Conservation implications of the evolutionary historyand genetic diversity hotspots of the snowshoe hare
ELLEN CHENG,* KAREN E. HODGES,† JOS �E MELO-FERREIRA,‡ PAULO C. ALVES*§ and
L. SCOTT MILLS*¶*Wildlife Biology Program, Department of Ecosystem and Conservation Sciences, University of Montana, 32 Campus Drive,
Missoula, MT 59812, USA, †Department of Biology, University of British Columbia Okanagan, 3333 University Way, Kelowna,
BC V1V 1V7, Canada, ‡CIBIO, Centro de Investigac�~ao em Biodiversidade e Recursos Gen�eticos, Universidade do Porto, InBIO -
Laboratorio Associado, Campus Agrario de Vair~ao, 4485-661 Vairao, Portugal, §Departamento de Biologia, Faculdade de
Ciencias da, Universidade do Porto, Rua do Campo Alegre s/n, 4169-007 Porto, Portugal, ¶Fisheries, Wildlife and Conservation
Biology Program, Department of Forestry and Environmental Resources, North Carolina State University, Raleigh, NC 27695,
USA
Abstract
With climate warming, the ranges of many boreal species are expected to shift north-
ward and to fragment in southern peripheral ranges. To understand the conservation
implications of losing southern populations, we examined range-wide genetic diversity
of the snowshoe hare (Lepus americanus), an important prey species that drives boreal
ecosystem dynamics. We analysed microsatellite (8 loci) and mitochondrial DNA
sequence (cytochrome b and control region) variation in almost 1000 snowshoe hares.
A hierarchical structure analysis of the microsatellite data suggests initial subdivision
in two groups, Boreal and southwestern. The southwestern group further splits into
Greater Pacific Northwest and U.S. Rockies. The genealogical information retrieved
from mtDNA is congruent with the three highly differentiated and divergent groups
of snowshoe hares. These groups can correspond with evolutionarily significant units
that might have evolved in separate refugia south and east of the Pleistocene ice
sheets. Genetic diversity was highest at mid-latitudes of the species’ range, and genetic
uniqueness was greatest in southern populations, consistent with substructuring
inferred from both mtDNA and microsatellite analyses at finer levels of analysis.
Surprisingly, snowshoe hares in the Greater Pacific Northwest mtDNA lineage were
more closely related to black-tailed jackrabbits (Lepus californicus) than to other snow-
shoe hares, which may result from secondary introgression or shared ancestral poly-
morphism. Given the genetic distinctiveness of southern populations and minimal
gene flow with their northern neighbours, fragmentation and loss of southern boreal
habitats could mean loss of many unique alleles and reduced evolutionary potential.
Keywords: climate change, core-periphery, evolutionarily significant units, landscape genetics,
Lepus americanus, phylogeography
Received 23 May 2013; revision received 5 May 2014; accepted 5 May 2014
Introduction
Over the next century, North America’s southern boreal
forests are predicted to undergo rapid fragmentation
and loss due to climate change and human activities
(IPCC 2007). Understanding the conservation implica-
tions of southern habitat loss for boreal species requires
evaluating range-wide genetic structure of individual
species and assessing the generality of these patterns
across taxa. Specifically, insights into large-scale genetic
diversity and population differentiation would clarify
the relative importance of southern boreal populations
as hotspots of diversity and evolutionarily significantCorrespondence: L. Scott Mills,
E-mail: [email protected]
© 2014 John Wiley & Sons Ltd
Molecular Ecology (2014) 23, 2929–2942 doi: 10.1111/mec.12790
units (ESUs; Moritz 1994), with implications for adap-
tive potential.
Distribution of genetic diversity and structure across
a species’ range reflects historical range contraction and
recolonization from glacial refugia as well as current
habitat fragmentation and dispersal. During the Quater-
nary ice ages (approximately 2.6 mya–present), the
alternation of glacial and interglacial periods caused
repeated changes in species’ distributions. A frequently
invoked ‘southern refugia’ model in phylogeography
suggests that when ice sheets advanced, many boreal
species persisted primarily in refugia in southern lati-
tudes (Hewitt 1996). Interglacial periods enabled north-
ward range expansion, with leading-edge populations
carrying a subset of the genetic diversity of refugial
populations. Simultaneously, range contraction to
higher elevations in southern populations may have
reduced connectivity and increased local diversification
(Moritz et al. 2008). For some species in North America
and Europe, a pattern of decreasing genetic diversity
with increasing latitude (‘southern richness, northern
purity’) may reflect a dominant influence of historical
southern refugia on patterns of diversity (Pielou 1991;
Hewitt 1996; Soltis et al. 1997).
The generality of the southern refugia model has
been challenged for species with high dispersal and
large contemporary ranges (Hewitt 2000; Provan & Ben-
nett 2008). For many North American species, fossil evi-
dence and phylogeographic studies have identified
additional glacial refugia in eastern Beringia, the Cana-
dian Arctic, coastal British Columbia, the Maritimes,
and other northern locations (Soltis et al. 2006; Provan
& Bennett 2008; Godbout et al. 2010). When expanding
populations from separate refugia met in zones of sec-
ondary contact, they often created hotspots of genetic
diversity (Provan & Bennett 2008). Geographic and
genetic subdivision within major refugia further compli-
cate genetic patterns (Gomez & Lunt 2007).
Contemporary gene flow also impacts genetic struc-
ture and diversity. Many boreal species have popula-
tions occupying peninsular habitat extensions into
montane forests of the USA (Shugart et al. 2005). In
addition to natural habitat fragmentation, these south-
ern boreal forests are heavily impacted by logging and
habitat conversion (Hansen et al. 2010; Powers et al.
2012), which could lead to differentiation in remnant
habitats. The core-periphery hypothesis suggests that
connected populations in the boreal range core should
have higher genetic diversity than the fragmented pop-
ulations of the southern periphery (Eckert et al. 2008).
But while low gene flow and chronic genetic drift may
reduce genetic diversity in peripheral populations, these
processes may simultaneously facilitate genetic differen-
tiation and preserve unique alleles (Eckert et al. 2008).
Given the complex interplay of forces that shape
intraspecific distribution of genetic diversity, what are
the consequences of losing southern boreal populations?
In this study we examined range-wide genetic diversity
of the snowshoe hare (Lepus americanus) to address this
question.
Snowshoe hares are important prey for most boreal
carnivores, structuring food web dynamics as strong in-
teractors (Krebs et al. 2001). Fossil evidence suggests the
persistence of snowshoe hares in extensive refugia
south of the ice sheets and in the northern refugium of
Beringia during the Last Glacial Maximum (LGM;
FAUNMAP Working Group 1994). Snowshoe hare pop-
ulations near historical Beringia and in Montana har-
bour high genetic diversity and are genetically
differentiated from each other (Burton et al. 2002). Mor-
phological differences among snowshoe hare popula-
tions in and around the Pacific Northwest suggest
genetically differentiated populations (Dalquest 1942).
We analysed mtDNA and microsatellite data
throughout the contemporary range of snowshoe hares
to test the hypotheses that: (i) extant snowshoe hare
populations derive from Beringia and southern refugia;
and (ii) snowshoe hare populations near the core of the
range exhibit higher genetic diversity than populations
near the periphery. We predict highest genetic diversity
and uniqueness in the species’ southern range and near
the Alaska-Yukon border, that is, in likely refugia, with
reduced diversity in range-edge populations outside of
these areas.
We then discuss how anticipated loss of southern bor-
eal habitats might affect snowshoe hare genetic diversity.
First, we determine whether multiple snowshoe hare
ESUs (sensu Moritz 1994) are warranted. Second, we
examine genetic diversity and uniqueness across a latitu-
dinal gradient, with particular focus on populations
below 49°N, the approximate southernmost extent of the
LGM. Many scenarios of climate change predict the cli-
mate envelope for North America’s boreal ecosystem will
shift north of 49°N within a century (Koven 2013).
Finally, we discuss similarities in findings between snow-
shoe hares and other North American hare species.
Materials and methods
We analysed 975 snowshoe hare samples from 16 U.S.
states and 12 Canadian provinces and territories
(Appendix S1, Supporting information). Nearly all sam-
ples were ear tissue collected from road kill, game har-
vests, and live-trapping during 1989–2010. Ten samples
were faecal pellets collected in Isle Royale, Michigan, in
2009. Eleven samples were tissue from specimens col-
lected near Vancouver, British Columbia, from 1929 to
1970, held at the University of British Columbia Cowan
© 2014 John Wiley & Sons Ltd
2930 E. CHENG ET AL.
Museum. For phylogenetic analyses, we additionally
analysed one white-tailed jackrabbit (Lepus townsendii)
tissue sample obtained from GenBank (Accession no.
AY292729; Matthee et al. 2004) and seven black-tailed
jackrabbit (L. californicus) tissue samples collected from
three U.S. states (California, New Mexico, Nevada).
Microsatellite analysis
Samples were grouped into populations on the basis of
two geographic criteria: (i) no potential genetic barriers
such as large lakes or rivers, mountain ranges, or non-
forested regions bisecting populations (Burton et al.
2002; Shafer et al. 2010b); and (ii) a maximum of
260 km between any two samples in a population. The
second criterion represents a coarse spatial scale much
greater than the distance hares disperse (up to ~20 km,
Gillis & Krebs 1999) but within the scale of reported
gene flow in northern snowshoe hare populations
(~600 km, Burton et al. 2002). After grouping samples,
we limited genetic analyses to groups with at least
seven samples. This minimum threshold was arbitrary,
but has precedence in other population genetic studies
(Schwartz et al. 2003; Tracy & Jamieson 2011).
We selected eight polymorphic microsatellite markers
developed in the European rabbit, Oryctolagus cuniculus,
and successfully used with snowshoe hares (Burton
et al. 2002; Schwartz et al. 2007): 7L1D3 (Korstanje et al.
2003); SAT02, SAT12, SAT13, SAT16 (Mougel et al.
1997); SOL08, SOL30 (with ‘GTGTCTT’ tail added) (Rico
et al. 1994); and SOL33 (Surridge et al. 1997) (Appendix
S2, Supporting information). DNA extraction and geno-
typing methods are detailed in Appendix S3 (Support-
ing information).
Allelic dropout and false allele rates were calculated
with 10 000 search iterations in Pedant version 1.0.
(Johnson & Haydon 2007). For each population, we used
Genepop version 4.0.11 (Rousset 2008) to test for Hardy-
Weinberg Equilibrium (HWE) and linkage disequilib-
rium. Markov chain parameters for exact tests were set
at 10 000 dememorizations, 100 batches, and 5000 itera-
tions per batch (Raymond & Rousset 1995). We used the
false discovery rate approach (FDR; Benjamini & Hoch-
berg 1995) in the R software package ‘fdrtool’ (Strimmer
2008; http://cran.r-project.org/) to correct for multiple
significance testing type I error. Potential null alleles
and scoring errors due to stuttering and allelic drop-out
were identified by Monte Carlo simulation in Micro-
Checker version .2.2.3 (van Oosterhout et al. 2004).
We performed Bayesian analyses in STRUCTURE
version 2.3.3 (Pritchard et al. 2000) and Geneland
version 4.0.3 (Guillot et al. 2005) to partition microsat-
ellite data into genetic clusters, and to assign individu-
als to their likely cluster of origin. In STRUCTURE, we
applied an admixture model with the ‘locprior’ option,
using a burn-in period of 20 000 generations and
100 000 MCMC iterations after burn-in. We compared
results from correlated vs. uncorrelated allele fre-
quency models. To check for MCMC consistency, we
performed 20 replicates for each K (number of clus-
ters) from 1 to 40. The most likely K value was deter-
mined in Structure Harvester version 0.6.93 (Earl &
vonHoldt 2011) as the likelihood model with the high-
est DK (Evanno et al. 2005), unless the maximum lnP
(D) was for K = 1 (which would indicate no substruc-
ture). In the presence of substructure, the DK method
detects the highest hierarchical structure (Evanno et al.
2005). We assigned each individual to its most proba-
ble cluster and repeated the analysis for each cluster
separately, until further substructure could not be
detected. Cluster assignment was based on outcomes
from the run with highest lnP(D) among 20 replicates.
Following Coulon et al. (2008), only individuals with at
least 60% membership in a cluster were included in
subsequent analyses. For each subsequent analysis,
model parameters remained the same, but maximum
K was set at one greater than the number of sampled
populations in that cluster.
In Geneland, we evaluated results from three spa-
tially explicit model combinations. We examined both
correlated and uncorrelated allele frequency models
without filtering null alleles. We additionally examined
an uncorrelated frequency model while filtering null
alleles. For each of the three model combinations we
ran 20 independent replicates of 1 000 000 MCMC itera-
tions with a thinning of 1000 and burn-in of 200 000
iterations. K was allowed to vary from 1 to 40. Follow-
ing program recommendations for our sample size, we
set maximum rate of Poisson process = 853 and maxi-
mum number of nuclei in the Poisson-Voronoi tessella-
tion = 2559. We allowed a 15-km uncertainty in spatial
coordinates. MCMC convergence was assessed for each
model combination by comparing estimated K and clus-
ter assignments across replicate runs.
GENALEX version 6.3 (Peakall & Smouse 2006) was
used to calculate number of alleles and expected hetero-
zygosity (Nei 1978) for each population. For all pairs of
populations, we estimated Nei’s D (Nei 1972) and Weir
& Cockerham’s (1984) FST, with the latter calculated in
ARLEQUIN version 3.5.1.2 (Excoffier et al. 2005). Signif-
icance was determined with 1000 permutations of
samples among populations and FDR correction for
multiple comparisons.
We used rarefaction, implemented in HP-RARE ver-
sion 1.0 (Kalinowski 2005), to calculate private allelic
richness (PAR) for each population, standardized to
the smallest sample size (seven individuals) in this
study. To minimize biases due to uneven sampling
© 2014 John Wiley & Sons Ltd
SNOWSHOE HARE LANDSCAPE GENETICS AND CONSERVATION 2931
density, all populations within a 350 km radius were
excluded from the calculation of PAR for each sam-
pled population (we also tested 500 km and results
were similar; data not shown). We examined scatter-
plots of genetic metrics against latitude and longitude
to identify signatures of genetic drift at the current
range periphery and to understand geographic pat-
terns of diversity.
Mitochondrial DNA analysis
We amplified a 468 bp fragment of the mitochondrial
control region (CR) with primers LCRSEQ (Melo-Ferre-
ira et al. 2007) and LepD2H (Pierpaoli et al. 1999) in all
snowshoe hare samples. A fragment with 633 bp of the
cytochrome b (Cytb) gene was also sequenced in a sub-
set of 80 snowshoe hare and seven black-tailed jackrab-
bit samples, using primers LGCYF (Alves et al. 2003)
and LCYTBR (Melo-Ferreira et al. 2005), as detailed in
Appendix S3 (Supporting information). The Cytb subset
comprised at least one snowshoe hare sample from each
population and additional samples from regions of high
CR genetic structure. We visually aligned sequences in
CodonCode Aligner version 3.5.4 (CodonCode Corpora-
tion, Dedham, MA, USA).
Phylogenetic trees were constructed in BEAST ver-
sion 1.7.4 (Drummond et al. 2012) based on the Cytb
gene, which has a slower mutation rate and thus lower
tendency than CR for homoplasy over long timescales
(Baker & Marshall 1997). We used jModelTest version
2.1.3 (Darriba et al. 2012) and the Bayesian information
criterion to assess the best-fit model of sequence evolu-
tion. Posterior probabilities were determined from three
independent runs of 250 million generations, using the
selected mutation model, the Yule tree prior and a ran-
dom local clock (Drummond & Suchard 2010), exclud-
ing the initial 10% of each run as burn-in. The stability
of the runs and convergence of the MCMC were
assessed with Tracer version 1.5 (http://beast.bio.ed.ac.
uk/Tracer). Results from the three runs were concate-
nated in LogCombiner version 1.7.4 and trees anno-
tated using TreeAnnotator version 1.7.4. The annotated
phylogenetic tree and posterior probability estimates
were visualized in FigTree version 1.4.0 (http://tree.
bio.ed.ac.uk/software/figtree/). To estimate lineage
divergence times, we used a mutation rate of 0.02 sub-
stitutions per site per million years (Brown et al. 1979),
which has been used to estimate divergence in other
hare studies (Pierpaoli et al. 1999; Melo-Ferreira et al.
2007).
The demographic history of the major mtDNA lineages
was inferred from control region sequences using the
Bayesian Skyline Plot (BSP) (Drummond et al. 2005)
implemented in BEAST. Three replicate runs of 100
million generations were performed using the appropri-
ate mutation models (for Boreal, TrN+I+G; for Greater
Pacific Northwest and U.S. Rockies, HKY+G) selected
using the procedure described above and a random local
clock (Drummond & Suchard 2010). Tracer version 1.5
was used to assess stability of the MCMC and the initial
10% of each run was discarded as burn-in. We used Log-
Combiner version 1.7.4 to concatenate results of the three
replicate runs. A CR mutation rate of 0.156 substitutions
per site per million years (derived from Melo-Ferreira
et al. 2007) was used to calibrate the BSP.
We examined within-lineage structure with the CR
gene, because its fast rate of evolution makes it suitable
for intraspecific studies (Vigilant et al. 1991). An unroot-
ed median-joining network (NETWORK version 4.5.1.6,
http://www.fluxus-engineering.com/) was generated
from CR haplotypes identified in DnaSP version 5.10
(Librado & Rozas 2009). Transversions were weighted
three times as high as transitions, following software
recommendations. For K = 1–10, SAMOVA version 1.0
(Dupanloup et al. 2002) identified the partitioning of CR
haplotype variance due to differences among groups.
We ran SAMOVA with 500 initial population partitions
and 10 000 iterations for each K. Significance of vari-
ance components was evaluated by 1000 permutations
of populations among groups.
We used ARLEQUIN v.3.5.1.2 (Excoffier et al. 2005) to
calculate haplotype and nucleotide diversities. As with
microsatellite data, scatterplots were used to assess lati-
tudinal and longitudinal patterns in genetic diversity.
To evaluate genetic differentiation, we calculated pair-
wise control region FST. ARLEQUIN v.3.5.1.2 was used
to determine significance of tests with 10 000 bootstraps
and FDR control for multiple comparisons.
Results
Microsatellite analysis
With an average of 2.2 PCR replicates per sample, we
successfully genotyped eight microsatellite loci for 922
snowshoe hares. The mean allelic dropout rate per allele
was 0.0070 and mean false allele rate was 0.0035, for all
loci combined. After excluding populations with <7 indi-
viduals, 853 samples in 39 populations remained for
analyses (Fig. 1). Only 4% of 312 population-loci combi-
nations significantly deviated from Hardy-Weinberg
Equilibrium, generally due to heterozygote deficit.
Slightly over 5% of 1026 tests for linkage disequilibrium
were significant. Micro-checker identified potential null
alleles in 8% of 312 population-loci tests. However,
null alleles and deviations from HWE were not associ-
ated more frequently with any particular locus, and
genotypic disequilibrium was not consistently attributed
© 2014 John Wiley & Sons Ltd
2932 E. CHENG ET AL.
to a particular locus pair. Therefore we retained all loci
for subsequent analyses.
STRUCTURE analyses identified hierarchical popula-
tion division (Fig. 1 and Appendix S4, Supporting infor-
mation). In the first round of STRUCTURE runs, the
highest likelihood model (K = 2) identified a Boreal
cluster comprising the entire northern and eastern
range of the species, and a southwestern cluster com-
prising remaining populations. The second round of
STRUCTURE further splits the Boreal cluster into two
subclusters. However, proportion membership in the
subclusters transitioned from west to east (Appendix
S5, Supporting information), suggesting an effect of iso-
lation by distance rather than historical isolation (Meir-
mans 2012). A Mantel test (Mantel 1967), following
Rousset’s (1997) method, confirmed a significant corre-
lation between geographic and genetic distance in the
Boreal cluster (P < 0.001). Because other analyses and
markers also supported a single Boreal cluster, we did
not continue STRUCTURE analyses to further subset
the Boreal cluster. In contrast to the Boreal cluster, the
second round of STRUCTURE clearly divided the
southwestern group into two genetic clusters, corre-
sponding to the Greater Pacific Northwest region and
to the U.S. Rockies. Further rounds of the hierarchical
STRUCTURE analysis subdivided the Greater Pacific
Northwest and U.S. Rockies groups into many subclus-
ters. Ultimately, by the fifth round of analysis, all
southwestern populations were identified as distinct
subclusters except for WA1 and WA4 in Washington.
For these two populations, some individuals could not
be assigned to a cluster with at least 60% probability,
and other individuals grouped with other populations.
Hierarchical cluster patterns were identical for the
correlated and uncorrelated allele frequency models.
Using Geneland, all replicates of the uncorrelated fre-
quency models (with and without filtering null alleles)
consistently identified a single Boreal cluster and 4–7
distinct clusters in the species’ southwestern range.
With the uncorrelated model and null alleles filtered,
the highest mean posterior density across 20 replicates
was obtained for K = 5, with clusters almost identical to
those identified from the first three rounds of STRUC-
TURE hierarchical analysis (Appendix S6, Supporting
information). All replicates of the correlated frequency
model in Geneland inferred 39–40 genetic clusters,
likely due to known instabilities of this model in the
presence of isolation by distance (Guillot 2008).
Measures of FST and Nei’s D were highly correlated
across populations (r = 0.93, P < 0.001). We found high
FST pairwise estimates (>0.20) between the three genetic
clusters identified in the first two rounds of STRUC-
TURE. These clusters were the most congruent across
markers (microsatellite and mtDNA) and analyses. Pair-
wise FST was high within the Greater Pacific Northwest
and U.S. Rockies clusters, but was usually below 0.20
within the Boreal cluster (Appendix S7, Supporting
information).
Most snowshoe hare populations were characterized
by high genetic diversity (Table 1). On average, popula-
tions in the Boreal cluster exhibited the highest allelic
richness and heterozygosity, but the lowest uniqueness
SouthwesternBoreal
STRUCTURE ROUND ONE
WA3
WA1
OR1
CA1
CA2 UT1
STRUCTURE ROUND TWO
BC1
OR2
WA4
MT1
WY1
CO1
AK4
AK2
AK6
BC4
BC2AB2
YK2
NWT1
NWT2
AB1 SK1 MB1
MN2MN1
IR1MI1
ON3 ON1
PA1
WV1
PA2
NY1
ME2ME1
QC4
NB1
QC3
Greater Pacific NorthwestU.S. Rockies
STRUCTURE ROUND THREE
Boreal
Fig. 1 Sampling locations and geographic
distribution of major snowshoe hare mi-
crosatellite clusters, as defined by
STRUCTURE version 2.3.3 hierarchical
analysis (Pritchard et al. 2000). The first
run of STRUCTURE distinguished Boreal
(circles) from southwestern (triangles)
populations. The second run further split
the southwestern cluster (grey shades).
By the third run, five distinct genetic
clusters were identified (dotted ovals).
The approximate southernmost latitude
of the LGM is marked with a grey
dashed line. Study results suggest extant
snowshoe hare populations likely
expanded from refugia south of this lati-
tude and from the current eastern hare
range.
© 2014 John Wiley & Sons Ltd
SNOWSHOE HARE LANDSCAPE GENETICS AND CONSERVATION 2933
(PAR). Greater Pacific Northwest populations exhibited
high diversity and the highest uniqueness of the three
major clusters.
Genetic diversity was highest at mid-latitudes (i.e.
near 49°N latitude; Fig. 2 and Appendix S8, Supporting
information) and increased from west to east across the
species’ range. For the Greater Pacific Northwest clus-
ter, PAR increased with latitude up to 49°N latitude
(Fig. 3). There were no apparent longitudinal trends in
PAR.
Table 1 Microsatellite diversities averaged across 8 loci for each of 39 sampled populations. Populations are grouped into three
genetic clusters identified by the first two rounds of STRUCTURE hierarchical analysis
Population N A AR H0 HE PAR
AB1 9 7.00 6.20 0.69 0.76 0.13
AB2 18 8.63 5.95 0.69 0.77 0.14
AK2 28 7.50 4.79 0.56 0.62 0.02
AK4 15 5.75 4.63 0.56 0.64 0.07
AK6 9 4.75 4.38 0.61 0.64 0.19
BC2 25 7.88 5.09 0.57 0.67 0.07
BC4 9 4.75 4.33 0.56 0.65 0.08
IR1 10 5.88 5.13 0.66 0.71 0.05
MB1 13 6.50 5.12 0.59 0.71 0.10
ME1 40 9.75 5.89 0.72 0.77 0.06
MI1 8 6.50 6.10 0.72 0.80 0.33
MN1 34 9.63 5.62 0.58 0.73 0.07
MN2 12 7.50 5.97 0.70 0.79 0.04
NB1 20 9.00 5.86 0.68 0.74 0.17
NWT1 9 6.50 5.83 0.69 0.74 0.11
NWT2 18 7.63 5.59 0.67 0.72 0.01
NY1 13 6.75 5.58 0.70 0.79 0.10
ON1 11 7.00 5.86 0.67 0.75 0.02
ON3 19 9.25 5.85 0.66 0.75 0.15
PA1 10 5.25 4.63 0.63 0.65 0.15
PA2 13 6.63 5.23 0.70 0.68 0.07
QC3 20 8.50 5.80 0.70 0.77 0.09
QC4 17 7.00 5.38 0.71 0.76 0.05
SK1 8 6.13 5.78 0.56 0.72 0.20
VT1 10 5.63 4.97 0.70 0.73 0.03
WV1 14 4.50 3.90 0.50 0.63 0.10
YK2 30 8.50 5.10 0.62 0.65 0.11
Boreal Total
N = 442
7.05 (1.50) 5.35 (0.61) 0.64 (0.06) 0.72 (0.06) 0.10 (0.07)
BC1 15 4.25 3.30 0.46 0.51 0.02
CA1 12 4.88 4.16 0.59 0.61 0.13
CA2 7 3.63 3.63 0.48 0.52 0.01
MT1 100 12.13 5.58 0.70 0.73 0.17
OR1 32 9.00 5.29 0.64 0.68 0.23
OR2 17 5.88 4.48 0.62 0.61 0.23
WA1 30 9.25 5.91 0.71 0.77 0.38
WA3 9 5.13 4.64 0.71 0.65 0.29
WA4 29 9.38 5.73 0.67 0.76 0.35
Greater
Pacific
Northwest
Total
N = 251
7.06 (2.94) 4.75 (0.94) 0.62 (0.09) 0.65 (0.10) 0.20 (0.13)
CO1 58 7.75 4.25 0.55 0.57 0.14
UT1 25 4.88 3.66 0.50 0.54 0.27
WY1 77 7.25 4.06 0.53 0.55 0.10
U.S. Rockies Total
N = 160
6.63 (1.53) 3.99 (0.30) 0.53 (0.03) 0.55 (0.02) 0.17 (0.09)
N, number of individuals; A, number of different alleles; AR, allelic richness; HO, observed heterozygosity; HE, expected heterozygos-
ity; PAR, population private allelic richness.
Cluster averages and standard deviations (in parentheses) are italicized.
© 2014 John Wiley & Sons Ltd
2934 E. CHENG ET AL.
Mitochondrial DNA analysis
The final data set for CR analyses comprised 893 snow-
shoe hare samples represented by 365 haplotypes. For
phylogenetic tree construction, the subset of 80 snow-
shoe hare Cytb sequences comprised 43 haplotypes.
The best-fit model of nucleotide substitution for Cytb
phylogeny was HKY+G. Three highly divergent lin-
eages were identified (Fig. 4), broadly corresponding
with the major genetic clusters identified in the first
two rounds of STRUCTURE analysis of microsatellites
(Fig. 1 and Appendix S4, Supporting information).
MtDNA analysis also identified two sublineages (with
>95% posterior probability) that corresponded with the
finer scale splitting of southwestern populations from
hierarchical STRUCTURE analysis: (i) WA3 in Olympic
National Park, Washington, was a sublineage of the
Greater Pacific Northwest lineage; and (ii) CO1 in
Gunnison, Colorado, was a sublineage of the U.S. Rock-
ies lineage. Within the Boreal lineage, a basal group
comprised samples from populations near the lineage’s
southern range (MT1, Montana; MB1, Manitoba; ON1,
Ontario; Fig. 4). The Cytb topology indicated that
snowshoe hares in the Greater Pacific Northwest line-
age are more closely related to black-tailed jackrabbits
than to other snowshoe hare populations.
Divergence estimation between the Boreal and U.S.
Rockies snowshoe hare lineages is 1.30 mya (95% CI
0.88–1.74 mya). The CO1 sublineage split off from the
U.S. Rockies major lineage more recently (0.78 mya;
95% CI 0.42–1.18 mya). The clade comprising most of
the snowshoe hares in the Greater Pacific Northwest
lineage diverged from BTJR about 0.59 mya (95% CI
0.34–0.90 mya; Fig. 4). The Bayesian skyline plot pro-
vided strong evidence of a recent demographic expan-
sion of the Boreal lineage (Fig. 5), whereas expansion of
the Greater Pacific Northwest and U.S. Rockies groups
was not supported.
NETWORK and SAMOVA analyses, based on the CR
gene, accorded with the Cytb phylogeny. In an unroot-
ed median-joining network, the Cytb lineages and sub-
lineages were reciprocally monophyletic and separated
from each other by ≥10 CR base pair substitutions
(Appendix S9, Supporting information). By these crite-
ria, the OR2 population of Malheur National Forest,
35 40 45 50 55 60 65 70
0.5
0.6
0.7
0.8
0.9
Exp
ecte
d he
tero
zygo
sity
Latitude (°N)
Boreal Greater Pacific Northwest U.S. Rockies
−180 −160 −140 −120 −100 −80 −60 −40
0.5
0.6
0.7
0.8
0.9
Exp
ecte
d he
tero
zygo
sity
Longitude (°W)
Fig. 2 For each sampled population, expected heterozygosity
plotted against latitude (top) and longitude (bottom). The grey
vertical bar marks the approximate southernmost latitude of
the LGM.
35 40 45 50 55 60 65 70
0.0
0.1
0.2
0.3
0.4
Priv
ate
alle
lic ri
chne
ss
Latitude (°N)
−180 −160 −140 −120 −100 −80 −60 −40
0.0
0.1
0.2
0.3
0.4
Priv
ate
alle
lic ri
chne
ss
Longitude (°W)
Boreal Greater Pacific Northwest U.S. Rockies
Fig. 3 For each sampled population, private allelic richness
(PAR) plotted against latitude (top) and longitude (bottom).
The grey vertical bar marks the approximate southernmost lati-
tude of the LGM.
© 2014 John Wiley & Sons Ltd
SNOWSHOE HARE LANDSCAPE GENETICS AND CONSERVATION 2935
Oregon, was also identified as a distinct sublineage of
the U.S. Rockies lineage.
Differences between the Greater Pacific Northwest
lineage and all other snowshoe hare lineages explained
77% of total genetic variation in the CR gene (SAM-
OVA; Dupanloup et al. 2002). If the Greater Pacific
Northwest snowshoe hare lineage is introgressed from
black-tailed jackrabbits, this deep genetic division when
K = 2 is an artefact of interspecific hybridization rather
than snowshoe hare demographic history. Therefore,
we also analysed the data with the Greater Pacific
Northwest lineage excluded. In this analysis, 66% of
variation was explained by differences between four
groups (K = 4, P < 0.001): the Boreal lineage, U.S. Rock-
ies lineage, and the sublineages CO1 and OR2. Control
region pairwise FST was high between and within all
lineages except within the Boreal lineage (Appendix
S10, Supporting information).
Only two populations (MT1 and WA4; Fig. 1) con-
tained haplotypes that could be ascribed to more than
one lineage, with both showing ties to Boreal and
Greater Pacific Northwest lineages. Most populations
were characterized by high haplotype and nucleotide
diversities, with exceptions in British Columbia (BC1,
near Vancouver) and in southern populations (CA1 and
CA2 in California; UT1 in Utah; WV1 in West Virginia;
Appendix S11, Supporting information). Haplotype and
nucleotide diversities increased with latitude up to
~49°N latitude (Appendices S12 and S13, Supporting
information). MtDNA diversity did not exhibit any
clear longitudinal pattern.
Discussion
In this range-wide study of a species distributed across
boreal North America, we found that snowshoe hares
formed three major genetic groups with well-defined
distributions, coincident with patterns observed in other
North American boreal species (Arbogast 1999; van Els
et al. 2012). The entire northern and eastern range of the
Fig. 4 Snowshoe hare phylogenetic relationships, as constructed in BEAST version 1.7.4 (Drummond et al. 2012), based on the Cytb
gene. Asterisks (*) indicate lineages and sublineage divisions with ≥95% posterior probability support. Map shows the geographic
distribution of the three major Cytb lineages. Filled symbols identify sublineages.
© 2014 John Wiley & Sons Ltd
2936 E. CHENG ET AL.
snowshoe hare, spanning 6000 km across Canada and
the eastern U.S., constituted a single Boreal group char-
acterized by high genetic diversity and gene flow. Two
geographically confined groups―in the Greater Pacific
Northwest and U.S. Rockies―exhibited lower gene flow
and high genetic uniqueness. The three major groups
are coherent from microsatellite and mtDNA analyses.
Both markers further identified genetic subdivision
within the Greater Pacific Northwest and U.S. Rockies,
of which the separation of CO1 (Colorado) from the
U.S. Rockies group was congruent across all markers
and analyses. Modern populations of snowshoe hares
likely derived from refugial populations that persisted
through the Quaternary ice ages in eastern and south-
ern refugia.
We found high genetic diversity in most sampled
populations, but reduced diversity at current range
edges, especially for populations at the species’ frag-
mented southern edge. Southern range populations
below 49°N had high genetic uniqueness with minimal
gene flow with their northern neighbours, suggesting
snowshoe hares could lose considerable genetic diver-
sity if southern boreal habitats are lost.
Evolutionary history and refugial origin
This work revealed strong genetic structure at different
hierarchical levels and a remarkable coincidence of the
inference of three major geographically explicit groups
of snowshoe hares based on mtDNA sequences and mi-
crosatellite data. Further, these markers also coincide in
the suggestion of additional genetic fragmentation in
the species’ southwestern range. Based on the mtDNA
phylogeny, we estimated divergence of the three major
groups to be 1.30–0.78 mya, long before the height of
the LGM ~18 kya. Regional mixing among groups was
sufficiently low during subsequent interglacial warm
periods, including the current one, that deep genetic
divisions are still maintained in the mtDNA and micro-
satellite data.
Many co-occurring forest species for which continent-
wide genetic data are available share this phylogeo-
graphic pattern―a large genetic cluster across Canada
and the eastern U.S. and one or more smaller genetic
clusters in the western USA. Examples include the gray
jay (Perisoreus canadensis, van Els et al. 2012), northern
flying squirrel (Glaucomys spp., Arbogast 1999), black
bear (Ursus americanus, Wooding & Ward 1997) and
hairy woodpecker (Picoides villosus, Klicka et al. 2011).
Genetic groups in these species diverged an estimated
2.97–0.69 mya, a range that encompasses our diver-
gence estimates for the snowshoe hare.
Boreal snowshoe hare lineage. The height of the LGM in
North America occurred ~18 kya, and by 6 kya, the gla-
ciers had largely disappeared (Pielou 1991). Given the
Boreal lineage diverged from other snowshoe hare lin-
eages an estimated 1.30 mya, colonization of newly
available boreal habitats after the LGM must have
occurred primarily from refugial populations within the
Boreal lineage. We had hypothesized a Beringian refu-
gium for snowshoe hares, as reported for several other
North American boreal species (Shafer et al. 2010b), but
we did not find genetic diversity or uniqueness patterns
indicative of a major Beringian refugium for snowshoe
hares. A few snowshoe hare fossils are documented
from Alaska and Yukon from 20 to 10 kya, but the
majority of hare fossils from this period are from the
lower 48 U.S. states (FAUNMAP Working Group 1994).
Thus, any relict snowshoe hare populations that sur-
vived the LGM in Beringia may have been too small or
isolated to be heavily represented in contemporary
snowshoe hare genetic structure.
Instead, genetic patterns suggest that the Boreal line-
age primarily expanded from refugia near the southern
0 20 40 60 80
0.0
0.5
1.0
1.5
2.0
2.5
Pop
ulat
ion
size
BOREAL
0 20 40 60 80
0.0
0.5
1.0
1.5
2.0
2.5
Pop
ulat
ion
size
GREATER PACIFIC NORTHWEST
0 20 40 60 80
0.0
0.5
1.0
1.5
2.0
2.5
Pop
ulat
ion
size
U.S. ROCKIES
Time from present (kya)
Fig. 5 Bayesian skyline plots (BEAST version 1.7.4; Drummond
et al. 2012) for three snowshoe hare genetic lineages. Relative
population sizes are in units of Ne x mutation rate. Grey lines
represent the 95% CI.
© 2014 John Wiley & Sons Ltd
SNOWSHOE HARE LANDSCAPE GENETICS AND CONSERVATION 2937
edge of the ice sheets and from eastern refugia. This
idea is supported by the basal position of snowshoe
hare Cytb haplotypes sampled from locations close to
current southern limits of the Boreal lineage. In addi-
tion, microsatellite diversity was highest in eastern pop-
ulations of the Boreal lineage. Results are consistent
with fossil pollen data, which indicate that at the LGM,
boreal forests in North America persisted in at least
two major pockets—the Pacific Northwest and the
southeastern USA (Williams et al. 1993).
The overall high genetic diversity through much of
the Boreal lineage, and the significant pattern of IBD
across the Boreal lineage range, suggests cross-continen-
tal expansion may have proceeded slowly or from a
broad refugial front (Hewitt 1996). Signals of demo-
graphic expansion revealed in the Bayesian skyline plot
indicate Boreal lineage expansion may have begun
~48 000 years ago.
Greater Pacific Northwest and U.S. Rockies snowshoe hare
lineages. The Greater Pacific Northwest and U.S. Rock-
ies lineages occur in the species’ southwest range,
which was largely ice free during the Pleistocene. The
high genetic uniqueness and strong genetic subdivisions
in mtDNA of these lineages indicate that they arose
from at least two discrete refugia. Comparative phylog-
eographic studies have identified the northwestern USA
as an area of exceptionally high genetic differentiation
for boreal and temperate species, due to the complex
physiography of the region and its relative stability as a
glacial refugium (Soltis et al. 1997; Swenson & Howard
2005; Shafer et al. 2010b). Our Bayesian estimates of
temporal fluctuations of effective population size sug-
gested these evolutionary groups remained relatively
stable through evolutionary time (Fig. 5).
The mtDNA analyses indicate the Greater Pacific
Northwest snowshoe hares are more closely related to
black-tailed jackrabbits than to other snowshoe hares.
Two competing hypotheses may explain this result: (i)
mitochondrial DNA introgression (through hybridiza-
tion) from Lepus californicus into Lepus americanus in the
southwestern range of the latter or (ii) retention of an
ancestral polymorphism shared between the two spe-
cies (Moore 1995). Extensive mtDNA introgression
occurs among other species of hares, resulting from
ancient or current contacts among species and some-
times causing extensive replacements of lineages (Alves
et al. 2008). Even though the geographic restriction of
the shared variants and the remarkably close phyloge-
netic relationship with current L. californicus variants
support the introgression hypothesis, the inference of
such phenomena would require reconstruction of the
speciation history of the taxa, using genealogical infor-
mation from nuclear loci (Melo-Ferreira et al. 2012). It is
nevertheless important to note that if introgression
caused this interspecific sharing of lineages, it was
remarkably pervasive and may have completely
replaced the mtDNA variation in the Greater Pacific
Northwest evolutionary group identified using micro-
satellites.
Genetic diversity of core vs. peripheral populations
Genetic diversity of snowshoe hares was highest in
mid-latitude populations, near the southernmost edge
of the LGM. From here, diversity clearly decreased
towards the south and less dramatically towards the
north. The southern range edge for snowshoe hares is
highly impacted by natural and anthropogenic habitat
fragmentation (Hansen et al. 2010). The observed
genetic pattern is consistent with the core-periphery
hypothesis, with populations in the fragmented south-
ern periphery exhibiting the lowest genetic diversity
and gene flow. Further, high amplitude population fluc-
tuations may promote gene flow and genetic diversity
(Ehrich et al. 2009). Snowshoe hare populations across
their northern range undergo large population cycles,
whereas southern populations may have reduced cyclic-
ity (Hodges 2000).
Anticipated genetic consequences of southernpopulation loss
Our study provides important insights on how potential
loss of southern hare populations (below 49°N) may
affect genetic diversity. The strong genetic subdivisions
and uniqueness of snowshoe hare populations suggest
that anticipated fragmentation and loss of these habitats
due to climate change and human activities may greatly
reduce overall species genetic diversity, with possible
negative implications for future adaptive potential. For
example, we identified at least three snowshoe hare
evolutionarily significant units (ESUs), using Moritz’s
(1994) criteria of reciprocal monophyly for mtDNA and
significant divergence in the frequencies of nuclear
alleles. Two ESUs occurred wholly in the species’
southern range. Three snowshoe hare sublineages,
reciprocally monophyletic and separated from each
other by ≥10 CR base pair substitutions, were also
found in the southern range. Additional ESUs may
occur in parts of the southern range not sampled in this
study: for example, we did not sample hares in New
Mexico or in northern Idaho, an area with high ende-
mism hypothesized to be the ‘Clearwater refugium’
(Daubenmire 1975; Soltis et al. 1997).
A limitation of this study is its reliance on neutral
genetic variation, without complementary information
on adaptive potential, for identifying ESUs (Funk et al.
© 2014 John Wiley & Sons Ltd
2938 E. CHENG ET AL.
2012). We identified ESUs on the basis of Moritz’s defi-
nition because it can be operationally applied from neu-
tral genetic markers (de Guia & Saitoh 2007). Other
definitions of ESU emphasize conserving adaptive vari-
ation, by incorporating adaptive genetic variation, life
history traits, morphology and species distribution
(Ryder 1986; Vogler & DeSalle 1994). An additional
question that should be addressed is, ‘How much
would loss of southern populations impact the species’
ability to adapt to global warming?’ Such studies would
require evaluation of quantitative genetic trait variations
directly linked to traits with adaptive value under
altered climate regimes. For snowshoe hares, adaptive
variation may include phenology of seasonal coat col-
our moult confronting decreased snow pack, especially
in the southern part of the range (Mills et al. 2013).
Concomitant with the predominantly southern distri-
bution of ESUs and sublineages, a large proportion of
snowshoe hares’ neutral private allelic richness (PAR)
occurs in the U.S. Rockies and Greater Pacific North-
west, where isolation and relative stability over
evolutionary time were likely responsible for their accu-
mulation of mutations and unique genetic structure. On
average, populations in the U.S. Rockies and Greater
Pacific Northwest lineages had almost twice the PAR of
populations in the Boreal lineage. In contrast to the
highly connected Boreal populations, loss of a popula-
tion in the U.S. Rockies and Greater Pacific Northwest
lineages could mean complete loss of many unique
alleles. At neutral markers, this loss would not be a
major conservation concern, but it portends an analo-
gous loss of diversity at evolutionarily significant loci.
The high genetic structure and uniqueness in the
southern range of the snowshoe hare reflect a common
phylogeographic pattern among North American spe-
cies. Regional comparative studies emphasize that the
Pacific Northwest and U.S. Rockies are hotspots of
genetic diversity for many species (Soltis et al. 1997;
Swenson & Howard 2005; Shafer et al. 2010a). Although
there are few rangewide studies for boreal species, they
typically corroborate the cryptic genetic distinctiveness
of these southern populations in the context of the spe-
cies’ entire North American range (Wooding & Ward
1997; Arbogast 1999; Arbogast & Kenagy 2001; Klicka
et al. 2011; van Els et al. 2012). Collectively, these find-
ings support Hampe and Petit’s (2005) call for prioritiz-
ing conservation of southern edge populations of boreal
species.
For snowshoe hares and many other boreal species in
North America, southern populations may already be
losing genetic diversity due to anthropogenic change
such as habitat fragmentation (desert bighorn sheep,
Ovis canadensis nelsoni; Epps et al. 2005) and climate
change (alpine chipmunk, Tamias alpinus; Rubidge et al.
2012). The range of snowshoe hares has contracted
northward throughout the previous century, primarily
related to habitat loss and conversion, with potential
contributions from harvest and climate change (Hodges
2000; NatureServe 2014). Populations in West Virginia,
North Carolina, Tennessee and Virginia have declined.
Snowshoe hares are extirpated from Ohio, New Jersey
and North Carolina and possibly extirpated from Mary-
land (NatureServe 2014). They are listed as critically
imperilled (S1) in Virginia, imperilled (S2) in New Mex-
ico and vulnerable (S3) in Pennsylvania, Utah and
Nevada. In California, the subspecies L. a. tahoensis is a
state-listed Species of Special Concern.
In the face of certain climate change with uncertain
impacts, it is difficult to predict how species conserva-
tion efforts can best be prioritized to maximize long-
term persistence. Although we cannot anticipate the
unforeseen, we can use our understanding of the pres-
ent to heed the advice of geneticist Otto Frankel (1974)
that ‘at this point of decision-making it may be our evo-
lutionary responsibility to keep evolutionary options
open so far as we can’. Using historical processes as a
guide, an emphasis on conserving southern edge popu-
lations seems prudent for this strongly interacting prey
species.
Acknowledgements
We thank the Canadian and U.S. hunting and trapping com-
munity, agency biologists, university researchers and private
citizens for donating more than 1000 snowshoe hare and black-
tailed jackrabbit genetic samples for this study. We are espe-
cially grateful to N. Berg, S. Carriere, J. Ivan, H. Jolicoeur, J.
MacCracken, B. McIntosh and P. Zevit for their efforts to fill in
critical sampling gaps. Biologists with the U.S. Forest Service
and state agencies provided invaluable help with permitting
and field logistics. Special thanks to D. Wager, J. Wager, C.
Brown and M. Strauser for their dedicated assistance in the
field and laboratory. This work was funded by the National
Science Foundation Grant 0817078, Natural Sciences and Engi-
neering Research Council (Canada), U.S. National Park Service,
and University of Montana, the Portuguese Fundac~ao para a
Ciencia e a Tecnologia (FCT) and the FEDER European Social
Fund (PTDC/BIA-EVF/115069/2009). J.M.-F. and PCA were
funded by Portuguese Foundation for Science and Technology
grants (SFRH/BPD/43264/2008 and SFRH/BSAB/1278/2012,
respectively, cofunded by the European Social Fund) and
Luso-American Development Foundation (FLAD).
References
Alves PC, Ferrand N, Suchentrunk F, Harris DJ (2003) Ancient
introgression of Lepus timidus mtDNA into L. granatensis and
L. europaeus in the Iberian Peninsula. Molecular Phylogenetics
and Evolution, 27, 70–80.
Alves PC, Melo-Ferreira J, Branco M et al. (2008) Evidence for
genetic similarity of two allopatric European hares (Lepus
© 2014 John Wiley & Sons Ltd
SNOWSHOE HARE LANDSCAPE GENETICS AND CONSERVATION 2939
corsicanus and L-castroviejoi) inferred from nuclear DNA
sequences. Molecular Phylogenetics and Evolution, 46, 1191–1197.
Arbogast BS (1999) Mitochondrial DNA phylogeography of
the new world flying squirrels (Glaucomys): implications
for pleistocene biogeography. Journal of Mammalogy, 80, 142–155.
Arbogast BS, Kenagy GJ (2001) Comparative phylogeography
as an integrative approach to historical biogeography. Journal
of Biogeography, 28, 819–825.Baker AJ, Marshall HD (1997) Mitochondrial control region
sequences as tools for understanding evolution. In: Avian
Molecular Evolution and Systematics (ed. Mindell DP), pp. 51–
82. Academic Press, San Diego, California.
Benjamini Y, Hochberg Y (1995) Controlling the false discovery
rate: a practical and powerful approach to multiple testing.
Journal of the Royal Statistical Society, Series B, 57, 289–300.
Brown WM, George JM, Wilson AC (1979) Rapid evolution of
animal mitochondrial DNA. Proceedings of the National Acad-
emy of Sciences, USA, 76, 1967–1971.Burton C, Krebs CJ, Taylor EB (2002) Population genetic struc-
ture of the cyclic snowshoe hare (Lepus americanus) in south-
western Yukon, Canada. Molecular Ecology, 11, 1689–1701.
Coulon A, Fitzpatrick JW, Bowman R et al. (2008) Congruent
population structure inferred from dispersal behaviour and
intensive genetic surveys of the threatened Florida scrub-jay
(Aphelocoma coerulescens). Molecular Ecology, 17, 1685–1701.
Dalquest WW (1942) Geographic variation in northwestern
snowshoe hares. Journal of Mammalogy, 23, 166–183.Darriba D, Taboada GL, Doallo R, Posada D (2012) jModelTest
2: more models, new heuristics and parallel computing. Nat-
ure Methods, 9, 772.
Daubenmire R (1975) Floristic Plant Geography of Eastern Wash-
ington and Northern Idaho. Brigham Young University Press,
Provo, Utah.
Drummond AJ, Suchard MA (2010) Bayesian random local
clocks, or one rate to rule them all. BMC Biology, 8, 114.
Drummond AJ, Rambaut A, Shapiro B, Pybus OG (2005)
Bayesian coalescent inference of past population dynamics
from molecular sequences. Molecular Biology and Evolution,
22, 1185–1192.Drummond AJ, Suchard MA, Xie D, Rambaut A (2012) Bayes-
ian phylogenetics with BEAUti and the BEAST 1.7. Molecular
Biology and Evolution, 29, 1969–1973.
Dupanloup I, Schneider S, Excoffier L (2002) A simulated
annealing approach to define the genetic structure of popula-
tions. Molecular Ecology, 11, 2571–2581.Earl DA, vonHoldt BM (2011) STRUCTURE HARVESTER: a
website and program for visualizing STRUCTURE output
and implementing the Evanno method. Conservation Genetics
Resources, 4, 359–361.Eckert CG, Samis KE, Lougheed SC (2008) Genetic variation
across species’ geographical ranges: the central-marginal
hypothesis and beyond. Molecular Ecology, 17, 1170–1188.
Ehrich D, Yoccoz NG, Ims RA (2009) Multi-annual density
fluctuations and habitat size enhance genetic variability in
two northern voles. Oikos, 118, 1441–1452.van Els P, Cicero C, Klicka J (2012) High latitudes and high
genetic diversity: phylogeography of a widespread boreal
bird, the gray jay (Perisoreus canadensis). Molecular Phylogenet-
ics and Evolution, 63, 456–465.
Epps CW, Palsboll PJ, Wehausen JD et al. (2005) Highways
block gene flow and cause a rapid decline in genetic diver-
sity of desert bighorn sheep. Ecology Letters, 8, 1029–1038.
Evanno G, Regnaut S, Goudet J (2005) Detecting the number of
clusters of individuals using the software STRUCTURE: a
simulation study. Molecular Ecology, 14, 2611–2620.Excoffier L, Laval G, Schneider S (2005) Arlequin ver. 3.0: An
integrated software package for population genetics data
analysis. Evolutionary Bioinformatics Online, 1, 47–50.
FAUNMAP Working Group (1994) FAUNMAP: a database
documenting late Quaternary distributions of mammal spe-
cies in the United States. Illinois State Museum Scientific
Papers, 25, 1–690.
Frankel OH (1974) Genetic conservation: our evolutionary
responsibility. Genetics, 48, 53–65.
Funk WC, McKay JK, Hohenlohe PA, Allendorf FW (2012)
Harnessing genomics for delineating conservation units.
Trends in Ecology & Evolution, 27, 489–496.Gillis EA, Krebs CJ (1999) Natal dispersal of snowshoe hares
during a cyclic population increase. Journal of Mammalogy,
80, 933–939.
Godbout J, Beaulieu J, Bousquet J (2010) Phylogeographic
structure of jack pine (Pinus banksiana; Pinaceae) supports
the existence of a coastal glacial refugium in northeastern
North America. American Journal of Botany, 97, 1903–1912.
Gomez A, Lunt DH (2007) Refugia within refugia: patterns of
phylogeographic concordance in the Iberian Peninsula. In:
Phylogeography in Southern European Refugia: Evolutionary Per-
spectives on the Origins and Conservation of European Biodiver-
sity (eds Weiss S, Ferrand N), pp. 155–188. Kluwer Academic
Publishers, Dordrecht, The Netherlands.
de Guia APO, Saitoh T (2007) The gap between the concept
and definitions in the Evolutionarily Significant Unit: the
need to integrate neutral genetic variation and adaptive vari-
ation. Ecological Research, 22, 604–612.Guillot G (2008) Inference of structure in subdivided popula-
tions at low levels of genetic differentiation. The correlated
allele frequencies model revisited. Bioinformatics, 24, 2222–
2228.
Guillot G, Mortier F, Estoup A (2005) Geneland: a program for
landscape genetics. Molecular Ecology Notes, 5, 712–715.Hampe A, Petit RJ (2005) Conserving biodiversity under climate
change: the rear edge matters. Ecology Letters, 8, 461–467.Hansen MC, Stehman SV, Potapov PV (2010) Quantification of
global gross forest cover loss. Proceedings of the National Acad-
emy of Sciences, USA, 107, 8650–8655.
Hewitt GM (1996) Some genetic consequences of ice ages, and
their role in divergence and speciation. Biological Journal of
the Linnean Society, 58, 247–276.Hewitt GM (2000) The genetic legacy of the Quaternary ice
ages. Nature, 405, 907–913.Hodges KE (2000) Ecology of snowshoe hares in southern bor-
eal and montane forests. In: Ecology and Conservation of Lynx
in the United States (eds Ruggiero LF, Aubry KB, Buskirk SW
et al.), pp. 163–206. University Press of Colorado, Boulder.
IPCC (2007) Climate Change 2007: Impacts, Adaptation, and
Vulnerability. Cambridge University Press, Cambridge, UK.
Johnson PCD, Haydon DT (2007) Maximum-likelihood estima-
tion of allelic dropout and false allele error rates from
microsatellite genotypes in the absence of reference data.
Genetics, 175, 827–842.
© 2014 John Wiley & Sons Ltd
2940 E. CHENG ET AL.
Kalinowski ST (2005) HP-RARE 1.0: a computer program for
performing rarefaction on measures of allelic richness. Molec-
ular Ecology Notes, 5, 187–189.
Klicka J, Spellman GM, Winker K, Chua V, Smith BT (2011) A
phylogeographic and population genetic analysis of a wide-
spread, sedentary North American bird: the hairy wood-
pecker (Picoides villosus). The Auk, 128, 346–362.
Korstanje R, Gillissen GF, Versteeg SA et al. (2003) Mapping of
rabbit microsatellite markers using chromosome-specific
libraries. Journal of Heredity, 94, 161–169.Koven CD (2013) Boreal carbon loss due to poleward shift in
low-carbon ecosystems. Nature Geoscience, 6, 452–456.Krebs CJ, Boonstra R, Boutin S, Sinclair ARE (2001) What drives
the 10-year cycle of snowshoe hares? BioScience, 51, 25–35.Librado P, Rozas J (2009) DnaSP v5: a software for comprehen-
sive analysis of DNA polymorphism data. Bioinformatics, 25,
1451–1452.
Mantel N (1967) The detection of disease clustering and a gen-
eralized regression approach. Cancer Research, 27, 209–220.
Matthee CA, van Vuuren BJ, Bell D, Robinson TJ (2004) A
molecular supermatrix of the rabbits and hares (Leporidae)
allows for the identification of five intercontinental exchanges
during the Miocene. Systematic Biology, 53, 433–447.
Meirmans PG (2012) The trouble with isolation by distance.
Molecular Ecology, 21, 2839–2846.
Melo-Ferreira J, Boursot P, Suchentrunk F, Ferrand N, Alves
PC (2005) Invasion from the cold past: extensive introgres-
sion of mountain hare (Lepus timidus) mitochondrial DNA
into three other hare species in northern Iberia. Molecular
Ecology, 14, 2459–2464.
Melo-Ferreira J, Boursot P, Randi E et al. (2007) The rise and
fall of the mountain hare (Lepus timidus) during Pleistocene
glaciations: expansion and retreat with hybridization in the
Iberian Peninsula. Molecular Ecology, 16, 605–618.
Melo-Ferreira J, Boursot P, Carneiro M et al. (2012) Recurrent
introgression of mitochondrial DNA among hares (Lepus
spp.) reveated by species-tree inference and coalescent simu-
lations. Systematic Biology, 61, 367–381.
Mills LS, Zimova M, Oyler J et al. (2013) Camouflage mismatch
in seasonal coat color due to decreased snow duration. Pro-
ceedings of the National Academy of Sciences, USA, 110, 7360–7365.
Moore WS (1995) Inferring phylogenies from mtDNA variation:
mitochondrial-gene trees versus nuclear-gene trees. Evolution,
49, 718–726.Moritz C (1994) Defining evolutionarily significant units for
conservation. Trends in Ecology & Evolution, 9, 373–375.Moritz C, Patton JL, Conroy CJ et al. (2008) Impact of a century
of climate change on small-mammal communities in Yosem-
ite National Park, USA. Science, 322, 261–322.
Mougel F, Mounolou JC, Monnerot M (1997) Nine polymorphic
microsatellite loci in the rabbit, Oryctolagus cuniculus. Animal
Genetics, 28, 58–71.NatureServe (2014) NatureServe Explorer: An Online Encyclopedia
of Life [web Application[. Version 7.1. NatureServe, Arlington,
Virginia. http://explorer.natureserve.org
Nei M (1972) Genetic distance between populations. The Ameri-
can Naturalist, 106, 283–292.
Nei M (1978) Estimation of average heterozygosity and genetic
distance from a small number of individuals. Genetics, 89,
583–590.
van Oosterhout C, Hutchinson WP, Wills DPM, Shipley P
(2004) MICRO-CHECKER: software for identifying and cor-
recting genotyping errors in microsatellite data. Molecular
Ecology Notes, 4, 535–538.Peakall R, Smouse PE (2006) GENALEX 6: genetic analysis in
Excel. Population genetic software for teaching and research.
Molecular Ecology Notes, 6, 288–295.
Pielou EC (1991) After the Ice Age: The Return of Life to Glaci-
ated North America. University of Chicago Press, Chicago,
Illinois.
Pierpaoli M, Riga F, Trocchi V, Randi E (1999) Species distinc-
tion and evolutionary relationships of the Italian hare (Lepus
corsicanus) as described by mitochondrial DNA sequencing.
Molecular Ecology, 8, 1805–1817.Powers RP, Coops NC, Morgan JL et al. (2012) A remote sens-
ing approach to biodiversity assessment and regionalization
of the Canadian boreal forest. Progress in Physical Geography,
37, 36–62.Pritchard JK, Stephens M, Donnelly P (2000) Inference of popu-
lation structure using multilocus genotype data. Genetics,
155, 945–959.
Provan J, Bennett KD (2008) Phylogeographic insights into cryp-
tic glacial refugia. Trends in Ecology & Evolution, 23, 564–571.
Raymond M, Rousset F (1995) An exact test for population dif-
ferentiation. Evolution, 49, 1283–1286.
Rico C, Rico I, Webb N, Smith DB, Hewitt G (1994) Four poly-
morphic microsatellite loci for the European wild rabbit,
Oryctolagus cuniculus. Animal Genetics, 25, 367.
Rousset F (1997) Genetic differentiation and estimation of gene
flow from F-statistics under isolation by distance. Genetics,
145, 1219–1228.Rousset F (2008) Genepop’007: a complete reimplementation of
the Genepop software for Windows and Linux. Molecular
Ecology Resources, 8, 103–106.
Rubidge EM, Patton JL, Lim M et al. (2012) Climate-induced
range contraction drives genetic erosion in an alpine mam-
mal. Nature Climate Change, 2, 285–288.Ryder OA (1986) Species conservation and systematics: the
dilemma of the subspecies. Trends in Ecology & Evolution, 1,
9–10.
Schwartz MK, Mills LM, Ortega Y, Ruggiero LF, Allendorf FW
(2003) Landscape location affects genetic variation of Canada
lynx (Lynx canadensis). Molecular Ecology, 12, 1807–1816.Schwartz MK, Pilgrim KL, McKelvey KS, Rivera PT, Ruggiero
LF (2007) DNA markers for identifying individual snowshoe
hares using field-collected pellets. Northwest Science, 81, 316–
322.
Shafer ABA, Cote SD, Coltman DW (2010a) Hot spots of
genetic diversity descended from multiple Pleistocene refu-
gia in an alpine ungulate. Evolution, 65, 125–138.
Shafer ABA, Cullingham CI, Cote SD, Coltman DW (2010b) Of
glaciers and refugia: a decade of study sheds new light on
the phylogeography of northwestern North America. Molecu-
lar Ecology, 19, 4589–4621.
Shugart HH, Leemans R, Bonan GB (2005) A Systems Analysis
of the Global Boreal Forest, pp. 565. Cambridge University
Press, Cambridge.
Soltis DE, Gitzendanner MA, Strenge DD, Soltis PS (1997)
Chloroplast DNA intraspecific phylogeography of plants
from the Pacific Northwest of North America. Plant Systemat-
ics and Evolution, 206, 353–373.
© 2014 John Wiley & Sons Ltd
SNOWSHOE HARE LANDSCAPE GENETICS AND CONSERVATION 2941
Soltis DE, Morris AB, McLachlan JS, Manos PS, Soltis PS (2006)
Comparative phylogeography of unglaciated eastern North
America. Molecular Ecology, 15, 4261–4293.
Strimmer K (2008) fdrtool: a versatile R package for estimating
local and tail area-based false discovery rates. Bioinformatics,
24, 1461–1462.Surridge AK, Bell DJ, Rico C, Hewitt GM (1997) Polymorphic
microsatellite loci in the European rabbit (Oryctolagus cunicu-
lus) are also amplified in other lagomorph species. Animal
Genetics, 28, 302–305.Swenson NG, Howard DJ (2005) Clustering of contact zones,
hybrid zones, and phylogeographic breaks in North Amer-
ica. American Naturalist, 166, 581–591.
Tracy LN, Jamieson IG (2011) Historic DNA reveals contempo-
rary population structure results from anthropogenic effects,
not pre-fragmentation patterns. Conservation Genetics, 12,
517–526.
Vigilant L, Stoneking M, Harpending H, Hawkes K, Wilson
AC (1991) African populations and the evolution of human
mitochondrial DNA. Science, 253, 1503–1507.Vogler AP, DeSalle R (1994) Diagnosing units of conservation
management. Conservation Biology, 8, 354–363.Weir BS, Cockerham CC (1984) Estimating F-statistics for the
analysis population structure. Evolution, 38, 1358–1370.Williams MAJ, Dunkerley DL, De DP, Kershaw AP, Stokes T
(1993) Quaternary Environments. Edward Arnold, New York.
Wooding S, Ward R (1997) Phylogeography and Pleistocene
evolution in the North American black bear. Molecular Biol-
ogy and Evolution, 14, 1096–1105.
E.C., L.S.M. and K.E.H. designed the research; E.C. con-
ducted field sampling and laboratory work; L.S.M. and
K.E.H. contributed some hare genetic samples; P.C.A.
and J.M.F. provided laboratory training and assistance;
E.C. and J.M.F. analysed data and wrote this work with
direction, assistance and editorial review from L.S.M.,
K.E.H. and P.C.A.
Data accessibility
All Cytb and CR sequences from this project have been
deposited in GenBank (Accession nos KF781351–
KF781437; KF804153–KF805042; HM771306–HM771308).
Sample details, sequence alignments, microsatellite geno-
types, and input files have been deposited in the Dryad
DigitalRepository (Conservation implicationsof theevolu-
tionary history and genetic diversity hotspots of the snow-
shoehare.http://dx.doi.org/10.5061/dryad.dh63p).
Supporting information
Additional supporting information may be found in the online
version of this article.
Appendix S1 Sources of genetic samples analysed in this
study.
Appendix S2 Microsatellite loci diversity.
Appendix S3 DNA extraction and genotyping methods.
Appendix S4 STRUCTURE hierarchical analysis results under
a model of admixture and uncorrelated allele frequencies.
Appendix S5 Analysis of the Boreal cluster in the second
round of STRUCTURE hierarchical analysis.
Appendix S6 Highest mean posterior density Geneland results
for a model with uncorrelated allele frequencies and null
alleles filtered.
Appendix S7 Pairwise FST and Nei’s D calculated across eight
microsatellite loci.
Appendix S8 Population allelic richness plotted against lati-
tude and longitude.
Appendix S9 Mitochondrial control region median-joining net-
work.
Appendix S10 Mitochondrial control region pairwise FST.
Appendix S11 Mitochondrial control region diversity statistics.
Appendix S12 Haplotype diversity plotted against latitude and
longitude.
Appendix S13 Nucleotide diversity plotted against latitude
and longitude.
© 2014 John Wiley & Sons Ltd
2942 E. CHENG ET AL.
6210 | Ecology and Evolution. 2017;7:6210–6219.www.ecolevol.org
Received: 25 January 2017 | Revised: 26 April 2017 | Accepted: 2 May 2017
DOI: 10.1002/ece3.3137
O R I G I N A L R E S E A R C H
Genetic sampling for estimating density of common species
Ellen Cheng1,2 | Karen E. Hodges3 | Rahel Sollmann4 | L. Scott Mills2
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.© 2017 The Authors. Ecology and Evolution published by John Wiley & Sons Ltd.
1Daniel B. Warnell School of Forestry and Natural Resources, University of Georgia, Athens, GA, USA2Wildlife Biology Program and Office of the Vice President for Research and Creative Scholarship, University of Montana, Missoula, MT, USA3Department of Biology, University of British Columbia Okanagan, Kelowna, BC, Canada4Department of Wildlife, Fish and Conservation Biology, University of California Davis, Davis, CA, USA
CorrespondenceEllen Cheng, Daniel B. Warnell School of Forestry and Natural Resources, University of Georgia, Athens, GA, USA.Email: [email protected]
Funding informationAnimal Welfare Institute; National Science Foundation, Grant/Award Number: 0841884, Division of Environmental Biology; National Park Service, Grant/Award Number: GLAC-2005-SCI-0047
AbstractUnderstanding population dynamics requires reliable estimates of population density, yet this basic information is often surprisingly difficult to obtain. With rare or difficult- to- capture species, genetic surveys from noninvasive collection of hair or scat has proved cost- efficient for estimating densities. Here, we explored whether noninvasive genetic sampling (NGS) also offers promise for sampling a relatively common species, the snowshoe hare (Lepus americanus Erxleben, 1777), in comparison with traditional live trapping. We optimized a protocol for single- session NGS sampling of hares. We compared spatial capture–recapture population estimates from live trapping to esti-mates derived from NGS, and assessed NGS costs. NGS provided population esti-mates similar to those derived from live trapping, but a higher density of sampling plots was required for NGS. The optimal NGS protocol for our study entailed deploy-ing 160 sampling plots for 4 days and genotyping one pellet per plot. NGS laboratory costs ranged from approximately $670 to $3000 USD per field site. While live trap-ping does not incur laboratory costs, its field costs can be considerably higher than for NGS, especially when study sites are difficult to access. We conclude that NGS can work for common species, but that it will require field and laboratory pilot testing to develop cost- effective sampling protocols.
K E Y W O R D S
density estimators, fecal pellets, noninvasive genetic sampling, snowshoe hare, spatial capture–recapture
1 | INTRODUCTION
Estimating animal density is central to most wildlife management and conservation decisions, to assess trend, evaluate numeric responses to natural and anthropogenic disturbances, and measure population per-sistence. Capture–mark–recapture (CMR) models applied to live- trap data have long been the standard for obtaining robust estimates of animal density (Pierce, Lopez, & Silvy, 2012; Pollock, Nichols, Brownie, & Hines, 1990). Unlike index- based sampling (e.g., track or pellet transects), live- trap data can be used to estimate detection probabil-ities from recapture histories of marked individuals, allowing statisti-cally rigorous estimates of animal density. However, live trapping is
invasive, is logistically daunting in remote areas, and can be difficult for some species.
Noninvasive genetic sampling (NGS) may be an effective alterna-tive to live trapping, to obtain data amenable to CMR analysis (Lukacs & Burnham, 2005; Mills, Citta, Lair, Schwartz, & Tallmon, 2000; Schwartz, Luikart, & Waples, 2007). NGS can be used to construct CMR capture histories through individual genotypes “captured” from noninvasively collected samples of scat or hair. NGS coupled with CMR has been used to estimate densities for many uncommon or difficult- to- capture species, including jaguars (Panthera onca Linnaeus, 1758; Sollmann et al., 2013), grizzly bear (Ursus arctos Linnaeus, 1758; Ciucci et al., 2015), and wolverine (Gulo gulo Linnaeus, 1758; Mulders,
| 6211CHENG Et al.
Boulanger, & Paetkau, 2007). For species that are not easily trapped, NGS often yields larger sample sizes and is more cost- effective than trapping efforts because scat and other sources of noninvasive DNA data are comparatively easy to collect (e.g., Hedges, Johnson, Ahlering, Tyson, & Eggert, 2013). The question we take up here is whether NGS can be as effective as live trapping for estimating density of relatively abundant and trappable species such as common raccoons (Procyon lotor Linnaeus, 1758), Virginia opossums (Didelphis virginiana Kerr, 1792), and snowshoe hares (Lepus americanus Erxleben, 1777); we use the latter as a tractable model for comparing live- capture results and NGS data.
For many species, the fieldwork needed to collect noninvasive samples is cheaper and faster than live trapping, when both methods are used to obtain data over the same number of temporal sessions. But NGS gains an additional (and often large) field cost advantage, especially with difficult- to- access sites, when the capture history for CMR analyses is built from DNA data collected during a single site visit. With live trapping, animals must be released from traps between each sampling session. In contrast, during a single NGS sampling ses-sion lasting days to weeks, animals can leave their genetic signatures at multiple “traps,” generating spatial capture histories analogous to the temporal capture histories from live trapping. Spatial capture–recapture (SCR; Efford, 2004; Borchers & Efford, 2008) models are extensions of CMR that can be used for analyzing spatial capture his-tories from single- session surveys.
Obtaining reliable population estimates with NGS depends not only on the number of genetic samples collected from the field, but also on the proportion of those samples that yield correct genotypes. To minimize genotyping errors and laboratory costs, field sampling must be designed to ensure genetic samples are fresh. Even then, two types of genotyping errors—false alleles and allelic dropout—can occur with the low quality and quantity of DNA available from NGS (Taberlet, Waits, & Luikart, 1999). Both error types can falsely inflate the number of unique individuals identified, leading to density esti-mates that are biased high (Waits & Leberg, 2000). These genotyping errors can be reduced through repeat amplification of genetic samples and reanalysis of samples with highly similar genotypes (Taberlet et al., 1996). Alternatively, density may be underestimated if the number and variability of molecular markers used in genotyping is inadequate to distinguish individuals, which may require sampling more loci to dif-ferentiate between even closely related individuals (Mills et al., 2000), again raising laboratory costs.
Laboratory costs for NGS studies are greater for common than rare species, because costs accrue based on the number of samples genotyped (Lukacs & Burnham, 2005; Marucco, Boitani, Pletscher, & Schwartz, 2011). Thus, NGS surveys for density estimation are primarily used for rare or difficult- to- capture species, where the method provides obvious advantages over live trapping. Few studies have assessed the reliability and cost of NGS surveys for common, readily catchable animals. To address this gap, we evaluated the cost- effectiveness of a single- session NGS approach for estimating den-sity of snowshoe hares, a common species in boreal forests of North America. We asked three questions to determine whether and when
this NGS approach is a viable alternative to live trapping: (1) Can we obtain sufficient genetic samples from the field while meeting single- session CMR assumptions of population closure and no un- modeled capture heterogeneity? (2) Can we achieve high genotyping success? (3) Are density estimates from live trapping and NGS comparable? We also consider the cost- effectiveness of the two methods for obtaining analyzable capture histories.
2 | MATERIALS AND METHODS
2.1 | Study species
Snowshoe hares are medium- sized herbivores widespread in montane and boreal forests of North America (Figure 1). In Canada and Alaska, the species’ northern range, population densities may reach six hares per ha during the high phase of 10- year population cycles (Hodges, 2000a). In these northern boreal forests, population sizes of Canada lynx (Lynx canadensis Kerr, 1792) and other major predators closely track the population cycles of snowshoe hares. In the contiguous United States, snowshoe hare populations are less cyclic and densi-ties rarely exceed 2.7 hares per ha (Hodges, 2000b). Snowshoe hares occupy overlapping home ranges that can cover 1.6–10.2 ha during the year. They are an important game species in many regions where they occur.
2.2 | Study area
This work was implemented in summers 2006 and 2009 at nine 20- ha sites in two study areas in Montana. Three sites were in Glacier National Park (Cheng, Hodges, & Mills, 2015) and six in Flathead National Forest west of Glacier NP (Hodges & Mills, 2008), both areas of more extensive sampling for other research questions. Five sites were used in a pilot study, and five sites (including one from the pilot study) were used in a field survey. We selected the sites to reflect a range of hare densities and forest types. Average monthly high tem-peratures for the study season ranged from 24° C to 27° C. Average monthly total precipitation ranged from 4.3 to 8.9 cm.
F IGURE 1 Snowshoe hare (Lepus americanus Erxleben, 1777). Photograph credit: Karen E. Hodges
6212 | CHENG Et al.
2.3 | Optimizing an NGS field protocol (pilot study)
First, we determined whether our NGS approach could yield sufficient fresh samples for reliable CMR- based population estimates. Because snowshoe hares deposit >500 pellets per day (Hodges, 1999), collec-tion from the forest floor would be difficult in terms of determining which pellets and how many to collect; aging pellets is subjective, pellets might be missed, and defining independent samples would be tricky. Instead, we established specific sampling stations where we deployed baited 0.5- m2 ground cloths that were left in the field for several days. Ground cloths ensured we obtained fresh hare pellets, which helped both with DNA amplification and with meeting the CMR assumption of “closure” of the population during sampling. Each 20- ha (400 × 500 m) study site was divided into an 8 × 10 grid with 50- m spacing between plots, for a total of 80 NGS plots per site. This sam-pling design echoed our survey method for live trapping snowshoe hares.
A hare can deposit multiple pellets during a single visit to an NGS plot. Therefore, on each NGS plot, only pellets from different hares (determined by pellet genotypes) can be considered indepen-dent captures for CMR analyses. In optimizing our NGS field proto-col, we therefore sought to increase the number of NGS plots with pellets, rather than to increase the number of pellets collected per plot. Independent captures from different plots could be increased by extending the number of sampling days, using attractive baits, or increasing the number of ground cloths per site. The first two options were evaluated during a pilot study, described below. The latter was assessed during our field survey, by doubling the number of sample plots at a subset of sites.
2.3.1 | Pellet accumulation pilot study
In summer 2006, we pretested survey methods at five pilot study sites. We examined impacts of sampling duration by counting how many NGS plots at each site had pellets after 1–5 days of sampling. To compare bait efficacy, at each site we randomly assigned one of three bait types (apples, oats, alfalfa) to each plot. We used a Kruskal–Wallis rank sum analysis to test for differences among bait types in the per-centage of plots that accumulated pellets.
2.3.2 | Pellet decomposition pilot study
Genotyping success declines with the time samples are left in the field, because DNA degradation increases with temperature, moisture, and exposure to ultraviolet radiation (Murphy, Kendall, Robinson, & Waits, 2007). An optimal NGS sampling duration would be long enough to collect many pellets, but short enough to ensure high genotyping suc-cess. Therefore, simultaneous with determining pellet accumulation rates in the field, we conducted a study of how quickly hare pellet DNA degrades with field exposure.
Degraded DNA can manifest as low PCR amplification success or high genotyping error rates. Estimating genotyping error rates often relies on comparison with reference genotypes from high- quality DNA
sources (Broquet & Petit, 2004). To quantify genotyping error rates for different- aged pellets, we collected an ear tissue sample (reference genotype) and pellets from each of 18 snowshoe hares captured at two pilot study sites. The pellets were obtained from the floors of live-traps that contained the captured hares. At the time of collection, these pellets could have been up to 12 hrs old (traps were open over-night). We transferred the pellets to a forest near our base camp. At 0, 2, 4, and 6 days postcapture, we selected up to three pellets from each hare’s pellet pile for genetic analysis. Once selected, pellets were stored in 95% alcohol in a −20°C freezer until extraction, which oc-curred within 6 months of collection.
We used Qiagen DNeasy Blood & Tissue Kits to extract DNA from the 18 hare tissue samples. The pellet samples were extracted with QIAamp DNA Stool Mini Kits, in a separate laboratory at University of Montana designated exclusively for low quality DNA samples collected noninvasively. All samples were genotyped at eight highly variable microsatellite loci originally developed in the European rab-bit, Oryctolagus cuniculus Linnaeus, 1758, and successfully used with snowshoe hares (Burton, Krebs, & Taylor, 2002; Cheng, Hodges, Melo- Ferreira, Alves, & Mills, 2014; Schwartz, Luikart et al, 2007; Schwartz, Pilgrim, McKelvey, Rivera, & Ruggiero, 2007). PCR amplifications were run on an ABI 3130xl Genetic Analyzer (Murdoch DNA Sequencing Facility; Missoula, MT) and scored with GENEMAPPER v. 3.7 (Applied Biosystems Inc., Foster City, CA, USA). We manually checked all mi-crosatellite genotypes to confirm allele calls. Sample extraction and PCR conditions are described in Cheng et al. (2014). PCR amplifica-tions of the eight loci were combined into three multiplex reactions, and each tissue and decomposition pellet sample was amplified four times across all loci.
For each pellet age class, we calculated nonamplification as the proportion of PCR attempts in which a sample locus did not yield any genotype. Nonamplification rates were averaged across all loci, sam-ples and PCRs. Using the corresponding tissue samples as reference genotypes, we calculated allelic dropout rates, false allele rates, and base shift error rates averaged across all loci, samples and PCRs for each pellet age class (Program GIMLET v1.3.3; Valiere, 2002). Allelic dropout was observed when a heterozygote was typed as a homozy-gote. A false allele occurred when slippage during PCR generated an additional erroneous allele. Base shift errors were small shifts in allele size—typically a one- base pair increase or decrease. Because calcula-tions used different denominators, error rates could not be summed to yield total error rate for each pellet age class (Broquet & Petit, 2004).
2.4 | Collecting live-trap and NGS data (field survey)
After optimizing an NGS protocol based on the pilot study, we con-ducted live trapping and NGS at two sites in Glacier NP in 2006 and at three sites in Flathead National Forest in 2009. At each site, live trap-ping and genetic sampling occurred sequentially within 2 weeks and in random order (i.e., live trapping then genetic sampling or the reverse). Significant differences in population estimates are attributable to dif-ferences in the survey methods rather than to changes in hare density within this short time.
| 6213CHENG Et al.
2.4.1 | Live trapping
At each site for the field survey, we placed 80 Tomahawk live-traps in the same 8 × 10 grid configuration described for the pilot study. Each site was trapped for three to five nights. Traps were opened every evening and baited with apple and alfalfa and then checked the fol-lowing morning. Captured hares were weighed, sexed, and ear tagged. We used sterile 3- mm biopsy punches to collect a small piece of ear tissue from each hare for genetic analyses. All hare handling was ap-proved by the University of Montana’s IACUC. Ear tissue samples were stored in silica gel until return from the field, at which point they were frozen to −20°C.
2.4.2 | Noninvasive genetic sampling
Noninvasive genetic sampling followed a refined protocol developed from our pilot work. At each of the 80 grid points, we baited a 0.5- m2 ground cloth with two to three commercially produced alfalfa cubes that were 2.5 to 5 cm per side. We returned 4 days later to collect all pellets that had accumulated on the ground cloths. At the three Flathead sites, we also tested the efficacy of using 160 cloths, placing the additional cloths halfway between the main sampling plots.
2.5 | Genotyping and error- checking genetic samples
We genotyped all tissue genetic samples collected from live- trapped hares. For each tissue sample, a “consensus genotype” was confirmed when two independent PCR amplifications yielded the same geno-type. From confirmed tissue genotypes, we estimated allele frequen-cies, heterozygosities, and probability of identity (PID and PIDsib) by locus, using Program GIMLET. PID is the probability that two individu-als (or siblings, for PIDsib) drawn from the population have identical confirmed genotypes. The PID and PIDsib were determined by multi-plying the eight locus- specific estimates.
Genotyping all pellets collected by NGS would have been expen-sive and likely redundant. Multiple pellets from one hare’s visit would be quite possible (Hodges & Sinclair, 2005). But if the multiple pel-lets on a single plot were instead from different hares, we would lose important capture information if we genotyped only one pellet per plot. As a compromise to minimize laboratory costs while maximizing potential capture rates, we genotyped up to four randomly selected pellets from each NGS plot. When plots had four or fewer pellets, all were genotyped.
For each NGS pellet selected, we determined a “consensus gen-otype” based on a stringent error- checking protocol. Pellet samples with <40% amplification success, across eight microsatellite loci, in the first two PCR runs were omitted from further analysis. We then used a three- stage approach and mismatch comparisons, modified from Waits and Paetkau (2005), to confirm a consensus genotype for each sample. A sample was designated a confirmed homozygote at a locus if it amplified as a clear homozygote in at least four PCRs with no discrepancies, and as a confirmed heterozygote if each allele amplified clearly in at least two PCRs with no discrepancies. If a genotype was
confirmed at all loci, or if the eight- locus genotype matched that of another genetic sample collected from the same site, the eight- locus genotype was the sample’s consensus genotype. Each sample was am-plified up to six times per locus.
2.6 | Comparing live-trap and NGS density estimates from field surveys
We applied maximum- likelihood spatial capture–recapture mod-els, implemented in the R package “secr” (Efford, 2016), to estimate densities separately from live- trap and NGS data for each study site. For SCR analysis, we assumed animal activity centers were distrib-uted according to a homogeneous point process and detection prob-ability followed a half- normal function. NGS plots were modeled as single occasion proximity detectors, which allowed individuals to be captured at multiple detectors during a survey occasion, but at each detector an individual was counted only once. We modeled live-traps as multicatch traps, a substitution that is minimally biased (Efford, 2016), because a single- catch likelihood is not available with maximum- likelihood estimation in “secr.” For live- trap data, we used Akaike’s Information Criterion for small sample sizes (AICc) to rank support for three detection models: The null model, a model that in-cluded capture- related behavioral effects (e.g., trap shyness), and a two- class mixture model allowing for two groups with different detec-tion probabilities (e.g., males vs. females or adults vs. juveniles). For the NGS data, we used AICc to rank support for the null and two- class mixture models. The behavioral model was not applicable because there was only one survey occasion. We set an integration buffer width of 600 m, which exceeds the minimum recommended width of three times the σ parameter (a spatial scale parameter that describes how quickly detection probability declines as the distance between a trap and an animal’s activity center increases) estimated from the data (Royle, Chandler, Sollman, & Gardner, 2013).
To examine how density estimates are influenced by the number of pellets genotyped per plot, we analyzed data from 500 iterations each of computer- generated subsampling of one or two pellets per plot, randomly selected from the four- pellet NGS dataset. We also tested three and four pellets per plot, but results were very similar to two pellets per plot, so are not reported. For each level of pellet sampling, the median of the density estimates and the median of the 95% confidence intervals from estimable iterations (capture histories with at least one recapture) were compared to estimates from live- trap data. When there are no recaptures, SCR model parameters are not identifiable.
2.7 | Comparing live trapping and NGS through simulation
To compare the accuracy and precision of estimates from live- trap versus NGS methods, for different hare densities and with different assumptions about the movement distances of hares, we simulated 500 iterations of 135 scenarios modeled after our study system. Using SCR model formulation, we simulated five sampling approaches: for
6214 | CHENG Et al.
live trapping, a single- trap detection model with four survey occa-sions, using the 80- plot trap grid of our empirical study (8 × 10 grid); for NGS, a proximity detection model with one survey occasion, using either an 80- or 160- plot grid and either one or two pellets sampled per plot. For each sampling approach, we simulated three hare densi-ties (0.2, 1.0, and 1.8 hares per hectare) × three detection probabilities at activity center (g0 = 0.05, 0.10, and 0.15) × three levels of σ (20, 50, and 80 m) with a half- normal detection function. The SCR parameters used in this simulation spanned the range of values estimated from 25 site-years of hare data collected from our long- term research in the Flathead NF (LS Mills & KE Hodges, unpublished data).
The 500 simulated capture histories for each scenario were an-alyzed as described above for the empirical data. We calculated the following summary metrics to compare density estimates from the simulated live-trap versus NGS approach:
1. Proportion of estimable iterations, measured as the proportion of 500 iterations with at least one recapture. Only estimable iterations were included in other summary metrics. If fewer than 10% of iterations were estimable for any simulated scenario, summary metrics were not calculated.
2. Root mean square error, a combined measure of bias and variance,
calculated as �
∑
( D−D)2
n, where D is estimated density, D is true den-
sity, and n is the number of iterations.3. Median coefficient of variation across iterations, with the coeffi-
cient of variation for each iteration calculated as the estimated standard error divided by estimated density.
4. Confidence interval coverage, which was the proportion of itera-tions in which the 95% confidence interval included the simulated (true) parameter value.
3 | RESULTS
3.1 | Optimizing an NGS field protocol (pilot study)
After 1 day of sampling, 13 ± 2.1% (SD) of plots at pilot study sites had hare pellets (Figure 2). The number of plots with pellets increased most rapidly during the first 2 days of sampling. By the fifth day of sampling, 37.2 ± 7.2% of plots had at least one pellet. Per- locus gen-otyping error rates increased with the number of days pellets were left in the field (Figure 3). Averaged across the eight loci, only a small percentage of 0- day- old pellets did not amplify at a locus (7%), but 71% of 6- day- old pellets did not amplify at a locus. When genotypes did amplify at a locus, allelic dropout was the most common type of genotyping error (Figure 3).
The three baits (apples, oats, and alfalfa) were similar at attracting hares to sampling plots (χ2 = 1.31, df = 2, p = .52). Alfalfa was easiest to handle and minimized disturbance to plots (most commonly, deer attracted to apples). Our final NGS protocol used 4 days of sampling with alfalfa as bait. The nonamplification rate for 4-day-old pellets was almost 50%, but a majority of the pellets collected after 4 sampling days would be <4 days old.
3.2 | Collecting live-trap and NGS data (field survey)
From 119 live- trap captures of snowshoe hares across the five study sites, we identified 72 unique individuals (Table 1). The average num-ber of live- trap captures per site was 23.8 (SD = 14.5), and the average number of unique individuals per site was 14.4 (SD = 10.0). We col-lected 488 snowshoe hare pellets from ground cloths (Table 2). Eighty percent of sampling plots had no hare pellets, but variation among sites was high. On average, 7% of sampling plots at each site had one pellet; 4%, two pellets; 2%, three pellets; and 7% had four or more pellets. At the three Flathead sites, the proportion of plots with pellets
F IGURE 2 Accumulation of snowshoe hare pellets on NGS sampling plots as a function of sampling duration. Each line represents a different 20- ha pilot site
0
20
40
60
1 2 3 4 5
Sampling duration (days)
% o
f plo
ts w
ith p
elle
ts
F IGURE 3 Per- locus genotyping error rate as a function of pellet age. Error rates are averaged across all loci, samples and PCRs. Nonamplification is when a sample- locus does not yield any genotype. Allelic dropout occurs when a heterozygote is typed as a homozygote. A false allele is an additional erroneous allele. Base shift errors are one- or two- base pair shifts in allele size
0
25
50
75
100
0 2 4 6
Pellet age (days)
Gen
otyp
ing
erro
r rat
e (%
)Non−amplification Allelic dropout False allele Base shift
| 6215CHENG Et al.
was similar when considering the 80 main plots versus all 160 plots. One site (Flathead 2) had 2 days of rain during the 4- day pellet accu-mulation period, but we did not observe a clear negative impact of rain on pellet numbers or PCR success (i.e., some other sites without rain had fewer pellets or lower PCR success than Flathead 2).
3.3 | Genotyping and error- checking genetic samples from field surveys
We obtained and successfully genotyped genetic samples from 85% (N = 61) of the 72 live- trapped hares. The 11 missing genetic sam-ples were primarily due to hares escaping before samples could be obtained. Based on tissue samples, average observed heterozygosity for the eight loci was 0.71 (SD = 0.19; Table 3). The number of alleles per locus ranged from four (SOL33) to 29 (SOL30). Overall PID, calcu-lated as the product of the locus- specific values, was 1.8 × 10−10 and PIDsib was 8.9 × 10−4.
Across the five study sites, 55% (N = 269) of the 488 pellets col-lected by NGS were genotyped, of which 210 (78%) amplified and yielded consensus genotypes. We conducted an average of 4.6 PCR runs per pellet. After error checking, 87% of pellet samples had geno-types that could be matched to another pellet or to a tissue genotype from a hare live-trapped at the same site. The consensus genotypes for three pellets (of 210) differed from another sample at only one locus. After additional independent PCRs, we concluded the consen-sus genotypes for these pellets represented unique individuals. For 3% of sampling plots, we confirmed pellet genotypes for two unique hares; for 1.1% of plots, three unique hares.
3.4 | Comparing live-trap and NGS density estimates
At all sites, for both live-trap and NGS data, the null model was the highest ranked SCR model, with at least 75% AICc weight. The 95% CI
for hare density estimates overlapped substantially among methods (Figure 4). Confidence intervals were usually largest for 80- plot NGS sampling. Unusually large 95% CI’s corresponded with recapture rates <20% (Table 4). Sampling one versus two pellets per plot for genotyp-ing had little influence on density estimates (Figure 4).
No SCR detection parameters were consistently higher or lower across sites for a particular sampling method (live- trap vs. NGS). Excluding Flathead2 80- plot NGS, the average g0 (detection probabil-ity at activity center) was 0.09 and ranged from 0.03 to 0.24 for all sites and methods. The parameter σ averaged 57.6 m and ranged from 22.1 to 98.4 m. SCR detection parameter estimates for Flathead2 80- plot NGS were unusual, with a very high g0 of 1.0 and small σ estimates (18.1–19.9), yet still yielding density estimates similar to live trapping.
3.5 | Comparing live trapping and NGS through simulation
In simulations, live trapping and 160- plot NGS produced more ac-curate and precise density estimates than did 80- plot NGS. As with the field data, the number of pellets genotyped had little influence on estimates. At the smallest σ (20 m), hare density often could not be es-timated, was biased low, or had a large 95% CI (Figure 5). Density es-timates were also poor when low detection (g0 = 0.05) resulted in few individuals captured or a low proportion of recaptures, even if simu-lated hare densities were moderate to high (Appendix S1). Regardless of survey method, when unique captures exceeded 20 individuals and at least 20% of captured animals were recaptured, density estimates were generally unbiased and close to true values (Appendix S1).
TABLE 1 Live- trap hare captures
SiteNumber of trap nights
Total number of captures
Number of unique hares captured
Glacier1 5 40 30
Glacier2 3 12 7
Flathead1 4 9 8
Flathead2 4 38 19
Flathead3 4 20 8
Site% of plots with pellets
Number of pellets collected
Number of pellets genotyped
% estimable genotypes
Number of unique hares
Glacier1 45 138 87 91 32
Glacier2 6 26 17 70 5
Flathead1 13/10 19/37 19/33 79/88 8/10
Flathead2 18/18 70/138 29/60 66/68 11/17
Flathead3 19/19 102/149 34/72 62/68 13/21
TABLE 2 NGS pellets collected and genotyped. For Flathead sites, the first value is based on 80 NGS plots; second, 160 NGS plots. We genotyped up to four randomly sampled pellets per plot. From a subset of genotyped pellets, we were able to obtain reliable consensus genotypes for individual identification (“% estimable genotypes”)
TABLE 3 Microsatellite diversities and probability of identity, by locus
Locus A Ho PID PIDsib
7L1D3 7 0.63 0.18 0.48
SAT02 26 0.86 0.01 0.29
SAT12 6 0.81 0.11 0.42
SAT13 6 0.67 0.20 0.49
SOL33 4 0.33 0.46 0.70
SAT16 8 0.62 0.11 0.43
SOL08 9 0.87 0.06 0.36
SOL30 29 0.87 0.00 0.28
A, number of different alleles; Ho, observed heterozygosity; PID, probabil-ity of identity; PIDsib, probability of identity for siblings.
6216 | CHENG Et al.
Root- mean- square error and coefficient of variation were often smaller with live trapping and 160- plot NGS compared to 80- plot NGS, except at the smallest σ simulated (Appendix S2). Root- mean- square error declined as density declined, and as g0 or σ increased. Coefficient of variation declined as density, g0, or σ increased. Confidence interval coverage was close to 95% except for 80- plot NGS at the lowest g0 or smallest σ.
4 | DISCUSSION
Single- session NGS was a viable alternative to multiple- session live trapping for estimating densities of snowshoe hares under a range of field and simulation conditions. When detection probability was very low or hare movements limited, additional sampling plots were required for NGS to yield density estimates comparable to live trap-ping. Increasing the density of NGS plots at a site is relatively easy and cheap, and an important benefit of NGS is that all genetic samples can
be collected in a single site revisit, compared to the multiple survey nights required with live trapping. NGS density estimates should also be improved by increasing the number of survey sessions, but this op-tion may be more expensive than increasing sampling plots and was not evaluated in this study.
4.1 | Optimizing an NGS protocol
Studies addressing effects of environmental exposure on DNA degra-dation have suggested that NGS samples should be collected within a few days to a week of deposition, but in some cold and dry environ-ments, samples up 1 month old still had reasonable (>60%) genotyping success (Murphy et al., 2007; Piggott, 2004; Stetz, Seitz, & Sawaya, 2015). We identified an optimal NGS sampling duration of 4 days for snowshoe hare pellets. Our 80% genotyping success was relatively high for NGS (Marucco et al., 2011). The proportion of sampling plots with pellets increased steadily over the 4 days. With a sampling den-sity of 160 plots per site, this duration usually yielded sufficient sam-ples for reliable population estimates. These results are specific to our study species, sampling design, and survey conditions (e.g., timing of sampling and weather), so we recommend that other researchers conduct similar presurvey testing prior to conducting NGS population estimates. For example, sampling duration may need to be reduced for surveys conducted in warmer and wetter months or sites, due to lower genotyping success.
Nonamplification rates increased rapidly with pellet age over 4 days of sampling. Genotyping success could be further improved by limiting sampling duration to 2 or 3 days, while increasing the density of sampling plots to maintain similar capture and recapture numbers. A shorter sampling duration could reduce per- sample laboratory costs, as fewer PCR runs could be required to obtain reliable consensus gen-otypes. Eliminating the most problematic loci (those with the lowest amplification success and highest error rates) could also reduce costs while potentially improving density estimates, provided the remaining loci are sufficiently variable to minimize error due to the “shadow ef-fect” (Mills et al., 2000).
4.2 | Comparing density estimates among survey methods
In field and simulation studies, we compared the accuracy and pre-cision of density estimates from live trapping, 80- plot NGS, and
F IGURE 4 Snowshoe hare density estimates and 95% confidence intervals, based on spatial capture–recapture analysis of live trapping and noninvasive genetic sampling (NGS) at five sites. NGS results are shown for two sampling densities (80 and 160 plots per site) and computer- generated subsampling of one or two pellets per NGS plot, each for 500 iterations. We present the median density estimates and median confidence intervals of estimable iterations
TABLE 4 At each site, the % of unique hares recaptured. With live trapping, a hare is recaptured if it is caught on more than one trap night. With single- survey NGS, a hare is recaptured if its genotype is confirmed from pellets collected from more than one NGS plot. For NGS, we present the median value of estimable iterations from subsampling one or two pellets per plot
Site Live- trap 80 plots, 1 pellet 80 plots, 2 pellets 160 plots, 1 pellet160 plots, 2 pellets
Glacier1 30 31 40 NA NA
Glacier2 57 33 40 NA NA
Flathead1 12 14 12 30 50
Flathead2 63 20 27 47 41
Flathead3 75 12 9 40 42
| 6217CHENG Et al.
160- plot NGS. The number of pellets genotyped per NGS plot had little influence on density estimates. This finding is not surprising, as multiple pellets on a single plot often arose from just one individual; we focus on one- pellet results hereafter.
Previous studies on multiple- session CMR surveys demonstrated that when trap spacing is more than twice the value of σ, SCR estimates may be poor (Sollmann, Gardner, & Belant, 2012; Sun, Fuller, & Royle, 2014). Our simulations corroborated these findings. At the lowest σ simulated (20 m), trap spacing exceeded individual hare movement distances, recaptures were rare, and all methods performed poorly. For higher σ, live trapping and 160- plot NGS density estimates were unbiased and precision was comparable. However, density estimates from 80- plot NGS were sometimes still biased and almost always had larger 95% CI’s than the other methods did.
In multiple- session live trapping, a recapture can occur at the same trap on different trap nights, although that kind of recapture alone does not provide the spatial information necessary to estimate the detection parameters in a SCR model. With single- session NGS, a re-capture occurs only when an animal is detected at different plots. This distinction may explain why a higher density of NGS plots (160 instead of 80) was required to achieve density estimates comparable with live trapping at low σ.
Differences in recapture and initial capture rates underpinned dif-ferences in density estimates among survey methods, most evident at low g0 or σ. Density estimates were highly variable, sometimes greatly exceeded true simulated densities, and had large 95% CI’s when there were few recaptures (<~20%). In simulations, recapture rates were al-most always lower for 80- plot NGS than for live trapping or 160- plot NGS. This result is intuitive, as live trapping occurred over multiple oc-casions, and sampling density in the 160- plot design was twice as high as in the 80- plot scenario. Field results were consistent with these
findings—the three cases with large 95% CI’s (one live-trap and two 80- plot NGS) had <20% recaptures.
Simulations also identified a target threshold for initial captures. When fewer than ~20 unique individuals were captured (unless low captures were due to low densities), estimated density was frequently biased low. Similarly, White, Anderson, Burnham, and Otis (1982) rec-ommended at least 20 unique individuals and 30% capture probability for reliable CMR estimates.
4.3 | Cost- effectiveness of live trapping versus NGS
When the primary variable of interest is a population estimate, single- session NGS may be a cost- effective alternative to live trapping, even for common species like snowshoe hares. NGS is especially advanta-geous when study sites are difficult to access, because noninvasive genetic samples are often much easier and cheaper to collect in the field compared to live- trap data.
Single- session NGS survey costs entail two site visits (deployment and collection), field supplies, and laboratory costs. It is difficult to re-duce the field costs any further; the primary variables that can be ad-justed are the number of cloths deployed (in USD, ~$15 for 80 cloths; $30 for 160 cloths) and the duration in the field. In the laboratory, there are significant cost and data- quality trade- offs concerning pellet freshness, numbers of attempted amplifications, and re- runs to estab-lish consensus genotypes. All of these costs increase as the number of pellets to analyze increases, and that number is a function of actual hare density and the field sampling design.
With our optimized 160- plot NGS protocol, the per- sample cost covering laboratory labor and genetic supplies in this study was $42. Using this protocol, estimated laboratory costs for our Flathead sites ranged from $672 (Flathead1: 16 plots with pellets X $42 per pellet)
F IGURE 5 Snowshoe hare density estimates and 95% confidence intervals, from simulations based on our snowshoe hare study system. For each of five sampling approaches (figure legend), we simulated three hare densities (0.2, 1.0, and 1.8 hares per hectare, in figure rows), three detection probabilities (g0 = 0.05, 0.10, and 0.15, in figure columns), and three levels of sigma (20, 50, and 80 m, on x- axis in figure cells). For each simulated scenario, we present the median density estimates and median confidence intervals of estimable iterations. If <10% of the 500 iterations were estimable, results are blank
6218 | CHENG Et al.
to $1260 (Flathead3: 30 plots with pellets × $42 per pellet). At our highest density site (Glacier1), we only deployed 80 NGS plots, but assuming the proportion of plots with pellets would be similar for 160 plots, the estimated laboratory cost would have been $2940 (70 plots with pellets × $42 per pellet).
Live trapping surveys have no laboratory costs but entail higher field costs, because of the high number of site visits (our protocol involved opening traps at night and checking traps the following morning, for three to five nights of trapping) and more time and labor required to deploy live-traps than NGS cloths. In our case, the genetic surveys would take about 8–12 person hours in the field (two people× two visits × two to three hours per visit), whereas trapping would take anywhere from 30 person hours on our easiest sites (nine person hours to set out traps, nine person hours to collect them, three evenings × two person hours to set traps, and three mornings × two person hours to check traps) to 80–1120 hrs for harder sites (more brush, deadfall, hill, or further from roads), sites with more hares to handle, or for more nights of trapping. If surveying backcountry study sites, the additional time and expense of overnight stays for live trapping could be even more considerable. At an hourly wage of $10, the costs would become $80 to $120 for the field time for the genetic survey, and anywhere from $300 to $1200 for live trapping each site, plus all the additional gasoline for the repeat site visits. These simple cost estimates suggest that when hares are very abundant or laboratory costs expensive, live trapping may be more cost- effective, but that harder sites to navigate or sites with few hares may be more efficiently sampled via the NGS pellet protocol.
In wildlife studies, issues other than cost and the reliability of population estimates are often important for evaluating the cost- effectiveness of survey methods. If studies require the additional data that live- captures afford (e.g., age or body mass information, opportu-nities for radio- collaring, or blood or tissue biopsies for disease work), then this approach to population estimation makes more sense than employing an NGS approach. On the other hand, NGS has the huge advantage of being noninvasive and less disruptive to wildlife popula-tions, and often less visible in the field, which is advantageous in areas with many tourists, such as national parks.
5 | CONCLUSIONS
Surprisingly few comparisons have been made between traditional trap- based and noninvasive estimates of density, and to our knowl-edge, none have asked whether noninvasive genetic methods can be cost- efficient for surveying relatively common and easily trappable species such as snowshoe hares. These questions are increasingly relevant because the downsides of noninvasive genetic sampling have been rapidly decreasing with improved laboratory and analytical techniques.
Our comparison of NGS and live trapping for snowshoe hares shows that NGS could indeed be viable. Our pilot work was essen-tial for determining an appropriate trade- off in the collecting pe-riod between acquiring more pellets and avoiding excessive pellet
degradation. We also found that increasing the sampling density (from 80 to 160 cloths per site) greatly improved NGS results. Both live trap-ping and NGS methods suffer when recapture rates are low. We en-courage researchers contemplating an NGS approach to calculate a cost comparison between methods for their study system; NGS does have much lower field costs, but those need to be weighed against laboratory costs, which increase with hare density.
ACKNOWLEDGMENTS
We thank the Animal Welfare Institute, the National Park Service, and NSF (Division of Environmental Biology Grant 0841884) for funding this research. We appreciate logistical support from staff at Glacier National Park, Flathead National Forest, and Stillwater State Forest.
CONFLICT OF INTEREST
None declared.
DATA ACCESSIBILITY
Microsatellite data are available for download at http://datadryad.org under DRYAD Repository entry https://doi.org/10.5061/dryad.s04 h8. Simulation R code is provided in Appendix S3.
REFERENCES
Borchers, D. L., & Efford, M. G. (2008). Spatially explicit maximum likeli-hood methods for capture- recapture studies. Biometrics, 64, 377–385.
Broquet, T., & Petit, E. (2004). Quantifying genotyping errors in noninvasive population genetics. Molecular Ecology, 13, 3601–3608.
Burton, C., Krebs, C. J., & Taylor, E. B. (2002). Population genetic structure of the cyclic snowshoe hare (Lepus americanus) in southwestern Yukon, Canada. Molecular Ecology, 11, 1689–1701.
Cheng, E., Hodges, K. E., Melo-Ferreira, J., Alves, P. C., & Mills, L. S. (2014). Conservation implications of the evolutionary history and ge-netic diversity hotspots of the snowshoe hare. Molecular Ecology, 23, 2929–2942.
Cheng, E., Hodges, K. E., & Mills, L. S. (2015). Impacts of fire on snow-shoe hares in Glacier National Park, Montana, USA. Fire Ecology, 11, 119–136.
Ciucci, P., Gervasi, V., Boitani, L., Boulanger, J., Paetkau, D., Prive, R., & Tosoni, E. (2015). Estimating abundance of the remnant Apennine brown bear population using multiple noninvasive genetic data sources. Journal of Mammalogy, 96, 206–220.
Efford, M. (2004). Density estimation in live- trapping studies. Oikos, 106, 598–610.
Efford, M. G. (2016). secr: Spatially explicit capture-recapture models. R package version 2.10.3. http://CRAN.
Hedges, S., Johnson, A., Ahlering, M., Tyson, M., & Eggert, L. S. (2013). Accuracy, precision, and cost- effectiveness of conventional dung den-sity and fecal DNA based survey methods to estimate Asian elephant (Elephas maximus) population size and structure. Biological Conservation, 159, 101–108.
Hodges, K. E. (1999). Proximate factors affecting snowshoe hare move-ments during a cyclic population low phase. Ecoscience, 6, 487–496.
Hodges, K. E. (2000a). The ecology of snowshoe hares in northern boreal forests. In L. F. Ruggiero, K. B. Aubry, S. W. Buskirk, G. M. Koehler, C. J.
| 6219CHENG Et al.
Krebs, K. S. McKelvey, & J. R. Squires (Eds.), Ecology and conservation of lynx in the United States (pp. 117–162). Boulder, CO: University Press of Colorado.
Hodges, K. E. (2000b). The ecology of snowshoe hares in southern boreal and montane forests. In L. F. Ruggiero, K. B. Aubry, S. W. Buskirk, G. M. Koehler, C. J. Krebs, K. S. McKelvey, & J. R. Squires (Eds.), Ecology and conservation of lynx in the United States (pp. 163–206). Boulder, CO: University Press of Colorado.
Hodges, K. E., & Mills, L. S. (2008). Designing fecal pellet surveys for snow-shoe hares. Forest Ecology and Management, 256, 1918–1926.
Hodges, K. E., & Sinclair, A. R. E. (2005). Browse site selection by snowshoe hares: Effects of food supply and predation risk. Canadian Journal of Zoology, 83, 280–292.
Lukacs, P. M., & Burnham, K. P. (2005). Review of capture- recapture meth-ods applicable to noninvasive genetic sampling. Molecular Ecology, 14, 3909–3919.
Marucco, F., Boitani, L., Pletscher, D. H., & Schwartz, M. K. (2011). Bridging the gaps between non- invasive genetic sampling and population pa-rameter estimation. European Journal of Wildlife Research, 57, 1–13.
Mills, L. S., Citta, J. J., Lair, K. P., Schwartz, M. K., & Tallmon, D. A. (2000). Estimating animal abundance using noninvasive DNA sampling: Promise and pitfalls. Ecological Applications, 10, 283–294.
Mulders, R., Boulanger, J., & Paetkau, D. (2007). Estimation of population size for wolverines Gulo gulo at Daring Lake, Northwest Territories, using DNA based mark- recapture methods. Wildlife Biology, 13, 38–51.
Murphy, M. A., Kendall, K. C., Robinson, A., & Waits, L. P. (2007). The impact of time and field conditions on brown bear (Ursus arctos) faecal DNA amplification. Conservation Genetics, 8, 1219–1224.
Pierce, B. L., Lopez, R. R., & Silvy, N. J. (2012) Estimating animal abun-dance. In N. J. Silvy (Ed.), The wildlife techniques manual (pp. 284–310). Besthesda, MA: John Hopkins Press, The Wildlife Society.
Piggott, M. P. (2004). Effect of sample age and season of collection on the reliability of microsatellite genotyping of faecal DNA. Wildlife Research, 31, 485–493.
Pollock, K. H., Nichols, J. D., Brownie, C., & Hines, J. E. (1990). Statistical inference for capture- recapture experiments. Wildlife Monographs, 107, 1–97.
Royle, J. A., Chandler, R. B., Sollman, R., & Gardner, B. (2013). Spatial cap-ture recapture. Waltham, Mass: Academic Press.
Schwartz, M. K., Luikart, G., & Waples, R. S. (2007). Genetic monitoring as a promising tool for conservation and management. Trends in Ecology & Evolution, 22, 25–33.
Schwartz, M. K., Pilgrim, K. L., McKelvey, K. S., Rivera, P. T., & Ruggiero, L. F. (2007). DNA markers for identifying individual snowshoe hares using field- collected pellets. Northwest Science, 81, 316–322.
Sollmann, R., Gardner, B., & Belant, J. L. (2012). How does spatial study design influence density estimates from spatial capture- recapture models? PLoS ONE, 7(4), e34575. https://doi.org/10.1371/journal.pone.0034575
Sollmann, R., Torres, N. M., Furtado, M. M., Jacomo, A. T. A., Palomares, F., Roques, S., & Silveira, L. (2013). Combining camera- trapping and noninvasive genetic data in a spatial capture- recapture framework im-proves density estimates for the jaguar. Biological Conservation, 167, 242–247.
Stetz, J. B., Seitz, T., & Sawaya, M. A. (2015). Effects of exposure on geno-typing success rates of hair samples from brown and American black bears. Journal of Fish and Wildlife Management, 6, 191–198.
Sun, C. C., Fuller, A. K., & Royle, J. A. (2014). Trap configuration and spacing influences parameter estimates in spatial capture- recapture models. PLoS ONE, 9(2), e88025. https://doi.org/10.1371/journal.pone.0088025
Taberlet, P., Griffin, S., Goossens, B., Questiau, S., Manceau, V., Escaravage, N., … Bouvet, J. (1996). Reliable genotyping of samples with very low DNA quantities using PCR. Nucleic Acids Research, 24, 3189–3194.
Taberlet, P., Waits, L. P., & Luikart, G. (1999). Noninvasive genetic sampling: Look before you leap. Trends in Ecology and Evolution, 14, 323–327.
Valiere, N. (2002). GIMLET: A computer program for analysing genetic indi-vidual identification data. Molecular Ecology Notes, 2, 377–379.
Waits, J. L., & Leberg, P. L. (2000). Biases associated with population esti-mation using molecular tagging. Animal Conservation, 3, 191–199.
Waits, L. P., & Paetkau, D. (2005). Noninvasive genetic sampling tools for wildlife biologists: A review of applications and recommendations for ac-curate data collection. Journal of Wildlife Management, 69, 1419–1433.
White, G. C., Anderson, D. R., Burnham, K. P., & Otis, D. L. (1982). Capture-recapture and removal methods for sampling closed populations. Los Alamos, NM: Los Alamos National Laboratory.
SUPPORTING INFORMATION
Additional Supporting Information may be found online in the supporting information tab for this article.
How to cite this article: Cheng E, Hodges KE, Sollmann R, Mills LS. Genetic sampling for estimating density of common species. Ecol Evol. 2017;7:6210–6219. https://doi.org/ 10.1002/ece3.3137