Multi-trait selection indexes for sustainable UK hill sheep production

Transcription

1 Animal Science 2001, 73: /01/ $ British Society of Animal Science Multi-trait selection indexes for sustainable UK hill sheep production J. Conington 1, S. C. Bishop 2, B. Grundy 1, A. Waterhouse 3 and G. Simm 1 1 Scottish Agricultural College, West Mains Road, Edinburgh EH9 3JG, UK 2 Roslin Institute, Roslin, Midlothian EH25 9PS, UK 3 Scottish Agricultural College, Auchincruive, Ayr KA6 5HW, UK j.conington@ed.sac.ac.uk Abstract Three selection indexes for the UK hill sheep sector are derived to suit the extremes of hill production systems. These are: (i) intensive, where all surplus lambs not required for breeding are finished for slaughter, (ii) extensive, where all surplus store lambs are sold to other farmers for finishing, and (iii) semi-intensive, which is intermediate between the two extremes, i.e. farms finish some lambs for slaughter and sell others as store lambs. Parameters for 12 breeding goal and index traits were estimated using a total of 3962 lamb records and 5944 ewe lambing records from Scottish Blackface sheep on two Scottish Agricultural College experimental hill farms. The breeding goal comprised carcass, maternal and survival traits. The evaluation of these indexes showed that improvements in maternal traits are possible, along with more modest improvements in carcass quality traits. Responses to selection are expected to be lower for the extensive farm in general, compared with the intensive farm. Evaluations of alternative indexes show that an index using measurements of fat and muscle on ewes rather than on lambs may be more cost-effective to implement in practice, compared with the original index, although this change results in a higher (i.e. undesirable) gain in mature size. Sensitivity analyses showed that in general, the indexes are robust to changes in economic values and to changes in heritability estimates. Keywords: carcass composition, maternal effects, selection index, sheep. Introduction Until relatively recently, only a small proportion of UK hill sheep flocks were engaged in performance recording. For the last two decades, the total number of lambs from hill breeds recorded per year has fluctuated between 5000 and 8800 (J. Southgate, unpublished results). This represents proportionally less than 0 01 of the total number of lambs from hill breeds born each year. Some hill breeds have no flocks registered with Signet, the organization that carries out performance recording and genetic evaluations for the sheep industry in the UK. The reasons why such a low level of participation in genetic improvement for production traits exists in the purebred hill flocks probably include the physical difficulty of pedigree recording in extensive 413 conditions, lack of awareness of the potential benefits of genetic improvement programmes, emphasis on perceived aesthetic qualities of the sheep and the perception that the environment has an overriding influence on sheep performance. The introduction of sheep quotas in the UK in 1992 imposed an upper limit to the number of sheep eligible for subsidy support payments. Improving performance must therefore come from improved productivity per ewe. This can be achieved by temporary means such as management or nutrition but improvement of production through genetic selection is permanent. As more hill flocks now lamb either indoors, or in enclosed paddocks, the physical difficulties of recording, in particular identifying a lamb to its dam at birth, can be overcome.

2 414 Conington, Bishop, Grundy, Waterhouse and Simm Purebred hill lambs tend to fall at the poorer end of the carcass quality scale used in the UK (Conington et al., 1998), hence there is considerable scope for improving carcass traits from hill lambs as well as maternal traits. An index derived by Simm and Dingwall (1989) incorporating measurements of lamb weight, ultrasonic fat depth and muscle depth to predict carcass lean and fat weights has been used successfully for terminal sires in the UK. Some hill flocks have used both a maternal index and this terminal sire index on their flocks in a two-stage selection process. However, to date, there are no estimated breeding values (EBVs) for traits associated with improved health, longevity or welfare and no overall index combining carcass and maternal traits. The objective of this paper, therefore, is to derive selection indexes for purebred hill sheep incorporating carcass and maternal traits and including traits that contribute to sustainable, increased performance in hill ewes. (In this paper, the term sustainable refers to increased performance without detrimental correlated decreases in animal welfare, as judged by lamb and ewe mortality.) As the hill sector in the UK is diverse, three indexes are derived to reflect the extremes of hill production systems. These are: (i) intensive, where all surplus lambs not required for breeding are finished for slaughter, (ii) extensive, where all surplus lambs are sold on to other farmers for finishing ( store lambs) and where physical constraints such as hill pasture quality and availability of improved pasture for twin rearing limit sheep performance, and (iii) semi-intensive, which is intermediate between the two extremes and where farms finish some lambs and sell others as stores. This study explores whether responses to selection are similar in the three production systems, which traits contribute the most to the indexes and which could be omitted from the index without affecting responses in other traits. The accuracy of the indexes and the predicted economic returns from selection using the indexes are examined. The extent to which the results from the index calculations are altered due to changes in both economic values and heritabilities of key traits are investigated through sensitivity analyses. Material and methods A selection index combines information from an individual s own and relatives phenotypic performance for multiple traits into an overall score. It is an effective way of selecting breeding stock when several traits are being evaluated. The best way to implement multiple trait selection is via EBVs from best linear unbiased prediction (BLUP). However, a multi-trait BLUP framework does not readily lend itself to decision making regarding which traits and measurements to include in an index and the relative importance of each trait. This may be achieved by selection index theory. In this case, selection indexes that mimic the typical contribution from relatives with BLUP were constructed to: (i) determine how each trait contributes to selection, (ii) determine which traits are important and which can be dropped from the index with little effect, (iii) calculate the expected genetic change in each trait with selection, and (iv) determine the overall accuracy of selection. Definition of the breeding goal and measured traits Ten breeding goal traits are evaluated for their inclusion into the selection indexes. They include Table 1 Breeding goal and index (measured) traits Breeding goal traits Ewe traits Mature size (weight) Longevity (age at culling or death) Lamb loss No. of lambs reared (litter size at weaning) Maternal ( average ) weaning weight Fleece weight Lamb traits Lamb weaning weight (direct) Carcass fat class Carcass conformation class Carcass weight Index traits Ewe traits Pre-mating live weight Age at culling Lamb loss from birth to weaning Litter size at weaning Average weight of lambs weaned Fleece weight Lamb traits Weaning weight Ultrasonic fat depth Ultrasonic muscle depth No. of lambs lost to weaning (including dead at birth, lambs fostered/taken off). Including lambs born dead. Own lambs only. Including weight of lambs fostered on.

3 maternal and lamb traits. To achieve the goals, nine recorded ( index ) traits are considered. The goal and index traits are shown in Table 1. Pre-mating live weight is used to predict mature size, and age at culling or death is used for longevity. Lambs lost from birth to weaning include lambs born dead and lambs fostered or taken off the dam. The number of lambs reared includes only the ewe s own lambs reared until weaning. Ewes that have reared fostered lambs are credited via the average weight of lambs weaned ( average weaning weight ), which is used as the indicator of maternal ability. Fleece weight is recorded as greasy fleece weight. The breeding goals for lamb traits are direct weaning weight (at an average age of 17 weeks), carcass weight, carcass fat class and carcass conformation score. Lamb traits, which assist in the prediction of these goal traits, are ultrasonic fat and muscle measurements and weaning weight. Data and genetic parameters In order to calculate the properties of the selection indexes, genetic and phenotypic parameter estimates (heritabilities and correlations) are required for all goal and index traits. The genetic and phenotypic parameters were estimated from data collected on Scottish Blackface sheep in two experimental flocks over a 6-year period, as part of an experimental programme described in detail by Conington et al. (1995 and 1998). The total number of lamb records used for parameter estimation was 3962 of which 1580 records were obtained on slaughter characteristics. A total of 4410 repeated ewe records over four parities were used together with single records on 1534 ewes. A total of 98 sires were used, 36 of which were used on both farms. Univariate heritabilities, means, standard deviations and repeatabilities of the traits used in index calculations are shown in Table 2. Multi-trait selection indexes for UK hill sheep 415 Heritability estimates measured on the ewe were obtained using the restricted maximum likelihood procedure (REML), fitting an animal repeatability model with a common environmental source of variance (c 2 ). For the lamb traits, however, the c 2 effect is a common litter effect. The data structure precluded estimation of maternal genetic effects. Meat and Livestock Commission conformation class was coded as E = 5, U = 4, R = 3, O = 2, and P = 1. These conformation scores were then transformed to corresponding standard deviation units on the underlying normal scale, with a mean of zero and standard deviation of 1. The parameter estimates were then obtained on this scale. Carcass fat classes were transformed to their corresponding subcutaneous fat proportions (ESF) before analysis, (i.e. 1 = 0 04, 2 = 0 08, 3L = 0 11, 3H = 0 13, 4L = 0 15, 4H = 0 17 and 5 = 0 20; Kempster et al., 1986). Genetic correlations were estimated using bivariate animal model analyses for all combinations of goal and index traits. All covariance component analyses were performed using the variance component estimation (VCE) program (Groeneveld, 1996). The fixed effects used in the models for the estimation of ewe traits were farm/year/heft (grazing location) interaction, dam birth rank, genetic line, parity and age/farm interaction. In addition, for maternal weaning weight, the interaction of sex of lamb and the number of lambs reared were included. The fixed effects used in the models for the estimation of lamb traits were farm/year/heft (grazing location) interaction, genetic line, birth/rearing type, dam age/farm interaction, age at scanning, and sex. The genetic and phenotypic correlations used in the index calculations are shown in Table 3. Phenotypic correlations between carcass traits and all other traits were not required for the index calculations. The same parameters were used for each selection index. Table 2 Heritabilities (h 2 ), means, phenotypic standard deviations (σ p ) and repeatabilities (r) used in the index calculations Trait h 2 Mean σ p r Mature size (kg) Longevity (days) N/A Lamb loss (lambs per ewe) No. weaned (lambs per ewe) Average weaning weight (kg) Fleece weight (kg) Fat class (ESF) N/A Conformation score (units ) N/A Carcass weight (kg) N/A Lamb weaning weight (kg) N/A Muscle depth (mm) N/A Average fat depth (mm) N/A N/A; not applicable. Threshold units on the underlying normal distribution scale. The mean value of 3 for conformation score is transformed to 0 on the underlying normal scale. Estimated subcutaneous fat proportions.

4 416 Conington, Bishop, Grundy, Waterhouse and Simm Table 3 Correlations between breeding goal and index traits Mature Long- Fleece Average No. No. Fat Muscle Fat Carcass size evity weight WW reared lost depth depth WW class CS weight Mature size Longevity Fleece weight Average WW No. reared No. lost Fat depth Muscle depth WW Fat class CS Carcass weight Genetic correlations above the diagonal, phenotypic below. WW = weaning weight; CS = conformation score. Economic values Economic values for each breeding goal trait were derived using a whole farm model approach for each farm system, and have been described by Conington (1999). All economic values have been expressed on a per 100 ewe flock basis, and differ for each farm system. They were derived assuming that all other traits were held constant. For example, the costs associated with increasing maternal weaning weight were calculated independently of the costs associated with increasing lamb weaning weight directly. The index calculations in this paper are based on the assumptions that relative economic values remain constant over the change in trait means likely to occur with genetic progress, and they do not change over time. Gene flow In any population under selection, different numbers of animals express genetic improvements in different traits at different times. For example, females over the age of 2 years express litter size every year. Females over the age of one year express fleece weight every year but longevity is only expressed once in the animal s lifetime. Accounting for the flow of genes across time through the flock overcomes this problem when deriving economic values. Gene flow methodology (Hill, 1974) was used in this study to count the expression of genes across years, using matrices describing flock age structures and the flow of genes through the flock. Expressions of traits were obtained by multiplying the vector of gene distribution at each time point by a vector describing the expression of the trait category by each age class. The goal traits may be classified into four categories of traits: (i) lamb performance, (ii) maternal performance, (iii) wool and mature size, and (iv) longevity. Selection on an index will lead to genetic superiority in each of these categories expressed as follows: Lamb performance traits. Genetic superiority in these traits is expressed once by all year 1 lambs destined for slaughter. Also, half of the genetic superiority is expressed once by the slaughter progeny of daughters. Using typical flock age structures, this occurs from years 3 to 6. A quarter of the genetic superiority is expressed once by slaughter grandprogeny of females, from years 5 to 12, etc. Litter size, lambs lost, maternal weaning weight. These are expressed by one-crop ewes in year 3, two-crop ewes in year 4, three-crop ewes in year 5, and fourcrop ewes in year 6. Thus, each ewe has up to four opportunities to express maternal traits but each year only a proportion of the flock is expressing the genetic superiority resulting from earlier selection decisions. Daughters of these ewes express half their genetic superiority, etc. Wool and mature size. The timing of expressions is the same as those in the category above, but expression starts 1 year earlier. Longevity. Longevity is expressed only once per ewe, with the subsequent normal flow of genes through the population. It is assumed that it is expressed at the mean flock age. Discounting. Economic benefits promised in the distant future are perceived to be of less value than benefits immediately available. Future benefits are commonly discounted to take account of this time

5 Multi-trait selection indexes for UK hill sheep 417 effect. Assuming a discount rate of x, then future benefits occurring in year t are scaled by 1/(1 + x) t to express them as current values. To compare all trait categories in an equitable way, the expressions of the benefits were counted across years (assuming a 15-year time horizon), discounted appropriately (at 5%) and then summed to get a discounted expression of genes in the flock per trait ( discounted gene expression coefficient ). The discounted gene expression coefficients, summed over the 15-year time horizon for all lamb performance traits are and those for maternal traits are As measurements of wool and mature size occur prior to those for other maternal traits, the discounted gene expression coefficient for these traits is slightly higher at Longevity is only expressed once in the animal s lifetime and hence it is lower at The discounted gene expression coefficients were then multiplied by the economic values calculated from Conington (1999) to obtain economic weights. Table 4 shows the economic values, discounted gene expression coefficients and economic weights used in the index evaluations. Index calculations The selection index used the following assumed data structure for lamb traits: (i) a measurement on the lamb itself, (ii) a measurement on the dam of the lamb, and (iii) a measurement on 60 paternal halfsibs of the lamb. Given the lamb, dam and paternal half-sib information, knowledge of the sire performance contributes negligible information. The data structure for the traits measured on the dam (ewe) of the lamb used : (i) measurements on the ewe herself, (ii) measurements on the dam of the ewe, and (iii) measurements on 15 female paternal halfsibs of the ewe. It is assumed that for ewes each trait has an average of two recordings (in separate parities) except for longevity, which only has one measurement. Longevity is defined as the of ewe age (in years) at death or culling. This data structure mimics that for an average animal subjected to multi-trait BLUP analyses in the selection flocks, once selection has been underway for several years. It is important to note that the maternal traits are being predicted for the lambs that are candidates for selection, rather than for their dams. The selection index calculations were programmed using GENSTAT (Lawes Agricultural Trust, 1983) and based on index theory outlined by Cunningham (1969). The index matrices and vectors were: X = vector of phenotypic observations for each index (measured) trait; v = vector of economic weights for the goal traits; b = vector of weighting factors to be used in the index (economic value discounted gene expression coefficient); P = m m matrix of phenotypic covariances between the measured traits X. In this case, P has dimensions of 27 27, as the nine traits are measured on three groups of animals (animal, dam and paternal half-sibs); G = m n matrix of genetic covariances between the m measured variables and the n traits in the breeding Table 4 Economic values for the three farm systems, discounted gene expression coefficients and the economic weights for each goal trait Economic values Economic weights ( per 100 ewe flock) Discounted gene ( per 100 ewe flock) expression Trait Extensive Semi-intensive Intensive coefficient Extensive Semi-intensive Intensive Mature size (per kg) Longevity (per day) Fleece weight (per kg) Average weaning weight (per kg) No. reared (per 1% increase) No. lost (per 1% increase) Weaning weight (per kg) Fat class (per % unit ESF) Conformation (per unit) Carcass weight (per kg) Economic values taken from Conington et al. (2000). For example, increasing no. of lambs weaned per 100 ewe flock from 120 to 121. For example, increasing the number of lambs lost per 100 ewe flock from 13 to 14%. For example, increasing estimated subcutaneous fat from 10 to 11%.

6 418 Conington, Bishop, Grundy, Waterhouse and Simm goal. The dimensions of this matrix are 27 10; C = n n matrix of genetic covariances between the goal traits, (a matrix); D = selection differential on a standardized normal distribution. The general solution to the index equations is: b = P 1 Gv. The variance of the index is: b Pb. The expected monetary returns from the index are: b Pb and the accuracy of the index is: b Pb. v Cv To derive the annual progress in each trait, a selection intensity of 1 and generation interval of 2 5 was assumed. The genetic gain for each trait per generation is the regression of each trait in the aggregate genotype on the index, i.e. the ith element of b G. b Pb One round of selection on the index will produce D σi units of change in the index, and therefore a vector b G D b Pb of units of change in each trait. The contribution of each index trait to the selection objective was calculated by measuring the reduction in the efficiency of the index, the correlation between the true breeding value and the index after each trait was individually dropped from the index. Sensitivity analysis The sensitivity of the index to changes in the economic weights, proportionally 0 5 above and below the base values, was investigated. Each goal trait was changed individually, and the effects on the expected genetic progress in all the goal traits, and the accuracy of the new index were determined. The changes in economic weights necessary to alter expected progress in the desired direction for mature size and number of lambs reared are also reported. This was achieved by substituting different economic weights for these traits in the programme until responses in the desired direction were obtained. As heritability estimates for ultrasonic fat depth and weaning weight may differ according to the environment in which lambs were reared (Conington et al., 1995), the sensitivity of the index to a change in these heritabilities was investigated. The heritability for fat depth was reduced from 0 25 to 0 12 and that for weaning weight reduced from 0 22 to All sensitivity analyses were performed using the intensive farm index. As well as sensitivity analyses for changes in economic weights and heritabilities, some alternatives to the indexes that might be appropriate for some, but not all, hill breeds were investigated for the intensive farm system. The aim was to investigate the consequences of: (i) excluding fleece weight from the breeding goal, (ii) leaving fleece weight in the breeding goal but not measuring it, and (iii) using fat and muscle measurements on the ewe, rather than on all lambs. The last option (iii) would provide more flexibility in particular for farms where lambs are relatively immature at the time of ultrasonic measurements (17 weeks), and show little variation in fat depths. It may also prove to be a cheaper and more practical alternative to scanning all lambs. The different parameters used for exploring this last option are as follows: heritability estimates, phenotypic standard deviations and repeatabilities for ewe fat depth and ewe muscle depth are 0 30 and 0 40, 1 8 mm and 2 9 mm and 0 45 and 0 51 respectively. Genetic correlations between mature weight and ewe fat and ewe muscle depths used were and respectively. Additionally, genetic correlations were and between fleece weight and ewe fat and ewe muscle depths respectively and between ewe fat and ewe muscle depths. The phenotypic correlation between ewe fat and ewe muscle depth used was and those between weaning weight and ewe fat and ewe muscle depths were and respectively. Results Progress in goal traits Predicted annual progress in each goal trait for the three farm types is shown in Table 5. In general, greater responses are seen in the intensive farm system for all traits compared with the semiintensive and extensive farm systems. With the exception of conformation score and perhaps mature

7 Multi-trait selection indexes for UK hill sheep 419 Table 5 Expected annual genetic change in each goal trait in the three farm systems Extensive Semi-intensive Intensive Mature size (g) Longevity (days) Fleece weight (g) Average weaning weight (g) No. of lambs reared per ewe No. of lambs lost per ewe Weaning weight (g) Fat class (ESF) N/A Conformation (units ) N/A Carcass weight (g) N/A N/A: not applicable because it is assumed that a store lamb producer would not benefit directly by changes in carcass traits. Threshold units on the underlying normal distribution scale. size, all predicted responses are in the desired direction. However, the changes in all responses as a proportion of mean values are modest, ranging between to 0 04 per generation ( to per annum). All weight traits (mature size, maternal weaning weight, direct weaning weight, carcass weight) are predicted to increase, despite mature size having a negative economic weight. Traits which are conventionally more difficult to improve, such as the number of lambs reared, number of lambs lost and longevity show smaller proportional responses. The expected response in number of lambs reared in the intensive is 1 15 times greater than the extensive farm system. No corresponding increases in the number of lambs lost are predicted. Small improvements (i.e. reductions) in fat class are predicted and conformation score is predicted to decrease by a negligible amount. In practice, these carcass quality traits will be held more or less constant whilst improvements in carcass weight are achieved. Contribution of traits in the index The importance of each measured trait in the overall index is shown in Table 6. The figures shown represent the genetic improvement obtained if a particular measurement was dropped from the index, compared with the genetic gain achieved by the full index. Therefore, the smaller the value, the more important that measurement is and the more it contributes to the index. For example, if mature size is dropped from the intensive index, the overall rate of progress is predicted to be proportionally 0 67 of that possible from selection on the full index. The results show that mature size is the most important trait for all three indexes. This is despite a negative economic weighting on mature size, and it is due to the relatively high correlations between mature size and the other weight traits, as well as the high heritability estimate (0 47) for mature size. Fleece Table 6 Relative contribution of each trait to the index, in the three farm systems Extensive Semi-intensive Intensive Mature size Longevity Fleece weight Average weaning weight No. of lambs reared No. of lambs lost Fat depth Muscle depth Fat + muscle depths Weaning weight The smaller the value, the more the trait contributes to the index. The efficiency of the index r TI is the correlation between the index and the true breeding value. The data presented show the relative efficiency of the index, measured by the reduction in r TI resulting from dropping each variable from the index.

8 420 Conington, Bishop, Grundy, Waterhouse and Simm weight, number of lambs lost and the carcass quality traits are the least important traits for the intensive farm system. With the exception of mature size, the number of lambs reared and weaning weight, most of the traits have a small effect on the index and, if removed individually, result in a reduction in the efficiency of the index of proportionally between 0 03 and Accuracy In general, when traits have low heritabilities, or breeding objectives conflict, the accuracy of the index is lower than if traits have higher heritabilities and are favourably correlated. Having many traits in the breeding goal increases the chance that some unfavourable correlations will exist among them. The accuracy for the intensive index is 0 41 and that for the semi-intensive and extensive farm types is only slightly lower at Monetary returns Selection indexes provide the optimal selection emphasis on different traits, based on their relative economic importance. They do not answer the question of whether a breeding programme is economically justified. To help breeders answer that question, the expected monetary return from using the selection indexes was determined. As the benefits from continuous genetic improvement are cumulative, these benefits become greater after several years of selection. The expected benefits of selection for a 100 ewe flock are 348, 238 and 220 per generation for the intensive, semi-intensive and extensive farms respectively, using the economic weights applied here which describe returns over a 15-year time horizon. Although implicit from the economic values, these figures confirm that the potential to improve farm efficiency is greater for the more intensive hill farms. Sensitivity analysis Proportional changes of ±0 5 of the economic weight of each goal trait were made to investigate the sensitivity of the intensive farm system index. The majority of the changes in economic weights resulted in proportional changes in each trait response of less than ±0 10. However, some exceptions to this rule were seen. The greatest change in predicted response was seen for fat class (a proportional decrease of 0 61), when the economic weight for the number of lambs reared was halved, and for longevity, which fell by proportionally 0 42 in response to a decrease in the economic weight of the number of lambs lost from 42 6 to The same reduction in the economic weight for the number of lambs lost led to proportional reductions in responses in weight traits (mature size, weaning weight and carcass weight) of between 0 20 and Changing the economic weight for carcass weight had a similar effect on the predicted response of fat depth to that of changing the economic weight for weaning weight. Although the change was in the same direction, it was smaller in magnitude. The accuracies of the indexes with the different economic weights were also computed as a result of the sensitivity analyses. In general, they were similar to the optimum index (0 41) but when the economic weight for weaning weight was , the accuracy rose to 0 44, the highest accuracy of all. When the economic weight for average weaning weight was 36, the accuracy went down to 0 37, the lowest accuracy. The effect of increasing the magnitude of the negative economic weight on mature size to constrain its increase was investigated. The expected progress in each trait and the accuracy of the index when the economic weight is 100 and when it is 155 were used. These economic weights were chosen after successively reducing the economic weight of mature size to substantially affect responses in mature size. When the economic weight for mature size is 100, mature size still increases but at a lower rate than in the original index. Responses in carcass weight, weaning weight and average weaning weight are all predicted to decline by up to two-thirds, with an accuracy of When the economic weight for mature size is 155, no increase in this trait per annum is predicted. Weaning weight and carcass weight are predicted to show only trivial increases (21 g and 13 g respectively) with a significant improvement in the number of lambs lost. The accuracy of this index is reduced to When the heritability for weaning weight was reduced from 0 22 to 0 11, the predicted responses in most traits changed. In particular, the expected response in weaning weight almost halved from 371 to 177 g per annum, that for carcass weight reduced from 121 to 93 g and that for mature size reduced from 654 to 552 g. Index options Excluding fleece weight from the breeding goal. Improving fleece production is currently not a priority for breeders of some hill sheep breeds, in particular for coarse wool breeds such as the Scottish Blackface. Including fleece weight in the breeding goal may therefore be an obstacle to uptake of these selection indexes. Index calculations were redone for the intensive farm system excluding fleece weight from the breeding goal. The results show that fleece

9 Multi-trait selection indexes for UK hill sheep 421 weight increases by proportionally 0 66 of the original response. This is due to the correlations between fleece weight and the other components of the breeding goal, in particular that of mature size. The predicted responses in weight traits are all predicted to increase by a factor of between 0 05 and The discounted returns of are less than the original index, and the accuracy is 0 39 (compared with 0 41). Not recording fleece weight, but retaining fleece weight as a goal trait. The physical and financial effort of recording weight of fleece at clipping could be a barrier to its improvement. Including it in the breeding goal but not recording it, will give a more reliable estimate of the economic benefit of the index compared with excluding it. The responses are very similar to those from taking fleece weight out of the goal altogether and the annual discounted returns are higher, being only 9 30 less than the original index. The accuracy is The expected responses in individual goal traits as a result of either leaving fleece weight out of the breeding goal, or not recording fleece weight are reported by Conington (1999), along with individual results for all the sensitivity analyses. Measuring fat and muscle on the ewe. Another modified index was investigated using ultrasonic measurements of fat and muscle from the ewes prior to mating, instead of measuring lambs at weaning. The results from this index shift the emphasis in favour of most carcass traits, at the expense of some of the maternal traits. In particular, a proportional increase of 0 65 in mature size above the original increase of 654 g per annum may be an unacceptably high price to pay for using this index. This index shows extra improvements in weaning weight, average weaning weight, carcass weight and conformation, whilst holding fat class and number of lambs lost more or less constant (with a tendency for them to increase). This index has an expected annual monetary benefit of and an accuracy of Discussion The results from this work show that multi-trait selection indexes can theoretically improve flock productivity and economic returns for UK hill sheep for a range of different production levels and production systems. Improvements in maternal characteristics can be achieved alongside increasing lamb weaning and carcass weights, with more modest changes in lamb carcass quality traits. Lamb survival, as a trait of the dam, has been shown to be both repeatable and heritable (Atkins, 1980; Donnelly, 1982; Haughey and George, 1982; Piper et al., 1982; Matos et al., 2000) and increasing lamb survival through genetic improvement would help to alleviate the seemingly intractable problem of perinatal mortality (Haughey, 1983). It is also more cost-effective to farm ewes with the ability to rear more of their lambs, rather than increasing prolificacy per se. This is because the costs of supplementary feeding ewes that are carrying higher litter sizes during winter would be lower if the selection emphasis were to be placed on ewes ability to rear lambs they have already given birth to, rather than increasing litter size born. Improving longevity has been documented as being important for livestock profitability and evaluations of this trait are currently used in breeding programmes in the UK for dairy cows (Veerkamp et al., 1995; Brotherstone et al., 1997). This is the first time that it has been considered in the breeding goal for UK sheep. Improving longevity of the flock with the multi-trait index ensures that higher flock productivity will not be achieved at the expense of shorter ewe lifespan. Recording ewe longevity and the number of lambs lost in breed improvement programmes is inexpensive relative to the benefits gained through genetic improvement. The small responses in fat class, in the desired direction, and a small response in conformation in the undesired (i.e. poorer) direction, reflect the low heritabilities for these traits, the very weak correlation between muscle depth and conformation score and the positive (i.e. unfavourable) genetic correlation between conformation and fatness. Even though farmers in the UK are paid on the basis of these two classification scores (as well as carcass weight), true carcass meat value to the processor could probably be determined if objective measurements of fat and muscle mass were defined as goal traits. These goal traits could be predicted by either ultrasound or computed tomography. The results from these indexes show that the importance of carcass quality traits for hill flocks in terms of genetic improvement is less than is generally perceived in the industry because the selection criteria are not very effective (conformation score and carcass fat class) and there is not much genetic variation in ultrasonic fat depth. However, they remain important from both a farm and strategic point of view. Firstly, it is important to increase lamb carcass weight without a corresponding increase in fat, so that the desired carcass grades are achieved. Secondly, a hill sheep industry that is breeding to improve carcass quality will benefit all levels of the stratified breeding pyramid and, ultimately, the consumer. However, the results from the index calculations show that the current industry payment structure for lamb carcasses is inadequate to provide

10 422 Conington, Bishop, Grundy, Waterhouse and Simm clear signals to the farmer for reducing fatness in meat, in line with current government objectives. This has also been discussed by Lewis et al. (1996) who suggested that progress in reducing fat in market lambs is being slowed by the current carcass assessment system in the UK. The differences among the three farm types show that the monetary returns and, in general, the economic weights for all traits, are smaller for the extensive, than the other two farm systems but the predicted responses in goal traits are not. The higher monetary gain from the intensive index reflects the greater output per ewe from intensive systems. The intensive index predicts slightly greater progress in lamb production traits compared with the other two farming systems. The alternative selection indexes described for the intensive hill farm situation explore the implications of (i) different breeding goals (omitting fleece weight), and (ii) using a different strategy to meet the same goals (by not measuring fleece weight, or by using ewe ultrasonic measurements of fat and muscle). Using indexes which achieve less than optimum economic response may well be the most profitable to the breeder, if the cost of implementation is less than that of the optimum index. For example, measuring fat and muscle on the ewe instead of on the lamb removes the need to measure all lambs at, or just prior to, weaning. As weaning is normally in August for hill flocks in the UK and the major store and breeding lamb sales commence in September, lambs are frequently sorted into groups for slaughter, for store sales or for breeding at this time. Thus measuring them at this stage leaves little time for evaluation and selection. Recording fat and muscle on lambs much earlier for many hill flocks is inappropriate, as lambs may only be gathered once or twice after lambing, and they may be too immature to have measurable variation in fatness. Using the index based on measuring fat and muscle on the ewe may well be the most economic selection index, at least for the breeder. At a current cost of 2 50 per scan, and based on the numbers of lambs reared per 100 ewes for the intensive farm type (120), compared with the number of replacement ewes (29), the difference in cost between using the original index and that using measurements of fat and muscle on the ewe, is 227 in favour of using the index based on measuring fat and muscle on the ewe. However, this index predicts a high response in mature weight and perhaps may not be recommended to the industry in general. The economic benefits from genetic improvement depend on the level of production of individual farms when embarking on genetic improvement programmes, together with their ability to accommodate increased productivity on the farm. In some instances, unrealistic breeding objectives, such as increasing the number of lambs reared beyond the capacity of the farm to provide suitable grazing, results in negative economic weights and returns from genetic improvement in some hill situations (Conington, 1999; Conington et al., 2000). The use of these indexes is expected to lead to an increase in the mature size of the sheep. For some flocks, this may not be acceptable. However, improvements in lamb weaning and carcass weights, as well as the predicted economic returns, would be smaller if the increase in mature size was constrained through the use of a desired gains, or target production index approach (Newman and Ponzoni, 1994). The results in this study from applying a greater negative economic weight on mature size confirm that lamb weaning weights and carcass weights will not increase if mature size is constrained. These results highlight the need for a flexible approach to breeding objectives and breed improvement in hill flocks to suit the farm production system and the physical limitations to genetic improvement. Such flexibility may well pave the way for a greater involvement by the hill sector in genetic improvement programmes. The indexes described here are a first step towards improving sustainability and flock performance simultaneously, through genetic selection. By including lamb losses in the breeding goal, the wastage of lambs from birth to weaning will be addressed. Including longevity in the breeding goal will reduce ewe wastage as ewes will live longer and fewer replacement ewes will be required. This paper predicts that gains in productivity can be made without compromising ewe or lamb survivability. In the future, other traits important for improved sustainability of the sheep system, such as improving resistance to parasites, lambing ease and maternal behaviour important for neonatal survival, could be considered for inclusion into wider breeding goals for hill sheep. A selection index which identifies sheep which live longer, lose fewer lambs, are resistant to disease, have good maternal behaviour and few lambing problems, may well appeal to flock owners whose custodial rôle of hill environments is likely to become increasingly important in the future. Acknowledgements The authors gratefully acknowledge the Meat and Livestock Commission, Scottish Executive Rural Affairs Department and the British Wool Marketing Board for funding this research. Support from MAFF is also acknowledged. The technical support of Maureen Shanks,

11 Multi-trait selection indexes for UK hill sheep 423 Nicola Lambe, Sue Murphy and Tom Baillie is most appreciated, as is the constructive criticism from Peter Amer. References Atkins, K. D Selection for skin folds and fertility. Proceedings of the Australian Society of Animal Production 13: 174. Brotherstone, S., Veerkamp, R. F. and Hill, W. G Genetic parameters for a simple predictor of the lifespan of Holstein-Friesian dairy cattle and its relationship to production. Animal Science 65: Conington, J The genetic improvement of carcass and maternal traits in Scottish Blackface sheep. Ph.D. thesis, University of Edinburgh. Conington, J., Bishop, S. C., Waterhouse, A. and Simm, G A genetic analysis of early growth and ultrasonic measurements in hill sheep. Animal Science 61: Conington, J., Bishop, S. C., Waterhouse, A. and Simm, G A comparison of growth and carcass traits in Scottish Blackface lambs sired by genetically lean or fat rams. Animal Science 67: Conington, J., Bishop, S. C., Waterhouse, A. and Simm, G A bio-economic approach to estimating economic values for UK hill sheep. Proceedings of the British Society of Animal Science, 2000, p. 39 (abstr. ). Cunningham, E. P Animal breeding theory. Internordic licentiat course notes in quantitative genetics, Norway. Donnelly, F. B A practical attempt to breed for better lamb survival. Proceedings of the Australian Society of Animal Production 14: 30. Groenveld, E REML VCE a multivariate multi model restricted maximum likelihood (co)variance estimation package version 3 2 users s guide. Federal Research Centre of Agriculture, Mariensee, Germany. Haughey, K. G Selective breeding for rearing ability as an aid to improving lamb survival. Australian Veterinary Journal 60: Haughey, K. G. and George, J. M Lifetime rearing performance of Merino ewes and its relationship with pelvic size and early rearing status. Proceedings of the Australian Society of Animal Production 14: 26. Hill, W. G Prediction and evaluation of responses to selection with overlapping generations. Animal Production 18: Kempster, A. J., Cook, G. L. and Grantley-Smith, M National estimates of body composition of British cattle, sheep and pigs with special reference to trends in fatness: a review. Meat Science 17: Lawes Agricultural Trust GENSTAT, a general statistical program. Numerical Algorithms Group Ltd, Oxford. Lewis, R. M., Simm, G., Dingwall, W. S. and Murphy, S. V Selection for lean growth in terminal sire sheep to produce leaner crossbred progeny. Animal Science 63: Matos, C. A. P., Thomas, D. L., Young, L. D. and Gianola, D Genetic analyses of lamb survival in Rambouillet and Finnsheep flocks by linear and threshold models. Animal Science 71: Newman, S. and Ponzoni, R. W Experience with economic weights. Proceedings of the fifth world congress on genetics applied to livestock production, Guelph, vol. 18, pp Piper, L. R., Hanrahan, J. P., Evans, R. and Bindon, B. M Genetic variation in individual and maternal components of lamb survival in Merinos. Proceedings of the Australian Society of Animal Production 14: 26. Simm, G. and Dingwall, W Selection indices for lean meat production in sheep. Livestock Production Science 21: Veerkamp, R. F., Hill, W. G., Stott, A. W., Brotherstone, S. and Simm, G Selection for longevity and yield in dairy cows using transmitting abilities for type and yield. Animal Science 61: (Received 23 November 2000 Accepted 18 June 2001)