40 Years of Evolution: Darwin’s Finches on Daphne Major Island

Box 3.1 Research Strategy

We studied reproduction in every year for 16 years: 1976–91. We used maps to record nest locations (appendix 3.1). Most finches nest in Opuntia cactus bushes, with less than 5 percent nesting in Bursera trees or in Portulaca, Chamaesyce, and Sesuvium (perennial) plants close to the ground. Except for a couple of cactus bushes on a sea cliff in which finches may have nested, all nests were accessible at heights of less than 2.5 m. The goal was to determine clutch size, hatching success, and fledging success at every nest throughout each breeding season from 1978 to 1991; the coverage was approximately 50% in 1976, and only a few pairs of scandens bred in 1977. Nests were missed very rarely (Boag and Grant 1984b). As the study progressed, the fraction of breeding pairs that had been banded and measured increased to 99.5%; in early 1992 the last scandens, a male, was captured and banded, so at that time every adult of both species was identifiable. Nests were checked every 2–4 days, and nestlings were weighed and banded at day 8, counting the day of hatching as day 0. The first two broods produced in 1992 and 1993 were studied with complete coverage. In subsequent years breeding was monitored in an opportunistic way and for a shorter time, except that in 1998, a year of El Niño, breeding was studied again throughout the season (Grant et al. 2000).

Starting in 1988, we took a drop of blood for DNA analysis from the brachial vein of each finch, both adults and nestlings (fig. 2.13). In the absence of a buffer we developed our own method of storage with very good results. Samples were stored on EDTA-soaked filter paper in Drierite^® in the field and at –80°C in the laboratory, and they yielded high-quality DNA. A decade later it became possible to use these samples to identify parentage unambiguously with 16 microsatellite DNA markers (Petren 1998). Our identification of mothers by observations at nests was confirmed in all cases, but paternity was sometimes misidentified by observation. The 1988 and subsequent samples were dominated by birds hatched in 1987, but there were some old birds among them, including one that had hatched in 1975! Most offspring of known (genotyped) parents in our samples hatched in 1991.

This chapter explains how we estimated the genetic component of phenotypic variation in body-size and beak-size traits, and what we found after taking into account possible biases in the estimates that arise from misidentified paternity due to extra-pair mating, maternal effects, and correlations between genotypes and environments. The overall conclusion we reach is that a large fraction of the measurable variation can be attributed to genetic factors: all morphological traits are highly heritable. This applies to both fortis and scandens, although there are small differences between them. Therefore both species on Daphne possess a strong potential for evolution in all size traits.

A phenotypic trait is heritable to the degree that measured variation in the trait reflects variation in genotype. Family members are genetically more similar to each other than to nonfamily members. This fact is used to estimate the amount of phenotypic variation that can be attributed to the additive effects of genes. In formal terms the heritability (h²) of a trait in the narrow sense is the additive genetic variance (Va) as a proportion of the phenotypic variance (Vp): h² = Va/Vp. In practical terms it is estimated most simply by regressing measurements of offspring on the average of the measurements of their two parents. If there is no genetic component to the phenotypic variance, the slope will be 0. The theoretical maximum is 1.0. Most studies of morphological variation in birds find values between 0 and 0.5, sometimes higher (Grant and Grant 2000a, Merilä and Sheldon 2001). There are more statistically refined ways of estimating heritable variation by taking into account genetic effects from relatives other than parents when extensive pedigrees are available, and by estimating and excluding certain sources of environmental variation (reviewed in Kruuk et al. 2008). These give greater precision to the estimates, but even if only one parent is known, an estimate can be obtained by doubling the slope of the regression of offspring on single-parent measurements (Falconer and Mackay 1995, Lynch and Walsh 1998). The main caveat in all these studies of natural populations is the estimates may be biased by uncontrolled factors. For example, heritability estimates may be inflated as a result of parents and offspring growing up in similar environments that affect their final sizes, with different families experiencing different environments (genotype × environment correlations).

Our first estimates of heritabilities were remarkably high (fig. 3.1); in fact some of them were statistically indistinguishable from 1.0 (Boag and Grant 1978, Boag 1983). However, the standard errors of the estimates were also large. Sampling variance is large because our samples were relatively small and not because of unreliability in the measurements. In fact measurements of beak traits were highly repeatable (table 3.1, appendix 3.2). We minimized effects of measurement error by averaging repeated measurements of the same bird taken at different times.

The high heritability estimates were confirmed in a follow-up study in 1978 (Boag 1983). The slopes of regressions of offspring on mid-parent values in 1976 and 1978 were almost identical (Boag 1983), despite a strong difference in rearing conditions. Finches bred at high density and produced one brood in 1976, and bred at low density and produced three broods in 1978. Offspring that hatched in 1978 were closer in size to their parents, on average, than were the 1976 offspring, suggesting density-dependent suppression of growth in 1976: slower, incomplete, or earlier cessation. The fact that the regression slopes were identical in years of different conditions shows that large adult size is not a conspicuous function of the particular rearing environment experienced during growth (genotype × environment interactions). Growth is stunted when conditions are poorer than in 1976, but then the offspring do not survive long enough to be caught and measured as adults (Price 1985) and therefore do not enter the analysis.

We estimated heritabilities from five more years of breeding (1981, 1983, 1984, 1987, and 1991). Number of families ranged from 51 in 1981 to 101 in 1987, and number of offspring varied from 99 in 1981 to 467 in 1987 (Grant and Grant 2000a). The two main results were (a) beak trait heritabilities exceeded those for body-size traits, and (b) for beak depth the estimates were equal to or greater than 0.50 in all seven years (fig. 3.2). Relatively little annual variation in the estimates means relatively little influence of annual variation in growth conditions.

Our overall conclusion is there is substantial genetic variation in all six traits. It supports the hypothesis of evolutionary change of morphological traits on Daphne. However, heritability estimates may be biased, and possible biases need to be examined before the conclusion can be considered secure.

TABLE 3.1

Repeatabilities: the degree to which repeated measurements of the same individual are the same, as assessed by analysis of variance (Lessells and Boag 1987)

Note: Body-size traits have lower repeatabilities (R) than beak-size traits, due to inherently higher measurement error and higher variation within an individual over time; wing feathers abrade and are replaced, and body mass fluctuates daily and seasonally. Measurement error, as a proportion of the phenotypic variance, is 0.06–0.21 for fortis traits and 0.11–0.29 for scandens traits (Grant and Grant 1994). Abbreviation: s.e. is standard error of the repeatability estimate. Changes from first to second year only are illustrated in fig. A.3.2.

These associations could give biased heritability estimates if fortis adult size varied systematically across the three habitats (fig. 1.4); they did differ in some size traits according to where they hatched. We tested for differences in size between habitats by two-way analysis of variance (ANOVA), including sex as a factor because males are larger than females on average. After the sex difference was removed, there were no significant differences between habitats in any of the PC indices of size and shape, nor were there any differences within each of the habitats between those that bred in their natal habitat and those that bred in another one (all p > 0.1).

We found substantial extra-pair young effects upon heritability estimates, inconsistent evidence for maternal effects, and no evidence that the estimates are inflated by genotype × environment correlations. Two other pieces of information support the last conclusion. First, regressions of extra-pair young on their apparent, social (cuckolded) fathers were all small and nonsignificant, indicating little or no inflation from common environment effects (Keller et al. 2001). Second, experimental manipulation of clutches in seven studies of birds elsewhere all failed to find an effect of foster-rearing environment on adult size (Grant and Grant 2000a).

Given these findings, we asked if high heritabilities are solely a feature of fortis or if they apply to other species as well. Specifically, can the absence of an evolutionary trend in scandens beak depth (fig. 1.7) be explained by lack of genetic variation? The answer is no. G. scandens heritabilities were estimated in three years when family sizes exceeded 20 (Grant and Grant 2000a): in 1981 (n = 37), 1983 (n = 68), and 1987 (n = 25) but before we were able to identify EPY. Heritability estimates for the six traits uncorrected for EPY averaged 0.53, which is lower than an average of 0.70 for fortis (table 3.2, fig. 3.3). For beak depth, the trait shown in figure 1.7, the heritability estimate is 0.55 (table 3.2). Heritabilities for scandens traits corrected for EPY should be almost the same as the uncorrected ones (Keller et al. 2001), because EPY rate for this species is only 8%–10% (Petren et al. 1999a, Grant and Grant 2011a).

The combined results stimulated an analysis of heritable variation in a third species, G. conirostris on Genovesa Island (Grant and Grant 1989). Heritabilities were found to be high, almost identical to those of fortis, and with the same average of 0.70. Therefore all three species of Darwin’s finches studied in detail have highly heritable morphological variation and therefore a strong potential for evolutionary change. Evolutionary change as a result of natural selection is the subject of the next chapter.