ORIGINAL ARTICLE
Landscape genetics informs mesohabitat preference and
conservation priorities for a surrogate indicator species in a
highly fragmented river system
J Lean
1
, MP Hammer
2,3
, PJ Unmack
4
, M Adams
2,5
and LB Beheregaray
1
Poor dispersal species represent conservative benchmarks for biodiversity management because they provide insights into
ecological processes influenced by habitat fragmentation that are less evident in more dispersive organisms. Here we used the
poorly dispersive and threatened river blackfish (Gadopsis marmoratus) as a surrogate indicator system for assessing the effects
of fragmentation in highly modified river basins and for prioritizing basin-wide management strategies. We combined individual,
population and landscape-based approaches to analyze genetic variation in samples spanning the distribution of the species in
Australia’s Murray–Darling Basin, one of the world’s most degraded freshwater systems. Our results indicate that G. marmoratus
displays the hallmark of severe habitat fragmentation with notably scattered, small and demographically isolated populations
with very low genetic diversity—a pattern found not only between regions and catchments but also between streams within
catchments. By using hierarchically nested population sampling and assessing relationships between genetic uniqueness and
genetic diversity across populations, we developed a spatial management framework that includes the selection of populations in
need of genetic rescue. Landscape genetics provided an environmental criterion to identify associations between landscape
features and ecological processes. Our results further our understanding of the impact that habitat quality and quantity has on
habitat specialists with similarly low dispersal. They should also have practical applications for prioritizing both large- and
small-scale conservation management actions for organisms inhabiting highly fragmented ecosystems.
Heredity (2017) 118, 374–384; doi:10.1038/hdy.2016.111; published online 23 November 2016
INTRODUCTION
Freshwater environments are naturally and anthropogenically frag-
mented because of physical features such as mountain ranges, water-
falls, cascades, dams and agriculture. The degree to which
fragmentation affects species is directly influenced by their life-
history traits and habitat preferences (Frankham et al.,2010;
Osborne et al., 2014). For example, fragmentation can have a
negligible effect in species with high dispersal abilities that are able
to cross unsuitable landscapes between habitat patches (Boizard et al.,
2009; Faulks et al., 2010). However, in species with low dispersal
abilities, migration between patches can become impossible (Bilton
et al., 2001) that can result in severe demographic reductions and local
extinctions, especially when habitat fragmentation is accompanied by
habitat loss and degradation (Fahrig, 2003; Fagan et al.,2005;
McCraney et al., 2010). Such scenarios are expected to lead to loss
of genetic variation and inbreeding depression in populations of poor
dispersers, negatively affecting their adaptive potential and long-term
persistence (Hanski and Gaggiotti, 2004).
Landscape genetics contributes to the management of fragmented
metapopulations by providing an environmental angle to identify
associations between landscape features and ecological and
evolutionary processes. Landscape genetics combines information
from spatial statistics and population genetic data sets to identify
factors affecting population connectivity (Manel et al., 2003). As such,
it can clarify relationships between a species and its contemporary
environment. Riverine environments provide ideal systems for land-
scape genetic surveys, as dispersal of obligate aquatic organisms is
necessarily restricted to the river network. Thus, dispersal corridors
and the associated habitat features can be better defined compared
with many terrestrial and marine study systems (Wang et al., 2009).
Landscape genetics allows assessment of connectivity, both func-
tional (dispersal and gene flow) and structural (distance between
patches and differing habitats), enabling statistical assessment of
models of population structure. One such model is the stream
hierarchy model that predicts that dispersal of riverine organisms is
influenced by riverine distances between populations, and is therefore
hierarchically constrained by the dendritic nature of the river basins
(Meffe and Vrijenhoek, 1988; Fausch et al., 2002; Hughes et al.,2009).
Being able to test predictions from spatial models of population
structure can greatly improve our knowledge of the movements and
habitat preferences of a species. This allows us to better direct
1
Molecular Ecology Laboratory, School of Biological Sciences, Flinders University, Adelaide, South Australia, Australia;
2
Evolutionary Biology Unit, South Australian Museum, North
Terrace, Adelaide, South Australia, Australia;
3
Curator of Fishes, Museum and Art Gallery of the Northern Territory, Darwin, Northern Territory, Australia;
4
Institute for Applied
Ecology and Collaborative Research Network for Murray-Darling Basin Futures, University of Canberra, Canberra, Australian Capital Territory, Australia and
5
School of Biological
Sciences, University of Adelaide, Adelaide, South Australia, Australia
Correspondence: Professor LB Beheregaray, Molecular Ecology Laboratory, School of Biological Sciences, Flinders University, Adelaide, South Australia 5001, Australia.
E-mail: Luciano.Beheregaray@flinders.edu.au
Received 7 February 2016; revised 27 September 2016; accepted 27 September 2016; published online 23 November 2016
Heredity (2017) 118, 374– 384
&
2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved 0018-067X/17
www.nature.com/hdy
conservation initiatives, predict metapopulation scenarios and plan for
future threats (see, for example, Hopken et al.,2013).
To date, most landscape genetic studies on freshwater fishes (that is,
riverscape genetics) have focused on medium- to large-bodied species
with relatively high migration potential (Wofford et al.,2005;Whiteley
et al.,2006;Faulkset al., 2010). Very few studies have been conducted
on benthic, low dispersal species that display traits that make them
more susceptible to habitat fragmentation (McCraney et al., 2010;
Hopken et al.,2013;Robertset al., 2013). In addition, most studies
have predominantly focused on the impact that barriers or landscapes
variables have on population connectivity, or gene flow, with only a
few assessing the impact of the local environment on population
genetic diversity (Cena et al., 2006; Faulks et al., 2010, 2011). Yet, it is
these landscape genetic studies that identify which species-specific
habitats are best able to support larger, and consequently, healthier
populations that are likely to provide the most valuable information to
guide basin-wide conservation efforts.
Here we use population and landscape genetic approaches to study
a surrogate indicator species. Species-based biodiversity surrogates,
such as indicator species, provide valuable alternatives for assessing the
complex impacts of habitat fragmentation on ecosystems (Rodrigues
and Brooks, 2007). Indicator species can be treated as proxies that
monitororreflect the presence, abundance or richness of other
elements of biota, critical ecosystem processes or particular environ-
mental impacts (Lindenmayer and Likens, 2011). We assess
the influence of riverine configuration and habitat features on the
population structure and genetic diversity in a poor dispersal and
threatened freshwater fish from Australia’svastMurray–Darling Basin
(MDB). The MDB is Australia’s most important agricultural region,
accounting for 41% of the national agricultural income and 60% of its
water use (CSIRO, 2008), despite being characterized by low and
variable rainfall and severe droughts. Reflecting these paradoxical
attributes, the MDB is one of the world’s most degraded freshwater
systems (Palmer et al., 2008). Large modification to flow regimes,
extensive agriculture, land clearing, invasive species, in-stream barriers
to fish passage and habitat modification throughout the MDB have
together led to a highly fragmented environment (Jackson and
Williams, 1980; Allison et al., 1990; Lintermans, 2007). To alleviate
the anthropogenic pressures placed on threatened species in the MDB,
an understanding of the key environmental features influencing
population demography is clearly needed before sound conservation
initiatives can be designed and implemented.
Our study species is the river blackfish Gadopsis marmoratus
(Percichthyidae), and specifically the small-growing form found in
the MDB (size o350 mm total length; candidate species ‘NMD
’, sensu
Hammer et al.,2014).AllmembersoftheG. marmoratus species
complex are regarded as a habitat specialist, with a strong preference
for microhabitats with low levels of sediment, slow-flowing, deep
pools and high structural integrity from undercut banks, woody debris
and large rocks (Khan et al., 2004; Koster and Crook, 2008).
Furthermore, they have a relatively small home range, low fecundity
and demersal larvae that inhabit dense cover (Jackson, 1978; Khan
et al., 2004; Koster and Crook, 2008). Populations of G. marmoratus
across the MDB have declined dramatically in their extent of
occurrence and abundance since European settlement, especially for
heavily regulated large river environments and areas with less reliable
rainfall (Lake, 1971; Lintermans, 2007). This decline is formally
recognized at the northern and southern edges of the MDB with the
species protected under fisheries legislation in Queensland and South
Australia, respectively, with other regional populations outside the
MBD also listed as threatened (that is, upper Wannon River, Victoria
and Snowy River, New South Wales). Several genetic data sets are
available for the G. marmoratus complex (Sanger, 1984; Ovenden
et al.,1988;Milleret al., 2004; Ryan et al., 2004; Hammer et al.,2014),
but these were not directed at analyses of population genetic variation,
connectivity and landscape genetics.
Limited dispersal capabilities, threatened status and anthropogeni-
cally altered habitat make G. marmoratus an ideal surrogate indicator
species on which to use a landscape genetics approach to explore how
habitat features and fragmentation influence ecosystem-wide contem-
porary genetic architecture. To provide the appropriate detailed
genetic resolution, we employed a suite of recently developed
microsatellite markers (Arias et al., 2013) and a comprehensive
sampling of catchments and localities across the MDB. We expect
G. marmoratus populations to exhibit low genetic diversity and high
levels of population structure. We further predict that genetic diversity
will be positively influenced by the frequency of slow-flowing pools
and perenniality as measures of microhabitat availability (see, for
example, Bond and Lake, 2005), and negatively influenced by
sedimentation and riparian damage (habitat limiting). The anthro-
pogenic pressures currently placed on G. marmoratus in the MDB are
likely to persist into the future, a scenario exacerbated because of
global climate warming (Balcombe et al., 2011). Therefore, we use our
findings to develop targeted management recommendations that
prioritize conservation resources across the basin while maintaining
and enhancing population genetic diversity. Our results should shed
further light on the influence that habitat quality and quantity imposes
on habitat specialists with similarly low dispersal, plus provide
practical guidelines for prioritizing conservation management actions,
at a range of spatial scales, for organisms inhabiting highly fragmented
aquatic ecosystems.
MATERIALS AND METHODS
Study system and samples
The MDB is an expansive river system, occupying most of south-eastern
Australia (41millionkm
2
). The broad climate shifts from cool temperate in
the south to semi-arid in the north and west, with a fringe of cooler upland
habitat in two areas of topographic relief: (1) the Great Dividing Range in the
east and south, the third longest land-based ridge in the world; and (2) the
smaller Mount Lofty Ranges (elevation 200–480 m). The majority of stream
flow is generated in these upland areas (Leblanc et al., 2012), with a series of
major tributaries arising along the Great Dividing Range draining into two
main rivers, the Murray and the Darling. The lower Murray River exits into the
Southern Ocean via the Lower Lakes. Small tributary streams flow off the
eastern Mount Lofty Ranges into the Lower Lakes and lower Murray River
(Figure 1).
A total of 332 G. marmoratus individuals (whole fish in liquid nitrogen or fin
clips in 100% ethanol) were collected between 1999 and 2013 from 29 sites
encompassing 14 catchments and 3 broad regions of the MDB, namely the
Upper Darling, Upper Murray and lower Murray (Figure 1 and Supplementary
Table S1). Five sites in the lower Murray, the subject of regular threatened
species monitoring (Hammer et al., 2013), were represented by multiple
temporal-sampling events to reach a suitable sample size (MAR, ROD, ARS,
NAN, TOO: Supplementary Table S1). In three small stream systems, three
samples represented a composite of spatially proximate sites (o1km apart:
TOO, ARS, KBC). All genotypes from the 29 sites were retained for individual-
level analyses (Supplementary Table S1), whereas population-level analyses
were carried out only on 12 site samples that had ⩾ 8 individuals (Table 1).
Genetic variation
Genomic DNA was extracted using a modified salting out protocol (Sunnucks
and Hales, 1996). Twelve polymorphic microsatellite loci developed specifically
for the MDB G. marmoratus were amplified via the PCR: Gama01, 02, 03, 04,
05, 06, 07, 08, 09, 10, 12 and 13 (Arias et al., 2013). Reagent concentrations and
Riverscape genetics of a poor dispersal fish
JLeanet al
375
Heredity
the touchdown PCRs follow Beheregaray and Sunnucks (2000). The PCR
products were screened on an ABI 3730xl automated DNA sequencer (Applied
Biosystems, Foster City, CA, USA) and alleles called using GENEMAPPER 4.0
(Applied Biosystems). MICRO-CHECKER 2.2.3 (Van Oosterhout et al., 2004)
was used to assess scoring errors and null alleles.
Fisher’s exact test for linkage disequilibrium between all pairs of loci and an
exact test for deviations from Hardy–Weinberg equilibrium were carried out in
GENEPOP 4 (Raymond and Rousset, 1995), adjusted for multiple comparisons
using sequential Bonferroni (Rice, 1989). Rarefied allelic richness (Ar) and
rarefied private allelic richness (Pa) (controlled for a population size of five) was
calculated using HP-RARE (Kalinowski, 2005). The latter uses rarefaction to
compensate for sampling bias due to a larger number of alleles expected in large
sample sizes compared with small sample sizes (Leberg, 2002). The observed
(H
O
) and expected heterozygosity (H
E
) per population and per locus was
calculated in ARLEQUIN 3.5.1.3 (Excoffier and Lischer, 2010). The inbreeding
coefficient (F
is
) was calculated using FSTAT 2.9.3 (Goudet, 2001) for each
Figure 1 Sampling sites for the river blackfish G. marmoratus in the MDB. The top inset shows the location of the MDB study region in Australia, the lower
insets show further detail on sites in (a) the Mount Lofty Ranges and (b) central Victoria. Sites are grouped into one of three major regions in the MDB (see
key) and three letter site codes follow Supplementary Table S1.
Riverscape genetics of a poor dispersal fish
JLeanet al
376
Heredity
population. A nonparametric Wilcoxon signed-rank test was used to assess
differences in genetic diversity (Ar and H
E
) between populations.
Genetic structure and assignment tests
The allele size permutation test in SPAGeDi 1.3d (Hardy and Vekemans, 2002)
indicated that F
st
was more appropriate than R
st
as an index for assessing
population differentiation. Thus, population pairwise comparisons were carried
out in Arlequin using F
st
that was also used to assess temporal variation in
population structure at sites with multiple samples (Supplementary Table S1).
The latter disclosed no significant differentiation and thus samples from
different years were pooled for further analyses. In addition, an analysis of
molecular variance (AMOVA) based on F
st
was conducted in ARLEQUIN to
assess the fit of predictions from the stream hierarchy model by partitioning
genetic variation by differences between sites within streams, as well as streams
within catchments and catchments within regions. The significance levels for all
F
st
and AMOVA comparisons were assessed using 10 000 permutations.
STRUCTURE 2.3.4 (Pritchard et al., 2000) was used to determine the most
likely number of population clusters using the correlated allele frequency
model, and not using sampling locations as priors. Twenty STRUCTURE runs
were carried out, each with 20 separate iterations and 1 million Markov chain
Monte Carlo replicates after an initial burn-in period of 300 000. The most
likely number of clusters (K) was chosen following Evanno et al. (2005) and
assessed with STRUCTURE HARVESTER (Earl and vonHoldt, 2012). The
results were combined using CLUMP1.1.2 (Jakobsson and Rosenberg, 2007)
and visualized using DISTRUCT 1.1 (Rosenberg, 2004). All above analyses were
conducted using 12 population samples.
Individual multilocus assignment tests were carried out using the Bayesian
approach of Rannala and Mountain (1997). Assignment tests were used to
determine the population of origin of fish from sites with small sample sizes
and to assess the proportion of individuals assigned to their sampling
population. The latter assignment is based on higher than average probability
of being born locally and provides a measure of natal site philopatry (see, for
example, Möller and Beheregaray, 2004). This was carried out in GENECLASS
2 (Piry et al., 2004) by simulating 10 000 individuals and using an assignment
threshold of 0.01
Spatial population structure
Isolation by distance (IBD) predicts genetic differentiation as a function of
geographic distance and is considered the baseline for evaluating more complex
metapopulation scenarios that may exceed the ability of IBD to represent spatial
pattern in genetic structure (Jenkins et al., 2010). We assessed IBD at regional
scales with Mantels tests in IBDWS (Bohonak, 2002) by comparing population
pairwise matrices of F
st
with stream distance matrices calculated in ARCMAP
10.1 (ESRI 2012, Redlands, CA, USA). Three separate regional IBD tests were
carried out: across all sites, across the Upper Murray and across the lower
Murray. The IBD test was not carried out for the Darling region because it was
represented by only three sites.
Analysis of migration
Contemporary levels of connectivity were assessed by estimating migration rates
between inferred population clusters using BAYESASS 3.0 (Wilson and Rannala,
2003). BAYESASS estimates recent (last two generations) migration rates between
population pairs. The assumptions associated with this analysis (that is, no
linkage between loci, F
st
values 40.05) hold for our population samples. Ten
million iterations were carried out for the Markov chain Monte Carlo analysis,
with the initial 1 million being discarded as burn-in. Samples intervals of 100
were used to estimate posterior probabilities. Delta values for migration rates,
allele frequencies and inbreeding coefficients were set at 0.15, 0.46 and 0.65,
respectively. Different random seeds were used to carry out six independent runs.
Influence of genetic differentiation on genetic diversity
This was assessed following Coleman et al. (2013), who suggested that
prioritizing management on differentiation alone may be detrimental to the
maintenance of genetic diversity because of the strong negative relationship
usually found between population differentiation and diversity. Population-
specific F
st
was calculated using a hierarchical Bayesian approach in Geste 2.0
(Foll and Gaggiotti, 2006) to determine the level of differentiation of each
population relative to other populations. Linear regression was then used to
assess the relationship between genetic uniqueness (population-specific F
st
)and
diversity (Ar and H
E
). The associated normality assumptions were tested using
the Shapiro–Wilk test and independent histograms in IBM SPSS Statistics 2.1
(Chicago, IL, USA). Linear regression was carried out on Ar and H
E
and
associated significance tested in R 3.0.1 (Vienna, Austria). As Ar and H
E
were
Table 1 Summary statistics for Gadopsis marmoratus in the MDB based on 12 microsatellite DNA markers
Site (code) N Na (± s.d.) % Poly loci Ar Pa F
IS
HWE H
o
H
E
Pop.-specific F
st
Upper Darling Browns Ck (BRN) 37 1.92 (0.38) 50% 1.65 0.00 0.04 0.002 0.47 0.49 0.689
Molong Ck (MOL) 6 2.00 (0.17) 83% 1.92 0.08 0.19 0.751 0.35 0.43 0.516
Shawns Ck (SCC) 35 2.83 (0.41) 83% 1.79 0.33 0.17 0.000 0.21 0.25 0.503
Upper Murray Abercrombie R (ABE) 5 2.17 (0.21) 92% 2.09 0.17 0.61 0.047 0.20 0.47 0.488
Catherines Ck (CAT) 9 2.42 (0.31) 92% 2.10 0.00 0.59 0.000 0.16 0.38 0.498
Stony Ck (SCP) 5 3.08 (0.36) 83% 2.90 0.25 0.49 0.001 0.36 0.66 0.252
Scrubby R (SCR) 22 2.58 (0.34) 83% 2.15 0.08 0.14 0.062 0.43 0.50 0.501
Hughes Ck (HUG) 30 3.42 (0.49) 92% 2.38 0.50 −0.03 0.128 0.48 0.46 0.428
King Parrot Ck (KPC) 26 2.08 (0.38) 58% 1.57 0.17 0.16 0.005 0.28 0.33 0.641
Birch Ck (BIT) 5 3.08 (0.40) 92% 2.88 0.42 0.12 0.450 0.51 0.57 0.250
Creswick Ck (CRT) 5 2.67 (0.376) 75% 2.54 0.17 − 0.04 0.198 0.69 0.66 0.323
Avoca R (AVL) 10 3.33 (0.36) 92% 2.55 0.17 0.42 0.000 0.29 0.49 0.290
Lower Murray Marne R (MAR) 8 2.83 (0.35) 83% 2.42 0.08 0.03 0.957 0.55 0.57 0.336
Angas R (ARS) 32 2.33 (0.26) 83% 1.63 0.08 0.14 0.389 0.28 0.33 0.613
Rodwell Ck (ROD) 29 3.08 (0.40) 83% 2.25 0.17 0.19 0.000 0.40 0.49 0.417
Nangkita Ck (NAN) 18 3.33 (0.50) 100% 2.26 0.25 0.33 0.000 0.26 0.38 0.386
Tookayerta Ck (TOO) 24 3.67 (0.61) 92% 2.30 0.25 0.37 0.000 0.28 0.45 0.390
Average 18 2.75 (0.10) 83% 2.20 0.19 0.23 0.37 0.39 Global F
st
= 0.45
Abbreviations: HWE, Hardy–Weinberg equilibrium; MDB, Murray–Darling Basin.
Number of individuals genotyped at each site (N), average number of alleles per locus (Na), percentage of polymorphic loci (% poly loci), allelic richness (Ar), proportion of private alleles (Pa),
inbreeding coefficient (F
IS
), HWE P-values, observed heterozygosity (H
O
) and expected heterozygosity (H
E
) and the population-specific F
st
.
Bold P-values indicate significant deviations after Bonferroni correction. Only sites with ⩾ 5 individuals are included (see Supplementary Table S1 for details).
Riverscape genetics of a poor dispersal fish
JLeanet al
377
Heredity
highly correlated (results not shown), only Ar was used for subsequent analysis,
as it is more sensitive to recent changes in population size and genetic diversity
(Nei et al., 1975). The significance and strength of the relationship between F
st
and Ar was also determined as above.
Landscape genetics
Landscape genetics was used to determine the impact of five habitat features on
genetic variation (Ar) of demes from the 12 sites with larger sample sizes.
Surrogates were used to represent meaningful habitat features based on expert
opinion and field observations. These were the sum of the average proportion
of grid cells used for forestry, intensive animal and plant production (that is,
land use) which were used as a surrogate for sedimentation (Trimble and
Mendel, 1995; Croke and Hairsine, 2006). The average difference in altitude
over each upstream grid cell (that is, slope) was used as a surrogate for the
frequency of deep slow-flowing pools. The distance between the nearest
upstream and downstream dam (that is, barrier) was used as a surrogate for
potential river area. The percentage flow contribution to mean annual discharge
by the six driest months of the year (that is, perenniality) was used to represent
fluctuations in available habitat. Finally, the percentage of the grid cells where
industrial, agricultural, urban and point source pollution exists (that is,
modified land) was used as an indicator of overall human impacts. Data were
sourced from the National Catchment and Stream Environment Database 1.1.5
(Stein, 2012), available as part of the Geofabric products associated with the 9 s
Digital Elevation Model (DEM)-derived streams and the National Catchment
Boundaries. The segment number from the GEODATA national 9 s DEM that
corresponds to sample sites was identified in ARCMAP. These were used to
identify environmental characteristics linked to the stream layer in the database
using ARCMAP. The mean value from all upstream grid cells was used for the
predictor variables as it is likely that habitat features upstream from the
sampling site have an impact on habitat in the downstream segments.
Normality of all response variables was assessed as outlined above and
transformations were made where necessary. Collinearity assumptions were
assessed by examining the significance of pairwise correlations between all
predictor variables. Perenniality was significantly correlated with both slope and
land use (Supplementary Table S2) and modified land could not be
transformed to meet normality. These two variables were removed from
further analyses.
The log likelihoods of the linear modes for all possible combinations of the
three candidate environmental variables influencing genetic variation was
calculated in the all.regs function of R hier.part package (Walsh and
Macnally, 2008). The best models were then selected using AIC
c
(Akaike’s
information criterion adjusted for small sample sizes; Akaike, 1973) and the
hierarchical partitioning function in the hier.part package used to assess the
individual contribution of each variable to the full model. Recent criticism has
been made about the use of linear regression and AIC
c
in landscape genetics
because of the non-independence of explanatory variables (Legendre and
Fortin, 2010; Van Strien et al., 2012). The new approaches suggested to
overcome these issues relate to comparison of distance matrices (Van Strien
et al., 2012) that is not suitable for our analysis based on Ar. An analysis of
variance was carried out on each of our four highest ranked models to test
whether models significantly explained the variation in Ar across sites. Finally,
hierarchical linear regression was carried out to determine which variables
significantly contributed to the predictive power of each model. We considered
a model to be meaningful if it had a positive or negative regression coefficient,
significantly accounted for variation in Ar (analysis of variance Po0.05) and
each variable entered had a significant contribution to the predictive power of
the model (Po0.05).
RESULTS
Data quality and low levels of genetic variation
Missing data accounted for only 0.3% of the genotypes and these were
spread evenly over populations and loci. No significant linkage
disequilibrium was identified after sequential Bonferroni correction.
There was no evidence of scoring errors. MICRO-CHECKER sug-
gested null alleles in 8 loci, but the results were not consistent across
the 12 population samples. Gama03 was the only locus with evidence
for null alleles in two populations (SCC and TOO). All analyses were
run with and without Gama 03 and produced similar results.
Deviations from Hardy–Weinberg equilibrium were evident in 7 out
of 17 populations; these were mostly because of excess of homozygotes
(Table 1).
Overall genetic diversity was strikingly low with an average Ar of
only 2.2 alleles (Table 1). Average number of alleles within populations
was also low, ranging from 1.92 (BRN) to 3.67 (TOO). There were
significant differences in Ar (Po0.001) and H
E
(Po0.001) between
populations. Allelic richness ranged from 1.57 (KPC) to 2.9 (SCP) and
H
E
ranged from 0.25 (SCC) to 0.66 (CRT and SCP). In particular,
populations BRN and KPC showed lowest diversity, including lowest
percentage of polymorphic loci (50% and 58%, respectively). High
and mostly significant F
IS
values, suggestive of inbreeding, were found
in most populations (overall average F
IS
= 0.23; Table 1).
Population structure
The global F
st
was very high (F
st
= 0.45, Po0.001), indicating marked
population differentiation for G. marmoratus across the MDB. All
pairwise population comparisons were significant, with F
st
ranging
from 0.13 to 0.67 (Supplementary Table S3).
AMOVA identified basin-level differentiation only between the
Upper Darling and lower Murray (P = 0.019), suggesting that popula-
tions are not structured mainly as a consequence of regional
differences. On the other hand, both the catchment- and the
regional-level AMOVAs identified that the largest amount of the total
variation was accounted for by differences between populations, both
between catchments and within streams within catchments (Po0.01).
This indicates that populations within catchments are more related
and that catchment hierarchy explains a significant level of differ-
entiation (Table 2).
In terms of IBD results, the Mantel tests using pairwise population
F
st
and pairwise stream distances matrices identified no correlation
between the two variables and therefore did not support a pattern of
IBD. This result was consistent across all regional scales tested
(Supplementary Table S4).
STRUCTURE analysis indicated a value for K of 11 populations
(Figure 2). Populations were well defined, with either low or no
admixture between clusters. Only one of the inferred clusters was
represented by more than one sampling site (streams NAN and TOO);
these are the two closest sites in our study, separated by only 11 km.
Samples from these sites were pooled together for the BAYESSASS
migration analysis. A few admixed individuals are also apparent in the
STRUCTURE graph (Figure 2), but these were generally found
between sites with smaller sample sizes (for example, AVL and
MAR) that probably relates to our reduced ability to estimate allele
frequency for these populations (Evanno et al.,2005).Allpotential
cases of admixture were successfully assigned to their sampled
populations via assignment tests. A high proportion of individuals in
all 11 populations (97%; range of 89–100%) had multilocus genotypes
that assigned to their sampled population based on GENECLASS
results (Supplementary Table S5). Fish from the undersampled
locations (n ⩽ 5) were not assigned to any of the 11 population
clusters (Supplementary Table S5).
BAYESASS did not identify significant migration between groups at
any of the hierarchical levels (that is, within catchments, between
catchments and between regions; Table 3). The majority of tests
indicated ˂2% of migrant ancestry from other populations, with
confidence intervals with a lower bound overlapping zero. The only
exception to this is the unidirectional gene flow of 17% from MAR to
SCR, but this could be related to MAR having a small sample (n = 8)
Riverscape genetics of a poor dispersal fish
JLeanet al
378
Heredity