrosses was substantially The variation among mauritiana P insert lines may be different thanin pure mauritiana flies. (1)The recombi- due to genetic variation within the original host strain nation map for mauritiana is -1.4 times longer than that was differentially distributed to the insert lines dur- the standard simulans map (TRUEet al. 1996a). (2) Re- ing their construction. In addition, it is possible that combination rates between P inserts in different intro- the P inserts themselves have caused some variation by gression lines are, on the average, only about one-half inducing new mutations. In that case, some variation the corresponding rates in pure mauritiana (TRUEet al. among introgression lines also may be due to P-insert 1996b). (3) A likelihood analysis of molecular markers mutations rather than theeffects of wild-type mauritiana in the introgression lines indicates that,during the alleles at introgressed QTL. However, it appears that backcrossing procedure, the effective level of recombi- little (if any) of the introgression line variation is attrib- nation was only about one-fourth that in pure mauri- utable to the Pinsertsthemselves, since the correlations tiana and, moreover, there is evidence for heterogeneity between mauritiana line means and the means of intro- in recombination rates among different chromosomal gression lines with the same Pinsert are small and not regions (TRUEet al. 1996b). Thus, we concludethat significantly different from zero (0.06 for PC1, 0.07 for quantitative localization of QTL using a likelihood ap- ADJPC1 and 0.16 for area). proach is not feasible with the introgression line data Variation among introgression control limes:After 15 since the necessary recombination rates have not been generations of backcrossing without selection (as in the estimated with sufficient accuracy. control introgression lines), theprobability of retaining In spite of uncertainty regarding the precise pattern the mauritiana allele at a particularlocus is only 0.00003. and levelof recombinationduring backcrossing, we Therefore, we expect the control introgressions to have have used molecular marker data from the introgres- an essentially pure simulans genome. Consistent with sion lines to estimate the average amount of mauritiana this expectation, the average PC1 value of the 21 inde- genetic material introgressed into the simulans back- pendent control introgressions is 0.0288, which is not ground (see appendixof TRUEet al. 1996b). An experi- significantly different from the average value of the sim- mental determinationof the lengthsof individual intro- ulans 13w stock, 0.0302 (using a one-tailed t value for gressed segments throughout the chromosome would a = 0.05 in a linear contrast). Moreover, the control be very difficult, but inferences about the mean length sublines are not significantly different from each other, can be made by sampling a molecular marker locus in as determined by analysis of variance. However, all but a number of different introgression lines and asking two of the 21 controls have a PC1 value less than (ie., whether it has the mauritiana or simulans allele. This more mauritiana-like) than simulans 13w, indicating that was done for threewell spaced loci on thethird chromo- there may be some minor variation among introgres- some (hat 66D, Antp at 84D and jan at 99E), yielding sion lines due to loci that are unlinked to the selected a maximum likelihood estimate of the effective level of Pmarker. Therefore, in the following, we use the mean recombination during introgression. This analysis sug- of the control introgression lines as a standard forcom- gests that the genome of an introgression line after 15 parison with the P-marked introgression lines. or more generations of backcrossing is essentially pure Variation among P-marked introgressionlines: Varia- Introgression Analysis in Drosophila 343 tion among the introgression sublines can be due to A. Introgression variation in the location of the Pinsert, as well as varia- M .. tion in the length of introgressed segments associated with the P insert (i.e., variation among independent, replicate introgressions). Analyses of variance show that both sources of variation are highly significant (P < 0.0001) for all three posterior lobe variables (PC1, ADJPCl and area). In addition,both insert location and sublines within location are significant for tibia length (P < 0.0001 and P < 0.001, respectively). Correlations between tibia length and theposterior lobe variables, taken over the 164 introgression subline means, are small (-0.003 to 0.10) and not significantly different from zero. Thus, it appears that B. Interval mapping general body size variation contributes little to varia- Y. h Ira A~lp MI# Db Ald jam Muka tion in the posterior lobe. This resultis consistent with 40. T TT TT T our previous analysis of several strains from each of 35 A the two species, mauritiana and simulans, which revealed no significant difference in tibia length between species that might contribute to the posterior lobe difference (LIUet ul. 1996). As expected, the posterior lobe variables are significantly correlated with each other over the subline means. PC1, which reflects both size and shape variation, has a correlation of 0.98 (P< 0.0001) with area and 0.64 (P < 0.0001) withADJPC1, which reflects mainly shape variation. Similarly,ADJPCl andarea have a correlation of 0.52 (P< 0.0001). These results support our previous conclusion that genetic and developmental controls on posterior lobe size and shape are tightly coupled (LIU et al. 1996). For simplicity, only results for PC1 arepresented in the following discussions. The results and conclusions for the ADJPC1 and area variables are very similar to those presented for PCl. Figure 4A displays the mean PC1 value of the 164 FIGURE 4,"Estimation of QTL effects from two different introgression sublines in relation to the cytological loca- types of experiment. (A) The effects of introgressed segments tions of their P inserts. The means are scaled so that in relation to the cytological location of their P-insert marker. the average of the control introgressions (represented Each point represents the mean value of a marked introgression subline. The raw PC1 values were scaled by subtracting by the dotted line)is zero. In addition, thescaled means the mean of the controlintrogression lines and thenexpressed are expressed as a percentageof the difference between as a percentage of the pure species difference to get PC1%. the pure mauritiana and simulans strains used to initiate Zero represents the mean value of the control introgressions the experiment. Thus, the ordinate value represents and points above the dashed line are significantly different the percentage effect of an introgressed segment on from the control mean. Theband indexis a measure of cytological position along the chromosome. For reference to B, A at the trait value. Means above the dotted line are more the bottom of the figure represent the positions of molecular mauritianu-like and those below the line are more sim- markers used in the interval mapping experiment. (B) Sum- ulunclike thanthe introgression control mean. The mary of third chromosome data from the interval mapping figure also shows a critical value (dashed line), which experiment of LIU et al. (1996). Molecular marker positions is the Bonferroni 95% confidence interval of the differ- are indicated by a V above the panel. LOD scores are plotted against test position (in cM). LOD scores above the critical ence between an introgression subline mean and the value ( * - * ) provide evidence for the existence of a QTL control mean (adjusted for making 164 simultaneous within the interval. The LOD score given is for thejoint analysis comparisons). of backcrosses to both parental species. Locations of putative A large fraction of the introgression sublines are sig- QTL and their estimated effects are given below the plot. The nificantly more mauritiana-like than the controls (ie., additive effect is the average of effect estimates from both backcrosses. When the effects estimated from the two back- 77 of 164 points are above the dashed line), strongly crosses were significantly different, a nonadditive effect and indicating that the introgressed segment in those lines the two separate additive effect estimates (QTL,,,,, and QTI,,,,) contains one or more QTL affecting the trait. Unfortu- are also given. Data from LIU et al. (1996). 344 C. C. Laurie et nl. nately, there is little information about the number or of any individual QTL. However, thisfigure is very likely possible locations of the QTL, since the lengths of indi- to be an overestimate, since sublines with the largest vidual introgressed segments are unknown and ex- effects probably contain unusually long introgressions pected to average roughly 18% of the chromosome. that span multiple QTL. However, these data do contain information about the The resolution of both experiments is very limited. magnitude of QTL effects, which provides an important Clearly, it ispossible that intervals contain multiple comparison with the effects estimated from our previ- QTL and that no individual QTL has an effect as great ous interval mapping experiment (LIU et al. 1996). as 10%.Even if there are QTLwitheffects thislarge, the Comparison with interval mapping results. The like- introgression results suggest that there arealso multiple lihood ratio (LOD) profile from the interval mapping QTL with small effects on the order of a few percent experiment (Figure 4B)shows highly significant re- or less. The evidence for this suggestion is the observa- gions throughout the third chromosome, particularly tion that a majority of introgressions with small effect on X (left of Ant@),which is consistent with the obser- estimates aremore mauritiana-like thanthe controls vation that introgressed segments located throughout and these are distributed along the lengthof the chro- the third chromosome have significant effects. The re- mosome. Among the 83 sublines that have an effect gion of lowest LOD score is the hb/Ald interval, which between -7.2% and +7.2% (the critical value), 66 are also seems to have relativelyfew introgressions with sig- between 0 and+7.2%. Although effectswithin this nificant effect. range are not statistically significant, the fact that many In the interval mapping experiment, all estimated more are positive than negative provides a strong indi- effects have the same sign, indicating that substitution cation that small positive effects do exist. of a mauritiana for a simulans allele changes PC1 in the Visualizing QTL effects in the 10-15% range: In the same direction for all QTL. Similarly, nearly all of the interval mapping experiment, inbredlines of each spe- significant effects of the introgression sublines are in cies were hybridized to generate backcross populations the same direction. There areonly two sublines (having for QTL analysis. The difference in PC1 between the the same P insert at band index 680) that have large inbred parents is 30.4 environmental SD (based on the effects in the opposite direction.These two unusual variance among individuals within a genetically homog- sublines suggest the possibility that this particular Pin- enous inbred line). Thus, when a single QTL of 10% sert has an effect on the trait in mauritinna as well as in effect is segregating in an isogenic genetic background, the introgressions. However, the pure mauritiana line the means of the two homozygous classes are expected with this insertion does not have an unusual value of to lie -3 SD apart, showing little overlap in phenotype PCI. Thus, there may be a QTL with negative effects (assuming normal distributions).The introgression line in the immediate vicinity of band index 680. data support this expectation by providing a means of Comparison of the magnitude of effectsfrom the two visualizing the degree of overlap between the pheno- experiments requires a comparable scale of measure- typic distributions of alternative genotypes. ment. The introgression effects in Figure 4A represent Figure 5 shows (1) thedistribution ofPC1 for all the difference between homozygous segments, ex- individuals from introgression sublines with effects in pressed as a percentage of the difference between the the 10-15% range and (2) the distribution of all indi- pure mauritiana and simulans parental strains. Similarly, viduals from the control sublines. The variance among the QTL effects in Figure 4B represent the difference individuals within the control class is larger than the between a heterozygote and a homozygote expressed variance among individuals within an inbred line (ratio as a percentage of one-half of the parental difference. of 1.4), as expected since there is probably intraspecific These two measures are expected to be comparable, polymorphism in the simulans 13w stock and there may since inheritance of the PC1 trait is largely additive (LIU also be some residual mauritiana alleles not removed et al. 1996). In addition, interval sizes in the interval by repeated backcrossing. The introgression individuals mapping experiment are similar to the average length have a higher variance than the controls (also a ratio of introgressed segments. In the interval mapping ex- of 1.4), as expected since they derive from multiple periment, theintervals average -14% of the third chro- sublines having a range of effects. Using the standard mosome map length, while the expected length of in- deviation of the control individuals as a measure of trogressions is -18%. background (and environmental) variation, the differ- The magnitudes of effect estimated from the two ex- ence in means between the control and the introgres- perimentsare consistent. Interval mapping indicates sion groups is 3.0 SD. The introgression individuals are five significant intervals with effectsranging from 10.4% homozygous for mauritiana alleles in a small chromo- to 15.9%. There are introgressions based in each of somal segment and the controls are homozygous for the same intervals that have effects within the 10-15% simulans alleles throughout. There is rather little over- range. Assuming that the mauritiana allele of all QTL lap between the distributions. If individuals were as- affects PC1 in the same direction, the highest introgres- signed to two groups based on whether they had a PC1 sion effect (26.4%) places an upper limit on the effect value greater or less than the midpoint between the Introgression Analysis345 in Drosophila

A. introgression sublines with 10-15% effect unit area so that the eyemay focus on shape rather than size differences. (In every case, the introgression specimens have a smaller area than the standard simuluns before scaling.) The figure shows 21 types of speci- men: pure simuluns 13w, control introgressions and 19 introgression types, each representing a different P-insert location. The specimens displayed for each group were selected as those having the PC1 value(s) closest to the mean of their respective type. Figure 6 reveals that the posterior lobes of introgression specimens have shapes similar to pure simuluns, but in most cases there is a noticeable difference in curvature between the naturaltip and theartificial base- PC 1 line. Introgressions having their Pinserts at several dif- B. control sublines ferent chromosomal locations tend to have an altered I curvature such that thetip appears to have been pushed upward relative to that of pure simuluns. This impres- sion is reinforced by an examination of the average shapes of the control introgressions and those introgressions with effects in the 10-15% range (Figure 7). An average shape is constructed by averaging each of the 100 Fourier coefficients over individuals within a type and then plotting the outline specified by the set of 100 average coefficients. The size-adjusted outlines in Figure 7 emphasize the difference in shape, while those based on raw data illustrate the difference in size. Both components differ between these two groups of PC 1 individuals, as expected because of their high degree FIGURE5.-The distributions of PC1 for individuals from of correlation. (A) introgression sublines with effects in the 10-15% range It appears that differentQTL have similar effects on and (B) control sublines. The sample size,mean and SD are posterior lobe shape. Introgressed segments having P 164, 0.0221 and 0.0026 for (A) and 126, 0.0288 and 0.0022 for (B). elements situated atopposite ends of the third chromosome are very unlikely to overlap, yet they have similar effects on the tip of the lobe. (Compare shapes with means of thecontrol and introgression classes,only low us. high band index numbers in Figure 6). This 8.3% of individuals would be misclassified. However, result suggests that multiple QTL may participate in a adding an intermediateheterozygous class at a 2:l ratio common developmental process that specifies the de- would generate anessentially unimodal, continuousdis- gree of curvature from the tip to the base of the lobe. tribution of phenotypic values. Thus, a single QTL effect in the range of 10-15% DISCUSSION for the PC1 trait is substantial in the sense that it is readily detectable statistically and alternative homozy- One of the most important aspects of the genetic gotes show little phenotypic overlap on a relatively uni- architecture of a quantitative traitdifference is the form genetic background. However, it is not expected magnitude of effects of individual QTL, but accurate to produce qualitatively distinct phenotypic classes in a estimates of these effects are difficult to obtain. In a segregating population, especially if other QTL affect- simple marker-linkage approach, effect estimates are ing the trait are also segregating simultaneously. based on the difference in mean value between geno- The introgression lines also can be used to visualize types at a linked marker locus, which confounds the the effects of small chromosomal segments on posterior magnitude of effect with the genetic distance between lobe shape. Figure 6 shows the posterior lobe outlines the marker andQTL, leading to biased estimates (ZENG of a representative sample of individuals from introgres- 1994). Interval mapping (LANDER and BOTSTEIN1989) sion sublines with effects on PC1 in the l0-15% range. solves this particular difficulty through a likelihood ap- For comparison, samples of pure simuluns and control proach thatprovides effect estimates for themost likely introgression individuals are also shown. Each outline position of a QTL within an interval bounded by two is overlaid on a standard simuluns outline (the individ- markers. However, effect estimates from simple inter- ual with its PC1 value closestto the meanof the simuluns val mapping are based on the assumption of no more 13w group). All of the outlineshave been scaled to have than oneQTL per chromosome, so they may be biased 346 C. C. Laurie et al.

sim sim sim sim sim FIGURE 6.”Size-adjusted outlines of the posterior lobe of individuals from pure simuluns (sim), control introgression lines (con) and a set of P-marked intro- con con con con con gression lines with effects on PC1 of l0-15% (labeled with the band indexof their P insert). Each outline is overlaid on a standard simuluns, selected as the speci- L? L? men with its PC1 value clos- 308 349.5 107.5 206 223.5 est to the mean of the pure simuluns parental line. This standardoutline (std) is shown by itself in the lower right corner. The five sim and conspecimens given 454.5 526.5 J9554.5 581.5 730.5 are those withPC1 values closest to the means of their respective groups.Among the markedintrogression sublines with effects in the 10-15% range,one speci- 852.5 893.5 941.5 1010.5 1030.5 men representing each cytological location is shown; these specimens were cho- sen to have their PC1 values closest to the mean for that Q location. The outlines shown 1177.5 1753.5 1861 1978 Std here have been size-adjusted to have unit area,to focuson shape variation. The origin of each outline is its centroid and they are all oriented with Q parallel baselines. by the existence of QTL outside of the test interval. dent information about effect estimates. In this study, This problem is greatly diminished by composite inter- we have compareddirect estimates of the effects of val mapping, which utilizes a multiple regression ap- introgressed chromosome segments with estimates from proach to condition on markersoutside of the test composite interval mapping for marker-defined inter- interval ( ZENG 1994). vals that aresimilar in size to the introgressed segments. Although simulation studies supportthe utility of In the introgression experiment, the effect estimate is composite interval mapping (ZENG 1994), it also is im- simply the difference in mean value between genotypes portant to have empirical studies that provide indepen- that have a uniform genetic background(from simuluns), but are homozygous for either a simuluns or muuritiuna chromosome segment. In the interval mapping experiment, the effects of QTL in multiple intervals are estimated from backcross individuals that are segregating at all QTL simultaneously by using a com- plicated statistical procedure. Nevertheless, these two very different experiments are consistent in suggesting that several QTL on the third chromosome may have effects in the range of 10-15% of the parental difference and that nearly all QTL have effects in the same raw data size-adjusted data raw data direction. FIGURE7.-Average outlines for individuals from two However, both theintrogression and composite inter- groups of lines: all controlintrogressions (gray, n = 126) val mapping estimates may be biased by the existence and marked introgressions with effects in the 10-15% range of multiple QTL within an interval or introgressed seg- (black, n = 164). Theraw outlines illustrate a significant difference in size, while the size-adjusted outlines illustrate an ment. Therefore, the 10-15% effect estimates must be average shape difference. The origin of each outline is its considered tentative. Improvements in resolution can centroid and they are all oriented with parallel baselines. be obtained by adding more markers to the composite Introgression Analysis in Drosophila 347 interval mapping approach or by additional backcross- other QTL is likely to produce a phenotypic variance ing of introgression lines. However, single locus resolu- considerably larger than the environmental variance, tion is very difficult to achieve by recombination and which might cause substantial overlap between the two other approaches probably will be necessary to define homozygotes at the locus of interest. Nevertheless, an the effects of a single locus unambiguously. For exam- effect as large as 15% may at least border on being a ple, in Drosophila, it may be possible to associate QTL major factor. However, we must emphasize again that with candidate genes through complementation tests effects estimated from both the introgression and in- (MACKAY 1996) and thentest for the effects of different terval mapping experiments may be upwardly biased wild-type alleles by P-element transformation. We are by the occurrence of multiple QTL within an interval in the process of improving the resolution of the inter- or introgressed segment. Therefore, the possibility of val map, as well as identifylng candidate genes through major gene effects on the posterior lobe cannot be mutations that affect the posterior lobe. eliminated by our results to date, butrequires further A frequently asked question about QTL effects is investigation by studies with higher resolution. whether they are “major” or “minor” in magnitude. Substitution of a mauritiana for a simulans allele ap- The significance of this question relates to models of pears to decrease PC1 (as well as posterior lobe area) quantitative trait evolution that range from the “infini- for all (or nearly all) QTL detected. This observation tesimal” model in which a very large number of loci, of consistency in the direction of allelic effects is un- each with very small effect, control a given trait, to a usual. In previous work on domesticated crop plants, “saltational” model in which trait differences are due the majority of QTL detected have effects predicted by to one ora few loci of large effect. Since most empirical the direction of the parental difference, butmost stud- results appear to lie somewhere in between these two ies also report some allelic effects that are opposite in extremes (TANKSLEY1993), the definition of major ef- sign to the majority (i.e., a mixture of plus and minus fect is generally unclear or it may vary depending on effects) (TANKSLEY 1993).For example, a study of 11 the issue (s) being addressed. traits in a cross between two tomato species revealed 74 One of the issues to which the major us. minor gene QTL and 36% of those had allelic effects opposite in question has relevance is the cause of the “punctuated sign to the parental difference (DEVICENTEand TANK- equilibrium” pattern of morphological evolution in the SLEY 1993). In addition,we have done QTL analysis of fossil record, in which most evolutionary change ap- differences between D. mauritiana and D. simulans in pears to occur in relatively rapid bursts separated by bristle number onvarious structures and find a mixture longperiods ofstasis (ELDREDGEand GOULD 1972; of plus and minus effects for two of those structures, GOULD 1980).One of the possible causes of this pattern male anal plate and sex comb (unpublished data). is the rare occurrence of a favorable mutation with ma- The consistency in direction of allelic effects on the jor effect that rapidly becomes fixed in the population, posterior lobe has implications for the evolutionary his- causing a relatively sudden shift in morphology in a tory of this structure. One hypothesis is that selection time series of fossils. In this context, one might define has operated to produce divergence in a consistent di- a major effect as one for which alternative homozygotes rectionthroughout the period of isolation between have phenotypic distributions with little or no overlap mauritiana and simulans, which is estimated to be 0.6- (say <5% misclassification using a midpoint criterion). 0.9 million years (HEY andKLIMAN 1993). If extended For two normal distributions of equal variance, the periods of reverses in the direction of selection or of mean values must differ by 3.28 SD to have a classifica- stabilizing selection had occurred, or if random drift tion accuracy of 95%. Thus, amajor gene effect would had played an important part in the evolution of this be produced by alternative homozygotes that differ by trait, we might expectto find a mixture of allelic effects. at least 3.28 SD, based on the variance within a geno- Thus, the failure to observe a mixture of effects on the typic class. The minimum value this variance will take posterior lobe limits the number of reasonable evolu- is the environmental variance (among genetically ho- tionary scenarios relative to other traits that do show mogeneous individuals) and it will often be larger, de- such a mixture. pending on other segregating loci that affect the trait. The type of selective force(s) that may operate on Our genetic analysis of the posterior lobe has pro- the posterior lobeis unknown, butit is likely that sexual vided effect estimates as great as 15%, which corre- selection through female choice is involved. EBERHARD sponds to 4.6 environmental SD. Two populations that (1985) has provided strong indirectevidence that diver- differ by a fixed effect of this magnitude would have gent male genitalia frequently evolve through female distributions with little overlap (1% misclassification), preferences implemented by sperm displacement and provided there were no othersegregating QTL to sup- remating frequency. Theoretical models of sexual selec- plement the environmental variance. However, it is tion have shown that genetic correlation between fe- important to realize thatintermediate phenotypes male preference and the selected male trait can lead to would be present as heterozygotes during the process a runaway process with the potential to produce rapid of fixation, and,more importantly,segregation at evolutionary divergence between isolated populations 348 C. C. Laurie et al.

(LANDE 1981). In addition, the evolutionary dynamics 1995 Genetic mapping of floral traits associated with reproduc- tive isolation in monkeyflowers (Mimulus).Nature 376: 762-765. of this process may be cyclical, leading to periodic DEVICENTE,M. C., and S. D. TANKSLEY,1993 QTL analysis of trans- changes in the direction of sexual selection operating gressive segregation in an interspecific tomato cross. Genetics on the male trait (IWASAand POMIANKOWSKI1995). If 134: 585-596. DW:BI.EY,J., A. STECand C. GLJSTUS,1995 teosinte branched1 and either orboth the simulans and mauritiana populations the origin of maize: evidence for epistasis and the evolution of have been subject to cyclical selection on the posterior dominance. Genetics 141: 333-346. lobe, we expect the genetic architecture of the species EI.DREDGE,N., and S. J. CAXJLD,1972 Punctuated equilibria: an alternative to phyletic gradualism, pp. 82- 115 in ModeZs in Paleobiol- difference to reveal a mixture of allelic differences with ogy, edited by T. J. M. SCHOPF.Freeman, Cooper and Co., San positive and negative effects, since it is likely that differ- Francisco. Goulu, S. J., 1980 Is a new and general theory of evolution emcrg- ent sets of QTL respond to selection at different times, in$? Paleobiology 6: 119-130. depending on the availability of suitable genetic varia- HEY,J.. and R.M. JSI.IMAN,1993 Population genetics and phyloge- tion . Alternatively, it is possible that only a limited netics ofDNA sequence variation at multiple loci within the Drosophila melanogaster species complex. Mol. Biol. Evol. 10: number of loci are capable of responding to selection 804-822. and that successive fixations have occured at each of IWASA,Y., and A. POMIANKOWSKI,1995 Continual change in mate those loci, such that the most recent episode of selec- preferences. Nature 377: 420-422. KUHI.,F. P., and C. R. GIARDINA,1982 Elliptic Fourier features of a tion removes the evidence of previous episodes. How- closed contour. Comp. Graphics Image Process. 18: 236-258. ever, it seems unlikely that suitable genetic variation LAKDE, R., 1981 Models of speciation by sexual selection on poly- would be present at each locus during each episode genic traits. Proc. Natl. Acad. Sci. USA 78: 9721-3725. LANDER, E. S., and D. BOTSTEIN,1989 Mapping Mendelian Factors of selection. Thus, the results presented here do not underlying quantitative traits using RF'LP linkage maps. Genetics support the notion of an evolutionary history that in- 121: 185-199. LEMEUNIER,F., and M. ASHBURNER,1976 Studies on the evolution volved periodic changes in the direction of selection of the mrlanogastrr species subgroup of the genus Drosophila (So- on the posterior lobe. phophora). 11. Phylogenetic relationships of six species based upon polytene chromosomcbanding patterns. Proc. R. Soc. We are very gratefulto SAMAN7HA GASSONfor her careful and Lotld. B 193: 275-294. patient technical assistance with the difficult job of constructing and LIU,J.,J. M. MERCER,I.. F. STAM,G. GIBSONand C.C. LAL~RIE,1996 Genetic analysis of a morphological shape difference in the male analyzing the introgression lines, as wellas ROBERTWARD, MERY1. genitalia of Drosophila simulans and D. mauritiana. Genetics 142: CARTER,DARKEN PERKINS andLYNN STAM for additional technical 1129-1145. assistance. We also thank ZHAO-BANGZEN<; for helpful comments MAWAY, T. F. C., 1996 The nature of quantitative genetic variation on the manuscript. The Duke University Morphometrics Laboratory revisited: lessons from Drosophila bristles. BioEssays 18: 113-121. provided equipment and software for morphometric analyses. This TANKSI.L.Y,S. D., 1993Mapping polygenes. Annu. Rev. Cknet. 27: work was supported hy U.S. Public Health Service grant GM47292. 205-253. TRLTF,,J. R., J. M. MLRCEKand C. C. IAAI~RIE, l996aDifferences in crossover frequency and distribution among threesibling species of Drosophila. Genetics 142: 507-523. LITERATURE CITED TRUE,J. R., B. S. WEIRand C. C. LAURIE, 1996b A genome-wide sur- vey of hybrid incompatibility factors by the introgression of BIER,E., H. VAESSIN,S. SHEPHERD,K. LEE, K. MCCAI.~.et al., 1989 marked segmentsof Ilrosophila mauntiann chromosomes into Drr~ Searching for pattern and mutation in the Drosophila genome sophila simulans. Genetics 142 819-837. with a P-ladvector. Genes Dev. 3: 1273-1287. ZEN(;, 7,-B., 1994 Precision mapping of quantitative trait loci. Genet- BOWERMAN,B. L., and R. T. O'CONNELL,1990 Linear Statzstzcal Mod- ics 136: 1457-1468. els. PWSKENT Publishing Co., Boston. BRAI)SHAW,H. D., S. M. WII.BERT,K. G. OTTO and D. W. SCHEMSKE, Communicating editor: M. J. SIMMONS