Altitudinal and seasonal variation in microsatellite allele frequencies of Drosophila buzzatii


Volker Loeschcke, Department of Biological Sciences, Ecology and Genetics, Aarhus Centre for Environmental Stress Research (ACES), Aarhus University, Ny Munkegade, 114-116, DK-8000 Aarhus C, Denmark. Tel.: +45 89 423 268; fax: +45 89 422 722; e-mail:


Variation in climate, particularly temperature, is known to affect the genetic composition of populations. Although there have been many studies of latitudinal variation, comparisons of populations across altitudes or seasons, particularly for animal species, are less common. Here, we study genetic variation (microsatellite markers) in populations of Drosophila buzzatii collected along altitudinal gradients and in different seasons. We found no differences in genetic variation between 2 years or between seasons within years. However, there were numerous cases of significant associations between allele frequencies or expected heterozygosities and altitude, with more than half showing nonlinear relationships. While these associations indicate possible selection and local altitudinal adaptation, direct tests gave strong evidence for selection affecting two loci and weaker evidence for five other loci. Two loci that are located within an inversion (including the one with strongest evidence for selection) show a linear increase in genetic diversity with altitude, likely due to thermal selection. Parallel associations with altitude here and with latitude in Australian populations indicate that selection is operating on chromosomal regions marked by some of the loci.


Climate is an important variable determining the distribution and abundance of species (Andrewartha & Birch, 1960). Climatic variables change with latitude, and concomitantly, habitat conditions and biotic and other environmental variables also may change. Many studies have been conducted that relate variation in morphological and life history traits, as well as in allele or inversion frequencies, to changes in latitude (David et al., 1983; James & Partridge, 1995; van’t Land et al., 1999; Weeks et al., 2002; Laugen et al., 2003). In effect, latitude is here a surrogate for the climatic variables, with temperature often considered a major causal factor (‘thermal selection’). But temperature also has an effect on morphology and life history traits owing to phenotypic plasticity and may shape these traits independently of genetic differences (David et al., 1997; Angilletta et al., 2004).

To establish that observed variation has an adaptive significance, organisms can be collected in the wild at different latitudes and compared in common garden-type experiments in the laboratory, or organisms can be translocated across environments, as for example represented by different latitudes (Iraeta et al., 2008). Such studies have often provided evidence for local adaptation, particularly when similar patterns were observed at independent geographic locations (Hoffmann et al., 2003a,b; Balanya et al., 2006). But it is not always easy to exclude some nonadaptive explanation for the observations (Santos et al., 2005; Gienapp et al., 2008; Merila, 2009). Where the differentiation is along a cline that is highly correlated with an environmental gradient, adaptive differences may be more likely, but again the cline may have resulted from nonadaptive demographic processes such as migration between two populations that were originally genetically different and isolated. However, an observation of parallel independent clines for the same trait greatly strengthens the argument for natural selection affecting variation in the trait. That is, consistency of the independent clines suggests responses to similar selection pressures. Early evidence for selection affecting allozyme frequencies came from the finding of complementary latitudinal clines in Asia, Australia and North America for allele frequencies at the alcohol dehydrogenase and glycerol-3-phosphate dehydrogenase loci in D. melanogaster (Oakeshott et al., 1982). Furthermore, latitudinal clines for inversion frequencies in Drosophila species are often associated with climatic variables and show similar patterns in different continents (Krimbas & Powell, 1992).

Temperature also changes with altitude and hence may be a significant factor affecting changes in the genetic composition of altitudinal populations and local adaptation. Although differences in temperature may be significant for both latitudinal and altitudinal variations, other abiotic and biotic variables may be as important, or even more important, and care is needed in making any adaptive hypotheses. Where altitude is the primary variable of interest, but the populations are spread over a wide geographical area, altitude and latitude (or geographical location) are confounded (Sørensen et al., 2005; Norry et al., 2006; Sambucetti et al., 2006), although their effects may be partitioned to some extent statistically (Folguera et al., 2008). Alternately, altitudinal effects may be studied using paired low- and high-altitude populations from a range of latitudes (Collinge et al., 2006). In contrast, where populations are from various altitudes on a single mountain, geographical variation is minimized, but with potentially higher gene flow.

Whatever method is used, however, parallel latitudinal and altitudinal clines provide strong evidence for selection. Australian D. buzzatii populations show significant latitudinal variation in thorax and wing size traits (Loeschcke et al., 2000) and in a number of stress-related traits (Sarup et al., 2006). In Argentinian populations, altitudinal variation has been found for some stress resistance traits (Sørensen et al., 2001, 2005), thorax length, wing length and oviposition activity rhythms (Dahlgaard et al., 2001), developmental time (Folguera et al., 2008; Mensch et al., 2010), longevity, senescence rate and early fecundity (Norry et al., 2006), and on La Gomera (Canary Islands) for pupal Hsp70 expression and stress resistance traits (Sarup et al., 2009). In these studies, latitudinal and altitudinal clines are concordant for body size traits, as are some of the stress-related traits – knockdown resistance, duration of sterility at 25 °C and Hsp70 expression in immature stages. In addition, inversion frequencies in D. buzzatii populations in Argentina show similar latitudinal and altitudinal clines (Hasson et al., 1995). The altitudinal studies of stress resistance traits in Argentina (Sørensen et al., 2005) used populations at different altitudes but spread over a wide geographical area, whereas the La Gomera study (Sarup et al., 2009) used ones on a single mountain side. Of the traits that were in common between the two studies, more showed significant population differentiation on La Gomera, even though gene flow among the populations would be expected to be greater for these populations. Furthermore, only two of the common traits (starvation resistance and Hsp70 expression) had parallel linear altitudinal trends. In the Argentina populations, the relative genetic isolation may have allowed differential selection owing to factors related to geographical location (e.g. day length or species composition) to mask or reduce effects of altitude.

Little is known about the genetic basis of these adaptive latitudinal and altitudinal clines, although some relationships with thermal selection have been identified. McKechnie et al. (2010) have shown that polymorphism in the Dca (Drosophila cold acclimation) gene in D. melanogaster controls a substantial proportion of the latitudinal cline in wing size variation, whereas altitudinal differentiation in allele frequencies at the PGI locus of Lycaena tityrus is likely due to thermal selection affecting PGI or a closely linked gene (Karl et al., 2009).

If altitudinal populations differ because of adaptation to the different local conditions along the gradient (Bubliy & Loeschcke, 2005; Sarup et al., 2009), and if local population sizes are reduced in the winter months, then we expect altitudinal patterns of genetic variation in autumn (developed during the generations in the summer months) to be significant and similar across years. In spring collections, these altitudinal patterns would be lost, or at least less obvious and not consistent across years, owing to genetic drift resulting from the smaller effective population size over the winter generations. However, if effective population size in the winter months is not strongly reduced, and if selection is tracking seasonal variation, one should expect that populations from the same season in different years would be more alike genetically than populations from different seasons within years.

Here, we study whether populations collected along altitudinal gradients on two of the Canary Islands, La Gomera and Tenerife, show altitudinal variation in microsatellite allele frequencies, and whether this variation is similar on the different islands. On one of the islands (La Gomera), collections were made in two seasons (spring and autumn) in each of 2 years, and we also ask whether there is seasonal variation in allele frequencies. There were no differences between the 2 years or between seasons within years, but significant altitudinal patterns of genetic variation commonly were nonlinear. Differences between the two islands are likely due to genetic drift.

Materials and methods

Origin of flies

Flies were collected in the Canary Islands. On La Gomera, we used six altitudinal collection sites, and four temporal collections were made. Collections were made in late March (spring) 2003, mid to late November (autumn) 2003, late March – early April (spring) 2004 and late November (autumn) 2004. Locations, elevations and descriptions of the five lowest sites are given in Bubliy & Loeschcke (2005). At the highest site (locality 6, 956 masl, 28°08.3′N, 17°18.34′W), collections were made only in 2004. No flies were collected at the 620-m site in autumn 2003. For the Canary Islands, the mean annual temperature lapse rate (decrease per 1000 m height for an atmospheric variable) is 4.22 ± 0.48 °C, with a lower rate in the warmer months and a higher in the colder months (Meyer, 1992). Temperature differences among localities varied most at night, with minimum temperatures usually being 6–8 °C different between low- and high-altitude populations. The two lowland localities 1 and 2 as well as the two highland localities 5 and 6 had rather similar extreme and average temperatures, with the other localities being somewhat intermediate with similar steps of reduced temperature with increasing altitude. On Tenerife, one collection was made in late October (autumn) 2004 at nine altitudinal sites (for collection locations, altitudes and site descriptions, see Sarup & Loeschcke, 2010).

Collected flies were either put immediately into absolute ethanol or kept alive in vials with Carolina medium until return to the laboratory, where they were frozen at −80 °C. In total, 587 flies were genotyped from the 21 collections on La Gomera and 265 from the nine sites on Tenerife, with a range per collection from 11 to 36 and an average of 28.5.

DNA extraction and microsatellite analysis

DNA extraction and genotyping methods were performed as given in Frydenberg et al. (2002) for 10 microsatellites – (annealing temperature in parentheses): Db052 (50 °C), Db087 (45 °C), Db122 (55 °C), Db142 (55 °C), Db223 (60 °C), Db225 (58 °C), Db290 (45 °C), Db411 (60 °C), Db493 (60 °C), Db681 (40 °C) and in Barker et al. (2009) for five additional ones: Db003 (60 °C), Db013 (60 °C), Db034 (60 °C), Db090 (52 °C) and Db109 (60 °C).

Allele frequency and heterozygosity

Genotype and allele frequencies were estimated using genepop Version 3.4 (Raymond & Rousset, 1995) (genepop data file for all samples available as Table S1), and alleles per locus and observed and expected heterozygosity (gene diversity) using geneclass2 (Piry et al., 2004). Tests for deviations from Hardy–Weinberg equilibrium were performed using the exact tests of genepop (default values for the Markov chain method). Significance levels for each test were determined by applying to the probability estimates calculated by genepop, the sequential Bonferroni procedure (Hochberg, 1988; Lessios, 1992) over loci within each population.

Population differentiation

As only one locus in one population showed a significant deviation from Hardy–Weinberg equilibrium (see Results), allelic differentiation among populations was tested using genepop, for overall and pairwise differentiation (default values for the Markov chain method). The sequential Bonferroni procedure was applied over population pairs for the latter in determining significance levels. F-statistics (Weir & Cockerham, 1984) and their significance were determined using fstat Version 2.9.3 (Goudet, 2002), not assuming Hardy–Weinberg equilibrium and with 5000 iterations, and the sequential Bonferroni procedure was applied over loci to determine significance levels. In addition, we used the hierfstat package (Goudet, 2005; de Meeus & Goudet, 2007) to estimate hierarchical FST statistics for the La Gomera data (years, seasons, altitudes) and for the La Gomera and Tenerife data for the autumn 2004 collections (islands, altitudes).

Regressions (linear and quadratic) of the mean number of alleles, mean observed (Ho) and mean expected heterozygosity (He) at each altitude were calculated separately for each season on La Gomera and for Tenerife, and for each locus, allele frequencies and expected heterozygosity on altitude using the SPSS Inc. (Chicago, IL, USA) Statistics 17 package. While expected heterozygosity will change as allele frequencies change, we consider both as there can be significant change in allele frequencies without significant change in expected heterozygosity, or vice versa. For diallelic loci, frequencies of the more common allele only were used. For loci with three or more alleles, any allele whose average frequency was about 0.1 or less was not considered. Regressions were accepted as linear or quadratic depending on which explained a higher proportion of the variance (R2 value), or as linear if the R2 values were not significantly different.

We used the method for detection of selection in a hierarchically structured population that is implemented in the software package arlequin3.5 (Excoffier et al., 2005). This test for selection is based on FST, and following Excoffier et al. (2009), we assume the SMM model and compute FST as ρST. An obvious structure would be to define the altitudinal populations as demes within each of the five groups – the four collections on La Gomera and the one on Tenerife. But collections on La Gomera were seasonal, approximately six months apart, and thus structured temporally, and not spatially as specified by the model. Two of the collections, however, were made at essentially the same time (autumn 2004), one on La Gomera and one on Tenerife. These data were defined as two groups, and we ran 50,000 simulations, with 100 demes per group and 30 groups. We carried out the same analysis of Australian populations from Barker et al. (2009), where we have data for the same 15 microsatellites in nine populations. As these populations are largely significantly different from each other (pairwise FST), different combinations of five or six groups were run.


All loci were polymorphic in all populations, except for La Gomera – Db052 in one population (autumn 2004, 886 masl – designated Au04_886), Db122 in a different population (Au04_956) and Db223 in three populations (Au03_886, Au04_20, Sp04_620) and for one Tenerife (935) population. Allele frequencies for all populations are available in Table S2 (La Gomera) and Table S3 (Tenerife) and number of alleles at each locus in each population in Table S4. Means of measures of genetic variability for each population (Table 1) are generally similar for the two islands, the four seasonal collections on La Gomera and the various altitudes. Nevertheless, the quadratic regression of observed heterozygosity on altitude was significant for the autumn 2003 populations on La Gomera (P < 0.05, R2 = 0.993) and for the Tenerife populations (P = 0.05, R2 = 0.339). As in these two cases, variability tended to be higher at low to intermediate altitudes for all seasons on La Gomera and on Tenerife. Mean measures of variability across loci (Table 2) are not significantly different between the two islands (t-tests, results not shown), but for some loci (e.g. Db087, Db109 and Db681), the number of alleles is much smaller on Tenerife, whereas for Db034 and Db052, some alleles found on Tenerife were not present on La Gomera.

Table 1.   Mean over loci of the number of alleles and of heterozygosity for each population (SD in parentheses).
Population*Number of allelesHeterozygosity
La GomeraObservedExpected
  1. *Populations designated by metres asl.

Spring 2003
 202.800 (0.775)0.379 (0.181)0.423 (0.203)
 2243.133 (1.187)0.469 (0.163)0.477 (0.160)
 3743.067 (0.961)0.398 (0.210)0.448 (0.188)
 6203.200 (1.014)0.456 (0.183)0.460 (0.173)
 8862.867 (0.915)0.452 (0.254)0.435 (0.209)
Autumn 2003
 202.667 (0.724)0.440 (0.188)0.460 (0.177)
 2243.067 (1.033)0.440 (0.157)0.477 (0.167)
 3743.000 (1.000)0.439 (0.183)0.442 (0.176)
 8862.867 (0.915)0.397 (0.177)0.432 (0.192)
Spring 2004
 202.800 (0.775)0.443 (0.209)0.451 (0.194)
 2243.067 (1.100)0.451 (0.194)0.449 (0.186)
 3743.133 (1.125)0.426 (0.178)0.458 (0.191)
 6202.867 (0.990)0.430 (0.195)0.458 (0.199)
 8862.867 (0.834)0.493 (0.204)0.464 (0.183)
 9563.000 (1.000)0.445 (0.223)0.457 (0.207)
Autumn 2004
 202.800 (1.014)0.465 (0.218)0.454 (0.194)
 2243.067 (1.163)0.437 (0.194)0.463 (0.196)
 3743.067 (1.100)0.434 (0.181)0.445 (0.196)
 6202.667 (0.617)0.470 (0.220)0.485 (0.185)
 8862.800 (1.014)0.448 (0.269)0.435 (0.218)
 9562.733 (0.884)0.440 (0.243)0.419 (0.218)
 1092.800 (0.676)0.471 (0.197)0.479 (0.191)
 1332.733 (0.704)0.419 (0.181)0.461 (0.199)
 3183.200 (0.941)0.482 (0.183)0.484 (0.169)
 5552.933 (0.594)0.502 (0.162)0.492 (0.167)
 8252.867 (0.640)0.489 (0.197)0.475 (0.191)
 9352.733 (0.799)0.493 (0.180)0.484 (0.188)
 10682.800 (0.676)0.467 (0.172)0.477 (0.181)
 11422.800 (0.676)0.453 (0.182)0.460 (0.188)
 12712.733 (0.594)0.469 (0.180)0.488 (0.162)
Table 2.   Total observed number of alleles and mean number of alleles, and mean observed and expected heterozygosity per locus on La Gomera and Tenerife.
LocusLa GomeraTenerife
Number of allelesHeterozygosityNumber of allelesHeterozygosity

Significant deviations from Hardy–Weinberg equilibrium were observed for various loci in different populations, but only Db034 in the Sp03_374 population on La Gomera was significant after Bonferroni correction.

Allelic differentiation among the four seasonal collections on La Gomera (pooling altitudinal populations within each) was not significant over all loci, but Db223 showed significant differences (P < 0.05). Within seasons, there were a number of significant allelic differences among altitudinal populations, but only the following were significant after Bonferroni correction: spring 2003 – Db290 (P < 0.05); autumn 2003 – Db493 (P < 0.05); spring 2004 – Db034 (P < 0.01); autumn 2004 – Db052 (P < 0.01), Db122 (P < 0.05) and Db142 (P < 0.01). F-statistics analyses of population differentiation within seasons gave the same results.

Hierarchical FST analysis of the La Gomera data (Table 3) found that over all loci, between years and among seasons within years, were not significant, whereas among altitudes within seasons was close to significance (P = 0.066). For individual loci at this last level, three (Db034, Db052 and Db142) were significant or close to significance.

Table 3.   Hierarchical FST analysis of the 21 populations (seasons by altitudes) of Drosophila buzzatii from La Gomera (significance from permutation tests in the hierfstat package).
LocusBetween yearsBetween seasons/yearsAmong altitudes/seasons
  1. P < 0.1, **P < 0.01.


Differentiation among the Tenerife altitudinal populations was greater than on La Gomera, even among the six lower populations in the same range of altitudes as those on La Gomera. Overall allelic differentiation was highly significant (P < 0.001) and significant for five loci (P values in parentheses): Db034 (P < 0.001), Db087 (P < 0.01), Db122 (P < 0.05), Db290 (P < 0.05) and Db411 (P < 0.01). F-statistics results (Table 4) are essentially the same, but with Db090 also significant (P < 0.05). Both allelic estimates and Θ indicate pairwise differentiation (Table 5), with the three lowest and two highest altitudes showing differentiation and the intermediate altitudes with little differentiation.

Table 4. F-statistics analysis of the nine altitudinal populations of Drosophila buzzatii from Tenerife (significance from permutation tests in the fstat program).
Locus†F (FIT)Θ (FST)F (FIS)
  1. †Standard deviation in parentheses – estimate from jackknife over populations.

  2. ‡Standard deviation in parentheses – estimate from jackknife over loci.

  3. *P < 0.05, **P < 0.01, ***P < 0.001.

Db003−0.001 (0.054)0.015 (0.012)−0.016 (0.052)
Db013−0.032 (0.035)0.008 (0.010)−0.041 (0.036)
Db034−0.027 (0.067)0.049 (0.040)***−0.022 (0.068)
Db0520.013 (0.044)0.002 (0.008)0.011 (0.040)
Db087−0.005 (0.029)0.015 (0.009)*−0.019 (0.029)
Db090−0.011 (0.069)0.001 (0.007)*−0.012 (0.069)
Db109−0.021 (0.041)−0.000 (0.010)−0.021 (0.035)
Db1220.049 (0.048)0.019 (0.012)*0.031 (0.048)
Db1420.033 (0.037)−0.006 (0.003)0.039 (0.038)
Db223−0.025 (0.005)−0.007 (0.005)−0.018 (0.006)
Db2250.079 (0.034)0.001 (0.012)0.078 (0.041)
Db290−0.010 (0.033)0.007 (0.004)*−0.018 (0.035)
Db4110.059 (0.037)0.010 (0.009)**0.050 (0.034)
Db493−0.030 (0.051)0.003 (0.006)−0.033 (0.048)
Db6810.102 (0.039)*0.013 (0.008)0.090 (0.042)*
Mean‡0.021 (0.012)0.009 (0.004)***0.012 (0.013)
Table 5.   Pairwise population differentiation (Θ) among the altitudinal populations on Tenerife, with significance from the permutation tests in the fstat program and adjusted for multiple comparisons.
  1. *Significant Θ estimates –P < 0.05.

  2. Significant pairwise allelic differentiation is also shown – indicated by Θ values in bold (P < 0.001), italic (P < 0.01) and underlined (P < 0.05).

133 0.0200.018*0.0130.0100.0130.030*0.026*
318  −0.001−0.0050.0010.0150.0130.004
555   −0.007−0.0070.0080.0100.002
825    −0.0090.0060.0060.006
935     0.0010.001−0.001
1068      0.0080.018*
1142       0.014*

For the La Gomera and Tenerife collections made in autumn 2004, hierarchical FST analysis (Table 6) shows highly significant differences between the islands and among altitudes within islands. Five loci showed significant differences between islands, with three of these also significant among altitudes within islands. However, the latter differences among altitudes within islands are for an average effect over the two islands, and as noted earlier, more loci show significant altitudinal differences for Tenerife alone.

Table 6.   Hierarchical FST analysis of the altitudinal populations of Drosophila buzzatii from La Gomera and Tenerife collected in autumn 2004 (significance from permutation tests in the hierfstat package).
LocusBetween islandsAmong altitudes/islands
  1. *P < 0.05, **P < 0.01, ***P < 0.001.


Regressions on altitude for allele frequencies and expected heterozygosity were significant for a number of loci in each of the four seasonal collections on La Gomera and the one on Tenerife (Table 7). More than half of the significant regressions were curvilinear, with either maximum or minimum values at intermediate altitudes. Six loci had significant regression coefficients in two or more of the five data sets, but the nature of the relationships was not necessarily the same in each, with varying patterns that were linearly increasing or decreasing, or convex or concave quadratic. Nevertheless, over all the analyses the patterns were often similar between years (i.e. samples about 10 generations apart). For the spring collections, five of 21 alleles (0.238) and five of 15 expected heterozygosities (0.333) showed similar altitudinal patterns. For the autumn collections, there were similar patterns for six of 21 alleles (0.286) and four of 15 expected heterozygosities (0.267). Over the five data sets, 16.4% (26/159) of the regression analyses of allele frequencies, and 16.0% (12/75) of those of expected heterozygosities were significant, clearly more than would be expected by chance. Within data sets, the percentage significant ranged from 6.7% to 28.6%. The proportions of variation explained by the regressions (R2) ranged from 0.701 to 1.000 (mean ± SD = 0.932 ± 0.073) for the La Gomera data and from 0.455 to 0.767 (0.617 ± 0.096) for Tenerife.

Table 7.   Regressions of expected heterozygosity and allele frequencies at each locus on altitude, separately for each of the four seasons on La Gomera and for Tenerife (significant regressions only are shown).
La GomeraTenerife
Spring 2003Autumn 2003Spring 2004Autumn 2004Autumn 2004
  1. L, linear; Q, quadratic.

  2. Entries in bold identify loci that were significant in more than one set.

  3. *P < 0.05, **P < 0.01, ***P < 0.001.

 Db003/252** Q   
 Db013/He* Q   
Db034/106* L    
   /He* L   Db034/He* Q
       /He* L
   Db087/165** QDb087/167* Q
 Db087/177* Q    /177* Q   /177* Q
      /He* Q 
Db090/185* Q    
   /He** QDb090/193* L   
Db109/167* L    
  Db223/227** Q  
     /He** Q  
 Db225/He* Q   
 Db290/179* L  Db290/168* Q
       /He* Q
Db411/234* L Db411/234* LDb411/He** Q 
   /242* L    
Db493/165** L    
   /He** L    
Db681/259** L   Db681/243** L
       /247** L
       /263* Q
       /He* Q

The Arlequin test for selection showed seven loci to be potentially under selection – directional selection at Db034 (P = 1 × 10−7) and balancing selection at Db225 (P = 0.009) and at five other loci (Db087, Db109, Db290, Db411 and Db493 – all P < 0.05). For the Australian populations, the different groupings all gave the same results – balancing selection at Db087 and directional selection at Db225 (both P < 0.05).


We used regression analyses and F-statistics to study (i) the extent of altitudinal variation in microsatellite allele and genotype frequencies on two of the Canary Islands (La Gomera and Tenerife), (ii) whether there are clinal patterns of genetic variation, (iii) whether any patterns of variation on the two islands are similar, and (iv) for La Gomera only, whether there are year and seasonal effects.

Contrary to our predictions regarding possible seasonal and year effects on genetic variation, we found no differences between the 2 years or between seasons within years (Table 3). Thus, as far as can be determined from our data, effective population sizes of the altitudinal populations are not significantly different. A longer time series may detect year effects, while seasonal differences on the oceanic La Gomera are likely too limited to cause detectable genetic changes. The significant differences between the islands for five loci and overall (Table 6) are attributed to differences in allelic content and frequencies and are the likely result of founder events or drift since colonization.

Regression analyses show numerous cases of significant association between allele frequencies or expected heterozygosities and altitude. Six loci were significant in two or more of the data sets (Table 7), with three of these indicating seasonal effects: Db087 in both autumn collections on La Gomera and in the Tenerife autumn collection; Db290 in one autumn collection on La Gomera and in the Tenerife autumn collection; and Db411 in the two spring collections on La Gomera. But in addition to these significant effects, the patterns of change with altitude were often consistent within each season. While these consistent patterns may be expected owing to temporal autocorrelations of allele frequencies, conditions for any given season may vary from year to year, so that consistent trends over years for some loci, but not others, are another indicator of possible selection. These regression analyses, however, can only be suggestive of selection and altitudinal adaptation. On the other hand, the Arlequin hierarchical analysis provides direct and strong evidence for selection affecting Db034 and Db225. Of the five other loci that were significant in this test (for balancing selection –P < 0.05), three (Db087, Db290 and Db411) were also significant in the regression analyses. As here, loci identified by the Arlequin analysis as being affected by selection will not necessarily show an altitudinal cline, or vice versa, as different selective forces may be operating. Finally, significant latitudinal and altitudinal variation for the same locus further strengthens the case for that locus being affected by selection and at least marking a chromosomal region with candidate genes for local adaptation. Allele frequencies at Db090, Db142, Db223, Db411, Db493 and Db681 in Australian populations show significant associations with latitude (Barker et al., 2009), and all also show significant associations with altitude in one or more of the Canary Islands data sets. However, although Db034 shows some clinal altitudinal variation and strong evidence for directional selection in the Canary Islands, there is no indication of selective effects in Australian populations. In contrast, although the Arlequin analysis indicated selection affecting Db225 in both the Canary Islands and Australia, the selection mode was balancing in the former and directional in the latter. These differences between the Canary Islands and Australia may reflect differences in the selective forces operating, or differences in background selection following founder effects at colonization of each region (Barker et al., 2009). On the other hand, Db087 was significant for balancing selection in both the Canary Islands and Australia and in two Argentinean populations (Barker et al., 2009).

As altitude increases, temperature variables (mean, maximum and minimum) are expected to decrease linearly (but not necessarily at the same rate), whereas thermal amplitude (daily maximum – daily minimum) will generally increase. Monthly mean temperatures at sea level in the Canary Islands range over about 18–24 °C ( and will average about 5 °C less at our highest collection sites. Thus on the criteria of temperature variables only, the populations will be exposed to less favourable conditions as altitude increases, with expected smaller population sizes and decreased genetic diversity. Yet of the 12 significant regressions of expected heterozygosity on altitude (Table 7), only one (Db493 in Sp03) shows a linear decrease, one (Db223 in Sp04) a significant concave relationship, but with little difference in He among the three highest altitudes, and one (Db411 in Au04) convex but with little difference among the three lowest altitudes. Of the remaining significant regressions, maximum values are found at lower and higher altitudes (concave quadratic – five cases) or at intermediate values (convex quadratic – three cases). Clearly, the thermal environment is not the only factor affecting genetic diversity, and such nonlinear relationships should not be unexpected. For example, direct sun and wind exposure of the feeding and breeding habitat (cactus rots) may well differ among altitudinal sites, owing to differences in aspect and vegetation cover, so that other climatic and biotic variables will not necessarily change in a linear fashion with changes in altitude. About half of the significant altitudinal changes in allelic frequencies also were nonlinear (Table 7). Differentiation among altitudinal populations was much stronger on Tenerife (Table 4), possibly a function of the larger number and spatially wider separation of collection localities extending to higher altitudes.

Some of the significant differentiation among altitudinal populations or the significant regression coefficients for particular allele frequencies or heterozygosity may be attributed to variation in inversion frequencies, as a highly significant difference has been found in the frequencies of the second chromosome inversion arrangements between a highland (2460 m) and a lowland (596 m) population in Argentina (Dahlgaard et al., 2001). Among the loci that show significant variation with altitude, two (Db034 and Db052) are located within the 2j inversion, with one near but not very close to each breakpoint (Barker et al., 2009). The chromosome 4s inversion has been recorded in the Canary Islands (Fontdevila et al., 1981), and its frequency decreases with altitude in Argentina (Hasson et al., 1995). Db142 is located outside and near the distal breakpoint of this inversion (Barker et al., 2009). Thus for these three loci (and particularly for Db034), inversion selection may well be the driving force. Furthermore, it is likely to be primarily thermal selection, as (i) changes in inversion frequencies are related to direct or indirect effects of temperature shifts in three other species of Drosophila (Hoffmann & Rieseberg, 2008), and (ii) the significant associations of He and altitude for Db034 and Db052 were linear and positive, or for Db034 on Tenerife, convex but with He higher at higher altitudes.

Significant altitudinal variation in allele frequencies and expected heterozygosity for some loci indicate possible selection and local adaptation. The nonlinear patterns of genetic variation with altitude, as found here, may be quite common, but the selection forces operating are not known. Further analysis of previous studies of morphological and life history traits, which considered linear clines only, may assist in evaluating the selective forces and local adaptation. Finally, the parallel significant associations with latitude for some of these microsatellite loci provide additional evidence that these loci are marking chromosomal regions that are under selection. On the other hand, the significant differences between the two islands in genotypes at five loci and overall are more likely due to drift.


We are grateful to Kamilla Håkanson for help in the DNA laboratory, to Doth Andersen for help in the fly laboratory and to Deciding Editor Thomas Flatt and two anonymous reviewers for constructive comments on an earlier version of the paper. We thank the Sr. Consejero de Medio Ambiente, Excmo. Cabildo Insular de Tenerife and La Gomera for permission to collect flies. The study was supported by centre and frame grants to VL and by a grant from the Lundbeck Foundation and Carlsbergfondet to PS.