Enough work has now been done on reproductive barriers that a number of clear patterns have begun to emerge; one of the clearest is the important role played by conspecific sperm precedence in the reproductive isolation of closely related animal species (Howard 1999; Simmons 2001; Coyne and Orr 2004). Conspecific sperm precedence is defined as “the favored utilization of sperm from conspecific males in fertilization when both conspecific and heterospecific males have inseminated a female” (Howard 1999, pp. 110–111). The precedence may occur because conspecific sperm outcompete fertilization-competent heterospecific sperm or because of postinsemination incompatibilities between heterospecific males and females (i.e., noncompetitive gametic isolation). Consequently, there are many mechanisms that can underlie such heterospecific disadvantages.

Although barriers to fertilization operating at the level of sperm and egg have long been recognized as important in the reproductive isolation of broadcast-spawning marine invertebrates (Loeb 1915; Lillie 1921), the importance of postinsemination barriers to fertilization in terrestrial animals did not become apparent until the 1990s. It was not until this period of time that gamete competition studies were regularly incorporated into investigations of reproductive isolation. As a result of these analyses, we now recognize that conspecific sperm precedence isolates closely related species in groups as divergent as vertebrates and insects (Howard and Gregory 1993; Gregory and Howard 1994; Wade et al. 1994; Price 1997; Howard 1999; Price et al. 1999; Brown and Eady 2001; Simmons 2001; Fricke and Arnqvist 2004).

An insect group in which conspecific sperm precedence (CSP) has been particularly well studied is the ground cricket genus Allonemobius. Detailed studies over the course of many years (Benedix and Howard 1991; Howard and Gregory 1993; Howard et al. 1993; Gregory and Howard 1994; Doherty and Howard 1996; Gregory et al. 1998; Howard et al. 1998a,b; Britch et al. 2001) have demonstrated that the strong, but incomplete reproductive isolation between the closely related species Allonemobius fasciatus and A. socius is due to a single type of reproductive barrier—CSP. The simplicity of the system, a single barrier to gene exchange isolating two closely related species, is extremely rare among species pairs that have been thoroughly investigated (Coyne and Orr 2004). In this case, should we achieve an understanding of the genetic control of CSP, we will have achieved an understanding of the genetic changes that have given rise to new species.

Here we report in detail the results of QTL studies of CSP in A. fasciatus and A. socius, preliminary results of which were published earlier (Howard et al. 2002). Although several studies have looked at other isolating mechanisms through QTL analysis (Bradshaw et al. 1998; MacDonald and Goldstein 1999; Fishman et al. 2002; Tao et al. 2003; Nurnberger et al. 2003), this study is among the first to document QTL for conspecific sperm precedence (see also Civetta et al. 2002). The QTL approach allows us to estimate the number and location of genetic factors responsible for a difference between two species in a trait, as well as the magnitude of the effect of each QTL.

Although preliminary QTL analyses and a linkage map for A. fasciatus using amplified fragment-length polymorphisms (AFLPs) (Vos et al. 1995) were previously described (Howard et al. 2002), since the publication of those results, the Howard Lab has switched from an ABI 377 to an ABI 3100 automated sequencer. Given the ease and speed of analysis of the ABI 3100 and some discrepancies in fragment sizes, we reanalyzed all individuals in the A. fasciatus mapping population, and we analyzed individuals from an A. socius mapping population. In both cases, we used five primer combinations to create AFLP linkage maps and performed QTL analyses to document linked or unlinked single markers or groups of markers strongly associated with CSP in F₂-backcross females.

The QTL experiments for both species were similar in protocol and are described here in brief. Field-caught crickets from three focal populations in the East Coast (EC) transect (Figure 1), EC 49 and EC 60A (A. fasciatus, north of mixed populations), and EC 65 (A. socius, south of mixed populations), were brought to the lab and screened for species identity using allozymes (Howard 1983, 1986; Howard and Furth 1986). Screening was necessary due to the presence in both populations of introgressed individuals and cryptic congeners in low frequencies. Following the breeding design shown in Figure 2 we hybridized field-caught A. fasciatus males and A. socius females and backcrossed F₁ females with males from first-generation lab-reared parental lines to produce F₂-backcross females. For linkage mapping and QTL identification we developed AFLP markers (see below) unique to each species by tracing AFLP fragments through two generations (F₁ and F₂ backcross) that were absent in individuals of the species to which the backcrossing was done (Figure 2). The QTL trait of interest, CSP, was measured by documenting the frequency with which males of the species of interest (A. fasciatus when the introgressed genes in the F₂-backcross female were from A. fasciatus and A. socius when the introgressed genes were from A. socius) produced offspring when an F₂-backcross female was mated once each to an A. fasciatus male and to an A. socius male (Figure 3). Prior studies of sperm precedence among parental types demonstrated that order of matings has no significant effect on sperm utilization patterns (Howard and Gregory 1993; Gregory and Howard 1994); thus, for the sake of simplicity F₂-backcross females in both experiments were mated first to A. fasciatus males and second to A. socius males. All matings were closely observed to ensure that only a single spermatophore was transferred to the female by each male. A period of 24–48 hr was imposed between first and second matings.

Figure 1.— The inset shows a map of North America with the approximate ranges of A. fasciatus and A. socius. The dark patterned area is the estimated extent of the zone of overlap and hybridization between the species. The main map shows the eastern United States, (more ...)

Figure 2.— Design of pedigreed matings between A. fasciatus and A. socius for the breeding of F2-backcross females for use in both QTL experiments. Not shown are matings using field-caught crickets to produce parental lines of each species in both experiments. Parental (more ...)

Figure 3.— Mating trial design for measuring the CSP trait. F2-backcross females (Figure 2) are used in mating trials with males from parental lines of both species. Females are mated first to a male of A. fasciatus and must be mated to an A. socius male 24–48 (more ...)

Females that successfully mated to both males were placed in individual cages and left to oviposit in both soil and cotton media for 2 weeks, after which they were frozen at −80°. Soil medium is provided specifically for oviposition, but females also oviposit in approximately equal frequency (S. C. Britch, unpublished data) in water-soaked cotton provided for dietary moisture. Oviposition media were gradually cooled and exposed to an artificial overwintering period of 3 months in a 4° constant-temperature room. After overwintering, oviposition media were gradually warmed to room temperature and nymphs were left to emerge in individual family cages. When nymphs reached second or third instar they were frozen en masse at −80° to await paternity analysis, which was accomplished with allozyme phenotyping. The resulting data were used to determine the pattern of sperm utilization by each F₂-backcross female, in particular the frequency with which each male fathered offspring.

AFLP typing of parents, males, F₁ females, and F₂-backcross females was done post hoc, since many females were expected to either not complete a second mating or not mate at all; similarly, many males will not mate in the laboratory. Following restriction ligations of cricket genomic DNA, extracted using QIAGEN (Valencia, CA) DNEasy kits (no. 69504), we used the ABI Regular Genome mapping kit (no. 4303050) to do preselective and selective PCR amplification of samples, which were then run on the ABI 3100 sequencer. A prior survey (J. L. Marshall, unpublished data) identified five combinations of selective amplification primers that consistently produced ≥70 AFLP fragments per individual. We code these combinations here as B5, B6, G3, G7, and H5, but primer names as well as all molecular protocols are available from the corresponding author. Unique fragments were traced using the rules outlined in Figure 2 and tabulated by F₂-backcross female in a spreadsheet format suitable for Map Manager Qtxb19 (Meer et al. 2002). Linkage groups were created in Map Manager at the P = 0.001 level using the Kosambi map function, which assumes intermediate interference. At P = 0.001 the threshold of at least eight linkage groups was found in both data sets, the known haploid chromosome number in Allonemobius (Lim 1971). A chi-square test showed that several AFLP markers from both species showed patterns of segregation distortion, so we activated the feature accounting for segregation distortion in Map Manager. Once linkage groups were established we used the Ripple command, which refines the order of markers on all linkage groups by testing local permutations of the order (Meer et al. 2002). As a result of the breeding design (Figure 2) F₂-backcross females are heterozygous for all candidate markers, and so, despite dominance of AFLP fragments, data were treated as “codominant backcross” in Map Manager's linkage evaluation.

On the basis of trials using several standard transformations in Mapmaker/QTL 1.1b (Lincoln et al. 1993) we determined that square-root transformation of CSP trait data for both species was the closest approximation to the normal distribution. Raw trait data were then imported into Map Manager and square-root transformed. Before QTL analyses we had Map Manager hide redundant loci, i.e., markers mapping to a location already occupied (refer to results and Figures 4 and 5). We performed single-marker regression analysis for QTL in Map Manager under the additive regression model at the least restrictive significance level of P = 0.05. At P = 0.05 we were able to group significant single-marker associations manually instead of running several separate analyses at the five more restrictive levels of significance available in the program.

Figure 4.— Linkage groups of A. fasciatus. Linkage groups were calculated using Map Manager (Meer et al. 2002) at P = 0.001 using the Kosambi map function and accounting for segregation distortion. The linkage criterion of P = 0.001 yielded the number (more ...)

Figure 5.— Linkage groups of A. socius. Linkage groups were calculated and drawn with the criteria described in Figure 4.

Before interval QTL mapping we used the quick test to establish critical likelihood-ratio statistics (LRS). LRS are empirical significance level estimates across all linkage groups, built by randomly permuting markers and phenotypes (Hartl and Clark 1997) for the transformed trait values. This feature generated three values, suggestive, significant, and highly significant, with which we gauged the significance of the LRS produced by interval mapping of the trait on each linkage group (we also used quick test LRS values to filter results of single-marker regression analysis). As a general guideline for interpreting LRS scores, the underlying P-value for the LRS must be <0.0001 at one point on a linkage group for a genomewide P-value of 0.05, and the equivalent LRS at one point on a linkage group is ~15 for a backcross pedigree design (Meer et al. 2002). For readers accustomed to the LOD score, LOD = LRS/4.6; thus the LRS of 15 is roughly equivalent to the critical LOD of 3.0 often cited in QTL studies. However, the quick test adjusts the LRS score on the basis of the experimental data.

Interval mapping in Map Manager produces a graph of the LRS for the trait along each linkage group as well as the additive regression coefficient, which is positive if the presence of the marker tends to increase the trait value and negative if it tends to decrease it. The graph is marked with the three LRS thresholds calculated in the quick test. A table accompanies each graph with values at 1-cM intervals for the LRS and regression coefficient, as well as the percentage of the total trait variance explained by a QTL at that location. We activated the bootstrap test as a further means of localizing and assessing significant trait–marker associations on each linkage group. Bootstrap LRS values are displayed as a histogram on the graph, the width of which shows the confidence interval in centimorgans for the QTL on the linkage group. There is no accompanying table for bootstrap values. For each linkage group we ran two blocks of composite-interval mapping analyses that control for effects of linked and unlinked markers and reduce ambiguity of adjacent QTL. The first block controlled for genomewide background QTL with LRS ≥ 15 that had been identified with single-marker regression. The second block controlled for effects of adjacent QTL on the same linkage group. In the second block we limited analyses to linkage groups containing more than two markers (not including redundant markers) and controlled for each marker, in turn, as a background effect. We updated the location of a QTL only if controlling for a background QTL increased the actual (not bootstrapped) LRS at a marker or intermarker space on a linkage group; however, bootstrap values higher than significant were taken into account in borderline cases.

We produced 100 F₂-backcross females from two parental lineages (five F₁ families) for the A. fasciatus map and 111 F₂-backcross females from five parental lineages (17 F₁ families) for the A. socius map. Additionally, 13 F₂-backcross males to A. fasciatus and 20 to A. socius were included in the mapping data to increase the resolution of AFLP linkage maps. Lineages are based on a unique parental hybrid mating (Figure 2). Over 1000 AFLP fragments across five primer combinations were scored for the two A. fasciatus parents of which 116 were not found in A. socius individuals used to produce F₂ backcrosses in the A. fasciatus experiment. Of >2500 fragments scored for the five A. socius parents, 275 were not found in A. fasciatus individuals used to produce F₂ backcrosses in the A. socius experiment. Except for rare instances, all AFLP primer combinations yielded ≥70 fragments in all individuals. Linkage groups created at the P = 0.001 linkage criterion for A. fasciatus are shown in Figure 4 and those for A. socius in Figure 5. We mapped 65 A. fasciatus markers to 13 linkage groups (four redundant markers), spanning 833.9 contiguous cM with an average distance between markers of 17.7 cM (range: 4.4–45.9 cM); and we mapped 39 A. socius markers to 10 linkage groups (8 redundant markers, including 4 on 3 linkage groups made exclusively of markers mapping to the same location) spanning 361.1 cM with an average distance of 17.2 cM (range: 4.8–41.4 cM). The remaining 51 markers in A. fasciatus and 236 markers in A. socius were unlinked. The smaller number of markers mapping to fewer linkage groups and the large number of unlinked loci in A. socius were probably due to the large number of small families, sharing few markers, used in that experiment. The Ripple command affected only linkage group 2 of A. fasciatus—it reversed the order of G7.144.14 and G3.272.24 and gave rise to some changes in map distances on the linkage group. However, interval QTL mapping (see below) on both versions of A. fasciatus linkage group 2 did not produce greatly different LRS graphs or bootstrap histograms; therefore we retained the original marker order.

Although all 211 F₂-backcross females mated with conspecific and heterospecific males, not all produced offspring, and in some cases offspring paternity was not discernible due to degraded allozymes. Of the 100 F₂-backcross females used to generate the A. fasciatus map, 83 had sufficient trait data for QTL analysis. The corresponding number for A. socius was 95 of 111 F₂-backcross females. The trait data for A. fasciatus, the percentage of offspring sired by the A. fasciatus male, were bimodally distributed with an average of 0.39 (interquartile range, IQR 0.00–0.68). Trait data for A. socius, the percentage of offspring sired by the A. socius male, were also bimodal, with an average of 0.58 (IQR 0.15–1.00). The results of single-marker regression are given for A. fasciatus in Table 1 and for A. socius in Table 2. Although several additional markers in both species were found at the P = 0.05 level in single-marker regression we filtered results based on the quick test experimentwide LRS levels, requiring a minimum LRS of 5.6 for A. socius markers and 6.8 for A. fasciatus markers. Quick test critical LRS values were comparable between species (Tables 3 and 4).

TABLE 1 Single-marker QTL regression analysis results for A. fasciatus, calculated in Map Manager (Meer et al. 2002)

TABLE 2 Single-marker QTL regression analysis results for A. socius, calculated in Map Manager (Meer et al. 2002)

TABLE 3 Interval QTL mapping analysis results for A. fasciatus, calculated in Map Manager (Meer et al. 2002)

TABLE 4 Interval QTL mapping analysis results for A. socius, calculated in Map Manager (Meer et al. 2002)

Most significant QTL were associated with unlinked markers and significant QTL associated with linked markers were grouped onto a small number of linkage groups (Tables 1 and 2). As discussed in Howard et al. (2002) the relatively small size of the mapping populations is expected to reduce the number of significant QTL, but the QTL detected will be of large effect (Tanksley 1993; Bradshaw and Stettler 1995; Bradshaw et al. 1995, 1998). The results of these analyses reflect this observation. The percentages of trait variance explained by single-marker regression QTL ranged from 9 to 17% in A. fasciatus (Table 1) and from 6 to 29% in A. socius (Table 2). Although most markers in both species had a positive effect on conspecific sperm utilization, 4 of the 16 A. fasciatus markers and 4 of the 10 A. socius markers had a negative effect, meaning that the presence of the marker reduced the percentage of offspring fathered by the conspecific male. Positive and negative effects are determined by the average percentage of (conspecific) offspring column in Tables 1 and 2. In the A. fasciatus mapping population, positive effects are assigned to markers associated with an increase in fertilization by A. fasciatus males. In the A. socius mapping population, positive effects are assigned to markers associated with an increase in fertilization by A. socius males. In cases where the average percentages and/or IQRs are not notably different or ambiguous (e.g., B6.315.40 in A. fasciatus, Table 1; G7.255.32 or G7.260.15 in A. socius, Table 2), and to confirm all other cases, we use the sign of the additive regression coefficient (not shown). For B6.096.42 (Table 1), since the sample size was so small for the number of individuals possessing the fragment, we used the sign of the additive regression coefficient (negative).

Tables of results for the interval mapping analyses for A. fasciatus and A. socius are divided into markers (or marker intervals) associated with QTL with no control for background QTL (interval mapping) and those associated with QTL while controlling for background QTL (composite-interval mapping; Tables 3 and 4). In cases where a QTL is located with composite-interval mapping, the marker used for background control is noted. Every linkage group in both A. fasciatus and A. socius analyzed with simple interval mapping displayed at least one location with a significant to highly significant bootstrap value, yet rarely were these values associated with an LRS value above suggestive. We report only QTL associated with LRS values that were at least suggestive (Tables 3 and 4) and took high bootstrap values into account only to resolve two borderline cases, B5.190.43 and B6.069.43 in A. fasciatus (Table 3). Generally, linked markers strongly associated with QTL in simple interval mapping also showed high LRS values in single-marker analyses (Tables 1–4). One exception occurred in A. socius, where the interval between B5.304.15 and G3.324.66 on linkage group 1 had a strong positive effect on the trait variance (although supported only by a suggestive LRS; Table 4), yet the individual markers were not picked up by single-marker regression analysis.

In the first block of composite-interval mapping, single-marker regression had identified only one marker in A. fasciatus, B6.069.43 in linkage group 3 (Table 1), with an LRS ≥ 15 appropriate for use as a background QTL. We detected no effect of this marker on any linkage group, including linkage group 3. Single-marker regression revealed no markers with an LRS ≥ 15 in the A. socius data (Table 2). In the second block of composite-interval mapping, in which we controlled for each marker across each linkage group, two markers in A. fasciatus were found to be associated with the trait. Only one of these, B6.069.43, had already been identified with single-marker regression (Table 3), and the other, B5.190.45, had a borderline suggestive LRS. Additionally, composite-interval mapping had the effect of reducing the LRS of B6.069.43 from 15.1, the most significant LRS of the single-marker regression analyses, to the borderline suggestive value of 6.5. This was probably due to concerted background effects of the remaining markers on linkage group 3 (Figure 4). Overall, the range of the percentage of trait variance explained by markers found with interval mapping and composite-interval mapping in A. fasciatus was 9–19%, comparable to that with single-marker regression.

Turning to A. socius, composite-interval mapping revealed a new marker, H5.107.37, with only weak LRS support but moderate negative effect on sperm utilization (Table 4). One of the most surprising results came from the interval between B5.140.99 and H5.292.14 on A. socius linkage group 4 that gave a highly significant LRS of 24.6, the highest in any analysis of either experiment, and explained 69% of the trait variance (positive effect). Interestingly, B5.140.99 by itself had been detected by single-marker regression, as well as by simple interval analysis, but showed only a suggestive-to-significant LRS of 7.2, while still accounting for a very large 29% of the trait variance (positive effect). But by controlling the effect of the nearby H5.292.14 as a background QTL, a genetic factor between the two markers was permitted to show its full effect on the trait (Kearsey 1998). Even in the absence of this marker interval the range of the percentage of trait variance explained by markers found with interval mapping and composite-interval mapping in A. socius was 8–40%, higher than that found in single-marker regression.

Although the backcross mapping approach focuses on species-specific markers, there were 14 AFLP fragments shared between the 116 A. fasciatus and 275 A. socius fragment sets. These shared fragments along with the hundreds of others from which they were filtered underscore the close relatedness and recent divergence of the two species. Of these 14 shared fragments, only one, A. fasciatus marker B6.252.15 and A. socius marker B6.252.03, mapped to linkage groups in both species, but with no trait association: group 4 in A. fasciatus and group 10 in A. socius (Figures 4 and 5). The slight discrepancy in fragment sizes reflects small vagaries in the ABI 3100 sequencer's binning algorithms. Even though B6.252.15/B6.252.03 was a common marker between the species, we retained the linkage group numbering (4 in A. fasciatus and 10 in A. socius) output from Map Manager since linkage groups were derived separately in the A. fasciatus and A. socius QTL experiments. Another shared fragment, B6.088.93 in A. fasciatus and B6.088.98 in A. socius, appeared in A. socius linkage group 6 and was associated with the CSP trait in single-marker regression in A. fasciatus. One other fragment, B6.098.51 present in 31 A. fasciatus individuals and B6.098.65 present in 11 A. socius individuals, was associated with the CSP trait in single-marker regression in both species. In both sets of backcrosses, the presence of this marker increased the use of sperm from the species from which the marker was derived. This marker, although possessing a significant LRS and accounting for 12% of the trait variance in both experiments, appeared to have a stronger effect in A. socius where the average trait value associated with its presence was 76% fertilization by A. socius sperm (IQR 0.64–1.00) vs. 26% fertilization by A. fasciatus sperm (IQR 0.03–0.43) in the A. fasciatus experiment (Tables 1 and 2).

Compared to an earlier study (Howard et al. 2002) we were successful in increasing the number of species-unique AFLP markers available for creating linkage maps and for characterizing QTL in A. fasciatus. We raised 25 AFLP markers on 8 linkage groups (318.4 cM) found in the original experiment to 64 markers on 13 linkage groups (833.9 cM) in the new analysis (Figure 4). With the new analyses we also substantially increased the number of QTL detected in the A. fasciatus experiment (Tables 1 and 3). The preliminary A. fasciatus data had yielded 6 significant QTL (only 2 with a positive effect on conspecific sperm utilization), whereas in the new analysis we identified 14 QTL of positive effect and 4 of negative effect. One interesting outcome of the expanded analysis of the A. fasciatus data was that as more QTL markers were added, the genetic architecture of the trait remained fairly constant. Specifically, in the old and the new analyses of A. fasciatus we located QTL of positive and negative effect (although the proportion of negative QTL was greatly reduced in the new A. fasciatus analysis), QTL associated with linked markers were found on a small proportion of the total number of linkage groups, and linked and unlinked QTL accounted for a range of effects on the trait variances (i.e., some QTL with a small effect and some QTL with a more moderate effect). Of particular interest was the fact that the range of effects of positive QTL in A. fasciatus was stable as we increased the number of markers available for study: the previous A. fasciatus analysis had revealed 2 unlinked positive QTL explaining 7 and 24% of the trait variance; the new single-marker analysis identified 6 positive QTL explaining 9–17% of the trait variance. Moreover, the current interval analysis identified 4 positive QTL explaining 9–19% of the trait variance. Thus, as we added markers to the analysis we did not find that the trait variance became more divided between a greater number of QTL of smaller effect or was confined to fewer markers of large effect; rather, the pattern of QTL of moderate to large effect was simply expanded across more markers (Tables 1 and 3). This stability held true for the negative QTL in A. fasciatus as well, where, except for a single outlier (a negative QTL accounting for 38% of the trait variance), effects of negative QTL ranged from 8 to 18% in the old analysis, comparable to the 9–16% found in the new analysis.

The number of A. fasciatus linkage groups found in Howard et al.'s (2002) study was consistent with the haploid number of chromosomes in Allonemobius (N = 8; Lim 1971); however, in the present study, both for A. socius and A. fasciatus, the numbers of linkage groups are higher. The most straightforward explanation for the discrepancy is that the greater number of AFLP markers identified in the present study allowed clusters of linked markers to form that could not form in the previous study. Markers formed small groups rather than appending to larger groups because they represent segments of large chromosomes physically distant from more easily detected, closely adjacent groups of markers. The karyotype of Allonemobius is characterized by a disproportionately large X chromosome (Lim 1971) that may be the source of one or more of the larger linkage groups as well as one or more of the smaller linkage groups in both A. fasciatus and A. socius. The power to gain uniform coverage of AFLP markers throughout the genome decreases when the mapping population falls below 400 or 500 individuals (Lynch and Walsh 1998), and on a large chromosome such as the X chromosome there may be enough variation in the population that rare, physically isolated clusters of linked markers would be detectable only with either a larger mapping population or a larger repertoire of AFLP markers.

In four instances in the A. fasciatus linkage groups and six instances in the A. socius linkage groups two to three AFLP markers mapped to the same location. In only one case, on linkage group 2 in the A. fasciatus map, did a marker associated with a QTL, H5.311.13, map to the same location as another AFLP marker, G7.164.65. Aside from basic karyotypic work documented by Lim (1971) there are no studies of chromosome polymorphisms in Allonemobius that we are aware of. The fact that some markers map to the same location in both species may mark the presence of inversions, but if there are inversions there is little evidence in this study that QTL for conspecific sperm precedence are located within them or that they underlie the process of speciation in A. fasciatus and A. socius.

The genetic control of CSP in A. socius presents a slightly different picture from the one gleaned from the QTL analysis of CSP in A. fasciatus. In A. socius, AFLP loci accounting for CSP are fewer and of much larger effect than those in A. fasciatus. In addition, A. socius females possess a higher frequency of loci that hinder the use of conspecific sperm, but the loci tend to have weaker negative effects (Tables 2 and 4). The finding that QTL accounted for a higher percentage of the trait variance in the A. socius experiment may have been due to the Beavis effect. The Beavis effect refers to the expectation that additive genetic effects will be overestimated in QTL experiments with smaller sample sizes (i.e., n 500; Beavis 1994, 1998). The Beavis effect also states that QTL of small effect are unlikely to be detected in studies with small sample sizes. Thus, there may be many more loci involved with CSP in both A. fasciatus and A. socius than were detected in this study. In other words, whether the underlying genetics controlling CSP in females involve the accumulated effect of many small loci or a few loci of strong effect cannot be fully answered by the work presented here. However, the numbers of QTL found in both experiments, 18 in A. fasciatus and 13 in A. socius, are comparable to those summarized in Table 1 of Orr (2001), which shows a range of ~1–19 genes accounting for species differences across five groups of insects and plants. Also, the high phenotypic variances accounted for by the QTL (Tables 1–4) suggest that the apparent major effect of these QTL is real.

One finding worthy of discussion is the difference in sperm utilization patterns exhibited by backcross females in the A. fasciatus experiment and the A. socius experiment. Although second-male sperm precedence is a phenomenon observed in many insect species, the work of Howard and Gregory (1993) and Gregory and Howard (1994) showed that sperm mixing occurs in both A. fasciatus and A. socius (the second male fertilizes the same proportion of eggs as the first male in conspecific matings). Moreover, the conspecific male sires >90% of the offspring in matings between A. socius and A. fasciatus, regardless of whether the conspecific male mates first or second. The genomic content of a female offspring of a backcross to A. socius (in the A. fasciatus experiment) is 75% A. socius and 25% A. fasciatus. Assuming additive polygenic control of CSP, backcross females are expected to produce, on average, 25% A. fasciatus offspring if mated once to an A. socius male and once to an A. fasciatus male. The average percentage of offspring sired by A. fasciatus and produced by females from backcrosses to A. socius (i.e., in the A. fasciatus experiment) was 39%, with an IQR of 0.00–0.68. While not a close match to the expected percentage of 25%, more eggs were fertilized by A. socius males than by A. fasciatus males. In contrast, the average percentage of offspring sired by A. socius and produced by females from backcrosses to A. fasciatus (in the A. socius experiment) was 58% with an IQR of 0.15–1.00. In other words, in the A. socius experiment, backcross females, which were expected on average to preferentially use A. fasciatus sperm, were on average preferentially using A. socius sperm. In addition to this, the IQR was skewed to the high end, and many females produced 100% A. socius offspring (data not shown).

In most mating trials in the A. socius experiment, due to a shortage of males, both A. socius and A. fasciatus males were used in multiple matings (mean number of matings = 2.7, SD 1.6; range = 1–7 matings). Matings are physiologically costly to males in the A. fasciatus–A. socius system since males yield large protein-rich spermatophores to the female and allow the female to consume hemolymph. About 24 hr was allotted to males between matings to allow sperm supplies and hemolymph to regenerate. Moreover, if we used a multiply mated male in the first mating, we used a multiply mated male in the second mating that had been used in a similar number of matings as the first male. However, it may be the case that males of the two species recover at different rates, giving A. socius males an advantage in sperm competition because of concentration effects, effects of sperm quality, or effects of senescence. Also, although the target time period between first and second matings was 24 hr, once-mated females were often unwilling to mate with a second male in that time span. In cases of a more extended time between first and second matings, one would expect first-mated A. fasciatus to have a fertilization advantage. We used PROC GLM in SAS v.8 (SAS Institute, Cary, NC) to perform an analysis of variance between the frequency of A. socius offspring and the class variables “number of prior matings A. fasciatus male” and “number of prior matings A. socius males” and the continuous variable “number of days between first and second matings.” We tested for effects of each independent variable as well as interactions between independent variables. PROC GLM indicated that no relationships exist between any combination of mating experience of males of either species, or the lapse between first and second matings, and the variance in the number of A. socius offspring (ANOVA, F = 0.13–2.53, P = 0.12–0.95).

At the very least, the outcome of the ANOVA suggests that the use of males of both species multiple times did not influence our results in the A. socius experiment. However, this phenomenon of more fertilization of backcrosses to A. fasciatus by A. socius males than by A. fasciatus males carries implications for the A. fasciatus–A. socius hybrid zone. The zone has been moving north over 14 years of sampling, as inferred by the increased presence of A. socius in mixed populations and in A. fasciatus populations, sometimes to the point of extinction of pure A. fasciatus types (Britch et al. 2001; S. C. Britch and D. J. Howard, unpublished results). A. socius CSP alleles in hybrid females may have a stronger effect than CSP alleles from A. fasciatus, causing hybrid females to behave like A. socius females with regard to patterns of sperm utilization. Although changing climate has been implicated in the northward movement of the zone (Britch et al. 2001), if both types of backcross females in mixed populations preferentially utilize A. socius sperm in fertilizing eggs, A. socius will become the dominant species in mixed populations over time. Thus, skewed patterns of fertilization in backcross females may be another factor leading the zone of contact between the species to shift north. Patterns of flow across the hybrid zone of AFLP markers associated with CSP are being looked at in a separate study (S. C. Britch and D. J. Howard, unpublished data).

Some discussion of the negative QTL found in both species is warranted since they too were observed to exert a force on the phenotypic variances, albeit in favor of the heterospecific sperm (Tables 1, 2, and 4). The phenomenon of conspecific sperm precedence in both species is very strong (Howard and Gregory 1993; Gregory and Howard 1994; Howard et al. 1998a,b) and the efficacy of the barrier in nature is supported by the strong bimodality of character index scores observed in mixed populations of the two species (Britch et al. 2001). Why is it that not all QTL contribute to the effect and that some appear to have an antagonistic effect on sperm utilization? Howard et al. (2002) discussed one possible explanation for the negative QTL, that is, sexual conflict (reviewed in Howard 1999 and Panhuis et al. 2001). Briefly, the argument is that the multiple mating that is characteristic of Allonemobius females establishes sperm competition as an important selective pressure on males and leads to antagonistic coevolution between the two sexes in traits related to fertilization. This sexual conflict leads to females possessing genes that have antagonistic effects on the sperm of conspecific males. Thus, uncovering QTL with antagonistic effects on conspecific sperm may be seen as providing additional empirical support for the concept of sexual conflict and its importance in driving the evolution of barriers to fertilization between closely related species (Rice 1996, 1998a,b; Howard 1999; Gavrilets 2000; Knowles and Markow 2001; Wiklund et al. 2001; Miller and Pitnick 2003; Morrow and Arnqvist 2003; Orteiza et al. 2005).

The presence of negative QTL in both species could also be explained by epistasis (Howard et al. 2002). Epistatic interactions are clearly the basis of hybrid male sterility and inviability in Drosophila (Wu and Hollocher 1998; Coyne and Orr 1999). In the case of Drosophila, alleles that behave normally in the genetic background of their own species cause hybrid sterility and inviability when introduced into the genetic background of another species. In the present situation, the negative QTL could represent the effects of any kind of gene (i.e., not necessarily a locus that has anything whatever to do with the biochemistry of conspecific sperm precedence when present in its “home” genetic background) that has a negative fertilization side effect when bred into a genome that is 75% the opposite species (Figure 2).

Orr (2001) summarized important questions regarding the genetic architecture of species differences, two of which, the number of genes involved and the magnitude of their phenotypic effects, we have been able to address. Orr (2001) also noted that relative to natural selection, sexual selection may increase the complexity of genetic changes accompanying the evolution of species differences. As evidence for this point of view, Orr (2001) cited two Drosophila studies that found 19 and 11 QTL controlling divergence in male genitalia (a rapidly evolving trait in insects that is thought to be under strong sexual selection). This is a greater number of QTL than identified in Drosophila and other taxa for species-specific traits that are not subject to sexual selection. The numbers of CSP QTL that we found in both species are comparable to the numbers found in the studies cited in Orr (2001), namely 18 in A. fasciatus and 14 in A. socius. As far as the distribution of phenotypic effects among genes important in species differences (Orr 2001), we do not have the experimental resolution to distinguish whether a strong QTL is actually one gene or the combined effect of several adjacent genes that are individually weak. The majority of QTL in both A. fasciatus and A. socius appear to cover a small range of effects, from low to moderate; although a few QTL exist in A. socius with very large effects (Table 4). The QTL in both species are spread across many unlinked markers, yet the linked markers associated with QTL are on a relatively small number of linkage groups. Some evidence points to genes controlling prezygotic isolation being concentrated on the sex chromosomes (Servedio and Saetre 2003) and it may be that these few linkage groups in A. fasciatus and A. socius are actually fragments of the large sex chromosome characteristic of Allonemobius (Lim 1971).

We thank T. Parchman and S. Long for discussions regarding interpretation and troubleshooting of AFLPs, K. Hopper for discussions regarding QTL analysis, and K. Edwards and C. Johnson for expert help in the lab. This work was supported in part by a New Mexico State University Department of Biology Excellence in Research Fellowship to S.C.B. and in part by National Science Foundation grants DEB 011613 and DEB 0316194 to D.J.H.