Cranial morphology of Balkan and West Asian livestock guardian dogs

Abstract

Several large “shepherd” or livestock guardian dog (LGD) breeds were historically selectively bred to protect sheep and goat flocks in the Balkans, Anatolia, and the Caucasus regions. Although these breeds exhibit similar behavior, their morphology is different. Yet, the fine characterization of the phenotypic differences remains to be analyzed. The aim of this study is to characterize cranial morphology in the specific Balkan and West Asian LGD breeds. We use a 3D geometric morphometric in order to assess morphological differences regarding both shape and size between LGD breeds and compare this phenotypic diversity to close relative wild canids. Our results indicate that Balkan and Anatolian LGDs form a distinct cluster within a relatively large dog cranial size and shape diversity. Most LGDs display a cranial morphology that could be described as intermediate to the mastiff breeds and large herding dogs, except for the Romanian Mioritic shepherd which has a more brachycephalic cranium strongly resembling the bully‐type dog cranial morphotype. Although often considered to represent an ancient type of dog, the Balkan–West Asian LGDs are clearly distinguishable from wolves, dingoes, and most other primitive and spitz‐type dogs and this group displays a surprising cranial diversity.

In this study, skull morphological diversity of Livestock Guardian Dogs was examined and contributed information to the established literature. The Balkan‐West Asian Livestock Guardian Dogs are clearly distinguishable from wolves, dingoes, and most other primitive and spitz‐type dogs and this group displays a surprising cranial diversity.

1. INTRODUCTION

Today, there are hundreds of dog breed worldwide, displaying an incredible diversity of morphotypes (Schoenebeck & Ostrander, 2013). While dogs show little genetic variation, there is a wide variety of skull shape and size that has been previously described (Georgevsky et al., 2014). This morphological diversity is usually characterized using the cephalic index (CI) classification, a simple method used to differentiate dogs according to cranial proportions (Bognár et al., 2021). Dogs with short skull structures such as bulldog and boxer breeds are classified as brachycephalic while breeds characterized by long and narrow‐nosed skulls, such as greyhounds and collies, are classified as dolichocephalic breeds (Helton, 2009). Nevertheless, the fine characterization of the cranial morphology of many breeds and morphotypes (i.e., groups of dog breeds with common morphological characteristics, hereby based on the ICF classification) has not yet been undertaken. In addition to their functional importance in food intake and vocalization, cranial shape and size have also been related to brain functions and behaviors (Gácsi et al., 2009; McGreevy et al., 2004), although there are not direct correlations between CI and dog cognition (Selba et al., 2021), and the body mass (directly related to cranium size) seems to play a major role in this relationship (Stone et al., 2016). Even if only a partial correlation, this relationship between skull size/shape and cognitive abilities is not trivial as dog clubs or councils usually divide dog breeds into groups according to their morphometric appearance, but as well as their daily routines and lifestyles. For example, livestock guardian dogs (LGDs) are breeds used to protect livestock from various predators and protect farmers from potential thieves. While herd dogs are used to “manage” herds, LGDs were artificially selected for the purpose of “protecting” the herd from external threats (Rigg, 2001). As such, they are characterized by a large body mass and a powerful musculatures compared to other dog breeds and possess physical characteristics allowing them to resist difficult climatic conditions and maintain high activities without food intake for a long time (Rigg, 2001).

LGDs originate from Asia and Europe (Coppinger & Coppinger, 2007; Stone et al., 2008) with ancestors dating back 6000 years in the Balkan region (Rigg, 2001). The Balkans is the geographic and cultural region located in the southeastern part of the European continent and in the northwest of Anatolia. In this region, animal husbandry is widespread and predominantly consists of transhumant pastoralism. This mode of husbandry involves long‐distance movement of large flocks of sheep and goats across mountainous terrain, and long pasturing periods, often far away from populated settlements. The increased exposure of livestock to large predators such as wolves and bears during transhumance has created a demand for breeds of dog to protect them. As such, many breeds or types of LGDs have been artificiality selected within the broader Balkan and West Asian regions, which both share several basic similar physical characteristics (e.g., basis of geography, disparate climatic conditions). These breeds were therefore selected largely for their behavioral characteristics (e.g., obedience, stress management), in addition to some morphological elements such as large body size and robustness. This selection, based on both cognitive and morphological aspects, results in breeds that can vary significantly in terms of external phenotypes but are consistent in function, also known as a “land‐race.” For example, the Bernese mountain dog, an LGD of Swiss origin, is a good example of breed selected for its resistance to cold weather and endurance, while the Kangal shepherd dog, a traditional herd guard dog kept with sheep herds originating in Turkey (Atasoy et al., 2014; Yilmaz, 2007), presents a lighter color pattern with shorter hair and lighter build typical of warmer climates.

Despite observable differences, the morphological characteristics of these breeds are still largely unknown, and to our knowledge there is no study describing the skull differences and similarities of these breeds just as there is no comparison with other dog breeds or morphotypes. In this study, we propose to use 3D geometric morphometrics to characterize the skull shape and size of different LGDs. We aim to assess how diverse this group is compared to other dogs. Similarly, we aim to know if their morphology is closer to other dog morphotypes or to close relative wild species (i.e., Canis lupus and Canis dingo). We assume that LGDs present an underestimated morphological diversity compared to other morphotypes of dogs, and that they possess characteristics close to wild species.

2. METHODS

2.1. Materials

To characterize the morphology of LGDs we used a 3D geometric morphometric approach, using 3D models of the crania of dogs and other canids from the collections of the Osteoarchaeology Research Center, Istanbul or obtained via veterinary practices (Agricultural University of Tirana; Iasi University of Life Sciences; The University of Sydney) (Data S1). Our sampling includes 168 crania belonging to well‐identified breeds or dog morphotypes (e.g., “Terrier” breeds) as well as 35 crania of closely related wild species, including wild wolves (Canis lupus) and dingoes (C. dingo). The sampled dingoes are all from Fraser Island (Queensland), while the wolf sample is mixed, being predominantly from northern North America with some specimens from Eurasia. All specimens selected were adults including when possible both females and males. Among the dog breeds, we sampled seven LGDs (i.e., Bernese Mountain, Bucovina shepherd, Caucasian shepherd, Ilirian shepherd, Kangal shepherd, Aksaray Malaklısı, and Romanian Mioritic shepherd, see Figure 1) and nine morphotypes of other dog breeds to encompass the phenotypic cranial diversity focusing on mesocephalic dogs. We selected these breeds because we have information on their origin: the Bernese mountain is of Swiss origin. Bucovina shepherd and Romanian Mioritic shepherd are dog breeds of Romanian originally bred to guard herds of sheep and cattle. The Caucasian shepherd dog is a large guard dog native to countries in the Caucasus region (Yılmaz et al., 2015). Ilirian shepherd is a Kosovo‐specific LGD originating from the Balkan region (Jashari et al., 2022). The Kangal shepherd dog and Aksaray Malaklısı dog were traditional herd guard dogs kept with sheep herds originating in Turkey (Atasoy et al., 2014; Yilmaz, 2007) (Figure 1). Crania were digitized with the use of a Creaform Handyscan 700 or Shining Einscan SP with a camera resolution of 1.3 megapixels. This sample was completed by Siemens Somatom Scope vc30b CT scans acquired from veterinary care centers.

FIGURE 1.

Example of the cranium shape and size for the different livestock guardian dog breeds investigated in this study and their estimated geographic origins. Asterisk indicates that evidence of the geographic origin of this breed are very scattered and disputed.

2.2. 3D geometric morphometrics

Landmarking was performed using the program Stratovan Checkpoint (Stratovan Corporation), using a schema of 64 landmarks distributed on both sides of the cranium covering important anatomical features and regions (Figure 2). For the purpose of analysis, samples were classified into morphotypes on the basis of their general body form, which broadly (but not strictly) corresponds to traditional functional roles (Table 1) following FCI guidelines (https://www.fci.be/en/). It is important to note that most of the “Pinscher and Schnauzer” group here is comprised of the Molossoid subtype, which includes many mastiffs and other large dogs. Breeds with extreme morphologies such as highly brachycephalic bulldogs and “toy dog” morphotype were excluded from the sampling because no LGDs fit these morphotypes.

FIGURE 2.

The 64 anatomical landmarks used in this study used to encompass the cranial disparity of dogs and close relatives.

TABLE 1.

Study specimens and breeds.

Breed/morphotype	N specimens	N breeds
LGDs
Aksaray Malaklısı	2
Bernese mountain	1
Bucovina shepherd	1
Caucasian shepherd	1
Ilirian shepherd	17
Kangal shepherd	13
Romanian Mioritic shepherd	1
Other dogs
Gundog	38	9
Hound	3	3
Pinscher and Schnauzer	28	13
Primitive‐Spitz	22	8
Sheepdogs and Cattledogs	22	4
Sighthound	7	3
Terrier	10	5
Toy	2	1
Wild canids
Dingo (Canis dingo/Canis lupus dingo)	19
Wolf (C. lupus)	17
Totally	204

Abbreviation: LGD, livestock guardian dog.

Geometric morphometric analysis was performed using MorphoJ (Klingenberg, 2011). The landmark data were first transformed using a standard Procrustes fit to remove differences in position, orientation, and scale. Principal component analysis (PCA) was used to explore the morphology of the Balkan–West Asian LGDs and how it relates to other dogs and wolves, with the morphological trajectories of the most important clines visualized with wireframe plots. A regression of shape against the log centroid size (CS) of the cranium was then used to investigate the role of overall size in cranial form and to examine the importance of allometric changes for LGD compared to other dogs and wild relatives.

For visualization and discussion purposes, dogs were categorized into brachycephalic, mesocephalic, or dolichocephalic, based on the CI (CI = skull width/skull length × 100). In recent studies, those with low CI (<50) were classified as dolichocephalic, those with medium (50–60) as mesocephalic and those with high (>60) as brachycephalic (Bognár et al., 2021).

2.3. Discriminant function analyses

Discriminant function analyses (DFAs) were performed in order to highlight the morphological features that better characterized the LGDs compared to other dogs and wild canids (wolf and dingo), and to investigate differences between the two LGDs for which several specimens were available (Ilirian shepherd and Kangal shepherd). DFA uses pairwise comparisons to identify movements of specific landmarks and the variation that most strongly differentiate two given populations. Post‐hoc leave‐one‐out cross‐validation was performed with DFA tests to assess the reliability of their results, which can be affected by small sample sizes. The DFA and CV tests are complemented by Procrustes and Mahalanobis distances between groups. These are measures of Euclidean distance in the absolute shape (landmark configuration) deviation between groups and the distance between groups relative to within‐group variation, respectively. These measures provide a sense of the relative degree of differentiation between study populations, essential to identify the distinctiveness of the LGDs among other dogs and wild canids.

3. RESULTS

3.1. Principal components analysis—cranial form

In the principal component analysis, the first two components together represent approximately 54% of the total variation (Figure 3). We found that 11.49% of this form variability is predicted by the CS (p < 0.0001), indicating a modest allometric component as expected given the diversity of breed sizes in the sample. The first axis of the PCA describes a cline between brachycephalic and dolichocephalic dogs, with boxer‐like breeds and sighthound‐like breeds lying at the respective extremes of the axis. A significant but minor (4.25%) proportion of PC1‐related form change is predicted by CS (p = 0.0031). Form as PC2 increases is associated with decreasing overall cranial size, as change along this component is very strongly and significantly predicted by the CS (73.60%, p < 0.0001). Shape changes defined by PC2 are mostly defined by the braincase shape. In particular, the braincase is proportionally rounder in the maximum values of the PC2, as highlighted in the wireframe plots (Figure 4).

FIGURE 3.

Principal component analysis plot of cranial form (including cranial shape) corresponding to PC1 and PC2, with convex hulls representing the area occupied by the gun, herding, and mastiff dog morphotypes.

FIGURE 4.

Wireframe plots describing cranial shape changes between the minimum and maximum values of PC1 and PC2. Gray represents the mean position of each landmark; black represents landmark movement from the mean shape in the direction of each PC (at 0.1 Procrustes distance).

Most LGDs can be described as having mesocephalic crania, and fall at the intersection of the regions of morphospace occupied by mastiff‐like breeds (various generic and breed individuals), herding morphotypes (mostly German shepherds), and to a lesser extent the gun dog‐like breeds (mostly retrievers). The only exception is the Romanian Mioritic shepherd, which has a more brachycephalic cranium than other dogs, and occupied the same morphological space as the mastiff morphotype dogs (Figure 5). The two breeds with large samples, the Kangal shepherd and Ilirian shepherd, show closely overlapping ranges. Most LGDs are distributed outside the concentrated range of the wolves and dingoes, although some of the primitive and spitz types such as huskies are not far removed. As large dogs, the LGDs are distributed toward the negative part of the PC2 axis—particularly the Bucovina shepherd and Aksaray Malaklısı.

FIGURE 5.

Principal component analysis plot of cranial form (including cranial shape) corresponding to PC1 and PC2, with convex hulls representing the area occupied by the gun, herding and mastiff dog morphotypes.

Our analyses on cranial CS confirm that the crania of the Balkan–West Asian LGDs are the largest breeds in our sample. This was expected considering their overall body size and mass, although these vary considerably among and within the breeds (Table 2). In particular, the two Aksaray Malaklısı individuals are the largest canids sampled in this study, and both their CS scores (669.66 and 691.41, respectively) exceed the maximum scores for any other group. The Romanian Mioritic shepherd is the smallest LGD sampled, but otherwise, all groups have mean CS scores ranging between the Doberman and St. Bernard breeds, with CS scores close to those found for wolves.

TABLE 2.

Centroid sizes estimated for LGDs, selected other medium to large dogs and wild relatives. SD were computed when n ≥ 5.

Breed/morphotype	n	Minimum	Maximum	Mean
LGDs
Aksaray Malaklısı	2	669.66	691.41	680.53 ± 15.38
Bucovina shepherd	1	619.82	619.82	619.82
Caucasian shepherd	1	601.21	601.21	601.21
Ilirian shepherd	17	506.43	624.66	564.49 ± 31.87
Kangal shepherd	13	482.65	644.05	574.59 ± 48.73
Romanian Mioritic shepherd	1	501.50	501.50	501.50
Other
Doberman	6	493.19	583.94	544.67 ± 32.41
German shepherd	12	485.85	592.92	547.05 ± 34.76
Golden retriever	18	429.24	556.93	506.80 ± 29.30
St. Bernard	5	550.88	664.07	604.66 ± 44.68
Dingo	19	426.98	506.43	475.92 ± 18.81
Wolf	17	516.904	647.82	586.43 ± 43

Abbreviation: LGD, livestock guardian dog.

3.2. Discriminant function analysis

Our results of pairwise DFA and cross‐validation comparisons of LGDs with other dog morphotypes highlighted a similar morphology in the PCAs presented in Tables 3 and 4, respectively. Procrustes and Mahalanobis distances with significance are presented in Table 5. The rates of inter‐group misclassification are higher in the cross‐validation tests, but the overall patterns are mostly very similar between the two methods. LGDs are most frequently misclassified as belonging to the pinscher and schnauzer morphotype group (8.3% in DFA and 25% in cross‐validation), followed by the sheepdog and cattledog group (11.1% and 22.2%) and gundog morphotype group (2.8% and 19.4%). LGDs show a lower rate of misclassification to the primitive‐spitz morphotype group (0% and 11.1% misclassification), 0% rate of misclassification in DFA and cross‐validation with the dingoes, and 0% misclassification in DFA and just 5.5% in cross‐validation for wolves.

TABLE 3.

Results of DFA comparisons.

Group 1	Group 2	n group 1 allocated to 2	n group 2 allocated to 1
LGD	Gundog	1 (2.8%)	1 (2.6%)
LGD	Pinscher and Schnauzer	3 (8.3%)	0
LGD	Sheepdog and Cattledog	4 (11.1%)	1 (5.3%)
LGD	Primitive‐Spitz	0	0
LGD	Dingo	0	0
LGD	Wolf	0	0
Ilirian shepherd	Kangal shepherd	0	1 (6.6%)

Abbreviations: DFA, Discriminant function analyses; LGD, livestock guardian dog.

TABLE 4.

Results of cross‐validation comparisons.

Group 1	Group 2	n Group 1 allocated to 2	n Group 2 allocated to 1
LGD	Pinscher and Schnauzer	9 (25%)	12 (34.3%)
LGD	Gundog	7 (19.4%)	8 (21.1%)
LGD	Sheepdog and Cattledog Dog	8 (22.2%)	6 (31.6%)
LGD	Primitive‐Spitz	4 (11.1%)	6 (35.3%)
LGD	Dingo	0	0
LGD	Wolf	2 (5.5%)	1 (5.9%)
Ilirian shepherd	Kangal shepherd	4 (23.5%)	3 (20%)

Abbreviation: LGD, livestock guardian dog.

TABLE 5.

Procrustes and Mahalanobis distances and test significances to three significant figures.

Group 1	Group 2	Procrustes distance	p‐Value	Mahalanobis distance	T‐square statistic
LGD	Pinscher and Schnauzer	0.0492	<0.0001	3.437	0.0010
LGD	Gundog	0.0313	<0.0001	4.331	<0.0001
LGD	Sheepdog and Cattledog	0.0282	<0.0001	3.252	<0.0001
LGD	Primitive‐Spitz	0.0334	<0.0001	3.982	<0.0001
LGD	Dingo	0.0490	<0.0001	10.152	<0.0001
LGD	Wolf	0.0505	<0.0001	5.820	<0.0001
Ilirian shepherd	Kangal shepherd	0.0296	<0.0001	2.798	0.0010

Abbreviation: LGD, livestock guardian dog.

These patterns are generally reflected in the Procrustes and Mahalanobis distances, but it is interesting to note the relatively large Procrustes distance between LGDs and the pinscher and schnauzer group compared to the sheepdog and cattledog group. This may be driven by the more divergent or outlying brachycephalic mastiff specimens in the pinscher and schnauzer group evident in PCA (Figure 3), and this difference is not as strongly evident in the Mahalanobis distances, which are highly comparable between LGDs and both of these groups. These Mahalanobis distances also indicate stronger differences between LGDs and dingoes than between LGDs and wolves.

In comparing the well‐sampled Ilirian shepherd and Kangal shepherd LGD breeds, the rates of misclassification are high across both test methods, with approximately 20% to 25% of cases of each breed being misclassified as the other. Examination of the landmark movements identified as contributing to the difference between the two in DFA for suggested that the pterion (Landmarks 44 and 45) is slightly more posteriorly in the Kangal shepherd than it is in Ilirian shepherd breed. Accordingly the anterior (postorbital) portion of the calvarium is relatively narrowed and begins to broaden more posteriorly in Kangal shepherd dogs. Despite their high degree of distributional overlap in PCA form and shape‐space, the Ilirian shepherd and Kangal shepherd breeds are distinguishable from one another in cranial morphology with significant results for both Procrustes and Mahalanobis distances (Table 5).

3.3. Within‐group variability

Procrustes variances (i.e., intra‐group morphological variability) are presented in Table 6. Here, the dingo sample provides an example of a very conservative population. Indeed, all dingo samples are from the same location, and dingoes have limited morphological variability compared to dogs. In their distribution in the PCA, most Balkan–West Asian LGDs cluster within a relatively limited region of form and shape‐space, with the exception of the Romanian Mioritic shepherd breed. Accordingly, their variance score is lower than those of the gundogs, mastiffs, and primitive‐spitz morphotype groups, all of which consist of more morphologically divergent breeds, but higher than the herding group. The Kangal shepherd sample was found to be more variable than the Ilirian shepherd breed.

TABLE 6.

Procrustes variance in LGDs and other canids to three significant figures.

Group	n	Procrustes variance
LGD	36	0.00238
Ilirian shepherd	17	0.00183
Kangal shepherd	13	0.00237
Gundog	38	0.00301
Sheepdog and Cattledog	22	0.00233
Pinscher and Schnauzer	28	0.00708
Primitive‐Spitz	17	0.00299
Dingo	19	0.000701
Wolf	17	0.00161

Abbreviation: LGD, livestock guardian dog.

4. DISCUSSION

This study has identified some observations and trends of interest with regard to the cranial morphology of Balkan and Anatolian livestock guardian dogs. Although belonging to different climates and having geographic origins, LGDs are characterized by a strong resemblance in their cranial shape and size within mesocephalic dogs. This common morphology is defined by large size and robusticity of the cranium, and potentially reflects ongoing selection for the task of livestock guardianship against large predators (wolves and in some places bears) within the broader region. In addition, the geographic distribution of LGDs does not seems to be associated with skull shape variations. With this acknowledged, it is interesting that the LGDs do not show exclusive alignment to mastiff or Molossoid‐type dogs but also share similarities with members of the sheepdog and cattledog group, namely the German shepherds. The morphological proximity and distance of the dogs are similar in their behavioral characteristics. Molossoid‐type dogs are among the bravest examples of protective breeds (Starling et al., 2013). It has been reported that molossoid dogs show fewer stress symptoms than shepherd dogs against external dangers (Duranton et al., 2016) while shepherd dogs are chosen to herd livestock and watch their owners during communal activities. LGDs show behavioral similarities to this type of dog, as sheepdogs and cattledogs perform daily herd‐oriented tasks.

The largest CS recorded from studied LGDs was observed in the Aksaray Malaklısı. The closest LGD specimen to brachycephalic shape was the Romanian Mioritic shepherd. Ilirian shepherd and Kangal shepherd skull shapes were narrower and longer (closer to dolichocephalic shape) than other LGDs. LGD skull shapes were broadly closest to the pinscher and schnauzer, and sheepdog and cattledog specimens, not particularly close to the gundogs or primitive‐spitz dogs, and completely different in shape from the wolf and especially the dingo. These results were similar to those of Drake (2011) that the cranial shape of none of the modern breeds of dogs resembles the cranial shapes of adult or juvenile wolves.

We found that the cranial shape of LGDs was more similar to those of the behaviorally similar sheepdog and cattledog specimens than other breeds or species. There is a significant relationship between morphological features and behavior in dogs (Stone et al., 2016). For example, dolichocephalic breeds are known to be more efficient hunters compared to other breeds for tracking and killing game. This efficiency can be associated with the sight feature used during hunting, as the ratio of peak ganglion cell density to visual line in the area centralis had a high positive correlation with the CI (McGreevy et al., 2004). In addition, behavioral differences have been reported between dogs classified as large and small morphologically. Dogs with large body size generally exhibit behavior associated with anxiety or fear less frequently than small dogs (Martínez et al., 2011). Also, LGDs showed similar cranial morphology to one another, except for the Romanian Mioritic shepherd. The Romanian Mioritic shepherd has a more brachycephalic cranium than other LGDs. Although the Romanian Mioritic shepherd is similar to other LGDs in terms of its height at the withers (67.3 cm [Dronca et al., 2016], compared to 65.2 and 71.7 cm for the Kangal shepherd [Daskiran, 2007] and 73.0 cm for the Malaklı [Atasoy et al., 2014]), it differs from other LGDs in its hairy appearance. A more detailed study of the skeleton is required to understand the differences between LGDs more fully.

In this study, skull morphological diversity of LGDs was examined and contributed information to the established literature. LGDs showed similar skull shape variation compared to other dog morphotypes, especially dogs with similar behaviors such as the German shepherds. We suggest that geographic proximity had little effect on these shape variations. The morphological diversity among breed groups appears to be the result of genetic variation and different behavioral effects that are the result of selective breeding. In the future, we hope that studies combining genetics and morphology will shed some light on this topic.

AUTHOR CONTRIBUTIONS

Ozan Gündemir, Loukas Koungoulos, and Margot Michaud conceptualized and designed the study; Ozan Gündemir and Vedat Onar scanned the samples in Türkiye; Ozan Gündemir and Sokol Duro scanned the samples in Albania; Ozan Gündemir and Mihaela‐Claudia Spataru scanned the samples in Romania; Loukas Koungoulos scanned the samples in Australia; Vedat Onar provided supervision; Loukas Koungoulos and Margot Michaud drafted the manuscript and created the figures; Tomasz Szara provided edits and comments on the manuscript draft; All authors have approved the final version of the manuscript.

CONFLICT OF INTEREST STATEMENT

The authors have no conflict of interest to declare.

Supporting information

Data S1.

Gündemir, O. , Koungoulos, L. , Szara, T. , Duro, S. , Spataru, M.‐C. , Michaud, M. et al. (2023) Cranial morphology of Balkan and West Asian livestock guardian dogs. Journal of Anatomy, 243, 951–959. Available from: 10.1111/joa.13929

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the first author upon reasonable request.