Horn growth rate and longevity: implications for natural and artificial selection in thinhorn sheep (Ovis dalli)


John Loehr, University of Jyväskylä, Department of Biological and Environmental Science, PO Box 35, 40014, Jyväskylä, Finland.
Tel.: +358 14 260 1211; fax: +358 14 260 2321;
e-mail: johloeh@cc.jyu.fi


We used horn measurements from natural and hunted mortalities of male thinhorn sheep Ovis dalli from Yukon Territory, Canada, to examine the relationship between rapid growth early in life and longevity. We found that rapid growth was associated with reduced longevity for sheep aged 5 years and older for both the hunted and natural mortality data sets. The negative relationship between growth rate and longevity in hunted sheep can at least partially be explained by morphologically biased hunting regulations. The same trend was evident from natural mortalities from populations that were not hunted or underwent very limited hunting, suggesting a naturally imposed mortality cost directly or indirectly associated with rapid growth. Age and growth rate were both positively associated with horn size at death for both data sets, however of the two growth rate appeared to be a better predictor. Large horn size can be achieved both by individuals that grow horns rapidly and by those that have greater longevity, and the trade-off between growth rate and longevity could limit horn size evolution in this species. The similarity in the relationship between growth rate and longevity for hunted and natural mortalities suggests that horn growth rate should not respond to artificial selection. Our study highlights the need for the existence and study of protected populations to properly assess the impacts of selective harvesting.


Do male characteristics that aid in mating success also result in improved viability? This is a central question in the study of sexual selection (Kokko et al., 2003; Hunt et al., 2004) and has direct implications for life history theory (Stearns & Koella, 1986). Because life history and sexual selection are often interlinked, the study of how they affect each other can provide interesting evolutionary insights (Kokko, 1997; Höglund & Sheldon, 1998). In males traits that confer a mating advantage can bear survival disadvantages (Clinton & Le Boeuf, 1993), yet this may be counteracted by differences in male ‘quality’ (e.g. Nur & Hasson, 1984; Papadopoulos et al., 2003). If sexual and viability selection point in different directions, a larger set of males will have a genetic contribution to future generations than if sexual and natural selection both favoured the same type of males (Gray & Cade, 1999). Several studies have investigated male life history strategies when sexually selected traits trade-off with traits required for survival (e.g. sticklebacks: Candolin, 2000; scorpionflies Panorpa cognata: Engqvist & Sauer, 2002; Drosophila melanogaster: Cordts & Partridge, 1996; flycatchers Ficedula albicollis: Gustafsson et al., 1995; field crickets Teleogryllus commodus: Hunt et al., 2004). However, such work has typically considered species in which males display to females, and females subsequently choose their mates. Less attention has been paid to settings where life history trade-offs operate between viability and traits related to success in male–male competition (but see Stevenson & Bancroft, 1995). Here we test whether horn growth rate is positively or negatively associated with longevity in thinhorn sheep (Ovis dalli).

North American mountain sheep (Ovis spp.) provide excellent study organisms to investigate the relationship between growth rate and longevity. Mountain sheep rams grow large horns that serve as display, defence and attack organs in male–male conflicts (Geist, 1966a). Annual growth rings visible on horns make possible the very precise measurement of growth rates. Size of horns is associated with mating success (Coltman et al., 2002), and the onset of sexual reproduction is thought to be associated with higher mortality rates (Jorgenson et al., 1997).

Relevant to our research are two hypotheses regarding horn growth rate for mountain sheep. The first hypothesis proposes a trade-off between horn growth rate and longevity (Geist, 1971). In a horn collection study of an unhunted bighorn sheep O. canadensis population, Geist (1966a) found that sheep with rapid horn growth died at an earlier age than those with slower growth. Geist (1971) proposed that this trade-off could result in selection for rapid growth and longevity balancing each other out. However, this study was based on a small sample of 40 horns gathered over a short time period. Because it was unknown to what degree horn growth could be affected by environmental factors, and birth years were not presented (or known) for the sheep, it is difficult to assess whether this interpretation was confounded by a cohort effect. The second hypothesis is an assumption inherent to the work of Coltman et al. (2003, 2005) that proposes that males with rapid horn growth have an annual reproductive advantage and would also survive longer than males with slow growth. Although such a system could result in strong positive directional selection on horn growth rate, selection could be limited by covariance with many other traits under selection. This creates a dispersed target for selection and increased possibility for deleterious mutations to occur, which could allow for maintenance of genetic variance.

The relationship between growth rate and longevity also has broader implications for the study of human-induced selection on harvested populations (Palumbi, 2001; Ashley et al., 2003; Ernande et al., 2004; Olsen et al., 2004). Mountain sheep are particularly interesting in this respect because hunting regulations for adult males are frequently based on degree of horn curl and thus have a phenotypic bias. This creates a situation in which rams that grow horns quickly can be shot at an earlier age than rams that grow horns slowly. Such selective harvesting is especially a concern if it is phenotypically biased towards a trait that is highly heritable such as horn growth rate (Coltman et al., 2003). An evolutionary response because of phenotypically biased harvesting has been well documented in a controlled experiment in the Atlantic silverside (Menidia menidia) (Conover & Munch, 2002). In this case larval growth rate either increased or decreased based on the phenotypic bias of the harvest. Studies of wild populations have also suggested that phenotypic harvesting may result in an evolutionary response in horn growth rate in bighorn sheep O. canadensis (Coltman et al., 2003) and a change in maturation trends in northern cod Gadus morhua (Olsen et al., 2004).

In this study, we test whether rapid horn growth rate early in life is associated with increased or reduced longevity in thinhorn sheep O. dalli. If rapid growth is associated with a survival advantage we should find a positive correlation between growth rate and age at death in unhunted populations. This trend should be opposite to that which we find among hunted mortalities where growth rate should be negatively associated with longevity as a result of morphologically biased hunting regulations. If these predictions hold true then selective hunting will produce the opposite trend in longevity than what is expected in populations with no hunting. Such a result would be strong evidence to support the idea that the harvest system based on horn curl can result in an evolutionary response in growth rate. However, a second possibility is that rapid horn growth is also associated with reduced longevity in populations with no hunting. In this case the selective effect of hunting on growth rate may be limited if the harvesting regulations merely mimic trends in natural mortality.

Materials and methods

Analysis of horn measurements was undertaken using Yukon Territory government records from horns that were found 1969–2005 and hunted sheep 1980–2002. Hunted and natural mortality horns were gathered from the approximately 22 000 sheep in southern and central Yukon Territory, Canada (60–66°N, 132–141°W). Samples from Kluane National Park (KNP), Yukon, are from Sheep Mountain (61°0′N, 138°3′W) which is a population of about 220 individuals (Hoefs & Bayer, 1983).

Sheep can be legally hunted in Yukon Territory by resident and nonresident hunters once their horns have reached full curl or they are older than 7 years. A few individuals are also harvested each year as part of a subsistence hunt (not limited by sex or morphological characteristics) by aboriginal hunters. Although some sheep were harvested legally or illegally before full curl, nearly all sheep harvested are full curl individuals (96% of registered kills). Harvest rate in Yukon Territory measured from 1980 to 2002 is approximately 27% of legal rams (Yukon Government, unpublished).

In preliminary analysis we considered both horn length (from third to fifth annulus) and horn volume (from first to fifth annulus) for our measurement of growth rate. Horn measurements of bighorn sheep have usually been made using length, with the the first two annuli frequently omitted, because they can be worn or broken off over time. For our sample of thinhorn sheep, we found that both measurement methods produced similar results, however, length explained a lower proportion of variance in both the natural and hunted mortality samples. Horn volume has the advantage of better representing the total amount of horn that an individual grows because it takes into account the thickness of the horn, which is an important factor in mountain sheep collision conflict (e.g. Geist, 1971). Horn volume has been demonstrated to better reflect environmental fluctuation than horn length (Hik & Carey, 2000). Horn volume is also a more accurate measure of somatic investment than length.

Horn collection considerations

We initially had horns from 188 individuals available for analysis (see Table 1). We divided horns into four categories depending on where and what type of populations they were gathered from as well as knowledge of cause of death. For analysis of growth rate and age at death horns from 65 individuals were used from KNP. In KNP no hunting is allowed and a natural cause of death could be assumed in cases where only horns were found and cause of death could not be determined. In addition to this, data from 30 horns gathered from areas of Yukon Territory where hunting does not occur or is very infrequent were also used. These horns were gathered by the general public (GP) or conservation officers. Of the 30 individuals, a natural cause of death was confirmed in seven cases after inspection of carcases by conservation officers. We further divided these 30 individuals into two groups: the first group was composed of sheep (n = 16) from populations that had not been hunted in the 10 years prior to the horns being found, and sheep (n =3) that were known to have died of natural causes in populations where one or two sheep had been harvested by subsistence hunters (not limited by sex or horn morphology) in the 10 years prior to the horns being found. The second group was composed of sheep (n = 11) for which natural mortality could not be confirmed from a carcass yet natural mortality seemed very likely because they were from populations that had very little hunting activity. In these cases a minimum of one sheep and a maximum of four sheep were harvested from the population in the 10 years prior to each individual's horns being found. Our sample comes from horns gathered evenly over 36 years making the potential for cohort influence small. Horns were gathered from 11 different populations in 24 of the 36 years, and on average horns of 3.8 ± 0.6 (SE) individuals were gathered per year.

Table 1.   Description of horn collections used in analysis. Natural mortality analysis included horns from collections 1–3, for which natural mortality was known or could be reasonably assumed. Collection 4 was from populations that were frequently hunted, and was only used to calculate the effect of horn wear on volume calculations and assess the collecting habits of the general public vs. biologist. See Materials and methods for more information.
Horn collection originNo. individualsNo. populationsHarvest in population for 10 years prior to horns entering collection
Natural mortality
1. Kluane National Park651
2. Populations with no hunting (*) or very limited hunting and natural death confirmed (**)  16*4*
   3** 2**1–2**
3. Populations with very limited hunting but natural death not confirmed11 41–4
4. Frequently hunted populations93606–90
Hunted mortality
5. Very frequently hunted populations32976222–90

Horn collection studies have been criticized because of the opportunities for bias they present (e.g. Murphy & Whitten, 1976). Most of these criticisms concern the construction of synthetic cohort life tables from horn collection data because to be accurate an assumption is required that age structure remains constant over time. Here we compare growth rate of sheep across age groups which does not require this assumption. Relevant to our study, however, is the assumption that a random sample of all horns is present in the collection. For example, horn collection studies have been criticized for comparing ewes and rams because ewe horns are much smaller in size and more difficult to find (Murphy & Whitten, 1976). For natural mortalities we only used horns of sheep older than 4 years. We assume that at 5 years and older the size bias for finding horns is nonexistent because horns found for sheep aged 5 and older are large enough that all growth rates would be found in the sample. In their fifth year ram horns are on average 70.6 cm (± 8.4 SD). We checked for a minimum size threshold under which horns may not be found, however, we found that horns well under the minimum size for 5-year-olds have been found (Fig. 1).

Figure 1.

 Horn length and age at death for all thinhorn sheep natural mortality horns (collections 1–4; see Table 1) collected in Yukon Territory, Canada. Circles are those collected by the general public, and triangles are those collected by a biologist (M.H.).

We were concerned that there may be differences in the size of horns gathered by the GP or by the biologist. Sixty-three of the 65 horns found in KNP were gathered by a biologist (M.H.) for scientific research, and horns from outside the park were gathered by the GP. For this comparison, to improve the sample size of the GP horns, we also added GP horns found in hunted areas (Table 1, collection 4) to the analysis. If GP is selective in gathering habits it is highly unlikely that selectiveness in horn growth rate differs between hunted and unhunted areas. The biologist and GP gathered similar horn size at death as seen in Fig. 1. A more exact comparison may be found by comparing the growth rates of horns found by GP with the growth rates of horns found by the biologist for sheep aged 5 and 6. For these young age classes GP may discard smaller than average horns and keep larger ones. However, we found no evidence to suggest that GP preferred horns of greater growth rate: biologist (mean ± SD: 1277.5 ± 387.5 cm3, n = 8, minimum =808.2, maximum = 1706.8) and GP (1256.6 ± 349.9 cm3, n = 19, minimum = 395.3, maximum 1908.1). The minimum horn growth rate gathered by GP for a 5-year-old sheep ranked as the second lowest growth rate for any horn found regardless of age.

Finally, bias can be entered into horn collection data if there are systematic differences where individuals are found (Murphy & Whitten, 1976). This is not the case in our data; until about 3 years of age rams live in habitat more often used by ewes, but then move to ram habitat (Geist, 1971; Festa-Bianchet, 1991).

Measuring age and horn volume

The age of each sheep was determined by counting annular growth segments (Geist, 1966b), and the basal circumference and length of each segment was also measured. Lengths were measured using a flexible tape measure placed along the frontal surface of the longer horn. Measurements were made by conservation officers, wildlife biologists and technicians using standard methodology (Merchant et al., 1982; Barichello & Hoefs, 1984). To calculate horn volume, the volume of the first annular segment was calculated as a cone and the remainder of annuli segments as conical fustra (Heimer & Smith, 1975; Hik & Carey, 2000).

Statistical analyses

Linear models were constructed using the spss 10.0 statistics package (SPSS Inc. Chicago, IL, USA). We included the following variables in analysis. (i) Horn growth rate: measured in volume from the first five summers of horn growth. (ii) Horn wear: a dummy variable which either identified horns which still had all growth annuli intact or horns that had the first (28% of cases) or second summer (0.4% of cases) of growth worn off. We use this variable to control for the effect of horn wear on growth rate estimates. Bunnell (1978) recorded growth of Dall's sheep horns in KNP from live rams in their first year of life. Using these length measurements we calculated that the first summer of growth only represents about 1.0% of total horn growth in the first 5 years, and 0.4% of total horn volume at death. (iii) Cohort growth rate index: cyclical horn growth fluctuations occur along with a decadal climate cycle (Hik & Carey, 2000). We calculated the average growth for each birth cohort in the first five summers of growth to create a growth rate index for hunted horns. (iv) Age: age of sheep in years. (v) Horn collection: to assess whether there was any systematic differences between horns gathered in (a) KNP, (b) populations with no hunting or where natural mortality could be confirmed in populations with very little hunting pressure, and (c) populations where limited hunting occurs but natural death could not be confirmed (also see Table 1). (vi) Population: based on designated game management subzones in Yukon Territory. On average each subzone occupies about a 30 × 30 km area. This variable was entered in linear models for hunted horns only.

Based on values from hunted horns we found that horn growth rate is about 10% slower in northern Yukon (Ogilvie/Mackenzie Mountains) than in southern Yukon (Pelly/Coast Mountains) (linear regression, anova: F1,3297 = 23.8, independent variable ‘horn wear’β =−85.4, SE = 17.5, β coefficient = −0.09, P < 0.001). To account for this difference in the natural mortality horns we adjusted the growth rate values of horns found in northern Yukon (N = 9) based on a regression equation calculated from the hunted horns (northern growth rate = raw horn growth + 103.2).

We had three horn collections suitable for analysis of natural mortality (Table 1). To demonstrate the effect that combining horn collections had on our results we analysed horns from KNP first and then added collections 2 and 3 to subsequent analyses. By adding more horns to analysis we greatly increased the statistical power of our analysis. We calculated statistical power using the program gpower (Faul & Erdfelder, 1992), and assumed an effect size of 0.075 (based on a predicted R2 of 0.07), and α = 0.05. For n = 65, statistical power was calculated as 0.57. Power was improved to 0.76 for n = 95.

We were able to broadly control for growth rate differences between northern and southern Yukon (see above), however, we wanted to further determine to what degree the variance contained in population and cohort could have on our result from natural mortality. To do this we used a Monte Carlo simulation (Crowley, 1992) to test whether our results can be explained by random variation in cohort or population. Because Yukon Territory has kept very precise records of hunted sheep horn growth over a long time period it allows us to know what differences in populations and cohorts can exist. We can then apply this knowledge to ask the question: Is our result from natural mortality simply because of chance and explainable by our inability to directly control for variance introduced through cohort and population?

In the Monte Carlo simulation model we held growth rate constant for all individuals across age groups and randomized the relationship between individuals and populations based on our knowledge of the magnitude of cohort and population fluctuation. We defined the variance in horn growth rate as the additive effects of cohort and population. For cohort fluctuation we generated random numbers with a mean of zero from a uniform distribution limited by the maximum and minimum mean cohort growth rates generated by the decadal environmentally mediated cycle. This cycle has been confirmed to exist in sheep horns at least since 1963 (Hik & Carey, 2000; Yukon Government, unpublished). To simulate differences between population growth rates we generated random values for each of the 11 populations where horns were found. From the hunted horn data set we measured growth rate for 62 well-sampled populations from across Yukon Territory in the same geographical areas as the natural mortality sample. This sample contained 3297 hunted rams. Based on the hunted sheep data we knew that the mean growth rates of populations were normally distributed (Kolmogorov–Smirnov = 0.070, d.f. = 60, P = 0.200) with a standard deviation of 96.7. Using the standard deviation we generated 11 random numbers (one for each population in the natural mortality dataset) for each iteration of the model from a normal distribution with a mean of zero. We also probed the data for the possibility that magnitude of cohort fluctuation differed with population mean growth rate or age group but could find no trends to suggest this. The model was run for 1000 iterations. We calculated the one-tailed P value from the number of simulations that exceeded the value of F that we attained in the regression test of the natural mortality horns. The P value of the model was one-tailed because it was only necessary to consider those simulations that exceeded the F value and had a negative β value.

Age structure and population density may affect the relationship between growth rate and longevity. Age structure and density has undoubtedly fluctuated to some degree during our study, although we do know that in KNP these variables were basically constant for the time period of our study (Hoefs & Bayer, 1983, Parks Canada, unpublished). Coltman et al. (2003, 2005) studied the relationship between horn growth rate and longevity over a 30-year period in bighorn sheep. In their analysis the age structure and density of the population at the time of death was not accounted for. Age structure and density fluctuated during the work of Coltman et al. (2003, 2005) and horn curl hunting regulations were modified. These factors, in addition to variations in hunting effort could result in the strength of the cost of rapid growth differing from year to year. However, this approach is reasonable because the intention is to achieve an overall indication of the relationship between growth rate and longevity for a population that experiences fluctuations in factors relating to the variables of interest.


For the natural mortality sample we first controlled for the effect of horn wear on our measurement of growth rate using all natural mortality horn collections (Table 1, collections 1–4) gathered in the field (linear regression, anova: F1,185 = 5.9, independent variable ‘horn wear’β = −115.9, SE = 47.8, β coefficient = −0.18, P < 0.05). We then continued with analysis and tested whether the relationship between growth rate and longevity was positive or negative (Table 2). We first considered only the samples gathered from KNP (collection 1), and then added collections 2 and 3 in succession to analysis. For all tests the relationship between growth rate and longevity was negative. Statistical power was low for initial tests and the result for collection one was not significant. Statistical significance was found as sample size increased. Values of β also increased as collections were added to analysis; however, the increase was marginal once outliers (14-year-old ram from collection 1 and 15-year-old ram from collection 2) were removed (Fig. 2 and Table 2). In this analysis we could not directly control for the variance introduced by cohorts and populations. Therefore, we used a Monte Carlo model simulation designed to test the null hypothesis that our result from natural mortalities could be explained by variance from cohort and population. For the full data set (collections 1–3) the test was significant (P = 0.001) indicating that variance from cohorts and populations alone could not account for our result. With the outliers removed from the full data set the Monte Carlo model was also significant (P < 0.05).

Table 2.   Regression analysis showing negative relationship between horn growth rate and longevity for natural and hunted mortality in thinhorn sheep Ovis dalli. Horn growth rate is measured as the amount of horn growth in the first five summers of life. For a description of horn collections refer to Table 1.
Age at deathβSEβ coefficientFd.f.PR2
Natural mortality
Collection 1
 Horn growth rate−0.00090.0009−0.131.01, 630.3170.02
Collections 1 and 2
 Horn growth rate−0.00180.0007−0.286.81, 810.0110.08
 Horn growth rate, no outlier−0.00130.0007−0.203.41, 800.0690.04
Collections 1–3
 Horn growth rate−0.00230.0007−0.3311.41, 930.0010.11
 Horn growth rate, no outliers−0.00160.0007−0.234.91, 910.0290.05
Hunted mortality
Collection 5
 Horn growth rate−0.002130.0001−0.27257.81, 3295< 0.0010.07
Figure 2.

 Negative relationship between age at death and horn growth rate for natural (n = 95) and hunted mortality (n = 3297) in thinhorn sheep, Yukon Territory, Canada. Natural mortality is sampled from populations with very little or no hunting. Growth rate is measured as growth occurring in the first five summers of life. For hunted mortality growth rate measurements have been controlled for the effect of horn wear, population and environment, and for natural mortality the effect of horn wear has been controlled.

We also directly tested whether origin of collection (see Table 1) affected our result for natural mortality by constructing a linear model where we entered ‘growth rate’ as a covariate and ‘collection’ as a fixed factor (Table 3). Both ‘growth rate’ and ‘collection × growth rate’ were significant. Again we checked whether the presence of two possible outliers (14- and 15-year-old rams) affected this relationship. With the outliers removed ‘growth rate’ remained significant, but ‘collection × growth rate’ was no longer significant.

Table 3.   Linear model constructed to test the effects of collection origin on age at death for natural mortalities. Collections 1–3 (see Table 1) were used in analysis.
Age at deathSSd.f.MSFP
Natural mortality (model R2 = 0.21)
 Model101.7522.16.1< 0.001
 Intercept3938.013938.01211.1< 0.001
 Horn growth rate (HGR)80.0180.024.6< 0.001
 Collection × HGR44.6222.36.90.002
Natural mortality, outliers removed (model R2 = 0.10)
 Intercept2315.012315.0744.4< 0.001
 Horn growth rate (HGR)11.8111.83.80.054
 Collection × HGR5.

We tested whether our finding that growth rate is negatively associated with longevity is robust to the unlikely possibility that our assumption is false that there is no age-related size bias in the horns found (see Materials and methods). If we further limit the minimum age of horns to sheep over 10 years old we find that horn size and age are not correlated (n = 16, r = −0.251, P =0.348). Therefore, in this subset of the data there is no longer a concern that younger horns with slower growth are not being found. We found a clearly negative relationship between growth rate and age at death in this subset of the data (linear regression, dependent variable ‘age at death’, anova: F1,14 = 17.6, P = 0.001, independent variable ‘growth rate’β = −0.003, SE =0.001, β coefficient = −0.75, P = 0.001).

Growth rate also predicted age at death in the hunted sheep sample. Growth rate may be affected by population, cohort fluctuations in growth rate and horn wear. We controlled for these variables by first removing their effects on growth rate in a linear model with population entered as a random factor, growth rate index as a covariate and horn wear as a fixed factor (population: F61,3233 = 6.2, P < 0.001, growth rate index: F1,3233 =458.2, P < 0.001, horn wear: F1,3233 = 178.7, P < 0.001). Then we continued with regression analysis using the corrected values. We found a negative relationship between longevity and growth rate for hunted sheep (Table 2, Fig. 2) which was similar to that found for natural mortality sheep.

Finally, we tested whether age and growth rate could predict horn size at death. For both natural and hunted mortality horn size at death was predicted by both age and (residuals of) growth rate (Table 4, Fig. 3a,b).

Table 4.   Results for linear models for predictors of horn volume at death in thinhorn sheep. Data are from natural mortalities in populations with little or no hunting and hunted mortalities. Horn growth rate is measured as the amount of horn growth in the first five summers of life. Collection origin was defined according to Table 1, and collections one, two and three were used for natural mortality analysis.
Horn size at deathSSd.f.MSFPIncluded in final model?
Hunted mortality (model R2 = 0.57)
 Model2.76 × 108122.3 × 107360.0< 0.001 
 Intercept9.62 × 10819.62 × 10815052.5< 0.001Yes
 Horn growth rate1.84 × 10811.84 × 1082876.2< 0.001Yes
 Age1.66 × 108111.51 × 107234.5< 0.001Yes
 Error2.10 × 10832836.39 × 104   
 Total1.66 × 10103296    
Natural mortality (model R2 = 0.70)
 Model1.4 × 107110.1 × 10725.6< 0.001 
 Intercept14.7 × 107114.7 × 1073079.4< 0.001Yes
 Horn growth rate (HGR)1.0 × 10711.0 × 107201.5< 0.001Yes
 Age0.8 × 107100.1 × 10713.3< 0.001Yes
 Origin0.02 × 10720.01 × 1071.70.182No
 HGR × Origin0.02 × 10720.01 × 1071.10.346No
 Error0.6 × 107837.0 × 104   
 Total47.0 × 10795    
Figure 3.

 (a) Positive relationship between horn growth rate and horn volume at death (i), and age and horn volume at death (ii) for 3297 hunted thinhorn sheep rams in Yukon Territory, Canada. See Table 4 for analysis. (b) Positive relationship between horn growth rate and horn volume at death (i), and age and horn volume at death (ii) for 95 thinhorn sheep rams dying of natural mortality in populations with no or very little hunting activity in Yukon Territory, Canada. See Table 4 for analysis.


We found a negative relationship between horn growth rate and longevity for natural mortality in areas with little or no hunting (Fig. 2). This does not support the hypothesis that rapid growth provides a survival advantage (Coltman et al., 2003, 2005), but does support the alternative hypothesis of a trade-off between growth rate and longevity (Geist, 1971). For natural mortalities horn size at death was positively associated with age and growth rate (Fig. 3b), although growth rate explained more of the total variance (Table 4). The relationship between age and horn size at death appeared to be nonlinear with peak average horn size occurring at about 11 years of age (Fig. 3b). Although this suggests that the largest horn sizes may not be achieved by individuals with very slow growth rates, in general it can be stated that relatively large horn size can be attained by rapid growth (although this may come at an expense of reduced longevity), or by greater longevity, because horns grow each year and are not shed. An excellent example of the relationship between growth rate and ultimate horn size in our natural mortality data is that the ram with the lowest growth rate while young, died with larger than average horns because of longevity. The relationship between growth rate and longevity may have direct implications for life history strategies because large horn size has been demonstrated to affect mating opportunities and dominance status in thinhorn sheep (Geist, 1971; Loehr, 2006).

In the analysis of hunted mortalities it was also evident that sheep with faster horn growth died at a younger age (Fig. 2). We note that this analysis is not intended to measure a natural cost of rapid growth because hunting practices and natural mortality could both explain this result. However, it is interesting that growth rate explained similar amounts of variance (5–11%) in both the natural and hunting mortality models. The regression β coefficients for growth rate for natural and hunted mortality were −0.33 and −0.27 respectively. (We note that the β coefficient for natural mortality was reduced to −0.23 when two potential outliers were removed.) Our results are similar to those of Coltman et al. (2005) who found a correlation of −0.20 between horn growth rate and longevity for a hunted population of bighorn sheep. Instead of finding sharply contrasting β coefficients between hunted and natural mortalities, surprisingly, we have found that trends in growth rate and mortality are very similar.

In this study, density and age structure was not known. It is possible that the cost of rapid growth varies with age structure or density of the population. This may be especially true if the cost is socially imposed. However, a strength of the data set is that it has been gathered from 11 populations over a 36-year period. It seems reasonable that our data are indicative of the general relationship between growth rate and longevity over the time period studied. Thus we can conclude that in general, rapid growth is costly, however, fluctuations in the strength of the cost may have occurred during this time.

We have made an important assumption that there is no size bias for finding horns of sheep aged 5 years and over. If our assumption is false then our sample will be biased (i) by old horns with rapid growth and (ii) in each age group by the sheep with the most rapid growth. If the survival advantage hypothesis is true and the assumption false then this should still result in the acceptance of the survival advantage hypothesis. Thus our result showing a negative relationship between growth rate and longevity safely refutes the survival advantage hypothesis. It is also plausible that there is no correlation between growth rate and longevity. Given this scenario and our assumption is false then it is possible that a negative correlation between growth rate and longevity will be found when none is present (type I error). This would occur if horns that were young and had a low growth rate were not found. If this is the case we should still find some old horns with rapid growth in the data, however, these horns were distinctly missing (see Fig. 2). Furthermore, the chance of type I error appears low because using the horn collection method: (i) despite smaller sample sizes we have found variance in growth rates in young sheep that is similar to that in older age classes that have larger sample sizes (Fig. 2), (ii) it was possible to find sheep with low growth rates in young age classes (see Materials and methods), (iii) for sheep aged 5–14 years all horns are relatively large (see Materials and methods).

Ram survival and horn growth rate

The relationship between early growth rate and life expectancy and the underlying mechanisms are poorly understood (Metcalfe & Monaghan, 2001; Olsson & Shine, 2002). Physiological costs of early (usually juvenile) growth have been found in greater tumour production, increased incidence of coronary lesions, and higher susceptibility to external parasites (Eklund & Bradford, 1977; Saunders et al., 1992). Rapid growth may also result in greater predation risk (Munch & Conover, 2003). At the population level horn growth is positively associated with the resources (forage) available (Festa-Bianchet et al., 2004), and at the individual level there is a trade-off between predation risk and foraging because more abundant forage is located further away from escape terrain (Bleich et al., 1996). Individuals that are less vigilant have greater foraging efficiency, however, such behaviour may come at a cost of increased predation risk (Lima & Dill, 1990; Frid, 1997).

In mountain sheep another possible explanation may be found in the cost of reproduction, which is not mutually exclusive to those presented above. Research on polygynous male mammals has shown that the cost of reproduction for red deer (Cervus elaphus) males appears to increase with higher density (Mysterud et al., 2001) and in Himalayan tahr (Hemitragus lemjahicus) the depletion of energetic reserves during the rut is most pronounced for adult males (Forsyth et al., 2005). In bighorn sheep, rams that are most active during the rut incur a cost of higher parasite load (Pelletier et al., 2005). Reproductively successful individuals may pay an immediate survival cost while young as is the case in the northern elephant seal Mirounga angustirostris (Clinton & Le Boeuf, 1993). In Soay sheep O. aries early reproduction by males also bears a survival cost (Stevenson & Bancroft, 1995). However, it is also possible that successful individuals are more likely to survive. McElligott et al. (2002) found that male fallow deer (Dama dama) that reproduced in 1 year had a greater chance of surviving to the next year than males that did not.

In thinhorn sheep rapid horn growth may allow young individuals to compete successfully for mates (possibly resulting in a survival cost) or they may be prone to greater reproductive effort even if success rate is low (and incur greater costs by participating fully in the rut). Research suggests that the success rate of young individuals with rapid growth is low. Coltman et al. (2002) found that in bighorn sheep the mating advantage of rapid horn growth began to increase from about 6 years of age, however for these young rams success was still marginal compared with older individuals.

Geist (1971) suggested that if rapid growth were associated with a natural mortality cost a mechanism would be present to limit or slow horn size evolution. Our findings are consistent with Geist's idea. We have found that on average sheep with slower horn growth will live to an older age (Fig. 2); however, horn growth rate may still be a better predictor of ultimate horn size than age (Table 4). Interestingly, Coltman et al. (2002) found that both age and horn size are strong yet independent predictors of annual reproductive success in bighorn rams. The idea that a trade-off between growth rate and longevity can function to preserve genetic diversity is strongly supported by recent research on Soay sheep (Robinson et al., 2006). In this system a trade-off between male horn growth rate and longevity also exists and despite an annual mating advantage of large horns no effect of growth rate on lifetime reproductive success was found.

We point out that a limitation of our data is that we did not have information on body size which could possibly affect male longevity and mating success. However, horn growth and body mass are only weakly correlated (Festa-Bianchet et al., 2004). Coltman et al. (2005) found no correlation between weight and longevity in bighorn sheep, and no correlation was found between weight and lifetime reproductive success. Coltman et al. (2002) did find that annual reproductive success was correlated with weight in two of 6 years of study; however, age and horn size were overall much better predictors of annual reproductive success.

Implications for artificial selection

A reduction in horn growth over time might be expected in a system where growth rate and longevity are negatively correlated (e.g. Coltman et al., 2003). Yet, despite a negative association between growth rate and longevity and a harvest rate of approximately 27% of legal rams, mean annual growth of hunted horns has gradually increased in the Yukon from 1963 to 2004 (Yukon Government, unpublished). Horn growth is influenced by the Pacific Decadal Oscillation (PDO) climate cycle (also see Hik & Carey, 2000) which has shown a warming trend over this same period (Mantua et al., 1997). When PDO-influenced environmental factors (which explain over 40% of mean annual variation in horn growth) are controlled there is no change in mean horn growth during 1963–2004 (J. Loehr, J. Carey & D. S. Hik, unpublished). Importantly, the harvest system in Yukon Territory results in similar mortality trends to those found in populations without hunting (Fig. 2). It is likely that this trade-off between growth rate and longevity is a key aspect of the system because it can allow the maintenance of genetic variation in populations (e.g. Geist, 1971; Robinson et al., 2006). Therefore, a harvest system that is morphologically biased does not appear to result in selection against rapid horn growth, precisely because it mimics the natural mortality trade-off.

When interpreting our results we caution that regional or population-specific differences between bighorn and thinhorn sheep (e.g. Singer & Zeigenfuss, 2002) could still lead to differing selection regimes. Thus it may be possible that the assumption that rapid growth results in increased longevity (Coltman et al., 2003, 2005) in unhunted populations is valid for bighorn sheep, although the earlier work of Geist (1966a, 1971) may suggest otherwise.


Our study is one of the first to demonstrate a cost associated (directly or indirectly) with rapid growth in a natural setting (also see Olsson & Shine, 2002; Robinson et al., 2006). Such a trade-off between longevity and growth rate could allow for conservation of genetic variation in horn growth rate in mountain sheep populations (Geist, 1971; Robinson et al., 2006). It is also clear that in cases where artificial selection is suspected great care must be taken to consider how the system functions in its natural state. Without knowledge of how a system functions without harvesting it is possible that the mechanism for artificial selection can be misidentified. This highlights the great need for the existence and study of protected wildlife populations as a tool to understand the influences of humans.


We are very grateful for the funding provided by the Ella and Georg Ehrnrooth Foundation, University of Jyväskylä, Yukon Department of Environment, Town of Faro, and Ellen and Artturi Nyyssönen Foundation to J.L., and the Academy of Finland under the Finnish Centre of Excellence Programme during 2000–2005 (Project 44878) to J.S. and H.Y. The authors wish to thank J. Merilä, D.W. Coltman, T. Coulson, A. Mysterud as well as L. Kruuk and one anonymous referee for their helpful comments that greatly improved the manuscript. H. Kokko provided the initial inspiration to investigate the idea of a cost of rapid growth and gave valuable input as the manuscript developed. We especially thank Philip Merchant as well as conservation officers and wildlife technicians who have carried out the thousands of horn measurements necessary for this research.