Mixed-source reintroductions lead to outbreeding
depression in second-generation descendents of a
native North American fish
DAVID D. HUFF,* LOREN M. MILLER,* CHRISTOPHER J. CHIZINSKI* and
BRUCE VONDRACEK*†
*Department of Fisheries, Wildlife and Conservation Biology, University of Minnesota, 1980 Folwell Avenue, 200 Hodson Hall,
Saint Paul, MN 55108, USA, †United States Geological Survey, Minnesota Cooperative Fisheries and Wildlife Research Unit
Reintroductions are commonly employed to preserve intraspecific biodiversity in
fragmented landscapes. However, reintroduced populations are frequently smaller and
more geographically isolated than native populations. Mixing genetically, divergent
sources are often proposed to attenuate potentially low genetic diversity in reintroduced
populations that may result from small effective population sizes. However, a possible
negative tradeoff for mixing sources is outbreeding depression in hybrid offspring. We
examined the consequences of mixed-source reintroductions on several fitness surrogates
at nine slimy sculpin (Cottus cognatus) reintroduction sites in south-east Minnesota. We
inferred the relative fitness of each crosstype in the reintroduced populations by
comparing their growth rate, length, weight, body condition and persistence in
reintroduced populations. Pure strain descendents from a single source population
persisted in a greater proportion than expected in the reintroduced populations, whereas
all other crosstypes occurred in a lesser proportion. Length, weight and growth rate were
lower for second-generation intra-population hybrid descendents than for pure strain
and first-generation hybrids. In the predominant pure strain, young-of the-year size was
significantly greater than any other crosstype. Our results suggested that differences in
fitness surrogates among crosstypes were consistent with disrupted co-adapted gene
complexes associated with beneficial adaptations in these reintroduced populations.
Future reintroductions may be improved by evaluating the potential for local adaptation
in source populations or by avoiding the use of mixed sources by default when
information on local adaptations or other genetic characteristics is lacking.
Keywords: Cottus, fish conservation, fitness, hybridization, sculpin, translocation
Received 23 July 2010; revision received 21 July 2011; accepted 3 August 2011
Introduction
Reintroductions, which are intended to re-establish a
species within its former range, are a common practice
(IUCN 1998; Fischer & Lindenmayer 2000; Seddon et al.
2007) and are integral to a high proportion of recovery
plans for imperiled fishes in North America (Williams
et al. 1988; George et al. 2009). As habitats continue to
be degraded, there is concern that fragmented wild ani-
mal populations are vulnerable to inbreeding depres-
sion and reduced evolutionary potential (Keller &
Waller 2002; Jenkins 2003). Yet, reintroduction practices
may hasten a decline in genetic variability by reducing
the effective population size of the source and reintro-
duced populations (Lande & Barrowclough 1987; Grif-
fith et al. 1989; Wolf et al. 1996). Intentional
hybridization of genetically divergent source popula-
tions is a reintroduction approach that may alleviate
inbreeding depression (e.g., Pimm et al. 2006), restore
genetic diversity to historic levels and provide novel
Correspondence: David D. Huff, N.O.A.A., Southwest Fisheries
Science Center, Fisheries Ecology Division, 110 Shaffer Road,
Santa Cruz, CA 95060, USA, Fax: 1-831-420-3921;
E-mail: david.huff@noaa.gov
2011 Blackwell Publishing Ltd
Molecular Ecology (2011) 20, 4246–4258 doi: 10.1111/j.1365-294X.2011.05271.x
genetic combinations required for rapid evolutionary
change (Lewontin & Birch 1966; Stockwell et al. 2003).
An increase in fitness known as hybrid vigour that
may result from either the suppression of deleterious
recessive alleles or the beneficial overdominance has
been recognized as a useful application of outbreeding
(or hybridization) for centuries (Darwin 1876; Lynch
1991). However, intentional mixing of distinct source
populations poses risks (Lesica & Allendorf 1999; Jones
2003); among these is a decline in fitness among off-
spring that are the product of mating between geneti-
cally disparate individuals, known as outbreeding
depression (Lynch 1991; Tallmon et al. 2004). Two
mechanisms may lead to outbreeding depression. First,
interpopulation hybridization may decrease fitness
because introduced nonlocal alleles cause a population
to become less suited to local environmental conditions
by producing intermediate phenotypes (Hatfield & Sch-
luter 1999). Reduced fitness by this mechanism would
be apparent in the F
1
generation. Outbreeding depres-
sion may also occur by a second mechanism, the dis-
ruption of positive epistatic interactions (i.e., co-adapted
gene complexes), which would only occur beyond the
first generation because of recombination and segrega-
tion during meiosis in the F
1
generation. The resulting
F
2
genomes may contain genes with different evolution-
ary histories that have not undergone co-adaptive selec-
tion as a group (Brncic 1954; Templeton et al. 1986).
Although there is debate that inbreeding depression
may not be relevant to wild populations (Pusey & Wolf
1996), there is ample evidence that it occurs (Keller &
Waller 2002). Conversely, recent literature has high-
lighted the potential importance of outbreeding depres-
sion for conservation, but there is a relative paucity of
reports that document it (Edmands 2007; McClelland &
Naish 2007). Most outbreeding studies rely on con-
trolled crosses conducted in a laboratory, whereas stud-
ies in natural environments are rare, especially those in
wild fish populations (Edmands 2007; McClelland &
Naish 2007). While laboratory studies focus on purely
genetic components that require little environmental
context, field studies can provide opportunities to mea-
sure the genetic and ecological constituents of hybrid
fitness acting in concert. This environment-dependent
reduction in hybrid fitness is relevant to both conserva-
tion efforts and our understanding of the role of diver-
gent evolution as the basis of local adaptation and,
ultimately, reproductive isolation.
Outbreeding depression has been demonstrated in a
variety of organisms, including fish (Endler 1977; Ed-
mands 2007; McClelland & Naish 2007). Nevertheless, a
meta-analysis of the consequences of unrelated fish
population crosses by McClelland & Naish (2007)
demonstrated that overall the response to outbreeding
was positive and there was no relationship between the
genetic distance among parental populations and its
effect on life history, behavioural, morphological or
physiological traits. Yet many of the comparisons they
reported were inconclusive. Failure to detect positive or
negative outcomes of outbreeding may have been asso-
ciated with small sample sizes and a lack of power to
detect statistical significance, especially in the F
2
genera-
tion. Ultimately, they concluded that outbreeding conse-
quences may be difficult to predict and that it might be
unreasonable to make broad generalizations because of
the variety of processes by which an outbreeding effect
may occur. They also identified several gaps in the
understanding of outbreeding outcomes in fishes.
Among these were documenting responses to outbreed-
ing beyond the F
1
generation, responses of fitness
related traits within a single species, the influence of
environment on hybrid response and the relationship
between outbreeding depression and genetic distance.
We utilized a native fish reintroduction programme
that provided a unique opportunity to investigate out-
breeding effects in a natural setting. The timing, num-
ber and origin of reintroduced individuals were known,
the source populations were genetically distinct and
there were nine discrete reintroduced populations in
close geographic proximity that provided an unusual
degree of replication for a field study (See Huff et al.
2010 for additional details). We recognized the potential
for adaptive differences between source populations
because one of the source strains was far more persis-
tent in the reintroduced populations. We accordingly
hypothesized that the fitness surrogates body size,
growth rate and body condition would differ among
pure strain individuals in the reintroduced populations
depending on ancestral origin. Furthermore, we investi-
gated the consequences of outbreeding on fitness surro-
gates in first- and second-generation hybrid crosses and
backcrosses among young-of-the-year (hatched to age I)
and over-yearlings (older than age I). Our aim in
including second-generation crosstypes in the analysis
was to enable a greater understanding of the genetic
mechanisms underlying outbreeding if we detected a
difference among crosstypes in fitness surrogates.
Methods
Study organism and reintroduction project
The slimy sculpin (Cottus cognatus Richardson) is a
small (<130 mm), cryptic, freshwater fish that occupies
benthic habitats in lakes, rivers and small streams from
Virginia, USA, to Labrador in eastern Canada and
northwest across Canada to eastern Siberia (Scott &
Crossman 1979). Cottus spp. are often locally abundant
MIXED-SOURCE REINTRODUCTIONS AND OUTBREEDING DEPRESSION 4247
2011 Blackwell Publishing Ltd
and are frequently a prominent constituent of ecosys-
tems suited to trout and other cold-water fish (Petrosky
& Waters 1975; Goyke & Hershey 1992). Slimy sculpins
in the study region spawn once per year during the
early spring at age II, or rarely at age I, and live up to
6 years (Petrosky & Waters 1975).
The study area is located in the Driftless Region of
south-east Minnesota, USA (Fig. 1). Prior to major set-
tlement by European immigrants beginning in 1850,
nearly all of the spring-fed streams in the region pre-
sumably held slimy sculpins and brook trout (Salvelinus
fontinalis). In subsequent years, slimy sculpin and other
cold-water fish abundance declined because of severe
habitat degradation and overexploitation (Waters 1977;
Leopold & Sewell 2001). Since the 1940s, the Minnesota
Department of Natural Resources (MNDNR) and other
organizations completed hundreds of in-stream
improvement projects (Waters 1977; Thorn et al. 1997;
MNDNR 2003). In locations where they improved habi-
tat sufficiently, MNDNR personnel recently reintro-
duced slimy sculpins. The goal was to re-establish
viable, self-sustaining populations where native popula-
tions were likely present historically, but had been
extirpated. Nine recipient streams were stocked from
2003 to 2005 in mid-autumn with a mixture of sculpins
from three source streams. The source streams are all
small tributaries in separate sub-drainages that enter
the Mississippi River within approximately 40 river
kilometers of each other (Fig. 1). We focus our analysis
on only two of these source ancestries: Beaver Creek
(Beaver) and Garvin Brook (Garvin). Although three
source populations were originally reintroduced, for
this evaluation, we did not consider descendents from
one source, Cold Spring Brook, because initial analyses
indicated very low overall ancestry in most reintro-
duced populations and they were not stocked some
years in certain streams. A total of 1230 Beaver and
Garvin sculpins were stocked in equal proportions
across all nine reintroduction sites. Huff et al. (2010)
provide additional details for the reintroduction pro-
gramme.
Sampling
We conducted sampling in autumn 2007 at all source
and recipient sites except Pickwick Creek, which we
sampled in spring and autumn 2008 and autumn 2009.
We additionally sampled Little Pickwick Creek in
autumn 2008 and autumn 2009. We collected fish using
a Wisconsin Abp-3 pulsed DC backpack electrofisher
with power output settings adjusted to minimize effects
on the reintroduced fish (Cowx & Lamarque 1990). We
anaesthetized fish with tricaine methanosulfate (MS-
222) (Summerfelt & Smith 1990); then we weighed and
measured standard length for each fish. We clipped a
small amount of tissue from the left pelvic fin of each
fish and preserved it in 95% ethanol for genetic analy-
sis. After processing, we returned all fish to the streams,
except those captured in autumn 2008 and 2009 at Pick-
wick and Little Pickwick sites. We euthanized these fish
with a lethal dose of anaesthetic (1 g ⁄ L of MS-222) and
retained them for otolith analysis. We did not capture
any of the visibly marked, originally stocked fish dur-
ing these sampling events.
United States of America
MN
IA
WI
IL
Trout Brook
Hay Creek
Cold Spring Brook
Beaver Creek
Garvin Brook
Pickwick Creek
Klaire Creek
Gilbert Creek
Rock Creek
LiƩle Pickwick Creek
Latsch Creek
Sugar Loaf Creek
Mississippi River
Fig. 1 Source (closed circle) and recipi-
ent (closed triangle) sites in south-east
Minnesota. The Driftless Region where
study sites are located is shown in the
inset, indicated by the cross-hatched
area covering portions of Minnesota,
Wisconsin, Iowa and Illinois.
4248 D. D. HUFF ET AL.
2011 Blackwell Publishing Ltd
DNA extraction and amplification
For 2007 and 2008 samples, we initially used eight mi-
crosatellite loci developed for other Cottus species that
resolved genetic variation in C. cognatus: Cgo18, Cgo42,
Cgo310 and Cgo1033 (Englbrecht et al. 1999); Cott290,
Cott686 and CottES1 (Nolte et al. 2005); and Cba14 (Fiu-
mera et al. 2002). We extracted DNA for polymerase
chain reaction (PCR) amplification using a chelating
resin as described in Fujishin et al. (2009). Microsatellite
amplification was performed in 15 lL reactions contain-
ing 1· polymerase buffer (10 m
M Tris-HCl, 50 mM KCl,
0.1% Triton
X-100), 1.5 mM MgCl2, 0.2 mM each dNTP,
0.5 l
M of the forward and reverse primers, with the for-
ward primer labelled with a fluorescent dye 6FAM,
VIC, NED or PET, and 0.5 units Taq DNA polymerase
(Promega, Madison, WI, USA). Amplification was car-
ried out in a thermocycler (Hybaid Omn-E; Thermo-
Hybaid US, Franklin, MA, USA) with 35 cycles at the
following temperature profile: 95 C for 30 s, 50 C for
30 s and 72 C for 1 min; followed by a 20-min exten-
sion at 72 C. We submitted PCR products to the Bio-
medical Genomics Center (University of Minnesota, St.
Paul, MN, USA) for electrophoresis on an ABI Prism
3130xl Genetic Analyzer (Applied Biosystems, Foster
City, CA, USA). We scored alleles using the software
program
GENOTYPER 2.5 (Applied Biosystems 2001). For
samples that we determined to be advanced generation
crosses, we carried out a second round of amplification
with six newly developed microsatellite loci for C. cogn-
atus (Fujishin et al. 2009): Cco01, Cco09, Cco10, Cco14,
Cco15 and Cco17. Autumn 2009 samples from Little
Pickwick Creek and Pickwick Creek were genotyped at
12 loci, with the lower variation markers Cott290 and
Cott686 removed from the original set.
Crosstype assignment
We used multilocus genotype data to assign individual
fish to one of six crosstypes: parental (Beaver, Garvin),
first-generation hybrids (F
1
) or second-generation
hybrids [F
2
, backcrosses to Beaver (F
1
· B), or back-
crosses to Garvin (F
1
· G)]. First, we evaluated data
from three source populations in
MICROCHECKER v2.2.3 to
detect evidence of null alleles or scoring errors because
of large allele drop-out (Van Oosterhout et al. 2004).
Conformance with Hardy–Weinberg expectations and
linkage equilibrium was tested using
GENEPOP v4.0.4
(Raymond & Rousset 1995). We adjusted significance
values for both tests using sequential Bonferroni proce-
dures (Rice 1989).
We next removed fish with Cold Spring ancestry
from the dataset. The proportion of each individual’s
ancestry derived from the three source populations was
estimated using the Bayesian clustering algorithm
implemented in the program
STRUCTURE (V. 2.2.3; (Prit-
chard et al. 2000); also refer to http://pritch.bsd.uchica-
go.edu). The number of populations (K) was set to
three, which was the known number of genetically dis-
tinct source populations, with an admixture model and
correlated allele frequencies. We ran the program with
a 50 000 burn-in period followed by 100 000 Monte Car-
lo simulations. Baseline individuals were included in
the runs without population identification to assist reso-
lution of genetically differentiated clusters and deter-
mine the ability of
STRUCTURE to determine the ancestry
of known fish. Individuals with probable Cold Spring
ancestry (q > 0.125) were removed from the dataset,
and we conducted subsequent analyses with only Bea-
ver Creek and Garvin Brook descendents.
The software NewHybrids (Anderson & Thompson
2002) was used to classify individual fish to crosstypes
assuming no more than second-generation descendents
of founders were present. This assumption is reasonable
as reintroduced populations were sampled within
3 years of initial spawning and sculpins typically
mature at age 2 (Petrosky & Waters 1975). Individuals
from the two source populations were included as a
baseline in the analyses. Each run had a 50 000 burn-in
period followed by 150 000 simulations, using the Jeff-
reys prior option for allele frequencies and mixing pro-
portions. Runs were repeated using different seeds to
verify that consistent solutions were found. We classi-
fied individuals into a pure strain or hybrid category if
their probability of membership was ‡0.70; otherwise,
the classification of the individual was considered
uncertain. Second-generation hybrids were difficult to
distinguish, so we genotyped the previously mentioned
additional six loci for all individuals whose combined
probability of membership across all three-second-gen-
eration cross-types exceeded 0.70 and repeated the Ne-
wHybrids analysis.
Statistical analysis
Expected quantities of each strain within the reintro-
duced populations were estimated for the autumn 2007
sampling season using a two-generation multinomial
expansion of crosstypes based on the quantity of indi-
viduals from each strain that were stocked and assum-
ing null conditions: equal survival, equal reproduction
and random mating among lineages (see Epifanio &
Philipp 2000). The first and second generations com-
prised the total population in a 1 : 2 ratio, and origi-
nally stocked fish were subtracted from the totals for
each corresponding pure strain category. We considered
a 1 : 2 ratio a conservative approximation of overall
population growth based on population estimates (Huff
MIXED-SOURCE REINTRODUCTIONS AND OUTBREEDING DEPRESSION 4249
2011 Blackwell Publishing Ltd
2010) that indicate abundances from 2 to 10 times
greater than were originally stocked in the reintroduced
populations. This ratio would also tend to overestimate
expected pure strain and F
1
individuals because more
of these are produced in the first generation of admix-
ture than the second (Epifanio & Philipp 2000). Like-
wise, expected quantities of F
2
, F
1
· B and F
1
· G
would be underestimated by our chosen ratio. Statistical
assessment of the divergence between expected and
observed values for each category was made using the
median test (Zar 1999), a version of the Kruskal–Wallis
ANOVA that frames the computation in terms of a contin-
gency table. Pickwick and Little Pickwick populations
were stocked once in autumn 2005, so there were only
Beaver, Garvin and F
1
crosstypes present in 2007 and
spring 2008. Our estimate, therefore, included only first-
generation crosstypes (Beaver, Garvin, and F
1
) for the
two Pickwick sites based on the timing of the sampling
relative to when these sites were stocked.
We used a fixed effects model (Weisberg 1993; Weis-
berg et al. 2010) to analyse differences among cross-
types in incremental growth rates using otoliths
collected from 418 sculpins from Pickwick and Little
Pickwick reintroduction sites. Because extracting oto-
liths from fish is lethal, we killed fish from only two of
the populations. Otoliths were collected from these pop-
ulations in autumn of 2008 and 2009 to ensure that
there would be enough F
2
, F
1
· B and F
1
· G crosstypes
to develop a growth model. We modelled the growth
increments for each fish as a function of four fixed
effects: age (levels = 0, 1, and 2), stream (levels = Pick-
wick and Little Pickwick), crosstype (levels = Garvin,
Beaver, F
1
, F
2
, F
1
· B and F
1
· G) and sex (lev-
els = Male, Female and Unknown). Following specifica-
tions from Weisberg et al. (2010), we also included
Year, Year–Age interaction and Unique ID as random
effects because of year-to-year variation likely to occur
between the sequential years of sampling and natural
variation likely to occur among individual fish. The
Year–Age interaction was included in the model to
allow for separate year effects during each year of a
fish’s life. Models were fit using the function lmer (Bates
& Maechler 2009) in R (v 2.10.0) using maximum likeli-
hood procedures. Starting with the full model contain-
ing all fixed parameters and 2-level interactions, we
used backward model selection to select the most parsi-
monious model as determined by the lowest Akaike
Information Criterion (AIC) corrected for sample size
(Burnham & Anderson 1998). The lmer function does
not produce p-values for model parameters so these
were calculated using the pvals.fnc function in R (Baa-
yen 2009), which computes P-values and Markov chain
Monte Carlo (1000 iterations) confidence intervals for
mixed models. Post-hoc multiple comparisons of means
(Tukey’s) among crosstypes were calculated using the
glht function in R (Hothorn et al. 2008).
Because there are possible age differences between
first (Beaver, Garvin and F
1
) and second (F
2
, F
1
· B and
F
1
· G) generation crosstypes that could be because of
differential survival or recent stocking, we categorized
all fish into two age categories that included young-of-
the-year and over-yearlings. All young-of-the-year fish
were designated by length (<41 mm, n = 91), based on
age–length relationships from the Pickwick otolith data
and supplementary otolith data from Beaver (n = 38)
and Garvin (n = 39) source sites (see Appendix S1).
In addition to crosstype comparisons for weight and
length, we compared relative body condition, a trait
that is generally considered a good indicator of fitness
in fish (Danzmann et al. 1988; Rakitin et al. 1999; The-
len & Allendorf 2001) to corroborate potential differ-
ences in growth rate and body size. We estimated body
condition by calculating relative condition factor (K
n
),
which has previously been employed as a fitness
related trait in Cottus species (Knaepkens et al. 2002),
for each fish. Relative condition factor is defined as
K
n
= W ⁄ W
pred
, where W is the observed weight and
W
pred
is the predicted weight from a third-order poly-
nomial based on a weight–length relationship (LeCren
1951; Wootton 1998) developed for each reintroduced
population.
Results
Genetic marker s and error simulations
The microsatellite data indicated strong differentiation
between the source populations (F
ST
= 0.32, P < 0.05).
Within each population, all loci were in Hardy–Wein-
berg and linkage equilibrium. We detected no evidence
for null alleles or large allele dropout. Simulated geno-
types for the eight initial loci in NewHybrids estimated
error rates of 2–6% for Beaver, Garvin and F
1
cross-
types; these errors caused assignment to F
2
, F
1
· B and
F
1
· G, whereas 8–12% of F
2
, F
1
· B and F
1
· G cross-
types were erroneously assigned to Beaver, Garvin and
F
1
crosstypes. Simulations using all 12 loci estimated
error rates of 0–0.3% for Beaver, Garvin and F
1
cross-
types. For F
2
, F
1
· B and F
1
· G crosstypes, 2–5% erro-
neously assigned to Beaver, Garvin or F
1
, and 1–3% of
backcrosses erroneously assigned to F
2
, while 12% of F
2
erroneously assigned to backcrosses.
Persistence
Assignment of crosstypes for 1230 slimy sculpins
revealed that there were more sculpins of Beaver ances-
try (531; 43% of total) at the reintroduction sites than
4250 D. D. HUFF ET AL.
2011 Blackwell Publishing Ltd
