Molecular and quantitative genetic differentiation across Europe in yellow dung flies


Marco Demont, Zoological Museum, University of Zurich, Winterthurerstrasse 190, CH 8057 Zurich, Switzerland.
Tel.: +41 44 635 47 79; fax: +41 44 635 47 80; e-mail:


Relating geographic variation in quantitative traits to underlying population structure is crucial for understanding processes driving population differentiation, isolation and ultimately speciation. Our study represents a comprehensive population genetic survey of the yellow dung fly Scathophaga stercoraria, an important model organism for evolutionary and ecological studies, over a broad geographic scale across Europe (10 populations from the Swiss Alps to Iceland). We simultaneously assessed differentiation in five quantitative traits (body size, development time, growth rate, proportion of diapausing individuals and duration of diapause), to compare differentiation in neutral marker loci (FST) to that of quantitative traits (QST). Despite long distances and uninhabitable areas between sampled populations, population structuring was very low but significant (FST = 0.007, 13 microsatellite markers; FST = 0.012, three allozyme markers; FST = 0.007, markers combined). However, only two populations (Iceland and Sweden) showed significant allelic differentiation to all other populations. We estimated high levels of gene flow [effective number of migrants (Nm) = 6.2], there was no isolation by distance, and no indication of past genetic bottlenecks (i.e. founder events) and associated loss of genetic diversity in any northern or island population. In contrast to the low population structure, quantitative traits were strongly genetically differentiated among populations, following latitudinal clines, suggesting that selection is responsible for life history differentiation in yellow dung flies across Europe.


The partitioning of genetic variation across a species range can be the result of diversifying or directional selection, genetic drift, or some combination of the two. There is continued debate about the relative importance of each for population differentiation (e.g. Turelli et al., 1988; Merilä & Crnokrak, 2001; McKay & Latta, 2002; O’Hara, 2005; Roff & Mousseau, 2005; Whitehead & Crawford, 2006; Leinonen et al., 2008), and beyond dispute, determining what proportion of observed genetic variation can be attributed to selection or neutral processes lies at the heart of evolutionary biology. Direct comparisons of the genetic differentiation of quantitative traits (QST) to that observed at neutral loci (FST) is one very useful approach for evaluating the relative importance of natural selection vs. genetic drift in nature (but see Gockel et al., 2001). Studies comparing these two measures of genetic differentiation have shown that QST typically exceeds FST, suggesting a strong role for diversifying or directional natural selection in shaping genetic heterogeneity (reviewed by Merilä & Crnokrak, 2001; McKay & Latta, 2002; Leinonen et al., 2008). For example, Spitze’s (1993) pioneering comparison of body size and allozyme variation among populations of Daphnia obtusa revealed that natural selection was responsible for diversification of body size among populations. Nevertheless, in several cases the reverse pattern has been detected, with FST values exceeding QST values (Petit et al., 2001; Edmands & Harrison, 2003; Koch et al., 2004).

Quantitative trait variation across latitudinal clines is expected to occur because of responses to numerous correlated environmental variables (e.g. temperature or season length). Indeed, a recent review of latitudinal clines in arthropod body size and development time lists numerous cases where empirical data support this pattern (Blanckenhorn & Demont, 2004). To date very few studies have extended this by examining the relationship between latitude and quantitative traits variation in parallel with neutral molecular population genetic structure. One study of Drosophila serrata along the eastern cost of Australia revealed low levels of neutral genetic differentiation among populations, whereas quantitative trait differentiation was substantial, suggesting, as with D. obtusa, that selection is more important than neutral processes in shaping life history evolution (Magiafoglou et al., 2002). In contrast, work on body size clines of a related species (D. melanogaster) yielded an inconclusive picture because of conflicting estimates of gene flow among populations (Agis & Schlötterer, 2001; Kennington et al., 2003). Neither of these studies measured quantitative trait variation using the FSTQST comparison. In fact, FSTQST comparisons specifically assessing the effects of latitudinal clines are scarce. The few studies that have been completed tend to agree with the majority of FSTQST research to date: genetic drift is not sufficient to explain observed latitudinal patterns and selection is therefore invoked as the causal explanation (Storz, 2002; Palo et al., 2003).

This is not to say that neutral processes have had no influence on shaping latitudinal patterns of genetic structure and gene flow. The climatic changes that occurred during the quaternary period had a tremendous impact on the distribution of species, and consequently the current genetic structure of species’ ranges in temperate regions (e.g. Taberlet et al., 1998; Hewitt, 2000, 2001, 2004). For example, speciation in the Western Palearctic brown frogs was driven by the repeated post-Messinian climatic oscillations and the neutral processes associated with colonization during each interglacial period (Veith et al., 2003). This pattern of post-glacial colonization and population genetic structure has been especially well-studied in Europe: glacial refugia have been detected in Iberia, Italy, the Balkans and the Caucasus (Hewitt, 1999, 2000). Post-glacial expansions from these European refugia were associated with sequential founding events, and a general pattern of decreasing genetic diversity with increasing latitude can be seen in a number of taxa (Hewitt, 1996, 1999; Taberlet et al., 1998). Sequential founding events have also led to strong population genetic differentiation along the south-to-north gradient. A general assumption of decreased genetic variability in the north and strong latitudinal population genetic differentiation would, however, be presumptuous. Hewitt (2000) pointed out that more species need to be examined before one can relate particular species characteristics (e.g. generation time or dispersal ability) to colonization patterns inferred from genetic studies. Furthermore, large scale population genetic surveys of widely distributed species in Europe are still scarce and taxonomically biased towards vertebrates and plants (Palo et al., 2004).

The yellow dung fly (Scathophaga stercoraria: Diptera: Scathophagidae) is commonly used as a model organism for evolutionary, ecological and behavioural studies (e.g. Parker, 1970a; b; Simmons & Parker, 1992; Blanckenhorn, 1997, 1998a; Ward, 2000; Hosken et al., 2001; Martin et al., 2004). Much of the species’ life history is dependent on the availability of fresh cow dung, and possibly because of this dependence on domestic livestock, the species has become extremely widespread: S. stercoraria is common across the north-temperate regions of the New and Old Worlds. It is ubiquitous across central and northern Europe as far north as Spitzbergen, because of a general preference for cooler climes (Ward & Simmons, 1990; Blanckenhorn et al., 2001). During the hottest months of the year, flies are absent from the pastures of the warmer regions of central Europe (Parker, 1970b; Gibbons, 1987), a pattern not repeated at higher latitudes or altitudes (Otronen, 1995; Sigurjónsdóttir & Snorrason, 1995; Blanckenhorn, 1997). Studies of Swiss populations indicate that local adaptation to high altitude conditions is weak and phenotypic plasticity is extensive (Blanckenhorn, 1997, 1998a,b). Estimates of gene flow across the same scale showed no significant population differentiation, a factor presumed to be impeding the development of locally adapted phenotypes (Kraushaar et al., 2002).

If gene flow is substantial and population differentiation of adaptive life history traits is weak at smaller spatial scales, the possibility remains that the pattern may be reversed once larger spatial scales are considered. Adaptive differentiation may be more pronounced along a wide latitudinal gradient as fly summer residence patterns appear to be dictated by temperature regimes, and thermal and diurnal variation across latitudes exceeds that seen along altitudinal gradients in Switzerland. Furthermore, even with good dispersal capabilities, we may also expect to see a more structured pattern of gene flow among populations of yellow dung flies at a broader geographic scale (i.e. increased isolation by distance). The aim of this study was to assess genetic structure of S. stercoraria populations on a European scale, and compare the extent of neutral genetic differentiation (FST) to that exhibited by a number of quantitative life history traits (QST). To do this, we used multiple molecular marker data sets and examined variation of adaptive traits that should be affected strongly by latitudinal climatic variation (body size, development time, growth rate and aspects of diapause induction; Mousseau & Roff, 1989; Schmidt et al., 2005). We also addressed a number of other population genetics questions: Is there any evidence of founder effects and decreased genetic variability at northern latitudes? Is latitudinal differentiation more pronounced than previously measured altitudinal differentiation (Kraushaar et al., 2002)?

Materials and methods

Molecular data

We sampled a total of 223 flies from across Europe for the assessment of neutral genetic differentiation. Sampling locations ranged from the two latitudinal limits of the species’ range (northern edge, Iceland; southern edge, Switzerland). Acronyms for each location, sampling effort, and additional geographic information are provided in Table 1 and Fig. 1. We sampled a selection of field-collected adults (FIN, NL, CH[Z], CH[C] and CH[L]), hatched individuals from field-collected eggs (ICE, GER and UK), or laboratory-reared F1 offspring from field-collected parents (SWE and CH[B]). Samples were frozen at −80 °C and subsequently used for allozyme and microsatellite characterization.

Table 1.   Acronyms, geographical information on sampling sites, sample sizes, genetic diversity measured as mean gene diversity and mean allelic richness (±SE), and number of private alleles (PA) for the 10 European Scathophaga stercoraria populations sampled (arranged from north to south).
AcronymCountry, nearby cityLatitude*Longitude*Altitude* (m)NMean gene diversityMean allelic richnessPA
  1. *Source:

ICEIceland, Reykjavik64.15°N21.95°W 15220.783 ± 0.0320.236 ± 0.0868.92 ± 0.802.24 ± 0.2410
FINFinland, Jyväskylä62.23°N25.73°E 90230.800 ± 0.0430.287 ± 0.0599.81 ± 0.862.80 ± 0.424
SWESweden, Lund55.70°N13.18°E 25230.795 ± 0.0280.329 ± 0.0058.18 ± 0.622.56 ± 0.286
GERGermany, Bielefeld52.03°N8.53°E104180.771 ± 0.0360.220 ± 0.1468.17 ± 0.631.98 ± 0.565
NLNetherlands, Wageningen51.97°N5.67°E  8220.810 ± 0.0279.31 ± 0.947
UKUnited Kingdom, Oxford51.75°N1.25°W 68230.808 ± 0.0250.230 ± 0.1258.96 ± 0.672.22 ± 0.267
CH[Z]Switzerland, Zurich47.37°N8.55°E433240.801 ± 0.0240.201 ± 0.1068.67 ± 0.602.35 ± 0.503
CH[B]Switzerland, Boudevilliers47.03°N6.88°E746220.801 ± 0.0370.155 ± 0.1108.96 ± 0.932.17 ± 0.394
CH[C]Switzerland, Chur46.85°N9.50°E580220.813 ± 0.0340.208 ± 0.0769.14 ± 0.722.48 ± 0.247
CH[L]Switzerland, Lugano46.00°N8.97°E235240.786 ± 0.0370.240 ± 0.0779.13 ± 0.892.65 ± 0.030
Figure 1.

 Geographical distribution of Scathophaga stercoraria populations used for this study (circles). Populations from north to south: ICE, Iceland, Reykjavik; FIN, Finland, Jyväskylä; SWE, Sweden, Lund; GER, Germany, Bielefeld; NL, Netherlands, Wageningen; UK, United Kingdom, Oxford; CH[Z], Switzerland, Zurich; CH[B], Switzerland, Boudevilliers; CH[C], Switzerland, Chur; CH[L], Switzerland, Lugano.

For allozymes, frozen S. stercoraria individuals were ground in 60 μL mercaptoethanol-EDTA-buffer, centrifuged at 9500 g for 10 min at −5 °C and stored until electrophoresis at −80 °C. We used three allozymes previously shown to be informative at a smaller spatial scale, mannose-6-phosphate isomerase (MPI; EC, phosphoglucomutase (PGM; EC and glucose-6-phosphate isomerase (GPI = PGI; EC, Kraushaar et al., 2002). We used PHERO-celTM gels (BIOTEC-Fischer GmBH, Reiskirchen, Germany) and procedures for horizontal cellulose acetate gel electrophoresis as described by Hebert & Beaton (1989) with slight modifications and a Tris Glycine (TG) pH 8.5 running buffer. Stain recipes were also taken from Hebert & Beaton (1989) with two modifications: we increased the dose of MgCl2 and glucose-1-phosphate solution to six drops for PGM staining. Allelic variants were scored using successive numbers, with ‘1’ denoting the allele with the slowest relative mobility. To ensure that scoring among gels was consistent, we ran a diagnostic subsample of previously electrophoresed samples on subsequent gels for direct comparison.

For microsatellites, DNA was extracted from allozyme homogenates using the QIAamp® DNA Mini Kit (Qiagen AG, Hombrechtikon, Switzerland). Thirteen microsatellite loci were amplified using eight previously published primer pairs (Garner et al., 2000) and five newly developed pairs (Table 2). Polymerase chain reactions were carried out using the annealing parameters described in Garner et al. (2000) and in Table 2. Thermocycles followed those described in Garner et al. (2000) with the following modifications: overall reaction volume was reduced to 10 μL, the initial denaturation step was increased to 3.5 min and the number of amplification cycles was reduced by one. Products were electrophoresed on SpreadexTM EL-300 S-100 or SpreadexTM EL-600 S-50 gels (Elchrom Scientific AG, Cham, Switzerland), depending on allele sizes, using the SEA 2000TM advanced submerged gel electrophoresis apparatus (Elchrom Scientific AG). Gels were run at 100 V for 90–160 min, depending on allele sizes, and scored against the M3 marker ladder using the Q-EL™ 330 Digital Recording and Analysis System (Elchrom Scientific AG, Switzerland).

Table 2.   Primer sequence and related information for five new Scathophaga stercoraria microsatellite loci.
LocusPrimer Sequences (5′–3′)Repeat motifTA(°C)No. allelesSize (bp)HoHe
  1. n = 223 samples for each locus. Repeat motif and size of amplification product are based on the original sequenced clone (GenBank accession numbers: AY437870-4). N in repeat motif refers to any nucleotide.

  2. TA, annealing temperature; Ho, observed heterozygosities; He, expected heterozygosities.

  3. [Correction added on 30 September 2008, after first online publication: term Repeat motif, SsCa30, corrected to the above]


Statistical analyses

We used the software package genepop 3.4 (Raymond & Rousset, 1995a) to test for Hardy–Weinberg equilibrium, heterozygote deficiency or excess (for details on these methods see Guo & Thompson, 1992; Rousset & Raymond, 1995). Tests for genotypic linkage disequilibrium were conducted using fstat 2.9.3 (Goudet, 2001). Because of multiple testing, we applied sequential Bonferroni correction (Rice, 1989).

We measured genetic diversity at thirteen microsatellite and three allozyme loci as the total number of alleles detected, mean number of alleles detected per population, mean observed (HO) and mean expected (HE) heterozygosity across all populations. In addition, we generated an unbiased estimate of genetic variability (gene diversity: Nei, 1987) and allelic diversity (allelic richness: El Mousadik & Petit, 1996; Petit et al., 1998) within populations using fstat 2.9.3. We examined latitudinal trends in genetic variability by regressing the intra-population gene diversity and allelic richness on latitude for each locus as well as for the means (separately for microsatellites and allozymes). We also compared mean gene diversity and mean allelic richness using univariate anovas, with sampling scheme (e.g. field-collected, hatched and F1) as a factor to address any influence that sampling may have had on neutral genetic estimates. We searched for private alleles (those present in only one population: Glaubitz, 2004) using the software convert and estimated gene flow as the effective number of migrants (Nm) using Slatkin’s (1985) private allele method implemented in genepop 3.4. In addition, we sought evidence of genetic bottlenecks (e.g. founder effects) that may have occurred during the process of recolonization of northern Europe after the last glaciation (bottleneck 1.2; Piry et al., 1999). We did this assuming both a one-step stepwise mutation model and a two-phase mutation model and applied Wilcoxon signed-ranks tests (Piry et al., 1999; Maudet et al., 2002).

Single- and multi-locus FST values were computed over all 10 populations as well as for the five populations used in the common garden experiments (see below; Weir & Cockerham, 1984). We calculated pairwise FST values for allozymes and microsatellites separately, and for all 16 loci combined. Single- and multi-locus SE or confidence intervals were estimated using jackknifing and bootstrapping procedures. We could not perform jackknifing or bootstrapping with only three allozyme loci (Raymond & Rousset, 1995b), so no SE could be estimated for the allozyme multi-locus FST. We tested for isolation by distance by comparing genetic [FST and FST/(1 − FST)] against geographical [in km and in ln(km)] distances using Mantel tests. We repeated Mantel tests using only the microsatellite data and estimated the parameter ΦST (Rousset, 1996) instead of FST. Distances in kilometres between population pairs were computed using a calculator available on the internet (

FSTQST comparisons are only reasonable if applied markers are effectively neutral. We therefore verified neutrality of applied markers by considering the relationship of FST to heterozygosity as implemented in FDIST2 (Beaumont & Nichols, 1996). We ran the program using stepwise mutations, infinite alleles and 10, 50 and 100 islands (demes) that can be occupied, and carried out 50 000 realizations.

Quantitative traits

We performed common garden laboratory rearing experiments with a subsample of five populations (ICE, SWE, UK, CH[Z] and CH[L]; cf. Blanckenhorn & Demont, 2004; Demont & Blanckenhorn, 2008). Briefly, we produced full sib families by crossing flies in the laboratory and split them among three environments. The larvae were allowed to develop at 60% relative humidity and constant 12, 18 or 24 °C temperature to investigate phenotypic plasticity according to temperature (Blanckenhorn & Demont, 2004; Demont & Blanckenhorn, 2008). Temperature treatments were coupled with 12, 13 and 14 h photoperiod, respectively, as these variables are strongly correlated in nature. Larvae were reared in plastic containers with overabundant (i.e. > 2 g per larva), uniform cow dung. We checked the containers for emerged adults at least every other day, and calculated egg to adult development times for all emerged individuals. We also measured the hind tibia length of three emerged males and females per family. Flies raised at 12 °C partially entered diapause (cf. Blanckenhorn, 1998a,b; Demont & Blanckenhorn, 2008). Separation of directly and indirectly developing (i.e. diapausing) individuals (12 °C treatment) was done using their bimodal distribution of emergence times: development times < 69 days were considered as direct development and those ≥ 69 days considered as diapause (indirect) development (cf. Demont & Blanckenhorn, 2008). All data were expressed as family means (N = 12–18 replicate families per population and rearing temperature combination). More details on the rearing methods and plots of the different quantitative traits against latitude or winter length can be found in Blanckenhorn & Demont (2004) and Demont & Blanckenhorn (2008).

We computed population differentiation in quantitative traits (QST) according to the formula


[Correction added on 30 September 2008, after first online publication: incorrectly formatted equation corrected to the above] where inline image is the component of additive genetic variance between populations and inline image the component within populations (Spitze, 1993; Merilä & Crnokrak, 2001; McKay & Latta, 2002). inline image was estimated by VP, the observed component of variance between populations and inline image was estimated by VF, the observed component of variance between families assuming full-sib families: inline image = 2VF (Lynch & Walsh, 1998).

We carried out separate analyses for hind tibia length, development time and growth rate. For all analyses we fitted a linear mixed model with sex, temperature and their interaction as fixed effects, and population and family as random effects. As only flies reared at 12 °C partly went into diapause, linear mixed models for proportion of individuals in diapause (arcsin√x-transformed) and duration of diapause did not include temperature as a fixed effect. As our calculations are based on a full-sib design, the estimates are confounded by nonadditive genetic variances and maternal effects (Lynch & Walsh, 1998). We address this in the discussion.

Summarizing the variability in parameter estimates with symmetrical intervals is often inappropriate for variance components and QST estimates because their distribution can be quite asymmetrical. We therefore adopted a Bayesian approach and generated a sample from the posterior distribution of QST estimates. A recent theoretical study by O’Hara & Merilä (2005) has shown that Bayesian methods produce confidence intervals with the highest coverage (i.e. the proportion of times that a confidence interval contains the true value). We fitted our linear mixed-effects models using the lmer function in the R package lme4 (R Development Core Team, 2008). We used the mcmcsamp function from the same R package to generate a sample from the posterior distribution of the parameters of the fitted model using Markov Chain Monte Carlo (MCMC) methods. To simulate posterior QST estimates we used a slightly modified R code from Hall et al. (2007), which can be found at for each analysis, two independent MCMC chains were run after a burn-in of 10 000 iterations; every five of the next 50 000 iterations were taken to give a total of 2 × 10 000 draws from the posterior distribution (cf. Palo et al., 2003; Evanno et al., 2006). We used the coda package in R to perform output analysis and convergence diagnostics for the MCMC simulations. With Bayesian analysis the full posterior distribution of a statistic of interest is obtained: we therefore calculated QST point estimates (mean, median and mode) and the 95% central posterior interval of QST estimates from this posterior distribution. We preferred the 95% central posterior interval to the 95% highest posterior density region because the former has a direct interpretation as the posterior 2.5% and 97.5% quantiles (Gelman et al., 2004). We statistically assessed if the degree of differentiation in the five quantitative traits exceeds that attainable by genetic drift alone, by comparing QST 95% confidence intervals (95% central posterior intervals) with FST 95% confidence intervals. Nonoverlapping intervals indicate that the differentiation in quantitative traits exceeds neutral expectation, with directional natural selection favouring different phenotypes in different populations as the likely cause.


Hardy–Weinberg equilibrium, null allele frequencies and linkage disequilibrium

Four of the 13 microsatellite loci (SsCA28, Ss63T7, Ss28SP and SsCA30) were not in Hardy–Weinberg equilibrium (all P < 0.05 after sequential Bonferroni correction, k = 16) and exhibited a significant deficiency of heterozygotes (Table 3). Previous work using these loci for paternity analyses revealed the presence of null alleles at these loci (e.g. Hosken et al., 2001, 2003). Using the formula for null allele frequency r = D/(2 − D), where D = (HEHO)/HE (Chakraborty et al., 1992), average null allele frequencies (± 1 SE) across all populations were 0.156 ± 0.038, 0.109 ± 0.022, 0.167 ± 0.032 and 0.098 ± 0.016, respectively for the four loci. Heterozygote deficit at these loci, while significant over all populations, was driven by significant deficiencies in fewer than four of our 10 populations in all cases. Therefore, we included these loci in all subsequent analyses, after showing that removal of these loci from the analysis did not qualitatively change results (data not shown). We detected no evidence of genotypic linkage disequilibrium in any of the pairwise comparisons.

Table 3.   Genetic variability at thirteen microsatellite and three allozyme loci in 10 Scathophaga stercoraria populations: mean number of alleles detected per population (N), total number of alleles detected, and mean observed (HO) and expected (HE) heterozygosities (± SE) per population.
LocusN ± SETotalHO ± SEHE ± SE
  1. *Significant heterozygote deficiency at the 5% level after sequential Bonferroni correction (k = 16).

 SsCA310.9 ± 0.6190.761 ± 0.0230.853 ± 0.013
 SsCA1610.3 ± 0.5190.818 ± 0.0260.801 ± 0.011
 SsCA2410.9 ± 0.4180.822 ± 0.0150.823 ± 0.014
 SsCA269.7 ± 0.6160.855 ± 0.0270.842 ± 0.013
 SsCA28*9.9 ± 0.4190.619 ± 0.0440.831 ± 0.013
 Ss63T7*9.5 ± 0.3130.689 ± 0.0320.850 ± 0.010
 SsCA124.5 ± 0.370.532 ± 0.0480.526 ± 0.023
 Ss28SP*8.9 ± 0.3140.624 ± 0.0390.861 ± 0.008
 SsCA2115.0 ± 1.2370.896 ± 0.0240.878 ± 0.009
 SsCA30*11.3 ± 0.6260.724 ± 0.0240.877 ± 0.005
 SsCA113.0 ± 0.5210.891 ± 0.0200.893 ± 0.006
 SsCA175.7 ± 0.390.634 ± 0.0300.674 ± 0.022
 SsCA206.0 ± 0.380.595 ± 0.0380.652 ± 0.021
 MPI2.4 ± 0.230.383 ± 0.0400.398 ± 0.019
 PGM2.8 ± 0.240.179 ± 0.0360.188 ± 0.032
 GPI2.4 ± 0.350.106 ± 0.0400.116 ± 0.031

Genetic variability

Microsatellite polymorphism was substantial, with seven (SsCA12) to 37 (SsCA21) alleles detected at a given locus. Genetic variability of allozyme loci was much lower with three to five alleles detected per locus (Table 3). Intra-population genetic diversity was also high (Table 1). However, we detected no relationship between latitude and any measure of genetic variability. All regressions of measures of genetic diversity against latitude were not significant, irrespective of whether the analysis included a single locus or across-loci averages (gene diversity and allelic richness: microsatellites: all F1,8 ≤ 2.729, all P ≥ 0.137; allozymes: all F1,7 ≤ 3.400, all P ≥ 0.108). We also detected no evidence for recent bottlenecks (Wilcoxon signed-ranks tests, all populations: P ≥ 0.884). There was no obvious relationship between the number of private alleles and latitude (Table 1). The method of sampling also had no effect on measures of genetic variability (gene diversity and allelic richness: microsatellites: all F2,7 ≤ 2.172, all P ≥ 0.185; allozymes: all F2,6 ≤ 3.914, all P ≥ 0.082).

Population structure and differentiation

Global FST values were small but significant (Table 4). Single-locus FST values were comparable to multi-locus FST values, with one exception: single locus assessment for the allozyme locus GPI stood out as the largest overall (Table 4). However, all single-locus FST values except that for locus Ss63T7 were not significant (Table 4). Global tests of population differentiation based on allele frequencies (genic differentiation) revealed significant differences among populations (Fisher’s method, over all populations for microsatellites: χ226 = infinity, P < 0.00001; allozymes: χ26 = 30.728, P = 0.00003; both markers combined: χ232 = infinity, P < 0.00001). Specifically, the five microsatellite loci SsCA3, Ss63T7, Ss28SP, SsCA21, SsCA30 and the allozyme locus GPI showed significant allelic differentiation among populations (Table 4).

Table 4.   Single- and multi-locus FST values for all 10 populations and for the subsample of five populations used in the quantitative traits (common garden) experiments.
Marker lociAll populationsFive populations‡
  1. †SE obtained by jackknifing over populations or loci.

  2. ‡ICE, SWE, UK, CH[Z] and CH[L].

  3. *Significant genic differentiation at the 5% level after sequential Bonferroni correction (k = 19).

  4. **Significant genic differentiation at the 1% level after sequential Bonferroni correction (k = 19).

SsCA3* 0.006 ± 0.006 0.008 ± 0.009
SsCA16 0.009 ± 0.006 0.004 ± 0.010
SsCA24 0.002 ± 0.006 0.001 ± 0.008
SsCA26 0.005 ± 0.008 0.002 ± 0.013
SsCA28−0.004 ± 0.006 0.009 ± 0.010
Ss63T7* 0.016 ± 0.007 0.024 ± 0.012
SsCA12 0.007 ± 0.009 0.011 ± 0.011
Ss28SP* 0.010 ± 0.006 0.011 ± 0.009
SsCA21* 0.011 ± 0.012 0.021 ± 0.023
SsCA30* 0.007 ± 0.008 0.009 ± 0.014
SsCA1 0.005 ± 0.006 0.007 ± 0.009
SsCA17 0.010 ± 0.008 0.008 ± 0.008
SsCA20 0.002 ± 0.007 0.011 ± 0.013
MPI 0.002 ± 0.013−0.008 ± 0.010
PGM 0.015 ± 0.011 0.007 ± 0.016
GPI* 0.051 ± 0.048 0.083 ± 0.065
All markers combined** 0.007 ± 0.001 0.010 ± 0.002
Microsatellites** 0.007 ± 0.001 0.010 ± 0.002
Allozymes** 0.012 0.009

Pairwise FST values calculated using all 16 loci did not provide a straightforward picture of geographic population structure (Table 5). The population from Sweden was significantly different from all other populations, while the Iceland population was less so, but still more differentiated than any of the continental populations or the population from Finland. The probability test for genic differentiation confirmed this picture, but provided greater support for the distinctiveness of Iceland from the rest of the species distribution (Table 5). Iceland and Sweden had the strongest influence on overall population differentiation, as when both populations were removed from the analysis global FST was no longer significant (global FST = 0.003, 95% CI: −0.001 to 0.006). Slatkin’s private allele method yielded a migration rate of 6.2 migrants per generation (Nm), and no significant correlation between any measure of genetic differentiation (FST, ΦST and their associated transformations) and geographical distance [km and ln(km)] was detected (Mantel tests, all P ≥ 0.197).

Table 5.   Lower-triangular matrix of pairwise FST values and pairwise genic differentiation for all 16 marker loci combined.
  1. Values in bold are significant FST values obtained by bootstrapping over loci.

  2. *Significant genic differentiation at the 5% level after sequential Bonferroni correction (k = 45).


FSTQST comparison

FSTQST comparisons rely on neutrality of applied markers. FDIST2 analyses revealed that all microsatellites and two allozymes (MPI and PGM) were, independently of used parameters, always within the 95% confidence interval of neutral expectation (but see Ward, 2000, for an alternative explanation on PGM). In all our analyses only GPI, as indicated by its higher FST value (Table 4), showed marginally not significant P-values, or in one model (infinite alleles, 10 demes) a significant P-value, and hence weak indication of non-neutrality. As P-values for GPI were for the most part nonsignificant, GPI was included in the estimation of neutral differentiation. Note however, that omission of GPI would have magnified the discrepancy between FST and QST estimates (see below). Mean QST estimates calculated from body size, direct development time, growth rate, proportion of individuals that entered diapause (arcsin√x transformed) and duration of diapause were 0.136, 0.151, 0.139, 0.164, and 0.130, respectively (Fig. 2). As posterior QST distributions for all five quantitative traits were positively skewed (Fig. 2) it is perhaps more appropriate to consider modes (i.e. the most likely value) and medians (i.e. the central value): modes calculated from body size, direct development time, growth rate, proportion diapausing (arcsin√x transformed) and duration of diapause were 0.067, 0.070, 0.066, 0.077, and 0.058, and medians 0.103, 0.116, 0.104, 0.126, and 0.096, respectively (Fig. 2). Most importantly, differentiation of quantitative traits was significantly greater than differentiation measured at neutral marker loci, as all five confidence intervals for quantitative traits did not overlap with those for FST (Fig. 2). FST values calculated using data derived only from the five populations used for assessing quantitative variation did not differ from those calculated for all 10 populations (paired t-test: t15 = 1.356, P = 0.195) (Table 4).

Figure 2.

 Mode (solid circles), median (squares), mean (triangles) and 95% posterior confidence intervals (solid horizontal lines) for QST estimated from five quantitative traits. The dashed vertical lines mark the 95% confidence intervals for FST derived from the 16 molecular markers. Plotted QST and FST values are based on the five populations from the common garden experiment. Note that none of the confidence intervals overlap.


Clinal variation in quantitative traits is generally thought to be maintained by natural selection in spite of gene flow (Endler, 1977). The pattern we observed for yellow dung flies across the latitudinal range of the species supports this view. Differentiation in quantitative traits was substantial, with more genetic variance being attributed to among-population differentiation than expected by neutral genetic processes (i.e. genetic drift) alone. Additionally, evolutionary theory predicts that different quantitative traits will respond differently to selection depending on their genetic architecture, but in contrast to this expectation all five measured traits exhibited very similar among-population divergence. Parallel analysis of neutral molecular markers in contrast revealed little evidence of among-population structure across Europe. Neutral structure was primarily driven by the effects of two of the most northern populations (Iceland and Sweden), and when these populations were excluded from the analysis, no population structuring was detected. Therefore, no evidence of differentiation at neutral loci among populations of yellow dung flies was present at the continental scale, including the UK. Thus, our study provides evidence that differentiation of quantitative life history traits is more the result of selection than neutral genetic processes (Merilä & Crnokrak, 2001; McKay & Latta, 2002; Leinonen et al., 2008).

Our QST estimates are based on variation among full-sib families and, therefore, estimates of additive genetic variance may include dominance and/or environmental maternal effects (Lynch & Walsh, 1998). However, theoretical studies on the properties of QST concluded that nonadditive genetic effects are unlikely to bias QST estimates considerably (e.g. Goudet & Buchi, 2006). A recent comprehensive meta-analysis by Leinonen et al. (2008) supported this, as using broad- or narrow-sense estimates had an inconsiderable effect on QST estimation. Therefore, we do not think that our comparison is biased by our use of full-sib genetic estimates. Furthermore, use of five populations is within the common range of studies estimating QST (Leinonen et al., 2008) and at lower QST values, such as in our case, estimation bias due to the number of populations is minor (O’Hara & Merilä, 2005).

The latitudinal genetic differentiation and the specific responses of body size, development time, and aspects of diapause to the temperature treatments have been previously published (Blanckenhorn & Demont, 2004; Demont & Blanckenhorn, 2008). Briefly, body size increased (following Bergmann’s rule) and development time decreased (following the converse Bergmann rule) with increasing latitude. This implies faster growth rates of high latitude populations and hence adaptive countergradient variation. In addition, the number of individuals entering diapause and diapause length increased with increasing latitude (Demont & Blanckenhorn, 2008), a pattern detected in other laboratory investigations evaluating patterns of diapause (Mousseau & Roff, 1989; Schmidt et al., 2005). These results indicate that high latitude populations evolved a more ‘hard-wired’ (i.e. genetically fixed) temperature threshold to enter diapause and, because of longer winters towards the poles, longer diapause (Demont & Blanckenhorn, 2008).

Genetic variability, population structure and differentiation

Microsatellite polymorphisms can reveal post-glacial patterns of species range expansion (Zeisset & Beebee, 2001; Garner et al., 2004; Burg et al., 2006; Clauss & Mitchell-Olds, 2006; Pabijan & Babik, 2006), and patterns described by microsatellites are usually supported by mtDNA data and allozymes (Raeymaekers et al., 2005; Adams et al., 2006; Pertoldi et al., 2006; Wilson, 2006; this study). Distance from a glacial refugium is a commonly cited explanation for patterns of genetic variability detected across species ranges (e.g. Hewitt, 1996, 2000; Palo et al., 2004). Population-level neutral polymorphism across the range of S. stercoraria showed no clear pattern of either increasing or decreasing genetic variability along the latitudinal gradient, and we saw no evidence of genetic bottlenecks anywhere across the range. The lack of discernable pattern indicates that recolonization after glaciation was substantial and probably rapid, and may also be affected by more recent patterns of dung fly movement. This is further supported by the low levels of among-population differentiation we estimated using FST, and the absence of any discernable isolation by distance, even though substantial numbers of private alleles were detected in many of our populations. In the genus Drosophila, which in many ways has similar ecology to yellow dung flies (Blanckenhorn, 1999), population genetic studies at a broad geographic scale have described population structure similar to that presented here (Pascual et al., 2001; Magiafoglou et al., 2002). Migration in Drosophila occurs over large geographic distances (e.g. Coyne et al., 1982; Berry & Kreitman, 1993; Colson, 2002; Magiafoglou et al., 2002; Kennington et al., 2003; but see Agis & Schlötterer, 2001), and there is no doubt that the more robust yellow dung fly is equally capable of substantial geographic displacement. Previous studies have suggested that S. stercoraria have high mobility: Kraushaar et al. (2002) found no significant population genetic structuring among 34 Swiss populations and tests with tethered flies in the laboratory demonstrated maximal flight distances of 70 km and 90 km for females and males, respectively (C. Reim, unpublished). Migration aided by both wind and humans through the movement of livestock may also contribute to the observed pattern. Nevertheless, the lack of detectable population structure across inter-population distances of up to 2800 km is remarkable.

Flies from southern Sweden were found to be significantly different from the other populations sampled irrespective of which measure of differentiation (FST or genic) was used. Phylogeographic studies often show that Sweden was colonized after glaciation by routes that did not pass through western Europe (Zeisset & Beebee, 2001; Schönswetter et al., 2003; Heuertz et al., 2004; Emerson & Hewitt, 2005), but most studies group Sweden with Fennoscandian countries or the U.K. (Zeisset & Beebee, 2001; Carlsson et al., 2004; Tiedemann et al., 2004; Marion & Le Gentil, 2006). In our study, Swedish populations are distinct from both Finland and the U.K., which suggests that Sweden represents a distinct northern European lineage and that the western European populations and Finland are derived from a separate southern source that recolonized along two separate routes. Two routes of recolonization into Fennoscandia have been proposed for other taxa (Hewitt, 2000). Further sampling in Norway, Denmark, northern Germany and Poland could confirm whether Swedish flies are indeed derived from a distinct northern European lineage.

At first glance, the barrier posed by the thousands of kilometres of North Atlantic Ocean separating Iceland from the European continent seems a reasonable explanation for why Iceland is distinct, albeit less consistently than the Swedish population. However Iceland rarely proves to be phylogeographically distinct from Fennoscandia or the UK. (Hagen et al., 2001; Liebers & Helbig, 2002; Philipp & Siegismund, 2003; Ruokonen et al., 2005; but see Dalen et al., 2005), and the North Atlantic Ocean apparently poses little in the way of a barrier to insect dispersal. Of 150 coleopteran species found on Iceland, none are endemic (Sadler, 1998), and the North Atlantic insect fauna in general lacks endemics (Sadler, 1999). Even flightless collembolids have the ability to disperse over long reaches of cold, arctic waters (Coulson et al., 2002) and successfully colonized the recently formed island of Surtsey, just south of Iceland, a mere 10 years after the formation of the island (Moore, 2002). The inconsistent pattern of distinctiveness based on FST, the ease by which insects colonize Iceland and adjacent islands, and the possibility of two lineages in northern Europe may instead indicate that Iceland was colonized by flies from the western lineage and flies from the other northern lineage denoted in our data set by Sweden. Further, we cannot dismiss the possibility of northern refugia as an explanation for both Swedish and Icelandic distinctiveness (e.g. Tremblay & Schoen, 1999; Ehrich et al., 2000).


The low population differentiation of yellow dung flies in Europe can most likely be explained by high gene flow due to mobility in combination with large population sizes, weakening any isolating effects of genetic drift. The latter may relate to a relatively recent range expansion, as this fly is a cow dung specialist and may consequently have become so common because of human cattle-breeding. Clearly, northern populations are not depleted in terms of molecular genetic variation, most probably because this fly is well adapted to cold climates. Importantly, differentiation in quantitative traits was greater than differentiation in putatively neutral molecular markers, a result providing strong indirect evidence that the latitudinal clines in life history traits observed in S. stercoraria (Blanckenhorn & Demont, 2004; Demont & Blanckenhorn, 2008) are caused by selection despite considerable gene flow between populations. This conclusion is in agreement with other work arguing that random genetic drift is an unsatisfactory explanation for the common differentiation in quantitative traits observed in nature (Merilä & Crnokrak, 2001; McKay & Latta, 2002; Palo et al., 2003; Leinonen et al., 2008).


We would like to thank Oliver Y. Martin for helpful discussions and suggestions during the entire study and for reviewing several versions of the manuscript. Many thanks also to Lukas F. Keller and Peter Wandeler for advice with population genetic analyses, David Hall for providing R code, and Tony Wilson, Jason Kennington, Brent Emerson, Laurene Gay, Tracie M. Ivy, Luc F. Bussière, Stephanie Bauerfeind and Gioia Schwarzenbach for very helpful comments on the manuscript. Thanks particularly to Thomas Bucher, Barbara Vincenz, Caroline Henggeler and Pierina Maibach for their tireless help in the laboratory, and Noëmi Braem for providing assistance with Fig. 1. We acknowledge funding from the Swiss National Foundation and the Forschungskredit of the University of Zurich for this project.