Estimates of Natural Selection in a Salmon
Population in Captive and Natural Environments
MICHAEL J. FORD,∗§ JEFFREY J. HARD,∗BRANT BOELTS,† ERIC LAHOOD,∗
AND JASON MILLER∗
∗Conservation Biology Division, Northwest Fisheries Science Center, 2725 Montlake Boulevard E. Seattle,
WA 98112, U.S.A.
†Washington Department of Fish and Wildlife, 600 Capital Way N, Olympia, WA 98501-1091, U.S.A.
Abstract: Captive breeding is a commonly used strategy for species conservation. One risk of captive breeding
is domestication selection—selection for traits that are advantageous in captivity but deleterious in the wild.
Domestication selection is of particular concern for species that are bred in captivity for many generations
and that have a high potential to interbreed with wild populations. Domestication is understood conceptually
at a broad level, but relatively little is known about how natural selection differs empirically between wild
and captive environments. We used genetic parentage analysis to measure natural selection on time of
into captive and natural components. Our goal was to determine whether natural selection acting on the
traits we measured differed significantly between the captive and natural environments. For males, larger
individuals were favored in both the captive and natural environments in all years of the study, indicating
that selection on these traits in captivity was similar to that in the wild. For females, selection on weight was
significantly stronger in the natural environment than in the captive environment in 1 year and similar in
the 2 environments in 2 other years. In both environments, there was evidence of selection for later time
of return for both males and females. Selection on measured traits other than weight and run timing was
relatively weak. Our results are a concrete example of how estimates of natural selection during captivity can
be used to evaluate this common risk of captive breeding programs.
Keywords: artificial propagation, captive breeding, domestication selection, hatchery, Oncorhynchus kisutch,
Estimaciones de Selecci´ on Natural en una Poblaci´ on de Salm´ on en Ambientes Cautivos y Naturales
Resumen: La reproducci´ on en cautiverio es una estrategia usada com´ unmente para la conservaci´ on de
especies. Un riesgo de la reproducci´ on en cautiverio es la selecci´ on por domesticaci´ on—la selecci´ on de atrib-
utos que son ventajosos en cautiverio pero delet´ ereos en el medio silvestre. La selecci´ on por domesticaci´ on
es de inter´ es particular para especies que han sido reproducidas en cautiverio por muchas generaciones
y que tienen un alto potencial de entrecruzarse con poblaciones silvestres. La domesticaci´ on es entendida
conceptualmente a nivel general, pero se conoce poco sobre como difiere emp´ ıricamente la selecci´ on natural
entre ambientes silvestres y cautivos. Utilizamos an´ alisis parental gen´ etico para medir la selecci´ on natural
sobre el tiempo de migraci´ on, peso y morfolog´ ıa de una poblaci´ on de salm´ on Oncorhynchus kisutch que
fue subdividida en componentes cautivos y naturales. Nuestra meta fue determinar si la selecci´ on natu-
ral que act´ ua sobre los atributos que medimos difiere significativamente entre los ambientes cautivos y
naturales. Para machos, los individuos m´ as grandes fueron favorecidos tanto en los ambientes cautivos
como en los naturales, en todos los a˜ nos del estudio, lo que indica que la selecci´ on de estos atributos en
cautiverio era similar a la del medio silvestre. Para hembras, la selecci´ on de peso fue significativamente
mayor en el ambiente natural que en el cautivo en 1 a˜ no y similar en los 2 ambientes en los otros dos
a˜ nos. En ambos ambientes hubo evidencia de selecci´ on de un mayor tiempo de retorno tanto para machos
Paper submitted April 10, 2007; revised manuscript accepted December 17, 2007.
Conservation Biology, Volume 22, No. 3, 783–794
Journal compilation C ?2008 Society for Conservation Biology. No claim to original US government works.
Selection during Captivity
como hembras. La selecci´ on de los atributos medidos distintos al peso y el tiempo de corrida fue relativa-
mente d´ ebil. Nuestros resultados son un ejemplo concreto de c´ omo las estimaciones de selecci´ on natural
durante el cautiverio pueden ser utilizadas para evaluar este riesgo com´ un en los programas de reproducci´ on
Palabras Clave: an´ alisis parental, criadero de peces, propagaci´ on artificial, Oncorhynchus kisutch, repro-
ducci´ on en cautiverio, selecci´ on por domesticaci´ on
Understanding how populations evolve is becoming
an increasingly important part of conservation biology.
There are several reasons for this. Global warming and
associated climate change have brought attention to the
fact that many species live in environments that are
changes is an important factor in the extinction risk of
that species (e.g., Lynch & Lande 1993; Lande & Shannon
1996; Etterson & Shaw 2001). In addition to global warm-
ing, most species face a myriad of environmental changes
that alter the selective landscape they experience, such
as land-use changes and invasions by non-native species.
Another important form of environmental change occurs
during captive breeding. Many species of conservation
concern are propagated artificially in an attempt to in-
crease their abundance. Most captive breeding programs
are not intended to be permanent substitutions for viable
natural populations, however, and seek to minimize the
amount of evolution that occurs in captivity. Whether in-
tentional or not, selection for traits that are advantageous
or neutral in captivity but disadvantageous in the wild
(domestication selection) is expected to be a particular
problem if a species is propagated for many generations
Bryant & Reed 1999).
Seven species of Pacific salmon (Oncorhynchus spp.)
spawn in rivers on the west coast of North America,
and all contain populations that have declined to the
point of conservation concern (Nehlsen et al. 1991).
Twenty-seven evolutionarily significant units (ESUs)
of Pacific salmon have been listed as threatened or
endangered under the U.S. Endangered Species Act
(http://www.nwr.noaa.gov/). Salmon populations have
declined for several reasons common to other threat-
ened species, including loss and degradation of habitat
and overharvest (National Research Council 1996). An-
other potential factor for the decline of wild salmon pop-
ulations is the genetic and ecological effects of artificial
propagation. In much of their range, salmonids are ar-
tificially propagated on a very large scale. For example,
over 4 billion anadromous juvenile salmon are released
annually into the North Pacific Ocean from hatcheries
in North America and Asia (Beamish et al. 1997). Similar
hatchery programs and large-scale closed-pen fish farm-
ing operations exist for Atlantic salmon (Salmo salar) in
the North Atlantic Ocean and Baltic Sea.
Artificial propagation poses both ecological and ge-
netic risks to salmonids (reviewed by National Research
Council 1996; Waples & Drake 2004). For example, the
potential effects of supportive breeding on effective pop-
ulation size have been modeled extensively (Ryman &
Laikre 1991; Ryman et al. 1995; Wang & Ryman 2001).
Domestication selection may be another particularly im-
portant genetic risk because salmonids are often prop-
agated artificially for many generations. Several investi-
gators have found evidence for domestication in salmon
(e.g., Reisenbichler & McIntyre 1977; Fleming & Gross
1993; Araki et al. 2007a, 2007b), and evolutionary mod-
els suggest that domestication has the potential to cause
declines in the fitness of natural populations that are
partially supported by artificial propagation (e.g., Lynch
& O’Hely 2001; Ford 2002). The predictions of these
models are quite sensitive to their underlying parame-
ters, however, including the strength and form of natural
selection that occurs in the captive and wild environ-
ments. Obtaining an empirical understanding of how se-
lection differs between captive and natural environments
is therefore important for predicting the outcomes of
long-term supplementation programs (Hershberger et al.
1990; Ford 2002; Araki et al. 2007a).
We quantified differences in how selection operates in
the wild and in captivity for a salmon population with
components that experienced either a wild or a cap-
tive environment. Using genetic parentage assignment
methods, we estimated the number of progeny produced
by individual fish in either of 2 distinct environments, a
salmon hatchery or a natural stream. Between the 2 envi-
ronments, we then compared selection gradients (regres-
sions of relative progeny number upon standardized trait
value) for several traits, including time of return to the
stream, weight, and body shape. We were particularly in-
fered in wild and captive environments or whether selec-
tion in captivity was “relaxed” (weaker) compared with
what we observed in the wild. The natural population
we studied had already experienced many generations
of artificial propagation, so the situation we analyzed is
probably more typical of a captive population readapting
to the wild environment than a wild population adapting
to a captive environment.
Volume 22, No. 3, 2008
Ford et al.
Minter Creek is a small stream draining into Henderson
Bay in Puget Sound, Washington (U.S.A.; Fig. 1). Coho
salmon in Minter Creek have a life-history pattern typi-
cal for the species at this stream’s latitude (Sandercock
1991), with juveniles spending about 18 months in the
stream prior to smoltification and migration to salt water.
The fish then typically spend another 18 months at sea,
returning to freshwater as 3-year-olds. In some years a siz-
able fraction of the males return as 2-year-olds (“jacks”)
after approximately 6 months at sea (Salo & Bayliff 1958;
Sandercock 1991), and other male and female life-history
patterns also occasionally occur.
A hatchery operated by the Washington Department
of Fish and Wildlife (WDFW) is located near the mouth
of the creek. Coho salmon have been bred and released
from the hatchery since 1938, when WDFW selected the
stream as a study site representative of the hundreds of
small Pacific Northwest streams inhabited by coho (On-
corhynchus kisutch) and chum (O. keta) salmon (Salo &
Figure 1. Map of the Minter Creek, Washington, study
area showing the location of the Minter Creek
Hatchery (including the adult sampling weir) and the
juvenile (smolt) collections weirs.
Bayliff 1958). Production-scale (>1,000,000 smolts/year)
coho salmon releases were initiated in 1960 (WDFW,
Olympia, Washington). There have been no sustained
attempts to keep the hatchery and natural populations
isolated from each other, and in the years leading up to
our study an unquantified (probably >90%) fraction of
the natural spawning population consisted of hatchery-
produced fish. The natural population of coho salmon
in Minter Creek is therefore clearly not a “pristine” wild
population; rather, it is representative of many of the
natural salmon populations in the Pacific Northwest that
have experienced high levels of hatchery stocking for
many decades. In a previous study (Ford et al. 2006), we
found no significant differences in fitness between hatch-
ery and naturally produced fish in Minter Creek. There-
fore, hatchery and natural-origin fish were combined in
the analyses reported here.
In 2000 through 2004 all coho salmon were inter-
cepted at a weir located at the head end of tidewater
(river kilometer 0.8) as they returned from the ocean to
spawn, and a fraction were detained for sampling. Each
sampled fish was sexed, weighed, and photographed,
and each fish had a small piece (<1 cm2) of caudal fin
tissue removed for subsequent DNA analysis. Fish were
allowed to recover in fresh flowing water before either
being released upstream (all years) or into the hatchery
holding pond (2001 through 2003).
In addition to our study fish, each year several thou-
sand hatchery-origin fish were passed directly into the
hatchery holding pond without being sampled, but no
unsampled fish were allowed to pass above the weir to
spawn naturally. Wild fish were identified by the pres-
ence of an adipose fin (hatchery coho salmon released
in the Puget Sound area typically have their adipose fin
removed prior to release) and by scale analysis (see Ford
et al. 2006 for details). In 2001 between 26 September
and 31 October, every third sampled fish (whether of
wild or hatchery origin) was released into the hatchery
holding pond. Prior to 26 September and after 31 Oc-
tober, all but 10 fish were passed upstream so that fish
management spawning goals for wild fish would be met.
The mean run timing (day of return to the weir measured
as days after September 1) of males put into the pond was
a few days earlier than the mean run timing of the males
put into the stream (day 45.8 [SD 15.03] vs. day 48.3
[18.8], p = 0.08, t test). The mean run timing of females
put into the pond was approximately 5 days earlier than
the mean of the females put into the stream (day 45.1 [SD
13.6] vs. 51.6 [17.7], p < 0.001, t test). All sampled fish
released into the holding pond were tagged with num-
bered metal jaw tags so that they could be identified at
the time of spawning. We followed a similar protocol for
directing fish into the stream and hatchery pond in 2002
and 2003, but these fish were not included as part of
the parentage analyses because their progeny were not
Volume 22, No. 3, 2008
Selection during Captivity
on 6 days in November (5, 7, 14, 19, 20, and 27). Spawn-
ingdatesweredeterminedby the hatchery staffandwere
the same as for the general (nonexperimental) hatchery
population. We spawned the fish by randomly select-
ing sexually mature, tagged males and females from the
hatchery pond. Each female was spawned with one male,
and >95% of the males were spawned with a single fe-
male. We kept progeny from the sampled fish spawned
in the hatchery separate from the general hatchery pop-
ulations until they were large enough to be sampled and
tagged in summer of 2002. At that time all progeny from
sampled fish received a coded wire tag (CWT) implanted
into their snout so that they could be identified as our
study fish when they returned to the stream as adults in
Wild juvenile progeny were sampled at the fry (first
summer after emergence) and smolt (second spring/
summer after emergence) stages between March 2001
and June 2003 as described in Ford et al. (2006). Fry were
tershed, and smolts were sampled at weir traps through-
out their outmigration period. These juvenile samples
were a small fraction (1–5%) of the total juvenile pop-
ulation. Adult progeny from the 2000 and 2001 broods
were sampled at the weir in the fall of 2003 and 2004,
respectively. Progeny were assigned to parents with the
maximum likelihood method implemented in the pro-
gram FAMOZ (Gerber et al. 2003). See Ford et al. (2006)
for a detailed description of the parentage analysis.
We measured 6 morphological traits (Fig. 2) from the dig-
ital images taken at the time of sampling. Each trait was
digitized and measured to the nearest millimeter with
the computer program TPSDIG (Rohlf 2001). Weights
were converted to the linear scale of the other traits by
cube-root transformation. To reduce allometric correla-
tions amongtraits, weregressedeachmorphological trait
other than weight on transformed weight with model I
Figure 2. Morphological traits that were studied for
their effects on relative fitness of coho salmon
spawning in Minter Creek. Fish image source:
National Oceanic and Atmospheric Administration
(NOAA) Photo Library.
as the trait values in further analyses. Some fish had miss-
ing data because of occasional failures of the scale or
camera, especially for the 2000 sampling season. Time
of return was essentially independent of weight (correla-
tion coefficients ranged from −0.11 to 0.14, depending
on the sex and cohort), so time of return was analyzed
Selection and Fitness Analyses
All statistical analyses were conducted with the general
linear model (GLM) function in the SYSTAT computer
package (version 11, Systat Software, San Jose, Califor-
nia). Traitswere standardized withineachsexandcohort
by subtracting the mean and dividing by the standard
deviation. We made separate fitness estimates by count-
ing progeny at fry, smolt, and adult life stages. Absolute
fitness (progeny counts) within sexes, cohorts, and lo-
cations (stream vs. hatchery) was converted to relative
fitness by dividing by the mean fitness. In addition, we
estimated the minimum number of unique mates an in-
dividual had by summing the other unique parents of
the progeny assigned to that individual regardless of life
stage. We estimated standardized linear and quadratic se-
lection gradients (Lande & Arnold 1983) individually for
each trait with the following model: progeny = constant
+ trait + (trait)2. We tested differences in linear selection
terms between the hatchery and natural environments
with analysis of covariance with the model: progeny =
constant + trait + location ∗ trait. No attempt was made
to correct p values for multiple tests, but results were
interpreted with an appreciation that a large number of
tests were conducted. We focused primarily on results
that were both highly significant at the level of individ-
ual tests and that followed a consistent pattern for the
different life stages sampled.
As an additional measure of selection we compared
the mean trait values of the group of parents inferred to
have produced offspring versus those inferred to have
produced no sampled offspring. We calculated selection
differentials as the difference in trait means of the sam-
pled population and the subset of individuals that were
inferred to produce offspring. For fish in the hatchery
pond, we also estimated selection differentials indepen-
of spawned fish as recorded in the hatchery records ver-
sus all tagged fish put into the hatchery pond.
All progeny assignments were the same as described in
Ford et al. (2006), with the exception of the adult off-
which were conducted with the same methods but were
Volume 22, No. 3, 2008
Ford et al.
Table 1. Summary of parent and progeny sample sizes for genetic parentage analysis of Minter Creek coho salmon in stream and hatchery
Brood year, locationfemalesmales (jacks)a
2001, hatchery pond
aMales includes all males (age 2 and 3 years). Numbers in parentheses are the total number of males that were 2 years old (jacks).
bIncludes 901 3-year-old fish sampled in 2003 and 23 2-year-old fish sampled in 2002.
cIncludes 1657 3-year-old fish sampled in 2004 and 26 2-year-old fish sampled in 2003.
not reported in the previous paper. Sample sizes for both
parents and progeny are in Table 1. For the 2000 brood
95.4% of the potential parents sampled at the weir were
genotyped for parentage analyses. For the 2001 brood
92.6% of the potential parents in the stream and 98.1% of
the potential parents in the hatchery were genotyped for
parentage analysis. The majority of the potential spawn-
ers in each of the years were 3-year-old fish, with 2-year-
olds making up only 6% of the males in 2000 and 1.4%
of the males in 2001 (Table 1). Similarly, in 2002, 2003,
and 2004, when offspring from the 2 parental years in the
study returned to spawn, >97% of the returning adults
were 3-year-old fish. Depending on the life stage and year
sampled, 50–70% of the progeny were unambiguously
assigned to a single pair of spawning parents.
In the stream environment significant selection on
weight (size) was detected for both sexes and in both
years (Table 2). Linear selection gradients for weight
were positive, indicating that for both males and fe-
males larger individuals were more successful at produc-
ing offspring (Fig. 3). For both sexes selection gradients
were generally similar among the different life stages of
progeny that we assessed, although the smolt-stage as-
sessments produced the largest selection gradients (Ta-
ble 2). Selection on weight was similar between years.
After taking into account the effects of size, there was no
significant effect of male age on the number of offspring
produced (analysis of covariance, results not shown). Sig-
nificant selection was also detected on run timing, as
was reported previously by Ford et al. (2006). The lin-
ear terms were generally positive (indicating selection
for later than average return time), and quadratic terms
were generally negative, indicating stabilizing selection
against returning at the extremes of the run (Table 2).
The absolute values of the selection gradients on run tim-
ing differed slightly from the earlier study because in this
study fitness was standardized by dividing by the mean
fitness (Lande & Arnold 1983), whereas the earlier es-
timates were standardized by subtracting the mean and
dividing by the standard deviation. For males significant
in the 2000 cohort but not the 2001 cohort. In 2001 sig-
nificant quadratic selection was detected for males on
fin size, body depth, snout length, and caudal peduncle
size, but these results depended on a single outlying indi-
vidual with high relative fitness that was unusually light
for its size (data not shown). In general there were no
clear patterns of selection on traits other than weight and
run timing after correcting for correlations with weight
For males selection in the hatchery pond was similar to
that observed in the stream. In particular, linear selection
gradients for weight were positive and highly significant
(Table 2 & Fig. 3). Selection on weight in the hatchery
pond was significantly stronger than selection on weight
tinguishable from selection in the stream when progeny
were counted at the smolt or adult life stages (Table 2).
Selection on run timing was also positive and significant
and not statistically distinguishable from selection on run
ronment, later-returning males produced more progeny
than earlier-returning males (Table 2). In the pond, un-
like the stream, significant selection was also detected on
dorsal fin size, when selection was measured as number
of mates or progeny at the fry stage (Table 2).
For females in the hatchery pond, selection on weight
in 2001 was weaker than was observed in the stream.
In particular, linear selection gradients on weight were
not significantly greater than zero and were significantly
less than the selection gradients estimated for females
in the stream environment for 2 of the 3 measures of
fitness (Table 2). Selection on run timing was similar to
that observed in the stream, with later-returning females
producing more progeny than earlier-returning ones, al-
though the p values were generally less significant than
they were for the stream coefficients (Table 2).
The selection differentials, which were based only
on whether or not an individual was inferred to have
tion gradients, which incorporated both spawning suc-
weight were significant in both environments (Table 3).
For females the selection differential for weight in 2001
was significant in the stream environment but not in
the pond environment, which was consistent with the
Volume 22, No. 3, 2008
Selection during Captivity
Table 2. Linear (B) and quadratic (G) regression coefficients describing the relationship between relative fitness (number of inferred mates and number of sampled fry, smolt, and adult progeny
per individual standardized by the mean number of progeny per individual) and the traits that were measured on coho salmon sampled at the Minter Creek weir.
Volume 22, No. 3, 2008
Ford et al.
Table 2. continued.
aLocation refers to whether the sampled fish were allowed to spawn naturally in the stream (steam) or were put into the hatchery pond (pond). The B and G terms refer to the linear and
quadratic coefficients of the regression, respectively: relative progeny = constant + B (trait) + G (trait)2. Significance tests for the regression coefficients (p values, uncorrected for multiple
tests) are in parentheses.
bLife stage refers to the age at which progeny were sampled. Progeny were sampled and assigned to parents at the fry (< 1 year old), smolt (approximately 1.5 years old), and adult
(primarily 3 years old) life stages. Mates refers to the minimum estimate of the number of mates on the basis of all the sampled progeny.
cSignificance of interaction terms for year and location, respectively, by each measured trait, for linear terms only.
Volume 22, No. 3, 2008
Selection during Captivity
Figure 3. Standardized selection gradients
(regressions of relative fitness on standardized trait
value) for weight for male and female coho salmon
spawning in the natural environment (stream) or
artificially propagated in the hatchery environment
(pond) in 2001. Weight was standardized to a mean
of 0 and an SD of 1.
selection gradient for the same trait. Selection differen-
selection gradients and indicated that (with the excep-
tion of males in the stream in 2000) successfully breed-
ing individuals in either environment had average return
times from 3 to 9 days later than unsuccessful individuals
and females (Table 3). For fish put into the hatchery
pond in 2002 and 2003, selection differentials estimated
from hatchery records were similar to those estimated in
2001for malesbut were differentthan the 2001selection
differentials for females. In particular, unlike in 2001,
were either significant or nearly so, and were similar to
the selection differentials for female weight in the stream
in 2000 or 2001 (Table 3).
This study is, to our knowledge, the first to measure di-
rectly the strength of selection in both the captive and
natural components of an artificially supplemented pop-
ulation. Domestication selection has long been a concern
for captive breeding programs (Conway 1980) and is of
particular concern for salmon conservation efforts be-
cause hatchery production occurs on such a large scale
(Beamish et al. 1997; Waples & Drake 2004). Numer-
ous studies have documented behavioral, morphological,
or physiological differences between hatchery produced
and naturally produced salmon (e.g., Fleming & Gross
1989; Swain & Riddell 1990; Marchetti & Nevitt 2003;
Wessel et al. 2006), and several models have been devel-
oped that explore how selection is expected to operate
in a captive/natural system (e.g., Lynch & O’Hely 2001;
Tufto 2001; Ford 2002). The degree to which selection
acts differently in captive and natural environments is
often a critical parameter in these models. Therefore,
obtaining estimates of the relative strength of selection
for the same population inhabiting captive and natural
environments is important for ongoing efforts to make
hatchery programs less harmful to wild salmon popula-
tions (Mobrand et al. 2005).
The most striking aspect of our results was how sim-
ilarly selection operated in the 2 environments on the
traits we measured. For males in particular, there was
strong selection for larger size in both the natural stream
environment and the hatchery pond (Tables 2 & 3; Fig.
3). Sexual selection on male size has been observed pre-
viously for salmon (e.g., Fleming & Gross 1994; Quinn &
Foote 1994; Quinn & Buck 2001), although several stud-
ies in natural streams have also failed to find selection
on male size (e.g., Garant et al. 2001; Dickerson et al.
2005). In our study males that successfully mated and
produced sampled offspring in the stream were on av-
erage 20–24% heavier than those that did not, and the
selection gradient on weight in the stream ranged from
0.26 to 0.75, depending on the progeny stage sampled
(Table 2). Surprisingly, selection on males for larger size
was similar in the hatchery pond to what was observed
in the stream, with spawned males on average 38% heav-
ier than unspawned males and with selection gradients
of 0.27–0.89 (Table 2). On the basis of survival data (Ta-
ble 3), it was clear that this selection was largely due to
size-selective survival that occurred in the hatchery pond
prior to spawning.
In contrast to selection on males, size selection for
females differed markedly between the hatchery and nat-
ural environments for the 1 year (2001) in which we esti-
observed strong selection for larger-sized females, similar
to what we observed for males (Table 2 & Fig. 3). Com-
pared with males, however, size selection on females in
the hatchery in 2001 was significantly weaker than in the
stream, with selection gradients in the stream ranging
from 0.5 to 1.2, depending on the life stage of progeny
sampled, and selection gradients in the hatchery ranging
from 0.04 to 0.3 (Table 2). The selection differentials in
2001 in the hatchery for female weight calculated either
from the parentage analyses or from the mating records
Volume 22, No. 3, 2008
Ford et al.
Table 3. Inferred and estimated mortality rates and selection differentials of coho salmon for weight and run timing in the stream and hatchery pond environments.
Mean run timing,
Mean weight, g (SD)
days after 1 Sep. (SD)
aLocation refers to whether the sampled fish were allowed to spawn naturally in Minter Creek (stream) or were put into the hatchery pond (pond).
bMethod refers to whether membership of the spawned versus unspawned groups was inferred on the basis of results of the genetic parentage analyses (genetic) or from the hatchery mating
cMortality refers to the estimated prespawning mortality rate, calculated as the number of individuals known or inferred to have spawned divided by the total number of individuals in the
sample group (row). The prespawning mortality rates estimated from the genetic parentage results are maximum estimates because it is likely that some individuals in fact spawned but by
chance did not produce sampled progeny.
dStandardized selection differential (standardized difference between the total population trait mean and the trait mean of the individuals known or inferred to have spawned). The p
values were calculated from a 2-sample t test that compared the difference between the spawned and unspawned groups.
Volume 22, No. 3, 2008
Selection during Captivity
were also much smaller than the selection differentials
estimated for females in the stream (Table 3). These re-
sults suggest that selection for female size was relaxed
in the hatchery in 2001 compared with selection in the
time, however, and in 2002 and 2003 selection on female
weight in the hatchery was similar to that in the stream
in 2001 (Table 3).
Conceptually there could be several mechanisms lead-
ing to selection for larger size in the hatchery. Large size
might be related to good general health, for example.
Interactions among fish in the hatchery pond could also
lead to size selection. For example, the fish, particularly
males, in the hatchery pond may interact and fight with
each other in a manner similar to what they do in the
stream, resulting in selection for larger size in the hatch-
for large size when choosing fish to spawn.
The similar strength of selection in the 2 environments
in these years does not imply that the selective mecha-
nisms were necessarily the same in the 2 environments,
however. For example, selection for larger female size in
the wild has been related to a variety of behavioral fac-
tors that would have no opportunity to be expressed in
the Minter Creek Hatchery, such as ability to dig deeper
nests in the gravel, defend superior spawning locations,
Fleming & Gross 1994; Quinn et al. 2001). Although we
have clearly documented that selection on size occurs
in both environments, we have no direct data on the
selective mechanism in either environment.
Natural selection for run timing was qualitatively simi-
lar for both sexes in both environments, although the sta-
tistical significance varied considerably among compar-
isons. In both environments there was some evidence for
selection for later than average time of return, with suc-
cessfully breeding individuals returning anywhere from
several days to about a week later than unsuccessful in-
dividuals (Tables 2 & 3). These results should be inter-
preted in light of the long-term changes that have been
observed in the run timing of Minter Creek coho salmon
(Ford et al. 2006). Both hatchery and natural-origin fish
are returning to Minter Creek more than a month ear-
lier on average than they did at the time when records
were first kept in the 1940s (Salo & Bayliff 1958). The
likely cause of this trend toward earlier time of return is
previous selection for earlier-spawning fish in the Minter
We previously estimated selection gradients on run
timing for those salmon that spawned in the stream (Ford
et al. 2006) and hypothesized that the selection for later
run time indicated that the current, relatively early, run
time was evidence of hatchery-induced selection. It was,
therefore, initially surprising to find that selection for
later return time was just as strong in the hatchery as
it was in the wild. Nevertheless, for the last decade,
hatcheries throughout Washington have been subject to
review and reform efforts designed to make them more
2005). As a result, at the time we initiated our study, the
Minter Creek Hatchery began to deliberately reverse its
previous spawning protocols in favor of protocols that
would avoid selection for early-returning spawners (D.
Popochock, personal communication). The finding of se-
lection for later return time therefore appears consistent
with the current spawning protocols employed by the
hatchery. In addition, the high level of prespawning mor-
contributed to selection for later return time simply be-
cause few early-returning fish survived long enough to be
Consistent with an earlier study on coho salmon (Flem-
ing & Gross 1994), we found little evidence for strong
selection on the other morphological traits we measured
after correcting for overall size (Table 2). An important
only be estimated on the relatively small number of traits
we measured. Other traits, including behavior and phys-
iology, were likely to have been under selection in both
captive and natural environments, but would have gone
undetected in our study. In addition, in our study all traits
were measured at the weir, often on fish that had not yet
become fully sexually mature. Many of the morpholog-
ical traits we measured vary depending on the state of
sexual maturation (Sandercock 1991), and the measure-
ments made at the weir likely differ somewhat from what
they were at the time that selection actually occurred in
The Minter Creek coho salmon population has some
characteristics that should be taken into account when
considering the patterns of selection we observed. For
example, the low proportion of jacks during the years of
our study may explain why we found no evidence of dis-
previous studies (e.g., Gross 1985). The approximately
50% mortality rate of our study fish in the hatchery was
also much higher than the 1–10% that is more typical for
a coho salmon hatchery (D. Popochock, personal com-
munication), and the patterns of selection we observed
in the hatchery may therefore not be typical of hatchery
populations generally. In particular, relaxation of selec-
tion, at least on traits associated with breeding success,
may be more typical of hatcheries with very high pre-
spawning survival rates. The selection we measured in
the stream setting may also be different from what might
have been observed had the population not already been
affected by several decades of hatchery stocking.
Despite these caveats, our results provide an impor-
tant glimpse into how natural selection operates in cap-
tive and natural environments. Contrary to our initial
expectations, we found that patterns of selection were
similar between the hatchery and stream environments,
Volume 22, No. 3, 2008
Ford et al.
especially for males. The observation of similar outcomes
of selection in the 2 environments suggests that selection
on these traits in the hatchery would not be expected to
create a large selective load that contributes to poor pop-
ulation fitness in the natural environment. The case for
females is somewhat different, with evidence for relaxed
selection on weight in the hatchery compared with the
natural stream, at least for 1 of the 3 years studied. Differ-
ences in selection regimes between captive and natural
environments can lead to reduced fitness in the natural
1999; Lynch & O’Hely 2001; Ford 2002). It is not clear,
however, whether the degree of relaxation we observed
for selection on female weight in captivity will produce
a significant selective load on the natural population in
Minter Creek. Addressing the fitness consequences of
the selection we observed will require additional anal-
yses that take into account factors such as the genetic
basis of the trait and the rates of gene flow between the
captive and natural components of the population.
We particularly acknowledge the Minter Creek hatchery
crew, D. Popochock, R. Henderson, E. Kinney, S. Davis,
D. Watkins, and W. Campbell, whose kind assistance, pa-
tience, and unfailing good humor were essential to the
success of this project. R. Waples, H. Araki, I. Fleming,
M. Chilcote, P. Hulett, and L. Laikre provided useful com-
ments on an earlier version of this paper. This project
was funded in large part by a grant from the Hatchery
Scientific Review Group, whose assistance we gratefully
Araki, H., W. Ardren, E. Olsen, B. Cooper, and M. Blouin. 2007a. Re-
productive success of captive-bred steelhead trout in the wild: eval-
uation of three hatchery programs in the Hood River. Conservation
Araki, H., B. Cooper, and M. S. Blouin. 2007b. Genetic effects of cap-
tive breeding cause a rapid, cumulative fitness decline in the wild.
Beamish, R. J., C. Mahnken, and C. M. Neville. 1997. Hatchery and
wild production of Pacific salmon in relation to large-scale, natural
shifts in the productivity of the marine environment. ICES Journal
of Marine Science 54:963–964.
Bryant, E., and D. Reed. 1999. Fitness decline under relaxed selection
in captive populations. Conservation Biology 13:665–669.
Conway, W. G. 1980. An overview of captive propagation. Pages 199–
208 in M. E. Soul´ e and B. A. Wilcox, editors. Conservation biology:
an evolutionary-ecological perspective. Sinauer Associates, Sunder-
Dickerson, B., K. Brinck, M. Willson, P. Bentzen, and T. Quinn. 2005.
grounds to reproductive success. Ecology 86:347–353.
Etterson, J., and R. Shaw. 2001. Constraint to adaptive evolution in
response to global warming. Science 294:151–154.
Fleming, I. A., and M. R. Gross. 1989. Evolution of adult female life
history and morphology in a Pacific salmon (coho: Oncorhynchus
kisutch). Evolution 43:141–157.
Fleming, I. A., and M. R. Gross. 1993. Breeding success of hatchery and
wild coho salmon (Oncorhynchus kisutch) in competition. Ecolog-
ical Applications 3:230–245.
Fleming, I. A., and M. R. Gross. 1994. Breeding competition in a Pacific
Salmon (coho: Oncorhynchus kisutch): measures of natural and
sexual selection. Evolution 48:637–657.
Ford, M., H. Fuss, B. Boelts, E. LaHood, J. Hard, and J. Miller. 2006.
Changes in run timing and natural smolt production in a naturally
spawning coho salmon (Oncorhynchus kisutch) population after
60 years of intensive hatchery supplementation. Canadian Journal
of Fisheries and Aquatic Sciences 63:2343–2355.
Ford, M. J. 2002. Selection in captivity during supportive breeding may
reduce fitness in the wild. Conservation Biology 16:815–825.
Frankham, R. H., H O. Ryder, E. Cothran, M. Soul´ e, N. Murray, and M.
Snyder. 1986. Selection in captive populations. Zoo Biology 5:127–
mating system and determinants of individual reproductive success
Gerber, S., P. Chabrier, and A. Kremer. 2003. FaMoz: a software for
parentage analysis using dominant, codominant and uniparentally
inherited markers. Molecular Ecology Notes 3:479–481.
Gross, M. R. 1985. Disruptive selection for alternative life histories in
salmon. Nature 313:47–48.
Hershberger, W., J. Myers, R. Iwamoto, W. Mcauley, and A. Sax-
ton. 1990. Genetic changes in the growth of coho salmon (On-
corhynchus kisutch) in marine net-pens, produced by 10 years of
selection. Aquaculture 85:187–197.
Lande, R., and S. J. Arnold. 1983. The measurement of selection on
correlated characters. Evolution 37:1210–1226.
Lande, R., and S. Shannon. 1996. The role of genetic variation in adap-
tation and population persistence in a changing environment. Evo-
Lynch, M., and R. Lande. 1993. Evolution and extinction in response
to environmental change. Sinauer Associates, Sunderland, Mas-
of natural populations. Conservation Genetics 2:363–378.
Marchetti, M., and G. Nevitt. 2003. Effects of hatchery rearing on brain
structures of rainbow trout, Oncorhynchus mykiss. Environmental
Biology of Fishes 66:9–14.
Mobrand, L. E., J. Barr, L. Blankenship, D. E. Campton, T. T. P. Evelyn,
T. A. Flagg, C. V. W. Mahnken, L. W. Seeb, P. R. Seidel, and W. W.
Smoker. 2005. Hatchery reform in Washington State: principles and
emerging issues. Fisheries 30:11–33.
National Research Council. 1996. Upstream: salmon and society in the
Pacific Northwest. National Academy Press, Washington, D.C.
Nehlsen, W., J. E. Williams, and J. A. Lichatowich. 1991. Pacific salmon
at the crossroads: stocks at risk from California, Oregon, Idaho, and
Washington. Fisheries 16:4–21.
Quinn, T. P., and G. B. Buck. 2001. Size- and sex-selective mortality of
adult sockeye salmon: bears, gulls, and fish out of water. Transac-
tions of the American Fisheries Society 130:995–1005.
Quinn, T. P., and C. J. Foote. 1994. The effects of body size
and sexual dimorphism on the reproductive behavior of sock-
eye salmon, Oncorhynchus nerka. Animal Behavior 48:751–
Quinn, T. P., A. P. Hendry, and G. B. Buck. 2001. Balancing natural and
sexual selection in sockeye salmon: interactions between body size,
reproductive opportunity and vulnerability to predation by bears.
Evolutionary Ecology Research 3:917–937.
Reisenbichler, R. R., and J. D. McIntyre. 1977. Genetic differences in
growth and survival of juvenile hatchery and wild steelhead trout,
Volume 22, No. 3, 2008
794 Download full-text
Selection during Captivity
Salmo gairdneri. Journal of the Fisheries Research Board of Canada
Rohlf, F. J. 2001. TPSDIG Program for digitizing images for analysis by
thin plate splines. State University of New York, Stony Brook, New
York. Available from http://life.bio.sunysb.edu/morph/ (accessed
Ryman, N., and L. Laikre. 1991. Effects of supportive breeding on the
genetically effective population size. Conservation Biology 5:325–
Ryman, N., P. E. Jorde, and L. Laikre. 1995. Supportive breeding and
variance effective population size. Conservation Biology 9:1619–
Salo, E. O., and W. H. Bayliff. 1958. Artificial and natural production of
Washington Department of Fisheries, Olympia.
Sandercock, F. K. 1991. Life history of coho salmon (Oncorhynchus
kisutch). Pages 395–446 in C. Groot and L. Margolis, editors. Pa-
cific salmon life histories. University of British Columbia Press,
Snyder, N., S. Derrickson, S. Beissinger, J. Wiley, T. Smith, W. Toone,
and B. Miller. 1996. Limitations of captive breeding in endangered
species recovery. Conservation Biology 10:338–348.
Swain, D. P., and B. E. Riddell. 1990. Variation in agonistic behavior
between newly emerged juveniles from hatchery and wild popula-
tions of coho salmon, Oncorhynchus kisutch. Canadian Journal of
Fisheries and Aquatic Sciences 47:566–571.
Tufto, J. 2001. Effects of releasing maladapted individuals: a
demographic-evolutionary model. The American Naturalist 158:
Van Den Berghe, E. P., and M. R. Gross. 1989. Natural selection result-
ing from female breeding competition in a Pacific salmon (coho:
Oncorhynchus kisutch). Evolution 43:125–140.
Wang, J., and N. Ryman. 2001. Genetic effects of multiple gen-
erations of supportive breeding. Conservation Biology 16:1619–
stock enhancement: a Pacific salmon perspective. Pages 260–306
in L. M. Leber, S. Kitada, T. Svasand, and H. L. Blankenship, editors.
enhancement. Blackwell Publishing, Oxford, United Kingdom.
Wessel, M., W. Smoker, R. Fagen, and J. Joyce. 2006. Variation of ag-
onistic behavior among juvenile Chinook salmon (Oncorhynchus
tshawytscha) of hatchery, hybrid, and wild origin. Canadian Jour-
nal of Fisheries and Aquatic Sciences 63:438–447.
Volume 22, No. 3, 2008