differentiate between the effects of mutation rate or popu-
lation size ⁄ migration rate on heterozygosity, we make use of the fact that most studies employed either allozymes or microsatellites as neutral markers, and that microsatellites on average have much higher mutation rates than allo- zymes (Ellegren 2004). If the variation among studies in heterozygosity is at least partly driven by mutation rate, then heterozygosity should be lower in studies using allo- zymes. Instead, if variation in heterozygosity is only because of variation in demography among studies, then studies using either allozymes or microsatellites should not differ in heterozygosity, because we have no reason to sus- pect systematic differences in demography. In all cases we assume that the molecular markers used are neutral (as did the original studies). In principle, selection on molecu- lar markers could also induce a correlation between hetero- zygosity and QST–FST, but this would require that selection acted on the molecular markers in the majority of studies, so we refute this possibility a priori.
Methods and results
As a starting point for compiling published studies on the topic, we used the latest review of QST–FST comparisons by Leinonen et al. (2008). We then further checked all articles that cited Leinonen et al. (2008) for relevancy and indepen- dently extracted values of FST and QST when possible. Whenever we encountered a difference with values pub- lished in Leinonen et al. (2008) and the difference could not be resolved in favour of the values published in Lei- nonen et al. (2008), we retained the values extracted by us. In addition, we recorded the published values of within- population expected heterozygosity (HWITHIN) when pro- vided. Most studies used either allozymes or microsatel- lites ( = STR or SSR markers). To allow for a meaningful comparison among marker types with known different average mutation rates, and because the calculation of HWITHIN was sometimes methodologically quite different for certain (dominant) markers, the few studies using other types of markers (AFLP, RAPD, ESTP, RFLP, CAPS) were not further considered here. An overview of the final set of studies and values used in our analyses is presented in Table 1 (n = 44 studies, 40 species).
Figure 4 shows the relationship between the difference between QST and FST and within-population heterozygosity (Fig. 4). The average QST–FST difference across studies clearly lies above the QST–FST = 0 line, supporting the con- clusion of Leinonen et al. (2008) that QST is on average larger than FST. However, there is also a clear pattern in that the difference QST–FST is positively related to heterozygosity. Dividing the studies into groups with heterozygosities either greater or smaller than 0.5, QST–FST is significantly greater for studies with high heterozygosity (least squares means
0.261 ± 0.042 SE vs. 0.087 ± 0.038 SE; ANOVA: F1,42 = 9.34, P = 0.004, R2 = 0.18). Similarly, regression of QST–FST on het- erozygosity yielded a positive slope (b = 0.277 ± 0.123 SE,
(b = 0.391 ± 0.198 SE, F1,26 = 3.90, P = 0.059, R2 = 0.13).
We calculated a simplified expected relationship between heterozygosity and the QST–FST difference, using the data of these studies. For this, we assumed that the degree of quantitative and neutral divergence is constant across stud- ies, but that the actual value of FST decreases linearly with increasing heterozygosity. We established that QST is indeed not related to heterozygosity (linear regression: F1,42 = 0.084, P = 0.77, R2 = 0.00) and used the mean of QST as an estimate for quantitative divergence (0.341 ± 0.033
SE). As an estimate of mean neutral divergence, we used the intercept of the regression of FST on heterozygosity (0.320 ± 0.056 SE). This predicts the following relationship between heterozygosity and the QST–FST difference: QST– FST = 0.341 ) 0.320 * (1 ) HWITHIN). This quantitative pre- diction matched the observed regression slope remarkably well (Fig. 4).
With respect to our first alternative explanation for a relationship between heterozygosity and the QST–FST differ- ence, there was only a mild non-significant temporal trend for the difference QST–FST (F1,42 = 1.09, P = 0.30). Moreover, the inclusion of the continuous variable ¢year¢ in the model of the regression of QST–FST on heterozygosity explained virtually no variance in QST–FST (t41 = )0.044, P = 0.97), whereas heterozygosity remained a very influential explan- atory variable (t41 = 1.95, P = 0.058). With respect to our second alternative explanation, mean heterozygosity dif- fered substantially and significantly among studies using either microsatellites or allozymes (Fig. 3; microsatellites:
0.578 ± 0.035 SE, allozymes: 0.245 ± 0.046 SE, ANOVA:
F1,42 = 33.4, P < 0.0001, R2 = 0.44).
Discussion
Is there a bias?
Our main hypothesis is that much higher mutation rates in the neutral markers than in the quantitative genetic loci are causing a strong upward bias in QST–FST.
Indeed, we found the positive correlation between het-
erozygosity and the difference QST–FST that is predicted by this hypothesis. Moreover, the predicted quantitative rela- tionship between heterozygosity and QST–FST based on this hypothesis shows a very good fit with the observed one (Fig. 4), suggesting that this hypothesis is a sufficient explanation for the observed pattern.
We found no significant temporal trend in QST–FST, and time explained virtually no variation in QST–FST indepen- dent of the effect of heterozygosity. Thus, the idea that the correlation between heterozygosity and QST–FST is acciden- tally due to independent temporal trends (e.g. a shift towards markers with higher heterozygosity coinciding with a shift towards systems with higher QST values) is not supported.
Variation in population size and ⁄ or migration rate among studies could also create a correlation between heterozygos-
Table 1 Overview of studies and values included in our analyses. We only included studies based on allozymes (¢allo¢) or microsat- ellites (¢micro¢) and that provided estimates of FST (or a comparable measure of neutral population divergence), QST and within-pop- ulation heterozygosity (HWITHIN) based on the same, wild populations. G¢st is the unbiased estimator (G¢¢ST) of Meirmans & Hedrick
2010; k is the number of study populations or regions. Some studies have multiple entries because they included several independent
comparisons, e.g. among populations within regions, and among regions. We excluded the studies on Tigriopus californicus (Edmands
& Harrison 2003) and Arabidopsis thaliana (Kuittinen et al. 1997) in our statistical analyses because they yielded theoretically impossi- ble values of G¢st and ⁄ or Jost’s D > 1