Material and Methods

Horses—Five thousand one hundred fifty-seven Hanoverian Warmblood horses designated by the Society of Hanoverian Warmblood Breeders (Verband hannoversches Warmblutzüchter eV) to be sold at riding-horse auctions in Verden on the Aller, Germany, from 1997 to 2004 were included. Information was available for 141 stallions, 2,847 geldings, and 2,169 mares. Data for the horses were provided by the Society, and radiographic data were obtained from the horses' official veterinary records.

All horses were examined clinically and radiographically by the same veterinarian. Routine radiographic examination of the horses included dorsoproximo-palmarodistal projections (ie, upright pedal view¹⁰) of the navicular region of the forefeet, lateromedial (90°) projections of all 4 feet, and lateromedial (90°) and dorsolateral-plantaromedialoblique (45°) projections of the tarsocrural joints. Radiographs were independently evaluated by 2 experienced radiologists. Deviations from normal radiographic appearances of bony structures were recorded.

Definition of traits—Navicular bones were evaluated on the basis of their appearance in the upright pedal projections. Where applicable, lateromedial projections of the front feet or additional views that had been taken for clarification of equivocal findings were also considered. Variables recorded were the number, location, and shape of the CSs and the structure and contour of the forelimb navicular bones. Categorization of the RNB was as follows: 0 = no abnormal findings; I = few (1 to 4) short and conical CSs in the central portion of the distal border; II = several (≥ 5) short and conical CSs in the central portion of the distal border; III = few (1 to 4) elongated or deformed (ie, enlarged) CSs in the central portion of the distal border; IV = several (≥ 5) enlarged CSs; V = markedly deformed (enlarged, elongated, or branched) CSs in the central, medial, or lateral portion of the distal border; VI = alterations in the contour of the navicular bone (ie, new bone formation at the proximal or distal border or at the medial or lateral extremities of the bone); and VII = alterations in structure of the navicular bone (ie, irregular pattern of spongiosum and increased [sclerosis] or decreased [osteolysis] bone opacity). Categories were defined and the order of categories chosen in accordance with current official guidelines for interpretation of radiographic findings in horses.¹¹ Horses were assigned to one of the RNB categories according to the most distinct finding in either forelimb navicular bone.

The categories of radiographic appearance were used to define 2 quasi-linear traits (RNB_0–7 and RNB_0–3) and 5 binary traits (RNB_0/1a–e) describing the type and distribution of radiographic findings. Each category corresponded to an individual trait level of the most specific trait (ie, quasi-linear trait RNB_0–7). For definition of the second quasi-linear trait (ie, RNB_0–3), categories III to VII were combined. All-or-none traits were defined such that horses were considered affected if they were allocated to RNB category I (RNB_0/1a), category II (RNB_0/1b), categories III to VII (RNB_0/1c), or categories II to VII (RNB_0/1d) or if they had any radiographic changes involving the distal border of the navicular bones (RNB_0/1e).

Lateromedial projections of the distal portion of the limbs were used to detect horses with OFMs. Lateromedial and oblique projections of the tarsocrural joints were used to detect horses with OFTs or with DAT. Definitions of OFMs, OFTs, and DAT have been described elsewhere.^12,13 Parameter values of 1 were assigned to affected horses, and parameter values of 0 were assigned to the remainder of the horses.

Pedigree information—Pedigree data were provided by an animal ownership database.^j The 5,157 horses were descendants of 509 sires that had a mean ± SD of 10.13 ± 19.21 (range, 1 to 251) investigated offspring and of 4,145 dams with 1.24 ± 0.55 (range, 1 to 6) investigated offspring. The number of maternal grandsires was 729; of those, 257 appeared as both sires and maternal grandsires in the pedigrees. Maternal grandsires were represented by 7.05 ± 11.66 (range, 1 to 105) investigated offspring.

Model development—Models for genetic analysis were developed on the basis of results of simple and multivariate ANOVAs. The 2 quasi-linear RNB traits were analyzed in general linear models,^j and the binary traits were analyzed in generalized linear models with binomial distribution function and probit link function.^k Values of P < 0.05 were considered significant.

The following effects were tested for association with the distribution of radiographic findings: date of presumptive auction (48 auction dates with 107.4 ± 31.0 [range, 67 to 176] preselected horses/auction date), year of presumptive auction (years from 1997 to 2004; 644.6 ± 65.3 [range, 548 to 740] pre-selected horses/year), sex (141 stallions, 2,847 geldings, and 2,169 mares), age (1,090 3-year-olds, 2,685 4year-olds, 1,042 5-year-olds, and 340 horses ≥ 6 years of age), interaction between sex and age (10 levels: 49 3-year-old stallions; 92 stallions ≥ 4 years of age; 595 3-year-old, 1,534 4year-old, and 572 5-year-old geldings and 146 geldings ≥ 6 years of age; 446 3-year-old, 1,097 4-year-old, and 446 5year-old mares and 180 mares ≥ 6 years of age), and season of birth (1,807 horses born from November to March, 2,026 horses born in April, and 1,324 horses born from May to October).

The models of all radiographic appearance traits included the fixed effects that were significant in the simple and multivariate ANOVAs of RNB_0–7. Accordingly, effects that were significant in the simple and multivariate ANOVAs of OFMs, OFTs, and DAT were included in the models for the respective traits, according to the following equations:

View Expanded

Genetic analysis—Genetic variables were estimated in multivariate analysis in linear animal models via restricted maximum likelihood.^14,l Information on all traits (ie, RNB traits and OFMs, OFTs, and DAT) was available for all horses. For all horses, complete pedigree information for 3 ancestral generations was considered, resulting in a relationship matrix composed of 23,279 horses.

Additive genetic (σ²_a, cov_a) and residual (σ²_e, cov_e) variance and covariance estimates from bivariate analyses were used to calculate heritabilities (h²), additive genetic (r_g) correlations, and residual (r_e) correlations. Mean heritabilities were calculated from the 9 bivariate heritability estimates that were obtained per trait. Heritabilities and residual correlations of the binary traits were transformed onto the underlying liability scale according to described methods.^15,16

Results

Mean ± SD age was 4.14 ± 0.86 years; 73.2% of the horses were 3- to 4-year-olds. Females were significantly older (4.19 ± 0.92 years) than males (geldings, 4.10 ± 0.80 years; stallions, 4.04 ± 1.02 years). Mean ± SD height at the withers (highest point of the dorsal process of the thoracic vertebrae) was 166.71 ± 3.76 cm (range, 146 to 183 cm).

Distributions of radiographic findings were summarized (Table 1). Deviations from what was considered to be a normal radiographic appearance of the forelimb navicular bones were observed in 40.57% (2,092/5,157) of the horses, mostly in the form of few (20.48%, 1,056/5,157) or several (16.97% 875/5,157) conical CSs in the central part of the distal border. In 98.66% of horses with an abnormal RNB, abnormal findings involved the distal border of the navicular bone. Osseous fragments in the fetlock joints were diagnosed in 1,433 (27.79%) horses, FTs were diagnosed in 473 (9.17%) horses, and DAT was diagnosed in 397 (7.70%) horses.

The year of presumptive auction was significantly (P = 0.02 for OFT; P < 0.01 for all other traits) related to the distribution of all traits, as was the date of presumable auction (P < 0.001). Sex was significantly related to OFTs, DAT, and all RNB traits except for RNB_0/1a. Stallions had the highest risk, and mares the lowest risk, for having higher values of RNB_0–7 and RNB_0–3 and for being affected with OFTs, RNB_0/1b, RNB_0/1d, and RNB_0/1e (P < 0.001). Males were more likely to be affected with DAT (P < 0.01) and RNB_0/1c (P = 0.02) than females. Age was significantly related to OFTs and all RNB traits except for RNB_0/1a. The probability for having higher values for RNB_0–7 and RNB_0–3and for being affected with RNB_0/1b, RNB_0/1d, and RNB_0/1e increased significantly with age. Three-year-old mares born from November to March were least likely to have high values of RNB_0–7 or to be affected with RNB_0/1a, RNB_0/1c, and RNB_0/1d, whereas 4-year-old stallions born in April were most likely to have high values of RNB_0–7 and to be affected with RNB_0/1b or RNB_0/1d.

Variance estimates of all traits differed little among the various bivariate analyses. Ranges of estimated additive genetic and residual variances were summarized (Table 2). Comparison of heritabilities of the quasi-linear RNB traits revealed slightly lower estimates for RNB_0–7 (h² = 0.133 to 0.156; SE_h² = 0.020 to 0.023) than for RNB_0–3 (h² = 0.179 to 0.193; SE_h² = 0.019 to 0.025).

Mean heritabilities and additive genetic and residual correlations between the binary RNB traits were summarized (Table 3). Transformed heritabilities ranged from h² = 0.09 to 0.10 for RNB_0/1a, from h² = 0.24 to 0.26 for RNB_0/1b, from h² = 0.16 to 0.19 for RNB_0/1c, from h² = 0.22 to 0.25 for RNB_0/1d, and from h² = 0.33 to 0.34 for RNB_0/1e. Standard errors of heritabilities were usually larger for the binary RNB traits (SE_h², 0.023 to 0.095) than for the quasi-linear RNB traits. Additive genetic correlations among RNB_0/1a, RNB_0/1b, RNB_0/1d, and RNB_0/1e were highly positive (r_g ≥ 0.89; SEr_g ≤ 0.14). However, the additive genetic correlation between RNB_0/1c and the other binary RNB traits was r_g = 0.10 to 0.40 (SEr_g = 0.16 to 0.24). Residual correlations between the binary RNB traits ranged from r_e = −0.35 to 0.84 (SEr_e ≤ 0.02) before transformation. Transformation resulted in estimates of residual correlations that were partially outside the parameter space.

The distributions of horses with abnormal RNBs that had OFMs, OFTs, and DAT were summarized (Table 4). The number of coaffected horses ranged from 22 (RNB_0/1c and DATs) to 566 (RNB_0/1e and OFMs).

Transformed heritabilities were moderate for OFMs (h² = 0.14 ± 0.03), OFTs (h² = 0.34 ± 0.05), and DATs (h² = 0.18 ± 0.04). Results of the correlation analyses between RNB traits and OFMs, OFTs, and DAT were summarized (Table 5). Moderately negative additive genetic correlations were estimated between OFMs and RNB_0/1a; between OFTs and RNB_0–7, RNB_0–3, RNB_0/1a, RNB_0/1b, and RNB_0/1e; and between DATs and RNB_0/1a and RNB_0/1e (r_g = −0.40 to −0.11; SEr_g, 0.04 to 0.17). Moderately positive additive genetic correlations were estimated between OFMs and RNB_0/1e and between DAT and RNB_0/1c (r_g = 0.19 to 0.25), but the corresponding SEs were high (SEr_g, 0.13 to 0.20). Residual correlations between RNB traits and OFMs, OFTs, and DAT were close to 0, ranging from r_e = −0.04 to 0.05 (SEr_e, 0.02) before transformation and from r_e = −0.07 to 0.09 (SEr_e, 0.06) after transformation.

Discussion

The first objective of this study was to determine whether there were genetic correlations between different types of radiographic appearances of the navicular bones in the forefeet of Hanoverian Warmblood horses. The second objective was to determine whether there were genetic correlations between RNB and other radiographic findings that are frequently observed in the limbs of young riding horses.

Prevalences of 9.1% to 20.8% for osseous fragments in the fetlock and tarsocrural joints, deforming arthropathy in the tarsal joints, and pathologic changes in the navicular bones in young Hanoverian Warmbloods have been determined.^7,12,13,17 However, detailed information derived from radiographic findings in individual horses was not used in previous genetic analyses^5–9,12,i of the radiographic appearance of navicular bones. The fact that there was no distinction among individual radiographic findings in genetic analyses implies genetic homogeneity, but no studies have substantiated the assumption of genetic homogeneity or disproven the assumption of genetic heterogeneity for various types of radiographic findings in navicular bones. Because analyzing a combination of genetically heterogeneous conditions as 1 trait is likely to affect estimated genetic variables and biased variance and covariance estimates will result in over- or underestimation of heritabilities and genetic correlations, detailed radiographic data were used in this study.

Navicular bones were classified on the basis of the radiographic appearance of CSs and bone structure and contour. There is general agreement among investigators^2,11,18–20 that these are the criteria that should be considered in evaluation of navicular bone radiographs, and several radiographic classification schemes^{1,11,18,20,21,g} have been developed on that basis. However, a clear distinction between radiographic findings that are indicative and those that are not indicative of podotrochleosis has not been drawn.^a–g Although the risk for clinical manifestation of lameness may increase with the extent of radiographic changes, presumptively pathologic changes that are detected radiographically in the navicular bone do not necessarily correlate with clinical disease.

The shape of the proximal border of the navicular bone is postulated to be hereditary and associated with radiographic findings considered to be indicative of podotrochleosis.¹ By use of different classification schemes for categorizing the RNB, significant differences among paternal half-sibling groups have been reported.^1,5,22,i,m Genetic parameters for all-or-none traits resembling presence or absence of radiographic findings presumed to be associated with podotrochleosis have been estimated. Reported heritability estimates were low (h² = 0.06)⁹ when expressed on the observed scale, but were moderate to high (h² = 0.20 to 0.46) when expressed on the underlying liability scale.^6–8,12 In the present study, the transformed heritability estimate was low (h² = 0.09 to 10) for presence of few, short, conical CSs in the central portion of the distal border of the navicular bone. This radiographic finding is generally considered to represent physiologic variation,^{2, 3,11,19,20,h} and it is unlikely that it was included in previous genetic analyses. Transformed heritability estimates for the other binary traits were moderate (RNB_0/1b-e; h² = 0.17 to 0.34) and similar to the heritability estimates for the quasi-linear traits (RNB_0–7 and RNB_0–3; h² = 0.14 to 0.19).

If linear models are used for genetic analysis of binary traits, underestimation of heritabilities and residual correlations must be compensated for via transformation.^15,16 The applicability of transformation factors to our data was substantiated in a simulation study.¹⁷ Linear models were also used for analysis of the quasi-linear traits, although assumption of normality was clearly violated in our data. However, previous investigators^23–25 failed to detect relevant differences between results of linear and threshold-model analysis of ordered categoric data. Untransformed heritability estimates of the quasi-linear traits should therefore be as reliable as transformed heritability estimates of the binary traits.

Variation of trait definition, resembling separate versus combined genetic analysis of particular radiographic features of the navicular bones, was used to test the genetic equivalence of different types of radiographic findings. Radiographic findings in the navicular bones were fit in 1 of 7 mutually exclusive categories of radiographic appearance. Given the low prevalences of radiographic findings fitting into 5 of those 7 categories, it was not possible to derive 7 distinct binary traits or estimate genetic correlations among them. Accordingly, either the 2 large categories (ie, I and II) or combinations of different categories of RNB provided the basis for binary trait definition. Combination of different forms of radiographic findings in defining binary traits reflects the common practice of simplification in animal breeding. Inclusion of mutually exclusive and combined traits in 1 set of bivariate genetic analyses results in overlap and estimation of part-to-whole correlations.

To study the effects of simplification, we evaluated 3 individual binary traits (RNB_0/1a, RNB_0/1b, and RNB_0/1c) and 2 combined binary traits (RNB_0/1d and RNB_0/1e). Additive genetic correlations were mostly close to unity. However, radiographic findings that represented more extensive navicular bone remodeling indicated that there were exceptions; those radiographic findings were genetically correlated to the other traits with a maximal r value of 0.40. Because alterations in structure or contour were rarely seen in the navicular bones of the study horses, they were not analyzed individually but together with deformed CSs. Even so, the proportion of horses with these marked radiographic findings was < 5%, and the respective values for variance and covariance had large SEs. Although correlation estimates relating to these lowprevalence findings were therefore less meaningful than estimates relating to the more prevalent findings, they may be an indication of different genetic background. Choice of trait definition for genetic evaluation for RNB must account for possible genetic inhomogeneity. If breeding measures are to aim at improvement in the RNB, the choice of trait definition cannot be made exclusively on the basis of heritability estimates (ie, from estimating which trait would be expected to respond the most quickly to selection).

Osseous fragments in the fetlock and tarsocrural joints and deforming arthropathy in the tarsal joints were included in the correlation analyses because of their quantitative importance among young riding horses.^7,12,13,17 The mode of trait definition with respect to the radiographic appearance of the navicular bones had a considerable influence on the correlation estimates. The finding of few short and conical CSs in the central portion of the distal border of the navicular bone is typically interpreted as variation within physiologic range.^{2,3,11,19,20,h} Significant negative additive genetic correlations indicated that few short and conical CSs in the central portion of the distal border of the navicular bone were less likely to occur in horses affected with OFMs, OFTs, or DAT. There were further indications for a negative additive genetic correlation between most classifications of RNB of the forelimbs and OFTs. However, SEs of additive genetic correlations in many cases were as large as or larger than the respective correlation estimates.

Detailed radiographic information from larger numbers of horses is needed to obtain more reliable estimates of correlations among different types of radiographic findings in the navicular bones and between those and other relevant radiographic findings in horses. Furthermore, data from horses with clinical navicular disease as well as extensive radiographic changes should be included in future genetic analyses. Such information would likely help provide genetic justification for distinguishing between physiologic and pathologic radiographic findings. Until then, neither merely distinguishing between presence or absence of radiographic changes nor empiric distinction between normal and abnormal radiographic appearance appears to be a suitable basis for reliable genetic analyses, and a quasi-linear description of the RNB should be used instead. Such use of all available information should increase the reliability of genetic analyses, and use of reliably estimated genetic parameters is prerequisite for optimal use of the radiographic appearance of navicular bones in future breeding strategies. Genetic gain from selection may be greater with the use of RNB_0/1e (ie, distinction between existence and nonexistence of radiographic changes involving the distal border of the navicular bone). However, the importance of physiologic, clinically irrelevant variation in radiographic appearance is not yet known and should be determined in future long-term studies.

Distinction between existence and nonexistence of distinct radiographic findings in the navicular bones (trait RNB_0/1d in this study) has been used to demonstrate that it will be possible to decrease the prevalences of certain radiographic findings in the Hanoverian Warmblood horse population via genetic evaluation and sire selection.^12,26 Refined trait definition will increase selection efficiency and accelerate breeding progress with respect to the radiographic appearance of the forelimb navicular bones in Hanoverian Warmblood riding horses.