See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/221837800

Rapid evolution of an adaptive cyanogenesis cline in introduced North

American white clover (Trifolium repens L.)

Article in Molecular Ecology · February 2012

DOI: 10.1111/j.1365-294X.2012.05486.x · Source: PubMed

CITATIONS READS

46 249

2 authors:

Nicholas J Kooyers Kenneth M Olsen

University of Louisiana at Lafayette Washington University in St. Louis
106 PUBLICATIONS 464 CITATIONS 345 PUBLICATIONS 6,879 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Arabidopsis thaliana Kas-1 apetala 1 (APA) gene, partial cds View project

Allele frequency spectrum View project

All content following this page was uploaded by Nicholas J Kooyers on 27 February 2018.

The user has requested enhancement of the downloaded file.

Molecular Ecology (2012) 21, 2455–2468 doi: 10.1111/j.1365-294X.2012.05486.x

Rapid evolution of an adaptive cyanogenesis cline

in introduced North American white clover (Trifolium
repens L.)
N I C H O L A S J . K O O Y E R S and K E N N E T H M . O L S E N
Department of Biology, Washington University in St Louis, One Brookings Drive, Campus Box 1137, St. Louis, MO 63130,
USA

Abstract
White clover is polymorphic for cyanogenesis (HCN production after tissue damage), and
this herbivore defence polymorphism has served as a classic model for studying adaptive
variation. The cyanogenic phenotype requires two interacting biochemical components;
the presence ⁄ absence of each component is controlled by a simple Mendelian gene
(Ac ⁄ ac and Li ⁄ li). Climate-associated cyanogenesis clines occur in both native (Eurasian)
and introduced populations worldwide, with cyanogenic plants predominating in
warmer locations. Moreover, previous studies have suggested that epistatic selection may
act within populations to maintain cyanogenic (AcLi) plants and acyanogenic plants that
lack both components (acli plants) at the expense of plants possessing a single
component (Acli and acLi plants). Here, we examine the roles of selection, gene flow and
demography in the evolution of a latitudinal cyanogenesis cline in introduced North
American populations. Using 1145 plants sampled across a 1650 km transect, we
determine the distribution of cyanogenesis variation across the central United States
and investigate whether clinal variation is adaptive or an artefact of population
introduction history. We also test for the evidence of epistatic selection. We detect a clear
latitudinal cline, with cyanogenesis frequencies increasing from 11% to 86% across the
transect. Population structure analysis using nine microsatellite loci indicates that the
cline is adaptive and not a by-product of demographic history. However, we find no
evidence for epistatic selection within populations. Our results provide strong evidence
for rapid adaptive evolution in these introduced populations, and they further suggest
that the mechanisms maintaining adaptive variation may vary among populations of a
species.
Keywords: adaptive cline, balanced polymorphism, cyanogenesis, epistatic selection, parallel
evolution, Trifolium repens

Received 20 October 2011; revision received 25 November 2011; accepted 6 December 2011

useful model systems because the patchy complexity of

Introduction
landscapes is reduced into a simpler gradient where it
Clinal variation, a shift in phenotypes or allele frequen- is easier to associate adaptive traits and their underly-
cies over geographic space, has frequently been used as ing genetic variation with environmental variation.
a model to investigate how species respond to heteroge- Recent studies have examined the genetic basis of adap-
neous selection pressures in the presence of gene flow tive clines in diverse traits, including pigmentation
(e.g. Hiesey et al. 1942; Antonovics & Bradshaw 1970; (Hoekstra et al. 2004; Mullen & Hoekstra 2008; Gross
Lenormand et al. 1999; Saccheri et al. 2008). Clines are et al. 2009; Steiner et al. 2009; Rosenblum et al. 2010),
morphology (Huey et al. 2000; McKechnie et al. 2010),
Correspondence: Kenneth M. Olsen, Fax: 314 935 4432; life history (Caicedo et al. 2004) and physiology (Storz
E-mail: kolsen@wustl.edu et al. 2007; Zhen & Ungerer 2008). These studies have

 2012 Blackwell Publishing Ltd

2456 N . J . K O O Y E R S and K . M . O L S E N

provided new insights into the spatial and temporal study takes advantage of a well-studied polymorphism
scales over which local adaptation can take place. Spe- for cyanogenesis (HCN release in response to tissue
cifically, this research indicates that local adaptation can damage) in white clover (Trifolium repens L.) to examine
occur on short time scales (e.g. Umina et al. 2005; Sacc- the roles of coarse-grained and fine-grained selective
heri et al. 2008), that adaptive clines can evolve in par- factors, as well as the potential role of demographic his-
allel in different portions of a species range (e.g. tory, in shaping clinal variation.
Oakeshott et al. 1982; Huey et al. 2000; Phifer-Rixey
et al. 2008) and that this process of parallel cline evolu-
Cyanogenesis in white clover
tion can occur through multiple, independent molecular
mechanisms (e.g. Steiner et al. 2009; Rosenblum et al. The white clover cyanogenesis polymorphism is consid-
2010). ered a textbook example of selectively maintained adap-
For populations occurring along an adaptive cline, tive variation (e.g. Dirzo & Sarukhan 1984; Silvertown
genotype frequencies within a given population are & Charlesworth 2001). Cyanogenic and acyanogenic
potentially determined by two types of processes: those plants occur together in many natural populations, and
reflecting coarse-grained (regional-scale) factors and the relative proportions of the two phenotypes correlate
those reflecting fine-grained (within-population) closely with environmental variation. Specifically, there
dynamics. Coarse-grained factors operate in all adaptive is a higher frequency of cyanogenic plants in warmer
clines and are the primary focus of most studies of climates, creating a pattern of climate-associated clines
clinal variation; these include interpopulation gene flow in both native (Eurasian) and introduced populations
and selection associated with regional-scale environ- around the world. Previous studies have documented
mental heterogeneity. Fine-grained factors that can latitudinal cyanogenesis clines across Europe (Daday
affect adaptive polymorphism levels within populations 1954a) and North America (Daday 1958), and altitudinal
of a cline include disruptive (diversifying) selection and clines in the Swiss Alps (Daday 1954b), Wales (de Ara-
directional selection corresponding to fine-scale spatial ujo 1976), the French Alps (Till-Bottraud et al. 1988), the
or temporal environmental heterogeneity (e.g. Weinig Himalayas (Majumdar et al. 2004) and the Pacific Cas-
et al. 2003; Caicedo et al. 2004; Kroymann & Mitchell- cades (Ganders 1990). Although white clover occurs in
Olds 2005; Todesco et al. 2010; reviewed by Charles- temperate regions worldwide today, it has been intro-
worth 2006; Hedrick 2006; Hudson & Kane 2009). One duced into most of its current range within the last
particular genetic mechanism by which polymorphism 500 years. Given that this is a potentially long-lived
can be maintained within populations is epistatic selec- perennial species, the recurrent and rapid emergence of
tion, whereby selection favours specific allele combina- cyanogenesis clines suggests intense climate-related
tions at unlinked loci. For example, selection for selection associated with the cyanogenesis polymor-
alternative flowering time phenotypes in Arabidopsis phism. However, the population structure of white clo-
has been shown to arise through epistatic selection for ver in these regions is unknown, and a formal
specific allele combinations at two unlinked genes (Cai- comparison to neutral genetic variation is necessary
cedo et al. 2004). before concluding that cyanogenesis clinal variation is
In addition to coarse- and fine-grained selective pro- adaptive.
cesses, clines can also potentially reflect nonadaptive It is generally hypothesized that white clover cyano-
processes, such as independent population introduc- genesis clines are products of an ecological trade-off
tions and ⁄ or range expansion (Clegg et al. 2002; Vase- between selection for resistance to herbivores and selec-
magi 2006; Keller & Taylor 2008). Comparison of tion against the costs of cyanogenesis in colder climates
putatively selected genes with unlinked neutral markers (reviewed by Hughes 1991). Ecological studies con-
can be used to assess the degree to which observed ducted over the last several decades have provided sub-
clinal variation is actually adaptive vs. a product of stantial evidence that cyanogenic plants are
demographic history and resulting population struc- differentially protected from small generalist herbivores,
ture. including slugs, snails, insects and voles (e.g. Ennos
Despite a well-developed theoretical literature on 1981; Dirzo & Harper 1982a,b; Kakes 1989; Pederson &
cline evolution (e.g. Fisher 1937; Haldane 1948; Endler Brink 1998; Saucy et al. 1999; Viette et al. 2000). The
1977), very few studies have empirically examined the predominance of acyanogenic plants in colder climates
relative roles of coarse-grained and fine-grained selec- has been proposed to reflect the decreased benefits of
tive factors in shaping allele or trait frequencies along energetic investment in chemical defence in areas of
adaptive clines. To do so requires a well-characterized low herbivore pressure (Kakes 1989). Consistent with
adaptive polymorphism, preferably with a simple this hypothesis, acyanogenic plants in greenhouse and
genetic basis, which also shows clinal variation. This field experiments have been reported to flower earlier

 2012 Blackwell Publishing Ltd

 C L I N E E V O L U T I O N I N I N T R O D U C E D P O P U L A T I O N S 2457

and more prolifically than cyanogenic plants in the this hypothesis, some polymorphic mid-cline popula-
absence of herbivores (Daday 1965; Kakes 1989), sug- tions in Europe have been reported to show strong link-
gesting a competitive advantage under these conditions. age disequilibrium (LD) between the two unlinked
An alternative ‘autotoxicity’ hypothesis proposes that cyanogenesis genes, with a statistical excess of AcLi
cyanogenic plants may be directly selected against in and acli cyanotypes (i.e. positive LD) among adult
colder climates, because frost-induced tissue damage plants (Ennos 1982; Kakes 1987); this pattern potentially
could lead to cyanide poisoning within the plant suggests pervasive epistatic selection against Acli and
(Daday 1965; Brighton & Horne 1977; but see Olsen & acLi plants. If epistatic selection were a general phe-
Ungerer 2008). nomenon affecting cyanotype frequencies in white clo-
The cyanogenic phenotype in clover requires two ver populations, this fine-grained selective force would
components that are separated in intact tissue and be predicted to generate positive LD among adult
brought together with tissue damage that causes cell plants in most polymorphic populations.
rupture: cyanogenic glucosides (stored in the vacuoles In this study, we examine a continent-wide latitudinal
of photosynthetic tissue) and their hydrolysing enzyme transect of introduced U.S. white clover populations to
linamarase (stored in the cell wall) (reviewed by investigate the roles of coarse-grained environmental
Hughes 1991). Acyanogenic phenotypes can arise heterogeneity, fine-grained epistatic selection and demo-
through the absence of either or both of the two graphic history in shaping cyanogenesis cline evolution.
required components. Classical genetics studies (e.g. We address the following questions: (i) Is the U.S. lati-
Corkill 1942; Atwood & Sullivan 1943) revealed that tudinal cyanogenesis cline adaptive, or is it an artefact
two independently segregating Mendelian genes can of white clover’s introduction history into North Amer-
account for these biochemical polymorphisms. The gene ica? and (ii) If the cline is adaptive, can cyanogenesis
Ac describes the presence ⁄ absence of cyanogenic gluco- frequencies in U.S. populations be attributed solely to
sides, and Li describes the presence ⁄ absence of linamar- coarse-grained dynamics (interpopulation gene flow
ase. For both genes, a homozygous recessive genotype and regional-scale environmental gradients), or does
corresponds to the absence of the cyanogenic compo- epistatic selection within populations also play a role?
nent. Thus, plants with at least one dominant (func- We find evidence for striking clinal variation in these
tional) allele at both Ac and Li are cyanogenic (the AcLi introduced populations that is not a by-product of
cyanogenesis phenotype, or ‘cyanotype’), while the demographic history. However, unlike populations in
other cyanotypes (acLi, Acli and acli) lack one or both the native European range, we find no evidence for epi-
components and correspond to acyanogenic pheno- static selection or other fine-scale selective processes
types. The molecular bases of both biochemical poly- within these North American populations. These find-
morphisms have been recently described: Ac ⁄ ac ings suggest that coarse-grained factors alone are suffi-
corresponds to a gene presence ⁄ absence polymorphism cient to maintain the Ac ⁄ ac and Li ⁄ li adaptive
at the locus CYP79D15, which encodes the P450 protein cyanogenesis polymorphisms in North America.
catalysing the first step in the cyanogenic glucoside bio-
synthetic pathway (Olsen et al. 2008); Li ⁄ li corresponds
Materials and methods
to an unlinked gene presence ⁄ absence polymorphism at
Li, which encodes the linamarase glycoprotein precur-
Sampling
sor (Olsen et al. 2007). These loci are known to occur in
different linkage groups, and genetic mapping is in pro- Trifolium repens has been widely introduced from its
gress (K. Olsen, unpublished observations). native Eurasian range into mesic temperate regions
The fact that the clover cyanogenesis polymorphism worldwide (Kjærgaard 2003). It is cultivated as a forage
arises through two independently segregating simple crop and also occurs commonly in lawns, roadsides
Mendelian genes provides a unique opportunity to and other mowed or grazed areas. The species is a self-
study the role of epistatic selection in the evolution of incompatible perennial, which spreads vegetatively
this adaptive polymorphism. Plants that produce one through stolons. For this study, introduced white clover
cyanogenic component without the other may be mal- populations were sampled along a latitudinal transect
adaptive because they bear some of the energetic costs spanning the central United States (Fig. 1). Seven
of cyanogenesis but without the herbivore defence ben- regions were sampled: New Orleans, LA; Jackson, MS;
efits (Kakes 1989). Thus, in polymorphic populations, Grenada, MS; Memphis, TN; Cape Girardeau, MO;
there may be epistatic selection favouring plants that St. Louis, MO; and Wausau, WI. Collections in each
are either cyanogenic (AcLi cyanotype) or lacking in region were made within a geographical area of
both cyanogenic components (acli cyanotype) at the approximately 20 · 20 km (except in the New Orleans
expense of acLi and Acli cyanotypes. Consistent with region where sampled sites broadly surround Lake

 2012 Blackwell Publishing Ltd

2458 N . J . K O O Y E R S and K . M . O L S E N

populations were calculated by the same method but

using the geographical mean centre of subpopulations
WI
for each population, as determined with the MeanCen-
ter function in ArcMap 9.3.1 (ESRI, Redlands, CA).
Because cyanogenesis allele frequencies may be more
closely correlated with local climate than latitude (Da-
STL day 1965; Hughes 1991), we also estimated minimum
CG winter temperature for each subpopulation using GIS
data (extracted from bio6 in the BIOCLIM dataset; Hij-
MTN
mans et al. 2005). Minimum winter temperature for
MSG
each population was calculated by averaging values
MSJ
across subpopulations.
LA

Cyanogenesis gene analyses

Cyanotypes were determined for collected plants using
Feigl-Anger HCN detection assays (Feigl & Anger
Fig. 1 Locations of sampled clover populations. Each dot cor-
1966), following the protocol described by Olsen et al.
responds to a sampled subpopulation within one of seven pop-
ulations (south to north): New Orleans, LA (LA); Jackson, MS (2007, 2008). Briefly, fresh leaves were collected from
(MSJ); Grenada, MS (MSG); Memphis, TN (MTN); Cape Girar- each plant to provide six approximately identical leaf
deau, MO (CG); St. Louis, MO (STL); Wausau, WI (WI). The tissue samples (typically three leaves used per sample).
blue to red gradient indicates minimum winter temperature as Pairs of duplicate samples were then tested for HCN
determined from the BIOCLIM data set (Hijmans et al. 2005). production under one of three treatments: leaf tissue
tested alone (to identify AcLi cyanotypes); leaf tissue
plus exogenous cyanogenic glucoside solution (to iden-
Pontchartrain). In each region, plants were initially col- tify acLi cyanotypes); or leaf tissue plus exogenous lina-
lected in 15 separate sampling sites (‘subpopulations’), marase solution (to identify Acli cyanotypes). Leaves
each site encompassing an area of approximately were frozen, thawed and mashed with a pipet tip to
50 · 50 m; ten stolon cuttings were collected per sub- ensure cell rupture. Samples were incubated with test
population, yielding 150 individuals per region and paper for four hours at 37 C, as longer incubations can
1050 individuals (Table S1, Supporting information). potentially generate false positives for acLi cyanotypes
An additional 15 subpopulations were later sampled in (resulting from the gradual hydrolysis of exogenously
the St. Louis region, yielding a total of 120 subpopula- added cyanogenic glucosides; data not shown). We
tions (1200 individuals in total, 1145 of which survived tested for experimental error by rephenotyping 40
transplantation and were used in analyses). Most sam- plants and confirmed complete reproducibility. Any
pling sites were in public parks, schools and athletic plants showing discrepancies in results between the
fields; within each sampling site, sampled individuals two duplicate samples for each test treatment were
were separated by at least 3 m to prevent the collection retested until we were confident of the cyanotype. We
of multiple ramets from a single genet (Gliddon & Sa- also tested whether Feigl-Anger results were appropri-
leem 1985). Plants from New Orleans, St. Louis and ate proxies for the presence ⁄ absence of the CYP79D15
Wausau were collected in early summer 2008, and and Li loci using PCR assays for the presence ⁄ absence
remaining regions were collected in early summer 2009. of each gene (see Olsen et al. 2007, 2008 for PCR proto-
Collected stolon cuttings were planted in MetroMix360 cols); a sample of 54 plants representing all cyanotypes
soil medium in 4¢¢ pots and maintained on a mist bench was tested. We found a false-negative error rate (i.e.
in the greenhouse of Washington University for 3– failure to detect a cyanogenic component despite the
5 days, after which plants were grown under standard presence of the corresponding gene) of <5% (data not
greenhouse conditions. Because we observed no genetic shown). Thus, for the rest of this study, we use the clas-
differentiation among subpopulations within regions sical gene terminology (Ac ⁄ ac and Li ⁄ li) with the
(see Results), we refer to groups of subpopulations assumption that the dominant and recessive alleles cor-
within regions as ‘populations’ hereafter. respond at the molecular level to the presence and
Geographical distances among subpopulations were absence of each cyanogenesis gene.
calculated from GPS coordinates for each sampling site Cyanotype frequencies (AcLi, Acli, acLi and acli) and
using a modified Haversine Formula as implemented in inferred Ac and Li genotype and allele frequencies were
GenAlEx 6 (Peakall & Smouse 2006). Distances among calculated for individual subpopulations and for popu-

 2012 Blackwell Publishing Ltd

 C L I N E E V O L U T I O N I N I N T R O D U C E D P O P U L A T I O N S 2459

lations. Frequencies of homozygous recessive genotypes from Hardy–Weinberg equilibrium using exact tests
(acac or lili) can be calculated directly from cyanotype implemented in Arlequin 3.5.1.2 (Excoffier et al. 2005).
data, as they correspond to the absence of individual We used 95% confidence intervals followed by Bonfer-
cyanogenic components. Allele and genotype frequen- roni corrections as criteria for significant evidence of
cies for Ac ⁄ ac and Li ⁄ li were inferred using homozy- null alleles.
gous recessive genotype frequencies under the To assess the distribution of neutral variation among
assumption of Hardy–Weinberg equilibrium (p2 + 2pq + subpopulations and populations, we used both Fst-like
q2 = 1, where q2 is the frequency of homozygous reces- and Rst-like AMOVA settings in Arlequin 3.5.1.2 (Excoffier
sive genotypes). Calculations were performed both by et al. 2005) and tested for population differentiation
hand and using GenAlEx 6. based on a frequency distribution created from 1000
permutations of the data. For all subpopulations and
populations, we also calculated pairwise Fst (Reynolds
Neutral marker analyses
et al. 1983) and Rst (Slatkin 1995) in Arlequin 3.5.1.2
To assess population structure across the sampled tran- and pairwise upt using GenAlEx 6. These measures all
sect, we genotyped a subset of individuals from each provided quantitatively similar estimates of population
population using primer pairs for nine microsatellite differentiation (see Results). Unlike Fst and Rst, upt can
loci that have been found to be polymorphic in white be calculated for both codominant markers (such as mi-
clover (C. Brummer, pers. comm.; Noble Foundation, crosatellites) and dominant markers (such as the cyano-
Ardmore, OK) (Table S2, Supporting information). We genesis genes), and so we focused on this measure of
selected 4–7 subpopulations for each population at ran- differentiation in the comparisons described elsewhere.
dom and genotyped all sampled individuals that sur-
vived transplantation within them, yielding
Comparison of population differentiation at neutral
approximately 50 individuals genotyped per population
markers and cyanogenesis genes
(349 individuals in total, representing 38 subpopula-
tions; Table S3, Supporting information). DNA was To test for signatures of isolation by distance in micro-
extracted with DNeasy Plant Mini kits (Qiagen, Valen- satellite data and the cyanogenesis genes, we performed
cia, CA) following the manufacturer’s protocols using Mantel tests and partial Mantel tests using the program
50 mg young leaf tissue from each plant. PCR condi- zt (Bonnet & Van de Peer 2002). Statistical significance
tions followed the protocol developed by Brummer and of correlations between matrices was determined
colleagues (Noble Foundation), including use of a PIG- through Fisher’s exact tests following 100 000 permuta-
tail (Brownstein et al. 1996) and an M13 fluorescently tions of the matrices (see Bonnet & Van de Peer 2002
labelled tag (Schuelke 2000). Amplified products were for exact methods and formulas). We tested for isolation
multiplexed and diluted to 1:300 before being geno- by distance in neutral markers by comparing the geo-
typed on an ABI 3100 sequencer at Washington Univer- graphical distance matrix between subpopulations with
sity in St. Louis. All microsatellite alleles were scored a linearized pairwise upt matrix computed from micro-
using a ROX 400HD ladder, and the resulting fragment satellite genotype frequencies. We investigated clinal
lengths were analysed using GeneMapper version 3.7 variation in the cyanogenesis genes by comparing the
(Applied Biosystems, Foster City, CA). We hand- geographical distance matrix with linearized pairwise
checked every allele call and coded any sample with upt matrices constructed from cyanotype frequencies
more than two strong peaks as missing data. Microsat- and from the individual Ac ⁄ ac and Li ⁄ li genes. Because
ellite reactions that did not amplify were repeated up to we did not have microsatellite genotypes for every indi-
three times before being coded as missing data. All loci vidual for which we had cyanotype data, we used only
had <5% missing data. those populations where we had both microsatellite
Genotypes were exported from GeneMapper and con- genotypes and cyanotypes (38 subpopulations in total).
verted to GenePop format using GMCONVERT 0.32 To explicitly test for greater isolation by distance at the
(Faircloth 2006). Population tags were added to the cyanogenesis genes compared to neutral markers, we
GenePop file, and CONVERT 1.31 (Glaubitz 2004) was used partial Mantel tests to test for correlations between
used to produce Arlequin and revised GenePop files. geographical distance and differentiation at cyanogene-
CONVERT 1.31 was also used to tabulate the number sis loci while controlling for neutral marker genetic dis-
of private alleles in each subpopulation. We used tances. We also used a matrix of pairwise differences in
MICRO-CHECKER 2.2.3 (Van Oosterhout et al. 2004) to minimum winter temperature in place of the geographi-
check for the evidence of null alleles within each popu- cal distance matrix for comparison to the microsatellite
lation via 1000 Monte Carlo simulations of the alleles in and cyanogenesis gene matrices. Further, to test for any
each population. Further, we checked for deviations evidence of adaptive divergence at Ac ⁄ ac and Li ⁄ li with-

 2012 Blackwell Publishing Ltd

2460 N . J . K O O Y E R S and K . M . O L S E N

out regard to geography, we compared pairwise upt populations (Table 1; see also Table S1, Supporting
values between the cyanogenesis genes and the micro- information). Across the sampled transect, a strong lati-
satellite loci. Under neutral expectations, microsatellite tudinal cline is apparent in the frequency of cyanogenic
pairwise upt values should be equal to pairwise upt plants; AcLi cyanotype frequencies were 85.8% in the
values at each cyanogenesis gene; however, if the two New Orleans, LA population and steadily declined to
cyanogenesis genes are under selection for adaptive 11.0% in the Wausau, WI population (Table 1; Fig. 2).
divergence, we expect greater values for the cyanogene- Among the three acyanogenic cyanotypes, weak clinal
sis genes. variation is apparent for the Acli and acli cyanotypes,
with both showing an increase in frequency towards
higher latitudes. In contrast, the frequency of acLi cya-
Testing for epistatic selection
notypes is highest near the centre of the transect and
To investigate the potential role of epistatic selection in declines to the north and south (Fig. 2).
shaping cyanotype frequencies, we tested for linkage We used linear regressions to investigate spatial dis-
disequilibrium (LD) between the Ac ⁄ ac and the Li ⁄ li tributions of the Ac ⁄ ac and Li ⁄ li allele frequencies and
genes. A statistically significant excess of AcLi and acli the cyanogenic phenotype (AcLi cyanotype) as a func-
cyanotypes and a corresponding deficit of Acli and acLi tion of sampling site across the transect. Correlations
cyanotypes (i.e. positive LD) would be consistent with are highly statistically significant (P < 0.001 in all cases),
the patterns of epistatic selection previously reported both for regressions against geographical distance along
for some white clover populations (Ennos 1982; Kakes the transect and against minimum winter temperature
1987). Because LD between unlinked genes can also (Figs 3 and S1, Supporting information). As a further
arise as an artefact of pooling subpopulations that are assessment of clinal variation at the cyanogenesis genes,
genetically differentiated, we tested for LD both at the we performed Mantel tests to compare matrices of pair-
population level (pooling across subpopulations) and wise differences between sampling sites (using either
within individual subpopulations. LD between Ac and geographical location or minimum winter temperature)
Li was calculated from cyanotype frequencies using with matrices of pairwise differentiation at the cyano-
Hill’s (1974) maximum likelihood estimator (D), ^ follow- genesis genes (using upt values for Ac ⁄ ac, Li ⁄ li, or cya-
ing the method employed previously in white clover notype). All comparisons yielded statistically significant
(Ennos 1982; Kakes 1987). This method for calculating associations (P < 0.05 in all tests; Table 2).
LD does not depend on our inferred allele or genotype
frequencies, only on the empirically observed cyanotype
Microsatellite variation and population structure
frequencies.
All microsatellite loci were polymorphic in every sub-
population sampled (Table S3, Supporting information).
Results
The number of alleles per locus ranged from 3 to 16,
with 90 alleles detected in total. Additional summary
Cyanogenesis genes show latitudinal clinal variation
statistics are presented in Table S3 (Supporting infor-
across the United States
mation). Not all loci were in Hardy–Weinberg equilib-
Both the Ac ⁄ ac and Li ⁄ li genes were polymorphic in rium within every population. After implementing
most subpopulations sampled in the seven surveyed Bonferroni corrections for multiple tests, four of nine

Table 1 Summary of cyanogenesis polymorphism data by population

Latitude Average minimum AcLi Ac allele Li allele

Region N (N) winter temp. (C) cyanotype frequency frequency frequency ^
D

New Orleans, LA (LA) 141 30.22 4.59 0.86 0.72 0.73 )0.01 (ns)
Jackson, MS (MSJ) 146 32.36 1.22 0.47 0.46 0.41 )0.02 (ns)
Grenada, MS (MSG) 142 33.83 )0.8 0.42 0.33 0.44 )0.05 (ns)
Memphis, TN (MTN) 147 35.02 )1.26 0.33 0.29 0.43 0.00 (ns)
Cape Girardeau, MO (CG) 135 37.31 )5.36 0.24 0.34 0.27 0.03 (ns)
St. Louis, MO (STL) 298 38.65 )6.76 0.15 0.24 0.16 )0.01 (ns)
Wausau, WI (WI) 136 44.93 )16.77 0.11 0.30 0.13 0.01 (ns)

Allele frequencies for Ac and Li were calculated with the assumption of Hardy–Weinberg equilibrium within populations. Linkage
disequilibrium values were calculated directly from observed cyanotype frequencies. No linkage disequilibrium values are
statistically significantly different from zero.

 2012 Blackwell Publishing Ltd

 C L I N E E V O L U T I O N I N I N T R O D U C E D P O P U L A T I O N S 2461

this excess was consistent with the presence of null

alleles at these loci (Tables S3 and S4, Supporting infor-
mation). We therefore ran all subsequent analysis with
two data sets, the first including all microsatellites and
the second excluding the four loci with greatest evi-
dence of null alleles (TRSSRB03C06, TRSSRA05E05,
TRSSRA04G02, TRSSRA02B11; see Table S2, Supporting
information). There were no qualitative differences in
any analysis, and the quantitative differences did not
affect any of the conclusions of this study. Further,
when we used the null allele correction method of Cha-
puis & Estoup (2007), we observed no differences in
downstream analysis. Results reported below are for all
nine microsatellite loci.
Fig. 2 Cyanotype frequencies in each population along the
AMOVAs using both Rst-based genetic distances (which
sampled U.S. latitudinal transect. Population abbreviations cor-
respond to Fig. 1. incorporate information on repeat number differences
among alleles) and Fst-based measures (based on allele
identity only) reveal that there is little overall geograph-
loci displayed an excess of homozygotes in at least 15% ical structure in neutral marker variation across the
of the subpopulations surveyed. Analysis with MICRO- sampled North American transect. Almost all of the
CHECKER (Van Oosterhout et al. 2004) indicated that variation is distributed within individual subpopula-

Fig. 3 Regression of population frequencies of the Li allele (A), Ac allele (B) or the AcLi cyanotype (C) against average minimum
winter temperature for the region where samples were collected. Statistical significance of linear regressions is indicated as follows:
*0.01 < P < 0.05, **0.01 < P < 0.001 and ***P < 0.001.

Table 2 Results of Mantel and partial Mantel tests comparing differentiation at the cyanogenesis genes or microsatellite loci with
geographical and climatic data for sampled populations

Matrix 1 Matrix 2 Control Matrix r P

Geography Ac ⁄ ac — 0.239 0.0017

Geography Li ⁄ li — 0.345 0.0003
Geography AcLi cyanotype — 0.303 0.0020
Geography Microsatellites — 0.483 0.0001
Minimum Winter Temperature Ac ⁄ ac — 0.203 0.0051
Minimum Winter Temperature Li ⁄ li — 0.335 0.0004
Minimum Winter Temperature AcLi cyanotype — 0.280 0.0008
Minimum Winter Temperature Microsatellites — 0.476 0.0001
Geography Ac ⁄ ac Microsatellites 0.153 0.0152
Geography Li ⁄ li Microsatellites 0.185 0.0047
Geography AcLi cyanotype Microsatellites 0.164 0.0151
Minimum Winter Temperature Ac ⁄ ac Microsatellites 0.112 0.0570
Minimum Winter Temperature Li ⁄ li Microsatellites 0.176 0.0083
Minimum Winter Temperature AcLi cyanotype Microsatellites 0.139 0.0363

 2012 Blackwell Publishing Ltd

2462 N . J . K O O Y E R S and K . M . O L S E N

Table 3 Hierarchical analysis of molecular variance (AMOVA) based on nine microsatellite markers

Source of variation d.f. Sum of squares Variance components Percentage of variation

Among populations 6 932.05 1.23 5.19***

Among subpopulations within populations 31 1023.80 0.6 2.54
Within subpopulations 660 14478.88 21.93 92.27***
Total 697 16434.72 23.78

Variances were calculated using the ‘Rst-like’ AMOVA option in Arlequin 3.5.1.2 (Excoffier et al. 2005). Asterisks denote statistical
significance of variation components as follows: *0.01 < P < 0.05, **0.01 < P < 0.001 and ***P < 0.001.

tions rather than among subpopulations within popula-

tions or among populations (92.27% for Rst-based AM-
OVA; Table 3). Within populations, there is no more
variation partitioned among subpopulations than would
be expected by chance, indicating an absence of any
detectable population structure at this geographical
scale (Table 3). Lack of population structure within
populations is also evident from very low pairwise
genetic distances between subpopulations within popu-
lations (mean upt = 0.030). Pairwise genetic distances
among all subpopulations across the entire transect also
reveal very low differentiation, even between sites
located more than 1600 km apart (mean Rst = 0.066,
mean upt = 0.050; see Table S4, Supporting informa-
tion). Nonetheless, the AMOVAs reveal significantly more
Fig. 4 Regressions of linear pairwise upt values between sam-
variation distributed among populations than would be pling sites for microsatellite data, Ac ⁄ ac, Li ⁄ li and AcLi cyano-
expected by chance (5.19%, P < 0.0001 for Rst-based AM- type against pairwise minimum winter temperature data.
OVA; Table 3), a pattern consistent with some degree of Statistical significance of regressions is indicated as follows:
population differentiation on a continent-wide scale. In *0.01 < P < 0.05, **0.01 < P < 0.001 and ***P < 0.001.
addition, we find a statistically significant correlation
between pairwise genetic distances (upt) among sub- favouring adaptive divergence at the cyanogenesis
populations and geographical distances (Mantel’s genes. These findings indicate that the strong clinal var-
r = 0.473, P = 0.01), indicating some isolation by dis- iation observed at the cyanogenesis genes is not an arte-
tance across the transect. fact of demographic history.
To test whether the isolation by distance detected at
the microsatellite markers is sufficient to account for the
No evidence for epistatic selection within populations
cyanogenesis gene clinal variation, we used partial
Mantel tests, where a pairwise geographical distance To assess the potential role of epistatic selection in
matrix (or a pairwise minimum winter temperature dif- maintaining the cyanogenesis polymorphism, we mea-
ference matrix) was compared with pairwise linear upt sured the degree of linkage disequilibrium (LD)
matrices from Ac ⁄ ac, Li ⁄ li and AcLi cyanotype while ^ Hill
between the Ac ⁄ ac and Li ⁄ li genes (calculated as D;
controlling for a pairwise linear upt matrix from micro- 1974) for each sampled population and for individual
satellite data. We find that after controlling for popula- subpopulations within populations. In contrast to native
tion structure, there is still statistically significant clinal European populations, where a positive association
variation in Ac (Mantel’s r = 0.15, P = 0.015), Li (Man- between the dominant Ac and Li alleles has been
tel’s r = 0.19, P = 0.005) and the AcLi cyanotype (Man- reported for some populations (Ennos 1982; Kakes
tel’s r = 0.16, P = 0.015) (Table 2, Fig. 4). Similarly, we 1987), none of the North American populations have D ^
find that pairwise upt values among subpopulations values that are statistically significant from zero
are statistically significantly greater for the cyanogenesis (Table 1). At the level of individual subpopulations, 11
genes than for microsatellite loci (Ac: t = 16.0, d.f. = 702, of 120 sampled sites show statistically significant values
P < 0.0001; Li: t = 17.6, d.f. = 702, P < 0.0001; Fig. S2, ^ however, all of these have an excess of Acli and
of D;
Supporting information), consistent with selection acLi cyanotypes (i.e. negative linkage disequilibrium

 2012 Blackwell Publishing Ltd

 C L I N E E V O L U T I O N I N I N T R O D U C E D P O P U L A T I O N S 2463

(LD) between Ac and Li; Table S1, Supporting informa- long as interpopulation gene flow is not so great as to
tion), which is the opposite of the pattern expected overwhelm local adaptation (Endler 1977).
under the epistatic selection hypothesis proposed for Our study is not the first to document the North
white clover. American white clover cyanogenesis cline. Daday (1958)
surveyed 3203 plants from 38 populations between
southern Louisiana and Alaska, 36 of which were poly-
Discussion
morphic at both Ac and Li. That study revealed similar
The white clover cyanogenesis polymorphism has long clinal variation to the patterns we observe, although
served as a model system for understanding how adap- Daday observed universally higher Ac and Li allele fre-
tive variation is maintained within and between popula- quencies at a given minimum winter temperature, as
tions. Here, we have investigated the evolutionary well as greater variation in cyanogenesis frequencies
processes underlying the establishment and maintenance among the northern populations that he sampled.
of cyanogenesis clinal variation in introduced North It is important to note that observations of clinal varia-
American populations. In the native Eurasian species tion in putatively adaptive traits do not by themselves
range, both adaptation in response to coarse-grained spa- demonstrate adaptive cline evolution. For recently estab-
tial environmental heterogeneity and fine-grained epi- lished populations in particular, the history of population
static selection have been proposed to maintain the Ac ⁄ ac introduction can easily create clines that are reflections of
and Li ⁄ li cyanogenesis polymorphisms (Daday 1954a, demographic history rather than local adaptation (Caval-
1958, 1965; Ennos 1982; Kakes 1987). We find that while li-Sforza et al. 1993; Clegg et al. 2002; Vasemagi 2006;
the North American latitudinal cline is consistent with Keller & Taylor 2008; Keller et al. 2009). In the case of
adaptation and is not an artefact of introduction history white clover, the close correlation between climate and
(Figs 2 and 4; Table 3), there is no evidence for within- cyanogenesis in the native Eurasian range could have led
population epistatic selection as a factor in the mainte- to the differential introduction of acyanogenic, cold-
nance of cyanogenesis variation (Tables 1 and S3, Sup- adapted clover from northern Europe into the northern
porting information). Thus, our findings suggest that United States and warm-adapted, cyanogenic clover
coarse-grained spatial environmental heterogeneity alone from the Mediterranean into the southern United States.
is sufficient to explain the origin and persistence of the If this introduction pattern did in fact occur, a near-
cyanogenesis polymorphism in the United States. Below, panmictic level of subsequent gene flow has all but elimi-
we discuss the implications of our findings for studies of nated any initial population structure (see Tables 3 and
adaptive cline evolution and specifically for the clover S4, Supporting information). Thus, the cyanogenesis cline
cyanogenesis polymorphism. observed across our sampled latitudinal transect (Fig. 2)
can be most easily attributed to local adaptation along
the latitudinal gradient.
Rapid adaptive cline evolution
Historically, the presence of similar phenotypic clines,
White clover was introduced into the United States with either in different geographical portions of a species
European colonization within the last 500 years. Thus, range or in several different species over the same range,
the continent-wide cyanogenesis cline documented in has been taken as evidence of adaptation (e.g. Endler
this study has evolved in this perennial species in a rela- 1986; Huey et al. 2000). While several other climate-
tively few generations. Trifolium repens is an obligately associated cyanogenesis clines have been described in
outcrossing, self-incompatible species, and, as a common white clover populations around the world (Daday
human commensal, it has an exceedingly large effective 1954a,b, 1958; de Araujo 1976; Till-Bottraud et al. 1988;
population size in North America (Kjærgaard 2003). Ganders 1990; Majumdar et al. 2004), most of this docu-
These characteristics would all be expected to facilitate mentation occurred before the advent of molecular tech-
the rapid evolution of adaptive clinal variation in this niques that could be used for assessing population
species. A number of other studies have also documented structure with neutral markers. We suggest that under-
the establishment of adaptive clines in relatively short standing population structure is one of the necessary pre-
periods of time (Antonovics & Bradshaw 1970; Huey requisites for drawing any definitive conclusions about
et al. 2000; Umina et al. 2005; Dlugosch & Parker 2008; adaptive divergence in these populations.
Montague et al. 2008; Saccheri et al. 2008). Like white clo-
ver, these species generally have large effective popula-
No evidence for epistatic selection or other fine-grained
tion sizes, little population structure and outcrossing
processes within North American populations
mating systems (e.g. Drosophila melanogaster). All of these
characteristics are expected to lead to efficient responses If epistatic selection were acting to maintain cyanogene-
to selection within populations (Charlesworth 2009), as sis variation within North American white clover

 2012 Blackwell Publishing Ltd

2464 N . J . K O O Y E R S and K . M . O L S E N

populations, as has previously been proposed for other scale spatial differences in selection among the clover
populations (Ennos 1982; Kakes 1987), one would subpopulations, the selection is either too weak or the
expect a statistically significant excess of AcLi and acli gene flow too great to allow for local adaptation.
cyanotypes relative to Hardy–Weinberg expectations In order to compare our LD observations to previous
(i.e. positive LD between Ac and Li). We reject epistatic studies, we compiled data from the literature docu-
selection as a mechanism for maintaining polymor- menting cyanogenesis polymorphisms in white clover
phism within our sampled populations: we find no evi- populations from around the world, and we calculated
dence for significant LD in any population (Table 1). ^ for these populations. Across nine studies, which col-
D
Similarly, LD is not significantly different from zero in lectively examined cyanogenesis variation in 14 world
most subpopulations sampled, and those showing sig- regions, 12 of the surveyed regions contain one or more
nificant D^ values indicate negative rather than positive populations showing statistically significant LD
LD (Table S1, Supporting information); this pattern is (Table 4). Among these 12 regions, 11 show a bias
inconsistent with the epistatic selection hypothesis pro- towards positive LD, consistent with epistatic selection
posed for white clover. for AcLi and acli cyanotypes. However, fewer than one-
We also find no evidence of a role for other fine-scale third of polymorphic populations show statistically sig-
factors in the maintenance of the cyanogenesis polymor- nificant positive LD (29.4% of populations on average
phism. For example, if fine-scale spatial environmental across the 11 regions), and only two regions show a
heterogeneity were favouring adaptive divergence majority of populations with positive LD (Table 4). To
among subpopulations within populations, subpopula- test whether our LD estimates for these other studies
tions within a given population would be expected to might be biased downwards by the inclusion of popula-
show some degree of adaptive divergence at the cyano- tions with low polymorphism levels, we also calculated
genesis genes. However, no differentiation is detected ^ using only those populations where all Ac ⁄ ac and
D
on this spatial scale (Table S1, Supporting information). Li ⁄ li alleles were all present at a frequency of at least
Moreover, when subpopulations within populations are 0.15. This recalculation does not result in a statistically
pooled, we find no evidence of LD between the cyano- significant increase in the percentage of populations
genesis genes, which would be predicted with the pool- showing positive LD (t-test; P > 0.05). Comparisons of
ing of subpopulations that are adaptively diverged native vs. introduced populations also do not reveal
(Nordborg & Tavare 2002). Thus, if there are any fine- any obvious patterns with respect to LD; proportions of

Table 4 Summary of linkage disequilibrium detected between Ac ⁄ ac and Li ⁄ li within polymorphic populations in 10 studies. Posi-
tive LD corresponds to an excess of AcLi and acli cyanotypes over Hardy–Weinberg expectations

Populations Populations
Average N per with significant with significant
No. polymorphic polymorphic nonzero LD positive LD
Study Area* populations† population N, (%) N, (%)

Daday (1954a) Eurasia 33 133.5 7 (21.2) 7 (21.2)

Daday (1954b) Swiss Alps 5 115.6 0 (0) 0 (0)
de Araujo (1976) Wales 17 40.8 1 (5.9) 1 (5.9)
Daday (1958) Asia 8 143.5 2 (25.0) 2 (25.0)
Daday (1958) Africa* 7 84.9 0 (0) 0 (0)
Daday (1958) S. America* 5 119.8 1 (20.0) 1 (20.0)
Daday (1958) N. America* 36 85.5 4 (11.1) 3 (8.3)
Daday (1958) Australia* 27 103.7 8 (29.6) 8 (29.6)
Daday (1958) New Zealand* 3 72.3 2 (66.6) 2 (66.6)
Ennos (1982) England 10 94.2 2 (20.0) 2 (20.0)
Frasier (1986) Canada* 7 58.7 1 (14.3) 0 (0)
Kakes (1987) Netherlands 16 49.8 7 (43.8) 7 (43.8)
Till-Bottraud et al. (1988) French Alps 13 104.8 3 (23.1) 3 (23.1)
Majumdar et al. (2004) Himalayas 5 1258.2 5 (100) 3 (60.0)
Present study N. America* 7 163.6 0 (0) 0 (0)

*Indicates non-native (introduced) populations.

†
Sampled populations that were polymorphic at both cyanogenesis genes.

 2012 Blackwell Publishing Ltd

 C L I N E E V O L U T I O N I N I N T R O D U C E D P O P U L A T I O N S 2465

populations showing positive LD range from 0% to functions would thus be expected to affect how the cya-
‡60% for both categories of populations. Taken nogenesis polymorphisms are maintained. Consistent
together, these findings suggest that while epistatic with this explanation, our review of published white
selection may be a factor in the evolution of the cyano- clover cyanogenesis data indicates that very few popu-
genesis polymorphism in some populations, it does not lations are entirely fixed for the acyanogenic phenotype,
appear to be a universal phenomenon in white clover. even in extremely cold locations such as Northern Can-
For the specific case of North American populations, ada and Alaska (e.g. Fraser 1986). This observation sug-
the absence of any detectable LD in the present study gests that there may be undiscovered functions of the
differs somewhat from the observations of Daday cyanogenesis components that could be affecting cyano-
(1958), who surveyed 36 polymorphic North American type frequencies.
populations, three of which (8.3%) show evidence of
significantly positive LD (Table 4). Whereas Daday’s
Conclusions
North American sampling focused largely on popula-
tions occurring west of the Rocky Mountains, the pres- Understanding how polymorphism is maintained in
ent study sampled populations exclusively within the heterogeneous environments has been widely studied
Mississippi basin. This difference in sampling could both theoretically and empirically (Levene 1953; Hed-
potentially account for some of the differences in LD rick 2006). A major conclusion of these studies is that
observed between the two studies. Another possibility generalizations are rare because the complexity of nat-
is that the earlier North American study may have ural systems, including animal–plant interactions, will
unintentionally pooled some populations that were often produce multidimensional selection pressures
adaptively differentiated at the cyanogenesis genes, that vary across geographic mosaics (Linhart 1991;
thereby creating positive LD as an artefact of admixture Thompson 2005). This study identifies rapid evolution
(Nordborg & Tavare 2002). We judge this latter expla- of an adaptive cyanogenesis cline across the central
nation as unlikely because the population sampling United States and finds that the mechanism of main-
scheme employed by Daday in his classic studies is taining polymorphism may differ from that in the
very similar to our own (e.g. Daday 1954a, 1958). native region. Specifically, epistatic selection, previ-
More generally, the apparent presence of positive LD ously proposed as a force that maintains this chemical
in some, but not all, polymorphic white clover popula- defence polymorphism, does not appear to be operat-
tions (Table 4) suggests that the importance of epistatic ing in these North American white clover popula-
selection in the cyanogenesis polymorphism may vary tions. Our study raises interesting new questions
from one population to another. We propose two non- about the maintenance of adaptive polymorphism in
mutually exclusive explanations that could account for this classic ecological genetic system. Ultimately,
this pattern. First, differences in herbivore communities we suggest that alternative mechanisms of maintain-
and abundances may affect which components of the ing the Ac ⁄ ac and Li ⁄ li polymorphisms may be at
cyanogenic response are under selection in a given loca- play in different populations within this species, and
tion. For example, Ennos (1982) notes that some mol- that understanding the ecological differences between
lusks possess b-glucosidase enzymes in their guts that these populations (including both biotic and abiotic
could function in the same way as linamarase in hydro- factors) would provide much-needed insight into this
lysing cyanogenic glucosides and inducing a deterrent system.
effect. Thus, clover populations in a community rich in
mollusks may be equally well protected whether they
Acknowledgements
are Acli or AcLi cyanotypes. Under such conditions, no
epistatic selection would be expected. We appreciate the assistance of AgResearch New Zealand and
A second possibility is that the cyanogenic compo- specifically Brent Barrett and Andrew Griffith for providing
information regarding molecular work. We thank Yangling
nents may have functions other than in herbivore deter-
Wei and other members of the Brummer Laboratory (Noble
rence, and that there may be variation among
Foundation) for providing data on microsatellite markers and
populations in selective pressure on these functions. for technical advice in genotyping. We thank Mike Dyer and
Both linamarase and cyanogenic glucosides have been other staff of the Washington University greenhouse for assis-
proposed to serve as nitrogen storage and transport tance in growing clover samples. This study was greatly
compounds (Gleadow & Woodrow 2002; Piotrowski improved by Briana Gross, Jared Strasburg, Allison Miller and
2008; Møller 2010), functions whose importance could the Olsen laboratory group, who took time to review an earlier
version of this manuscript. Funding was provided by an NSF
vary tremendously depending on factors such as the
CAREER award to KMO (DEB-0845497), and by an NSF Doc-
length of the growing season or other climatic variables.
toral Dissertation Improvement Grant to NJK (DEB-1110588).
Regional differences in selective pressure for these other

 2012 Blackwell Publishing Ltd

2466 N . J . K O O Y E R S and K . M . O L S E N

Dirzo R, Harper JL (1982b) Experimental studies on slug-plant

References interactions IV. The performance of cyanogenic and
Antonovics J, Bradshaw AD (1970) Evolution in closely acyanogenic morphs of Trifolium repens in the field. Journal
adjacent plant populations. 8. Clinal patterns at a mine of Ecology, 70, 119–138.
boundary. Heredity, 25, 349–362. Dirzo R, Sarukhan J (1984) Perspectives on Plant Population
de Araujo AM (1976) The relationship between altitude and Ecology. Sinauer & Associates, Sunderland, Massachusetts.
cyanogenesis in white clover (Trifolium repens, L.). Heredity, Dlugosch KM, Parker IM (2008) Invading populations of an
37, 291–293. ornamental shrub show rapid life history evolution despite
Atwood SS, Sullivan JT (1943) Inheritance of a cyanogenic genetic bottlenecks. Ecology Letters, 11, 701–709.
glucoside and its hydrolyzing enzyme in Trifolium repens. Endler JA (1977) Geographic Variation, Speciation, and Clines.
Journal of Heredity, 34, 311–320. Princeton University Press, Princeton, New Jersey.
Bonnet P, Van de Peer Y (2002) zt: a software tool for simple Endler JA (1986) Natural Selection in the Wild. Princeton
and partial Mantel tests. Journal of Statistical Software, 7, 1–12. University Press, Princeton, New Jersey.
Brighton F, Horne MT (1977) Influence of temperature on Ennos RA (1981) Detection of selection in populations of white
cyanogenic polymorphisms. Nature, 265(5593), 437–438. clover (Trifolium repens L). Biological Journal of the Linnean
Brookfield J (1996) A simple new method for estimating null Society, 15, 75–82.
allele frequency from heterozygote deficiency. Molecular Ennos RA (1982) Association of the cyanogenic loci in white
Ecology, 5, 453–455. clover. Genetical Research, 40, 65–72.
Brownstein MJ, Carpten JD, Smith JR (1996) Modulation of Excoffier L, Laval G, Schneider S (2005) Arlequin (version 3.0):
non-templated nucleotide addition by tag DNA polymerase: an integrated software package for population genetics data
primer modifications that facilitate genotyping. analysis. Evolutionary Bioinformatics, 1, 47–50.
BioTechniques, 20, 1004–1006. Faircloth BC (2006) GMCONVERT: file conversion for
Caicedo AL, Stinchcombe JR, Olsen KM, Schmitt J, GENEMAPPER output files. Molecular Ecology Notes, 6, 968–
Purugganan MD (2004) Epistatic interaction between 970.
Arabidopsis FRI and FLC flowering time genes generates a Feigl F, Anger V (1966) Replacement of benzidine by copper
latitudinal cline in a life history trait. Proceedings of the ethylacetoacetate and tetra base as a spot-test reagent for
National Academy of Sciences, USA, 101, 15670–15675. hydrogen cyanide and cyanogen. Analyst, 91, 282–284.
Cavalli-Sforza LL, Menozzi P, Piazza A (1993) Demic Fisher RA (1937) The wave of advance of advantageous genes.
expansions and human-evolution. Science, 259, 639–646. Annals of Human Genetics, 7, 355–369.
Chapuis M, Estoup A (2007) Microsatellite null alleles and Fraser J (1986) Cyanogenesis in naturalised white clover
estimation of population differentiation. Molecular biology and (Trifolium repens L.) populations in Atlantic Canada. Progress
evolution, 24, 621–631. Report, Clovers and Special Purpose Legumes Research, 19, 12–14.
Charlesworth D (2006) Balancing selection and its effects on Ganders FR (1990) Altitudinal clines for cyanogenesis in
sequences in nearby genome regions. PLoS Genetics, 2, 379– introduced populations of white clover near Vancouver,
384. Canada. Heredity, 64, 387–390.
Charlesworth B (2009) Effective population size and patterns of Glaubitz JC (2004) CONVERT: a user-friendly program to
molecular evolution and variation. Nature Reviews Genetics, reformat diploid genotypic data for commonly used
10, 195–205. population genetic software packages. Molecular Ecology
Clegg SM, Degnan SM, Kikkawa J, Moritz C, Estoup A, Owens Notes, 4, 309–310.
IPF (2002) Genetic consequences of sequential founder Gleadow RM, Woodrow IE (2002) Constraints on effectiveness
events by an island-colonizing bird. Proceedings of the of cyanogenic glycosides in herbivore defense. Journal of
National Academy of Sciences, USA, 99, 8127–8132. Chemical Ecology, 28, 1301–1313.
Corkill L (1942) Cyanogenesis in white clover (Trifolium repens Gliddon C, Saleem M (1985) Gene-flow in Trifolium
L.). V. The inheritance of cyanogenesis. New Zealand Journal repens—and expanding genetic neighbourhood. In: Genetic
of Science and Technology, 23, 178B–193B. Differentiation and Dispersal in Plants (ed P. Jacquard,
Daday H (1954a) Gene frequencies in wild populations of G. Heim and J. Antonovics), pp. 293–309. Springer-Verlag,
Trifolium repens L. I. Distribution by latitude. Heredity, 8, 61– Berlin, Germany.
78. Gross JB, Borowsky R, Tabin CJ (2009) A novel role for Mc1r
Daday H (1954b) Gene frequencies in wild populations of in the parallel evolution of depigmentation in independent
Trifolium repens II. Distribution by altitude. Heredity, 8, 377– populations of the cavefish Astyanax mexicanus. PLoS
384. Genetics, 5, e1000326.
Daday H (1958) Gene frequencies in wild populations of Haldane JBS (1948) The theory of a cline. Journal of Genetics, 48,
Trifolium repens L. III. World distribution. Heredity, 12, 169– 277–284.
184. Hedrick PW (2006) Genetic polymorphism in heterogeneous
Daday H (1965) Gene frequencies in wild population of environments: the age of genomics. Annual Review of Ecology
Trifolium repens IV. Mechanism of natural selection. Heredity, Evolution and Systematics, 37, 67–93.
20, 355–365. Hiesey WH, Clausen J, Keck DD (1942) Ecological aspects of
Dirzo R, Harper JL (1982a) Experimental studies on slug-plant evolution: relations between climate and intraspecific
interactions III. Differences in the acceptability of individual variation in plants. The American Naturalist, 76, 5–22.
plants of Trifolium repens to slugs and snails. Journal of Hijmans RJ, Cameron SE, Parra JL, Jones PG, Jarvis A (2005)
Ecology, 70, 101–117. Very high resolution interpolated climate surfaces for

 2012 Blackwell Publishing Ltd

 C L I N E E V O L U T I O N I N I N T R O D U C E D P O P U L A T I O N S 2467

global land areas. International Journal of Climatology, 25, Nordborg M, Tavare S (2002) Linkage disequilibrium: what
1965–1978. history has to tell us. Trends in Genetics, 18, 83–90.
Hill WG (1974) Estimation of linkage disequilibrium in Oakeshott J, Gibson J, Anderson P, Knibb W, Anderson D,
randomly mating populations. Heredity, 33, 229–239. Chambers G (1982) Alcohol-dehydrogenase and glycerol-3-
Hoekstra HE, Drumm KE, Nachman MW (2004) Ecological phosphate dehydrogenase clines in Drosophila melanogaster
genetics of adaptive color polymorphism in pocket mice: on different continents. Evolution, 36, 86–96.
geographic variation in selected and neutral genes. Evolution, Olsen KM, Ungerer MC (2008) Freezing tolerance and
58, 1329–1341. cyanogenesis in white clover (Trifolium repens L. Fabaceae).
Hudson ME, Kane NC (2009) Plant genomes do a balancing International Journal of Plant Sciences, 169, 1141–1147.
act. Molecular Ecology, 18, 2743–2745. Olsen KM, Sutherland BL, Small LL (2007) Molecular evolution
Huey RB, Gilchrist GW, Carlson ML, Berrigan D, Serra L of the Li ⁄ li chemical defence polymorphism in white clover
(2000) Rapid evolution of a geographic cline in size in an (Trifolium repens L.). Molecular Ecology, 16, 4180–4193.
introduced fly. Science, 287, 308–309. Olsen KM, Hsu S-C, Small LL (2008) Evidence on the
Hughes MA (1991) The cyanogenic polymorphism in Trifolium molecular basis of the Ac ⁄ ac adaptive cyanogenesis
repens L (white clover). Heredity, 66, 105–115. polymorphism in white clover (Trifolium repens L.). Genetics,
Kakes P (1987) On the polymorphism for cyanogenesis in 179, 517–526.
natural populations of Trifolium repens L. in the Netherlands Peakall R, Smouse PE (2006) GENALEX 6: genetic analysis in
I. Distribution of the genes Ac and Li. Acta Botanica Excel. Population genetic software for teaching and research.
Neerlandica, 36, 59–69. Molecular Ecology Notes, 6, 288–295.
Kakes P (1989) An analysis of the costs and benefits of the Pederson GA, Brink GE (1998) Cyanogenesis effect on insect
cyanogenic system in Trifolium repens L. Theoretical and damage to seedling white clover in a bermudagrass sod.
Applied Genetics, 77, 111–118. Agronomy Journal, 90, 208–210.
Keller SR, Taylor DR (2008) History, chance and adaptation Phifer-Rixey M, Heckman M, Trussell GC, Schmidt PS (2008)
during biological invasion: separating stochastic phenotypic Maintenance of clinal variation for shell colour phenotype in
evolution from response to selection. Ecology Letters, 11, 852– the flat periwinkle Littorina obtusata. Journal of Evolutionary
866. Biology, 21, 966–978.
Keller SR, Sowell DR, Neiman M, Wolfe LM, Taylor DR (2009) Piotrowski M (2008) Primary or secondary? Versatile nitrilases
Adaptation and colonization history affect the evolution of in plant metabolism. Phytochemistry, 69, 2655–2667.
clines in two introduced species. New Phytologist, 183, 678–690. Reynolds J, Weir BS, Cockerham CC (1983) Estimation of the
Kjærgaard T (2003) A plant that changed the world: the rise co-ancestry coefficient: basis for a short-term genetic-
and fall of clover 1000-2000. Landscape Research, 28, 41–49. distance. Genetics, 105, 767–779.
Kroymann J, Mitchell-Olds T (2005) Epistasis and balanced Rosenblum EB, Roempler H, Schoeneberg T, Hoekstra HE
polymorphism influencing complex trait variation. Nature, (2010) Molecular and functional basis of phenotypic
435, 95–98. convergence in white lizards at White Sands. Proceedings of
Lenormand T, Bourguet D, Guillemaud T, Raymond M (1999) the National Academy of Sciences, USA, 107, 2113–2117.
Tracking the evolution of insecticide resistance in the Saccheri IJ, Rousset F, Watts PC, Brakefield PM, Cook LM
mosquito Culex pipiens. Nature, 400, 861–864. (2008) Selection and gene flow on a diminishing cline of
Levene H (1953) Genetic equilibrium when more than one melanic peppered moths. Proceedings of the National Academy
ecological niche is available. The American Naturalist, 87, 331– of Sciences, USA, 105, 16212–16217.
333. Saucy F, Studer J, Aerni V, Schneiter B (1999) Preference for
Linhart YB (1991) Disease, parasitism and herbivory: acyanogenic white clover (Trifolium repens) in the vole
multidimensional challenges in plant evolution. Trends in Arvicola terrestris: I. Experiments with two varieties. Journal of
Ecology & Evolution, 6, 392–396. Chemical Ecology, 25, 1441–1454.
Majumdar S, De KK, Banerjee S (2004) Influence of two Schuelke M (2000) An economic method for the fluorescent
selective factors on cyanogenesis polymorphism of Trifolium labeling of PCR fragments. Nature Biotechnology, 18, 233–
repens L. in Darjeeling Himalaya. Journal of Plant Biology, 47, 234.
124–128. Silvertown J, Charlesworth D (2001) Introduction to Plant
McKechnie SW, Blacket MJ, Song SV et al. (2010) A clinally Population Biology, 4th edn. Blackwell Science, Oxford, UK.
varying promoter polymorphism associated with adaptive Slatkin M (1995) A measure of population subdivision based
variation in wing size in Drosophila. Molecular Ecology, 19, on microsatellite allele frequencies. Genetics, 139, 457–462.
775–784. Steiner CC, Roempler H, Boettger LM, Schoeneberg T,
Møller BL (2010) Functional diversifications of cyanogenic Hoekstra HE (2009) The genetic basis of phenotypic
glucosides. Current Opinion in Plant Biology, 13, 338–347. convergence in beach mice: similar pigment patterns but
Montague JL, Barrett SCH, Eckert CG (2008) Re- different genes. Molecular Biology and Evolution, 26, 35–45.
establishment of clinal variation in flowering time among Storz JF, Sabatino SJ, Hoffmann FG et al. (2007) The molecular
introduced populations of purple loosestrife (Lythrum basis of high-altitude adaptation in deer mice. PLoS Genetics,
salicaria, Lythraceae). Journal of Evolutionary Biology, 21, 3, 448–459.
234–245. Thompson JN (2005) The Geographic Mosaic of Coevolution.
Mullen LM, Hoekstra HE (2008) Natural selection along an University of Chicago Press, Chicago, Illinois.
environmental gradient: a classic cline in mouse Till-Bottraud I, Kakes P, Dommee B (1988) Variable
pigmentation. Evolution, 62, 1555–1569. phenotypes and stable distribution of the cyanotypes of

 2012 Blackwell Publishing Ltd

 2468 N . J . K O O Y E R S and K . M . O L S E N

Trifolium repens L. in Southern France. Acta Oecologica, 9, Data accessibility

393–404.
Todesco M, Balasubramanian S, Hu TT et al. (2010) Natural Microsatellite Data—DRYAD doi:10.5061/dryad.7q3m05fm.
allelic variation underlying a major fitness trade-off in
Arabidopsis thaliana. Nature, 465, 632–636.
Umina PA, Weeks AR, Kearney MR, McKechnie SW, Supporting information
Hoffmann AA (2005) A rapid shift in a classic clinal pattern
in Drosophila reflecting climate change. Science, 308, 691–693. Additional supporting information may be found in the online
Van Oosterhout C, Hutchinson WF, Wills DPM, Shipley P version of this article.
(2004) MICRO-CHECKER: software for identifying and Table S1 Subpopulation information and cyanogenesis poly-
correcting genotyping errors in microsatellite data. Molecular morphism data.
Ecology Notes, 4, 535–538.
Vasemagi A (2006) The adaptive hypothesis of clinal variation Table S2 Details of microsatellite markers and observed poly-
revisited: single-locus clines as a result of spatially restricted morphism data.
gene flow. Genetics, 173, 2411–2414.
Table S3 Summary of microsatellite polymorphism data per
Viette M, Tettamanti C, Saucy F (2000) Preference for
subpopulation.
acyanogenic white clover (Trifolium repens) in the vole
Arvicola terrestris: II. Generalization and further Table S4 Pairwise population measures of genetic differentia-
investigations. Journal of Chemical Ecology, 26, 101–122. tion among subpopulations using all microsatellite data.
Weinig C, Dorn LA, Kane NC et al. (2003) Heterogeneous
Fig. S1 Regression of population frequencies of the Li allele
selection at specific loci in natural environments in
(A), Ac allele (B) or the AcLi cyanotype (C) against geographi-
Arabidopsis thaliana. Genetics, 165, 321–329.
cal distances among populations, using the southernmost pop-
Zhen Y, Ungerer MC (2008) Clinal variation in freezing
ulation as the reference point (0 km). Statistical significance of
tolerance among natural accessions of Arabidopsis thaliana.
linear regressions is indicated as follows: *0.01 < P < 0.05,
New Phytologist, 177, 419–427.
**0.01 < P < 0.001 and ***P < 0.001.

Fig. S2 Regressions of pairwise measures of population differ-

The authors’ research interests center on plant evolutionary entiation for (A) Ac ⁄ ac and (B) Li ⁄ li against population differ-
biology. N.K. is a PhD student in the EEPB graduate program entiation for nine microsatellite loci. Points above the diagonal
at Washington University in St. Louis. His research focuses on lines indicate greater differentiation between sampling sites for
understanding the genetic basis of adaptation and ecological the cyanogenesis genes than for neutral markers. Statistical sig-
trade-offs that maintain polymorphism across space and time. nificance of one to one plots is indicated as follows:
He is particularly interested in adaptations involving plant sec- *0.01 < P < 0.05, **0.01 < P < 0.001 and ***P < 0.001.
ondary compounds and plant-animal interactions. K.O. is an
associate professor in the Biology department of Washington Please note: Wiley-Blackwell are not responsible for the content
University. His laboratory’s research includes studies on or functionality of any supporting information supplied by the
genetic mechanisms of adaptation and the evolution of crops authors. Any queries (other than missing material) should be
and their relatives. directed to the corresponding author for the article.

 2012 Blackwell Publishing Ltd

View publication stats