The probability and success of a second brood, adult survival until and reproduction in the next season were then compared to the natural variation among control pairs.. Components of su
Trang 1q 2002 The Society for the Study of Evolution All rights reserved.
TIMING OF CURRENT REPRODUCTION DIRECTLY AFFECTS FUTURE
REPRODUCTIVE OUTPUT IN EUROPEAN COOTS
MARTINW G BRINKHOF,1,2 ANTONJ CAVE´,1 SERGE DAAN,3 AND ALBERT C PERDECK1
1Netherlands Institute of Ecology, Center for Terrestrial Ecology, P.O Box 40, NL-6666 ZG Heteren, The Netherlands
3Zoological Laboratory, University of Groningen, P.O Box 14, NL-9750 AA Haren, The Netherlands
Abstract. Life-history theory suggests that the variation in the seasonal timing of reproduction within populations
may be explained on the basis of individual optimization Optimal breeding times would vary between individuals as
a result of trade-offs between fitness components The existence of such trade-offs has seldom been tested empirically.
We experimentally investigated the consequences of altered timing of current reproduction for future reproductive
output in the European coot (Fulica atra) First clutches of different laying date were cross-fostered between nests,
and parents thereby experienced a delay or an advance in the hatching date The probability and success of a second
brood, adult survival until and reproduction in the next season were then compared to the natural variation among
control pairs Among control pairs the probability of a second brood declined with the progress of season Delayed
pairs were less likely and advanced pairs were more likely to produce a second brood These changes were quantitatively
as predicted from the natural seasonal decline The number of eggs in the second clutch was positively related to egg
number in the first clutch and negatively related to laying date Compared to the natural variation, delayed females
had more and advanced females had fewer eggs in their second clutch The size of the second brood declined with
season, but there was no significant effect of delay or advance Local adult survival was higher following a delay and
reduced following an advance The effect of the experiment on adult survival was independent of sex Laying date
and clutch size of females breeding in the next year were not affected by treatment The study demonstrates the
existence of a trade-off between increased probability of a second brood and decreased parental survival for early
breeding Timing-dependent effects of current reproduction on future reproductive output may thus play an important
role in the evolution of the seasonal timing of reproduction.
Key words Adult survival, Fulica atra, future reproduction, life-history evolution, multiple breeding, timing of
reproduction.
Received July 31, 2000 Accepted October 22, 2001.
Models of life-history evolution assume that the variation
in reproductive traits within populations may be explained
on the basis of trade-offs between fitness components A
major trade-off is between investment in current reproduction
and future reproductive output (Williams 1966; Lessells
1991; Roff 1992; Stearns 1992) This trade-off incorporates
the concept of a cost of reproduction: An increase in current
reproductive effort can only be at the expense of a reduction
in residual reproductive value Costs of reproduction are
prin-cipally expressed in survival and future fecundity (Bell 1980)
and should be investigated by experimental manipulation
within the natural range of variation of the trait concerned
(Linde´n and Møller 1989) The experimental evaluation of
trade-offs between current and future reproduction is a major
goal for understanding the evolution of life-history patterns,
such as the variation in timing of reproduction within animal
populations (Daan and Tinbergen 1997)
The seasonal timing of reproduction is a reproductive trait
that often has a large impact on fitness (Clutton-Brock 1988),
as has been shown in insects (e.g., Ohgushi 1991; Landa
1992; Cushman et al 1994), fish (e.g., Schultz 1993), reptiles
(e.g., Olsson and Shine 1997), and mammals (e.g.,
Festa-Bianchet 1988) Most descriptive and nearly all experimental
research on the topic has focused on birds (Nilsson 1999)
In most avian populations, early breeding individuals produce
larger clutches, more fledglings, and eventually more recruits
from their first clutch than late breeders (Klomp 1970; Perrins
2 Present address: University of Bern, Zoological Institute,
Di-vision of Evolutionary Ecology, Wohlenstrasse 50 A, CH-3032
Hin-terkappelen, Switzerland; E-mail: martin.brinkhof@esh.unibe.ch.
1970; Daan et al 1989; Rohwer 1992) Early breeders are also more likely to raise an additional brood in the same breeding season (e.g., Kluyver et al 1977; Smith et al 1987; Geupel and DeSante 1990; Hepp and Kennamer 1993; Ver-boven and Verhulst 1996) They also may molt earlier (Leaf-loor and Batt 1990) and use a longer post-breeding period
to recover, which may enhance their survival until next breed-ing (Nilsson and Svensson 1996) Thus, the overall picture
is one of early breeders having a higher fitness than late breeders
The apparent natural seasonal decline in fitness does not imply that there is directional selection for earlier breeding dates (Cave´ 1968; Price et al 1988) Two mechanisms have been proposed to explain seasonal variation in single fitness components: the parental quality hypothesis and the date hy-pothesis (Brinkhof et al 1993; Nilsson 1999) The parental quality hypothesis predicts that variation in the value of fit-ness components with date reflects differences in phenotype (e.g., age or breeding experience, Sæther 1990; Perdeck and Cave´ 1992) or environmental quality (e.g., territory quality, Alatalo et al 1986) between early and late breeders Differ-ences in reproductive output between early and late individ-uals would then not reveal the consequences of an alternative timing of breeding for the individual The date hypothesis predicts that earlier or later timing of breeding affects the value of a fitness component for all pairs alike However, variation in the trade-offs between individuals might result
in unforeseen effects on other fitness components and thereby
on total fitness (Daan et al 1990) Ultimately, the optimal breeding date may be determined by a trade-off between cur-rent and future reproduction, the outcome of which depends
Trang 2on parental and territory quality (Daan et al 1990; Daan and
Tinbergen 1997; Nilsson 1999) Individuals may thus adopt
different strategies to maximize fitness (individual optimal
date hypothesis) Individual optimization may explain the
absence of a relationship between breeding time and parental
survival in several species (e.g., Newton and Marquiss 1984;
Daan et al 1990; Winkler and Allen 1996) despite quality
differences between early and late breeders (Nilsson 1999)
Thus, the consequences of timing of breeding for an
indi-vidual can only be assessed experimentally
Reproductive costs have been experimentally
demonstrat-ed as a rdemonstrat-eduction in survival or future fecundity following
brood size enlargements (reviewed by Linde´n and Møller
1989; Dijkstra et al 1990) Experimental date manipulations
mostly indicate timing per se as the most important factor
for seasonal variation in reproductive success during the
pre-fledging and postpre-fledging period, supporting the date
hy-pothesis (Nilsson 1999) Thus, an experimental change in
breeding time generally alters the probability of nestling
sur-vival (e.g., Brinkhof et al 1993; Norris 1993; Wiggins et al
1994; Brouwer et al 1995) Even if an experimentally altered
timing has no direct effect on fledging success (e.g.,
Hatch-well 1991; Verhulst et al 1995; De Forrest and Gaston 1996),
future reproductive output may change through the effects of
breeding date on other stages of the life cycle, such as molt
(Nilsson and Svensson 1996) Few studies have examined
the effects of date manipulations beyond survival and growth
of first brood offspring (Nilsson 1994; Verhulst et al 1995;
Verboven and Verhulst 1996; Nilsson and Svensson 1996;
reviewed by Nilsson 1999) Most of these studies employed
artificial delays in breeding only, although experimental
ad-vances are crucial in the distinction between some models of
optimal timing (Daan and Tinbergen 1997) No study has
investigated the effect of an experimental delay as well as
advance on major components of future reproduction,
in-cluding the probability of producing a second brood as well
as adult survival
In this study, we examine how timing of reproduction
af-fects the incidence of second broods and parent survival of
European coots, Fulica atra We manipulated the timing by
exchanging first clutches of equal size, but differing in laying
date between nests, thereby creating delayed and advanced
pairs over most of the season (Brinkhof et al 1993) The
date hypothesis predicts that an experimental delay should
lead to a decline in the production of second broods, whereas
an advance should lead to an increase Under the parental
quality hypothesis we expect no effect of our manipulation
on the production of second broods Adult survival in coots
is independent of season, which may be the result of
indi-vidual date optimization, if parents refrain from breeding
earlier because survival costs outweigh the benefits for
cur-rent reproductive output Thus, experimentally delayed and
advanced pairs are predicted to show enhanced and reduced
adult survival, respectively
MATERIALS ANDMETHODS
General and Experimental Procedures
Fieldwork was conducted from 1988 to 1993 at the lake
Westeinderplassen (578189N, 48429E), 17 km southwest of
Amsterdam, The Netherlands For a description of the study area, see Cave´ and Visser (1985) The annual number of breeding pairs ranged from 135 to 157 Most adults were marked with steel leg bands and numbered plastic neck col-lars Throughout the breeding season (mid-March through mid-July) the area was searched at least once a week to locate nest sites, determine laying date and clutch size, and identify marked parents Laying date (of the first egg) was determined
by backdating, assuming a laying frequency of one egg per day Incubation starts before the clutch is completed, and eggs hatch asynchronously The mean hatching date of the nestlings in each brood was used in the analyses (Brinkhof
et al 1993) Successful clutches were defined as those that hatched at least one young The number of young surviving
in successful broods was determined weekly until at least 4 weeks of age Components of subsequent reproduction stud-ied included the probability and success of a second brood
in the same breeding season, adult survival until the following breeding season, and laying date and clutch size of females breeding the next season Calendar dates were expressed as the day of the year (e.g., 1 May 5 day 121)
The study uses both natural and experimentally induced variation in the timing of first broods To investigate the natural relationship between timing of the first brood and the probability and success of a second brood, we used data from all breeding pairs that successfully hatched a first brood in the years 1988–1993 (Table 1) In 1988, 1989, and 1991 we manipulated the timing of the brood care period for individual pairs by exchanging first clutches of equal size, differing by
10 days in laying date, between nests For further details on the experimental design, see Brinkhof et al (1993) The ex-periment confronted individual pairs with a delayed or an advanced hatching date of young, and thus induced them to start parental care for the foster brood at a later or earlier date than anticipated By comparing seasonal variation in fitness components between control and experimental pairs,
we can discriminate between the date and quality hypotheses Moreover, the experiment allowed us to investigate the costs
or benefits of an advanced or delayed breeding date in terms
of future reproduction at the individual level Within years, control and experimental pairs showed similar variation in clutch size and hatching date of the first brood (Table 1)
Second Broods
Second broods are those initiated after successful hatching
of a first brood within the same breeding season, irrespective
of the number of first-brood young eventually surviving The observed incidence of second broods is the result of initiation rate and finding probability Coots may switch nests between breeding attempts, and not all second broods initiated are found In particular, some nests may be predated prior to detection The observed incidence of second broods therefore underestimates the actual initiation rate We have no means
to estimate the finding probability of initiated second broods
in our population and assume that this probability was in-dependent of year, timing, success, and experimental manip-ulation of the first brood (see below) Second broods are readily discovered once the eggs have hatched, either visually
or acoustically by the vocalization of small chicks Thus,
Trang 3TABLE 1 Mean clutch size and original hatching date of the first clutch for control, delayed, and advanced pairs in 1988 – 1993 Within years, there were no significant differences in clutch size between control and experimental groups The number of individuals wearing a neck collar indicates the number used in survival analyses.
Clutch size
Original hatching date
No with neck collar Females Males
control
delayed
8 116 10
7.9 7.5 8.0
0.8 1.6 1.0
128.5 124.9 118.1
7.4 12.3 6.7
2 73 5
6 83 2
control
delayed
17 101 17
7.0 7.1 7.0
1.4 1.6 1.2
131.8 127.6 120.1
10.5 16.2 12.0
14 60 11
15 68 13 1990
1991
control
advanced
control
delayed
112 26 89 25
7.2 6.9 7.3 7.0
1.4 1.2 1.7 1.3
132.0 138.8 134.6 129.5
12.1 11.4 12.1 11.5
49 18 57 15
45 21 47 20 1992
1993
control
control
104 122
7.2 7.1
1.6 1.2
137.4 131.7
14.3 12.1
whether a breeding pair reared second-brood young until at
least 2 weeks of age was reliably determined in scheduled
weekly visits to each territory The assessment of brood size
was blind to treatment and accurate, because counts in
sub-sequent weeks rarely exceeded those of the previous one The
number of young surviving up to 2 weeks of age largely
determines the number of young raised to independence
(Brinkhof et al 1993) and was used in the analysis
Analysis of Breeding Parameters
Data analysis was performed using Statistix (Analytical
Software 1992) or GLIM (Francis et al 1993) by fitting
gen-eralized linear models using a stepwise backward-elimination
procedure (Crawley 1993) These models include multiple
regression models with continuous explanatory variables
(variates), analysis of variance (ANOVA) with categorical
variables (factors), as well as models with any mixture of
factors and (co)variates (ANCOVA) and their interactions
Logistic regression (binomial error) was used when analyzing
proportions, Poisson regression (Poisson error) for analyzing
count data, and significance was tested using the chi-square
test Linear regression (normal error) with F-tests was used
in other analyses All statistical tests are two-tailed
Basic explanatory variables in the analysis of breeding
parameters were year (as a factor), hatching date of the first
brood, and brood size 2 weeks after hatching In the analysis
of clutch or brood size of second broods, we also used the
laying date of the second clutch as a predictor variable The
maximal model (starting point for analysis) also included
two-way interactions and quadratic terms The size of the
first brood at 2 weeks after hatching was used to investigate
the effect of the success of the first brood on the probability
and success of second broods, because renesting normally
takes at least 2 weeks (i.e., the mean interbrood interval,
defined as the number of days between the hatching date of
the first brood and the laying date of the second clutch, was
21.7 days, SE5 1.3, n 5 70).
The earliest coots of the year tend to be the largest females,
who breed with the oldest males in the population (Perdeck
and Cave´ 1992; A.C Perdeck, unpubl data) on the largest
territories with abundant natural vegetation (Cave´ et al 1989;
A C Perdeck, unpubl data) These parameters of territory
or bird quality were not included in the analyses of breeding parameters and adult survival (next section) because we ran-domly assigned pairs to the experimental treatments, irre-spective of territory and bird quality The natural variation
in breeding parameters was taken as the starting point for the analysis of experimental variation The variation among experimental groups was initially examined using manipu-lation as a factor In case a breeding parameter varied with hatching date among controls, we tested manipulation as a variate (i.e., the number of days manipulated), first, relative
to the natural variation of the hatching date of the experi-mental pair’s original clutch (original hatching date, OHD) and, second, relative to the actual hatching date (AHD) of the fostered clutch This procedure allowed us to discriminate between the date hypothesis and the parental quality hy-pothesis (Brinkhof et al 1993; Verboven and Verhulst 1996)
Adult Survival
Survival estimates of adult birds were determined on the basis of capture-recapture data using the program MARK (Cooch and White 1998; White 1998; White and Burnham 1999) We excluded nests outside the main study area and nests with a failed first clutch, unknown hatching date, or extra food provisioning (Brinkhof and Cave´ 1997) Recapture was based on the visual observation of neck-collared indi-viduals Survival estimation is thus local and underestimates real survival, because some individuals may emigrate per-manently from the study area Overall mean real adult sur-vival, estimated from ringing recoveries of dead adult birds ringed in our study area from 1966 to 1978, was 0.70 (95% confidence interval: 0.65–0.75; Perdeck 1998) The overall estimate in the present study (data 1989–1993) was 0.59 (95% confidence interval: 0.54–0.63; data 1989–1993, control birds
of model f[m]P[s], see below) Thus, the ratio of local to
real survival is about 0.86, indicating that the most adults not locally recaptured in the study area are dead We therefore used local recaptures only, and, in discussing the results, we will assume that variation in local survival probability reflects genuine variation in survival
In 1988, the number of experimental manipulations in-volving neck-collared individuals was too few to obtain sur-vival estimates (Table 1) We therefore restricted the sursur-vival
Trang 4analysis to observations from 1989 through 1993, comparing
the survival of individuals with experimentally delayed or
advanced breeding in 1989 or 1991 with that of
unmanipu-lated controls in these years (Table 1) In the initial analysis
we considered three factors: time (t, year), sex (s), and
ma-nipulation (m; with three levels: control [c], delayed [d],
ad-vanced [a]) We applied the Cormack-Jolly-Seber (CJS)
method to model time dependency in both recapture and
sur-vival probability CJS models are denoted (Ft, Pt), whereFt
is the survival probability from year t 2 1 to year t (including
possible permanent emigration) and Pt is the probability of
recapture of birds alive in year t.
To avoid sparse data problems, we considered the data as
two three-encounter sets, that is, the first set contained the
capture data for 1989, 1990, and 1991 and the second set
those for 1991, 1992, and 1993 For each of the two sets of
encounters histories (referred to as replicates, r) six groups
were formed: (1) control females; (2) delayed females; (3)
advanced females; (4) control males; (5) delayed males; and
(6) advanced males A consequence of a three-encounter
anal-ysis is that survival and capture rate of the second year t1
1 cannot be estimated independently Note that an individual
can be present in both sets and is thus counted twice We
assumed that the manipulations had an effect on survival until
the year after the manipulation only
Traditionally, the analysis is carried out by a series of
likelihood-ratio tests (LRTs), starting with a model
contain-ing all factors of interest (the maximal model in GLIM
ter-minology) and removing one by one the nonsignificant terms
until a minimal adequate model is obtained This method
works for nested models only, and the Aikaike information
criterion (AIC) has been proposed as an alternative tool to
select the most parsimonious model (Cooch and White 1998)
Two main reasons for considering AIC as a general model
selection tool instead of LRT include: (1) the equal or better
performance of AIC when data meet or violate standard
mark-recapture assumptions, respectively; and (2) that AIC
pro-vides a criterion that is not affected by the number of tests
performed (i.e., no need for adjustment of a-values to the
number of tests performed, e.g., using Bonferroni method)
The use of AIC requires an a priori set of models from which
the one with the lowest AIC value is chosen as the best model
Comparisons with next-lowest models cannot be evaluated
by a statistical test Instead, the degree of support is given
using an index of relative plausibility known as AIC-weight
We used both LRTs and AIC criteria in our analysis
In defining the maximal model we made the following
considerations The main aim was to investigate the effect
of manipulation (m) on adult survival from the experimental
year (1989, 1991) to the following year Survival for the
experimental years may also differ between the sexes (s) and
the effects of m on survival rate could differ between the
sexes (interaction s.m) Capture probability may vary with
sex, manipulation, or both; therefore, s, m, and s.m were
included in the capture rate part of the model The survival
and capture rate for the second year cannot be estimated
independently in a three-encounter dataset We assumed that
the capture rate within replicates (r) was independent of year
(t), and obtained survival estimates for the two years of each
replicate separately To test for differences in survival rate
or capture rate between replicates, we included t, r, and the interaction t.r in the survival part of the model and r in the capture part (t and t.r are nonexistent for capture rate)
Fi-nally, because the effects of manipulation on survival and capture rates may differ between replicates, the interaction
m.r was included in both parts of the model Summarizing,
the maximal model we considered in the analysis was [f(t/
r 1 s/m 1 m*r) P(s/m 1 m*r)], in which t/r stands for t 1
t.r, s/m for s 1 s.m, and m*r for m 1 r 1 m.r.
To evaluate the goodness-of-fit of this model, we used the parametric bootstrap procedure provided in MARK (Cooch
and White 1998) A P-value for the goodness-of-fit test of
0.97 was obtained from 100 simulations The model showed
no overdispersion (cˆ5 0.973)
RESULTS
Detecting a Second Brood
Of all second broods recorded in the years 1988–1993 (n
5 94), 83% (n 5 78) were detected prior to hatching Over
successive years (1988–1993) the proportion of second
broods found prehatching was 0.72 (n 5 18), 0.65 (n 5 23),
1.00 (n 5 6), 0.95 (n 5 20), 0.93 (n 5 14), and 0.92 (n 5
13), respectively The probability of detecting a second brood before hatching (eggs found5 1, no eggs found 5 0) differed
significantly between years (logistic regression, (x25 12.58,
df 5 5, P 5 0.028), but was independent of hatching date
of the first brood (x25 0.91, df 5 1, P 5 0.34), size of the
first brood (x25 0.46, df 5 1, P 5 0.50), and experimental
manipulation (tested as factor, i.e., control, delayed, advance: (x2 5 3.25, df 5 2, P 5 0.20) We therefore assume that
within-year variation in the observed rate of second broods reflects the actual variation in initiation rate
Probability of a Second Brood
The natural variation in the probability of a second brood was analyzed among 644 unmanipulated pairs that success-fully hatched a first clutch in 1988–1993 The incidence of second broods varied significantly between years (tested as factor) and with hatching date and size of the first brood (Table 2A) The size of the first clutch did not affect the probability of a second brood (x25 2.56, df 5 1, P 5 0.11).
The probability of a second brood progressively declined with season (Fig 1A), and the slope of this decline was similar between years (for interaction year3 hatching date, (x2 5
9.45, df 5 5, P 5 0.092) In addition, the incidence of a
second brood gradually declined with the size of the first brood (Fig 1B)
The probability of a second brood among experimentally
delayed (n 5 51) and advanced (n 5 50) pairs was compared
with the expectation on the basis of the natural seasonal trend
found among controls in 1988, 1989, and 1991 (n 5 306)
In agreement with the general trends (i.e., 1988–1993), the incidence of second broods among control pairs declined sig-nificantly with hatching date and size of the first brood, whereas the variation between years was not significant (x2
5 5.35, df 5 2, P 5 0.07; cf Table 2A) The model containing
hatching date and size of the first brood was taken as basis for the analysis of the experimental variation
Trang 5TABLE 2 Logistic regression analysis of the natural (A) and experimental (B, C) variation in the occurrence of second broods Model A is
based on all unmanipulated broods (n5 644, data 1988 – 1993); for year, the mean constant over the five years is given Models B and C give
the result of the analysis among control (n 5 306) and experimental (delayed, n 5 52; advanced, n 5 51) pairs in 1988, 1989, or 1991 Model
B gives the effect of manipulation relative to the hatching date of the parent’s original first clutch; model C gives it relative to the actual hatching date of the adopted clutch Manipulation was tested as a variate, that is, the number of days delayed ( 1 values) or advanced (2 values).
(Increase in)
Coefficients estimate 6 SE
Final model
Constant
Year
Hatching date of first brood
Brood size of first brood
479.41 276.71 11.03 141.64 98.03
643 636 1 5 1 1
0.051 ,0.001 ,0.001
17.3 6 2.01 20.151 6 0.0175 20.934 6 0.120
B, C
B
Null model
Final model
Constant
Original hatching date
Brood size of first brood
Manipulation
344.56 182.59 100.02 73.64 7.18
408 405 1 1 1 1
,0.001 ,0.001 0.007
19.06 6 2.558 20.1496 6 0.0198 21.153 6 0.170 20.0979 6 0.0372
Constant
Actual hatching date
Brood size of first brood
Manipulation†
184.64 109.45 77.76 2.05
406 1 1 1 1
,0.001 ,0.001 0.152
18.36 6 2.475 20.1445 6 0.0190 21.123 6 0.1661
† Excluded from the model.
First, the experimental variation in the occurrence of
sec-ond broods was compared to the level expected on the basis
of original timing The probability of a second brood was
significantly related to OHD, brood size and manipulation
(Table 2B) The coefficients for OHD and brood size indicate
the established natural decline in the probability of a second
brood with hatching date (Figs 1C, D; solid line) and with
offspring number of the first brood The significance of
ma-nipulation indicates that the probability of a second brood
among experimental pairs differed significantly from that
ex-pected from the natural seasonal trend on OHD For a given
OHD, delayed pairs showed a lower frequency of second
broods and advanced pairs a higher frequency of second
broods (Table 2B, see coefficient) Thus, the experiment
af-fected the occurrence of second broods, which contradicts
the parental quality hypothesis
Second, to test the date hypothesis we compared the
ex-perimental variation in the incidence of second broods to the
level expected on the basis of the AHD Given the established
decline in the probability of a second brood with hatching
date (note that AHD equals OHD for controls) and size of
the first brood, the effect of the manipulation was not
sig-nificant (Table 2C) Thus, quality differences between early
and late breeders had no additional effect beyond the effect
of breeding time itself The incidence of second broods
among experimental pairs was explained by the AHD of the
first clutch: A delay in the hatching date of the first clutch
led to a decline, whereas an advance led to an increase in
the incidence of second broods (Figs 1C, D) This result is
consistent with the date hypothesis
Interbrood Interval
The length of the interbrood interval was determined in 70
of 79 control pairs that started a second brood in 1988–1993
In the remaining nine cases, the laying date of the second clutch was unknown The interbrood interval varied between
8 and 69 days, with a mean of 21.7 days (SE 5 1.3) The
interbrood interval was positively related to the size of the
first brood (linear regression; F1,685 15.17, P , 0.001; R2
5 0.19) and independent of year (F5,63 5 1.23, P 5 0.31)
and hatching date of the first brood (F1,675 2.26, P 5 0.14).
In particular, pairs that raised only one or no first-brood chick renested sooner than more successful pairs
To investigate the effect of the experiment on interbrood interval, data from 38 control, three delayed, and 10 advanced pairs were available (data 1989, 1991, 1991) The interbrood interval increased significantly with size of the first brood
(F1,495 16.71, P , 0.001) Manipulation was not significant
when tested as variate (F1,485 0.051, P 5 0.82) nor as factor
(control, delayed, advanced; F2,475 0.43, P 5 0.65) Thus,
the increase in the interbrood interval with success of the first brood among experimental pairs was similar to that among controls, suggesting that the relationship was inde-pendent of the timing of breeding
Clutch Size of the Second Brood
The second clutch (mean5 6.1 6 0.2 eggs) from
unma-nipulated pairs (data 1988–1993) was significantly smaller than that of the first clutch (mean 5 7.2 6 0.2 eggs; paired
t-test, t 5 24.64, df 5 52, P , 0.0001) Size of the second
clutch differed significantly between years, and within years there was a progressive decline in clutch size with laying date Females with a large first clutch also produced a large second clutch (Table 3A, see sign of coefficients) The effect
of hatching date of the first clutch was not significant (F1,44
5 0.51, P 5 0.48).
Sample sizes for analyzing the effect of the experiment on the egg number in the second clutch were small (data 1988,
Trang 6FIG 1 (A, B) Natural variation in the occurrence of second broods (data 1988–1993) with hatching date and second-week size of the first brood (C, D) Seasonal variation in the occurrence of second broods for experimentally delayed and advanced pairs The solid line gives the natural seasonal decline in the probability of a second brood as based on control pairs only Dots represent the mean initiation rate of second broods for experimental pairs, grouped according to the actual hatching date of the first brood For sample sizes, see the top of each graph The arrows give the effect of the manipulation Arrow tails give the probability of a second brood on the hatching date of experimental pair’s original first clutch; arrow heads point toward the mean probability found after delay or advance Under the date hypothesis, the probability of a second brood following delay or advance should be as predicted from the natural seasonal trend (solid line) The parental quality hypothesis predicts no effect of a 10-day delay or advance on the probability of a second brood The broken line indicates this prediction.
1989, 1991; delayed n 5 3, advanced n 5 9, controls n 5
27) Second clutches did not differ significantly in size
be-tween years in this dataset (F2,335 1.15, P 5 0.33), whereas
the decline in clutch size with laying date and the increase
in egg number with the size of the first clutch (P 5 0.06)
were confirmed (cf Tables 3A, B) Delayed and advanced
pairs produced, on average, 0.67 eggs more or less,
respec-tively, than expected on basis of laying date of the second
clutch and size of the first clutch (Table 3B)
Brood Size of the Second Brood
The number of young surviving two weeks after hatching
was analyzed using Poisson regression Among
unmanipu-lated pairs (data 1988–1993, n5 79), the size of the second
brood progressively declined with hatching date of the first
brood (x2 5 37.43, df 5 1, P , 0.0001; Fig 2) Effects of
year (x25 7.05, df 5 5, P 5 0.22) and size of the first brood
(x25 2.40, df 5 1, P 5 0.12) were not significant In addition,
the effect of second brood hatching date was not significant
(x2 5 0.39, df 5 1, P 5 0.53), indicating that the number
of second brood young was better explained by the timing
of the first brood than by that of the second brood
In agreement with the general pattern (data 1988–1993), brood size declined progressively with hatching date of the first brood among controls (x25 15.00, df 5 1, P , 0.001;
n 5 46; data from 1988, 1989, 1991) As predicted by the
parental quality hypothesis, the number of second brood
young of delayed (n 5 5) or advanced pairs (n 5 10) did
not deviate significantly from the number expected on the basis of their original timing (manipulation tested as variate: (x2 5 0.02, df 5 1, P 5 0.89; tested as factor: (x2 5 0.02,
df 5 2, P 5 0.99) However, brood size was also not
sig-nificantly different from the size expected for the actual hatching date (tested as variate: (x2 5 2.27, df 5 1, P 5
0.13; tested as factor: (x2 5 2.27, df 5 2, P 5 0.32) Thus,
the data do not distinguish between date and parental quality differences as the cause for the natural seasonal decline in size of the second brood, which may be due to the small sample sizes
Adult Survival
We started the analysis of adult survival by modeling re-capture probability The AIC-based selection indicated the recapture-model with sex only as the best model (Table 4A,
Trang 7TABLE 3 Linear regression models describing (A) the natural variation (data 1988 – 1993) and (B) the experimental variation (data 1988,
1989, 1991) in size of the second clutch Significance was tested using F-tests.
(Increase in)
Coefficients estimate 6 SE
Final model
Constant
Year
Laying date of second clutch
Clutch size of first clutch
80.53 41.26 12.90 11.43 8.56
52 45 1 5 1 1
0.027 0.001 0.004 20.0418 6 0.0118
0.249 6 0.0815
Final model
Constant
Laying date of second clutch
Clutch size of first clutch
Manipulation
49.59 37.65 6.77 4.13 4.57
38 35 1 1 1 1
0.017 0.058 0.047
9.48 6 1.92 20.0335 6 0.0133 0.185 6 0.0943 0.0669 6 0.0325
FIG 2 Natural seasonal variation in the number of second brood
young (data 1988–1993).
model 4) Of the five preconceived models, it had the lowest
AIC and 2.8 times the support of the nearest best model
(Table 4A, model 3) Recapture probability (p) was higher
in males (p 5 0.97, SE 5 0.02) than in females (p 5 0.76,
SE5 0.06) The lower recapture probability in females was
not due to an overall lower probability of identifying (i) the
female parent in relation to a recorded first brood in the study
area (data 1989–1993, range 162–178 broods; i 5 0.96, SE
5 0.01, both for males and females) The difference in
re-capture probability thus suggests that females are more likely
to skip a breeding season or to breed outside the study area
than males It is important to note that manipulation had no
effect on recapture probability In addition, p was
indepen-dent of r, indicating that recapture probability did not differ
between the two experimental years
Next we modeled survival by taking model 4, which
con-trolled for variation in recapture probability with sex, as a
starting point (Table 4B) The interactions s*m, m*r, and t*r
were not relevant to survival (Table 4B, see AIC-weight and
LRTs to model 4) Thus, the effect of manipulation on
sur-vival did not differ between the sexes or between replicates,
whereas among replicates there was no difference in survival
between the first and the second year In addition, the main
factors s and t were not relevant to survival (Table 4B, models
6 and 7, respectively) Thus, males and females had similar survival rates, and first-year survival was similar to the sec-ond-year survival within replicates The AIC-based selection
indicated model 8, which contained manipulation (m) and replicate (r), as the best model This model has nearly 1.5
times the support than the nearest best model (i.e., model 9), which contained manipulation only (Table 4B) Replicate ac-counts for the overall higher first-year survival shown by the
1991 group (Fig 3, best shown by controls) Removing ma-nipulation from the best model reduces the support by nearly five times (Table 4B, cf AIC-weights to models 8 and 10)
We therefore conclude that manipulation was the main factor explaining variation in survival rate among adults
To assess whether the effect of manipulation was due to the advance, the delay, or both treatments, we tested their effects separately Thus, we compared the current best model (model 8) with two distinct models, one with advanced pairs removed (Table 4C, model 12) and one with delayed pairs removed (Table 4C, model 11) AIC-based selection indi-cated model 8 as the best and minimal adequate model, with
a nearly two-fold support over the two alternative models Therefore, we conclude that individuals with experimentally delayed breeding dates showed a higher survival, whereas those with experimentally advanced breeding showed lower survival than controls of the same year (Fig 3)
We assumed that delayed and advanced parents did not differ in quality and that the effect of manipulation is limited
to the year following manipulation As a consequence of the experimental design, which requires early-laid clutches to be swapped with late-laid ones, on average, delayed pairs had earlier original hatching dates than controls, whereas ad-vanced pairs had later original hatching dates than controls (see Table 1) Thus, the possibility exists that delayed birds were of a higher quality than advanced birds, and this could
in principle also explain the effect of the manipulation on survival (an indirect or chronic effect; see Burnham et al 1987) To check this, we modeled survival with the inclusion
of the original hatching date of the first brood (h) as an
individual covariate In a set of modelsf(r 1 m)P(s), f(r 1
m 1 h)P(s) and f(r 1 m 1 h 1 h2)P(s) the AIC-weights were
0.52, 0.35, and 0.13, respectively The LRTs also showed
Trang 8TABLE 4 Capture-recapture models for coot parents For each model the Aikaike information criterion (AIC), the AIC weight (calculated
separately within each a priori set of models A, B, or C), the number of parameters (np) and the deviance (Dev) are given All models include
an intercept, both for survival (f ) and recapture probability (P) Linear predictors are year (t), replicate (r; datasets 1989, 1990, 1991 and
1991, 1992, 1993), and sex (s) The factor manipulation (m) has three levels: control (c), advanced (a), and delayed (d) Model algebra specification is according to GLIM (Crawley 1993) and m*r stands for m 1 r 1 m.r, t/r for t 1 t.r, and s/m for s 1 s.m Within each set of
analyses, models are numbered according to decreasing complexity, but ordered according to AIC The selected model in each set of analysis
as well as the key comparisons between models are presented in bold type.
Models compared, hypothesis tested, likelihood-ratio test
A Modeling recapture probability (P):
4. f(t/r1s/m1m*r)P(s)
3. f(t/r1s/m1m*r)P(s1r)
1. f(t/r1s/m1m*r)P(s/m1m*r)
2. f(t/r1s/m1m*r)P(m1s1r)
5. f(t/r1s/m1m*r)P(.)
1119.94
1121.99 1123.55 1124.57 1129.07
0.613
0.219 0.101 0.060 0.006
13
14 17 16 12
14.46
14.44 9.73 12.84 25.67
3 – 4, r, P5 0.88
2 – 3, m, P5 0.45
1 – 2, interactions, P5 0.08
4 – 5, s, P5 0.0008
B Modeling survival rate ( f ):
8. f(m1r)P(s)
9. f(m)P(s)
7. f(t1m1r)P(s)
10. f(r)P(s)
6. f(t1m1s1r)P(s)
4. f(t/r1s/m1m*r)P(s)
1112.07
1112.86 1114.07 1115.21 1115.40
1119.94
0.406
0.275 0.149 0.085 0.077 0.008
6
5 7 4 8 13
20.98
23.80 20.95 28.18 20.23 14.46
8 – 9, r, P5 0.09
7 – 8, t, P5 0.85
8 – 10, m, P5 0.03
6 – 7, s, P5 0.40
4 – 6, interactions, P5 0.33
C Checking delayed and advanced separately:
8. f(m1r)P(s)
12. f(d1r)P(s)
11. f(a1r)P(s)
1112.07
1113.09 1113.42
0.474
0.285 0.241
6
5 5
20.98
24.03 24.37
8 – 12, a 5 c, P 5 0.08
8 – 11, d 5 c, P 5 0.07
FIG 3 Mean local survival of advanced, control, and delayed adult
coots Survival estimates (with SE) are for female and male
com-bined and based on the best model (Table 4B, model 8).
that h (P 5 0.27) and h2 (P5 0.52) did not contribute
sig-nificantly to the survival probability Therefore, the natural
timing of breeding was not associated with adult survival and
the difference in original hatching date between control and
experimental pairs was apparently not responsible for their
different survival rates
The previous analyses included birds that were
manipu-lated in 1988 and thereafter used as controls or experimentals
Manipulation in 1988 may theoretically affect survival in
later years and the same may hold for individuals manipulated
both in 1989 and 1991 We therefore repeated the analyses
with individuals not manipulated previously (n 5 409, 53,
and 63 for control, delayed, and advanced birds, respectively,
as compared to 428, 59, and 68 birds in the main analysis)
The result was similar to that of the analyses previously
shown: Survival estimates for advanced, control, and delayed birds, respectively, were 0.40, 0.55, and 0.69 for 1989 and 0.47, 0.62, and 0.74 for 1992 This confirms our initial as-sumption that there is no indirect or chronic effect of ma-nipulation on survival
We also checked whether the difference in the initiation rate of second broods could explain the difference in survival probability between experimental groups We therefore mod-eled survival with the inclusion of an individual covariate
indicating the initiation of a second brood (br2) The
AIC-weights in this set of models f(r 1 m)P(s) and f(r 1 m 1
br2)P(s) were 0.62 and 0.25, respectively The LRTs for the
contribution of br2 was not significant (P 5 0.62) Thus,
differences in initiation rate of second broods did not explain the variation in survival among advanced, control, and de-layed birds
Reproduction in the Next Year
We investigated the effect of the experiment on laying date and size of the first clutch produced by females breeding in the next year To control for variation in breeding parameters between years, we calculated for each individual female the deviation in laying date and clutch size from the population mean for first clutches in each year Manipulation (delayed,
n 5 15; advanced, n 5 12) did not affect the change in relative
laying date (F2,98 5 0.39, P 5 0.68) or relative clutch size
(F2,985 0.51, P 5 0.48) between successive breeding seasons
as compared to control birds (n5 74) Thus, there were no
indications of long-lasting effects of date manipulations on reproductive performance of surviving females
DISCUSSION Life-history theory suggests that variation in the seasonal timing of reproduction within populations may be explained
Trang 9on the basis of trade-offs between fitness components (Daan
and Tinbergen 1997) Our experimental study provides strong
evidence for the existence of such trade-offs, because the
timing of current reproduction directly affected the future
reproductive output of adult coots The probability of starting
a second brood in the same breeding season declined
follow-ing an experimental delay of the hatchfollow-ing date of the first
brood, whereas an experimental advance raised the likelihood
of second broods The effect of the experiment was exactly
as predicted on the basis of the natural seasonal decline in
the probability of a second brood in the population, and we
therefore conclude that the result was consistent with the date
hypothesis Furthermore, the survival probability of adult
fe-males and fe-males with an experimentally advanced hatching
date of the first brood was significantly reduced compared to
that of control pairs, whereas delayed breeding pairs had a
significantly higher survival An experimental change in
tim-ing of breedtim-ing thus had oppostim-ing effects on second broods
and adult survival, the two major components of future
re-production in coots
Experimental Bias or Genuine Effects of Altered Timing?
The experimental design does not allow us to distinguish
between effects due to altered reproductive timing and
con-sequences of the prolonged or shortened incubation period
used to delay or advance the hatching date of the first brood,
respectively One might argue that delayed pairs are forced
to allocate more energy to incubation, whereas advanced pairs
enjoy reduced incubation costs, and that this affects their
future reproductive potential relative to that of control birds
Nevertheless, it is unlikely that manipulation of the length
of the incubation period severely affected the condition of
the adult coots and thereby determined the outcome of the
present experiment In European coots (Horsfall 1984) as well
as American coots (Ryan and Dinsmore 1979) male and
fe-male take about equal parts of the incubation load, leaving
each parent ample time to forage and to maintain energy
balance during the incubation (Alisauskas and Ankney 1985)
In Brunnich’s guillemots (Uria lomvia), another species with
shared incubation, body mass was not affected by an
exper-imentally prolonged incubation period (Gaston and Perin
1993)
The change in reproductive performance observed among
experimental pairs also argues against a major role for the
experimental bias First, it is unclear why the increased
num-ber of second broods by advanced pairs should be more
pro-nounced in the first than in the second half of the breeding
season (Fig 1D) In great tits (Parus major), with higher
initiation rates of second broods throughout the season, the
seasonal effect of delay on the incidence of second broods
similarly contradicted the uniform outcome predicted from
an experimental bias (Verboven and Verhulst 1996) Second,
local survival of experimentally delayed pairs, which
expe-rienced a longer incubation period, was higher than that of
experimentally advanced pairs It is important to note that
the difference in survival between control and experimental
pairs was not associated with the production of a second
brood Therefore, we cannot attribute the altered survival of
delayed or advanced pairs to a retarded effect of an
experi-mental bias on the production of second broods Thus, it is unlikely that the manipulation of the length of the incubation period played a role in the future reproductive performance
of coots
Link between Timing of the First Brood and the Probability
of a Second Brood
Consistent with the general pattern in birds (e.g., Smith et
al 1987; Geupel and DeSante 1990; Stouffer 1991; Hepp and Kennamer 1993; Verboven and Verhulst 1996), the incidence
of second broods in coots declined progressively with season
In addition, the probability of a second brood was negatively related to the number of first-brood young surviving the first two critical weeks after hatching A possible mechanism is indicated by the increase in the interbrood interval with size
of the first brood If the opportunities or benefits of raising
a second brood decline with the progress of season, whereas large first broods require a longer period of parental care, then pairs with many first-brood young might be selected to refrain from starting a second brood (Tinbergen and Van Balen 1988) This proposition is supported by experimental studies, which generally found a decline in the probability
of a second broods and an increase in interbrood interval following enlargement of the first brood, whereas the op-posite is found following brood reduction (Smith et al 1987; Tinbergen 1987; Linde´n 1988)
In the present study we manipulated the hatching date of the first brood and revealed a causal relationship between the timing of the first brood and the probability of a second brood (date hypothesis) Because we statistically controlled for the effect of the experiment on size of the first brood, this re-lationship most likely reflects a direct effect of altered timing
on the incidence of second broods Three experimental
stud-ies are available for comparison, all in Parus specstud-ies Nilsson (2000) found that female blue tits (P caeruleus) producing
their first clutch early in the season were most likely to initiate
a repeat clutch following removal of the first clutch Verhulst
et al (1995) also induced female great tits to produce a re-placement clutch and considered the rere-placement as a delayed brood None of the experimental pairs initiated an additional brood, in agreement with the natural seasonal decline in sec-ond broods in the population Verboven and Verhulst (1996) used a similar experimental approach as used in coots and obtained an identical result: As predicted on the basis of the natural seasonal variation, delay of the hatching date of the first brood caused a decline, whereas advance increased the incidence of second broods Thus, all four experimental stud-ies indicate that the seasonal decline in the probability of starting a second (or repeat) brood was causally related to the timing of the first brood
The seasonal decline in the probability of a second brood
is most likely related to the progressive decline in the avail-ability of insect food for raising young in the second half of the breeding season (Brinkhof 1997) and to the decline in the first-year survival of independent offspring with hatching date (Brinkhof et al 1997) Thus, the reproductive value of
a second brood in coots will decline with its initiation date Such a seasonal decline in brood reproductive value is com-mon in birds (Daan et al 1989; Rohwer 1992) Given that
Trang 10rearing a second brood may causally reduce reproductive
output of the female parent in subsequent years (Verhulst
1998), the seasonal decline in the likelihood of a second
brood most likely reflects a general decision rule that
opti-mizes the trade-off between successive attempts of future
reproduction (Verhulst 1998; Nilsson 2000)
Clutch Size of the Second Brood
Given the size of the first clutch and the laying date of the
second clutch, advanced females produced fewer eggs and
delayed females produced more eggs in the second clutch
than controls (Table 3B) Individual female European coots
differ consistently in laying date (Perdeck and Cave´ 1992)
These intrinsic quality differences also affect clutch size:
Consistently early-breeding females produced large clutches
for the laying date, whereas the consistently late-breeding
ones produced small clutches, independent of female age
(A.C Perdeck, unpubl data) Consistent individual
differ-ences in laying date and clutch size are common in birds
(reviewed by Klomp 1970; Findlay and Cooke 1983; Loman
1984; Hochachka 1993) In the European coot such
differ-ences are related to body size, with large females breeding
early and producing large first and second clutches
Adult Survival
Survival of adult coots of both sexes was independent of
timing of breeding An experimental delay of the hatching
date of the first brood by 10 days enhanced both female and
male survival In tits, effects of experimental delays on
sur-vival are either absent (Verhulst et al 1995) or inconsistent
between years (Nilsson and Svensson 1996), whereas
ex-perimental advances led to reduced female survival in blue
tits (Nilsson 1994)
In coots, most pairs for which hatching date was advanced
eventually raised more young, whereas fewer young were
raised following delay as compared to the number predicted
from original timing This is the combined result of the causal
effect of hatching date on the success of the first brood
(Brinkhof et al 1993) and the production of second broods
(present study) Brood care is provided by both sexes and
mainly involves food provisioning as well as brooding during
periods of rain or low ambient temperature (Brinkhof 1997)
The mortality of the parents is increased by advancing and
reduced by delaying the date, which suggests that the
ac-companying change in the number of young places a toll on
the parents Another possibility is that parental care is
pro-longed or shortened by advancing or delaying the date of
hatching, respectively One might surmise that both the
change in the number of young or in the length of the parental
care period result in a change in daily work rate, which could
explain the enhanced or reduced mortality following advance
or delay, respectively (Dijkstra et al 1990; Daan et al 1996)
However, we have no data to substantiate such a relationship
in the coot, which would require the direct comparison of
time and energy budgets during brood care between pairs
with natural and experimentally altered timing of breeding
The individual optimization hypothesis, which states that
variation in the timing of reproduction reflects
fitness-max-imizing strategies of individuals that differ in breeding
qual-ity, might give a comprehensive explanation for the observed natural and experimental variation in adult survival That advanced and delayed breeding resulted in reduced and en-hanced adult survival, respectively, whereas survival was in-dependent of season among controls, indicates that pairs with
a different laying date adjusted their timing to pay moderate survival-mediated costs to future reproduction Parents might thus maximize their lifetime fitness by trading off, with the advancement of laying date, the benefits of enhanced current reproduction (Brinkhof et al 1993, 1997) and future repro-duction on the basis of second broods (present study), against the reduction in future reproductive output Such a trade-off, based on a cost of early reproduction, might be especially important to fitness in long-lived species such as the Euro-pean coot To assess whether the variation in breeding date can be explained on the basis of individual optimization, the opposite effects of timing on different components of the current and future reproduction need to be integrated into one measure of fitness, for instance, using reproductive value (Daan et al 1990; Lessells 1991) The implementation of such integrative studies is essential to fully understand the evolution of breeding date in animal populations
ACKNOWLEDGMENTS
We thank F Hage, G Speek, J Visser, M Cave´, M Bal-lintijn, I Knevel, A Kooi, and E Troost for their help in the field Comments by B Crespi and two anonymous re-viewers greatly improved the manuscript
LITERATURE CITED Alatalo, R V., A Lundberg, and C Glynn 1986 Female pied flycatchers choose territory quality and not male characteristics Nature 323:152–153.
Alisauskas, R T., and C D Ankney 1985 Nutrient reserves and the energetics of reproduction in American coots Auk 102: 133–144.
Analytical Software 1992 Statistix Users manual Ver 4.0 An-alytical Software, St Paul, MN.
Bell, G 1980 The costs of reproduction and their consequences.
Am Nat 116:45–76.
Brinkhof, M W G 1997 Seasonal variation in food supply and
reproductive success in European coots Fulica atra Ardea 85:
51–65.
Brinkhof, M W G., and A J Cave´ 1997 Food supply and seasonal variation in breeding success: an experiment in the European coot Proc R Soc Lond B 264:291–296.
Brinkhof, M W G., A J Cave´, F J Hage, and S Verhulst 1993.
Timing of reproduction and fledging success in the coot Fulica
atra: evidence for a causal relationship J Anim Ecol 62:
577–587.
Brinkhof, M W G., A J Cave´, and A C Perdeck 1997 The seasonal decline in the first-year survival of juvenile coots: an experimental approach J Anim Ecol 66:73–82.
Brouwer, A., A L Spaans, and A A N De Wit 1995 Survival
of herring gull Larus argentatus chicks: an experimental analysis
of the need for early breeding Ibis 137:272–278.
Burnham, K P., D R Anderson, G C White, C Brownie, and K.
H Pollock 1987 Design and analysis methods for fish survival experiments based on release-recapture Am Fish Soc Monogr 5:1–437.
Cave´, A J 1968 The breeding of the kestrel, Falco tinnunculus
L., in the reclaimed area Oostelijk Flevoland Neth J Zool 18: 313–407.
Cave´, A J., and J Visser 1985 Winter severity and breeding bird numbers in a coot population Ardea 73:129–138.