Genetic relationships in Spanish dog breeds. I. The analysis of morphological characters

Summary - The relationships between 10 Spanish dog breeds have been studied using qualitative and quantitative analyses of data from 32 morphological characters. The average distance between breeds, measured as a morphological index, has a value of 4.228 (! 0.681), with extreme values of 1.732 between Mastin del Pirineo and Mastin Espanol, and of 5.099 for the Gos d’Atura - Sabueso Espanol pair. The morphological phylogeny obtained in this study confirms the classifications made previously by means of dental, cranial, historical and behavioral comparative criteria. The results suggest the formation of 2 large clusters; one formed by the breeds belonging to the ancestral trunks Canis fa7rciliaris intermedius and Canis familiaris inostranzewi, and the other which includes the members of the Canis familiaris leineri and Canis familiaris metris-optirrtae trunks. Spanish dog breeds / genetic distance / morphological character / dendrogram /


INTRODUCTION
Archaeological studies show the existence of differences within populations of prehistoric dogs in the same area. These studies also show that there were already distinguishable and separated classes of dogs about 5 000 years ago (Villemont et al, 1970).
Two main factors have determined the differentiation of canine breeds: natural selection in the environment and conscious selection by man. The length of time from prehistoric times to the present and the number of generations elapsed explain the proliferation of canine breeds. Added to this has been the modern tendency of selective breeding to produce specialist and distinguishable breeds, with strict definitions of desirable and undesirable traits for each breed.
Man first began to influence the classes of canines when he began to adapt them to his needs. Sheep farming, extensive throughout Eurasia, created the need for gentle, intelligent animals which would respond to orders from the shepherd and help manage the flock. Dogs were adapted for defence: here the desired traits were fierceness, toughness and suspicion of strangers. Dogs were also used for hunting: some would have to be very fast to catch their prey, others would track and flush the prey and others would retrieve the dead prey. Each had a specialist task. Finally, a general category of dogs served for defence, for company or merely for decoration.
The first known classification of dogs dates from 1486 and is found in the St Albar!s' Book, attributed to Juliana Barnes, prioress of the convent of Sopwell, England (Peters, 1969). But the systematic classification of different dog breeds began to have greater importance at the end of the 19th century with the creation of the Kennel Clubs in England and North America.
Despite the huge difficulties involved in the reconstruction of the phylogenies of the more than 400 dog breeds currently recognized, the systematic classification into groups, as closely related as possible, as well as the search for their phylogenic relationships has been an uninterrupted task. There have been studies based on archaeological findings (Olsen and Olsen, 1977;Clutton-Brock, 1984), historical studies (Gomez-Toldra, 1985), cranial, dental and skeletal morphology (Clutton-Brock et al, 1976;Wayne, 1986), comparative studies of behaviour (Scott, 1968), and immunological and electrophoretic studies of proteins and blood enzymes (Leone and Anthony, 1966;Tanabe et al, 1974).
Although part of the variation observed among morphological traits may have an environmental component, in general, the heritability values for morphological traits are relatively high. The differences observed among breeds therefore should be good indicators of the genetic relationships among them. So far, however, no studies have been published on the genetic relationships between Spanish dog breeds from the analyses of morphological characters. Since statistical methods and computing packages are available to perform such analyses (Felsenstein, 1986;Swofford, 1991), the present paper is a contribution to the study of the genetic relationships between Spanish canids from qualitative and quantitative analyses of data on morphological characters.

Breeds studied
We have studied 9 Spanish dog breeds recognized by the Federation Cynologique Internationale (FCI): Gos d'Atura, Mastin del Pirineo, Mastin Espanol, Perdiguero de Burgos, Galgo Espanol, Sabueso Espanol, Ca de Bestiar, Podenco Ibicenco and Podenco Canario, and a tenth breed not yet recognized, Podenco Ib6rico. The geographical distribution of the original breeds is shown in figure 1. There are several existing hypotheses about their origin (Jordana et al, 1990), which we summarize in the following way: Gos d'Atura (Catalonian Sheepdog) or Perro de Pastor Catalin Andreu (1984) points out that the Romans took and ancient Shepherd dog on their campaigns, which could have been the Bergamasco. This dog was adapted to the different climatic environments and types of shepherding, and was the basis of a large number of breeds existing today in Central Europe. Gomez-Toldra (1985) and Delalix (1986) agree with the opinion of the Roman origin of the Gos d'Atura breed, and placed the origin of the Bergamasco in the Polish Shepherd dogs, which might have descended from the old Eastern Shepherds.

Mastin Espanol and,-Mastin del Pirinea. (Spanish Mastiff and Pyrenean Mastiff)
These are breeds included in the &dquo;ortognated moloses&dquo; which seem to descend from the legendary Mastiff of Tibet (in central Asia). These dogs are supposed to have reached Spain by 2 routes: the Central European route and via the Mediterranean (Esquir6, 1982).

Sabueso Espanol (Spanish Bloodhound)
Several authors (Villemont et al, 1970;Gondrexon and Browne, 1982;Rousselet-Blanc, 1983;Gomez-Toldra, 1985) have attributed a Celtic origin to the Bloodhounds. Most of the European Bloodhound breeds seem to descend from the Saint Hubert, a modern-day Belgian breed, the direct descendant of the Segusius of the Celts and the Gauls, which the Greek historian Arrian of Nicomedia talks about in his Cinegetics (Villemont et al, 1970;Rousselet-Blanc, 1983).
Ca de Bestiar (Balearic Sheepdog): also called Perro de Pastor

Mallorquin and Ca Garriguer
The FCI includes this breed in the second group, within the molosoid breeds, together with the Boxer and the Dogo among others. Several authors (Guasp, 1982;Sotillo and Serrano, 1985;Delalix, 1986) agree that the origin of this breed seems to be the result of crossing between Podencos Ibicencos, Perdigueros (Ca NIe) and Rousselet-Blanc, 1983;G6mez-Toldrh, 1985) and that it was brought to Ibiza by the Phoenicians (Pugnetti, 1981;Maza, 1982;Delalix, 1986), even though other hypotheses state that it arrived much later, with the Moslems, at the same time as the Galgo (Villemont et al, 1970;Rousselet-Blanc, 1983).
Podenco Canario (Canary Hound) Certain hypotheses (Delalix, 1986) suppose that this hunter came from Egypt and that it was taken to the Canary Islands, probably by the Phoenicians, Greeks, Carthaginians or even by the Egyptians, but it is possible that Majorcan monks, forced to emigrate to these islands by the Vatican, introduced these dogs (Anonymous, 1982).

Qualitative and quantitative analyses
In an ideal specimen of each of 10 Spanish dog breeds, a total of 32 characters have been studied. Some of the characters were established by the official standards of the breed while the other characters came from data of a review (Avila, 1982; I Symposium Nacional de las Razas Caninas Espanolas, 1982;Gomez-Toldra, 1985 ;Sotillo and Serrano, 1985;Delalix, 1986). The numbers were assigned to each state of the different characters in an arbitrary manner. These numbers did not represent any specific weighting of the state. The number of states for each character was established depending upon the number of distinguishable phenotypic classes. The characters used and their states are shown in table I.

Qualitative analysis
For the qualitative analysis, discrete characters were recoded into a series of (0, 1) 2-state characters, denoting absence or presence of the character, respectively. Continuous quantitative characters (D and E characters in table I) may be divided into a small number of classes, each representing one of the states of the character in the data matrix. For recoding a character with several states we have used the following transformations (Sneath and Soka, 1973) : and so on. The original and recoded matrices of morphological resemblances are shown in tables II and III respectively.
The MIX program of the phylogeny inference package (PHYLIP) (Felsenstein, 1986) was used to construct the dendogram of Spanish breeds of dogs from qualitative data of morphological characters. This analysis is based upon the &dquo;parsimony&dquo; principle, and the criterion is to find the tree requiring the minimum number of changes. Two dendrograms can be obtained: the first, using Wagner parsimony (Farris, 1970), is used when the ancestral state of the character is unknown; the second, using Camin and Sokal's method (1965), presupposes the knowledge of the ancestrality. Several possible criteria have been proposed to infer the ancestral state of the character: the fossil record, the frequency criterion and outgroup analysis (Avise, 1983). Each of these criteria has been seriously and justifiably criticized (Stevens, 1980), although it has been recognized that the outgroup analysis provides a particulary compelling rationale for estimating the character state polarity (in our case, for example, the wolf, Canis lupus). We have chosen, however, the frequency criterion (the state of the character appearing most frequently in the group being examined) in order to make comparisons between these dendrograms and those obtained in a second study (Jordana et al, 1992) on the phylogenetic relationships among Spanish dog breeds derived from the analysis of biochemical polymorphisms. The reason for choosing the frequency criterion was the lack of adequate literature on electrophoretic results of any species of wolf candidate to be used as an outgroup. The tree generated by Wagner parsimony is unrooted, so we chose arbitrarily the Galgo Espanol breed as an outgroup in order to make comparisons with other dendrograms. An evolutionary tree generated by a parsimony criterion was also computed using the phylogenetic analysis using parsimony computer package (PAUP) (Swof ford, 1991). The resulting tree was rooted and the midpoint rooting method (Farris, 1972) was chosen to give the tree an evolutionary direction. The PAUP package allows us also to compute the confidence limits of the topology by means of a bootstrap analysis (Efron, 1979), adapted to the inference of phylogenies (Felsenstein, 1985). One hundred bootstrap replicates were made, and a consensus tree was obtained based upon the majority-rule method (Margush and McMorris, 1981). The minimum frequency of the bootstrap replicates-in which a group-issupported in order to be included in the bootstrap consensus tree was set to 50 (Conlevel = 50).

Quantitative analysis
For the quantitative analysis of morphological characters, qualitative data were transformed and introduced in the form of a matrix of distances. An Euclidean distance (Sneath and Sokal, 1973) was used to estimate distances between populations, under the assumption of independence between characters.
where: d!!,!! = value of the distance between the j and the breed k. The distance ranges from 0 to fl, where n is the number of traits; (J!,j &mdash; Xi k ) = alternative values (0, 1) for the differences between j and k breeds within the character i.
The mean character difference (MCD) proposed by Cain and Harrison (1958) was also calculated as a measure of taxonomic resemblance. MCD varies between 0 and 1.
Fitch and Margoliash's method (1967) was used to find the unrooted tree that would best adapt to the matrix (FITCH program in PHYLIP package). The tree that minimizes the sum of squares SS was searched for by means of the following expression: where: D jk = observed distance between populations j and k; d jk = expected distance between populations j and k, computed as the addition of tree segment lengths, from population j to population k (patristic distance).
Alternatively, a rooted tree was computed by applying the KITSCH program (PHYLIP package). In this method, a tree similar to that generated by the cluster analysis was computed and subsequently the topology of the tree was altered in order to improve its goodness-of-fit. By assuming: a), that the expected rates of change are constant through all lines; b), that all the subpopulations are contemporary; and c), that the phenotypes behave as an evolutionary clock, this method can be regarded as an estimator of the phylogeny (Felsenstein, 1984, 198G).

Qualitative analysis
The dendrograms resulting from the application of Wagner parsimony and Camin and Sokal's methods are shown in figures 2 and 3 respectively. Two large groups can be observed in each tree. One of the groups is formed by 4 breeds: Mastin del Pirineo, Mastin Espanol, Sabueso Espanol and Perdiguero de Burgos; the other group includes Podenco Ibicenco, Podenco Canario, Podenco Ib6rico and Galgo Espanol. In the dendrogram resulting from Wagner parsimony, the breeds Ca de Bestiar and Gos d'Atura are halfway between the 2 large groups, even though Gos d'Atura is nearer the greyhound group (Podencos and Galgo) and Ca de Bestiar is nearer the other group. The closeness of Gos d'Atura and Ca de Bestiar breeds to one group or the other is more evident in the three resulting from the application of Camin and Sokal's method. Gos d'Atura is placed halfway between 2 subgroups formed by Podenco Ibicenco-Podenco Canario and Podenco lb6rico-Galgo Espanol breeds. The Ca de Bestiar breed is more closely related to the Mastiffs than to the subgroup formed by Sabueso Espafiol and Perdiguero de Burgos. Both topologies are possible, even though the tree obtained by applying Wagner parsimony needed only 96 steps to rearrange the characters and to obtain the most parsimonious tree, while for the tree generated by Camin and Sokal's method, 101 steps were needed. This difference in the number of steps, however, probably reflects the differences in the assumptions of the kinds of changes used in both methods (Felsenstein, 1986), and consequently cannot be considered as a definitive criterion to infer the true relationships. Figure 4 shows a dendrogram of the Spanish dog breeds estimated according to the parsimony and midpoint rooting criteria (PAUP package). This dendrogram again shows the 2 groups described above. Branch and internodal distances are proportional to the number of character-stage changes required. The total length was 85 (versus 96 found in Wagner parsimony), and the consistency index (a measure of the homoplasy) was 0.671. Included within parentheses are the values of the number of replicates from the bootstrap analysis (loosely, the width of the confidence interval).

Quantitative analysis
The results of the morphological distance indexes between Spanish dog breeds are shown in The trees obtained using FITCH and KITSCH programs are shown in figures 5 and 6. The dendrograms obtained by the FITCH program are unrooted, so we arbitrarily used the Galgo Espafiol breed as an outgroup. Two hundred and twentysix possible trees were examined. Figure 5 shows the tree that best adjusts to the matrix of data. The sum of squares had a value of 0.183, whereas the average percent standard deviation was 4.5G%. In the tree in figure 5, the 2 groups previously described are again observed. The Greyhound cluster (Podenco Ibicenco, Podenco Canario, Podenco lb6rico and Galgo Espanol) additionally contains the Gos d'Atura breed. The Ca de Bestiar breed remains in an intermediate position, slightly closer to the Greyhound group.
In the resulting tree from the application of the KITSCH program, the 2 large clusters were observed again, Gos d'Atura and Ca de Bestiar being included in the greyhound group, even though an unresolved trichotomy is presented between Galgo Espauol, Gos d'Atura and Ca de Bestiar breeds. The sum of squares had a value of 0.248 and the average percent standard deviation was 5.31%.

DISCUSSION
In examining all the topologies of the trees resulting from the analysis of morphological characters, it is possible to verify some stable relationships among different groups of breeds. Sabueso Espanol and Perdiguero de Burgos form a separate cluster from Mastin Espanol and Mastin del Pirineo breeds. The last 2 clusters, in their turn, are related and form a new cluster. The bootstrap analysis (figure 4) confirms this grouping (79% of the bootstrap replicates). Podenco Ibicenco, Podenco Canario, Podenco Ib6rico and Galgo Espanol breeds are related in all trees. The bootstrap analysis, however, failed to confirm the relationship between Galgo