Drosophila santomea and D. yakuba are sister species that live on the African volcanic island of São Tomé, where they are ecologically isolated: D. yakuba inhabits low-altitude open and semiopen habitats while D. santomea lives in higher-elevation rain and mist forest. To determine whether this spatial isolation reflected differential preference for and tolerance of temperature, we estimated fitness components of both species at different temperatures as well as their behavioral preference for certain temperatures. At higher temperatures, especially 28°C, D. santomea was markedly inferior to D. yakuba in larval survival, egg hatchability, and longevity. Moreover, D. santomea females, unlike those of D. yakuba, become almost completely sterile after exposure to a temperature of 28°C, and conspecific males become semisterile. Drosophila santomea adults prefer temperatures 2–3°C lower than do adults of D. yakuba. Drosophila santomea, then, is poorly adapted to high temperature, partially explaining its restriction to cool, high habitats, which leads to extrinsic premating isolation and immigrant inviability. Rudimentary genetic analysis of the interspecific difference in egg hatchability and larval survival showed that these differences are due largely to cytoplasmic effects and to autosomal genes, with sex chromosomes playing little or no role.

Despite Drosophila's long history as a model for genetic analysis, the genus has lagged in the study of adaptation. This is because, except for cosmopolitan human commensals, we know very little about the ecology of the most well-studied fly taxa, such as the Drosophila melanogaster and Drosophila obscura groups. Further, many of the known adaptive differences between closely related Drosophila species occur in clades that are not genetically well characterized or involve taxa that cannot produce the fertile hybrids required for genetic analysis (e.g., Spicer and Jaenike 1996; Fogleman and Danielson 2001; Gibert et al. 2001; Simunovic and Jaenike 2006). One exception is the adaptive behavior and physiology of Drosophila sechellia, a species in the D. melanogaster subgroup that feeds and breeds on fruits toxic to all its closest relatives (R’kha et al. 1991; Amlou et al. 1998; Jones 2001, 2004; Matsuo et al. 2007; McBride 2007). The dearth of genetic studies of adaptations in Drosophila contrasts with the many studies involving interspecific differences in traits of unknown adaptive significance (e.g., bristle number), or of traits that probably evolved by sexual rather than natural selection (e.g., mating pheromones or genital shape, Liu et al. 1996; True et al. 1997).

Here we describe differences in temperature tolerance and preference between two sister species in the D. melanogaster subgroup: D. santomea and D. yakuba. These differences involve both physiological tolerance to and behavioral preference for temperature, might help explain the known ecological distribution of these species, and probably produce some premating reproductive isolation.

Drosophila yakuba, in the D. melanogaster subgroup, is widespread throughout sub-Saharan Africa, occupying a diversity of more or less open habitats including semiarid areas, savannas, montane grassland, and semidomestic habitats (notably coffee and cacao plantations). It is not found in rainforests. Drosophila yakuba has also extended its range to neighboring islands, including Madagascar and Zanzibar in the western Indian Ocean and the Gulf of Guinea islands in the eastern Atlantic Ocean, which include Annobon, Principe, and São Tomé (Lachaise et al. 1988, 2000).

Also inhabiting São Tomé—an 860 km2 volcanic island 250 km off the west coast of Gabon—is D. yakuba’s endemic sister species, D. santomea, discovered in 1998 (Lachaise et al. 2000). On the island, D. santomea occupies montane mist forests above 1150 m on the slopes of the main volcano, Pico de São Tomé (2024 m). In contrast, D. yakuba is found exclusively in towns, disturbed sites, cut-over areas, open plantations, and edges of the rainforest below 1400 m on the slopes of Pico de São Tomé.

The ranges of the two species overlap between the altitudes of 1150 m and 1400 m, forming a zone in which one finds a low frequency of hybrids (about 1%; Lachaise et al. 2000; Llopart et al. 2005a). Curiously, a second “anomalous hybrid zone” occurs above the range of D. yakuba: extending from about 1500 m to the top of the Pico de São Tomé. In this second area, one finds only a few D. santomea, but—surprisingly—a very high frequency of hybrids between the two parental species. (93% of all melanogaster-group flies are yakuba/santomea hybrids; Llopart et al. 2005a). These hybrids are exclusively male F1 offspring from a D. yakuba father and a D. santomea mother (Llopart et al. 2005a,b). The explanation of the anomalous hybrid zone remains a mystery; one suggestion is that males of the proper genotype prefer lower temperatures and preferentially move from the regular to the “anomalous” hybrid zone.

Molecular evidence puts the divergence between D. yakuba and D. santomea at about 400,000 years ago (Llopart et al. 2002). It is likely that D. santomea descends from a common ancestor with D. yakuba that colonized the island at that time, and that the present contact between D. santomea and D. yakuba reflects secondary colonization by D. yakuba from the African mainland. This secondary contact probably occurred during the last 500 years when the Portuguese colonists turned large sections of coastal rainforest into plantations (Llopart et al. 2005a). There is some gene exchange between these species, but it is limited to mtDNA (which, courtesy of a recent introgression from D. yakuba, is virtually identical in the two species) as well as two nuclear regions (Llopart et al. 2005b; Bachtrog et al. 2006).

Morphological differences between these species include pigmentation (both males and female D. santomea are yellow, lacking black abdominal pigmentation characteristic of other species in the D. melanogaster subgroup), the shape of male genitalia, and the size of the sex combs. These differences have been analyzed genetically (Coyne et al. 2004; Carbone et al. 2005; Moehring et al. 2006a,b; Jeong et al. 2008).

The species also show substantial intrinsic reproductive isolation, including strong sexual isolation in the laboratory, sterility of male hybrids, and various forms of postmating, prezygotic isolation (Coyne et al. 2002, 2005; Chang, 2004; Llopart et al. 2002, D. R. Matute and J. A. Coyne, unpubl. data). Additional “extrinsic” reproductive isolation stems from the species’ altitudinal zonation, which limits their contact. Here we study whether that zonation reflects interspecific differences in genetic tolerances to or preferences for different temperatures, or merely ecological competition that displaces the species to different locations.

Materials and Methods and Results


All stocks were reared on standard cornmeal/Karo/agar medium at 24°C under a 12 h light/dark cycle.

We used eight lines of D. santomea. Five of these (STO.7, STO.4, STO.10, STO15, and STO18) were derived from single females caught in 1998 by D. Lachaise in the forest of the Obo Natural Reserve on São Tomé Island. Two additional lines, CAR 1600.1 and CAR 1566.1, derived from single females collected in 2001 by JAC between 1490 m and 1600 m on the Pico de São Tomé, also in the Obo Reserve. To avoid artifactual results due to any inbreeding within isofemale lines, we also used a “synthetic” strain of D. santomea made by combining offspring from several lines. This stock was S2005SYN, initiated as a mixture of six isofemale lines collected in 2005 at the field station Bom Successo (elevation 1150 m) at the lower edge of the “regular” hybrid zone on São Tomé. This line was always kept in large numbers (at least 24 bottles) since initiation.

We used seven lines of D. yakuba. Six of these were isofemale lines, four derived from flies collected on São Tomé by D. Lachaise in 2001 (SJ3, from southern São Tomé outside the hybrid zone; São Nicolau 20, from the São Nicolau waterfall; SA1 from Obo Natural Reserve inside the lower hybrid zone; and BOSU1153.1 from the field station at Bom Successo). The other two isofemale lines came from mainland Africa, one from Cameroon (Cameroon 115, collected by D. Lachaise in 1967), and the other from the Ivory Coast (Täi 18, collected by D. Lachaise in 1981 from the Taï rainforest on the border with Liberia). We also used a “synthetic” line of D. yakuba, Y2005SYN, constructed by combining five stocks collected by JAC in January 2005 at elevation 880 m on the Pico de São Tomé; this stock was also kept in large numbers after initiation.


Data were analyzed using standard statistical procedures described in each section below. For analysis of variance (ANOVA), we ensured that the data conformed to the assumptions of normality and homoscedasticity, either before or after transformation.


We made large groups of pure species and hybrid matings for collecting eggs and larvae. For pure species, we collected newly eclosed males and females and kept them in groups of about 100 flies for three days. On day 4, all flies were transferred to bottles containing only moist yeast paste and moist filter paper, where they were held for 12 h to mature and retain eggs. Flies were then transferred to food tinted with blue vegetable coloring or grape juice for collection of eggs and larvae, respectively. The egg-laying period was 6 h. Eggs were collected immediately thereafter, and first-instar larvae were collected 24 h thereafter.

For collecting hybrid eggs or larvae, males and females were collected as virgins under CO2 anesthesia and kept for three days in single-sex groups of 20 flies in eight-dram, food-containing vials. On the morning of day 4 we placed males and females together at room temperature (21–23°C) to mate for 9 h en masse on corn media. Females and males were then transferred without anesthesia to bottles containing yeast paste and water, where they were retained for 12 more hours. They were then transferred to either blue-dyed or grape-juice-dyed food as described above.


Groups of 50 first-instar larvae, collected 24 h after egg-laying, were placed in individual vials. The vials were then divided among four temperatures: 15°C, 21°C, 24°C, and 28°C. After larvae developed into adults, we counted males and females. We measured the survival of between 850 and 2100 larvae for each line at each temperature.

Six wild-type isofemale lines each of D. santomea and eight lines of D. yakuba were tested for larval survival over four temperatures (15°C, 21°C, 24°C, and 28°C). Figure 1 shows the mean survival among lines for both D. yakuba and D. santomea. (Supporting Fig. S1A gives results for the individual lines).

Figure 1.

Larval survival of D. yakuba and D. santomea strains at different temperatures. Standard errors of the mean among strains tested are shown (values for individual strains shown in Supporting Fig. S1).

Given the unbalanced nature of the experimental results, we estimated missing values by resampling the data within the cells (this approach was also applied to genetic analysis of egg hatchability). For D. yakuba, ANOVA on the arcsine-transformed proportion of surviving larvae (see Supporting Table S1) showed that both main effects (line and temperature) as well as their interaction, were highly significant (respectively, F6,707= 6.36, F3,707= 84.23, F18,707= 2.67, all P < 0.0001). Survival was lowest (around 35–45%) at 15°C, rose to around 60–65% at 21°C and 24°C, and dipped slightly at 28°C. We used Tukey's HSD test (honestly significant difference; Sokal and Rohlf 1995) to determine post facto whether means from ANOVAs were significantly different at the probability level of 0.05. This test shows that for D. yakuba the temperatures of 15°C and 28°C produce significantly lower larval survival than that seen at intermediate temperatures.

The survival of D. santomea larvae (Fig. 1) showed a strikingly different pattern (Supporting Fig. S1B gives results for the individual lines). Although survival is lower at 15°C than at intermediate temperatures, unlike D. yakuba survival drops slightly between 21°C and 24°C, and then plummets to almost zero at 28°C. As in D. yakuba, the main effects of lines and temperature, as well as their interaction, were all highly significant under the ANOVA (respectively, F7,568= 2.91, P < 0.0001, F3,568= 969.68, P < 0.003, F21,568= 2.32, all P < 0.003; Supporting Table S1 shows the full analysis). According to the Tukey HSD test, mean larval survivorship at each temperature differed significantly from that at every other temperature. At no temperature did D. santomea larvae survive better than those of D. yakuba. Clearly, high temperature is a significant stressor of D. santomea larvae, as one might expect from the altitudinal distribution of the species on São Tomé.

Combining data from both species, an ANOVA (using arcsine-transformed data) shows that all main effects (temperature, species, and lines within species) were highly significant (respectively, F3,1275= 667.95, F1,1275= 679.45, F13,1275= 4.44, P= 0.0001; all P < 0.001; Supporting Table S1 shows the full analysis), as was the interaction of species with temperature (F3,1275= 427.46, P < 0.001). This confirms that the species differ significantly in their profile of temperature tolerance. Comparing the species’ survival at each temperature using Tukey's HSD, we see that D. yakuba larvae survive significantly better than those of D. santomea at all temperatures except 21°C, where there is no difference.

We performed two preliminary and rather crude genetic analyses of larval survival: one set involved comparing survival of the pure species and their reciprocal F1 hybrids at all four temperatures; the other involved comparing these genotypes as well as offspring from all four possible backcrosses at only the stress temperature of 28°C. Each set was run with two pairs of strains: D. yakuba Taï 18/D. santomea STO.4, and D. yakuba Y2005SYN/D. santomea S2005SYN.

Reciprocal crosses are usually used to determine the effect of cytoplasm and of X and Y chromosomes on a trait, because males from reciprocal crosses differ in the source of their sex chromosomes and cytoplasm, while F1 females from reciprocal crosses differ only in their cytoplasm. Unfortunately, we cannot distinguish the sexes as early larvae and so cannot directly compare survivorship of males and females. However, in none of the genotypes or crosses tested did the sex ratio of adults reared from the tested larvae differ significantly from 1:1 (Supporting Table S2). Such a difference would be expected if one sex died preferentially in the cross. That males and females survive equally well (or poorly) at all temperatures suggests that at no temperature do the X and Y chromosomes affect differences in survival.

Figure 2 shows the effect of temperature on larval survival of pure species and F1 hybrids for the two pair of lines we compared. In both sets of crosses the survival of pure species was nearly identical to that seen in the independent test of the same lines (see Supporting Table S1).

Figure 2.

(A) Larval survival of both pure species and reciprocal F1 hybrids, isofemale lines. (B) Larval survival of both pure species and reciprocal F1 hybrids, synthetic lines.

For both sets of crosses, ANOVA showed that the effects of genotype and temperature were highly significant, as was their interaction. (For the experiment involving all four temperatures: genotype: F3,772= 111.46; line: F4,772= 8.12; temperature: F3,772= 185.94, genotype × temperature: F3,772= 59.87; all P < 0.0001. For the experiment at 28°C: genotype: F7,400= 157.39; line: F8,400= 7.19; both P < 0.0001.) Because absence of distorted sex ratios among hybrids suggested that neither X nor Y chromosomes affected temperature-dependent survival, we compared the overall survival of reciprocal F1 hybrids as a test of cytoplasmic effects using the Tukey HSD. For both isofemale and synthetic strains, hybrids with D. yakuba cytoplasm survived better than those with D. santomea cytoplasm at 21°C, 24°C, and 28°C. At 15°C there was no difference for isofemale lines in the Taï 18/STO.4 comparison; but in the SYN cross, hybrids with D. santomea cytoplasm survived significantly better than those with D. yakuba cytoplasm. The pattern of response was messy, as hybrids were not intermediate to parental species at most temperatures. (See Supporting Table S1 for statistical analyses.)

While cytoplasm obviously plays a role in larval temperature tolerance—not surprising given that larval cytoplasm derives from the maternal parent—genes do as well. This is because the survival of hybrid larvae is not identical to that of pure-species larvae that have the identical cytoplasm. At 28°C, for instance, F1 hybrids with D. santomea cytoplasm survive significantly better than do pure D. santomea individuals, undoubtedly because these hybrids carry genes from D. yakuba. Further, these genes must be autosomal because of the apparent lack of an X or Y chromosome effect on survival.

Although F1 hybrids with D. santomea cytoplasm (the genotype found in the “anomalous hybrid zone”) survived significantly better than pure-species individuals at 15°C, this advantage was not large enough to explain the absence of the reciprocal-male genotypes from the two hybrid zones. Moreover, the lack of distorted sex ratios at any temperature suggests that female hybrids survived as well as males, so larval survival obviously cannot explain the absence of females in this area.

We retested the larval survival of pure species and F1 hybrids, and newly tested all four backcross offspring at 28°C. Figure 3 gives the survival of all genotypes in both comparisons, along with standard errors of mean survival (genotypes in this figure are numbered for reference in the discussion below). No cross showed a significant distortion of sex ratio, again suggesting a lack of X or Y chromosome effects on interspecific differences in survival.

Figure 3.

F1 and backcross analysis of pure-species and hybrid larval survival at 28°C: isofemale and synthetic lines. For F1 females used in backcrosses (shown in parentheses), female parent is given first.

An ANOVA on arcsine-transformed survivorship of all genotypes for the combined sets of crosses (each pair of compared genotypes being considered a “line”) shows that the main effect of genotype is significant, as is the effect of line within genotype (genotype: F7,400= 157.39; line: F8, 400= 7.19; both P < 0.0001). The results from the two sets of crosses thus differ significantly from one another. However, inspection of Figure 3 shows that the general trends are the same in each set, with a lower survival of F1 hybrids having D. santomea cytoplasm than hybrids having D. yakuba cytoplasm—just as we saw in the set of crosses displayed in Figure 2. As expected, F1s and backcrosses generally show temperature tolerances intermediate to those of parental species.

The effects of cytoplasm and autosomal genes were gauged by comparing, using Tukey's HSD, the survival of pairs of genotypes differing by only one of these factors. For example, comparing genotypes 3 and 4 from reciprocal F1 crosses shows the size of cytoplasmic effects on survival (these genotypes differ only by cytoplasm and sex chromosomes, but the absence of a distorted sex ratio implies no significant effects of the sex chromosomes). The Tukey test shows that in both sets of crosses, F1 larvae with D. santomea cytoplasm survive significantly worse than those with D. yakuba cytoplasm. Similarly, the comparison of genotypes 5 versus 7 and 6 versus 8 shows significant cytoplasmic effects: offspring in each of these comparisons are genetically identical but differ in their source of cytoplasm. In the overall analysis, all six independent comparisons show that D. santomea cytoplasm lowers larval survivorship at 28°C. Our observation that the cytoplasmic effects appear in backcrosses as well as in the F1s, despite the fact that maternal genotypes are different in these two sets of crosses, implies that these are true cytoplasmic effects (e.g., mitochondria or some other substance residing in the cytoplasm) rather than effects produced by the maternal genotype.

Similarly, we can test for the effect of chromosomes in backcrosses by comparing pairs of genotypes having identical cytoplasm but different chromosomes: genotypes 5 versus 8 and 6 versus 7. In both sets of comparisons (four total), Tukey's HSD shows that, as expected, D. yakuba genes confer significantly higher larval survival at 28°C than do D. santomea genes. Because there is no difference in sex ratios in any of these comparisons, we can again conclude that differences in survival are caused by autosomal genes rather than X- or Y-linked genes.


Adults collected as described above were transferred to blue-tinted medium for egg laying. Eggs were collected after 6 h and transferred in groups of 50 onto small squares of black filter paper, which were placed in vials containing regular medium. Flies were resequestered with yeast paste and then again placed on tinted food the next day. This procedure was repeated twice more, after which egg production by females dropped off significantly. Each collection of eggs was split into four groups and incubated at four temperatures: 15°C, 21°C, 24°C, and 28°C. After 12 h of incubation, we counted hatched and unhatched eggs, scoring them every 12 h until no more eggs hatched (this took up to four days at 15°C). To determine the sex ratio of hatched eggs, we transferred all the vials to 24°C and tended them until all adults emerged, scoring male and females. The number of eggs scored for hatchability in each line at each temperature varied between 750 and 2200. The survival from hatched eggs to adults was between 50% and 60% at 24°C.

Figure 4 shows means of the hatchability of eggs laid by lines of D. yakuba and D. santomea at different temperature (Supporting Fig. S2 shows data from the individual lines). For both species, the relationship between egg hatchability and temperature followed the same trend as did larval survival: a general superiority of D. yakuba over D. santomea, a general increase in hatchability from 15°C to 24°C, and a drop in fitness at 28°C that is far more severe for D. santomea than for D. yakuba. The overall ANOVA on arcsine-transformed data from both species together shows highly significant effects of species, lines within species, temperature, and the temperature × species interactions (respectively, F1,1177= 94.11, F14,1177= 4.00, F3,1177= 870.86, F3,1177= 146.20, all P < 0.0001). Clearly, the species differ significantly in the hatchability of their eggs over this array of temperatures. Drosophila yakuba does significantly better than D. santomea at all temperatures, but the difference is particularly striking at the highest and lowest temperatures. Among the four temperatures tested, the highest hatchability of D. santomea was seen at 21°C, and for D. yakuba at 21°C and 24°C, which were not significantly different.

Figure 4.

Egg hatchability of D. yakuba and D. santomea strains at different temperatures. Standard errors of the mean among strains tested are shown (values for individual strains shown in Supporting Fig. S2).

ANOVAs performed on each species separately showed highly significant main effects of temperature (D. yakuba: F3, 427= 607.12, D. santomea: F3, 612= 553.97, both P < 0.0001), and the effect of line was highly significant for D. santomea (F8, 612= 7.14, P < 0.001) and D. yakuba (F6, 423= 2.581, P= 0.0182). The line × temperature interaction was significant for D. santomea (F6, 612= 6.08, P < 0.0001) but not for D. yakuba (F18, 423= 0.70, P= 0.82). For both species, hatchability at each temperature was, according to Tukey's HSD, significantly different from hatchability at all three other temperatures.

For genetic analysis of interspecific differences in hatchability, we did two sets of comparisons, one using a pair of isofemale lines (Taï18 and STO4) and the other a pair of synthetic lines (Y2005SYN and S2005SYN). Interspecific F1 hybrids were obtained by making both reciprocal crosses within each pair of lines. Six to eight one-day-old virgin D. yakuba or D. santomea females were mated in vials to twice as many one-day-old virgin males from the other species. Flies were changed to fresh vials after two days to produce progeny.

We measured larval survival rate at 28°C in the progeny of eight different crosses (S =D. santomea, Y =D. yakuba, and female parent shown first): two intraspecific crosses (S × S, Y × Y), two interspecific crosses (S × Y, Y × S), and four backcrosses ([S × Y]× Y, [S × Y]× S, [Y × S]× Y, and [Y × S]× Y).

Egg hatchability and larval survival of pure species and F1s were measured at 15°C, 21°C, 24°C, and 28°C, and in a separate experiment these genotypes and offspring survival from the four reciprocal backcrosses were scored at the stress temperature of 28°C. Sample sizes for estimates of egg hatchability varied between 1000 and 2500 individuals per genotype or cross; sample sizes for studies of larval survival ranged between 850 and 1400 individuals.

Figure 5 shows egg hatchability for two isofemale lines (STO.4 and Täi 18; Fig. 5A) and synthetic lines (S2005SYN and Y2005SYN; Fig 5B) and of eggs laid from the two reciprocal crosses at four temperatures. An ANOVA of arcsine-transformed proportions showed highly significant effects of genotype (F3,644= 249.9972, P < 0.0001), line nested within genotype (F3,644= 5.1951, P= 0.0015), and temperature (F3,644= 273.9647, P < 0.0001). The general pattern for the pure species resembles that shown for pure species in Figure 4: D. yakuba generally does better than D. santomea, and the hatchability of D. santomea eggs drops sharply at 28°C. At 15°C and 28°C, the hatchability of D. santomea eggs is significantly lower than that of D. yakuba, but there is no difference at two intermediate temperatures. Interspecific hybrids are not always intermediate in hatchability.

Figure 5.

(A) Egg hatchability of both pure species and reciprocal F1 hybrids, isofemale lines. (B) Egg hatchability of both pure species and reciprocal F1 hybrids, synthetic lines.

The sex ratios of flies hatched from these eggs did not differ from 50:50 at any condition, so that differences between the hatchability of eggs laid by females from reciprocal crosses can be attributed to cytoplasmic effects rather than to differences in the sex chromosomes of developing embryos. Comparing the means of reciprocal F1 hybrids using Tukey's HSD test gives similar results across both sets of crosses. At 15°C, hybrid eggs with D. santomea cytoplasm have significantly higher hatchability than do eggs with D. yakuba cytoplasm, whereas the reverse is true at the other three temperatures. The hatchability of hybrid eggs is not consistently in between the values for their parental species. Besides the cytoplasm, the autosomes also affect the trait difference, for the hatchability of eggs laid by F1 hybrids differs from that of pure-species eggs having the same cytoplasm.

Figure 6 shows the hatchability of eggs in pure species, interspecific crosses, and backcrosses at 28°C for isofemale lines Taï 18 and STO.4 (white bars) and 2005SYN lines from both species (black bars). The trends for both the pure species and reciprocal F1s are similar to those shown in Figure 5. In general, the hatchability of eggs from D. santomea is lower than that of D. yakuba, the hatchability of eggs from the F1 interspecific cross is lower then either of these, and the hatchability of eggs produced by backcross hybrids is intermediate to these.

Figure 6.

F1 and backcross analysis of egg hatchability of pure species F1-hybrid, and backcross females at 28°C: isofemale and synthetic lines.

An ANOVA on arcsine-transformed data shows that for the combined data for both sets of lines, there is a significant effect of strain pair used (F8,224= 3.847, P= 0.0003), as well as of genotype within strain pair (F7,224= 137.75, P < 0.0001). According to the Tukey's HSD test, there is no difference between hatchability of two pure D. yakuba strains or between the two pure D. santomea strains.

Because there was no significant deviation from a 50:50 sex ratio in any cross, we can get information about cytoplasmic effects from reciprocal F1 crosses. Comparing the hatchability of genotypes 3 and 4 using the Tukey's HSD test, we find that for both sets of crosses, hybrid eggs having D. santomea cytoplasm show significantly lower hatchability than those with D. yakuba cytoplasm, exactly as seen in Figure 5.

We can also compare cytoplasmic effects by looking at hatchability of eggs laid by hybrid females in backcrosses to pure species. Comparing genotypes 5 and 7 using Tukey's HSD test, we find a significant difference in hatchability in the comparisons of both isofemale and “synthetic” strains, with D. yakuba cytoplasm conferring higher hatchability in both cases. No significant effects, however, were found for the two comparisons of genotypes 6 and 8.

The effects of genes on egg hatchability can be seen by comparing pairs of genotypes having different chromosomes but identical cytoplasm. Using Tukey's HSD test, genotypes 5 versus 6 were significantly different for both comparisons, with D. yakuba chromosomes conferring significantly higher egg hatchability. For genotypes 7 versus 8, D. yakuba chromosomes conferred significantly higher hatchability in the comparison of isofemale lines but not of the SYN lines.

All tests show consistent effects of the cytoplasm on egg hatchability at 28°C, with cytoplasm from D. yakuba conferring significantly higher hatchability than that from D. santomea. Likewise, genes from D. yakuba—likely autosomal genes because no effect of sex chromosomes was evident in the reciprocal crosses—confer significantly higher hatchability at this temperature.


Because high temperature proved lethal to nearly all eggs and larvae of D. santomea but not of D. yakuba (see below), we wanted to determine if high temperature also affected male or female fertility of both species. Preliminary experiments showed that for D. yakuba, adults reared and collected as virgins as 24°C produced copious offspring when transferred to 28°C, but that D. santomea treated identically produced virtually no offspring. We therefore treated D. santomea males and females separately with high temperature to determine which sex was sterilized. We also treated D. yakuba in the same manner.

Males and females raised at 24°C were collected as virgins and immediately placed individually in food-containing vials at either 24°C or 28°C for three days. This constituted the premating temperature treatment. On day 4, flies were transferred into empty vials for mating observations. These tests involved flies that had undergone four treatments: females kept at 24°C × males kept at 24°C; females kept at 24°C × males kept at 28°C; females kept at 28°C × males kept at 24°C; and females kept at 28°C × males kept at 28°C.

Mating observations were conducted at room temperature (21–23°C) starting within the first hour of the 12-h light cycle. One female and one male were transferred without anesthesia into a fresh food vial for observation; the time was recorded at the beginning and end of each copulation. Up to 60 vials were observed simultaneously. We discarded females who did not mate after 1 h of observation. Males were removed from the vial by aspiration after copulation. Egg laying and hatchability were then scored at 24°C for all mated females. Each female was allowed to oviposit for 24 h in a vial, after which the number of eggs laid was scored and the females were transferred to a new vial. This was done daily for 10 days. Between 15 and 22 flies were tested for each type of mated female in each of the four temperature combinations. Figure 7 shows the results for D. santomea, D. yakuba, and the three outgroup species. ANOVAs for heterogeneity of fertility among pair treatments are significant for all strains and species except D. tessieri (D. santomea STO4: F3,96= 701.50, P < 0.0001; D. santomea SYN2005: F3,96= 388.04, P < 0.0001; D. yakuba Tai18: F3,36= 42.88, P < 0.0001; D. yakuba SYN2005: F3,120= 36.53, P < 0.0001; D. mauritiana SYN: F3,72= 13.60, P < 0.0001; D. simulans FC: F3,76= 16.83, P < 0.0001; D. tessieri: F3,76= 2.34, P= 0.08).

Figure 7.

Egg production of pure species from crosses using flies reared at different temperatures. Phylogeny shows the evolutionary relationships of these species in the D. melanogaster subgroup.

On each set of bars in Figure 7 are superimposed the results of Tukey's HSD test for the different temperature combinations (bars bearing identical letters are not significantly different). The most striking result is that the fertility of D. santomea females is severely reduced when they are treated for three days at 28°C: such females show roughly 5–10% of the fertility of females raised at 24°C. Males of D. santomea show a much smaller but still significant effect of temperature: when mated to females reared at 24°C, high-temperature males show a fertility about 74–83% that of control males raised at 24°C. In contrast, treatment of D. yakuba females at 28°C also significantly lowers their fertility, but to a smaller extent than seen in their sister species: fertility drops from 57% to 65% of the value observed at 24°C. Moreover, ANOVA shows no significant effect of high-temperature rearing on the fertility of D. yakuba males. Clearly, reproduction is impeded by high temperature much more severely in D. santomea than in D. yakuba.

To determine whether the temperature-dependent female sterility of D. santomea (compared to D. yakuba) was a derived or ancestral trait, we tested the three outgroup species D. mauritiana, D. simulans, and D. tessieri. The D. simulans Florida City (FC) isofemale line was collected by JAC in 1985 in Florida City, Florida and maintained in large populations. The D. mauritiana synthetic (SYN) stock was a wild-type line synthesized by combining six isofemale lines collected in Mauritius by O. Kitagawa in 1981. The D. tessieri line BRZ8 was collected by J. David in Brazzaville, Congo. Between 15 and 22 male–female pairs from each species were tested for fertility at each of the four combinations of sex and temperature (24°C and 28°C).

Figure 7 gives the phylogeny for these species. Drosophila tessieri is the closest outgroup species, with D. simulans and D. mauritiana being equally and more distantly related to the sister pair D. santomea/D. yakuba. ANOVAs among all three outgroup species showed significant heterogeneity among temperature treatments for D. simulans and D. mauritiana but not for D. tessieri. We used Tukey's HSD test to examine differences between the intraspecific pairings in egg output. For D. tessieri, no pairing was significantly different from any other, indicating no temperature-dependent sterility of either sex. For D. simulans, rearing at 28°C significantly reduced the fertility of females but not males. Conversely, in D. mauritiana high temperature significantly (but only slightly) reduced the fertility of males whereas that of females was unaffected.

Overall, the extreme sterility of D. santomea females treated at 28°C is not seen in any outgroup species, and thus appears to be a derived character. The smaller amount of temperature-dependent male sterility in D. santomea also appears to be derived, as it is not consistently seen in any other species. Drosophila santomea appears to have evolved an extreme sensitivity of fertility to high temperature, possibly the result of its adaptation to a cooler, higher-elevation habitat.


We measured the longevity of D. santomea and D. yakuba adults at two temperatures: 21°C and 28°C. At each temperature we made two comparisons, each involving a different pair of strains as well as their reciprocal F1 hybrids. One comparison consisted of D. yakuba Taï 18 and D. santomea STO.4 and their hybrids; the other D. yakuba YSYN2005 and D. santomea SSYN2005 and their hybrids. Males and females were tested separately, so that each comparison of longevity involved eight genotypes.

All flies whose longevity was tested were raised to eclosion at 24°C. For each test, 10 eight-hour-old virgin individuals of a single genotype were placed in an 8-dram, food-containing vial placed on its side in an incubator. Each day for 10 successive days, we supplemented the experiment by setting up eight vials with the two pure species and their F1 hybrids. Every vial of flies was scored once daily to determine how many were still alive, with flies were transferred to fresh food as needed (usually every two to three days). If flies showed any movement, even if they were stuck to the food and twitching feebly in the throes of death, they were scored as “live.” Dead flies were not replaced in a vial. The experiment continued until all flies had died. We began with 100 flies of each genotype and sex, but due to escapees the final numbers among tests ranged between 59 and 100.

Differences between types of crosses in longevity were analyzed by fitting a random linear mixed model (Pinhiero and Bates 2000) for the median longevity per vial with the main effects of genotype (two pair of pure species + two hybrids), line (nested within genotype), gender, temperature, and the interactions between these factors. We took the median longevity (in days) per vial as the response in the model. The differences between vials were considered random effects.

Table 1 shows the mean longevity, sample sizes, and significance level (using the Tukey–Kramer HSD test) for each comparison of the two pure species and their two F1 hybrids. As expected, flies lived considerably longer at 21°C than at 28°C (two to four times as long, depending on the genotype). Generally, males lived longer than females at 21°C, whereas the reverse was true at 28°C.

Table 1.  Mean longevity (days) at 21°C and 28°C (standard errors in parentheses). Sample sizes are shown below the longevities. Results of Tukey–Kramer HSD are given by letters, which should be compared among the four cells in each column (cells with identical letters do not differ significantly). Each of the two comparisons between different D. yakuba and D. santomea strains involves four tests (two sexes × two temperatures), each corresponding to one column of four cells. In crosses, female parents are given first.
Comparison 1
  san STO.461.90 (2.50)80.24 (2.21)15.19 (0.68)13.58 (0.59)
  yak Taï 1861.05 (2.33)72.20 (2.16)35.71 (0.73)31.35 (0.62)
  F1 STO.4 × Taï 1884.10 (2.48)115.42 (2.44)40.07 (0.78)37.68 (0.91)
  F1 Taï 18 × STO.473.57 (2.58)89.65 (2.37)40.60 (0.62)35.68 (0.54)
Comparison 2
  S2005SYN80.75 (2.79)82.04 (2.83)24.73 (0.83)22.29 (0.82)
  Y2005SYN84.22 (2.75)92.21 (2.85)34.94 (1.17)32.26 (1.27)
  F1 S2005SYN ×96.22 (2.76)115.10 (3.44)40.93 (0.76)39.09 (0.74)
  Y2005 SYN98598782
  F1 Y2005SYN ×95.57 (3.00)90.96 (2.87)40.36 (0.87)38.59 (0.61)

In the linear model for this design, we found that in the combined data for both lines there is a significant effect of all main effects taken together (F10,139= 3,46, P= 0.0004), as well as significant effects of genotype, gender, and temperature (respectively F3,139= 88.74, F1,139= 22.01, F1,139= 2450.53; all P < 0.0001). The most important source of the variation at 28°C is clear: the longevity of both D. santomea males and females is significantly lower than that of their D. yakuba counterparts in both lines, while hybrids live significantly longer than both pure species.

The difference in longevity between D. santomea and D. yakuba at 28°C does not reflect an inherent, temperature-independent difference, for (with the exception of one significant difference in the opposite direction between females of D. santomea STO.4 females and D. yakuba Taï 18 females) the species showed no significant difference at 21°C. The significant heterogeneity among genotypes at this lower temperature is due to the greater longevity of both hybrid genotypes. The difference in longevity between the two species at 28°C is in the direction predicted by the species’ distributions, for D. yakuba is much more likely to experience extreme high temperatures than is D. santomea.

The greater longevity of F1 hybrids than of pure species at both temperatures is surprising. This cannot be attributed to simple inbreeding of the pure-species strains, for at least in the 2005SYN comparisons, both strains were outbred mixtures of isofemale lines. It is notable that F1 male hybrids with D. santomea mothers had the highest longevity of all genotypes at 21°C, for this is the only genotype seen in the high-altitude “anomalous” hybrid zone. Nevertheless, this increased longevity seems unlikely to explain this hybrid zone, for these male hybrids live only about 20–40% longer than female hybrids from the same mother or than male hybrids from the reciprocal cross, genotypes completely absent from that hybrid zone.

The species difference in longevity, which holds for both strains, is clearly genetic, but can we discern anything of its genetic basis from the hybrids? First, hybrid females from the two reciprocal crosses are genetically identical, differing only in the source of their cytoplasm, which they get from their mother. As expected, save for one comparison they do not differ significantly in longevity (Table 1). The F1 hybrid males from reciprocal hybrid crosses differ genetically only in their X chromosomes, Y chromosomes, and cytoplasm. And in both comparisons of these genotypes at 21°C and one of the comparisons at 28°C, they differ significantly in the same way: males with the D. santomea cytoplasm and X chromosome live significantly longer than males from the reciprocal cross. This suggests that the X chromosome and/or cytoplasm carry genes affecting species-specific longevity. Little can be made of this, however, given that D. santomea do not consistently live longer than D. yakuba at 21°C, and live significantly less long at 28°C.


Thermal preference was estimated by allowing flies to distribute themselves along a thermal gradient in a “thermocline” apparatus, and then scoring their position in this gradient after a fixed time. The design of our thermocline was based on that of Sayeed and Benzer (1996). Figure 8 shows the apparatus, a rectangular plexiglas chamber with dimensions 12 cm × 45 cm × 1 cm, having an aluminum floor. The inside of the apparatus can be closed off into seven isolated chambers, each 10.5 × 6 × 1 cm, by pushing a rod connected to six plexiglas partitions. The removable plexiglas lid on the apparatus has a small hole drilled above the center of each segment for inserting a thermocouple probe to take the temperature of the aluminum floor. To keep flies off the ceiling and sides of the apparatus, these sections were lightly dusted with the repellant quinine sulfate (Sayeed and Benzer 1996). The entire apparatus was cleaned with ethanol and allowed to dry between each run. Between runs the apparatus was rotated 180 degrees so that temperature would not be conflated with directional cues such as light.

Figure 8.

Thermocline used for estimating temperature preferences of adults. Left end is over hot plate, right end is over cold plate.

Before flies were introduced into the gradient, one end was placed on a thermoelectric cold plate and the other on a temperature-controlled hot plate. After 20 min a stable equilibrium temperature was established that ran from 18°C at the cold end to 30°C at the hot end (temperatures were measured at the center of the end chambers using a digital thermometer with a with an accuracy of 0.1°C). These temperatures are likely to be encountered by flies in the field, although the lower temperature is probably experienced only by D. santomea at high altitudes (see below). It is possible that a fly's final position in the gradient reflects its response to humidity rather than temperature differences (e.g., higher temperature = lower humidity). We could not measure humidity at various points inside the apparatus without disturbing any such gradient. However, because temperature and humidity are correlated in the areas in which these flies live (that is, on a gradient from exposed, open habitat to cool, moist mist forest), we could still test whether position in the gradient correlated with our expectations under an “adaptive” distribution in nature.

Temperature preferences of pure species and reciprocal hybrids were tested on the same two groups of strains described above (Taï 18/STO.4 and YSYN2005/SSYN2005), with the two pure-species strains tested along with their reciprocal F1 hybrids. Males and females were tested separately. Each run used 60 four-day old virgin flies of a single sex, reared at 24°C. This sample included 15 flies of each of the two species and 15 flies from each of the two reciprocal F1 crosses between them. To distinguish F1 hybrids from the pure species, the two hybrid genotypes were placed on media dyed with vegetable food coloring 24 h before testing; we used blue dye for one genotype and red for the other, and colors were alternated between genotypes. Ingestion of the food by the flies leads their bellies to take on the color of the dye (Wu et al. 1995). Preliminary experiments (results not shown) confirmed that hybrids placed on dyed food did not differ in thermal preference from hybrids reared on undyed food.

To measure preference, flies were introduced into the middle segment of the apparatus under light CO2 anesthetization with the sliding doors opened so that flies could distribute themselves along the gradient. After 30 min, the partitions were quickly closed, isolating each fly within its chosen area. Flies were then anesthetized, removed from the apparatus, and their species and/or hybrid genotype was determined by inspection under a microscope. The temperature of each segment was taken at both the beginning and end of each run, and every fly in a chamber was assigned the average temperature of that chamber.

For the santomea STO.4/D. yakuba Taï 18 comparison, 18 runs were done for each sex, and for the 2005SYN D. santomea/D. yakuba comparison 13 runs were done for each sex.

To analyze temperature preference, we analyzed the data using a mixed linear model with three fixed effects: line nested within genotype, genotype, gender, and the interaction between them. Runs were considered a random factor. Table 2 shows the temperature preference, sample sizes, and statistical significance (using the Tukey–Kramer HSD test) for each comparison between genotypes (see Supporting Table S1 for the full statistical analysis). The results were remarkably similar between males and females, which did not differ in preference (F1,3667= 0.1833, P= 0.698), while significant heterogeneity in preference was detected among genotypes (F3,3667= 176.29, P < 0.0001) and line nested within genotype (F1,3667= 11.18, P= 0.0008). In all cases, D. santomea preferred temperatures slightly below 23°C, whereas D. yakuba preferred temperatures between 26°C and 27°C. This 3–4°C difference in preference between the species is highly significant in all comparisons, and is in the direction expected if species prefer the temperatures under which they normally live, because D. santomea lives in cooler mist forests, and D. yakuba in warmer and lower semi-open habitat on São Tomé and the African mainland. This interpretation presumes, of course, that species has evolved preferences for those habitats in which they are most fit, and that each of these species is fitter in its respective environment—a supposition for which we lack evidence. Field observations and manipulations will be required to determine whether this behavioral difference is indeed adaptive.

Table 2.  Mean temperature preference (in °C) of genotypes. Standard errors are given in parentheses, and sample sizes are shown below mean preference. Crosses are given with the female parent first (i.e., S×Y are the offspring of crosses between D. santomea females and D. yakuba males). Letters show the results of the Tukey–Kramer HSD test for comparing the four cells in each column (cells with identical letters are not significantly different). “2005SYN yak/san” refers to both the D. yakuba and D. santomea synthetic strains collected in 2005.
STO.4/Tai 182005SYNyak/sanSTO.4/Tai 182005SYNyak/san
D. santomea22.85 (0.25)22.96 (0.21)22.86 (0.26)22.94 (0.22)
D. yakuba26.24 (0.25)26.31 (0.21)26.62 (0.25)26.50 (0.23)
F1 (S × Y)23.11 (0.25)24.01 (0.22)23.44 (0.26)24.17 (0.23)
F1 (Y × S)22.76 (0.24)24.29 (0.22)23.36 (0.25)23.42 (0.23)

In all eight tests, the temperature preference of F1 hybrids was intermediate to those of the pure species, but was generally closer to D. santomea than D. yakuba. Hybrid preferences were statistically indistinguishable from those of D. santomea in six of the eight Tukey–Kramer tests. In none of the eight comparisons did the two reciprocal hybrids genotypes differ in temperature preference, suggesting no effect of the X or Y chromosome (males) or cytoplasm (females) on the interspecific difference in this behavior. Thus, the intermediacy of F1 hybrids indicates that the difference between the species in temperature preference resides largely on the autosomes.

The intermediate preferences of F1 hybrids are notable for one reason: F1 males with the D. santomea X chromosome are the only individuals found at very high altitudes (Llopart et al. 2005a). We expect that if this anomaly were due to behavioral preference of that genotype, it would show the strongest affinity for low temperature. This was not the case: these males were intermediate in preference to both parental species.


Our work shows a consistent pattern for both physiological tolerance and behavior: D. yakuba deals better than D. santomea with high temperature. Their temperature-related differentiation involves five aspects of fitness:

  • 1Larval viability. In contrast to D. yakuba, the viability of D. santomea larvae is effectively zero at 28°C.
  • 2Egg hatchability. Likewise, the egg hatchability of D. yakuba is high at all temperatures but that of D. santomea drops 40% at 28°C.
  • 3Adult fertility. Compared to D. yakuba and three outgroup species, the fertility of D. santomea individuals is markedly reduced at 28°C, a reduction far more severe for females than for males.
  • 4Adult longevity. At lower temperatures adults of D. santomea and D. yakuba have similar longevity, but at 28°C the longevity of D. santomea males and females is severely reduced.
  • 5Adult temperature preference. Given a choice of temperatures between 18°C and 30°C, D. santomea adults consistently choose temperatures 3–4°C cooler than do D. yakuba adults.

Because we tested different strains of each species, with results that were remarkably consistent, these appear to be real, genetically based differences between the species. Moreover, they are in the direction expected based on the known ecology of these species. Drosophila santomea, limited to higher-altitude rain and mist forest, is less tolerant to higher temperature and prefers lower temperatures than does the lower-elevation, open-habitat species D. yakuba. The temperatures we used also appear to be biologically realistic—that is, temperatures encountered by the species in nature. Temperatures routinely exceed 29°C at lower altitudes on São Tomé except during the dry season (June-September), but temperatures at altitudes above 700 m can drop to well below 10°C in the evening.

Could our results be an artifact of keeping fly strains in the laboratory for different periods of time? Strains adapting to laboratory conditions may undergo either novel selection or relaxed selection, both of which could alter stress resistance manifested in the wild. Hoffman et al. (2001), for example, found strong changes in desiccation and starvation resistance in D. melanogaster over 4 years of laboratory adaptation. On the other hand, Krebs et al. (2001) showed no changes in response to heat shock in wild populations of D. melanogaster cultured for 50 generations (about 2 years) in the laboratory.

Our D. santomea strains were collected between 1998 and 2005, the D. yakuba strains in 1967 (one strain) and between 1981 and 2005. The data, however, suggest that our major conclusions are not affected by adaptation in the laboratory. This is because the variation among lines within a species is small compared to the differences between species (see Figs. 1 and 4, and Supporting Figs. S1 and S2), regardless of when strains were collected. For example, egg hatchability of D. yakuba lines at 28°C ranged from 0.924 to 0.978, and in D. santomea from 0.547 to 0.786. Likewise, larval survival at 28°C ranged from 0.513 to 0.632 in D. yakuba lines, and from 0.000 to 0.229 in D. santomea. Temperature preference was nearly identical in two lines of D. santomea caught 7 years apart, and in two lines of D. yakuba collected 24 years apart (Table 2). Finally, adaptation to identical laboratory conditions should make the species more similar, erasing any differences that exist in the wild. We are confident, then, that we have uncovered real physiological differences between the species. Nevertheless, it is possible that our estimates of temperature-dependent sterility in D. simulans, D. tessieri, and D. mauritiana are not representative of wild strains because we used only one line of each of these species.

It should be noted that, with the exception of adult preference, these “adaptive” differences are in one direction only: that is, D. santomea has lower fitness than D. yakuba at high temperature, but the situation is not reversed at low temperature. Indeed, at 15°C D. santomea larvae survive less well than those of D. yakuba, and the hatchability of D. santomea eggs is lower than that of D. yakuba. In other words, there is no evidence for trade-offs of fitness with respect to temperature, trade-offs that would make each species best adapted to its thermal environment. This result is not unique. In tests of different strains in other species, one sometimes sees trade-offs in fitness (e.g., Hoffmann et al. 2002), so that populations that are most heat resistant are least cold resistant and vice versa. But this is not always the case: some populations are uniformly better at all temperatures (e.g., Hallas et al. 2002; Santos 2007). Selection, then, may favor broader thermal niches in some species, perhaps those that occupy a variety of habitats and so experience temperatures at both extremes (e.g., D. yakuba in our study).

Whether these interspecific differences are adaptive or not, they can clearly lead to reproductive isolation if the fitness differences we observe in the laboratory act in the same way in nature. The difference in temperature preference alone might be responsible for the altitudinal/spatial segregation of the species on São Tomé, and the lowered fitness of D. santomea at high temperature will lead them to reproduce more poorly than D. yakuba in the latter's preferred open, higher-temperature habitat. This produces a form of ecological isolation that has been called “immigrant inviability” (Nosil et al. 2005).

Because intrinsic “adaptations” to temperature appear to be one way, with D. yakuba being fitter at high temperatures and not generally less fit at low temperatures, we can ask what keeps these species altitudinally segregated. Because D. santomea is less fit at low altitudes, but D. yakuba is not less fit at higher altitudes, why does D. yakuba not expand its range upwards, displacing D. santomea? One reason may be the difference in temperature preference of the adults, which may lead them to seek out different habitats. But this begs the question of why D. yakuba prefers higher temperature than D. santomea although it is generally fitter than D. santomea at all temperatures. We strongly suspect that there are ecologically relevant differences in fitness between the two species that we have not measured, such as altitudinally segregated breeding or feeding sites. (Despite extensive effort, we have not been able to find natural breeding sites for either species, although D. yakuba does breed on fruits in local plantations).

Alternatively, D. yakuba may indeed be expanding its range, ultimately to displace D. santomea. Island endemics are, of course, often poorly adapted to resist new competitors, and other tests of competitive ability (J. Coyne, unpubl. data) show that D. yakuba outcompetes D. santomea under all laboratory conditions.

Our tests of temperature tolerance and preference have failed to uncover an explanation for the anomalous hybrid zone: the predominance of hybrid males (F1 hybrids with D. santomea mothers) above 1900 m on the Pico de São Tomé. Although hybrids did have somewhat higher fitness than pure species at low temperature for some fitness components, this difference was neither large enough to explain the presence of only one genotype of hybrids at high altitudes, nor was it consistently in favor of male hybrids having a D. santomea mother. Further, we see no evidence for hybrid adults preferring lower temperatures; indeed, they prefer temperatures intermediate to those of parental species. Further tests of geotaxis (data not shown) also show no tendency of hybrids to seek higher ground than the pure species. The existence of this anomalous hybrid zone remains a mystery.

The genetic analyses in this study were crude, limited to discerning, by examining reciprocal F1 hybrids and backcrosses, whether the sex chromosomes, cytoplasm, and autosomes (as a group) affected the trait differences. In general, we found a dearth of sex chromosome effects, with most of the differences in thermal tolerance attributable to cytoplasmic effects or autosomal genes. For larval viability, cytoplasmic effects and autosomes affect larval viability in the expected direction, with D. yakuba autosomes and cytoplasm conferring higher viability at higher temperatures. The results are similar for egg hatchability at the stress temperature of 28°C. Finally, there was no effect of cytoplasm or sex chromosomes on the interspecific difference in temperature preference; the intermediacy of hybrid males and females for this trait suggests that genes affecting its difference reside on the autosomes.

It is not surprising that cytoplasm affects differences in egg hatchability and larval survival, for early developmental stages of fertilized eggs and first-instar larvae are known to be affected by the mother's genotype. Two possibilities are that the cytoplasmic effects we see result from mitochondrial genes or from infection with the endosymbiontic bacterium Wolbachia carried by many species of Drosophila. In some insect species, Wolbachia produces temperature-dependent sterility (Hoffmann and Turelli 1997; Werren 1997). Neither of these possibilities, however, seem likely. The mtDNAs of D. yakuba and D. santomea are virtually identical, suggesting a recent introgression event from the former into the latter (Llopart et al. 2005b), and DNA analysis of two of the strains showing cytoplasmic effects in crosses (D. yakuba Täi 18 and D. santomea STO.4) show that they do not carry Wolbachia (M. Noor, pers. comm.). Nevertheless, whatever factor is responsible for these effects is obviously heritable, because it persists from the F1 hybrids into the backcross females.

How do our results comport with other studies of thermotolerance in Drosophila?Hoffman et al. (2003) review these studies, involving both heat and cold stress. Most work has estimated survivorship of heat or cold “shocks” of adults transferred rapidly from ambient to extreme temperatures, but other studies test survivorship of acclimated adult flies, as well as other life stages. In general, the studies show adaptive differences within species: populations from hotter areas are more resistant to heat and those from colder areas more resistant to cold. This has been shown for both clinal variation across large areas (e.g., Coyne et al. 1983; Hoffmann et al. 2002), over altitudinal transects (Dahlgaard et al. 2001), and in different microenvironments in the same area (Nevo et al. 1998). Similar adaptive differences among strains have also been found for temperature-dependent sterility (Rohmer et al. 2004). Nevertheless, the results are not uniform: “adaptive patterns” are often not found, or results are inconsistent in different tests (e.g., Mitrovski and Hoffman 2001).

There are many fewer papers comparing the thermal tolerance of different species, and many of these are not sister species, or even closely related (e.g., D. melanogaster vs. D. serrata). And again the results are mixed; sometimes adaptive differences are seen between species that have different geographic distributions (e.g., Gibert et al. 2001), while sometimes there are no obvious adaptive patterns (Stratman and Markow 1998).

Only one paper has examined thermal tolerance in several species of the D. melanogaster subgroup. Testing six of the nine species in this group (not including D. santomea), Stanley et al. (1980) found that D. yakuba and D. erecta were most sensitive to cold, heat, and desiccation stress for both survivorship and male sterility. The cosmopolitan species D. melanogaster was superior in all fitness measures, followed by the cosmopolitan D. simulans, implying that the evolution of weedy species may involve broadening their range of temperature tolerance. As in other species (see above), there was no evidence for trade-offs in performance under extreme heat versus cold.

Although the “one-way” differences in thermal tolerance between D. santomea and D. yakuba are not obviously adaptive, our observation of interspecific differences in adult behavior may well be. The only comparable situation in Drosophila is Nevo et al.'s (1998) observation that females of both D. simulans and D. melanogaster from the warmer south-facing slope of Evolution Canyon in Israel preferred to oviposit at higher temperatures than did females from the cooler north-facing slope. Although questions remain about the adaptive significance of our species’ differences in thermal tolerance—questions that can be resolved only with a much better understanding of their ecology—these differences clearly impose an “extrinsic” form of reproductive isolation in these species, adding to the many isolating barriers already known to separate D. yakuba and D. santomea.

Obviously, there is far more room for work on physiological differences and reproductive isolation between these species, one of the very few pair of Drosophila species that display a hybrid zone in nature. And we clearly need much more information on the ecology of these species on São Tomé. We need to know what temperatures are encountered by both larvae and adults in their natural habitats, and the variation among species in the microhabitats. It is thus best to regard our results as preliminary—as a guide to further ecological studies of flies on São Tomé. Nevertheless, the ability of these species to produce fertile hybrids may eventually allow us to pinpoint genes responsible for differences in thermal preference and tolerance that may contribute to reproductive isolation.

Associate Editor: D. Presgraves


We thank J. Gladstone and I. Butler for technical assistance. This work was funded by NIH grant R01GM058260 to JAC.