C. E. Adams, University Field Station, Institute of Biomedical and Life Sciences, University of Glasgow, Glasgow G63 0AW, UK; e-mail: firstname.lastname@example.org
Abstract – There is a very high degree of discrete variation in phenotype between populations of Arctic charr. This takes the form of variation not only in morphometric and meristic characters traditionally used to distinguish species, but also in characteristics of life-history, behaviour, coloration and ecology. This variability has a number of consequences, one of these is that there is a strong case for the conservation of populations with extreme phenotypes. However, if variation is discrete between populations but continuous across many populations, this poses difficulties in separating those populations of high conservation value from those of lower conservation value. In this paper we describe a statistical technique which enables populations on the extreme edges of the range of phenotypic variation to be identified and apply this to the morphometric characters of charr from 25 populations from across Scotland and Ireland. The technique enables the identification of any proportion of the most extreme phenotypes. When applied to our data, one population (Loch More) was in the top 2 percentile of the most extreme phenotypes from across the range of all populations included. Three populations were within the top 10% most extreme phenotypes (Lochs More, Uaine and Earn) and a further five within the top 20% (previously mentioned lochs plus Lochs Eck, Merkland, Uaine, Talla and Lough Nalughraman). This technique can potentially be used on any species and on any suite of characteristics as an objective measure of conservation value of a population within a continuous phenotypic range.
Variations in phenotypic traits across the geographic range of a species are commonplace. Where the habitats occupied by a species are spatially disjointed or where mobility of the species is limited, then limitations on free gene flow and/or local environmental variation operating upon phenotypic plasticity can result in significant discontinuities in the phenotype throughout the range of a species (Jenkins 1997). Amongst the salmonid fishes, it is presumed that micro-evolutionary processes working over small scales combined with restricted gene flow between populations breeding in different freshwater systems, have led to the diversity of form across many species. In the Brown trout (Salmo trutta L.) for example, between-population variation in form, most obviously in colour pattern, but also in a wide range of other traits, has long been recognised (Behnke 1986;Ferguson 1989). In this species, phenotypic differences between populations are underpinned by distinct genetic differences between populations resulting from postglacial isolation of populations in different fresh waters (Ferguson 1989;Prodöhl et al. 1994;Hynes et al. 1996).
Defining conservation policy and practice that provides for appropriate levels of protection for extremely variable species can be highly problematic (Meffe 1987; Ferguson 1989). Where a species exhibits a between-population mosaic of variation in phenotypic characters and the genetic structuring within the species that supports this, this presents a series of very significant challenges to wildlife management and conservation. Because the species is not the same in different populations, it may be inappropriate to develop management or conservation policy aimed at the level of the species. However, if there are a large number of populations, each exhibiting a different suite of character variants underpinned by genetic variation, it is unlikely to be practically possible to allocate sufficient resources to allow for positive management or protection of all of the individual populations.
One element of the dilemma of attempting to balance allocation of resources against the need for management is that some populations may be significantly more divergent within the species than others and thus might be considered as of potentially greater conservation value. Thus some means of quantifying the degree of divergence of populations from each other; a measure of their ‘uniqueness’, could be used to prioritise conservation actions.
Here we develop a methodology which leads to a measure of population ‘uniqueness’ based on variability in phenotypic characteristics. We apply this to data derived from Arctic charr populations from across Scotland and Ireland to attempt to identify populations that have divergent phenotypes and thus may be of a high conservation significance.
Materials and methods
Arctic charr were collected by gill-netting from 25 populations across Scotland and Ireland (Table 1). Standard Nordic survey gill nets (Lundgrens, Sweden), 60 m long and 1.5 m deep comprising 10 panels of 8–45 mm half-mesh size, were set in four habitat types (the littoral, sub-littoral, profundal and pelagic zones) of each lake overnight. Up to 50 Arctic charr from each site (see Table 1) were removed from the nets, stored on ice and photographed in lateral view for morphometric analysis within 24 h.
Table 1. Population name, location and the number of individual charr included in the analysis.
The phenotypic characteristics chosen for quantification in the collected charr were related to the morphology of the head. Head morphology in charr is known to have a functional role in foraging (Adams & Huntingford 2002b) and has been shown to be related to ecological differences between populations (Skúlason & Smith 1995).
Eleven morphometric variables relating to head morphology together with pectoral fin length, were measured digitally directly from photographs (Fig. 1). Because these linear measures of morphology varied with fish size, to derive a measure of morphology that was independent of the size of individual fish, each of the univariate morphology measures was regressed upon fish fork length (for all fish across all populations pooled) and regression residuals used as a measure of that variable, free from the effect of body size (Reist 1986; Adams et al. 2003).
To determine overall variation in head morphology, the univariate body size-independent morphometric measures were reduced to two variables of shape using Principal Components Analysis. Principal Components scores were derived for all fish from all populations from the first and second principal components (PC1 and PC2, respectively) (for a more complete explanation of the technique see Adams et al. 2003). To determine the degree of phenotypic deviation between populations, the population mean principal component scores for PC1 and for PC2 were calculated for each population separately. A mean of population mean values was used as the phenotype centroid position of the complete data set and Pythagoras’ theorem was used to calculate the two-dimensional vector distance of each individual population from the phenotype centroid for all populations.
There was considerable variation in expressed head morphology between populations from across Scotland and Ireland. Shown plotted on the first two dimensions of a principal component analysis (Fig. 2), the 12 size corrected morphometric variables show a broad scattered pattern of distribution across both dimensions. The distribution of the scatter of data did not differ significantly from a normal distribution in the PC1 dimension (Shapiro–Wilk test: SW = 0.968; 25 d.f.; P = 0.597) nor in the PC2 dimension (S-W = 0.979; 25 d.f.; P = 0.864). In both dimensions, there was a significant difference in head morphology between populations for both PC1 (F1,24 = 29.5; P < 0.001) and for PC2 (F1,24 = 22.9; P < 0.001).
To identify the populations exhibiting the most extreme head morphologies, the vector distance for each population mean from the mean of population mean values was calculated in the two PC dimensions using Pythagoras’ theorem. The mean of vector distances (±SE on the mean) for the 25 populations included in this study was 1.55 ± 0.176.
The z distribution provides a mechanism to derive thresholds above which one would predict a particular percentile of the population would lie within a normal distribution of a variable of known mean and standard deviation. Thus Y, the variable threshold is given by:
where z is the standard normal (derived from tables) SD the variable standard deviation and x the variable mean (Moore & McCabe 2002).
Applying this technique to the data presented here, 1% of the most extreme phenotypes would have a phenotype vector score in excess of 3.56; the populations exhibiting a phenotype representing the 5% most extreme phenotypes would have a phenotype vector score in excess of 2.98. Figure 3 shows the threshold values for the percentile deviations from the mean.
In the data examined here, there were no populations within the 1% of the most extreme phenotypes of head morphology. Loch More, however, was within the range of 2% of the most extreme head morphology phenotypes (Table 2). An additional population (Loch Earn) was within the 5% most extreme of head morphologies and Loch Uaine joined Earn and More as being with the most extreme 10% of head morphologies. There were seven populations in total within the 20% of the most extreme head morphologies and eight within the top 25% (Table 2).
Table 2. Arctic charr populations within the extreme ranges of phenotypic deviation from a centroid of the 25 populations examined.
For species where there is between-population variation in expressed phenotype or the underlying genotype across that species, the identification of conservation requirements are complex. Faced with a large number of populations potentially requiring management (and possibly different management strategies), some means of ranking the importance of populations is required in practice. For genetic approaches to this difficulty, techniques have been developed to determine genetic distances between populations (Nei 1972). Here, we describe a technique that would provide an approach based on phenotypic traits.
The technique we apply is based on the assumption that the degree to which populations differ in their traits can be measured numerically and that this difference reflects the ‘uniqueness’ of the population's traits. This application assumes that the degree of variation between populations itself varies and that this variation is normally distributed; as it is in this case study (however, this approach could be adjusted if an alternative distribution were more appropriate).
Here, we apply a simple statistical technique to measure the degree of ‘uniqueness’ of each population and to identify populations with the most extreme levels of ‘uniqueness’. Although here we applied this technique to phenotype data based on head morphology, it could also be applied to any numerical phenotype data and also to genetic data. One assumption of this approach is that all the characteristics included in the analysis are of equal value in assessing ‘uniqueness’. This might not always be true for all characteristics, however with different applications of this approach characters could be weighted to account for features that might be more important than others. One advantage of this technique is that it allows a threshold for ‘uniqueness’ to be set at any level. It would be inappropriate to consider this approach as a sole means of identifying populations of high conservation value. The outcome of approach may depend upon the suite of phenotypes examined and some populations may exhibit extreme variation resulting from population bottlenecks in the past. However as one tool in decision making, the advantage of this technique is that it allows for numeric comparisons of populations.
Applying this technique to a range of 25 Arctic charr populations from across Scotland and Ireland, shows that the charr population of Loch More has a phenotype with the most extreme head morphology. Charr from an additional two other lochs (Uaine and Earn) would seem to be within the most extreme 10% of the phenotype range. On the basis of these characteristics we argue that populations representing phenotype extremes should therefore be considered as of potential conservation importance and that this technique could provide a mechanism to identify these populations.
We thank Dr G. Alexander and A. Grant for their assistance with much of the fieldwork. An early version of the manuscript was greatly improved by Prof. Peter Maitland. The work was funded by EC Grant FAIR CT96–1981 to CEA.