MEC1204.

fm Page 593 Tuesday, March 13, 2001 10:35 AM

Molecular Ecology (2001) 10, 593 – 602

Sustaining genetic variation in a small population:

Blackwell Science, Ltd

evidence from the Mauritius kestrel

R I C H A R D A . N I C H O L S , * M I C H A E L W. B R U F O R D † and J I M J . G R O O M B R I D G E ‡
*School of Biological Sciences, Queen Mary University of London, London E1 4NS, UK †Cardiff School of Biosciences,
Cardiff University, PO Box 915, Cathays Park, Cardiff, CF10 3TL, UK ‡Institute of Zoology, Regent’s Park, London
NW1 4RY, UK

Abstract
We obtained measures of genetic diversity in 10 kestrel species at a suite of 12 microsatellite
loci. We estimated the relative effective size (Ne) of the species using a Markov chain Monte
Carlo (MCMC) approach, which jointly estimated the locus specific mutation rates as nuisance
parameters. There was surprisingly high genetic diversity found in museum specimens of
the Mauritius kestrel. Being an endemic species on a small island, it is known to have a long
history of small population size. Conversely, kestrels with a continental distribution had Ne
estimates that were only one order of magnitude larger and similar to each other, despite
having current population sizes that were between one and three orders of magnitude
larger than the Mauritius kestrel. We show how many of the theoretical results describing
the effective size of a subdivided population can be captured in terms of three rates which
describe the branching pattern of the gene genealogy, and that they are useful in estimating
the time to migration-drift and mutation-drift equilibrium. We use this approach to argue
that population subdivision has helped retain genetic diversity in the Mauritius kestrel,
and that the continental species’ genetic diversity has yet to reach equilibrium after the
range changes following the last ice age. We draw parallels with Hewitt’s observation that
genetic variation seems to survive species’ range compression and is rather vulnerable to
range expansion.
Keywords: coalescent, conservation genetics, glacial refuge, metapopulation, range expansion
Received 4 June 2000; revision received 25 September 2000; accepted 25 September 2000

conservation effort since 1973 (Jones 1987 ). Contemporary

Introduction
In this paper we compare the genetic variation in 10 differ-
ent kestrel species to evaluate the effects of past and present
population size on genetic diversity. Our study species have
population ranges that differ by three orders of magnitude,
and we examine the differences in genetic diversity at
the same suite of 12 microsatellite loci in each. We pay
particular attention to the Mauritius kestrel (Falco punctatus).
The Mauritius kestrel
The Mauritius kestrel is the last surviving endemic raptor
on Mauritius, and has been the focus of an intensive
Correspondence: Dr Richard A. Nichols. Fax: 020 8983 0973;
E-mail: r.a.nichols@qmw.ac.uk
records indicate that the kestrel suffered a severe but short-
lived population bottleneck. Prior to human settlement
in 1750, the Mauritius kestrel may have been widely
distributed over the island wherever suitable forest existed
(McKelvey 1977; Jones & Owadally 1985). The population
decline seems to have paralleled the destruction of the
native forest (Jones 1987), and by 1950 the species was rare
and approaching extinction (Hachisuka 1953; Greenway
1967). Further drastic decline, as a result of widespread
organochlorine pesticide use, reduced the population to
one known breeding pair in 1974 (Safford & Jones 1997).
Since then, the managed population rapidly recovered to
222–286 individuals in 1994 (Jones et al. 1995), and has
continued to increase, with a population believed to be in
excess of 400 birds in 1997 (Safford & Jones 1997 ).
Figure 1 shows the historical decline of distribution and
population size for the Mauritius kestrel, and the capture

© 2001 Blackwell Science Ltd

MEC1204.fm Page 594 Tuesday, March 13, 2001 10:35 AM
594 R . A . N I C H O L S , M . W. B R U F O R D and J . J . G R O O M B R I D G E
Fig. 1 Distribution and population decline of the Mauritius kestrel.
locations of the museum specimens used to estimate
ancestral variation. The museum specimens, dated between
1829 and 1940, predate the reduction of the kestrel popula-
tion to extremely low numbers. These samples, therefore,
allow a direct estimate of ancestral genetic variation. The
ancestral population was itself restricted in area, as the
island is only 25 km in radius. Furthermore, it appears to
have been effectively isolated from neighbouring islands.
A molecular phylogeny based on cytochrome b indicates
that the Mauritius kestrel diverged from the Seychelles
kestrel Falco araea two million years ago (Groombridge 2000).
Interpreting the genetic data
Hewitt (1999) has pointed out that genetic differentiation
between populations of many European species seems
to arise because different regions have been populated
from different glacial refugia. Instead of arguing that the
populations went through population bottlenecks, as they
became restricted to small refugia, he turns the usual
argument on its head. He points out that bottlenecks could
occur as ranges expand and small numbers of migrants
arrive in virgin territory. One major strand of evidence for
© 2001 Blackwell Science Ltd, Molecular Ecology, 10, 593 – 602
MEC1204.fm Page 595 Tuesday, March 13, 2001 10:35 AM
H O W S M A L L W E R E PA S T P O P U L AT I O N S ? 595
this interpretation is that areas in southern Europe, that are
postulated as refuges, typically harbour more genetic
variation than more northern latitudes.
We will search for similar patterns in the kestrel data.
Persistent island populations might sustain genetic vari-
ation in a similar manner to glacial refuges. On the other
hand, large continental kestrel populations may have lost
genetic variation during range expansions. As in the case of
populations in northern Europe, there may have been
insufficient time for genetic diversity to be replenished by
mutation.
The paper is in three parts. First, we conduct a preliminary
data exploration, which summarizes the genetic variation
in the kestrel species by estimating heterozygosities. This
analysis exposes evidence that there are significant dif-
ferences in effective size between the populations and in
mutation rate between the loci. These results invalidate
the calculation of averages to combine information across
loci or populations. In the second part we, therefore, make
use of a Markov chain Monte Carlo (MCMC) analysis that
can allow for different mutation rates and summarize the
difference between species as difference in effective
population size.
The third part of the paper deals with the interpretation
of differences in effective size. The relationship between
population size, subdivision and dynamics and effective
size is not straightforward. For example, Wakeley (1999)
argued that human genetic data suggests a reduction in
effective population size. This seems at odds with the
increase in size suggested by the palaeontological and
archaeological record. He pointed out, however, that
models of subdivided populations show that an increase
in migration rate between human populations could lead
to the reduced effective size. Whitlock & Barton (1997)
developed more sophisticated models that also incorporated
fluctuating population size and came to the opposite
conclusion, that subdivided populations will often have
reduced effective size. Here we point out that the results of
these and other models can be captured by three rates that
determine branching patterns of the gene genealogy. This
approach allows a ready application of ecological insight to
evaluate which of the theoretical approaches might be
most applicable to the kestrel species, and to assess the rate
of approach to equilibrium.
Materials and methods
Extant samples and DNA extraction
Blood sampling was carried out over 3–4 breeding seasons
between 1994 and 1998 from each of the three populations
shown in Fig. 1, with the majority of the samples being
obtained from wild offspring bled at the nest. Although 350
birds have been sampled, these include several offspring
© 2001 Blackwell Science Ltd, Molecular Ecology, 10, 593–602
for many of the nests. In order to obtain independent
samples, we restricted the analysis to adults from one year:
1996. In cases where we had not bled either member of
a breeding pair, we included one of their chicks, thereby
sampling half the adult’s genes. These criteria reduced the
sample to 75 individuals.
Between four and 10 fresh blood samples from the
Seychelles kestrel, Greater kestrel Falco rupicoloides, Lesser
kestrel F. naumanni, South African rock kestrel F. tinnunculus
rupicolus and Canary Islands kestrel F. tinnunculus canariensis
were collected from the brachial wing vein, and stored
in a standard Tris:EDTA:SDS storage buffer (10 mm
Tris, 100 mm EDTA, 2% w/v SDS; pH 8.0). Two separate
sources of material were sampled for the European kestrel
F. tinnunculus tinnunculus: (i) muscle preserved in alcohol
from captive stock in Mauritius (supplied by C.G. Jones,
Mauritius Wildlife Foundation); and (ii) previously extracted
DNA from a UK source (supplied by J. Wetton, Forensic
Science Service, UK). Primary feather samples were obtained
for the Madagascar kestrel F. newtoni and the Kenyan
kestrel F. tinnunculus rufescens.
Total DNA was extracted from blood samples using pro-
teinase K digestion, phenol–chloroform purification, NaCl
extraction and ethanol precipitation. Methods followed
Ausubel et al. (1989), but were modified for avian blood
following Miller et al. (1988). Total DNA was stored in
500 µL 10 mm Tris-HCl: 1 mm EDTA (TE) buffer (pH 8.0) at
−20 °C. Methods for feather DNA extraction were similar, but
total extraction volumes were reduced from 5.0 mL to 500 µL.
Museum samples and museum DNA extraction
A total of 26 Mauritius kestrel museum specimens of
different tissue types were used in this study, including
tissue from the underside of the footpad (proximal phalanx),
dermal skin samples, muscle preserved in alcohol and
feather tips, depending on the collection from which they
were obtained. The museum specimen tissue types, iden-
tification details and catalogue numbers are recorded at
www.qmw.ac.uk/~ugbt112, and those samples with location
information are included in Fig. 1.
To optimize DNA extraction techniques for the various
tissue types, trials were carried out using material from the
European kestrel of similar antiquity (100 –170 years), in
order to compare DNA yield and microsatellite amplifica-
tion from the same tissues. Samples were extracted using
the commercially available GFX Genomic DNA Purification
Kit (Pharmacia Biotech, UK). DNA was eluted into 200 µL
of TE buffer which had been preheated to 80 °C, and stored
at −20 °C. The most reliable tissue type both for DNA yield
and polymerase chain reaction (PCR) amplification was
footpad skin, as has been found in another study on the
genotyping of avian museum specimens from different
tissue sources (Mundy et al. 1997).
MEC1204.fm Page 596 Tuesday, March 13, 2001 10:35 AM
596 R . A . N I C H O L S , M . W. B R U F O R D and J . J . G R O O M B R I D G E
Microsatellite amplification of extant DNA
The DNA samples were screened using a set of 10 poly-
morphic dinucleotide microsatellite markers cloned from
the Peregrine Falcon (M. Nesje, GenBank accession nos
AF118420 –118434: Locus 5, 13, 31, 46–1, 79–1, 82–2, 86–2,
89, 92–1, 107), and a further two loci, a hexanucleotide and
tetranucleotide, from the European kestrel (J. Wetton,
unpublished; Locus Fu1, Fu2).
PCR amplification of the extant DNA was carried out
following end-labelling of the forward primer with
[γ32P]-dATP according to Ellegren (1991). PCR (10 µL)
was performed in 96-well microtitre plates using the
following reaction mix; 0.75 Units Taq polymerase (Gibco,
UK), 100 µm of each dNTP, 3.0 mm NH4 buffer (Gibco,
UK), 1.5 mm MgCl2, 0.5 pmol of each primer and approx-
imately 50 ng of DNA per reaction in a Hybaid thermal
cycler (Hybaid, UK). Cycling parameters were as follows:
94 °C for 3 min, then 30 cycles of 94 °C for 45 s, 50 °C for
45 s and 72 °C for 45 s, finishing with 72 °C for 10 min.
Products were mixed with 95% deionized formamide
and xylene cyanol/bromophenol-blue dye mixture,
preheated to 94 °C and loaded onto 6% denaturing
polyacrylamide gels to resolve the PCR fragments. The
sizes of alleles were initially determined by electro-
phoresis alongside the product of a DNA sequencing reac-
tion, and from then on by internal comparison of known
size PCR products. Gels were dried at 80 °C and the prod-
ucts visualized by autoradiography for between 12 and
48 h.
Microsatellite amplification of museum DNA
PCR amplification of museum DNA extractions was
carried out using the same method, but with the following
adjustments; all reagents and tubes were exposed to
ultraviolet light for 25 min prior to assembly, to degrade
any contaminating DNA. Annealing temperatures for all
reactions were lowered to 48 °C (36 cycles) in order to
increase the likelihood of PCR amplification from DNA. To
confirm PCR amplification and control for allelic drop-out,
each PCR was repeated at three different DNA dilutions
(1:1, 1:5 and 1:10), and run in consecutive lanes separated
by a blank lane. PCR products of extraction blanks were
also included for each set of reactions, with similar dilution
treatments. A negative PCR control was also included
for each set of reactions. PCR products were visualized
by autoradiography using exposures of 2–10 days. PCR
amplification of all museum DNA extractions was
repeated at least twice. Correct identification of alleles, as
opposed to spurious PCR product, was confirmed by the
presence of ‘slippage’ products typical of microsatellite
amplification by Taq polymerase (Litt & Luty 1989; Luty
et al. 1990).
Estimates of heterozygosity
An initial data exploration was conducted by estimating
heterozygosities. Likelihood curves were obtained for each
combination of locus and species using a simple recursive
function of F, the complement of heterozygosity (eqn 1
below). The same function has been used to estimate the
parameter FST in a subdivided population (Balding & Nichols
1997). For a variety of models of mutation, the probability
that the (n + 1)th gene drawn from the population is an
allele of type a depends on F and the number of a alleles
previously seen in the sample, xa.
xa F + ( 1 – F ) pa
P (a xa, n) = ----------------------------------- (1)
nF + (1 – F)
where pa is the probability that a mutation event generates
an a allele. Balding & Nichols (1995) pointed out that
setting pa = 0 gives the Ewens’ sampling formula, suitable
for the infinite alleles model of mutation (Ewens 1979);
whereas pa = 1/k gives the probabilities for the multinomial
Dirichlet distribution, appropriate for the k-alleles model
(Wright 1969). The probability for each sample, given F, can
be obtained by arranging the genes in arbitrary order and
calculating the product of eqn 1 for each gene in the sequence.
This probability can be obtained for values of F between
0 and 1, giving a likelihood curve for each locus in each
population. A preliminary evaluation of differences in
heterozygosity between species or between loci can be
obtained by combining the likelihood curves. Curves for
each species were obtained by multiplying the likelihoods
for every locus for each value of F. This procedure would
be appropriate for unlinked loci with the same mutation
rate. The result is shown in Fig. 2(a) (for the infinite alleles
model). The curves are standardized to have the same area,
so that the heights can be treated as probability densities
given a uniform prior. As expected there is a general trend
for the species thought to have larger population size to
have higher heterozygosities.
In order to evaluate differences in mutation rate between
loci, the curves for the continental species were combined
for each locus. Differences in mutation rate can be detected
as curves with minimal overlap. Any differences could be
obscured if the effective size of the different species were
substantially different, as this would reduce the precision
of the locus-specific estimates. This is why we restricted the
comparison to the continental species, which appear to
have similar effective population size (Ne) as indicated by
their coincident curves in Fig. 2(a) (similar heterozygosity
at the same loci implies similar Ne ).
Estimating relative population size
The preliminary analysis provided evidence of significant
deviations in heterozygosity between loci and between
© 2001 Blackwell Science Ltd, Molecular Ecology, 10, 593 – 602
MEC1204.fm Page 597 Tuesday, March 13, 2001 10:35 AM
H O W S M A L L W E R E PA S T P O P U L AT I O N S ? 597
species (see below). In addition, there were missing data
for some combinations of loci and species that may have
biased the estimates. We, therefore, make use of a method
that can deal with these problems. It was used (but not
described) by Groombridge et al. (2000), to identify the
restored Mauritius and the Seychelles populations as
outliers having heterozygosities substantially lower than
would be predicted from the size of the species’ range. The
method deals with differences between loci and species by
obtaining joint estimates of the size, Ne , for each species
(plus the restored population) and the mutation rate, µ, for
each locus. It is not possible to estimate the absolute Ne
values accurately without knowing the mutation rates
accurately, because the likelihood calculated using eqn 1
depends on the product Neµ. For example, in the case of the
infinite alleles model of mutation, the product enters eqn 1 as
1 recorded. The relative densities of the parameter ∆Ni are
F = ----------------------- . Nevertheless, it is possible to obtain accurate
1 + 4N e µ
estimates of the relative Ne, because the same loci were
typed across species.
The analysis, therefore, estimated a parameter that spe-
cified the relative population size of species i, ∆Ni. This
parameter is defined in terms of a deviation from an aver-
age across species, N , such that the effective size of each
species is given by log(Nei) = N + ∆Ni . Similarly the relative
mutation rate for each locus j, ∆µj, was estimated, where
the locus specific mutation rate is given by log(µj) = µ +
∆µj. The parameters N , µ and the deviations ∆Ni for each
population and ∆µj for each locus were estimated using the
Metropolis algorithm (Metropolis et al. 1953; Gelman et al.
1995). This is an MCMC method that starts from an arbi-
trary combination of parameter values and steps through
parameter space. The stepping rule is designed so that
parameter values are chosen with a probability propor-
tional to their likelihood. The ∆Ni and ∆µj were allowed to
vary by ±5 loge units from their respective means. At each
© 2001 Blackwell Science Ltd, Molecular Ecology, 10, 593–602
Fig. 2 (a) Heterozygosity estimates for each
population. Dark-shaded curves are the
mainland kestrel populations; from left to
right, Madagascan, Kenyan, Greater, Canary
Islands, European, South African, Lesser
kestrel. The white curve is for the Seychelles
population. ( b) Heterozygosity estimates
for each locus (mainland populations only).
iteration, all the parameter values were perturbed to new
values drawn from a uniform distribution centred on the
old values. The width of each distribution was one tenth of
the parameter’s range. The likelihood, Lnew, was calculated
according to eqn 1 for the new parameter combination. The
total likelihood was obtained as the product of the values
for each locus and each population. The species repres-
ented by only single individuals (Kenya and Madagascar)
were excluded.
If Lnew was greater than the likelihood for the previous
combination, Lold, then the algorithm stepped to the new
values. If it was smaller, then the step to the new values
was accepted with probability Lnew/Lold, otherwise the old
values were retained.
After a burn-in period of 25 000 iterations, the distribu-
tion of parameter values for the next 250 000 iterations was
shown for each species in Fig. 3. They are equivalent to
the posterior distributions, assuming a uniform prior on
the ∆Ni.
Results and Discussion
Genetic diversity
Table 1 summarizes the results as sample size, mean allelic
diversity and observed heterozygosity (H) for each popu-
lation. The total number of alleles per locus and the size
ranges are given in Table 2. We detected seven polymorphic
loci from the ancestral samples of the Mauritius kestrel but
only three in the restored population. Comparable data for
the continental species (Table 1), shows polymorphism for
a higher proportion of loci.
There is an effect of sample size, most noticeable in the
estimates of mean allele number from those species rep-
resented by a single individual. Figure 2(a) deals with this
MEC1204.fm Page 598 Tuesday, March 13, 2001 10:35 AM

598 R . A . N I C H O L S , M . W. B R U F O R D and J . J . G R O O M B R I D G E

Fig. 3 Markov chain Monte Carlo estimates

of ∆Ne: the deviation in Log(Ne) from the
population average. The curves for each
species are in the same sequence as in Fig. 2
except that those based on the small samples
from Madagascar and Kenya have been
excluded. The results shown here were
calculated using the likelihood function for
the k-alleles model.

Table 1 Summary of genetic diversity in each sample

Mauritius Ancestral Seychelles Madagascar SA Rock European Kenyan Canary Is. Greater Lesser

Sample number n = 75 n = 26 n=4 n=1 n = 10 n = 10 n=1 n=8 n = 10 n=8

Loci polymorphic 3 7 3 4 11 12 6 12 9 11
Mean no. alleles 1.41 3.10 1.25 1.33 5.00 5.50 1.50 4.41 4.50 5.41
Observed heterozygosity 0.099 0.231 0.062 0.333 0.525 0.608 0.500 0.583 0.400 0.614
No. unique alleles — 3 2 0 8 7 1 2 12 17

Table 2 Allelic diversity and size range of each microsatellite locus

Locus 5 13 31 46–1 79–1 82–2 86–2 89 92–1 107 Fu1 Fu2

Alleles 9 7 9 13 7 6 8 7 17 27 10 31
Size* 94 –118 91–111 124 –144 115–137 132–142 126–138 132–146 119–131 100–138 187 – 291 96 –140 168 – 360

*allele size as total PCR amplified fragment.

bias by estimating the underlying genetic diversity in the
species rather than that of the samples. The mode (peak) of
each curve is a maximum likelihood estimate. The esti-
mates from Mauritius and the Seychelles are clearly lower
than the continental species. The differences are significant
using Edwards’ (1972) support limits test (the estimates for
two species are considered different if there is no single hetero-
zygosity value at which the heights of both curves are within
two loge likelihood units of their peaks). The continental
species provide broadly similar estimates with overlapping
likelihood curves despite their very different population
sizes. This result was unexpected, as larger populations are
expected to carry greater heterozygosity at equilibrium.
The locus-specific curves, in Fig. 2(b), indicate that two
loci have significantly higher heterozygosities than most of
the others, suggesting higher mutation rates. This result
demonstrates a violation of one of the assumptions that
went into the construction of the species curves in Fig. 2(a):
that the mutation rates are equal. This finding was the
main reason for using the MCMC method to assess the
validity of the initial conclusions.
The estimates of ∆Ne obtained from the MCMC method
are shown in Fig. 3 for the k-alleles model (the results for
the infinite-alleles model are essentially indistinguishable).
As reported by Groombridge et al. (2000), the genetic data
show a markedly lower effective size in the restored popula-
tion, consistent with the historical record of a severe bottle-
neck, and evidence of a similar event on the Seychelles.
More important, from the point of view of this paper, is the
relative size of the ancestral Mauritius and continental
populations.
The continental species have Ne estimates only 10 times
larger than the ancestral Mauritius kestrel (note the scale of
Fig. 3 is in natural logarithms). Theoretical descriptions of

© 2001 Blackwell Science Ltd, Molecular Ecology, 10, 593 – 602

MEC1204.fm Page 599 Tuesday, March 13, 2001 10:35 AM
H O W S M A L L W E R E PA S T P O P U L AT I O N S ? 599
Table 3 Geographic range of each population (km2)*
Lesser Kestrel 25 000 000
Greater Kestrel 500 000
European Kestrel 42 000 000
Kenyan Kestrel 8 000 000
South African Rock Kestrel 4 000 000
Canary Islands Kestrel 91 000
Madagascar Kestrel 600 000
Mauritius Kestrel 1 865
Seychelles Kestrel 288
*Data from Village (1990).
populations at equilibrium suggest that Ne is proportional
to the total species population size (see eqn 3 below) at
equilibrium. This relationship is not seen with current area
of the species range given in Table 3, which are up to three
orders of magnitude larger. These areas only give a rough
guide to the expected Ne at equilibrium as the species’ bio-
logy and distributions differ in several ways. For example,
the nesting pattern varies from regular spacing in the
Greater Kestrel (Falco rupicoloides) to colonial nesting in the
Lesser Kestrel (F. naumanni). Nevertheless, the European
Kestrel’s range is nearly two orders of magnitude larger
than the Greater Kestrel’s and, additionally, estimates of its
density are 10 times higher (2 vs. 0.2 per km2; Kemp 1978;
Village 1990). It does not seem possible to explain the sim-
ilarity of the Ne estimates in terms of the current population
size and distribution.
The calculations of ∆Ne involved some assumptions. For
example, there was an implicit uniform prior probability
for each species’ ∆Nj , that the loci have the same mutation
rate in each species, the mutation model and that the
information from different loci and species is independent
(conditional on the values of the population sizes and
mutation rates). The most questionable of these is the
mutation model, as the mutation process for microsatel-
lites is not exactly known. The broad picture of a 10-fold
difference seems robust, however. For example, another
calculation is to substitute the maximum likelihood
estimates for heterozygosity from Fig. 2 into the stand-
ard expressions for the mutation parameter θ (θ = 4Ne µ).
(1 − H) = (1 + θ)–1 for the infinite alleles model or (1 + 2θ )– 0.5
for the ladder model (a stepwise mutation model, Kimura
1983). Under the assumption of equal mutation rates, the
ratio of θ values between species will be the ratio of Ne
values. The ratio between median continental values and
ancestral Mauritius is approximately 14 for the ladder
model and eight for the infinite-alleles model, correspond-
ing closely with our estimate of 10.
There are two aspects of the results that call for explana-
tion: the small difference in Ne estimates between Mauritius
and the continental species and the even smaller differ-
ences between the continental species. Is the amount of
© 2001 Blackwell Science Ltd, Molecular Ecology, 10, 593–602
Fig. 4 The three rates that characterize the genealogical structure
of a finite island model. The population is composed of D demes
and N diploid individuals. Each individual migrates with
probability m per generation. For simplicity the expressions for
rates have excluded multiple events in one generation.
genetic variation observed in the ancestral Mauritius
samples unusually high? If we substitute a mutation rate
for dinucleotide repeats of 2 × 10 – 4 (Cavalli-Sforza 1998)
into the expression for heterozygosity we obtain an Ne
of around 430. In many wild species the population size
is around 10 times the Ne (Frankham 1995), so this would
correspond to a population of over 4000. In the most
populated areas today, the density is around one pair
per square kilometre, so this would require the whole
1865 km2 of the island to be populated at peak density;
which seems unlikely, especially over extended periods.
The mutation rate was estimated for human dinucleotide
loci and so this absolute population size estimate must
be treated with caution. Rates are known to vary between
species and even alleles at the same locus ( Jin et al. 1996).
When it comes to the continental species, however, there
are clear discrepancies between Ne estimates and current
population size.
Such discrepancies can arise if the population is sub-
divided, or prone to extinction of local demes, or has not
reached equilibrium between genetic drift and mutation.
In the following section we show how many of the existing
theoretical results describing these effects can be captured
in terms of three rates which describe the branching of the
gene genealogy. We find that this approach helps us apply
our knowledge of the species ecology to interpret the
genetic patterns we have observed in the kestrel data.
Characterizing gene genealogies in subdivided
populations
Figure 4 illustrates the island model of population sub-
division. We can characterize the effects of subdivision by
considering the events that occur as we trace the genealogy
of a pair of genes back through time. The pattern is specified
by three rates: the rate at which lineages from the same
deme coalesce (a), or are separated by a migration event (b)
and the rate at which lineages from different demes arrive
in the same deme (c).
MEC1204.fm Page 600 Tuesday, March 13, 2001 10:35 AM
600 R . A . N I C H O L S , M . W. B R U F O R D and J . J . G R O O M B R I D G E
Because genes from the same deme have a higher prob-
ability of sharing a common ancestor than those from
different demes, there will be genetic differentiation between
the demes. The magnitude of the differentiation is deter-
mined by the relative probability, Φ, that two genes from
the same deme can trace their ancestry directly back to a
common ancestor in that deme. This probability is deter-
mined by the relative size of a and b. Simply substituting in
the expressions from Fig. 4 gives an expression for Φ in
terms of deme size and migration rate:
a 1
φ = ---------- = -------------------- . (2)
a + b 1 + 4Nm
The parameter Φ can be thought of as the value estim-
ated by the familiar statistic FST. Wright (1969) shows
that, at migration-drift equilibrium, FST estimates the
right-hand term if the mutation rate is low and the number
of islands high.
For the purposes of explaining the kestrel data, we are
interested in explaining the effective size of the different
species. This can be thought of as the idealized panmictic
population that would retain the same amount of genetic
diversity. We can use the parameter Φ to quantify the
effect of subdivision on the effective size, Ne. In the canon-
ical panmictic population, the loss of genetic diversity is
1 last generation, or the higher rate (1/2k) if it did and thereby
characterized by the coalescence rate, ---------, for of a pair of
2N e
genes. In the subdivided population coalescence between
genes within demes is rapid, and so the longer-term effects
are dominated by the coalescence rate for two lineages in
different demes. The coalescence rate is approximately
given by the rate at which they trace back to the same deme,
multiplied by the probability that they then coalesce: cΦ.
We can, therefore, obtain an expression for Ne by sub-
stituting the values of a, b and c given in Fig. 4 into
1 Charlesworth (1999) who investigated the effect of extinc-
--------- = cφ;
2N e
(3)
⇒N e = ND1 + ------------ .
1 ing the extinction rate can either increase or decrease Fst
 4Nm
This expression was obtained by Wakeley (1999), by a
different route. We find it helpful to formulate it in terms a,
b and c because we can use our ecological knowledge of
kestrels to evaluate these rates in less idealized situations,
and also reason about the approach to equilibrium when
population structure is disrupted (below). The arrangement
in eqn 3 emphasizes that Ne increases with population
subdivision (i.e. reduced m), and is equivalent to a result
found by Wright: Ne = ND/(1-FST). Could the unexpectedly
high genetic diversity of the ancestral kestrel population be
explained by population subdivision?
Whitlock & Barton (1997 ) are doubtful about the effective-
ness of subdivision in promoting diversity. They examined
more general models of metapopulation structure; extending
the simple island model to quantify the effects of variation
in deme size, deme extinction and recolonization. They
conclude Ne will commonly decrease with population
structure rather than increase as implied by eqn 3. Their
results can be understood by looking at the implications of
more realistic population structure on the rate cΦ.
Extinction and recolonization can be introduced using
an approach proposed by Slatkin (1977 ). Briefly, a propor-
tion, e, of demes go extinct each generation leaving D′
extant demes. The empty demes are recolonized by k
diploid founders and instantaneously grow up to full size.
The founders can be specified to be from the same or differ-
ent source demes, or some combination. Migration and
drift occur as before. The rates a, b and c can be recalculated
for this model. For example, in the case where founders
are drawn at random from all extant demes
1–e e
a ≈ ---------- + ----- ,
2N 2k
b ≈ 2m + (1 – 2m)e  --------------   --------------  ,
2k – 2 D′ – 1
 2k – 1   D′ 
1 – [(1 – e)(1 – m)]
2
c ≈ ----------------------------------------------.
D′
The expression for a is composed of two parts, the normal
coalescent rate if the deme did not experience an extinction
experienced a bottleneck of size k. Similarly the rate at
which one of a pair of lineages traces back to a different
deme is the normal expression for migration (2m) plus the
rate at which they trace back to different demes through
an extinction event. Finally, at rate c lineages in different
demes can trace back to the same deme either through
migration or extinction events.
Substituting these expressions into eqn 2 and eqn 3 gives
close approximations to the curves produced by Pannell &
tion on Fst and Ne in this model (one difference seems to be
due to our assumption of diploid recolonization). Increas-
(depending on k) and yet decrease Ne substantially in this
case because of the increase in c. Although Ne is readily
reduced below the value for an idealized panmictic popu-
lation, this is not to say that population subdivision reduces
Ne. The opposite is arguably the case: Ne is a decreasing
function of migration so, if the population dynamics are
unaltered, increased isolation increases Ne.
Whitlock & Barton (1997) found a similar reduction of Ne
in a model where demes vary in their contribution to future
generations. In that case c is inflated by a term proportional
to the variance. The rate is higher because there is a greater
chance that migrants and colonists trace back to the same
deme as they are more likely to both be descendants of one
of the demes with higher productivity. It remains the case
that reduced migration increases Ne. The converse is,
however, found in some cases where migration changes
© 2001 Blackwell Science Ltd, Molecular Ecology, 10, 593 – 602
MEC1204.fm Page 601 Tuesday, March 13, 2001 10:35 AM
H O W S M A L L W E R E PA S T P O P U L AT I O N S ? 601
the population dynamics. They give a combination of
parameters for population growth rate and carrying capa-
city for which increased migration increases the census size
of the population, because more demes are extant, and
decreases the variance (in the contribution of each deme)
by causing their sizes to fluctuate in synchrony. Both effects
contribute to increasing Ne with increased m. In theory,
then, population subdivision could either increase or
decrease effective population size. In practice, the outcome
will depend on a species population dynamics.
Interpretation of kestrel data
These theoretical results help interpret the unexpectedly
high ancestral genetic diversity in the Mauritius kestrel
population. We have ecological data that suggests popu-
lation subdivision. The Mauritius kestrel shows poor dispersal,
even across suitable habitat (Jones et al. 1995). Long-term
ringing studies have not detected mixing between the
three current populations on Mauritius (Fig. 1). The theory
suggests that subdivision is unlikely to have increased Ne
above the census size, but it could have helped sustain
more genetic diversity than in a comparable undivided
population. On the contrary, reduced diversity would be
expected if subdivision affected the population dynamics
by reducing the total number of kestrels or increasing the
variability in success of demes. These effects appear unlikely
in the case of the kestrel’s ecology: in the re-established
populations (which are limited in size and isolated)
population density appears strongly regulated by the
availability of nest-sites and territories.
The difference between the island and continental spe-
cies may reflect the time it takes for the pattern of genetic
diversity to reach equilibrium. In the case of differentiation
among demes, the equilibrium is between coalescence and
movement between demes. These occur at rates a and b
so the expected time for one or other to occur is (a + b)–1
generations. In an island model this will be of the order of
deme size, and Slatkin & Barton (1989) show that FST
reaches equilibrium over this time for a variety of popula-
tion structures. Similarly heterozygosity is determined by
the balance between mutation, occurring at rate ≈2µ, and
coalescence between lineages in different demes, at rate
≈cΦ. Equilibrium is therefore attained in roughly (cΦ + 2µ )–1
generations, which will be of the order of 2Ne.
For the island species we have estimated this to be a few
hundred generations, a period over which the population
structure may have been relatively constant. The continental
species have much larger populations that are unlikely to
have reached equilibrium since the perturbations intro-
duced by the climatic fluctuations associated with the ice
ages. The species range will have changed dramatically,
with retreats into refugia and subsequent expansions. The
example of the Mauritius species suggests that small refugial
© 2001 Blackwell Science Ltd, Molecular Ecology, 10, 593–602
populations are capable of sustaining considerable vari-
ation. Dramatic loss of variation can, however, occur during
range expansion (Nichols & Hewitt 1994; Ibrahim et al.
1996; Hewitt 1999), especially if the expansion involves long
distance migrants into virgin territory. In that case, a few
far-dispersing individuals can establish a population that
will expand to cover a large area in the virgin territory. A
small number of founders or slow growth rate can lead
to genetic impoverishment: Nichols & Beaumont (1996)
found that a population with k founders growing at rate r
under a rain of M migrants per generation has inbreed-
ing coefficient ≈(1 + r)/(1 + kr + 4M). If subsequent range
expansion is from such impoverished populations, genetic
variation will be rapidly lost. Genetic diversity will re-
establish at a rate limited by mutation. The similar genetic
diversities for the continental kestrel populations with very
different current sizes can be explained by a comparable
postglacial history from which they have yet to recover.
The trends that are found in these kestrel species appear
to be part of a broader tendency. Hewitt (1999) reviewed
DNA sequence evidence from a range of European plant
and animal species that had glacial refugia in southern
peninsulas. He suggests that genetic variation can persist
over several ice ages in the refugia. Rather than being
vulnerable to range compression, genetic diversity seems
to be lost during expansion into northern latitudes. The
persistence of diversity in refugia may be due, in part,
to the ability of populations to track suitable habitat by
moving up and down in mountainous regions as the
climate fluctuates. The mountainous topography will also
lead to population subdivision, which can play a part in
sustaining diversity. This proposal might be verified by
molecular phylogenetic surveys within refugial regions,
which could provide evidence of ancient but local geo-
graphical differentiation.
Acknowledgements
Primers were kindly provided by David Parkin, John Wetton
(University of Nottingham) and Marit Nesje ( Norwegian School
of Veterinary Science, Norway). Access to museum samples was
kindly provided by the British Museum of Natural History
(Robert Prys-Jones), Cambridge University Museum of Zoology
(Adrian Friday and Mike Brookes), Museum Nationale D’Histoire
Naturelle, Paris, Museum D’Histoire Naturelle, Geneva, Switzer-
land, Mauritius Institute, Mauritius, and Naturhistorisches Museum
Wien (Vienna), Austria. Anthony van Zyl, Welcome Glen, South
Africa collected most of the mainland kestrel population samples.
Funding: Durrell Wildlife Conservation Trust, Jersey; Institute of
Zoology, London; Mauritius Wildlife Foundation (Dr Carl Jones);
and The Peregrine Fund, USA.
References
Ausubel FM, Brent R, Kingston RE et al. (1989) Current Protocols in
Molecular Biology. John Wiley & Sons, New York.
MEC1204.fm Page 602 Tuesday, March 13, 2001 10:35 AM
602 R . A . N I C H O L S , M . W. B R U F O R D and J . J . G R O O M B R I D G E
Balding DJ, Nichols RA (1995) A method for quantifying differenti-
ation between populations at multi-allelic loci and its implications
for investigating identity and paternity. Genetica, 96, 3–12.
Balding DJ, Nichols RA (1997) Significant genetic correlations
among Caucasians at forensic DNA loci. Heredity, 78, 583–589.
Cavalli-Sforza LL (1998) The DNA revolution in population
genetics. Trends in Genetics, 14, 60 –65.
Edwards AWF (1972) Likelihood: an Account of the Statistical Concept
of Likelihood and its Application to Scientific Inference. Cambridge
University Press, Cambridge.
Ellegren H (1991) DNA typing of museum birds. Nature, 354, 113.
Ewens WJ (1979) Mathematical Population Genetics. Springer-
Verlag, Berlin.
Frankham R (1995) Effective population size / adult population size
ratios in wildlife: a review. Genetical Research (Camb.), 66, 95 –107.
Gelman A, Carlin JB, Stern HS, Rubin DR (1995) Bayesian Data
Analysis. Chapman & Hall, London.
Greenway JC (1967) Extinct and vanishing birds of the world, 2nd
edn. Dover, New York.
Groombridge JJ (2000) Conservation genetics of the Mauritius kestrel,
pink pigeon and echo parakeet. Unpublished PhD Thesis, QMW
University of London, London.
Groombridge JJ, Jones CG, Bruford MW, Nichols RA (2000)
‘Ghost’ alleles of the Mauritius kestrel. Nature, 403, 616.
Hachisuka M (1953) The Dodo and kindred birds or the extinct birds of
the Mascarene Islands. HF & G Witherby, London.
Hewitt GM (1999) Post-glacial re-colonization of European biota.
Biological Journal of the Linnean Society, 68, 87–112.
Ibrahim KM, Nichols RA, Hewitt GM (1996) Spatial patterns of
genetic variation generated by different forms of dispersal
during range expansion. Heredity, 77, 282–291.
Jin L, Macaubas C, Hallmayer J, Kimura A, Mignot E (1996) Muta-
tion rate varies among alleles at a microsatellite locus: Phylogenetic
evidence. Proceedings of the National Academy of Sciences of the
USA, 93, 15285 –15288.
Jones CG (1987) The larger land-birds of Mauritius. In: Studies
of Mascarene Island Birds (ed. AW, Diamond), pp. 208 – 300.
Cambridge University Press, Cambridge.
Jones CG, Heck W, Lewis RE, Mungroo Y, Slade G, Cade T (1995)
The restoration of the Mauritius kestrel Falco punctatus popu-
lation. Ibis, 137, 173 –180.
Jones CG, Owadally AW (1985) The status ecology and conserva-
tion of the Mauritius kestrel. In: Conservation Studies on Raptors
(eds I, Newton RD, Chancellor), pp. 211–222. International
Council for Bird Preservation, Cambridge.
Kemp AC (1978) Territory maintenance and use by breeding
Greater Kestrels. In: Proceedings of the Symposium on African
Predatory Birds (ed. Kemp AC), pp. 71–76. Northern Transvaal
Onithological Society, Pretoria.
Kimura M (1983) The Neutral Theory of Molecular Evolution. Cam-
bridge University Press, Cambridge.
Litt M, Luty JA (1989) A hypervariable microsatellite revealed
by invitro amplification of a dinucleotide repeat within the
cardiac-muscle actin gene. American Journal of Human Genetics,
44, 397–401.
Luty JA, Guo Z, Willard HF, Ledbetter DH, Ledbetter S, Litt M
(1990) Five polymorphic microsatellite VNTRs on the human
X-chromosome. American Journal of Human Genetics, 46, 776 –
783.
McKelvey DS (1977) The Mauritius kestrel some notes on its
breeding biology behaviour and survival potential. Hawk Trust
Annual Report, 8, 19–21.
Metropolis N, Rosenbluth AW, Teller AH, Teller E (1953) Equation
of state calculations by fast computing machines. Journal of
Chemical Physics, 21, 1087.
Miller SA, Dykes DD, Polesky HF (1988) A simple salting out pro-
cedure for extracting DNA from human nucleated cells. Nucleic
Acids Research, 16, 12–15.
Mundy NI, Unitt P, Woodruff DS (1997) Skin from feet of museum
specimens as a non-destructive source of DNA for avian geno-
typing. Auk, 114, 126–129.
Nichols RA, Beaumont MA (1996) Is it ancient or modern history
that we can read in the genes?. In: Aspects of the Genesis and
Maintenance of Biological Diversity (eds Hochberg ME, Clobert J,
Barbault R), pp. 69–87. Oxford University Press, Oxford.
Nichols RA, Hewitt GM (1994) The Genetic Consequences of
Long-Distance Dispersal During Colonization. Heredity, 72,
312–317.
Pannell JR, Charlesworth B (1999) Neutral genetic diversity in a
metapopulation with recurrent local extinction and recoloniza-
tion. Evolution, 53, 664–676.
Safford RJ, Jones CG (1997) Did organochlorine pesticide use
cause declines in Mauritian forest birds? Biological Conservation,
6, 1445–1145.
Slatkin (1977) Gene flow and genetic drift in a species subject to
frequent local extinction. Theoretical Population Biology, 12, 253 –
262.
Slatkin M, Barton NH (1989) A comparison of 3 indirect methods
for estimating average levels of gene flow. Evolution, 43, 1349 –
1368.
Village A (1990) The Kestrel. T & AD Poyser Publishers, Berkham-
stead, Herts.
Wakeley J (1999) Nonequilibrium migration in human history.
Genetics, 153, 1863–1871.
Whitlock MC, Barton NH (1997) The effective size of a subdivided
population. Genetics, 146, 427–441.
Wright S (1969) Evolution and the Genetics of Populations, Vol. 2: the
Theory of Gene Frequencies. University of Chicago Press, Chicago.
RAN was a PhD student and postdoctoral researcher with
Godfrey Hewitt. JJG was cosupervised by RAN, MWB and Carl
Jones of the Mauritius Wildlife Foundation.
© 2001 Blackwell Science Ltd, Molecular Ecology, 10, 593 – 602