See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/271693018 Identifying Targets of Selection: A Multivariate Analysis of Reproductive Traits in the Great Tit ARTICLE in OIKOS · APRIL 1997 Impact Factor: 3.44 · DOI: 10.2307/3545622 CITATIONS 22 READS 12 5 AUTHORS, INCLUDING: Peeter Hõrak University of Tartu 70 PUBLICATIONS 2,731 CITATIONS SEE PROFILE Raivo Mänd University of Tartu 68 PUBLICATIONS 1,158 CITATIONS SEE PROFILE All in-text references underlined in blue are linked to publications on ResearchGate, letting you access and read them immediately. Available from: Peeter Hõrak Retrieved on: 04 February 2016
Identifying targets of selection: a multivariate analysis of reproductive traits in the great tit
Peeter Lirak, Raivo Mand and Indrek Ots
Hbrak, P., Mand, R. and Ots, I. 1997. Identifying targets of selection: a multivariate analysis of reproductive traits in the great tit. - Oikos 78: 592-600.
In order to identify the true targets of phenotypic selection, we carried out a multivariate analysis on three reproductive traits in an urban population of great tits in Tartu, south-east Estonia, using data collected during 1987-1994. Individuals laying large eggs consistently recruited more offspring into the breeding population, independent of the simultaneous effects of clutch size and laying date. However, the mean egg size of a clutch was not related to the occurrence of embryonal or nestling mortality, which points to the possibility that the relationship between egg size and recruitment rate was due to the effects of female quality. Early breeders tended to recruit more offspring into the breeding population, but the effect of laying date disappeared when egg size was incorporated into the model. This fact indicates that the primary target of selection was not laying date per se but some property of the female affecting both laying date and egg size. Selection on clutch size fluctuated in direction between years. Selection against large clutches operated above all on individuals with small eggs, suggesting that the fitness consequences of similar reproductive decisions are different for individuals of different quality.
P. Hbrak, R. Mdnd and L Ots, Animal Ecology Section, Inst. of Zoology and Botany, Univ. of Tartu, Riia 181, Tartu EE-2400 Estonia ([email protected]).
Measuring phenotypic selection means comparing the fitness of individuals differing from each other in respect of some trait values. Because phenotypes are affected both by genes and by their past and current environ- ment, it is often not clear why individuals with some particular trait value have higher fitness than others.
The crucial importance of distinguishing whether a given trait is the primary target of selection or is correlated with such a target trait is illustrated by the models of selection on avian breeding traits by Price et al. (1988) and Price and Liou (1989). Introducing the concept of selection on some state-dependent quality factor of individuals, which simultaneously (but through separate pathways) affects both the expression of some reproductive trait and fitness, these models describe how a correlation between heritable traits and fitness can persist at an evolutionary equilibrium (Fig.
1). The concept of different pathways of selection thus suggests that females laying early and/or large clutches may be the fittest because selection operates on some state-dependent quality variable (often termed as nutri- tional state, health, or individual condition), rather than on clutch size or laying date per se. Analogous mechanisms have been proposed to explain the lack of response to selection on avian body size (van Noord- wijk et al. 1988, Alatalo et al. 1990) and egg size (Bolton 1991). State-dependent life-history models (e.g. McNamara and Houston 1992, Morris 1996) are based on similar reasoning.
Given the large number of factors (such as health status, age, experience, amount of physiological wear and tear, foraging efficiency, capability to monopolize resources and subdue competitors, parasite load, mate and territory quality) integrated in a hypothetical state-
Accepted 20 August 1996
Copyright ? OIKOS 1997 ISSN 0030-1299 Printed in Ireland - all rights reserved
592 OIKOS 78:3 (1997)
This content downloaded from 193.40.5.245 on Thu, 05 Mar 2015 11:40:20 UTCAll use subject to JSTOR Terms and Conditions
dependent quality variable of an individual, models of state-dependent selection appear difficult to apply to phenotypic selection in the wild. Recent attempts, how- ever, have demonstrated that this is possible if the question one wishes to ask in a particular study is specified unambiguously and the natural history of the study object is known in detail (see e.g. Morris 1996).
Our purpose with the present study was to apply the concept of different selection pathways on the process of phenotypic selection in a great tit (Parus major) population. We therefore needed to estimate the rela- tive importance of different pathways of selection. In an ideal case, such a problem may be solved by substi- tuting the path symbols in Fig. 1 with partial regression coefficients and estimating the relative importance of the different causal links. In practice, this task is unreal- istic since it demands exact measurements of the state- dependent quality variable, assumed to affect both the given reproductive trait and the fitness. Thus, given the practical difficulties of quantifying the role of the state variables, the problem will have to be solved by com- paring alternative qualitative models. We propose the following approach for pinpointing the true targets of selection affecting avian reproductive traits.
Assuming the causal links between the main compo- nents of the path scheme in Fig. 1 to be of different importance, we decompose the original model of Price et al. (1988) into three alternative scenarios of selection (Fig. 2). These scenarios correspond to different combi- nations of reproductive trait, individual state, and fitness, and they can be distinguished on the basis of their specific predictions in respect to univariate and multivariate analyses of selection. In a natural situation we expect these scenarios to represent a continuum resulting in a smooth transition from one to another, as one link between the components of the model becomes
Residual effect \
Genes -* Reproductive trait- +, Fitness
Condition Fig. 1. Relationship between breeding trait, individual state (condition) and fitness (after Price et al. 1988). Arrows connect dependent variables (arrowheads) with independent variables. A breeding trait is determined by an additive genetic compo- nent, state-dependent quality factor (condition), and a residual effect. 13b denotes direct selection on the trait, while As is a direct effect of the condition on fitness. Thus, fitness is affected by the direct effect of reproductive traits as indirectly influ- enced by genes and condition (ti), but also by direct effects of the condition (Ic). Breeding traits can therefore be potentially subjected to constant directional selection without any evolu- tionary response because selection acts on the environmentally determined condition of individuals rather than on reproduc- tive traits. For related path models, see also Larsson (1992), Schluter and Gustafsson (1993), Moller (1994).
T W 1. Model of apparent selection
C
T W 2. Model of plain selection ? C
T
W 3. Model of interactive selection
C
Fig. 2. Path models of relationships between the breeding trait (T), individual condition (C) and fitness (W) presented as three scenarios of selection. Pathways of the genetic and residual effect are as in Fig. 1, and since they are similar in all cases they are not shown here. Arrows connect dependent variables (arrowheads) with independent variables. The lack of connec- tion between variables denotes situations in which the causal link between some components of the path scheme presumably is of so small practical importance (as compared to more important pathways) that it can be disregarded. The condition term refers to that component of state-dependent quality factor which persists throughout the breeding cycle.
more (or less) important as compared to the others. Therefore, while we do not assume that in each particu- lar case an exact match between the data and one of the three models can be found, we do expect that it will usually be possible to distinguish which of the three scenarios matches the pattern in the data most closely.
In our models we call the state-dependent quality variable of an individual its condition. We define it as a property of an individual which, depending of the relative importance of causal links between the compo- nents of the model, has a potential to affect either reproductive traits, fitness, or both. If reproductive traits are condition-dependent, we assume an individual to lay early and/or many eggs and/or large eggs when in good condition, and to lay late and/or few eggs and/or small eggs when in poor condition. We also assume that part of the variation in reproductive trait values is due to genetic differences between individuals. If condition affects fitness, we assume that, all other things being equal, individuals in good condition are likely to recruit more offspring than those in a poor condition. In our terminology, condition is considered to persist throughout the entire breeding period (in the sense that its variation is parallel for all individuals). This means that, if condition affects both reproductive traits and fitness, we expect that individuals which are in a better condition than others during clutch forma-
OIKOS 78:3 (1997) 593
This content downloaded from 193.40.5.245 on Thu, 05 Mar 2015 11:40:20 UTCAll use subject to JSTOR Terms and Conditions
tion are also more efficient than others at brood rear- ing. Thus, individual condition, as defined here, denotes a persistent component of the state-dependent quality factor, rather than the momentary physiological state of an individual.
Scenarios of selection 1. According to the model of apparent selection (Fig. 2), both fitness and breeding traits (like clutch size and laying date) depend on the condition of the reproduc- tive individual, which results in a correlation between breeding trait and fitness as a by-product. The selection on breeding traits is, however, only apparent because each trait has no independent effect on reproductive success.
The model predicts that the largest clutches and/or earliest breeders are the fittest because correlations between the individual's condition and these particular reproductive traits and between condition and fitness lead to directional selection favouring individuals laying early and/or large clutches (which can be detected in a univariate analysis). However, because the selection on breeding traits is only apparent, clutch size or laying date become insignificant if some other trait which is more strongly correlated to individual condition is in- corporated into the model in a multivariate analysis of selection. This is because the contribution of different reproductive traits to fitness has an overlapping compo- nent which is caused by their covariance with the individual's condition. As a result, the most condition- dependent trait which is included in the multiple analy- sis will eliminate the condition-dependent fitness effects of other traits.
2. According to the model of plain selection, reproduc- tive traits of an individual are affected by its condition, but condition has no independent effect on fitness. Therefore, any kind of selection (directional, stabilizing or disruptive) on reproductive traits may occur. Be- cause condition has no direct effect on fitness, the model predicts that fitness-effects of traits like clutch size and laying date will remain significant in the multi- ple analysis involving other, more condition-dependent traits as predictor variables.
3. According to the model of interactive selection, the values of breeding traits are not determined by the individual's condition. At the same time, both the individual's condition and its breeding traits affect fitness, which generates interactive effects between con- dition and breeding trait. This means that the fitness value of clutch size (laying date) depends on the condi- tion of the reproductive individual: some individuals do well because they make correct reproductive decisions,
while others do well because they are in superior condi- tion. Birds which are good in both aspects do best, while those in poor condition and making wrong deci- sions do worst.
If some trait closely correlated to the condition of the reproductive individual is incorporated in the analysis, this scenario predicts a significant 'clutch size x condi- tion-related-trait' (or 'laying date x condition-related- trait') interaction term, because the fitness value of clutch size (laying date) depends on the individual's condition.
In this paper we attempt to distinguish between the targets of phenotypic selection by performing a multiple analysis of fecundity selection in the great tit. We examine the relationship between the local recruitment rate (as an estimate of fitness) and three reproductive traits - laying date, clutch size and egg size, seeking for the best match between the pattern in the data and the predictions of the three alternative scenarios of selec- tion described above. We proceed from the assumption that the female's average egg size in the great tit population under investigation reflects some persistent component of her state. This assumption is based on our previous knowledge about the correlation (r = 0.43) between egg size and female residual weight (in relation to body size) about one month after the eggs were laid (H6rak et al. 1995). It is also supported by similar findings in numerous other studies (see e.g. Ojanen et al. 1979, Jlrvinen and Vaisanen 1983, Murphy 1986, Mind 1988, Jirvinen and Pryl 1989, Leblanc 1989, Nilsson and Svensson 1993, Smith et al. 1993, Nager and Zandt 1994 and references in these), and by accu- mulating evidence for a close relationship between egg size and various aspects of parental quality (Slagsvold and Lifjeld 1989, Langston et al. 1990, Reid and Boersma 1990, Wiggins 1990, Bolton 1991, Sydeman and Emslie 1992, Williams et al. 1993, Brouwer and Spaans 1994, Sandercock and Pedersen 1994, Amund- sen 1995). Therefore, we have reason to expect that inclusion of egg size as a condition-related trait in the analyses of selection on laying date and clutch size will help clarifying the role of individual condition in the selection on these traits. To our knowledge, the present study is the first to use as rigorous a fitness measure as recruitment rate in assessment of the selection on egg size in a passerine bird species.
Methods The study was conducted during the period 1987-1994 in an urban great tit population breeding in Tartu, south-east Estonia (58022'N, 26043'E). The study area consisted of two large and two small parks (about 22 ha) and of avenues with a total length of 9 km. The main tree species were Tilia cordata, Acer platanoides,
594 OIKOS 78:3 (1997)
This content downloaded from 193.40.5.245 on Thu, 05 Mar 2015 11:40:20 UTCAll use subject to JSTOR Terms and Conditions
Betula pendula, Quercus robur, and Populus suaveolens. The number of great tit pairs breeding in the study area varied from 43 to 93 (first clutches only) in different years. The birds bred in nestboxes, mounted at a height of about 2.5 m. The dimensions of the box cavity were approximately 11 x 11 x 30 cm and the diameter of the entrance hole was 3.5 cm. Old nest material was re- moved every year.
Nestboxes were checked regularly to determine clutch size and laying date (=date of laying the first egg), assuming that one egg is laid every day. Unhatched eggs were classified as infertile if no embryo develop- ment had occurred. Adults were captured and nestlings ringed in the second half of the nestling period. For measuring egg size, whole clutches were photographed after the sixth day of incubation using a stand described in Mind et al. (1986). A graphics digitizer was used for the input of egg contours from photographs and a special program OMELETTE (Maind et al. 1986) for smoothing data and for estimating egg dimensions and volume. The volume of an egg was calculated from the contour using trapezoidal integration instead of deriv- ing it from linear measurements. Thus, individual dif- ferences in egg shape did not influence the accuracy of volume estimation. The clutch means of egg volume were used in all analyses. To minimize the influence of a few aberrant eggs on clutch means, one egg per clutch, namely the one most dissimilar to the others in size, shape, or appearance, was excluded before the clutch mean was calculated. The rejection was based on immediate impression of the researcher, not on calcu- lated egg measurements. Only data on first clutches were used and depredated nests were excluded from the analysis.
Local recruitment rate was used as a measure of individual fitness. Egg measurements were recorded during 1987-91, but recruitment was recorded up to 1994. Hence the length of the study period did not affect the chances of recapturing nestlings ringed in different years. In all analyses of selection, clutch size, laying date and clutch mean egg size were standardized within years (mean = 0, standard deviation = 1). Aver- age trait values for broods recruiting young to the breeding population (weighted by the number of re- cruits per brood) was thus equal to the selection differ- ential (s) in standard deviation units (Falconer 1983). The significance of selection differentials was tested by t-tests, comparing the means of individuals with and without recruitment (see e.g. Linden et al. 1992). Be- cause the distribution of the number of recruits ap- proached a Poisson distribution (see Appendix), we used regression models with Poisson error distribution and logarithmic link function when analysing net selec- tion over five years (SAS INSIGHT, SAS Institute 1993, 1994). Year term was included as a factor in all analyses to eliminate the effect of yearly differences in recruitment rate. Occurrence of stabilizing and disrup-
tive selection was tested by comparing variances in trait values between the individuals with and without re- cruitment, as well as in regression models including squared trait values. When testing the significance of squared or interaction terms, the main effects were kept in models irrespective of their significance. All signifi- cance levels are for two-tailed tests. Subscripts used in connection with t-tests refer to sample sizes.
Results
Selection on egg size
Females laying large eggs consistently recruited more offspring to the breeding population. The standardized net selection differential (s) for egg size was positive and significantly different from zero (s = 0.38 SD, t53,179 =-2.35, p = 0.020). Within individual years, we found a statistically significant positive selection diff- erential for egg size in 1989 (s = 0.62, t16,23 = 2.94, p = 0.006). The positive effect of egg size on recruit- ment rate was also detected in the regression analysis adjusting for year effects (Table 1). The effect of egg size was consistent in direction during all the years since the 'egg size x year' interaction term did not improve the model significantly. We found no sign of stabilizing nor disruptive selection on egg size when variances in egg size of females with and without recruitment were compared (both in individual years and pooled data). Neither did adding the egg size squared term improve the model in Table 1.
Because our previous study of the same great tit population (HMrak et al. 1995) has revealed a remark- able heritable component in egg size variation, we tested for the occurrence of selective response in egg size. For this purpose, we compared the average egg size of the population in 1989 (when highly significant positive selection differential for egg size was detected) with that of the females born in 1989 and breeding in 1990. In order to eliminate the effect of annual differ-
Table 1. Effect of egg size (ES) on local recruitment rate in regression analysis with Poisson error terms and logarithmic link function. Null model includes the constant only. Signifi- cance of main effects is tested by likelihood ratio tests, adding variables one after another to the null model. Significance of squared and interaction terms is tested by adding those terms one at time to the model with main effects of year and egg size.
Parameter Deviance DF AD ADF p
Null model 166.64 215 ? Year 148.71 211 17.94 4 0.001 ? ES 143.12 210 5.59 1 0.018
+ ES2 141.98 209 1.14 1 0.286
? ES x Year 140.37 206 2.74 4 0.602
OIKOS 78:3 (1997) 595
This content downloaded from 193.40.5.245 on Thu, 05 Mar 2015 11:40:20 UTCAll use subject to JSTOR Terms and Conditions
ences in environmental conditions we corrected the difference between the annual means with the mean difference in egg size of all individuals that bred in both years (see e.g. van Noordwijk et al. 1981). The response to selection (1.68 + 0.14 cm3 vs 1.64 + 0.11 cm3) did not differ significantly from zero (t26,68 =-1.36, p = 0.177).
To distinguish between the components of breeding success related to egg size variation, we examined the relationships between egg size, hatching success and prefledging nestling mortality. We found no effects of egg size on any one of these traits (Table 2).
Selection on laying date
Compared to late breeders, early breeding females re- cruited significantly more offspring into the breeding population in 1988 (s = -0.77, t17,51 = 2.89, p = 0.005). The net selection differential was not significant (s = -0.26, t52,173 =-1.50, p = 0.134). The date effect, however, was nearly significant in the regression analy- sis including the year term (Table 3). We found no evidence for a fluctuating selection pressure on laying date since the 'laying date x year' interaction term in Table 3 was not significant. As in the case of egg size, we could not detect stabilizing or disruptive selection on laying date neither by comparing the variances, nor in the regression analysis (Table 3).
Selection on clutch size
Selection favoured large clutches in 1987 (s = 0.69, t9 3= 3.54, p = 0.002), while small clutches did best in 1989 (s = -0.45, t16,23 =-2.43, p = 0.020). The net selection differential did not differ significantly from zero (s=-0.04, t52,178 =-0.81, p = 0.420), and clutch
size did not affect recruitment rate significantly in the regression analysis accounting for the year effect (Table 4). The 'clutch size x year' term was close to signifi- cance in the model.
As was the case for egg size and laying date, we found no sign of stabilizing or disruptive selection on clutch size (no difference in variances and insignificant effect of adding clutch size squared to the model in Table 4).
Selection on trait combinations
Although both laying date and clutch size were in- significant in the univariate analysis, the laying date term became significant in the model after inclusion of clutch size (comparison of models Year,CS,LD vs Year,CS: AD = 4.28, ADF = 1, p = 0.039). Fig. 3 re- veals that this was because selection favoured small early clutches. Large early clutches were not more successful than small late clutches.
We detected no effect of clutch size on recruitment rate when laying date and year were incorporated in the model (comparison of models Year,CS,LD vs Year,LD: AD = 1.95, ADF = 1, p = 0.163). Inclusion of the 'clutch size x laying date' interaction term did not sig- nificantly affect the model with main effects of year, date and clutch size (AD = 0.11, ADF = 1, p = 0.743). Adding egg size significantly improved the model (com- parison of models Year,CS,LD,ES vs Year,CSLD: AD = 4.13, ADF = 1, p = 0.042), indicating that the effect of egg size on recruitment rate was not con- founded by simultaneous effects of laying date and clutch size.
Clutch size was not a significant component of the model, so we dropped this term and refitted the model with year, laying date and egg size only. In this model,
Table 3. Effect of laying date (LD) on local recruitment rate in Table 4. Effect of clutch size (CS) on local recruitment rate in regression analysis with Poisson error terms and logarithmic regression analysis with Poisson error terms and logarithmic link function. Significance is tested analogously to Table 1. link function. Significance is tested analogously to Table 1.
Parameter Deviance DF AD ADF p Parameter Deviance DF AD ADF p
Null model 166.64 215 Null model 166.64 215 + Year 148.71 211 17.94 4 0.001 + Year 148.71 211 17.94 4 0.001 + LD 145.82 210 2.88 1 0.090 + CS 148.15 210 0.55 1 0.458
Fig. 3. Selection of laying date and clutch size in combination. 'Small' clutches and 'early' dates are smaller (earlier) than the yearly average while 'large' ('late') ones are greater (later) than average. Number of clutches is weighted by number of re- cruits.
the effect of laying date on recruitment rate disappears (comparison of models Year,LDES vs Year,ES: AD = 1.87, ADF = 1, p = 0.171), indicating that the effect of laying date on recruitment rate was due to the same factors as affected the relationship between egg size and recruitment rate. Adding the 'laying date x egg size' interaction term to the model with both the main effects and year term did not significantly affect the model (comparison of models Year,LD,ES,LD x ES vs Year,LD,ES: AD = 0.06, ADF = 1, p = 0.799). Fig. 4
shows that the effect of egg size on recruitment rate was more notable than the influence of laying date.
Finally, we examined whether the effect of egg size on recruitment rate was similar for individuals with different clutch sizes. The marginally significant 'egg size x clutch size' interaction term in Table 5 indicates
36
o, 30 ~~~ 12
16
Fig. 4. Selection on laying date and egg size in combination. Legend as in Fig. 3.
0, 30 C 9.
Fig. 5. Selection on egg size and clutch size in combination. Legend as in Fig. 3.
that this was not necessarily so. Fig. 5 reveals that individuals with large clutches and small eggs had disproportionately low breeding success.
Discussion
Selection on egg size
Of the three breeding traits investigated, we found a clear and consistently positive selection differential only for egg size. Great tits laying large eggs recruited more offspring to the breeding population, and the effect of egg size on recruitment was independent of simulta- neous effects of clutch size and laying date. The rela- tionship between egg size and local recruitment rate might have occurred either because the egg size effect per se affected recruitment, or because some property of a female affected both egg size and recruitment rate. The second possibility seems more likely for the follow- ing reasons.
First, we found no evidence that mean egg size affected either hatching success or nestling mortality (Table 2). Therefore, if in fact egg size per se affected offspring quality, then this effect must have been hid- den during the embryonal and nestling period, and influenced only the postfledging survival of young. The
Table 5. Joint effect of egg size (ES) and clutch size (CS) on local recruitment rate in regression analysis with Poisson error terms and logarithmic link function. Significance of main effects is tested by adding variables one after another to the null model.
Parameter Deviance DF AD ADF p
Null model 166.64 215 + Year 148.71 211 17.94 4 0.001 + ES 143.12 210 5.59 1 0.018 + CS 142.65 209 0.46 1 0.498 + CS x ES 139.87 208 2.79 1 0.095
OIKOS 78:3 (1997) 597
This content downloaded from 193.40.5.245 on Thu, 05 Mar 2015 11:40:20 UTCAll use subject to JSTOR Terms and Conditions
mechanism behind such a phenomenon is difficult to imagine since the majority of studies have found that the effect of egg size on nestling growth and survival decreases rapidly with nestling age (see e.g. Ricklefs 1984, Greig-Smith et al. 1988, Williams 1994, Smith et al. 1995 and references in these).
Second, our previous study in the same population (Hdrak et al. 1995) revealed a relationship between egg size and some component of female physiological con- dition. We found that egg size correlated positively with the residual weight of females (adjusted for body size) about one month after the onset of laying. Thus, fe- males with large eggs either were initially heavier, or/ and lost less weight during breeding. In either case, they may be classified as being in superior condition.
Third, the episode of strong selection in favour of large egg size in 1989 was not followed by a selective response, indicating that selection affected the condi- tion-determined part of egg-size variation. This result is notable given the high repeatability (=0.6) and herita- bility (h2 = 0.8) of egg size in this great tit population (H6rak et al. 1995).
We therefore conclude that it was some aspect of parental quality, rather than egg size per se, that caused females with large eggs to produce fledglings with better prospects of survival and/or subsequent territory estab- lishment. This conclusion is consistent with abundant previous evidence (see e.g. Ojanen et al. 1979, Jirvinen and Vaisanen 1983, Murphy 1986, Mind 1988, Jirvinen and Pryl 1989, Leblanc 1989, Slagsvold and Lifjeld 1989, Langston et al. 1990, Reid and Boersma 1990, Wiggins 1990, Bolton 1991, Sydeman and Emslie 1992, Nilsson and Svensson 1993, Smith et al. 1993, Williams et al. 1993, Brouwer and Spaans 1994, Nager and Zandt 1994, Sandercock and Pedersen 1994, Amundsen 1995, Perrins 1996) about the close relation- ship between egg size and different aspects of female quality. When fitted to scenarios of selection as outlined above, our data most closely match with the model of apparent selection. According to this scenario, the ef- fect of egg size on recruitment rate emerged as a by-product of the effects of female condition on both egg size and recruitment rate.
At present we cannot offer an explanation of the mechanisms generating the positive correlation between egg size and parental brood-rearing ability. It might be possible that females in good prebreeding condition acquire mates with superior brood-rearing abilities. An- other explanation could be the effect of blood parasites on reproducing females. Infection with some haemato- zoan taxa has been shown to have a negative effect on egg size in the great tit (Dufva 1994, 1995; own unpub- lished data), probably because parasites compete with the female for energy and nutrients during egg forma- tion (Korpimaki et al. 1993, Allander and Bennett 1995). If these parasites have a negative effect on parental brood-rearing ability, a positive correlation
between egg size and recruitment rate would result. To examine this possibility, parasitological studies of our great tit population are in progress.
Selection on laying date
In one out of five years, early breeding females re- cruited significantly more offspring into the breeding population than late breeders. Regression analysis ac- counting for the year effects revealed a marginally significant effect of laying date on recruitment rate also in the pooled data. This effect became significant below the 5% level when the effect of clutch size was ac- counted for. However, the effect of early laying on recruitment rate disappeared when the egg size term was incorporated into the model. Our interpretation is that the primary target of selection was not laying date per se but some property of a female, which had an effect on both laying date and egg size. Comparing this result with the scenarios of selection described in Fig. 2, the closest match can be seen to be with the apparent selection model. Our conclusion is consistent with the 'quality hypothesis' for explaining the seasonal decline in reproductive success (see Askenmo 1982, Wiggins et al. 1994, Verhulst et al. 1995), which states that repro- ductive success declines seasonally because 'high qual- ity' birds breed earlier than 'low quality' birds. An alternative explanation would be that the contribution of egg size and laying date to breeding success covaried because of the strong genetic correlation between these traits, but this possibility cannot be evaluated on basis of available data.
Selection on clutch size
Selection on clutch size fluctuated in sign during the five year period, the net selection differential (s = -0.04) being close to zero. If clutch size was closely related to the condition of a laying female, we would expect large clutches to be the most productive ones, as found e.g. by Bryant (1975), De Steven (1980), Richter (1984), Haydock and Ligon (1986), Boyce and Perrins (1987), Rockwell et al. (1987), Gibbs (1988), Gustafsson and Sutherland (1988), and Briskie and Sealy (1989). Contrary to this expectation, the selection differential for clutch size in our study sometimes even had a relatively large negative value (s= -0.45 SD in 1989), indicating that individuals with large clutches were not always the most efficient in terms of brood rearing.
One possible explanation for the lack of consistent positive selection on clutch size is the gene flow hypoth- esis (cf. Perrins and Moss 1975, Dhondt et al. 1990). Nestling mortality in our urban great tit population was notably higher than in a neighboring rural popu- lation (H6rak 1993), while adaptive brood reduction
598 OIKOS 78:3 (1997)
This content downloaded from 193.40.5.245 on Thu, 05 Mar 2015 11:40:20 UTCAll use subject to JSTOR Terms and Conditions
was unlikely (HMrak 1995). This suggests that the aver- age clutch size of our urban great tit population was too large and did not match the average parental brood-rearing ability. This in turn could be explained by persistent immigration from rural areas where the laying of large clutches is not selected against (Horak 1993). Another not mutually exclusive explanation would be that laying females possess more unreliable information about the future food situation in the urban as compared to the rural environment, which might lead to errors in the clutch size decision process.
Selection against large clutches mainly hit individuals with small eggs (Fig. 5). Since small eggs are likely to reflect poor condition of the female, our result indicates that selection on clutch size may have been mediated through the effects of selection on female condition. This fits with the scenario of interactive selection in Fig. 2. If large egg size reflects good individual condition and small clutch size good adjustment to locally pre- vailing conditions, then it is likely that some individuals were successful because they were in good condition (birds with large eggs, including those with large clutches in Fig. 5), while others achieved success be- cause they made correct reproductive decisions under the given circumstances (birds with small clutches, in- cluding those in a poor condition).
Implications
Our results demonstrate the advantages of multivariate analysis for identifying the real targets of selection on avian reproductive traits. Thus, inclusion of several fitness components as predictor variables simulta- neously in the models indicated that the effect of laying date on recruitment rate was different for small and large clutches, and that the date effect was due to the same factors that affected the relationship between egg size and recruitment rate. The interactive effect of clutch size and egg size suggests the fitness value of clutch size to be state-dependent, while the direct fitness effect of egg size could be separated from the poten- tially confounding effects of laying date and clutch size.
A possibility to interpret these results is offered by the framework of three alternative scenarios of selec- tion, as described above. When seeking for the best match between the data and the predictions of the different scenarios, we have proceeded from the as- sumption that the component of individual condition which affects both breeding traits and fitness is reflected in egg size variation between individuals. However, it is important to note that this assumption relies on the relationship between egg size and female residual weight in the later stages of breeding. A crucial ques- tion here is to what extent the residual weight of a female in the nestling phase reflects her condition in the sense of 'integrated state-dependent quality factor'. Our
unpublished data suggest that a female's residual weight during the nestling phase is related to her pre- reproductive condition (correlation with clutch size: r=0.29; correlation with laying date: r= -0.23; for both p < 0.001 and N = 282; data from 1991-95), al- though the reproductive effort made operates to deteri- orate her condition (female's residual weight also correlates negatively with the number of nestlings fledged (r = -0.25, p <0.001; the same data set as above). Our assumption is therefore legitimate if egg size covaries with that component of female weight which is affected by her pre-reproductive condition.
In principle, we cannot exclude the possibility that similar results of a multivariate analysis of selection could emerge because of the genetic correlations be- tween the reproductive traits examined. Environmental correlations between these traits and fitness, however, seem to offer a more parsimonious explanation.
Acknowledgements - We thank Margus Ots for the assistance in the field and Krista Lapp and Thnu M6ls for the advice in data analysis. Toomas Tammaru and Staffan Ulfstrand pro- vided valuable comments on the manuscript. Peeter H6rak was supported by the Kone Foundation (Helsinki), by the Rudolf S6mermaa stipendiary fund within the corporation Ugala in Sweden, and by the Swedish Institute. The study was partially funded by the Estonian Science Foundation grant No. 872. Upgrading of the program OMELETTE was sup- ported by the University of Halle.
References Alatalo, R. V., Gustafsson, L. and Lundberg, A. 1990. Pheno-
typic selection on heritable size traits: environmental vari- ance and genetic response. - Am. Nat. 135: 464-471.
Allander, K. and Bennett, G. F. 1995. Retardation of breeding onset in great tits (Parus major) by blood parasites. - Funct. Ecol. 9: 677-682.
Amundsen, T. 1995. Egg size and early nestling growth in the snow petrel. - Condor 97: 345-351.
Askenmo, C. 1982. Clutch size flexibility in the pied flycatcher. - Ardea 70: 189-196.
Bolton, M. 1991. Determinants of chick survival in the lesser black-backed gull: relative contributions of egg size and parental quality. - J. Anim. Ecol. 60: 949-960.
Boyce, M. S. and Perrins, C. M. 1987. Optimizing great tit clutch size in a fluctuating environment. - Ecology 68: 142-153.
Briskie, J. V. and Sealy, S. G. 1989. Determination of clutch size in the least flycatcher. - Auk 106: 269-278.
Brouwer, A. and Spaans, A. L. 1994. Egg predation in the herring gull Larus argentatus: why does it vary so much between nests?. - Ardea 82: 223-231.
Bryant, D. M. 1975. Environmental influences of growth and survival of nestling house martins Delichon urbica. - Ibis 120: 271-283.
De Steven, D. 1980. Clutch size, breeding success, and parental survival in the tree swallow (Iridiprocne bicolor). - Evolution 34: 278-291.
Dhondt, A. A., Adriaensen, F., Matthysen, E. and Kempe- naers, B. 1990. Nonadaptive clutch sizes in tits. - Nature 348: 723-725.
Dufva, R. 1994. Effects of blood parasites on egg size in two bird species. - J. Ornithol. 135: 10.
- 1995. Parasites, reproductive success and health status in birds. - Ph.D. Diss., Uppsala Univ.
OIKOS 78:3 (1997) 599
This content downloaded from 193.40.5.245 on Thu, 05 Mar 2015 11:40:20 UTCAll use subject to JSTOR Terms and Conditions
Falconer, D. S. 1983. Introduction to quantitative genetics, 2nd ed. - Longman, New York.
Gibbs, H. L. 1988. Heritability and selection on clutch size in Darwins medium ground finches (Geospiza fortis). - Evo- lution 42: 750-762.
Greig-Smith, P. W., Feare, C. J., Freeman, E. M. and Spencer, P. L. 1988. Causes and consequences of egg-size variation in the European starling Sturnus vulgaris. - Ibis 130: 1-10.
Gustafsson, L. and Sutherland, W. J. 1988. The costs of reproduction in the collared flycatcher Ficedula albicollis. - Nature 335: 813-815.
Haydock, J. and Ligon, J. D. 1986. Brood reduction in the Chihuahuan raven: an experimental study. - Ecology 67: 1194-1205.
H6rak, P. 1993. Low fledging success of urban great tits. - Ornis Fenn. 70: 168-172.
- 1995. Brood reduction facilitates female but not offspring survival in the great tit. - Oecologia 102: 515-519.
-, Mdnd, R., Ots, I. and Leivits, A. 1995. Egg size variation in the great tit Parus major: individual, habitat and geo- graphic differences. - Ornis Fenn. 73: 97-114.
Jarvinen, A. and Vdisanen, R. A. 1983. Egg size and related reproductive traits in a southern passerine Ficedula hy- poleuca breeding in an extreme northern environment. - Ornis. Scand. 14: 253-262.
- and Pryl, M. 1989. Egg dimensions of the great tit Parus major in southern Finland. - Ornis Fenn. 66: 69-74.
Korpimaki, E., Hakkarainen, H. and Bennett, G. F. 1993. Blood parasites and reproductive success of Tengmalm owls: detrimental effects on females but not on males. - Funct. Ecol. 7: 420-426.
Langston, N. E., Freeman, S., Rohwer, S. and Gori, D. 1990. The evolution of female body size in red-winged black- birds: the effects of timing of breeding, social competition, and reproductive energetics. - Evolution 44: 1764-1779.
Larsson, K. 1992. Ecological and quantitative genetic pro- cesses in a population of the Barnacle Goose (Branta leucopsis). - Ph.D. Diss., Uppsala Univ.
Leblanc, Y. 1989. Variation in size of eggs of captive and wild Canada geese. - Ornis. Scand. 20: 93-98.
Linden, M., Gustafsson, L. and Part, T. 1992. Selection on fledging mass in the collared flycatcher and the great tit. - Ecology 73: 336-343.
Mind, R. 1988. Intrapopulational variation of avian eggs (In Russian with English summary). - Valgus, Tallinn.
- , Nigul, A. and Sein, E. 1986. Oomorphology: a new method. - Auk 103: 613-617.
McNamara, J. M. and Houston, A. I. 1992. State-dependent life-history theory and its implications for optimal clutch size. - Evol. Ecol. 6: 170-185.
Moller, A. P. 1994. Parasites as an environmental component of reproduction in birds as exemplified by the swallow Hirundo rustica. - Ardea 82: 161-172.
Morris, D. W. 1996. State-dependent life histories, Mount- ford's hypothesis, and the evolution of brood size. - J. Anim. Ecol. 65: 43-51.
Murphy, M. T. 1986. Body size and condition, timing of breeding, and aspects of egg production in eastern king- birds. - Auk 103: 465-476.
Nager, R. G. and Zandt, H. G. 1994. Variation in egg size in great tits. - Ardea 82: 315-328.
Nilsson, J.-A. and Svensson, E. 1993. Causes and conse- quences of egg mass variation between and within blue tit clutches. - J. Zool. 230: 469-481.
Ojanen, M., Orell, M. and Vaisanen, R. A. 1979. Role of heredity in egg size variation in the great tit Parus major and the pied flycatcher Ficedula hypoleuca. - Ornis. Scand. 10: 22-28.
Perrins, C. M. 1996. Eggs, egg formation and the timing of breeding. - Ibis 138: 2-15.
- and Moss, D. 1975. Reproductive rates in the great tit. - J. Anim. Ecol. 44: 695-706.
Price, T. and Liou, L. 1989. Selection on clutch size in birds. - Am. Nat. 134: 950-959.
- , Kirkpatrick, M. and Arnold, J. 1988. Directional selec- tion and evolution of breeding date in birds. - Science 240: 798-799.
Reid, W. V. and Boersma, P. D. 1990. Parental quality and selection on egg size in the Magellanic penguin. - Evolu- tion 44: 1780-1786.
Richter, W. 1984. Nestling survival and growth in the yellow- headed blackbird, Xantocephalus xantocephalus. - Ecology 65: 597-608.
Ricklefs, R. 1984. Components of variance in measurements of nestling European starlings (Sturnus vulgaris) in southern Pennsylvania. - Auk 101: 319-333.
Rockwell, R. F., Findlay, C. S. and Cooke, F. 1987. Is there an optimal clutch size in snow geese. - Am. Nat. 130: 839-863.
Sandercock, B. K. and Pedersen, H. C. 1994. The effect of renesting ability and nesting attempt on egg-size variation in willow ptarmigan. - Can. J. Zool. 72: 2252-2255.
SAS Institute. 1993. SAS/INSIGHT user's Guide, Version 6, 2nd ed. - SAS Inst. Inc., Cary, NC.
- 1994. SAS/INSIGHT Software: Changes and Enhance- ments, Release 6.10. - SAS Inst. Inc., Cary, NC.
Schluter, D. and Gustafsson, L. 1993. Maternal inheritance of condition and clutch size in the collared flycatcher. - Evolution 47: 658-667.
Slagsvold, T. and Lifjeld, J. T. 1989. Constraints on hatching asynchrony and egg size in pied flycatchers. - J. Anim. Ecol. 58: 837-849.
Smith, H. G., Ottosson, U. and Ohlsson, T. 1993. Interclutch variation in egg mass among starlings Sturnus vulgaris reflects female condition. - Omis. Scand. 24: 311-316.
- , Ohlsson, T. and Wettermark, K.-J. 1995. Adaptive signifi- cance of egg size in the European starling: experimental tests. - Ecology 76: 1-7.
Sydeman, W. J. and Emslie, S. D. 1992. Effects of parental age on hatching asynchrony, egg size and 3rd-chick disadvan- tage in western gulls. - Auk 109: 242-248.
van Noordwijk, A. J., van Balen, J. H. and Scharloo, W. 1981. Genetic environmental variation in the clutch size of the great tit (Parus major). - Neth. J. Zool. 31: 342-372.
- , van Balen, J. H. and Scharloo, W. 1988. Heritability of body size in a natural population of the great tit Parus major and its relation to age and environmental conditions during growth. - Genet. Res. 51: 149-162.
Verhulst, S., van Balen, J. H. and Tinbergen, J. M. 1995. Seasonal decline in reproductive success in the great tit: variation in time or quality? - Ecology 76: 2392-2403.
Wiggins, D. A. 1990. Sources of variation in egg mass of tree swallows Tachycineta bicolor. - Ornis. Scand. 21: 157-160.
- , Part, T. and Gustafsson, L. 1994. Seasonal decline in collared flycatcher Ficedula albicollis reproductive success: an experimental approach. - Oikos 70: 359-364.
Williams, T. D. 1994. Intraspecific variation in egg size and egg composition in birds: effects on offspring fitness. - Biol. Rev. Camb. Philos. Soc. 69: 35-59.
- , Lank, D. B., Cooke, F. and Rockwell, R. F. 1993. Fitness consequences of egg-size variation in the lesser snow goose. - Oecologia 96: 331-338.
Appendix. Distribution of the number of recruits per brood
![TIT E IRISH IN CANA,''nq ; `]'III; I~i.I'Olt'l'~1'l'IOV](https://static.documents.pub/doc/80x56/631994471a1adcf65a0eb707/tit-e-irish-in-cananq-iii-iiioltl1liov.jpg)
