Dissection of X chromosome dosage compensation for quantitative traits in sheep using different statistical models

Abstract

Since males and females have different number of X chromosome, different mechanisms have evolved to equalize dosage of gene products from the X chromosome between XX females and XY males. The aim of this study was to study X chromosome dosage compensation for growth rate (GR), Kleiber ratio (KR), efficiency of growth (EF) and relative growth rate (RGR) in Zandi sheep. A two steps procedure was adopted to analysis data. In the first step, each trait was analyzed with a series of 6 animal models including different combinations of direct and maternal effects. Using Akaike’s Information Criterion (AIC) the best model (Model I) was selected for each trait. In the second step, five additional models were fitted by adding X chromosome effects to the Model I, considering 5 strategies for modeling X chromosome dosage compensation: (1) no global dosage compensation (ngdc), (2) random inactivation in the homogametic sex (hori), (3) doubling of the single shared sex chromosome in the heterogametic sex (hedo), (4) halving expression of both sex chromosomes in the homogametic sex (hoha) or (5) inactivation of the paternal sex chromosome in the homogametic sex (hopi). Predictive ability of models was measured using the mean squared error of prediction (MSE) and Pearson’s correlation coefficient between the real and predicted values of records $(\text{r}(y,\widehat{y})).$ Correlations between traits due to autosomal- and X-linked genetic effects were estimated by bi-variate analyses. For GR and KR, models including X-linked effects lead to a much better fit of data, expressed by the strong decrease in the AIC criterion. Models including X-linked effects had also better predictive ability as they provided smaller MSE and higher $\text{r}(y, \widehat{y}).$ For GR and KR, although all strategies for modeling X chromosome dosage compensation improved general properties of the model, the model “ngdc” fitted the data significantly better than other models. Including X-linked genetic effects in the model led to 10% (GR, KR) decrease in the autosomal additive variance, and 7% (KR) to 19% (GR) decrease in the residual variance. Estimates of autosomal heritability (${\mathbf{h}}_{\mathbf{a}}^{2}$), were 0.15 ± 0.03, 0.13 ± 0.03, 0.09 ± 0.03 and 0.13 ± 0.03 for GR, KR, EF and RGR, respectively. X-linked heritability (${\mathbf{h}}_{\mathbf{s}}^{2}$) was 0.08 ± 0.03 for GR and 0.04 ± 0.03 for KR, respectively. Maternal heritability (${\mathbf{h}}_{\mathbf{m}}^{2}$) were 0.02 ± 0.01, 0.01 ± 0.01, 0.03 ± 0.02 and 0.03 ± 0.02 for GR, KR, EF and RGR, respectively. For GR and KR, the Spearman’s correlation between breeding values obtained from the best model and model I deviated from unity, indicating re-ranking of top animals across models. The X-linked additive genetic correlation and autosomal additive genetic correlation were similar in terms of sign and magnitude in a way that they were all positive and high. As considering X-linked genetic effects resulted to an improvement in the general properties of the model and possibility of re-ranking of top animals, including these effects in the model, considering dosage compensation on the X chromosome was recommended.

Lamb growth rate is a key component of profitable sheep farming systems. Faster growing lambs are likely to be slaughtered earlier and/or be heavier at slaughter than their slower growing counterparts and therefore, brings a quicker income to the farmer. More importantly, faster growing lambs are more efficient user of feed because of the lower ratio of maintenance to total feed requirements. A wide range of factors interact to affect lamb growth rate, including ewe live weight and body condition, ewe milking ability, pasture quality and quantity and the genetic ability of the lamb to grow¹. Since, part of the phenotypic variation in growth rate and efficiency-related traits among animals has genetic origin, this genetic variation can be exploited to improve these traits genetically. However, a genetic improvement program requires knowledge of variance components and genetic parameters to predict selection response, compare breeding plans, and predict the breeding values of selection candidates². The heritability of growth rate and efficiency-related traits has been estimated by several authors including Sing et al.³, Javanrouh et al.⁴ and Mokhtari et al.⁵. However, in most studies, the effects of X chromosome are ignored (see Sing et al.³ and references therein).

In mammals, sex is determined by the heteromorphic sex chromosomes: XY chromosomes in males and XX chromosomes in females. There are more than 1000 genes on the X chromosome that is virtually identical across species. The Y chromosome is very much smaller, having lost all but a few of these genes and so is relatively gene poor⁶. The SRY gene on the Y chromosome underlies sex determination by initiating testis formation⁷. Most genes on the X chromosome do not have a Y chromosome counterpart and do not have sex-specific functions. Therefore, X copy number in males and females is different. This presents a challenge to placental mammals, who need to (roughly) equalize expression from the X chromosome in male (XY) and female (XX) cells. Deviation from diploidy may induce detrimental consequences. For example, gene duplications or deletions can induce cancer and chromosome monosomy or trisomy usually causes fetal lethality⁸. Ohno⁹ suggested upregulation of the X-linked genes in the heterogametic sex (XY) to maintain their expression to the levels of the diploid autosomes. Another mechanism is inactivating one X chromosome in females during early development by assembling a distinctive form of chromatin that largely silences gene expression¹⁰. Upregulation of the X-linked genes and inactivating one X chromosome in females balance the X chromosome gene dosage between males and females. These mechanisms have been actively studied in mice and humans but lag behind in domestic species⁸.

Although suitable approach to construct sex-chromosome additive genetic relationship matrix (S-matrix) for using in the mixed animal models to estimate BLUPs of autosomal and X-chromosomal additive effects was developed¹¹, due to the lack of suitable software, the use of this approach was delayed until recently. By developing software such as WOMBAT¹² and ASReml¹³ that were able to consider X-chromosome additive genetic relationship matrix (S-matrix) in the animal model framework, several authors including Maraveni et al.¹⁴, Latifi¹⁵, Ghafouri-Kesbi and Abbasi¹⁶, Kargar Borzi¹⁷, Noorian et al.¹⁸ and Bahri-Binabaj et al.¹⁹ estimated the ratio of phenotypic variation in economic traits of sheep which caused by activity of loci on the X chromosome. These authors pointed out that including the X-linked genetic effects in the model of genetic evaluation can help to better understanding of autosomal way of inheritance and to increase the accuracy of genetic evaluation. However, none of these studies considered dosage compensation on the X chromosome for analyzing complex traits of livestock. To inverse S-matrix, different mechanisms for dosage compensation on the X chromosome have been proposed including no global dosage compensation, random inactivation in the homogametic sex, doubling of the single shared sex chromosome in the heterogametic sex, halving expression of both sex chromosomes in the homogametic sex, or inactivation of the paternal sex chromosome in the homogametic sex²⁰. Linear models in BLUP that separate autosomal and X-linked additive effects and different mechanisms for gene dose compensation can be used to identify significant random effects and effective mechanisms of dosage compensation on the X chromosome²¹. In sheep, the effects of X chromosome dosage compensation on growth and efficiency related traits have not been studied so far. Therefore, the aim of this study was to estimate the autosomal and X-linked genetic components for growth rate and efficiency-related traits in Zandi sheep considering different models for dosage compensation on the X chromosome.

Materials and methods

Data

The phenotype and pedigree data was provided by the Zandi sheep breeding station located in Tehran, Iran. The station was established by the ministry of agriculture of Iran with the aim of improving the whole population of Zandi sheep. The pedigree included 5930 animals distributed over 10 generations (Table 1). The errors in pedigree were detected and corrected using CFC software²². The errors were (1) repeated animals in the pedigree (2) animals which were registered as their own sire or dam and (3) presence of loop in the pedigree.

Table 1 Pedigree structure of the Zandi sheep.

Birth weight (BW) and weaning weight (WW) were extracted from data files. To account for the differences among animals with different ages, weaning weights were adjusted to standard 90 days. The increase in body weight from birth to weaning was used to calculate growth rate (GR) by dividing total gain by the number of days in the period. The estimate of GR was then used to calculate the corresponding Kleiber ratio²³ as KR = GR/WW^0.75. Body weights were also used to calculate the efficiency of growth¹⁶ as EF = ((WW − BW)/BW) × 100. In addition, the relative growth rate²⁴ from birth to weaning (RGR) was calculated as RGR = Log_e(WW) − Log_e(BW)/90. Summary statistics of the studied traits are shown in Table 2.

Table 2 Characteristics of the data structure.

Statistical analysis

Fixed effects including year of birth, age of dam at lambing, sex of lambs and type of birth were analyzed with the GLM function in R²⁵. These fixed effects were significant (p < 0.05) for all traits and were subsequently included in the linear mixed models.

Estimation of variance components and genetic parameters was done in a two-step process. In the first step, each trait was analyzed with 6 univariate animal models, including various combinations of animal and maternal effects (Table 3). Model 1, which included only random animal effects, was the simplest. On the contrary, Model 6, which was the complete animal model, included animal additive genetic, maternal permanent environmental, maternal additive genetic and direct-maternal additive genetic covariance. The general representation of the Model 6 was as follow:

$${\mathbf{y}} \, = \, {\mathbf{X}\mathbf{\beta} } + \, {\mathbf{Z}}_{{\mathbf{1}}} {\mathbf{a}} \, + \, {\mathbf{Z}}_{{\mathbf{2}}} {\mathbf{c}} \, + \, {\mathbf{Z}}_{{\mathbf{3}}} {\mathbf{m}} \, + \, {\mathbf{e}},{\text{Cov}}\left( {{\mathbf{a}},{\mathbf{m}}} \right) \, = {\mathbf{A}}\upsigma_{{{\text{a}},{\text{m}}}}$$

where y is the vector of observations. β is the vector of fixed effects fitted with design matrix X. Z₁ is the design matrix for animal additive genetic effects. Z₂ and Z₃ are incidence matrices relating observations to maternal permanent environmental and maternal additive genetic effects, respectively. A was the numerator relationship matrix and a, c, m and e were vectors for direct additive genetic, maternal permanent environmental, maternal additive genetic, and residual effects, respectively. Cov(a,m) was direct-maternal additive genetic covariance.

Table 3 The random (co)variance components used in the six models.

To estimate variance components and genetic parameters, the WOMBAT¹² program was used. The information criterion of Akaike²⁶ (AIC) was computed to rank the models according to their power to fit the data. Let p denotes the number of random (co)variance parameters to be estimated, and log L is the maximum log-likelihood, then the information criterion is defined as AIC = − 2 log L + 2p. The model yielding the smallest AIC fits the data best. Predictive ability of models was measured using the mean squared error of prediction (MSE) and Pearson’s correlation coefficient between the real and predicted values of records $(\text{r}(y,\widehat{y})).$

In the second step, X-linked effects were added to the best model selected in the step one (from now on, referred to as Model I). Five models (models 7 to 11) were fitted for considering X chromosome dosage compensation. These models were the same as Model I but allowed an X-linked genetic effect as follows:

$${\text{Models 7-11}}:{\mathbf{y}} = {\mathbf{I}} + {\mathbf{Z}}{\mathbf{4}}{\text{s}},$$

The X-linked additive genetic relationship matrix (S) and its inverse (S⁻¹) were constructed using the nadiv package²⁰ in R software²⁵. This package utilizes equations presented in the Fernando and Grossman¹¹. The S⁻¹ was then supplied externally to WOMBAT software¹² using the GIN option. In order to identify the effective dosage compensation mechanism on the X chromosome, the inverse of the S was constructed considering different models of dosage compensation²⁰: (1) no global dosage compensation (ngdc), (2) random inactivation in the homogametic sex (hori), (3) doubling of the single shared sex chromosome in the heterogametic sex (hedo), (4) halving expression of both sex chromosomes in the homogametic sex (hoha) and (5) inactivation of the paternal sex chromosome in the homogametic sex (hopi). BLUPs of autosomal additive genetic, X-chromosomal additive genetic, maternal permanent environmental and maternal additive genetic can be obtained using mixed model equations considering covariance matrices of random effects a, s, c and m. (Co)variance matrix for the random effects was as follow:

$$Var \left[\begin{array}{c}\varvec{a}\\ \varvec{s}\\ \varvec{c}\\ \varvec{m}\\ \varvec{e}\end{array}\right]=\left[\begin{array}{c}\mathbf{A}{\updelta }_{\mathbf{a}}^\mathbf{2} \\ \mathbf{0}\\ \mathbf{0}\\ {\mathbf{A}\updelta }_{\text{am }}\\ \mathbf{0}\end{array}\begin{array}{c}\mathbf{0} \\ \varvec{S}{\updelta }_{\text{s}}^{2}\\ \mathbf{0}\\ \mathbf{0}\\ \mathbf{0}\end{array}\begin{array}{c}\mathbf{0}\\ \mathbf{0}\\ {\mathbf{I}}_{\text{c}}{\updelta }_{\text{c} }^{2}\\ \mathbf{0}\\ \mathbf{0}\end{array}\begin{array}{c}{ \mathbf{A}\updelta }_{\text{am} }\\ \mathbf{0}\\ \mathbf{0}\\ {\mathbf{A}\updelta }_{\text{m}}^{2}\\ \mathbf{0}\end{array}\boldsymbol{ }\boldsymbol{ }\boldsymbol{ }\begin{array}{c}\mathbf{0}\\ \mathbf{0}\\ \mathbf{0}\\ \mathbf{0}\\ {\mathbf{I}}_{\text{e}}{\updelta }_{\text{e}}^{2}\end{array}\right]$$

where a, s, c, m and e are vectors for autosomal additive genetic, X-linked additive genetic, maternal permanent environmental, maternal additive genetic and residual effects, respectively. ${\updelta }_{\text{a}}^{2}$, ${\delta }_{s}^{2}$, ${\updelta }_{\text{c}}^{2}$, ${\updelta }_{\text{m}}^{2}$, and ${\updelta }_{\text{e}}^{2}$ are autosomal additive genetic variance, X-linked additive genetic variance, maternal permanent environmental variance, maternal additive genetic variance, and residual variance, respectively. σ_a,m was direct-maternal additive genetic covariance. A was the numerator relationship matrix and I_c and I_e were identity matrices of appropriate dimensions.

Estimated additive breeding values (EBVs) for individuals for studied traits were derived using the best linear unbiased prediction procedure (BLUP). The effect of inclusion of X-linked effects in the model on additive breeding values was tested by estimating the Spearman’s correlation between breeding values obtained from the best model and model I (best model without X-linked effects). In addition, change in the ranking of top 10 and top 100 animals based on their EBVs obtained from the best model and model I was monitored by calculating the number of animals that dropped from the top 10 or top 100 animals after including X-linked effects in the model.

Correlations between traits were estimated by bi-variate analyses. The models applied in the bi-variate analyses were those selected as best for each of the underlying traits in the uni-variate analyses. The matrix notation for the bivariate model was as follow:

$$\left[\begin{array}{c}{\mathbf{y}}_\mathbf{1}\\ {\mathbf{y}}_\mathbf{2}\end{array}\right]=\left[\begin{array}{cc}{\mathbf{X}}_\mathbf{1}& \mathbf{0}\\ \mathbf{0}& {\mathbf{X}}_\mathbf{2}\end{array}\right]\left[\begin{array}{c}{\mathbf{b}}_\mathbf{1}\\ {\mathbf{b}}_{2}\end{array}\right]+\left[\begin{array}{cc}{\mathbf{Z}}_{\mathbf{a}\mathbf{1}}& \mathbf{0}\\ \mathbf{0}& {\mathbf{Z}}_{\mathbf{a}\mathbf{2}}\end{array}\right]\left[\begin{array}{c}{\mathbf{a}}_{1}\\ {\mathbf{a}}_\mathbf{2}\end{array}\right]+\left[\begin{array}{cc}{\mathbf{Z}}_{\mathbf{s1}}& \mathbf{0}\\ \mathbf{0}& {\mathbf{Z}}_{\mathbf{s2}}\end{array}\right]\left[\begin{array}{c}{\mathbf{s}}_\mathbf{1}\\ {\mathbf{s}}_\mathbf{2}\end{array}\right]+\left[\begin{array}{cc}{\mathbf{Z}}_{\mathbf{m1}}& \mathbf{0}\\ \mathbf{0}& {\mathbf{Z}}_{\mathbf{m2}}\end{array}\right]\left[\begin{array}{c}{\mathbf{m}}_{1}\\ {\mathbf{m}}_\mathbf{2}\end{array}\right]+\left[\begin{array}{c}{\mathbf{e}}_\mathbf{1}\\ {\mathbf{e}}_\mathbf{2}\end{array}\right],$$

where y₁ and y₂ denote traits 1 and 2, respectively.

Results

In the step one, for all traits studied, model 3 which included direct and maternal additive genetic had the lowest AIC and, therefore, selected as the best model (Model I). In the step 2, for GR and KR, adding X-linked effects to the best model already selected in step one (Model I), led to a substantially better data fit stated by the significant decrease in the AIC values. For GR and KR, model 7 in which no global dosage compensation (ngdc) was used to model X-chromosome dosage compensation was superior to other models expressed by lower AIC values. This model had also better predictive ability compared to model I as it provided lower MSE and higher r($y$, $\widehat{y}$) (Table 4). For EF and RGR, adding X-linked effects to the model I, didn’t led to an additional improvement in the likelihood of the model. In addition, regarding measures of predictive ability, there was no superiority over model I. Therefore, for EF and RGR, Model I was selected as the final best model.

Table 4 Predictive ability of the best model and model I measured by $\mathbf{r}({\varvec{y}},\widehat{{\varvec{y}}})$ and MSE.

Tables 5, 6, 7, and 8 present estimates of variance components and genetic parameters for the studied traits. Including X-linked genetic effects in the model led to 10% decrease in autosomal additive genetic variance for GR and KR. In addition, accounting for X-linked genetic effects in the model decreased residual variance by 19% (GR) and 7% (KR). Based on the best models, estimates of additive heritability (${\mathbf{h}}_{\mathbf{a}}^{2}$), were 0.15 ± 0.03, 0.13 ± 0.03, 0.9 ± 0.03 and 0.13 ± 0.03 for GR, KR, EF and RGR, respectively. X-linked heritabilities (${\mathbf{h}}_{\mathbf{s}}^{2}$) were 0.08 ± 0.03 and 0.04 ± 0.03 for GR and KR, respectively. Maternal heritability (${\mathbf{h}}_{\mathbf{m}}^{2}$) were 0.02 ± 0.01, 0.01 ± 0.01, 0.03 ± 0.02 and 0.03 ± 0.02 for GR, KR, EF and RGR, respectively.

Table 5 Estimates of variance components and genetic parameters for growth rate (best model in bold).

Table 6 Estimates of variance components and genetic parameters for Kleiber ratio (best model in bold).

Table 7 Estimates of variance components and genetic parameters for efficiency of growth (best model in bold).

Table 8 Estimates of variance components and genetic parameters for relative growth rate (best model in bold).

The Spearman’s correlation between breeding values obtained from the best model and model I was 0.94 for GR and 0.95 for KR. The number of animals that remained in their cohorts (top 10 and top 100 animals) after including X-linked effects in the model is shown in Table 9. Out of 10 top animals, for GR and KR, respectively, 7 and 8 animals (on average 8 animals) were common when EBVs of Zandi lambs were estimated using the model I and model 7 (I + ngdc). For 100 top animals, 73 and 78 animals (on average 76 animals) were common, indicating re-ranking of top animals after including X-linked effects in the model.

Table 9 The number of animals that were common between the best model and model I based on their additive breeding values.

Table 10 shows correlations between traits. Estimates of X-linked additive genetic correlations (r_s) and autosomal additive genetic correlations (r_a) were all positive and high. Phenotypic correlations between traits were also positive but lower than r_s and r_a.

Table 10 Correlations between studied traits.

Discussion

Genetic analysis of complex traits in mammals is frequently limited to autosomes, with the X chromosome excluded because of its hemizygosity in males. Broman et al.²⁷ developed a method for appropriate treatment of the X chromosome for QTL mapping in experimental crosses. Accordingly, some efforts focused on determination of the QTLs on the X chromosome underlying economic traits in livestock. These studies demonstrated the importance of the X chromosome in the genetic determinism of complex traits in livestock and highlighted new functional candidate genes and variants for these traits²⁸ (see for example Sanchez et al.²⁸).

Among strategies used to model dosage compensation on the X chromosome, a model including no global dosage compensation on the X chromosome (ngdc) fitted the data best. Dosage regulation of the sex chromosomes can be viewed as either global, i.e. employing mechanisms that modify most—but not all—genes on an entire chromosome, or local, i.e. acting on individual genes. This distinction is somewhat fluid as the number of dosage-compensated genes on a given sex chromosome varies between tissues, and also depends on methods of analysis²⁹. Our results showed lack of mechanisms that modify all genes on the X chromosome and, therefore, dosage compensation in sheep may be partial, i.e., not all genes on the X chromosome in females are modified. Nevertheless, to have a clear cut verdict about this finding, more research is needed.

In partial compensation, the male: female ratios of X-linked gene expression vary between 1 (full compensation) and 2 (absence of compensation)³⁰. Both complete and partial compensation have been evidenced in different species. Complete dosage compensation by upregulation of the male X chromosome in Drosophila melanogaster³¹, while in another insect, Heliconius butterfly, there is only partial dosage compensation³². In a plant (Silene latifola) with newly evolved sex chromosomes no global dosage compensation between sexes was detected³³. In addition, in chicken, Kaviani et al.²¹ analysed body weight at different ages with models similar to ours, and selected a model with no global dosage compensation on the X chromosome as best model for body weights at 8 and 12 weeks of age.

Estimation of the proportions of heritability due to autosomes and X chromosome for economic traits in sheep has a short history, beginning in 2018. Latifi¹⁵ in Mehraban sheep reported estimates of ${\mathbf{h}}_{\mathbf{s}}^{2}$ for GR and KR as 0.10 and 0.07, respectively. In Kermani sheep, Kargar-Borzi¹⁷ reported estimates of ${\mathbf{h}}_{\mathbf{s}}^{2}$ as 0.03 and 0.01 for GR and KR, respectively. For EF, Noorian et al.¹⁸ reported estimated value of ${\mathbf{h}}_{\mathbf{s}}^{2}$ as 0.02. In addition, for body weight at different ages, estimates of ${\mathbf{h}}_{\mathbf{s}}^{2}$ ranged from 0.01 for body weight at birth in Baluchi sheep¹⁹ to 0.14 for body weight at 6 months of age in Lori-Bakhtiari sheep¹⁴. Our findings, together with previous reports, show that part of phenotypic variation in growth- and efficiency-related traits in sheep caused by the activity of genes on the X chromosome. Therefore, a decrease in the autosomal additive genetic variance was expected after distangling this effect. However, our results showed that by including X-linked effects, not only autosomal additive genetic variance but also residual variance decreased which is an evidence for improvement in the general properties of the model. Our results are in agreement with Larsen et al.³⁴ who reported 31% decrease in ${{\varvec{\updelta}}}_{\mathbf{a}}^{2}$ following inclusion of X-linked genetic effects in the model for plumage spot diameter in barn owls. In addition, Noorian et al.¹⁸ who worked on efficiency of growth in Baluchi sheep reported 20% decrease in the additive genetic variance when X-linked effects were included in the model.

Ghafouri-Kesbi and Abbasi¹⁶ reported that inclusion of X-linked genetic effects in the animal models when they are really present resulted to increase in the accuracy of genetic evaluation. Information in the Table 9 shows that out of top 10 animals, about 20% of them were dropped from the group when the X chromosome effect was included in the model. It was 25% for top 100 animals, i.e., ranking of top animals will be changed by including X-linked effects in the genetic evaluation model in agreement with Vatankhah et al.³⁵, who reported that animal models that were able to partition X-linked effects from total additive genetic variation could enable more effective genetic selection to improve economic traits in sheep.

The estimated value of ${\mathbf{h}}_{\mathbf{a}}^{2}$ for pre-weaning GR (0.15) was in the reported range from 0.03 in the Sangsari breed³⁶ to 0.42 in the Moroccan Timahdit breed of sheep³⁷. For pre-weaning KR, literature estimates of ${\mathbf{h}}_{\mathbf{a}}^{2}$ ranged from 0.04 in the Arman breed⁵ to 0.17 in the Makuei sheep³⁸. For pre-weaning EF, literature estimates of additive heritability ranged from 0.05 in the Arman breed⁴ to 0.07 in the Moghani sheep³⁹. For pre-weaning RGR, literature estimates of ${\mathbf{h}}_{\mathbf{a}}^{2}$ ranged from 0.06 in the Arman breed¹⁸ to 0.15 in the Afshari breed⁴⁰. These reports show that efficiency-related traits have low heritability which means that response to selection will be limited. However, since these traits have a significant influence on the profitability of sheep production, genetic improvement in these traits should be considered.

Genetic correlation is an informative metric to quantify the overall genetic similarity between complex traits, which provides insights into their polygenic genetic architecture. There are no reports on the estimates of the X-linked additive genetic correlation (r_s) for growth- and efficiency-related traits in sheep which makes it impossible to compare the results. However, estimates of the autosomal additive genetic correlation (r_a) were close to those reported by Ghafouri-Kesbi and Gholizadeh⁴¹, Jafaroghli et al.³⁹ and Javanrouh et al.⁴. Surprisingly, the X-linked additive genetic correlation (r_s) and autosomal additive genetic correlation were similar in terms of sign and magnitude. Positive r_a and r_s shows that two traits share similar genes on autosomal and X chromosomes. Strong r_a and r_s between GR and KR indicate that animals with higher GR are also efficient user of feed and vice versa, and that this positive r_a and r_s between traits guarantee the success of multi-trait selection programs including growth rate and efficiency-related traits.

Conclusions

X-linked genetic effects contributed to phenotypic variation of studied traits up to 8%. Among strategies tested to model dosage compensation on the X chromosome, no global dosage compensation (ngdc) strategy was superior to other strategies, possibility indicating partial dosage compensation on the X chromosome. Estimates of X-linked additive genetic correlation indicated that GR and KR share similar genes on the autosomal and X chromosomes. Since inclusion of X-linked effects in the genetic evaluation model lead to a substantially better data fit, using a model including X-linked effects for estimating variance components and prediction of breeding values was recommended. Nonetheless, as our study was the first attempt to investigate X chromosome dosage compensation in sheep, further research is needed to validate our findings.