High Genetic Diversity and Low Differentiation Reflect The Ecological Versatility of The African Leopard

Article
High genetic diversity and low differentiation reflect

the ecological versatility of the African leopard
Graphical Abstract Authors
nerová, Genı́s Garcia-Erill,
Patrı́cia Pec
Xiaodong Liu, ..., Rasmus Heller,
Ida Moltke, Kristian Hanghøj
Correspondence
rheller@bio.ku.dk (R.H.),
ida@binf.ku.dk (I.M.),
kristianhanghoej@gmail.com (K.H.)
In Brief
Pecnerová et al. analyze whole-genome
data from 53 African leopards and identify
genetic features that set the African
leopard apart from the Amur leopards, as
well as the other big cats. This includes
long-term high effective population size
and exceptionally high levels of genetic
diversity and connectivity.
Highlights
d African and Amur leopards have markedly different
demographic trajectories
d Among big cats, African leopards have the highest genetic

diversity
d Gene flow on a continent-wide scale maintains low genetic

differentiation
d Broad dietary and habitat niche likely explain the

extraordinary genetic makeup
Pecnerová et al., 2021, Current Biology 31, 1–10

May 10, 2021 ª 2021 Elsevier Inc.
https://doi.org/10.1016/j.cub.2021.01.064 ll
nerová et al., High genetic diversity and low differentiation reflect the ecological versatility of the African leopard,
Please cite this article in press as: Pec
Current Biology (2021), https://doi.org/10.1016/j.cub.2021.01.064
ll
Article
High genetic diversity and low
differentiation reflect the ecological
versatility of the African leopard
nerová,1,5,7 Genı́s Garcia-Erill,1,5 Xiaodong Liu,1,5 Casia Nursyifa,1,5 Ryan K. Waples,1,5 Cindy G. Santander,1,5
Patrı́cia Pec
Liam Quinn,1 Peter Frandsen,1,2 Jonas Meisner,1 Frederik Filip Stæger,1 Malthe Sebro Rasmussen,1
Anna Brüniche-Olsen,1,3 Christian Hviid Friis Jørgensen,1 Rute R. da Fonseca,4 Hans R. Siegismund,1
Anders Albrechtsen,1,6 Rasmus Heller,1,6,* Ida Moltke,1,6,* and Kristian Hanghøj1,6,8,*
1Department of Biology, University of Copenhagen, Ole Maaløes Vej 5, 2200 Copenhagen N, Denmark
2Copenhagen Zoo, Research and Conservation, Roskildevej 32, 2000 Frederiksberg, Denmark
3Department of Forestry and Natural Resources, Purdue University, 610 Purdue Mall, West Lafayette, IN 47907, USA
4Center for Macroecology, Evolution and Climate (CMEC), GLOBE Institute, University of Copenhagen, Universitetsparken 15, 2100
Copenhagen, Denmark
5These authors contributed equally
6Senior authors
7Twitter: @PatriciaChrzan
8Lead Contact
*Correspondence: rheller@bio.ku.dk (R.H.), ida@binf.ku.dk (I.M.), kristianhanghoej@gmail.com (K.H.)

https://doi.org/10.1016/j.cub.2021.01.064
SUMMARY
Large carnivores are generally sensitive to ecosystem changes because their specialized diet and position at
the top of the trophic pyramid is associated with small population sizes. Accordingly, low genetic diversity at
the whole-genome level has been reported for all big cat species, including the widely distributed leopard.
However, all previous whole-genome analyses of leopards are based on the Far Eastern Amur leopards
that live at the extremity of the species’ distribution and therefore are not necessarily representative of the
whole species. We sequenced 53 whole genomes of African leopards. Strikingly, we found that the genomic
diversity in the African leopard is 2- to 5-fold higher than in other big cats, including the Amur leopard, likely
because of an exceptionally high effective population size maintained by the African leopard throughout the
Pleistocene. Furthermore, we detected ongoing gene flow and very low population differentiation within Af-
rican leopards compared with those of other big cats. We corroborated this by showing a complete absence
of an otherwise ubiquitous equatorial forest barrier to gene flow. This sets the leopard apart from most other
widely distributed large African mammals, including lions. These results revise our understanding of trophic
sensitivity and highlight the remarkable resilience of the African leopard, likely because of its extraordinary
habitat versatility and broad dietary niche.
INTRODUCTION to human settlements, labeling it as a habitat and dietary gener-

alist.8,9 Given that genetic diversity is related to ecological resil-
Apex predators are more sensitive to climate change and other ience, the leopard should have higher genetic diversity than
ecological disturbances than species at lower trophic levels.1–3 other big cats, yet this was not found in previous genome-wide
This makes them more prone to demographic fluctuations and studies.5 However, the only published whole-genome data
thus more susceptible to population crashes and even extinc- from the leopard5,10 originates from the Amur leopard (P. p. ori-
tion.4 Given that genetic diversity is linked to demographic fluc- entalis), which lives at the eastern extremity of the species’
tuations, apex predators are expected to have on average lower geographic distribution (Figure S1). This population has a history
genetic diversity than their herbivorous prey species. A rapidly of severe range and population contractions, making it the most
accumulating body of evidence from whole-genome studies5,6 critically endangered leopard subspecies with less than 60 indi-
has confirmed this in, among others, all of the big cats, i.e., the viduals surviving in the wild.11
species belonging to Panthera,7 a genus of typical apex The leopard is thought to have originated in eastern Africa
predators. approximately 2–4 million years ago (mya).12,13 The divergence
Among the big cats, the leopard, Panthera pardus, stands out between African and Asian leopards has been dated to 710
because of its extensive geographic distribution8 and its ability to thousand years ago (kya) on the basis of a mitogenome phylog-
persist under a wide range of ecological conditions, even close eny calibrated by using historical and ancient samples.12
Current Biology 31, 1–10, May 10, 2021 ª 2021 Elsevier Inc. 1
ll
Article
Figure 1. Sampling locations
Locations of origin of the samples analyzed in this
study. Terrain and country borders are shown for
context. Sample sizes at each location are shown
within the colored circles. Information on the cur-
rent range of P. p. pardus was drawn from the IUCN
Red List of Threatened Species 2020.11 The tropical
rainforest is highlighted based on the extent in-
ferred in Mayaux et al.18 See also Figure S1 and
Table S1.
(Figure 1; Table S1). We targeted 2–53

depth of coverage for 47 samples and
deeply sequenced (to 15–203 depth of
coverage) the remaining 6 samples. Map-
ping and rigorous quality filtering of the
reference genomes and the samples (see
STAR methods and Tables S2 and S3) re-
sulted in a final dataset of 1,374,856,842
genomic sites when mapped to the leop-
ard reference genome and 916,801,099
sites for the domestic cat (Felis catus)
reference for 41 samples (36 low coverage
Subsequently, the divergence between Asian and Pleistocene- and 5 high coverage). The 12 discarded samples were removed
era European leopards was dated to 483 kya, and if we assume due to either low DNA quality, species mis-labeling, or sample
that this divergence reflects the out-of-Africa event, this sets the duplication (Tables S2–S5). Our final dataset contained two pairs
likely timing of leopards entering Eurasia to 710–483 kya.12 of first-degree relatives (Table S5), of which one individual per
Because of this long period of separation, Asian and African pair was removed for analyses where relatedness can confound
leopards might differ markedly in their evolutionary trajectories, the results. For all analyses including any low-depth samples, we
including effective population size and population connectivity. accounted for the genotype uncertainty by using methods based
This, combined with the Amur leopard’s very limited census pop- on genotype likelihoods or single-read sampling instead of called
ulation size and the fact that population contractions reduce ge- genotypes (see STAR methods).
netic diversity, suggests that the results of genomic analyses of
Amur leopards might not represent the entire leopard species Population structure within African leopards
and in particular not the African leopards. We first explored the population structure of African leopards by
Although a few genetic studies have been performed on the performing a principal component analysis (PCA) of 39 African
African leopard, they have all been based on microsatellites leopards (excluding 1 sample from each of the 2 pairs of first-de-
and/or mitochondrial data. Based on these studies, which iden- gree relatives) by using PCAngsd.19 We grouped samples by
tified low population differentiation, all African leopards have country of origin, except for Tanzania, which we divided into
been classified as a single subspecies P. p. pardus.14–16 On three groups (North, West, and East). The resulting plot roughly
the contrary, a recent fine-scale sampling of 182 African leop- mirrors the geography (Figure 2A). Specifically, the first principal
ards by using 611 base pairs of mitochondrial DNA suggested component (PC) captures a cline of genetic variation from the
substantial population structure and deep divergence within Af- north (Ghana and Tanzania North) to the southwest (Namibia),
rica.17 Therefore, our genetic understanding of leopards in their whereas the second PC captures a cline across the northern lo-
proposed continent of origin remains conflicted and scarce. cations. Notably, sampling locations are not discretely clustered
In this study, we investigated population structure, demo- along these PC axes; instead, we detected a pattern consistent
graphic history, and genetic diversity in African leopards by with continuous genetic variation across the populations, with
sequencing and analyzing 53 whole genomes, covering most the exception of the Ghana population.
of their current range in Africa. The aim was to learn more about Next, we estimated per-individual admixture proportions by
the evolutionary dynamics of the leopard in its biogeographic using NGSadmix.20 We excluded the single sample from Uganda
cradle and to assess whether the generalist ecology of the leop- (n = 38) because an imbalanced sample size can create biases in
ard sets it apart from other big cats. admixture models.24 We found that a model with four ancestral
source populations (K = 4) provides the best fit to the dataset
RESULTS (Figures 2B and S2). In this model, most sampling locations are
assigned to homogeneous ancestry groups with two exceptions:
We generated whole-genome sequencing data for 53 African first, Zambia is modeled as a mixture of the Namibia and
leopards from 10 locations in sub-Saharan Africa and a total of Tanzania West clusters; and second, Tanzania West and East
5 countries: Ghana, Namibia, Tanzania, Uganda, and Zambia are grouped together, which might be caused by a small sample
2 Current Biology 31, 1–10, May 10, 2021

ll
Article
A B Figure 2. Population structure

(A) PCA of genetic variation in 39 individuals by
using PCAngsd,19 showing PC1 against PC2.
Samples are colored by sampling locations. See
also Figure S2A for plots showing PC1–PC4.
(B) Ancestry proportions estimated in NGSad-
mix20 for K = 4. The single sample from Uganda
has been excluded from the analysis. See also
Figure S2B for K = 2 and K= 3. The asterisk de-
notes individuals with distinct ancestry profiles
compared with those of the rest of samples from
the same population in (B) and (C).
(C) NJ tree based on an identity-by-state (IBS)
matrix of 39 individuals calculated in ANGSD21
and plotted in the R package ape.22 Scale bar
C shows the genetic distance.
(D) Evaluation of the admixture proportions shown
D
in (B) inferred by using the evalAdmix program.23
The upper diagonal shows the correlation of re-
siduals between individuals, and the lower diag-
onal shows the mean correlation within pop-
ulations. A positive correlation of residuals is
reflective of a bad model fit. Correlation values
above and below the color scale are set to dark
red and dark blue, respectively. See also Fig-
ure S2C for evalAdmix results with K = 2 to K = 4.
these groupings as populations and

quantify the magnitude of genetic differ-
entiation between them by using FST.
However, before doing so, we first
confirmed that this population grouping
was justifiable by performing a popula-
size and/or little differentiation between them. However, we note tion homogeneity test using D-statistics (ABBA-BABA)25,26 on
that an evaluation of the admixture model fit using evalAdmix23 the low-coverage samples that were not inferred to be admixed
(Figure 2D) indicates that the discrete clustering imposed by in the NGSadmix analysis. Because of low sample sizes,
NGSadmix is not a good fit to the data, as shown by the presence Tanzania East (n = 2) and Uganda (n = 1) were excluded. We
of numerous non-zero residuals within clusters, consistent with tested for relative differences in genetic affinity between all pairs
the genetic continuity across geographic space suggested by of individuals within a location (H1 and H2) and another sampling
the PCA (Figure 2A). location (H3), while using the Amur leopard reference (Pan-
To get a better sense of how the samples are genetically clus- Par1.0)5 as an outgroup (H4). We found strong signals of sub-
tered, we built a neighbor-joining (NJ) tree based on the 39 indi- structure in Zambia, where all D-statistics were significant (sug-
viduals from the PCA, as well as two previously published Amur gesting gene flow with the other sampling locations), in contrast
leopards (IDs 3042211 and 3042212).5 We found that all African to Ghana, Namibia, Tanzania North, and Tanzania West, where
leopards form a cluster with a branching pattern consistent with we found zero or one significant test (|Z-score|>3) (Figure 3).
the north-to-south cline observed in the PCA; however, it is This result is consistent with the population structure analyses
evident from the NJ tree that there is very low differentiation be- (Figure 2) and suggests that Ghana, Namibia, Tanzania North,
tween the different sampling locations with short internal branch and Tanzania West are meaningful population groupings. We
lengths. Ghana was the most divergent group, whereas the other therefore quantified the magnitude of genetic differentiation be-
sampling locations formed a separate cluster with a substructure tween populations by estimating pairwise FST. Overall, we find
representing the sampling locations (with the exception of low levels of differentiation, and the highest values of FST were
Zambia) in the following order of divergence: Uganda; Tanzania 0.118–0.145 (Table 1, lower triangle) between Ghana and the re-
North, East, and West; Zambia; and Namibia (Figure 2C). One maining populations, again consistent with the population struc-
sample (ID 5180) stands out as being labeled Tanzania West ture results (Figure 2). In addition, we estimated FST between all
and clustering with Tanzania North; however, this result is pairs of individuals to factor out any downward bias from poten-
consistent with NGSadmix results, indicating that the sample is tial substructure within the populations. Note that we used the
a mixture of the Tanzania West and North ancestral sources (Fig- Reich FST estimator27 because simulations show that it is robust
ures 2B and 2C). to low sample size, even down to a single sample from each pop-
Given that the estimated NJ tree largely supports our initial ge- ulation (Figure S7B). Because this analysis was based on called
ography-based grouping of the samples, we decided to treat genotypes, we restricted it to our five high-coverage samples
Current Biology 31, 1–10, May 10, 2021 3

ll
Article
African mammals with sub-Saharan distribution, including the

lion (Panthera leo), for which the dense rainforest is a barrier
(Figure 4).
Migration between Africa and Asia

We also tested for signals of post-divergence gene flow between
the Amur and African leopards by using D-statistics with the do-
mestic cat as an outgroup (H4). We found that all other popula-
tions (H2) contrasted against Ghana (H1) show a positive D-sta-
tistic (Figure 5), suggesting an excess of gene flow in the eastern
and southern African lineages with a source represented by, but
not necessarily equal to, the Amur leopard (H3). This is consis-
tent with the West African samples being more genetically
distant from the Amur leopard than leopards from East and
Southern Africa.
Demographic history across the big cats

To investigate the demographic history of the African leopard,
we inferred the effective population size through time by using
pairwise sequentially Markovian coalescent (PSMC)30 (Fig-
ure 6A). Because PSMC requires called genotypes, we restricted
the analysis to the five high-coverage African leopards. We
scaled the population size assuming a mutation rate of 108
per base per year, which has previously been used in studies
of big cat species,10 and a generation time of 7.5 years.11 For
comparison, we applied the same procedure to genomes repre-
senting all the other big cats for which we could find publicly
available data.5,10,31 We detected a clear signal of divergence
between the African and Amur leopards (Figure 6A), given that
their population size trajectories start to differ markedly around
300–400 kya, likely representing the out-of-Africa event. Our
PSMC results further show that after the divergence, the African
leopards maintained a high effective population size, in contrast
to the Amur leopards that went through a bottleneck. Notably,
the maintenance of high effective population sizes above
20,000 throughout the Pleistocene is unique for the African leop-
ard in comparison with all other big cats, which all show
decreasing effective population sizes in recent times.
Genetic diversity
Figure 3. Population homogeneity
Finally, we inferred the present-day genetic diversity of African
D-statistics (ABBA-BABA) testing for homogeneity of designated groups. Each
panel is titled by the population being tested (H1 and H2), and the y axis is leopards by estimating the genome-wide heterozygosity, i.e.,
labeled with the respective remaining African populations (H3). Dashed lines the proportion of heterozygous sites per sample. We inferred
mark the significance thresholds for Z-scores. on average 2 heterozygous sites per 1,000 base pairs (bp),
with little variation between the various populations of African
(Table 1, upper triangle). The results show a similar range in FST leopards (Figure 6B). However, we note that the samples from
(0.12–0.17) between Ghana and the remaining populations as Ghana were within the higher end of the error rate distribution,
the population-based FST analysis. which could slightly upward-bias the estimates. To investigate
the minor differences in heterozygosity between the different
Migration within Africa populations, we correlated the estimated heterozygosity levels
The migration patterns between the different sampling locations against tracts of runs of homozygosity (ROH) for the five high-
were further investigated by using Estimation of Effective Migra- coverage samples. We found a negative linear correlation be-
tion Surfaces (EEMS)29 (Figure 4). The results fit with an isolation- tween the total length of ROH and heterozygosity (Figure S4B).
by-distance model in most parts of the range (Figure S3), sug- This indicates that the majority of variability in genetic diversity
gesting that roughly distance-dependent gene flow is a major among African leopard populations can be explained by differ-
determinant of the observed genetic differences. In addition, ences in recent inbreeding rather than differences in demo-
we find that African leopards show only weak signatures of graphic histories.
migration barriers across the latitudinal cline, in particular across In line with the distinctive demographic trajectories inferred
the equatorial tropical rainforest. This is in sharp contrast to other from PSMC (Figure 6A), the estimated heterozygosity levels for

ll
Article
Table 1. FST between populations and pairs of individuals leopard was previously severely underestimated due to being
characterized in Amur leopards, which have an unrepresentative
Ghana Namibia TanzaniaN TanzaniaW TanzaniaE
population history for the species. We show that, in addition to
Ghana N/A 0.159 0.169 0.120 0.125
suffering recent population declines,32 Amur leopards also carry
Namibia 0.144 N/A 0.134 0.076 0.080 a genetic legacy of the out-of-Africa event occurring in the mid-
TanzaniaN 0.145 0.100 N/A 0.069 0.076 dle Pleistocene (Figure 6A) and possibly a number of subsequent
TanzaniaW 0.118 0.067 0.051 N/A 0.019 founder events. Lower genetic diversity is also frequently
TanzaniaE N/A N/A N/A N/A N/A observed at the edge of species ranges because of somewhat
Lower triangle shows the estimated FST between pairs of populations. extreme demographic histories.33
The upper triangle shows the estimated FST between pairs of high- In contrast to the Amur leopards, African leopards retained
coverage individuals. Tanzania East (TanzaniaE) was excluded from the large effective population sizes after the split. Furthermore, Afri-
population-based analysis. Figure S7B shows that the FST between pairs can leopards consistently maintained much higher population
of individuals is not biased. The population based FST was based on 2d- sizes than all other big cats throughout the Pleistocene (Fig-
SFS from all individuals from each population estimated with realSFS,28 ure 6A). Declining population sizes were also recently inferred
whereas the individual FST was based on the 2d-SFS from called geno- in the puma34 and the cheetah (Acinonyx jubatus).35 Consistent
types from a high-coverage individual from each population.
with these differences in effective population size histories, the
African leopards have by far the highest genetic diversity not
only among big cats (Figure 6B) but among wild cats in general,
the African leopards are 3-fold higher than those in the Amur matched only by the leopard cat, a small felid mesopredator spe-
leopards. They are also markedly higher than for any other big cies with a wide distribution in southern and eastern Asia.34,36
cat species, and most felid species in general (Figure 6A; Table Interestingly, our results suggest that after the divergence, the
S6), with the exception of two distantly related felids: the leopard African populations remained genetically connected with the
cat (Prionailurus bengalensis) and to a lesser degree the puma Asian populations, given that we detected gene flow with a pop-
(Puma concolor). ulation represented by the Amur leopard lineage and the African
leopards (Figure 5). The extent of Asian ancestry is not equal be-
DISCUSSION tween African populations, suggesting either ancestral popula-
tion structure before the split or a backflow into the ancestral
Our results provide new insights into the evolutionary history and branch of the eastern and southern African populations. Uphyr-
molecular ecology of the leopard in its biogeographic cradle, the kina et al.14 found that the South Arabian leopard (P. p. nimr)
African continent. We found that the genetic diversity of the could not be consistently placed with either African or Asian
Figure 4. Reconstruction of migration surfaces

Estimation of effective migration surfaces (EEMS)29 was used to identify barriers to gene flow and regions of sustained genetic connectivity. The color scale of
blue to dark orange represents the log10-transformed effective migration rates (m); the blue indicates areas with relatively higher migration, and orange indicates
areas with relatively lower migration. Circles represent the demes (subpopulations) rather than the predefined sampling locations. The lion and elephant plots
have smaller scales to make the coloring clearer because of a lower resolution in the microsatellite data. The equatorial rainforest is highlighted with gray shading
(according to Mayaux et al.18) to show that, unlike in the other species, it does not constitute a barrier for gene flow in leopards. See also Figure S3.

ll
Article
H1 H2 H3 Cat H1 H2 H3 Cat Figure 5. D-statistics with low coverage

samples and Amur leopard
All pairs of individuals assigned to each possible
pair of African locations are grouped as H1 and H2
with the Asian leopard sample with ID 3042211 as
H3. The domestic cat was used as an outgroup
(H4). Comparisons with a significant D-statistic
TanzaniaE, Zambia, Asia (absolute Z score above 3) are plotted as golden
Namibia, Zambia, Asia |Z score| < 3 points.
TanzaniaN, Zambia, Asia |Z score| > 3
TanzaniaN, TanzaniaW, Asia
TanzaniaE, Namibia, Asia
TanzaniaW, Zambia, Asia
out as having low levels of population
TanzaniaN, Namibia, Asia
structure compared with that of the ma-
TanzaniaN, TanzaniaE, Asia
jority of other species in the African
TanzaniaW, Namibia, Asia
TanzaniaW, TanzaniaE, Asia large-mammal savanna guild that show
Uganda, Zambia, Asia strong regional structuring.38–41 The
Uganda, TanzaniaW, Asia average pairwise FST between popula-
Uganda, TanzaniaN, Asia tions was 0.10, which is considerably
Uganda, Namibia, Asia lower than corresponding values for
Uganda, TanzaniaE, Asia other co-distributed African carnivores
Ghana, Uganda, Asia (cheetah: 0.2942 and lion: 0.2543).
Ghana, Zambia, Asia Although genome-wide data can be ex-
Ghana, TanzaniaW, Asia pected to capture finer-scale structure,
Ghana, TanzaniaN, Asia our results are consistent with early
Ghana, Namibia, Asia studies based on microsatellite data
Ghana, TanzaniaE, Asia and partial mitochondrial sequences,
which reported limited genetic structure
within the African continent.14,15 Howev-
−0.04
−0.02
0.00
0.02
0.04
er, a recent and densely sampled study

contrastingly found strong structure in
D statistic Africa.17 This discrepancy is possibly ex-
plained by the use of mitochondrial
data,17 which only reflects the female
leopards, which would be in accordance with genetic connectiv- lineage and fails to capture the male-biased dispersal in
ity between a proto-eastern African leopard population and leopards.44
those of the Arabian Peninsula and the Middle East. Our results As typical apex predators, the big cats of the genus Panthera
therefore suggest Pleistocene-era intercontinental genetic con- are specialized hypercarnivores depending on large prey spe-
nectivity comparable to the level observed in another versatile cies.45,46 Being at the top of the trophic pyramid,3,4 carnivores
carnivore, the spotted hyena.37 tend to have lower effective population sizes, and are thus
The topology of our genome-wide NJ tree is consistent with a more affected by environmental fluctuations and habitat frag-
southward expansion of the eastern African lineage after diver- mentation in periods when the climatic conditions are not favor-
gence with Ghana (Figure 2C); however, the internal branch able.47,48 As a consequence, carnivores are prone to increased
lengths are very short, suggesting that any such pattern is genetic drift and low genetic diversity compared with herbi-
only tentative and obscured by pervasive gene flow. Overall, vores,5,6,49 leading to an increased risk of extinction,4,50 with
we found signatures of an isolation-by-distance model in Holliday and Steppan45 labeling highly specialized carnivory as
most parts of the range, but this pattern is attenuated by exten- an evolutionary ‘‘dead end.’’ In stark contrast to its congeners,
sive gene flow on a continent-wide scale within Africa (Fig- we find that the African leopard is an exceptionally adaptable
ure S3). Concordantly, we found that African leopards show apex predator. High mobility,51,52 habitat versatility,8,9 and die-
at best weak signatures of dispersal barriers across the latitudi- tary generalism13,53 have buffered the long-term high effective
nal cline (Figure 4). Although the character of sampling and low population sizes in the African leopards by making them less
density of samples from western and northern Africa prevent us sensitive to habitat fragmentation and environmental fluctua-
from resolving genetic connectivity patterns on a finer scale, tions during the Pleistocene climatic cycles. In this light, our re-
the continent-level pattern in leopards clearly stands out sults are surprising, but they are in line with the biology of the
compared with that of other co-distributed African mammals leopard being defined by a broader dietary and habitat niche
by presence of gene flow across the equatorial rainforest. The than any other big cat. We argue that the African leopard might
equatorial tropical rainforest is one of the main drivers of diver- constitute an evolutionary anomaly with a better chance of
sification for many African mammals,38 but we found no evi- long-term survival than other Panthera species. Future studies
dence of it being a barrier for leopards, underlining their excep- involving more extensive sampling throughout the leopard range
tional habitat tolerance. Accordingly, the leopard also stands will resolve how current genetic diversity is connected with

ll
Article
Figure 6. Demographic history and genetic diversity

(A) Effective population sizes were reconstructed by using pairwise sequentially Markovian coalescent (PSMC30) assuming a mutation rate, mu, of 108 and a
generation time, g, of 7.5 years. African leopards have a distinct demographic history compared with that of the Amur leopard and the other big cats. See also
Figure S7A for PSMC with a generation time of 5 years.
(B) Estimates of genome-wide heterozygosity inferred in ANGSD21 are compared between the African leopard populations, and also between all felid species with
available whole-genome estimates. See also Figures S4, S6, S7A and Table S6.
demographic history. We speculate that the Amur leopard will Despite their resilience, leopards have lost 48%–67% of their
most likely not be representative of other Asian leopard subspe- original distribution in Africa within the last 300 years13 (Fig-
cies because of its recent bottleneck5 and its position at the ex- ure S1), and their range is contracting at a similar rate to some of
tremity of the leopard geographic range (Figure S1). the other big cats.54,55 Even though our results highlight that the

ll
Article
leopards might be among the most adaptable of apex predators, B Sample quality filtering: Identification of related and
and therefore relatively robust to environmental change and distur- duplicated samples
bance, anthropogenic activities can pose an unprecedented threat B Genotype likelihoods, SNPs, and genotype calling
to their survival. Until recently, these have not had a major effect on B PCA analyses
leopard populations in Africa, but their potential effect is evident in B NGSadmix
the other leopard subspecies, particularly the critically endangered B Runs of homozygosity (ROH) analysis
Amur leopard, which has been strongly affected by the lack of B D-statistics
available prey, illegal hunting, and fragmentation of habitat.11 In B Site frequency spectrum
addition, unlike species that went through periods of low popula- B SFS-based analyses: FST estimation
tion size, African leopards have had constantly high population B SFS-based analyses: Estimation of heterozygosity
sizes and have not endured bottlenecks, which would have purged B PSMC
strongly deleterious variation from the gene pool. African leopards B Identity-by-state (IBS)
might therefore harbor a larger number of strongly deleterious mu- B IBS-based analyses: Neighbor-joining (NJ) tree
tations at low population frequencies. These can increase in fre- B IBS-based analyses: Estimation of Effective Migration
quency as a result of population contractions, placing the African Surfaces (EEMS)
leopard at risk of inbreeding depression.56–58
In conclusion, we correct the existing bias in the assessment SUPPLEMENTAL INFORMATION
of leopard genetic diversity and show that African leopards
have exceptionally high genetic diversity for their trophic posi- Supplemental Information can be found online at https://doi.org/10.1016/j.
cub.2021.01.064.
tion, coupled with high continent-wide genetic connectivity in
one of their biogeographic strongholds. Also, on the basis of
ACKNOWLEDGMENTS
our observations of low population differentiation and lack of
migration barriers across Africa, we hypothesize that these un- The authors would like to thank Peter Arctander for providing access to his ex-
usual genetic features are caused by the resilience, adaptability, isting collection of leopard samples that were analyzed in this study. We also
and high mobility of the leopard. This study represents a rare thank David Moyer for contributing samples to the collection. The wildlife man-
positive narrative in conservation genetics and, to our knowl- agement authorities of Ghana, Namibia, Tanzania, Uganda, and Zambia are
edge, one of the most convincing couplings of a species’ gener- thanked for collaborating with Peter Arctander in the sample collection. P.F.
was supported by the Innovation Fund Denmark, Candys Foundation, and
alist ecology with its correspondingly exceptional long-term
the Alfred Benzon Fund. R.H. was supported by a Villum Foundation Young
evolutionary dynamics and genetic features. investigator grant (VKR023447) and a Danish Research Council Sapere
Aude grant (DFF8049-00098B). G.G.E., J.M., A.A., and K.H. were supported
by a Lundbeck Foundation grant (R215-2015-4174). I.M. and X.L. were sup-
ported by a Villum Foundation Young Investigator grant (19114), R.R.d.F.
STAR+METHODS
thanks the Danish National Research Foundation for its funding of the Center
for Macroecology, Evolution, and Climate (DNRF96).
Detailed methods are provided in the online version of this paper
and include the following: AUTHOR CONTRIBUTIONS
d KEY RESOURCES TABLE

H.R.S., A.A., R.H., and I.M. conceived and designed the study. P.P., G.G.E.,
d RESOURCE AVAILABILITY X.L., C.N., R.K.W., C.G.S., L.Q., P.F., J.M., F.F.S., A.B.-O., C.H.F.J.,
B Lead contact R.R.d.F., H.R.S., A.A., R.H., I.M., and K.H. analyzed the data. P.P., A.A.,
B Materials availability R.H., I.M., and K.H. wrote the manuscript with input from all authors.
B Data and code availability
d EXPERIMENTAL MODEL AND SUBJECT DETAILS DECLARATION OF INTERESTS
d METHOD DETAILS
The authors declare no competing interests.
B DNA extraction
B Sequencing
Received: September 8, 2020
B Mapping Revised: November 13, 2020
d QUANTIFICATION AND STATISTICAL ANALYSES Accepted: January 19, 2021
B Reference Quality filtering: Mappability and Repeat- Published: February 25, 2021
Masker
B Reference Quality filtering: Global Sequencing Depth REFERENCES
filter
1. Voigt, W., Perner, J., Davis, A.J., Eggers, T., Schumacher, J., Ba €hrmann,
B Reference Quality filtering: Autosomal and sex-linked
R., Fabian, B., Heinrich, W., Köhler, G., Lichter, D., et al. (2003). Trophic
scaffold identification levels are differentially sensitive to climate. Ecology 84, 2444–2453.
B Reference Quality filtering: Excessive heterozygosity
2. Cheng, B.S., Komoroske, L.M., and Grosholz, E.D. (2017). Trophic sensi-
filter tivity of invasive predator and native prey interactions: integrating environ-
B Reference Quality filtering: Summarizing reference mental context and climate change. Funct. Ecol. 31, 642–652.
quality filters 3. Stier, A.C., Samhouri, J.F., Novak, M., Marshall, K.N., Ward, E.J., Holt,
B Sample quality filtering: Error rates estimation R.D., and Levin, P.S. (2016). Ecosystem context and historical contin-
B Sample quality filtering: BLAST analyses of mtDNA gency in apex predator recoveries. Sci. Adv. 2, e1501769.

ll
Article
4. Purvis, A., Gittleman, J.L., Cowlishaw, G., and Mace, G.M. (2000). 24. Puechmaille, S.J. (2016). The program structure does not reliably recover
Predicting extinction risk in declining species. Proc. Biol. Sci. 267, the correct population structure when sampling is uneven: subsampling
1947–1952. and new estimators alleviate the problem. Mol. Ecol. Resour. 16, 608–627.
5. Kim, S., Cho, Y.S., Kim, H.-M., Chung, O., Kim, H., Jho, S., Seomun, H., 25. Green, R.E., Krause, J., Briggs, A.W., Maricic, T., Stenzel, U., Kircher, M.,
Kim, J., Bang, W.Y., Kim, C., et al. (2016). Comparison of carnivore, omni- Patterson, N., Li, H., Zhai, W., Fritz, M.H.-Y., et al. (2010). A draft sequence
vore, and herbivore mammalian genomes with a new leopard assembly. of the Neandertal genome. Science 328, 710–722.
Genome Biol. 17, 211. 26. Durand, E.Y., Patterson, N., Reich, D., and Slatkin, M. (2011). Testing for
6. Dı́ez-Del-Molino, D., Sánchez-Barreiro, F., Barnes, I., Gilbert, M.T.P., and ancient admixture between closely related populations. Mol. Biol. Evol.
Dalen, L. (2018). Quantifying Temporal Genomic Erosion in Endangered 28, 2239–2252.
Species. Trends Ecol. Evol. 33, 176–185. 27. Reich, D., Thangaraj, K., Patterson, N., Price, A.L., and Singh, L. (2009).
7. Nowak, R.M., and Walker, E.P. (1999). Walker’s Mammals of the World Reconstructing Indian population history. Nature 461, 489–494.
(JHU Press). 28. Nielsen, R., Korneliussen, T., Albrechtsen, A., Li, Y., and Wang, J. (2012).
8. Nowell, K., and Jackson, P. (1996). Wild cats: status survey and conserva- SNP calling, genotype calling, and sample allele frequency estimation
tion action plan (Switzerland: IUCN Gland). from New-Generation Sequencing data. PLoS ONE 7, e37558.
9. Hunter, L., Henschel, P., and Ray, J.C. (2013). Panthera pardus. In 29. Petkova, D., Novembre, J., and Stephens, M. (2016). Visualizing spatial
Mammals of Africa Volume V: Carnivores, Pangolins, Equids and population structure with estimated effective migration surfaces. Nat.
Rhinoceroses, J. Kingdon, D. Happold, T. Butynski, M. Hoffmann, M. Genet. 48, 94–100.
Happold, and J. Kalina, eds. (Bloomsbury Publishing PLC London, UK), 30. Li, H., and Durbin, R. (2011). Inference of human population history from
pp. 159–168. individual whole-genome sequences. Nature 475, 493–496.
10. Figueiró, H.V., Li, G., Trindade, F.J., Assis, J., Pais, F., Fernandes, G., 31. Cho, Y.S., Hu, L., Hou, H., Lee, H., Xu, J., Kwon, S., Oh, S., Kim, H.-M.,
Santos, S.H.D., Hughes, G.M., Komissarov, A., Antunes, A., et al. (2017). Jho, S., Kim, S., et al. (2013). The tiger genome and comparative analysis
Genome-wide signatures of complex introgression and adaptive evolution with lion and snow leopard genomes. Nat. Commun. 4, 2433.
in the big cats. Sci. Adv. 3, e1700299.
32. Uphyrkina, O., Miquelle, D., Quigley, H., Driscoll, C., and O’Brien, S.J.
11. Stein, A.B., Athreya, V., Gerngross, P., Balme, G., Henschel, P., Karanth, (2002). Conservation genetics of the Far Eastern leopard (Panthera pardus
U., et al. (2020). Panthera pardus (amended version of 2019 assessment). orientalis). J. Hered. 93, 303–311.
The IUCN Red List of Threatened Species 2020, e.T15954A163991139.
33. Eckert, C.G., Samis, K.E., and Lougheed, S.C. (2008). Genetic variation
12. Paijmans, J.L.A., Barlow, A., Förster, D.W., Henneberger, K., Meyer, M., across species’ geographical ranges: the central-marginal hypothesis
Nickel, B., Nagel, D., Worsøe Havmøller, R., Baryshnikov, G.F., Joger, and beyond. Mol. Ecol. 17, 1170–1188.
U., et al. (2018). Historical biogeography of the leopard (Panthera pardus)
34. Saremi, N.F., Supple, M.A., Byrne, A., Cahill, J.A., Coutinho, L.L., Dale n,
and its extinct Eurasian populations. BMC Evol. Biol. 18, 156.
L., Figueiró, H.V., Johnson, W.E., Milne, H.J., O’Brien, S.J., et al. (2019).
13. Jacobson, A.P., Gerngross, P., Lemeris, J.R., Jr., Schoonover, R.F., Anco, Puma genomes from North and South America provide insights into the
C., Breitenmoser-Würsten, C., Durant, S.M., Farhadinia, M.S., Henschel, genomic consequences of inbreeding. Nat. Commun. 10, 4769.
P., Kamler, J.F., et al. (2016). Leopard (Panthera pardus) status, distribu-
35. Dobrynin, P., Liu, S., Tamazian, G., Xiong, Z., Yurchenko, A.A.,
tion, and the research efforts across its range. PeerJ 4, e1974.
Krasheninnikova, K., Kliver, S., Schmidt-Küntzel, A., Koepfli, K.-P.,
14. Uphyrkina, O., Johnson, W.E., Quigley, H., Miquelle, D., Marker, L., Bush, Johnson, W., et al. (2015). Genomic legacy of the African cheetah,
M., and O’Brien, S.J. (2001). Phylogenetics, genome diversity and origin of Acinonyx jubatus. Genome Biol. 16, 277.
modern leopard, Panthera pardus. Mol. Ecol. 10, 2617–2633.
36. Corbet, G.B., Hill, J.E., and Hill, J.E. (1992). The Mammals of the
15. Miththapala, S., Seidensticker, J., and O’Brien, S.J. (1996). Indomalayan Region: A Systematic Review (Oxford University Press on
Phylogeographic Subspecies Recognition in Leopards (Panthera pardus). Demand).
Molecular Genetic Variation. Conserv. Biol. 10, 1115–1132.
37. Westbury, M.V., Hartmann, S., Barlow, A., Preick, M., Ridush, B., Nagel,
16. Kitchener, A.C., Breitenmoser-Würsten, C., Eizirik, E., Gentry, A., D., Rathgeber, T., Ziegler, R., Baryshnikov, G., Sheng, G., et al. (2020).
Werdelin, L., Wilting, A., Yamaguchi, N., Abramov, A.V., Christiansen, P., Hyena paleogenomes reveal a complex evolutionary history of cross-con-
Driscoll, C., et al. (2017). A revised taxonomy of the Felidae: The final report tinental gene flow between spotted and cave hyena. Sci Adv 6, eaay0456.
of the Cat Classification Task Force of the IUCN Cat Specialist Group (Cat
38. Lorenzen, E.D., Heller, R., and Siegismund, H.R. (2012). Comparative phy-
News).
logeography of African savannah ungulates. Mol. Ecol. 21, 3656–3670.
17. Anco, C., Kolokotronis, S.-O., Henschel, P., Cunningham, S.W., Amato,
39. de Manuel, M., Barnett, R., Sandoval-Velasco, M., Yamaguchi, N., Garrett
G., and Hekkala, E. (2018). Historical mitochondrial diversity in African
Vieira, F., Zepeda Mendoza, M.L., Liu, S., Martin, M.D., Sinding, M.S.,
leopards (Panthera pardus) revealed by archival museum specimens.
Mak, S.S.T., et al. (2020). The evolutionary history of extinct and living li-
Mitochondrial DNA A. DNA Mapp. Seq. Anal. 29, 455–473.
ons. Proc. Natl. Acad. Sci. USA 117, 10927–10934.
18. Mayaux, P., Pekel, J.-F., Descle e, B., Donnay, F., Lupi, A., Achard, F.,
40. Nyakaana, S., Arctander, P., and Siegismund, H.R. (2002). Population
Clerici, M., Bodart, C., Brink, A., Nasi, R., and Belward, A. (2013). State
structure of the African savannah elephant inferred from mitochondrial
and evolution of the African rainforests between 1990 and 2010. Philos.
control region sequences and nuclear microsatellite loci. Heredity 89,
Trans. R. Soc. Lond. B Biol. Sci. 368, 20120300.
90–98.
19. Meisner, J., and Albrechtsen, A. (2018). Inferring population structure and
41. Fennessy, J., Bidon, T., Reuss, F., Kumar, V., Elkan, P., Nilsson, M.A.,
admixture proportions in low-depth NGS data. Genetics 210, 719–731.
Vamberger, M., Fritz, U., and Janke, A. (2016). Multi-locus analyses reveal
20. Skotte, L., Korneliussen, T.S., and Albrechtsen, A. (2013). Estimating indi- four giraffe species instead of one. Curr. Biol. 26, 2543–2549.
vidual admixture proportions from next generation sequencing data.
42. Prost, S., Machado, A.P., Zumbroich, J., Preier, L., Mahtani-Williams, S.,
Genetics 195, 693–702.
Guschanski, K., Brealey, J.C., Fernandes, C., Vercammen, P., Godsall-
21. Korneliussen, T.S., Albrechtsen, A., and Nielsen, R. (2014). ANGSD: Bottriell, L., et al. (2020). Conservation genomic analyses of African and
Analysis of next generation sequencing data. BMC Bioinformatics 15, 356. Asiatic cheetahs (Acinonyx jubatus) across their current and historic spe-
22. Paradis, E., and Schliep, K. (2019). ape 5.0: an environment for modern cies range. bioRxiv, 2020.02.14.949081.
phylogenetics and evolutionary analyses in R. Bioinformatics 35, 526–528. 43. Smitz, N., Jouvenet, O., Ambwene Ligate, F., Crosmary, W.-G., Ikanda, D.,
23. Garcia-Erill, G., and Albrechtsen, A. (2020). Evaluation of model fit of in- Chardonnet, P., Fusari, A., Meganck, K., Gillet, F., Melletti, M., and
ferred admixture proportions. Mol. Ecol. Resour. 20, 936–949. Michaux, J.R. (2018). A genome-wide data assessment of the African

ll
Article
lion (Panthera leo) population genetic structure and diversity in Tanzania. 61. Li, H., Handsaker, B., Wysoker, A., Fennell, T., Ruan, J., Homer, N., Marth,
PLoS ONE 13, e0205395. G., Abecasis, G., and Durbin, R.; 1000 Genome Project Data Processing
44. Fattebert, J., Balme, G., Dickerson, T., Slotow, R., and Hunter, L. (2015). Subgroup (2009). The Sequence Alignment/Map format and SAMtools.
Density-dependent natal dispersal patterns in a leopard population recov- Bioinformatics 25, 2078–2079.
ering from over-harvest. PLoS ONE 10, e0122355. 62. Pockrandt, C., Alzamel, M., Iliopoulos, C.S., and Reinert, K. (2020).
45. Holliday, J.A., and Steppan, S.J. (2004). Evolution of hypercarnivory: the GenMap: ultra-fast computation of genome mappability. Bioinformatics
effect of specialization on morphological and taxonomic diversity. 36, 3687–3692.
Paleobiology 30, 108–128. 63. Smit, A.F.A. (2004). Repeat-Masker Open-3.0. http://www.repeatmasker.
46. Shrestha, B., Reed, J.M., Starks, P.T., Kaufman, G.E., Goldstone, J.V., org. Available at: https://ci.nii.ac.jp/naid/10029514778/ [Accessed August
Roelke, M.E., O’Brien, S.J., Koepfli, K.-P., Frank, L.G., and Court, M.H. 17, 2020].
(2011). Evolution of a major drug metabolizing enzyme defect in the do- 64. Morgulis, A., Coulouris, G., Raytselis, Y., Madden, T.L., Agarwala, R., and
mestic cat and other felidae: phylogenetic timing and the role of hypercar- Scha€ffer, A.A. (2008). Database indexing for production MegaBLAST
nivory. PLoS ONE 6, e18046. searches. Bioinformatics 24, 1757–1764.
47. Frankham, R. (1995). Effective population size/adult population size ratios 65. McKenna, A., Hanna, M., Banks, E., Sivachenko, A., Cibulskis, K.,
in wildlife: a review. Genet. Res. 66, 95–107. Kernytsky, A., Garimella, K., Altshuler, D., Gabriel, S., Daly, M., and
DePristo, M.A. (2010). The Genome Analysis Toolkit: a MapReduce frame-
48. Frankham, R., Ballou, J.D., and Briscoe, D.A. (2010). Introduction to
work for analyzing next-generation DNA sequencing data. Genome Res.
Conservation Genetics, 2nd ed. (Cambridge: Cambridge University Press).
20, 1297–1303.
https://doi.org/10.1017/CBO9780511809002.
66. Purcell, S., Neale, B., Todd-Brown, K., Thomas, L., Ferreira, M.A.R.,
49. Hansen, C.C.R., Hvilsom, C., Schmidt, N.M., Aastrup, P., Van Coeverden
Bender, D., Maller, J., Sklar, P., de Bakker, P.I.W., Daly, M.J., and
de Groot, P.J., Siegismund, H.R., et al. (2018). The muskox lost a substan-
Sham, P.C. (2007). PLINK: a tool set for whole-genome association and
tial part of Its genetic diversity on its long road to Greenland. Curr. Biol. 28,
population-based linkage analyses. Am. J. Hum. Genet. 81, 559–575.
4022–4028.
67. Li, H. (2011). A statistical framework for SNP calling, mutation discovery,
50. Van Valkenburgh, B., Wang, X., and Damuth, J. (2004). Cope’s rule, hyper-
association mapping and population genetical parameter estimation
carnivory, and extinction in North American canids. Science 306, 101–104.
from sequencing data. Bioinformatics 27, 2987–2993.
51. Fattebert, J., Dickerson, T., Balme, G., Slotow, R., and Hunter, L. (2013).
68. Köster, J., and Rahmann, S. (2012). Snakemake–a scalable bioinformatics
Long-distance natal dispersal in leopard reveals potential for a three-
workflow engine. Bioinformatics 28, 2520–2522.
country metapopulation. Afr. J. Wildl. Res. 43, 61–67.
69. Pontius, J.U., Mullikin, J.C., Smith, D.R., Lindblad-Toh, K., Gnerre, S.,
52. Ray-Brambach, R.R., Stommel, C., and Rödder, D. (2018). Home ranges,
Clamp, M., Chang, J., Stephens, R., Neelam, B., Volfovsky, N., et al.
activity patterns and habitat preferences of leopards in Luambe National
(2007). Agencourt Sequencing Team; NISC Comparative Sequencing
Park and adjacent Game Management Area in the Luangwa Valley,
Program Initial sequence and comparative analysis of the cat genome.
Zambia. Mamm. Biol. 92, 102–110.
Genome Res. 17, 1675–1689.
53. Hayward, M.W., Henschel, P., O’brien, J., Hofmeyr, M., Balme, G., and 70. Wei, L., Wu, X., Zhu, L., and Jiang, Z. (2011). Mitogenomic analysis of the
Kerley, G.I.H. (2006). Prey preferences of the leopard (Panthera pardus). genus Panthera. Sci. China Life Sci. 54, 917–930.
J. Zool. (Lond.) 270, 298–313.
71. Meisner, J., and Albrechtsen, A. (2019). Testing for Hardy-Weinberg equi-
54. Mahmood, T., Younas, A., Akrim, F., Andleeb, S., Hamid, A., and Nadeem, librium in structured populations using genotype or low-depth next gener-
M.S. (2019). Range contraction of snow leopard (Panthera uncia). PLoS ation sequencing data. Mol. Ecol. Resour. 19, 1144–1152.
ONE 14, e0218460.
72. Orlando, L., Ginolhac, A., Zhang, G., Froese, D., Albrechtsen, A., Stiller,
55. Quigley, H., Foster, R., Petracca, L., Payan, E., Salom, R., and Harmsen, M., Schubert, M., Cappellini, E., Petersen, B., Moltke, I., et al. (2013).
B. (2017). Panthera onca (errata version published in 2018). The IUCN Recalibrating Equus evolution using the genome sequence of an early
Red List of Threatened Species 2017, e.T15953A123791436. Middle Pleistocene horse. Nature 499, 74–78.
56. Robinson, J.A., Ra €ikkönen, J., Vucetich, L.M., Vucetich, J.A., Peterson, 73. Waples, R.K., Albrechtsen, A., and Moltke, I. (2019). Allele frequency-free
R.O., Lohmueller, K.E., and Wayne, R.K. (2019). Genomic signatures of inference of close familial relationships from genotypes or low-depth
extensive inbreeding in Isle Royale wolves, a population on the threshold sequencing data. Mol. Ecol. 28, 35–48.
of extinction. Sci Adv 5, eaau0757.
74. Manichaikul, A., Mychaleckyj, J.C., Rich, S.S., Daly, K., Sale, M., and
57. Abascal, F., Corvelo, A., Cruz, F., Villanueva-Cañas, J.L., Vlasova, A., Chen, W.-M. (2010). Robust relationship inference in genome-wide asso-
Marcet-Houben, M., Martı́nez-Cruz, B., Cheng, J.Y., Prieto, P., ciation studies. Bioinformatics 26, 2867–2873.
Quesada, V., et al. (2016). Extreme genomic erosion after recurrent demo- 75. Busing, F.M.T.A., Meijer, E., and Leeden, R.V.D. (1999). Delete-m jack-
graphic bottlenecks in the highly endangered Iberian lynx. Genome Biol. knife for unequal m. Stat. Comput. 9, 3–8.
17, 251.
76. Wasser, S.K., Brown, L., Mailand, C., Mondol, S., Clark, W., Laurie, C., and
58. Prado-Martinez, J., Sudmant, P.H., Kidd, J.M., Li, H., Kelley, J.L., Lorente- Weir, B.S. (2015). CONSERVATION. Genetic assignment of large seizures
Galdos, B., Veeramah, K.R., Woerner, A.E., O’Connor, T.D., Santpere, G., of elephant ivory reveals Africa’s major poaching hotspots. Science 349,
et al. (2013). Great ape genetic diversity and population history. Nature 84–87.
499, 471–475.
77. Bertola, L.D., Tensen, L., van Hooft, P., White, P.A., Driscoll, C.A.,
59. Gaspar, J.M. (2018). NGmerge: merging paired-end reads via novel empir- Henschel, P., Caragiulo, A., Dias-Freedman, I., Sogbohossou, E.A.,
ically-derived models of sequencing errors. BMC Bioinformatics 19, 536. Tumenta, P.N., et al. (2015). Autosomal and mtDNA Markers Affirm the
60. Li, H., and Durbin, R. (2009). Fast and accurate short read alignment with Distinctiveness of Lions in West and Central Africa. PLoS ONE 10,
Burrows-Wheeler transform. Bioinformatics 25, 1754–1760. e0137975.

ll
Article
STAR+METHODS
KEY RESOURCES TABLE
REAGENT or RESOURCE SOURCE IDENTIFIER

Chemicals, Peptides, and Recombinant Proteins
RNase A QIAGEN Cat# 19101
Critical Commercial Assays
DNeasy Blood & Tissue Kit QIAGEN Cat# 69506
AxyPrep MAG PCR Clean-Up Kit Axygen MAG-PCR-CL-50
QIAquick Gel Extraction kit QIAGEN Cat# 28706X4
Deposited Data
Raw sequencing reads This study ENA accession number: PRJEB41230
Software and Algorithms
NGmerge Gaspar59 https://github.com/jsh58/NGmerge
BWA-mem v0.7.17 Li and Durbin60 http://bio-bwa.sourceforge.net/
Samtools v1.9 Li et al.61 http://www.htslib.org/
GENMAP v1.2.0 Pockrandt et al.62 https://github.com/cpockrandt/genmap
RepeatMasker v4.0.8 Smit63 http://www.repeatmasker.org/
ANGSD Korneliussen21 http://www.popgen.dk/angsd/index.php/ANGSD
19
PCAngsd v0.973 Meisner and Albrechtsen http://www.popgen.dk/software/index.php/PCAngsd
megaBLAST Morgulis et al.64 https://blast.ncbi.nlm.nih.gov
GATK McKenna et al.65 https://gatk.broadinstitute.org/hc/en-us
NGSAdmix Skotte et al.20 http://www.popgen.dk/software/index.php/NgsAdmix
EvalAdmix Garcia-Erill and Albrechtsen23 https://github.com/GenisGE/evalAdmix
PLINK v1.9 Purcell et al.66 https://www.cog-genomics.org/plink/
realSFS v0.931 Nielsen et al.28 http://www.popgen.dk/angsd/index.php/RealSFS
Bcftools v1.9 Li67 http://samtools.github.io/bcftools/
PSMC Li and Durbin30 https://github.com/lh3/psmc
R package ape Purcell et al.66 https://cran.r-project.org/package=ape
EEMS Petkova et al.29 https://github.com/dipetkov/eems
RStudio v1.2.5033 RStudio Team https://rstudio.com/
29
make_eems_plots script Petkova et al. https://github.com/dipetkov/reemsplots2
Snakemake Köster and Rahmann68 https://snakemake.readthedocs.io/en/stable/
Other
Panthera pardus reference genome PanPar1.0 Kim et al.5 GenBank: GCF_001857705.1
Domestic cat reference genome Felis_catus_9.0 Pontius et al.69 GenBank: GCA_000181335.4
Panthera pardus mitogenome Wei et al.70 GenBank: NC_010641.1
RESOURCE AVAILABILITY
Lead contact
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Kristian
Hanghøj (kristian.hanghoej@gmail.com).
Materials availability
This study did not generate new unique reagents
Data and code availability

The accession number for the sequencing data reported in this paper is ENA: PRJEB41230. The code used for analyses in this project
is available on the project’s github: https://github.com/KHanghoj/leopardpaper. Some pipelines were developed using
snakemake.68
Current Biology 31, 1–10.e1–e5, May 10, 2021 e1

ll
Article
EXPERIMENTAL MODEL AND SUBJECT DETAILS
We analyzed fifty-three tissue samples of African leopards kindly provided from an existing collection of Peter Arctander. These sam-
ples were collected between 1993–1998 in Ghana, Namibia, Tanzania, Uganda, and Zambia (Table S1), with the assistance of the
local wildlife management authorities and in compliance with the local and international legislation. Most samples were from skin tis-
sue and were kept in a DMSO buffer in the field and stored at 20 C as soon as possible. The samples were subsequently transferred
to a 80 C freezer for long-term storage.
METHOD DETAILS
DNA extraction
For DNA extraction, we used the QIAGEN DNeasy Blood & Tissue Kit (QIAGEN, Valencia, CA, USA), following the manufacturer’s
protocol. 10uL RNase A was added to get RNA-free genomic DNA. DNA concentrations were measured with a Qubit 2.0 Fluorometer
and with a Nanodrop. We subsequently used gel electrophoresis to ascertain genomic DNA quality.
Sequencing
After DNA extraction, 1 mg genomic DNA was randomly fragmented by Covaris (350 bp on average), followed by purification by Ax-
yPrep Mag PCR clean up kit. The fragments were end repaired by End Repair Mix and purified afterward. The repaired DNAs were
combined with A-Tailing Mix, then the Illumina adaptors were ligated to the DNA adenylate 30 ends, followed by product purification.
Size selection was performed targeting insert size of 350 base pairs. Several rounds of PCR amplification with PCR Primer Cocktail
and PCR Master Mix were performed to enrich the Adaptor-ligated DNA fragments. After purification, the libraries were assessed by
the Agilent Technologies 2100 Bioanalyzer and ABI StepOnePlus Realtime PCR System.
All samples were sequenced in paired-end 2x150 bp mode, 47 of these to approximately 2-5X depth of coverage on the Illumina
NovaSeq platform, and six samples were sequenced to approximately 15-20X depth of coverage on the Illumina HiSeq2500 plat-
form. In total, 7.2 billion raw reads were generated and analyzed (Table S2).
Mapping
Cleaned paired-end reads were collapsed when read termini overlapped by at least 11 base pairs (bp) using NGmerge59 (options: -b
11 -g -f). Collapsed reads and uncollapsed paired-end reads were mapped separately with BWA-MEM (v0.7.17; default settings)60
to: 1) a Panthera pardus reference genome (PanPar1.0; GenBank: GCF_001857705.1,5 representing a female zoo specimen of an
Amur leopard), and 2) a reference genome from a female domestic cat (Felis_catus_9.0; GenBank: GCA_000181335.4).69 Read du-
plicates (markdup, default settings) were marked and removed (-F 3852) using Samtools (v1.9).61 We retained mapped collapsed
reads, and properly paired and mapped paired-end reads. All reads with mapping quality below 30 were excluded from all down-
stream analysis. For mapping statistics, see Table S2.
QUANTIFICATION AND STATISTICAL ANALYSES
Reference Quality filtering: Mappability and RepeatMasker

We estimated the mappability of each site in the reference genomes using GENMAP (v1.2.0).62 Mappability scores were computed
with 100 bp k-mers, allowing two mismatches (-K 100 -E 2), with remaining settings set to default. We excluded all sites with a mapp-
ability score less than one. We also excluded low complexity and repeat sequences as identified with RepeatMasker (v.4.0.8, sen-
sitive mode, -engine wublast -s -no_is -cutoff 255 -frag 20000, http://www.repeatmasker.org/).63 Lastly, we removed all scaffolds
smaller than 1 MB.
Reference Quality filtering: Global Sequencing Depth filter

We calculated the global depth (read count across all samples) for each site using ANGSD.21 Next, we calculated the lower and upper
1% percentile global depth threshold and excluded all sites outside this range. This analysis was done for all samples as well as sepa-
rately for the low-coverage and high-coverage samples (Figure S5A).
Reference Quality filtering: Autosomal and sex-linked scaffold identification

To identify sex-linked scaffolds, we first calculated the average depth for each scaffold for each sample normalized by the average
depth of the five largest scaffolds. Based on the normalized depths, we then performed a principal component analysis (PCA) and
identified two main clusters of samples, likely representing males and females (Figure S5B). Finally, considering that in leopards fe-
males have two X chromosomes and males have a single X chromosome, we designated and excluded putative sex-linked scaffolds
where the mean difference was greater than 0.4 between the two clusters. In addition, we excluded all scaffolds with an average
normalized depth greater than 1.1 or lower than 0.9 across all samples (Figure S5C). This resulted in 23 scaffolds being identified
as X chromosome-linked (Figure S5D) and excluded from further analyses.
e2 Current Biology 31, 1–10.e1–e5, May 10, 2021

ll
Article
Reference Quality filtering: Excessive heterozygosity filter

Several factors related to genomic repeats, like the presence of copy number variation between the reference genome and the
analyzed samples, can lead to mis-mapping of reads originating from multiple genome locations to a single location. Any differences
in these locations will result in inferred sites with an excess of heterozygosity that can be used to identify and mask these regions. We
computed a preliminary genotype likelihoods file for common polymorphic sites (MAF R 0.05 and SNP p value < 106) restricted to
data with a base quality of at least 30.21 Using these genotype likelihoods as input to PCAngsd,19 we calculated per-site inbreeding
coefficients (F), ranging from 1 where all samples are heterozygous to 1 where all samples are homozygous, and performed a
Hardy-Weinberg equilibrium (HWE) likelihood ratio test accounting for population structure.71 To obtain individual allele frequencies
in PCAngsd, three principal components that account for population structure were considered. Based on the per-site inbreeding
coefficients, windows of 100kb around sites with significant excessive heterozygosity estimates (F < 0.95 and p value < 106)
were excluded (Figure S5E). In addition, entire scaffolds where either 20% or more of their total sequence was removed or scaffolds
with average F value across all sites F < 0.1 were excluded.
Reference Quality filtering: Summarizing reference quality filters

The union of all masks of the reference genome (scaffold size, mappability, RepeatMasker, aligned depth, autosome identification,
excessive heterozygosity), resulted in 1.3 and 0.9 billion accessible bases in the leopard and cat reference genomes, respectively
(Table S3), henceforth referred to as strictref. The strictref filter was applied to all analyses unless otherwise noted.
Sample quality filtering: Error rates estimation

To identify and exclude highly error-prone samples from the dataset, we estimated error rates in ANGSD21 using the ‘‘perfect-indi-
vidual’’ approach as described in Orlando et al.72 Sample 3241 was selected as the ‘‘perfect-individual’’ sample for the analysis
based on the overall mapping statistics. A consensus sequence was generated for this sample with the -doFasta 2 option in
ANGSD,21 taking the most commonly observed base (-doAncError 2) as the consensus. Since the ‘‘perfect-individual’’ approach as-
sumes equal genetic distances of the samples to the outgroup, we used the domestic cat reference genome as the ancestral (out-
group) sequence. Error rates were then measured as an excess/deficit of derived alleles from the outgroup compared to the ‘‘perfect-
individual’’ sample. Both for generating the consensus sequence and for the error estimation, we restricted the data to a base quality
of at least 30. Based on the estimated error rates, we removed four samples from the dataset (sample IDs 6350, 6352, 7465, 7466),
which differed considerably from the ‘‘perfect-individual’’ estimate (Figures S6A and S6B).
Sample quality filtering: BLAST analyses of mtDNA

To ensure that all samples were leopard samples, we investigated the genetic similarity between the mitochondrial genome for each
sample and the leopard mitogenome. Mitochondrial genomes for each sample were reconstructed by creating consensus se-
quences using ANGSD21 option -doFasta 2 which selected the most common base observed per position on the NC_010641.1 scaf-
fold.70 We then used BLAST to compare the mitogenomes of our African leopards to the non-redundant BLAST nucleotide sequence
database using megaBLAST64 with a threshold of 90% identity and 75% query cover. All samples were inspected and summary sta-
tistics for the samples with extreme results are available in Table S4. Since these overlapped with the samples with high error rates
this analysis did not lead to removal of any additional samples.
Sample quality filtering: Identification of related and duplicated samples

Using the methodology described in Waples et al.,73 we identified and removed closely related and potentially duplicated samples by
first inferring two-dimensional site frequency spectrum (2d-SFS) for each pair of samples. Then we calculated three statistics directly
from the 2d-SFS for each pair of samples: R0,73 R1,73 and KING-robust kinship74 that can be used to identify close familial relatives73
without estimates of population allele frequencies.
Using this approach, we identified 12 duplicated samples. All of these had KING-robust kinship values > 0.470, R1 values > 10, and
R0 values < 106. Within each set of duplicated samples, we removed the sample with the highest error rate, thus removing eight
samples for further analyses (sample IDs 6342, 6344, 6348, 6349, 6353, 6556, 6357, 6359). All duplicated samples originated
from Zambia (Table S2). We also identified pairs of related samples, down to second-degree relatives (expected kinship = 0.125),
based on R0, R1, and KING-robust kinship following the criteria from Manichaikul et al.74 and Waples et al.73 In total, we found
two pairs of first-degree relatives: one parent-offspring pair and one full-sibling pair, and seven second-degree pairs of relatives (Ta-
ble S5). For some of the analyses we removed one of the samples from each of the pairs of first-degree relatives (samples 8647 and
7943).
For a summary of the sample sizes used for each analysis, see Table S3.
Genotype likelihoods, SNPs, and genotype calling

We used ANGSD21 to estimate genotype likelihoods using the genotype likelihood (GL) model from GATK (-gl 2),65 inferring major and
minor from GL data (-doMajorminor 1), estimating allele frequencies from the GL data (-doMaf 1), and applying the strictref filter
(-sites). We restricted the analysis to bases with base quality of at least 30.
For SNP calling we used a likelihood ratio test (p value = 106) as implemented in ANGSD.21 We only included common alleles (MAF
> 0.05).

ll
Article
For the five high-coverage samples, we called genotypes using bcftools (v.1.9).67 We called the genotypes per sample using the
multiallelic caller methodology (-m). We disabled base alignment qualities (BAQ; -B) and restricted the analysis to base qualities of at
least 30. In addition to the strictref filter (-T), we also excluded sites with a depth of coverage below 10 and heterozygous calls with
less than 3 reads support for each allele using plugin setGT to reduce genotype calling errors. These additional genotype calling filters
were validated based on heterozygous calls in runs of homozygosity (ROH) in these five samples (Figure S4A).
PCA analyses
We performed PCA (Figure 2A) on 39 individuals based on genotype likelihoods using PCAngsd,19 which accounts for missingness in
low-coverage samples. We excluded the two first-degree relatives (sample IDs 8647 and 7943) from the analysis. We used three ei-
genvectors to model population structure in the iterative procedure of PCAngsd19 as anything higher would reflect genetic variation
within populations (Figure S2A).
NGSadmix
We estimated admixture proportions for 38 individuals with NGSadmix20 (Figure 2B) based on genotype likelihoods. Besides
removing the two first-degree relatives, we also removed the only individual sampled in Uganda (sample ID 8540), as unbalanced
sample sizes prevent from accurately characterizing the genetic composition of its population of origin. We ran NGSadmix from
K = 2 to K = 6. For each K we performed several independent optimization runs either until convergence, defined as having the 3
top maximum likelihood results within 2 log-likelihoods unit of each other, or until a limit of 100 runs was reached without conver-
gence. For the values of K where the results converged (Figure S2), we assessed the model fit of the resulting admixture proportion
with evalAdmix, which estimates the pairwise correlation of the residual between individuals.23 Individuals within a population with a
bad model fit show a positive correlation of their residuals (Figure 2D).
Runs of homozygosity (ROH) analysis

We estimated ROH with the filtered genotypes of high coverage samples (see STAR Methods). We used the command ‘–homozyg’ in
PLINK v1.966 with default settings, except using a minimum of 500 kilobases to detect a ROH (–homozyg-kb 500).
D-statistics
We used the D-Statistics (ABBA-BABA)25,26 method to test for population homogeneity and gene flow. We used ANGSD21 option
-doAbbababa 1, which samples a random base at each position to estimate the counts of ABBA and BABA sites between each triplet
of H1, H2 and H3 individuals in blocks of a predefined size of 5 MB. We calculated the Z-score based on jackknife procedure75 with
475 blocks for data mapped to domestic cat reference and 567 blocks for samples mapped to Amur leopard.
First, we explored the homogeneity of the designated groupings (Figure 3). For each sampling location group, we tested all low-
depth individuals in a pairwise manner as (H1, H2) against the remaining African groups (H3). We excluded all individuals from Uganda
and Tanzania East due to low sample size, and three individuals with distinct ancestry profiles in the NGSadmix analyses with K = 4
compared to the rest of samples in their group (marked with an asterisk in Figure 2). The Amur leopard reference (PanPar1.0) was
used as an outgroup (H4). Second, we tested for relative differences in gene flow between the African groups (H1, H2) and the Asian
leopard (H3; ID 3042211). As an outgroup, we used the cat reference genome (H4). We excluded one Zambian sample (sample ID
2523) due to low mapping quality, which the D-statistics is sensitive to when placed in H1 or H2, and the same three individuals
with distinct ancestry profiles in the NGSadmix analyses that were also excluded from the D-statistics to test for population
homogeneity.
Site frequency spectrum

We estimated the site frequency spectrum (SFS) with the realSFS28 program (v0.931) within ANGSD,21,28 using default settings,
based on saf files generated with ANGSD (-dosaf 1 -gl 2 -minQ 30). This procedure was used to estimate one-dimensional (1d)
SFS per individual and per population and two-dimensional (2d) SFS for all pairs of individuals and populations.
SFS-based analyses: FST estimation

A genome-wide estimate of FST (Table 1) for each pair of populations was computed directly from the 2d-SFS, estimated with
realSFS,28 using the Reich estimator.27 To estimate FST between a pair of high-coverage individuals, we used as input the
2d-SFS based on called genotypes.
SFS-based analyses: Estimation of heterozygosity

We estimated heterozygosity for each sample in ANGSD21 as the proportion of heterozygous loci in the obtained per-individual 1d-
SFS. For the population and species comparison (Figure 6B), we only used the final dataset of 41 samples, minus two samples with
slightly elevated error rates, which show a linear trend between heterozygosity and error rates (Figure S6C).
For genetic diversity comparisons, we obtained whole-genome-based estimates from the literature (see Table S6 for details). To
ensure that these were comparable to our estimates we also re-estimated those for the genomes that we re-analyzed for our PSMC
analyses.
e4 Current Biology 31, 1–10.e1–e5, May 10, 2021

ll
Article
PSMC
We used PSMC30 to estimate the population size trajectory back through time for our five high coverage African leopards. We fol-
lowed the recommended procedure for generating a diploid sequence per individual using bcftools (v 1.9)67 and -c for calling geno-
types. PSMC was applied with default settings. For scaling, we used a mutation rate of 108 and generation time of 5 and 7.5
years.5,10,11 Five years is the generation time most commonly used in similar studies, but the IUCN11 lists the leopard generation
time at 7.42 years. Hence, we consider 7.5 more appropriate (Figure 6A), but we also performed the analysis with a generation
time of 5 years (Figure S7A) for consistency with previous studies.
For PSMC comparisons, we re-analyzed whole genome data from published studies of other big cats, including Amur leopards,5
tiger,31 lion,31 jaguar,10 and snow leopard.31 Briefly, we mapped the sequencing data to the cat reference genome using the same
pipeline as for the African leopard sequencing data. Unless otherwise noted, we also applied the strictref filter to all analyses.
Identity-by-state (IBS)
We estimated the pairwise genetic distance matrix (IBS) between all individuals by randomly sampling a single read from each po-
sition from each sample applying the strictref filter, using -doIBS in ANGSD. This determines the allelic distance (0 or 1) in sites with
information from both individuals and calculates the average identity-by-state (IBS) across all sites.
IBS-based analyses: Neighbor-joining (NJ) tree

We used the IBS matrix to construct a NJ tree (Figure 2C) using the R package ape22 (https://cran.r-project.org/package=ape). The
NJ tree included samples with distinct ancestry profiles as inferred from NGSadmix, denoted with an asterisk.
IBS-based analyses: Estimation of Effective Migration Surfaces (EEMS)

We used the pairwise IBS matrix as input to EEMS29 to estimate the relative migration rates (Figure 5, Figure S3). For each sampling
locality we visually inferred the centroid of its geographical coordinates and used this as the coordinate input to EEMS. EEMS was run
using the runeems_snps program and default settings for 5 million steps and a burn-in of 1 million steps, with 400 demes and the
number of sites (n = 1,374,592,127) based on the IBS matrix. For comparison, we ran EEMS with the same settings for other African
mammals with available SNP datasets - waterbuck,19 African buffalo (Nursyifa et al., in prep.), and common warthog (Jørgensen
et al., in prep.). Furthermore, we included the savanna African elephant76 and African lion,77 for which microsatellite datasets with
continent-wide sampling were available. For the microsatellite data, the runeems-sats was run using the genotypes and individual
sample coordinates as input and the same settings as above. The elephant data by Wasser et al.76 has been previously used for
the estimation of relative migration rates in the original EEMS publication.29 Unlike Petkova et al. we used the full savannah elephant
dataset (n = 1001), including samples assumed to be hybrids; however, the results are very similar.
For all species, the same habitat outline was used and was selected to encompass the sampling localities of all included species.
The outline was drawn manually using the tool available here: http://www.birdtheme.org/useful/v3tool.html. The results were visu-
alized in RStudio v1.2.5033 using the make_eems_plots script (available here: https://github.com/dipetkov/reemsplots2) as the
log10-transformed migration rates with posterior probabilities > 0.90. An outline of the equatorial rainforest was added based on
the spatial distribution of African rainforests in Mayaux et al.18

Current Biology, Volume 31
Supplemental Information
High genetic diversity and low

differentiation reﬂect the ecological
versatility of the African leopard
Patrícia Pecnerová, Genís Garcia-Erill, Xiaodong Liu, Casia Nursyifa, Ryan K.
Waples, Cindy G. Santander, Liam Quinn, Peter Frandsen, Jonas Meisner, Frederik Filip
Stæger, Malthe Sebro Rasmussen, Anna Brüniche-Olsen, Christian Hviid Friis
Jørgensen, Rute R. da Fonseca, Hans R. Siegismund, Anders Albrechtsen, Rasmus
Heller, Ida Moltke, and Kristian Hanghøj
Figure S1. The leopard species range and distribution of our sampling localities. Related to
Figure 1 and Table S1.
Nine currently recognized leopard subspecies [S1] are depicted by different coloring, while the
extent of historical range is indicated in gray shading. The Amur leopard subspecies, which is the
only one with whole-genome data available to-date, is highlighted by a red circle. Dots indicate the
sampling locations of African leopards used in this study.
Figure S2. Pairwise PCA plots and admixture analysis. Related to Figure 2 and STAR
Methods.
(A) Pairwise PCA plots, generated using PCAngsd [S2], displaying the top four inferred PCs in 39
individuals using PCAngsd. The percentage of variance explained by each PC is listed along the
diagonal. (B) and (C) Results of fitting an admixture model to the data assuming K=2-4 ancestral
populations. Convergence was not reached for higher K. (B) The admixture proportions estimated
with NGSadmix [S3]. (C) Evaluation of admixture proportions with evalAdmix [S4] as the
correlation of residuals; the upper diagonal shows the correlation of residuals between individuals,
and the lower the mean correlation within populations. A positive correlation of residuals is
reflective of a bad model fit.
Figure S3. Observed genetic dissimilarities between demes vs geographic distances
between demes. Related to Figure 4 and STAR Methods.
These estimates are calculated in EEMS [S5] and visualized in R using the reemsplots2 package
(https://github.com/dipetkov/reemsplots2/). The plot shows demes, instead of individual samples
and their precise geographic locations. The demes are assigned within the EEMS analysis on the
basis of a triangular grid with a pre-assigned density (here nDemes = 400) being laid over the
defined area (here approximately sub-Saharan Africa) and samples being allocated to the nearest
deme. As a result, a deme can contain one or more samples depending on their proximity. The
observed strong linear relationship between the great circle distance and the genetic distance
between pairs of demes suggests a good fit with the isolation-by-distance model. Demes
representing Ghana samples, highlighted in pink in the top-right corner, are the most genetically
and geographically distinct, yet they still follow the pattern.
A B
(bp)
(bp)
Figure S4. Genotype call filtering for the 5 samples (3241, 4343, 4354, 7547, and 7942)
sequenced to high depth (A), and the total length of runs of homozygosity (ROH) vs
heterozygosity (B). Related to Figure 6 and STAR Methods.
(A) In runs of homozygosity (ROH), heterozygous genotype calls are likely spurious. We identify
the source of these genotype calling errors by investigating the nature of heterozygous calls in
ROH. We show that the stringent filtering criterion reduces the number of heterozygous calls (red
bars) in regions of ROH (blue bars), thereby reducing the false positive heterozygous genotype
calls. For each of the samples, the results of two filtering criteria are shown for the seven largest
scaffolds. The filtering criteria used are from bottom (lenient) to top (stringent) row for each
scaffold: 1) Minimum depth of coverage 10 and minimum one read support for each allele for
heterozygous calls, 2) Minimum depth of coverage 10 and minimum three read support for each
allele for heterozygous calls. Blue color indicates genomic regions where heterozygous genotypes
are not observed, red are regions with heterozygous and homozygous calls. (B) The total length of
ROH is calculated as the cumulative length of genomic regions identified as ROH by PLINK [S6]
for each of the samples (indicated by blue dots and sample ID in the box). Heterozygosity is
calculated based on per-sample 1d-SFS in ANGSD [S7] and the values for all samples are listed in
Table S6. A linear regression model suggested a significant relationship between the total length of
ROH and heterozygosity with coefficient t-value=-9.789 and p-value=0.00265. The coefficient is
represented by the black line.
Figure S5. Reference filtering excluding sites with extreme depth (A), sex scaffolds (B-D),
and excess of heterozygosity (E). Related to STAR Methods.
(A) Excluding sites with extreme sequencing depth. In order to filter out sites with extreme
sequencing depth, we calculated the global depth for each site using ANGSD [S7] and then
estimated the lower and upper 1% percentile global-depth threshold and excluded all sites outside
this range. Shown in red are the thresholds excluding the 1% sites with the lowest global depth and
1% sites with the highest global depth. The number of sites with a given global depth for either all
samples, only high-depth samples, or only low-depth samples. (B-D) Autosome and sex scaffold
identification. (B) PCA of normalized depth across scaffolds from 49 samples. Projected samples
are colored by putative sex groups (group 1 and group 2). We used this grouping as sex estimate
and also an initial step of identifying sex-linked scaffolds to exclude these from our downstream
analyses. (C) Average normalized depth within the two groups for the putative sex-linked scaffolds.
We expected that mean difference between groups is around 0.5 (putative female group will have
average depth two times higher than the putative male group). (D) Mean normalized depth across
samples highlighting outlier scaffolds (red colored) and putative sex-linked scaffolds (blue colored).
The scaffolds are ordered according to their size and the error bars indicate standard error across
individuals. We developed a script to identify sex-chromosome related and outlier scaffolds. It is
available on github: https://github.com/KHanghoj/leopardpaper. See STAR Methods for further
explanation. (E) Example of the excess-of-heterozygosity filter in three different scaffolds. We used
this filter to mask regions with an excess of heterozygosity caused by mismapping of reads to
repetitive regions. Each cross represents the F value of a SNP, which is a measure of the deviation
of the SNP from Hardy-Weinberg proportions, with negative values indicating an excess of
heterozygotes and positive values an excess of homozygous. Golden crosses have a significant
excess of heterozygosity (p-value < 10-6 and F < -0.95) while grey crosses are not significant. The
black lines indicate the local mean F value in sliding windows of 100 SNPs. Regions delimited by
red lines and shaded in grey are excluded due to having an excess of heterozygosity. F values and
p-values were estimated with PCAngsd [S2,S8].
Figure S5. Reference filtering excluding sites with extreme depth (A), sex scaffolds (B-D),
and excess of heterozygosity (E). Related to STAR Methods.
(A) Excluding sites with extreme sequencing depth. In order to filter out sites with extreme
sequencing depth, we calculated the global depth for each site using ANGSD [S7] and then
estimated the lower and upper 1% percentile global-depth threshold and excluded all sites outside
this range. Shown in red are the thresholds excluding the 1% sites with the lowest global depth and
1% sites with the highest global depth. The number of sites with a given global depth for either all
samples, only high-depth samples, or only low-depth samples. (B-D) Autosome and sex scaffold
identification. (B) PCA of normalized depth across scaffolds from 49 samples. Projected samples
are colored by putative sex groups (group 1 and group 2). We used this grouping as gender
estimate and also an initial step of identifying sex-linked scaffolds to exclude these from our
downstream analyses. (C) Average normalized depth within the two groups for the putative sex-
linked scaffolds. We expected that mean difference between groups is around 0.5 (putative female
group will have average depth two times higher than the putative male group). (D) Mean
normalized depth across samples highlighting outlier scaffolds (red colored) and putative sex-
linked scaffolds (blue colored). The scaffolds are ordered according to their size and the error bars
indicate standard error across individuals. We developed a script to identify sex-chromosome
related and outlier scaffolds. It is available on github: https://github.com/KHanghoj/leopardpaper.
See STAR Methods for further explanation. (E) Example of the excess-of-heterozygosity filter in
three different scaffolds. We used this filter to mask regions with an excess of heterozygosity
caused by mismapping of reads to repetitive regions. Each cross represents the F value of a SNP,
which is a measure of the deviation of the SNP from Hardy-Weinberg proportions, with negative
values indicating an excess of heterozygotes and positive values an excess of homozygous.
Golden crosses have a significant excess of heterozygosity (p-value < 10-6 and F < -0.95) while
grey crosses are not significant. The black lines indicate the local mean F value in sliding windows
of 100 SNPs. Regions delimited by red lines and shaded in grey are excluded due to having an
excess of heterozygosity. F values and p-values were estimated with PCAngsd [S2,S8].
Figure S6. Sequencing error rates estimated using the "perfect-man" approach. Related to
Figure 6 and STAR Methods.
The error rates were estimated using -doAncError 2 in ANGSD [S7] as excess of derived alleles
compared to a high-quality “perfect-man” sample. (A) Including all 53 samples. (B) Excluding four
outliers (6350, 6352,7465, 7466) to allow a better-resolution view of the other samples. These four
outlier samples were also removed from all downstream analyses. The remaining samples have a
relatively low error rate below 0.05%. (C) A comparison of error rates and heterozygosity, both
estimated in ANGSD [S7]. Based on the comparison of heterozygosity and error rates, two Ghana
samples - IDs 7548 and 7549 - were excluded from the heterozygosity analyses, which then
consisted of 39 samples. Zambian sample ID 2523 showed a slightly elevated heterozygosity
compared to other samples, which is likely caused by the higher contamination rate in this sample.
A
Figure S7. Demographic history reconstructed assuming a generation time of 5 years

instead of 7.5 years, and pairwise FST estimated from simulated data for a varying number of
individuals. Related to Figure 6, Table 1 and STAR Methods.
(A) Reconstruction of effective population sizes using PSMC [S9] assuming a mutation rate, mu, of
10-8 and a generation time, g, of 5 years. In the main Figure 6A we used a generation time of 7.5
instead following IUCN [S10], however several previous studies have used a generation time of 5
years [S11, S12], hence this plot is included to make comparisons to previous studies easier. (B)
FST values were estimated based on 10M SNPs for all combinations of N=1 to 10 individuals from
each of two unadmixed populations. The genotypes were simulated using the R BNPSD package
[S13] using the draw_all_admix function assuming intermediate subpopulation FST values of 0.1 for
population 1 and 0.05 for population 2. Note the scale of the color bar, which suggests that the
range of values is very small and thus that the estimates for just one sample is useful.
Sample Date
Country Location Population Lat Lon Material
ID collected
2469 Zambia Kafue_N Zambia -14.5 26.2 Dried skin 95-07-11
2523 Zambia Luangwa_N Zambia -11.9 32.3 Dried skin 95-08-08
3241 Tanzania Maswa_(Mbono) TanzaniaN -3.2 34.6 Dried skin 95-07-31
3243 Tanzania Maswa_(Kimali) TanzaniaN -3.2 34.6 Dried skin 95-08-13
3244 Tanzania Maswa_(Mbono) TanzaniaN -3.2 34.6 Dried skin 95-07-29
4343 Tanzania Ugalla_W TanzaniaW -5.8 31.9 Skin 93
4346 Tanzania Ugalla_W TanzaniaW -5.8 31.9 Skin 93-04
4352 Tanzania Selous TanzaniaE -7.8 38.2 Skin 94
4354 Tanzania Selous TanzaniaE -7.8 38.2 Skin 94
4443 Tanzania Ugalla_W TanzaniaW -5.8 31.9 Skin 95-10-27
5180 Tanzania Ugalla TanzaniaW -5.8 31.9 Skin 96-08-30
5519 Tanzania Maswa TanzaniaN -3.2 34.6 Skin 97-07-25
6342 Zambia Luangwa_E Zambia NA NA Skin NA
6344 Zambia Nsumbu_N Zambia NA NA Skin NA
6346 Zambia Mumbwa_C Zambia -15 27 Skin NA
6348 Zambia Mfuwe_E Zambia NA NA Skin NA
6349 Zambia Zambezi_C Zambia NA NA Skin NA

6350 Zambia Zambezi_C Zambia NA NA Skin NA
6351 Zambia Zambezi_C Zambia -15.3 29.6 Skin NA
6352 Zambia Kasempa_NW Zambia NA NA Skin NA
6353 Zambia Kasempa_NW Zambia NA NA Skin NA
6354 Zambia Lunga_Luswishi_NW Zambia -13.3 26.6 Skin NA
6355 Zambia Livingstone_S Zambia -8.8 31.1 Skin NA

6356 Zambia Mpulungu_Port_N Zambia NA NA Skin NA
6357 Zambia Lochnivar_S Zambia NA NA Skin NA

6358 Zambia Kabwe_C Zambia -14.4 28.5 Skin NA
6359 Zambia Chiawa_C Zambia NA NA Skin NA
7246 Ghana Jang-N/R Ghana 10.2 -2.5 Dried skin (finger) 98-02-13
7465 Ghana Oku-A/R Ghana NA NA Dried skin 98-05-09
7466 Ghana Chierese-A/R Ghana NA NA Dried skin 98-05-11
7547 Ghana Kananto-N/R Ghana 9.2 -2 Dried skin (tail) 98-07-31
7548 Ghana Grupe-N/R Ghana 9.2 -2.2 Dried skin 98-08-04
7549 Ghana Grupe-N/R Ghana 9.2 -2.2 Dried skin 98-08-04
7934 Namibia Otjiworongo Namibia -20.4 16.7 NA 98-06-20
7943 Namibia Okahondja Namibia -22 16.9 NA 98-04-04
7944 Namibia Okahondja Namibia -22 16.9 NA 98-08-22
7946 Namibia Outjo Namibia -20.1 16.2 NA 96-12-30
7949 Namibia Winchoek Namibia -22.6 17.1 NA 97-07-09
8540 Uganda MFNP-Kulu_Niang Uganda 2.2 31.8 NA 96-01-15
8647 Tanzania Selous_la1 TanzaniaE -7.8 38.2 NA 98-10-24
Table S1. Sample information. Related to Figure 1 and STAR Methods.

A summary of details concerning the samples used in this study, including the country, location,
latitude (Lat), longitude (Lon), and date of collection. NA = not available.
Reads
Relatedness Fst grouping
Sequen- Sequen- mapped
Mapped % Mismatch Error rate Passed Assigned Admixture
ID cing cing Raw reads after Bases mapped
reads mapped rate in % QC population With K5 Reason to
technology depth collapsing Related Group
reads samples exclude
Population
2469 NovaSeq Low 99,045,128 98,633,702 99.58% 59,805,491 10,895,522,300 0.00699 0.0256 yes Zambia - - Clean -
D-statistics
Population
2523 NovaSeq Low 144,399,588 41,447,445 28.70% 21,043,884 3,867,756,348 0.00732 0.0275 yes Zambia - - Clean -
D-statistics
3241 HiSeq High 399,794,298 396,523,074 99.18% 235,435,378 44,107,730,686 0.00872 0.031 yes TanzaniaN 2. degree 3243 Clean TanzaniaN -
3243 NovaSeq Low 108,612,006 107,642,113 99.11% 58,336,532 11,078,206,167 0.00761 0.0287 yes TanzaniaN 2. degree 3241 Clean TanzaniaN -
4343 HiSeq High 391,370,766 387,335,417 98.97% 236,848,839 43,828,160,937 0.00732 0.0296 yes TanzaniaW - - Clean TanzaniaW -
4346 NovaSeq Low 93,204,174 92,817,742 99.59% 55,315,691 10,188,024,603 0.00697 0.0261 yes TanzaniaW - - Admixed TanzaniaW -
Too few
4352 NovaSeq Low 121,806,380 117,150,149 96.18% 72,794,903 12,995,777,125 0.00681 0.0204 yes TanzaniaE - - Clean -
samples
Too few
4354 HiSeq High 346,752,754 341,887,336 98.60% 199,720,046 37,583,702,283 0.00772 0.0298 yes TanzaniaE 1. degree 8647 - -
samples
4443 NovaSeq Low 98,873,124 98,379,526 99.50% 56,370,772 10,552,311,327 0.00736 0.0285 yes TanzaniaW - - Clean TanzaniaW -
5180 NovaSeq Low 100,065,918 98,062,666 98.00% 48,893,274 9,431,226,499 0.00796 0.0324 yes TanzaniaW - - Admixed - Admixed
5519 NovaSeq Low 96,590,796 96,074,719 99.47% 49,588,882 9,718,191,785 0.00794 0.0288 yes TanzaniaN - - Clean TanzaniaN -
5520 NovaSeq Low 99,025,642 98,565,487 99.54% 52,699,438 10,288,085,357 0.00715 0.0215 yes TanzaniaN - - Admixed - Admixed
5521 NovaSeq Low 87,514,464 86,545,744 98.89% 50,200,796 9,400,473,070 0.00711 0.0222 yes TanzaniaN - - Clean TanzaniaN -
6342 NovaSeq Low 107,875,464 107,402,929 99.56% 59,947,055 11,392,629,710 0.00735 0.0324 yes Zambia Duplicated 6342, 6358 - - Duplicated
6344, 6353,
6344 NovaSeq Low 109,918,710 109,329,150 99.46% 65,945,744 12,178,108,250 0.00720 0.048 yes Zambia Duplicated - - Duplicated
6354, 6359
Population
6346 NovaSeq Low 118,853,072 116,445,804 97.97% 70,643,222 12,975,135,203 0.00683 0.0304 yes Zambia Duplicated 6346, 6357 Clean -
D-statistics
6348, 6349,
6351
6348, 6349,
6351
6350 HiSeq High 424,583,038 229,046,746 53.95% 71,783,715 13,531,285,821 0.09552 7.7881 no - - - - - Sample QC
6348, 6349, Population

6351 NovaSeq Low 119,735,088 119,262,314 99.61% 70,586,833 13,079,820,626 0.00709 0.0252 yes Zambia Duplicated Clean -
6351 D-statistics
6352 NovaSeq Low 123,720,258 67,383,531 54.46% 21,921,352 4,111,481,895 0.09113 7.0646 no - - - - - Sample QC
6344, 6353,
6354, 6359
6344, 6353, Population
6354 NovaSeq Low 101,252,214 100,685,518 99.44% 63,299,496 11,468,018,275 0.00705 0.0386 yes Zambia Duplicated Clean -
6354, 6359 D-statistics
Population
6355 NovaSeq Low 112,682,960 110,973,803 98.48% 67,173,225 12,338,864,537 0.00688 0.03 yes Zambia Duplicated 6355, 6356 Admixed -
D-statistics
Population
6358 NovaSeq Low 81,057,162 80,646,298 99.49% 47,846,781 8,901,306,461 0.00726 0.0274 yes Zambia Duplicated 6342, 6358 Clean -
D-statistics
6344, 6353,
6354, 6359
7246 NovaSeq Low 110,795,818 110,107,570 99.38% 60,105,818 11,549,659,304 0.00737 0.0368 yes Ghana - - Clean Ghana -
7465 NovaSeq Low 87,120,618 85,401,667 98.03% 47,803,274 9,057,757,862 0.02969 -0.3104 no - - - - - Sample QC
7466 NovaSeq Low 104,826,330 101,082,320 96.43% 57,545,426 10,742,248,130 0.02968 -0.3109 no - - - - - Sample QC
7547 HiSeq High 363,332,682 360,346,512 99.18% 218,677,029 40,478,206,028 0.00824 0.0377 yes Ghana - - Clean Ghana -
7934 NovaSeq Low 92,867,046 92,400,836 99.50% 59,901,160 10,665,192,765 0.00676 0.0191 yes Namibia - - Clean Namibia -
7942 HiSeq High 386,306,320 384,192,975 99.45% 250,417,029 45,610,377,180 0.00831 0.0284 yes Namibia - - Clean Namibia -
7943 NovaSeq Low 113,128,612 112,625,847 99.56% 71,788,704 12,727,914,883 0.00702 0.0345 yes Namibia 1. degree 7944 - - Related
7944 NovaSeq Low 121,669,576 121,066,706 99.50% 77,725,221 13,610,045,948 0.00676 0.0183 yes Namibia 1. degree 7943 Clean Namibia -
7946 NovaSeq Low 109,472,012 108,969,285 99.54% 69,774,591 12,452,674,508 0.00692 0.0185 yes Namibia - - Admixed - Admixed
Too few
8540 NovaSeq Low 112,792,116 112,232,959 99.50% 61,152,842 11,687,049,416 0.00728 0.0309 yes Uganda - - Clean -
samples
8647 NovaSeq Low 104,450,130 100,266,356 95.99% 54,824,786 10,304,502,084 0.00865 0.0258 yes TanzaniaE 1. degree 4354 - - Related
Table S2. Mapping statistics. Related to STAR Methods.

Details on sequencing and the subsequent data processing, along with a detailed description of which samples have been filtered out of the
dataset and the reasons.
Proportion of the reference genome retained by filters
Percentage Percentage
Total length Total length
Name of reference of reference
(Leopard) (Cat)
(Leopard) (Cat)
Total 2,578,019,207 100.0% 2,521,863,845 100,0%
Scaffolds > 1Mb 2,491,448,152 96.6% 2,460,251,910 97.6%
Autosomes > 1Mb 2,326,901,936 90.3% 2,329,694,901 92.4%
Mappability 2,214,217,472 85.9% 2,208,371,323 87.6%
Inbreeding 2,270,669,704 88.1% 1,709,718,730 67.8%
Depth 2,203,509,920 85.5% 2,152,296,588 85.3%
Repeats 1,761,482,501 68.3% 1,431,490,556 56.8%
Combined (strictref) 1,374,856,842 53.3% 916,801,099 36.4%
Overview of sample sizes and reference genomes used

# samples Mapped to:
for each analysis
Initial dataset 53 Cat, Leopard
After error-rate filtering 49 Cat
After removing potentially duplicated samples 41 Leopard
After removing closely related individuals 39 Leopard
Inbreeding filter 41 Leopard
Sex determination 49 Leopard
Genotype likelihoods 41 Leopard
PCA 39 Leopard
ngsAdmix (minus Uganda) 38 Leopard
Heterozygosity 41 Cat
EEMS 39 Leopard
IBS 39 Leopard
D-statistics: population likeness (only low depth, minus:
29 Leopard
admixed, Uganda, TanzaniaE)
D-statistics: Asian introgression (only low depth) 34 + 2 Amur Cat
SFS, FST 26 Leopard
Genotype-call-based: ROH 5 Leopard
Genotype-call-based: PSMC 5 + published data Cat
Table S3. Proportion of the reference genome retained by filters, and an overview of
the sample sizes and reference genomes used for each analysis. Related to STAR
Methods.
In the upper part, we calculated the number of bases and proportion of the reference
genome (leopard and domestic cat) retained by each filtering step. In the lower part, we
indicated the initial number of samples, the numbers of samples left after each filtering step,
and the number of samples used as input for each analysis.
Sample ID: 3241 6350 6352 7465 7466
Panthera Panthera Panthera Leptailurus Panthera

Species with top hit
pardus pardus pardus serval pardus
Percent identity to
97.4 % 91.7 % 93.4 % 94.2 % 92.6%
top hit
Range of species for
7 2 5 2 23
hits in other Felidae
Proportion of hits to
0% 3.1 % 0% 0% 0%
non-Felidae species
Passed initial QC? Yes No No No No
Total # of hits 411 613 380 490 386
Table S4. Summary of BLAST hits for the mitochondrial genomes. Related to STAR
Methods.
Sample 3241 is included as an example of a sample that passed QC with a percent identity
higher than 95% to Panthera pardus. The other four samples analysed are all the samples
with a percent identity lower than 95% to Panthera pardus or the top BLAST hit was to
another species. Those four samples also failed the QC because of deviations in mapping
statistics, mismatch rate, and error rates. Bold cells indicate the reason for exclusion.
KING -robust Inferred
ID 1 ID 2 Country R0 R1
kinship relationship
7943 7944 Namibia 0.0006 0.4954 0.2486 PO
4354 8647 TanzaniaE 0.0635 0.6256 0.2470 FS
7548 7549 Ghana 0.0227 0.2265 0.1493 HS
3244 5521 TanzaniaN 0.1591 0.3555 0.1465 HS
3241 3243 TanzaniaN 0.1831 0.3527 0.1362 HS

7547 7549 Ghana 0.1698 0.2085 0.0997 HS
7936 7944 Namibia 0.2613 0.3177 0.0977 HS

5525 5181 TanzaniaW 0.2084 0.2213 0.0924 HS
7943 7939 Namibia 0.2703 0.2940 0.0895 HS
Table S5. Relatedness estimation. Related to STAR Methods.

Pairs of samples inferred to be second degree relatives or closer. The kinship coefficients
are based on the estimated 2d-SFS (see STAR Methods) using the methodology developed
by Waples et al. [S14]. PO = parent offspring, FS = full siblings, HS = half sibling, or other
second-degree relative.
Heterozygosity
Data Our
Sample Heterozygosity Population Species estimate Mapped to
reference estimate
reference
2469 0.002045 Zambia African Leopard This study This study Domestic cat
3241 0.001771 TanzaniaN African Leopard This study This study Domestic cat
4343 0.001962 TanzaniaW African Leopard This study This study Domestic cat
4352 0.001784 TanzaniaE African Leopard This study This study Domestic cat
7246 0.002104 Ghana African Leopard This study This study Domestic cat
7934 0.001788 Namibia African Leopard This study This study Domestic cat
8540 0.002070 Uganda African Leopard This study This study Domestic cat
African Lion1 0.0009 African Lion African Lion [S15] [S15] Domestic cat
African Lion2 0.00074 African Lion African Lion [S16] [S11] Domestic cat 0.00076
African Lion3 0.00148 African Lion African Lion [S11] Domestic cat
White Lion 0.00063 White Lion White Lion [S16] [S11] Domestic cat
Asiatic Lion 0.000276 Asiatic Lion Asiatic Lion [S17] [S17] Domestic cat
Snow Leopard 0.00043 Snow Leopard Snow Leopard [S16] [S11] Domestic cat 0.00033
Cheetah 0.00044 Cheetah Cheetah [S18] [S11] Domestic cat
Amur Tiger 0.00104 Amur Tiger Amur Tiger [S16] [S11] Domestic cat
Bengal Tiger 0.00093 Bengal Tiger Bengal Tiger [S16] [S11] Domestic cat 0.00101
White Tiger 0.00087 White Tiger White Tiger [S16] [S11] Domestic cat 0.00098
Amur Leopard1 0.00047 Amur Leopard Amur Leopard [S11] [S11] Domestic cat
Malayan Tiger 0.0008 Malayan Tiger Malayan Tiger [S19] [S15] Domestic cat
Leopard Cat1 0.00216 Leopard Cat Leopard Cat SRP059496 [S11] Domestic cat
Leopard Cat2 0.00173 Leopard Cat Leopard Cat [S11] [S11] Domestic cat
Jaguar 0.00119 Jaguar Jaguar [S12] This study Domestic cat
PumaBR338 0.00155 Puma Puma [S20] [S20] Puma
PumaBR406 0.00166 Puma Puma [S20] [S20] Puma
PumaEVG21 0.00121 Puma Puma [S20] [S20] Puma

PumaCYP47 0.00033 Puma Puma [S20] [S20] Puma
PumaCYP51 0.00034 Puma Puma [S20] [S20] Puma
PumaYNP198 0.0009 Puma Puma [S20] [S20] Puma
PumaSMM12 0.00079 Puma Puma [S20] [S20] Puma
PumaSMM22 0.00059 Puma Puma [S20] [S20] Puma
PumaSC29 0.00062 Puma Puma [S20] [S20] Puma
PumaSC36 0.00049 Puma Puma [S20] [S20] Puma
Iberian Lynx 0.000102 Iberian Lynx Iberian Lynx [S21] [S21] Iberian Lynx
Eurasian Lynx
0.000315 Eurasian Lynx Eurasian Lynx [S22] [S22] Iberian Lynx
Carpathians
Eurasian Lynx
0.000358 Eurasian Lynx Eurasian Lynx [S22] [S22] Iberian Lynx
Kirov
Eurasian Lynx
Latvia 0.000374 Eurasian Lynx Eurasian Lynx [S22] [S22] Iberian Lynx
Eurasian Lynx
Mongolia 0.000463 Eurasian Lynx Eurasian Lynx [S22] [S22] Iberian Lynx
Eurasian Lynx
Poland 0.000286 Eurasian Lynx Eurasian Lynx [S22] [S22] Iberian Lynx
Eurasian Lynx
Norway 0.000266 Eurasian Lynx Eurasian Lynx [S22] [S22] Iberian Lynx
Eurasian Lynx
Primorsky 0.000425 Eurasian Lynx Eurasian Lynx [S22] [S22] Iberian Lynx
Eurasian Lynx
Tuva 0.000489 Eurasian Lynx Eurasian Lynx [S22] [S22] Iberian Lynx
Eurasian Lynx
Urals 0.000395 Eurasian Lynx Eurasian Lynx [S22] [S22] Iberian Lynx
Eurasian Lynx
Yakutia 0.00045 Eurasian Lynx Eurasian Lynx [S22] [S22] Iberian Lynx
Table S6. Heterozygosity estimates. Related to Figure 6 and STAR Methods.

Heterozygosity was estimated for each sample as the proportion of heterozygous loci in the
obtained per-individual 1d-SFS inferred using ANGSD [S7]. For this purpose, we used the data
mapped to the domestic cat reference, to make it comparable with estimates previously published
for several other felid species. Moreover, we estimated heterozygosity from previously published
jaguar data [S12] and we re-analyzed heterozygosity in four big cat samples with publicly available
data, showing that our methodology yields estimates very close to the ones previously reported.
Supplemental References
S1. Kitchener, A.C., Breitenmoser-Würsten, C., Eizirik, E., Gentry, A., Werdelin, L., Wilting, A.,
Yamaguchi, N., Abramov, A.V., Christiansen, P., Driscoll, C., et al. (2017). A revised
taxonomy of the Felidae: The final report of the Cat Classification Task Force of the IUCN
Cat Specialist Group. Cat News.
S2. Meisner, J., and Albrechtsen, A. (2018). Inferring population structure and admixture
proportions in low-depth NGS data. Genetics 210, 719–731.
S3. Skotte, L., Korneliussen, T.S., and Albrechtsen, A. (2013). Estimating individual admixture
proportions from next generation sequencing data. Genetics 195, 693–702.
S4. Garcia-Erill, G., and Albrechtsen, A. (2020). Evaluation of model fit of inferred admixture
proportions. Mol. Ecol. Resour. 20, 936–949.
S5. Petkova, D., Novembre, J., and Stephens, M. (2016). Visualizing spatial population structure
with estimated effective migration surfaces. Nat. Genet. 48, 94–100.
S6. Purcell, S., Neale, B., Todd-Brown, K., Thomas, L., Ferreira, M.A.R., Bender, D., Maller, J.,
Sklar, P., de Bakker, P.I.W., Daly, M.J., et al. (2007). PLINK: a tool set for whole-genome
association and population-based linkage analyses. Am. J. Hum. Genet. 81, 559–575.
S7. Korneliussen, T.S., Albrechtsen, A., and Nielsen, R. (2014). ANGSD: Analysis of next
generation sequencing data. BMC Bioinformatics 15, 356.
S8. Meisner, J., and Albrechtsen, A. (2019). Testing for Hardy–Weinberg equilibrium in
structured populations using genotype or low-depth next generation sequencing data. Mol.
Ecol. Resour. 19, 1144–1152.
S9. Li, H., and Durbin, R. (2011). Inference of human population history from individual whole-
genome sequences. Nature 475, 493–496.
S10. Stein, A.B., Athreya, V., Gerngross, P., Balme, G., Henschel, P., Karanth, U., Miquelle, D.,
Rostro-Garcia, S., Kamler, J.F., Laguardia, A., et al. (2020). Panthera pardus (amended
version of 2019 assessment). The IUCN Red List of Threatened Species 2020
e.T15954A163991139.
S11. Kim, S., Cho, Y.S., Kim, H.-M., Chung, O., Kim, H., Jho, S., Seomun, H., Kim, J., Bang,
W.Y., Kim, C., et al. (2016). Comparison of carnivore, omnivore, and herbivore mammalian
genomes with a new leopard assembly. Genome Biol. 17, 211.
S12. Figueiró, H.V., Li, G., Trindade, F.J., Assis, J., Pais, F., Fernandes, G., Santos, S.H.D.,
Hughes, G.M., Komissarov, A., Antunes, A., et al. (2017). Genome-wide signatures of
complex introgression and adaptive evolution in the big cats. Sci. Adv. 3, e1700299.
S13. Ochoa, A., and Storey, J.D. (2016). FST and kinship for arbitrary population structures I:
Generalized definitions. bioRxiv 083915.
S14. Waples, R.K., Albrechtsen, A., and Moltke, I. (2019). Allele frequency-free inference of close
familial relationships from genotypes or low-depth sequencing data. Mol. Ecol. 28, 35–48.
S15. Armstrong, E.E., Taylor, R.W., Miller, D.E., Kaelin, C.B., Barsh, G.S., Hadly, E.A., and
Petrov, D. (2020). Long live the king: chromosome-level assembly of the lion (Panthera leo)
using linked-read, Hi-C, and long-read data. BMC Biol. 18, 3.
S16. Cho, Y.S., Hu, L., Hou, H., Lee, H., Xu, J., Kwon, S., Oh, S., Kim, H.-M., Jho, S., Kim, S., et
al. (2013). The tiger genome and comparative analysis with lion and snow leopard genomes.
Nat. Commun. 4, 2433.
S17. Mitra, S., Sreenivas, A., Sowpati, D.T., Kumar, A.S., Awasthi, G., Milner Kumar, M.,
Tabusum, W., and Gaur, A. (2019). De novo assembly and annotation of Asiatic lion
(Panthera leo persica) genome. bioRxiv 549790.
S18. Dobrynin, P., Liu, S., Tamazian, G., Xiong, Z., Yurchenko, A.A., Krasheninnikova, K., Kliver,
S., Schmidt-Küntzel, A., Koepfli, K.-P., Johnson, W., et al. (2015). Genomic legacy of the
African cheetah, Acinonyx jubatus. Genome Biol. 16, 277.
S19. Armstrong, E., Khan, A., Taylor, R.W., Gouy, A., Greenbaum, G., Thiéry, A., Kang, J.T.L.,
Redondo, S., Prost, S., Barsh, G., et al. (2019). Recent evolutionary history of tigers
highlights contrasting roles of genetic drift and selection. bioRxiv, 696146
S20. Saremi, N.F., Supple, M.A., Byrne, A., Cahill, J.A., Coutinho, L.L., Dalén, L., Figueiró, H.V.,
Johnson, W.E., Milne, H.J., O’Brien, S.J., et al. (2019). Puma genomes from North and South
America provide insights into the genomic consequences of inbreeding. Nat. Commun. 10,
4769.
S21. Abascal, F., Corvelo, A., Cruz, F., Villanueva-Cañas, J.L., Vlasova, A., Marcet-Houben, M.,
Martínez-Cruz, B., Cheng, J.Y., Prieto, P., Quesada, V., et al. (2016). Extreme genomic
erosion after recurrent demographic bottlenecks in the highly endangered Iberian lynx.
Genome Biol. 17, 251.
S22. Lucena-Perez, M., and Marmesat, E. (2020). Genomic patterns in the widespread Eurasian
lynx shaped by Late Quaternary climatic fluctuations and anthropogenic impacts. Mol. Ecol.
29, 812– 828.

High Genetic Diversity and Low Differentiation Reflect The Ecological Versatility of The African Leopard

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

High Genetic Diversity and Low Differentiation Reflect The Ecological Versatility of The African Leopard

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

High Genetic Diversity and Low Differentiation Reflect The Ecological Versatility of The African Leopard

Uploaded by

Copyright:

Available Formats

Article

High genetic diversity and low differentiation reflect

d Among big cats, African leopards have the highest genetic

d Gene flow on a continent-wide scale maintains low genetic

d Broad dietary and habitat niche likely explain the

Pecnerová et al., 2021, Current Biology 31, 1–10

*Correspondence: rheller@bio.ku.dk (R.H.), ida@binf.ku.dk (I.M.), kristianhanghoej@gmail.com (K.H.)

INTRODUCTION to human settlements, labeling it as a habitat and dietary gener-

(Figure 1; Table S1). We targeted 2–53

2 Current Biology 31, 1–10, May 10, 2021

A B Figure 2. Population structure

these groupings as populations and

Current Biology 31, 1–10, May 10, 2021 3

African mammals with sub-Saharan distribution, including the

Migration between Africa and Asia

Demographic history across the big cats

4 Current Biology 31, 1–10, May 10, 2021

Figure 4. Reconstruction of migration surfaces

Current Biology 31, 1–10, May 10, 2021 5

H1 H2 H3 Cat H1 H2 H3 Cat Figure 5. D-statistics with low coverage

er, a recent and densely sampled study

6 Current Biology 31, 1–10, May 10, 2021

Figure 6. Demographic history and genetic diversity

Current Biology 31, 1–10, May 10, 2021 7

d KEY RESOURCES TABLE

8 Current Biology 31, 1–10, May 10, 2021

Current Biology 31, 1–10, May 10, 2021 9

10 Current Biology 31, 1–10, May 10, 2021

KEY RESOURCES TABLE

REAGENT or RESOURCE SOURCE IDENTIFIER

Data and code availability

Current Biology 31, 1–10.e1–e5, May 10, 2021 e1

EXPERIMENTAL MODEL AND SUBJECT DETAILS

QUANTIFICATION AND STATISTICAL ANALYSES

Reference Quality filtering: Mappability and RepeatMasker

Reference Quality filtering: Global Sequencing Depth filter

Reference Quality filtering: Autosomal and sex-linked scaffold identification

e2 Current Biology 31, 1–10.e1–e5, May 10, 2021

Reference Quality filtering: Excessive heterozygosity filter

Reference Quality filtering: Summarizing reference quality filters

Sample quality filtering: Error rates estimation

Sample quality filtering: BLAST analyses of mtDNA

Sample quality filtering: Identification of related and duplicated samples

Genotype likelihoods, SNPs, and genotype calling

Current Biology 31, 1–10.e1–e5, May 10, 2021 e3

Runs of homozygosity (ROH) analysis

Site frequency spectrum

SFS-based analyses: FST estimation

SFS-based analyses: Estimation of heterozygosity

e4 Current Biology 31, 1–10.e1–e5, May 10, 2021

IBS-based analyses: Neighbor-joining (NJ) tree

IBS-based analyses: Estimation of Effective Migration Surfaces (EEMS)

Current Biology 31, 1–10.e1–e5, May 10, 2021 e5

High genetic diversity and low

Figure S7. Demographic history reconstructed assuming a generation time of 5 years

2469 Zambia Kafue_N Zambia -14.5 26.2 Dried skin 95-07-11

2523 Zambia Luangwa_N Zambia -11.9 32.3 Dried skin 95-08-08

3241 Tanzania Maswa_(Mbono) TanzaniaN -3.2 34.6 Dried skin 95-07-31

Pecnerová et al., 2021, Current Biology 31, 1–10