If anatomical or physiological factors limit simultaneous optimization of two relevant performance criteria, a compromised phenotype results that meets both criteria to some extent, but achieves excellence in neither [23
]. Such trade-offs are inevitable in evolution. The concept of functional trade-offs is often used to explain why species or populations differ in niche, reproductive strategy, performance criteria and behavior [25
]. Genetic analysis of Portuguese water dogs raised the possibility that the organization of genes controlling the morphology of the canine skeleton enables rapid phenotypic change in response to selection for function along a trade-off continuum between speed and power of a dog [15
] (Box 3
Box 3. Different dog breeds can be used for reducing haplotype length
Humans have managed the dog population for longer than any other domesticated animal, providing ample time and opportunity to select for novel phenotypic variations regulated by new alleles that were increased in frequency by drift and selection. Geographic isolation and selection for diverse tasks such as herding, guarding, hunting, retrieving, drafting and companionship created specialized subtypes within the species (Figure I). During the past two centuries these geographic isolates and morphological ‘types’ have become what we recognize today as the modern breeds: reproductively closed populations often derived from a small number of founder animals.
Not surprisingly, each breed although genetically differentiated has limited genetic diversity and common haplotypes are shared among breeds. Thus, Parker et al.
] found that breed membership accounted for 27% of total genetic variation and dogs could be assigned to their correct breed solely on the basis of genetic data. Significant differences in genetic composition separated four breed clusters: (i) Asian breeds including the Akita and chow-chow that grouped with the wolf; (ii) large, heavy, working breeds such as the boxer, mastiff and Newfoundland; (iii) a mixture of herding breeds such as Shetland sheepdog and Belgian Tervuren and non-herders like the greyhound and whippet; and (iv) a heterogeneous mix comprising mostly more recent European breeds. Such breeds can be used to reduce haplotype length around a QTL (Figure II).
Figure I. Dog breeds resulting from selection for energy-efficient speed (left) or power (right). Prototypic breeds exemplifying extremes from these groups are greyhounds or pit bulls (left and right center).
. Haplotype reduction based on common sequence shared by haplotypes regulating body size in different extreme breeds. Segregation of size in Portuguese water dogs (see Box 2
) enables the identification of a QTL containing the size haplotype shown at left. Regions of common haplotype nucleotide sequence can be identified in four other breeds (large, bullmastiff and mastiff; or small, boxer or bulldog) (right) reducing the haplotype to a much smaller region of common sequence.
Metrics of the canine skeleton, obtained either by direct measurement or from radiographs, present a complex array of data (e.g. in Chase et al. [15
] identifies 72 metrics). Chase and others [15
] have used principal component (PC) analysis [27
] to reduce the complexity of this array while retaining most of the information in the dataset obtained from radiographs of Portuguese water dogs. This technique transforms a set of correlated variables into independent sets of variables, vectors known as the principal components (PCs). The first PC explains the largest amount of skeletal variation (around 50% in canids), the second the next largest (no more than 15%) and so on. An unexpected result was that the PCs lent themselves to a biological interpretation: the first PC represents size as an overall average of the more than 70 different radiographic bone metrics taken from each animal (see size data on Portuguese water dogs in Box 2
). It explains ~50% of the total skeletal variation, and is highly correlated with body weight. (This is true of different dog breeds as well. Thus, all the bones of a Pekingese are small, those of an Irish wolfhound are large).
Figure 2 Skeletal shapes that represent functional trade-offs between speed and power. Dog shapes can vary along different independent trade-off axes, defined by principal components . (a) A trade-off between the size of the skull and the postcranial body. (more ...)
The remaining PCs represented aspects of shape and were more complex, with variation due to some traits (loadings) inversely correlated with variation contributed to the PC by other traits. These PCs represent a trade-off between power and speed as illustrated in Box 3
. Thus, in , smaller head and larger postcranial body (left) is associated with speed (e.g. greyhound), whereas a large head and smaller postcranial body (right) represents more power (e.g. pit bull). This shape variation is characterized by one principal component [15
]. Another PC describes an independent component of shape variation that affects metrics of length versus width [26
]. This is illustrated in in which the dimensions of a limb bone associated with speed (long and thin – on left) are part of a trade-off axis in which shorter, thicker bones support a more powerful morphology (, right). Similar PCs have been obtained for the fox [28
], separated by 10 million years of evolution from the dog [29
]. Recent radiographic data have identified several more such trade-offs common to both species. These are also components of shape consistent with the power versus speed trade-off [26
] (L. Trut et al.
PCs are phenotypes subject to genetic analysis [15
]. Every animal has a value for every PC that characterizes the variation in the population (this is illustrated by Figure I in Box 2
, graphing the values for PC1 of male and female Portuguese water dogs). Several nonlinked QTLs have been identified that regulate variation in size (PC1) and shape (PC2, PC3, etc.). PCs derived from the radiographic metrics of the skeleton have been associated with a large number of QTLs distributed throughout the genome of the Portuguese water dog ().
Figure 3 Map of the canine genome showing the position of QTLs regulating PCs derived from metrics of the Portuguese water dog pelvis and limb bones. The X chromosome and 38 autosomes are noted on the abscissa, chromosome length on the ordinate is in megabases. (more ...)
A more detailed analysis of these QTLs has examined the effects of a QTL on specific traits involved in the trade-off it regulates [15
]. Thus, the PC involving a trade-off between length and width of limb bones [26
] is regulated by a QTL associated with marker FH3585
on CFA 12. Specific haplotypes at this locus regulate both the length and width of limb bones. For example, in (K. Chase and K. Lark, unpublished), it can be seen that the D haplotype results in a wider but shorter radius. Similarly, haplotypes specifically regulate both the length of the snout (skull) and the width of the humerus [15
]; or the size of the pelvis and the dimensions of limb bones [26
]. In each case, a particular haplotype is regulating multiple aspects of the skeletal anatomy. Either a single gene informs several skeletal phenotypes (e.g. a signal interacting with different receptors) or several linked genes within the haplotype perform that function. Interestingly, the phenotypic characters involved are aspects of functional morphology involved in the trade-off between power and speed.
Figure 4 A QTL on CFA 12 in the Portuguese water dog regulates both the length (upper panel) and the width (lower panel) of limb bones (K. Chase and K.G. Lark, unpublished). The values shown are residuals of radius length after correction for overall size. Individual (more ...)
The observed patterns of variation in skeletal metrics might reflect the ontogenetic transition from the juvenile to the adult state [31
]. In mammals, newborns require anatomical specializations that result in relative uniformity of newborn body shape [34
]. Adult behaviors associated with feeding, locomotion, reproduction and sociality require changes in shape and proportion of the skeletal system during postnatal growth in most or all species of mammals [36
]. Genetic components that regulate the sets of inversely correlated characters of the shape PCs could account for much of this transformation. For example, appropriate temporal activation of different genes could produce the short, broad face and limbs of newborn dogs on the one hand, followed by the development of the relatively longer and narrower face and more gracile limbs of adults on the other. This suggestion is consistent with analyses of skeletal dimensions in canids [33
], which indicate that allometry (relative scaling of shape metrics) among adults of different species is often nearly identical to the shapes encountered during the course of postnatal growth.
The expression of multiple forms of phenotypic variation associated with single QTLs can explain in large part how it has been possible to select so rapidly for breeds with different functional morphologies [38
]. Such an explanation requires that the genes involved are ancient and probably central to a network that regulates many different aspects of postnatal growth. The fact that the same PCs are found in the fox provides an opportunity to determine if this is true. The fox is an outgroup for modern canids and its phylogenetic lineage has been separated from that of the dog for about 10 million years [29
]. Fox populations exhibit similar PCs to the dog [28
], and at least one population has all of the prerequisite pedigree information to identify QTLs by establishing associations between genetic markers and PC phenotypes [28
]. This process is currently underway (L. Trut et al.
, unpublished) and preliminary information has demonstrated that in the fox at least one trade-off axis of variation (shape PC) is regulated by a particular QTL haplotype.
An intriguing aspect of the dog vs fox comparison currently underway is that PC 21, derived from a matrix using 21 limb-bone traits, is common to both species. This PC involves a trade-off between the humerus and tibia on the one hand and the radius on the other and accounts for a small amount of skeletal variation (~0.2% of the variation in the fore and hind limbs). It is heritable (~40%) in both species and associated significantly with a dog QTL on autosome CFA 21 (see ). Why has the relationship between these skeletal elements not become fixed? How has an axis of variation, involving such a small change, survived?
Opposing selections might have maintained the variation during the intervening 10 million years despite bottlenecks that occurred during domestication and breed selection [42
]. Alternatively, the variation might involve a hypermutable gene that has maintained multiple alleles in the population. Fondon and Garner [43
] have proposed that repeat sequences within genes could produce the hypervariability required for the rapid evolution of morphology observed in dogs. They argue that simple sequence repeat (SSR) expansion and contraction occurring in coding sequences of developmental genes [44
] can better explain the diverse morphological types that are observed than can mutations such as single nucleotide polymorphisms (SNPs). Such SSR mutations could occur at rates up to 100 000 times greater than those characterizing SNP creation [44
]. This model is supported by their findings that many developmental genes in dogs carry a wide array of distinct alleles for simple sequence repeats within coding sequences [43
], some of which correlate with changes in morphology.
Such ‘slippery’ genes, where mutations can increase or decrease the length rapidly, could maintain allele variation over the millennia that separate the fox and the dog. Isolation of the genes responsible for the effects of QTLs should distinguish between these hypotheses and between hypotheses that explain the multiple action of shape QTLs: does one gene regulate multiple processes or are sets of closely linked, coselected genes responsible for the trade-offs associated with shape QTLs? Such questions can be resolved with the identification of the relevant gene(s), within a QTL, that inform the phenotype.