Inter-annual and spatial climatic variability have led to a balance between local fluctuating selection and wide-range directional selection in a perennial grass species

Abstract

Background and aims:
The persistence of a plant population under a specific local climatic regime requires phenotypic adaptation with underlying peculiar combinations of alleles at adaptive loci. The level of allele diversity at adaptive loci within a natural plant population conditions its potential to evolve, notably towards adaptation to a change in climate. Investigating the environmental factors that contribute to maintain adaptive diversity in populations is thus worthwhile. Within-population allele diversity at adaptive loci can be partly driven by the mean climate at population site but also by its temporal variability.

Methods:
The effects of climate temporal mean and variability on within-population allele diversity at putatively adaptive QTLs were evaluated using 385 natural populations of Lolium perenne L. (perennial ryegrass) collected right across Europe. For seven adaptive traits related to reproductive phenology and vegetative potential growth seasonality, the average within-population allele diversity at major QTLs (HeA) was computed.

Key results:
Significant relationships were found between HeA of these traits and temporal mean and variability of local climate. Those relationships were consistent with functional ecology theory.

Conclusions:
Results indicated that temporal variability of local climate has likely led to fluctuating directional selection, which has contributed to maintain allele diversity at adaptive loci and thus potential for further adaptation.

Full text

Annals of Botany 128: 357–369, 2021
doi: 10.1093/aob/mcab057, available online at www.academic.oup.com/aob
© The Author(s) 2021. Published by Oxford University Press on behalf of the Annals of Botany Company.
All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.
Inter-annual and spatial climatic variability have led to a balance between
local uctuating selection and wide-range directional selection in a perennial
grassspecies
T.Keep1, S.Rouet1, J.L.Blanco-Pastor1, P.Barre1, T.Ruttink2, K.J.Dehmer3, M.Hegarty4, T.Ledauphin1,
I.Litrico1, H.Muylle2, I.Roldán-Ruiz2, F.Surault1, R.Veron1, E.Willner3 and J.P.Sampoux1,*
1INRAE, Centre Nouvelle-Aquitaine-Poitiers, UR4 (UR P3F), F-86600 Lusignan, France, 2Flanders Research Institute for
Agriculture, Fisheries and Food (ILVO) - Plant Sciences Unit, Caritasstraat 39, 9090 Melle, Belgium, 3Leibniz Institute of Plant
Genetics and Crop Plant Research (IPK), Inselstr. 9, 23999 Malchow/Poel, Germany and 4IBERS-Aberystwyth University, Plas
Goggerdan, Aberystwyth, UK
*For correspondence. E-mail jean-paul.sampoux@inrae.fr
Received: 9 April 2021 Returned for revision: 9 February 2021 Editorial decision: 27 April 2021 Accepted: 29 April 2021
Electronically published: 5 May 2021
• Background and Aims The persistence of a plant population under a specic local climatic regime requires
phenotypic adaptation with underlying particular combinations of alleles at adaptive loci. The level of allele diver-
sity at adaptive loci within a natural plant population conditions its potential to evolve, notably towards adaptation
to a change in climate. Investigating the environmental factors that contribute to the maintenance of adaptive di-
versity in populations is thus worthwhile. Within-population allele diversity at adaptive loci can be partly driven
by the mean climate at the population site but also by its temporal variability.
• Methods The effects of climate temporal mean and variability on within-population allele diversity at putatively
adaptive quantitative trait loci (QTLs) were evaluated using 385 natural populations of Lolium perenne (perennial
ryegrass) collected right across Europe. For seven adaptive traits related to reproductive phenology and vegetative
potential growth seasonality, the average within-population allele diversity at major QTLs (HeA) was computed.
• Key Results Signicant relationships were found between HeA of these traits and the temporal mean and vari-
ability of the local climate. These relationships were consistent with functional ecology theory.
• Conclusions Results indicated that temporal variability of local climate has likely led to uctuating directional
selection, which has contributed to the maintenance of allele diversity at adaptive loci and thus potential for fur-
ther adaptation.
Key words: Allele diversity, climatic adaptation, adaptive diversity, uctuating selection, genome-wide
genotyping, grassland, Lolium perenne, natural genetic diversity, perennial ryegrass, intra-specic variability.
INTRODUCTION
The natural diversity of a plant species present over a wide
geographical range with diverse environments includes popu-
lations that are submitted to various environmental selection
pressures. Local adaption to specic environmental condi-
tions can lead to divergent genetic evolution of populations
(Sober and Wilson, 2011). Phenotypic polymorphism appears
when the different habitats are large and stable enough to in-
duce habitat specialization and tness optimization; otherwise
a monomorphic compromise phenotype endures (Rosenzweig,
1987). When phenotypic polymorphism is observed, it is likely
that tness has been improved in one habitat by relinquishing
some ability in another. Indeed, the intra-specic diversity of
many widespread plant species includes local adaptations to a
range of environments rather than just an adaptable all-purpose
phenotype (Van Tienderen, 1990; Balfourier and Charmet,
1991a; Weyl and Coetzee, 2016). Local adaptations have been
shown to emerge in reaction to small-scale and large-scale en-
vironmental variations (Clausen et al., 1940; Burdon, 1980;
Linhart and Grant, 1996; Macel etal., 2007). At large spatial
scales, climate is an important source of environmental vari-
ation and largely determines the composition of plant commu-
nities (Woodward, 1987). Relations between mean climatic
conditions and population genetic or phenotypic characteris-
tics have been widely studied for plant species (Balfourier and
Charmet, 1991a; Casler, 1995; Hancock etal., 2011; Bessega
etal., 2015). It is considered that a few traits are exposed to
stronger selection than most others and play a greater role
in the adaptation to various environments (Rieseberg et al.,
2004; Carnicer etal., 2012). Functional traits related to sea-
sonal vegetative growth potential and reproductive phenology
are regarded as the most important components of species t-
ness (Chuine, 2010; da Silveira Pontes etal., 2015). The inter-
population variability of such traits strongly contributes to local
adaptation across broad environmental gradients and is possibly
mainly determined by allele variability at a few major-effect
genes (Castède et al., 2014; Hill and Li, 2016). Indeed, loci
with a large effect on adaptive traits are likely those subject to
Downloaded from https://academic.oup.com/aob/article/128/3/357/6265263 by guest on 08 May 2023

Keep etal. — Local uctuating and wide-range selection in a perennial grass
358
the strongest selection pressures as their allele frequencies are
expected to vary the fastest and to pull population phenotypic
differentiation (Rieseberg etal., 2004; Storz, 2005).
Local climatic adaptation is notably facilitated by tempor-
ally stable constraining local climatic conditions (Kawecki
and Ebert, 2004). Accordingly, the probability of xation of
adaptive alleles, and hence the allele diversity at adaptive loci,
notably depends on the selection intensity imposed by the
local climatic constraints and on the stability of the selection
pressure through time. If the selection intensity varies along a
climatic gradient, the range of possible phenotypic values that
allow sufcient adaptation is expected to decrease as selection
intensity increases (Dostál etal., 2016; Barghi etal., 2020) and
the allele diversity at adaptive loci is also expected to follow
the same trend. Furthermore, at a given position along a mean
climate gradient, climatic conditions can uctuate over time
and lead to uctuating selection pressures (Bell, 2010). In that
case, the xation of alleles adapted to mean climatic conditions
is impeded. Allele frequencies at large-effect adaptive loci are
likely to be the most affected by uctuating selection pres-
sures (Bell, 2010). The greater the magnitude and frequency
of the local climatic uctuation, the greater the local allele di-
versity at these loci is likely to be. Consequently, inter-annual
climatic variability could maintain local genetic variance for
adaptive traits. Inter-annual uctuating selection has already
been observed in various plant species (Scopece etal., 2017;
Zahn, 2018; Busoms etal., 2018). For instance, an investiga-
tion into a natural shrub population performed in a Mexican
forest found that directional selection gradients were opposite
in sign between two consecutive years for owering initi-
ation date due to inter-annual differences in rainfall patterns
(Domínguez and Dirzo, 1995). Within a plant species popula-
tion, high levels of genetic diversity maintained by weak se-
lection intensity and/or local climate variability could improve
the population’s chances of adaptation to climate change and
reduce the risk of local disappearance of the species (Herben
etal., 2003; Schierenbeck, 2017).
Our study aimed to assess the relative impact of mean cli-
matic conditions and of uctuating inter-annual climatic con-
ditions on the adaptive diversity within natural populations of
Lolium perenne (perennial ryegrass) across Europe. This C3
diploid perennial grass is a highly outcrossing species whose
natural range covers the entire European continent, the Near
East and northern Africa. It has been adopted as a model spe-
cies for the genetics of temperate forage grasses, and accord-
ingly advances in genotyping technologies have been made
in this species, including the release of a whole genome ref-
erence sequence (Byrne etal., 2015). Our study was based on
a sample of 385 perennial ryegrass populations that origin-
ated from a large biogeographical range across Europe and
the Near East and was representative of the area of primary
expansion of the species (Blanco-Pastor et al., 2019). These
populations were grown in three common gardens in which
various traits related to agronomic value and environmental
adaptation were recorded at population level. Genotyping
based on next-generation sequencing technologies was ap-
plied to population DNA pools and provided allele frequen-
cies per population for a high number of nuclear genome SNP
markers. The climate at sites of origin of populations was
documented with norms and inter-annual standard deviations
of various climatic variables. Based on this experimental de-
sign, Keep etal. (2020) demonstrated that genome-wide as-
sociation study (GWAS) models can be efciently applied
at population level to discover major quantitative trait loci
(QTLs) in the natural diversity of perennial ryegrass. Blanco-
Pastor etal. (2020) highlighted loci and phenotypes involved
in adaptation to mean climate at sites of origin of populations.
Keep etal. (2021) pointed out the major functional trade-offs
and more specically the crucial role of vegetative growth sea-
sonality and reproductive phenology in the adaptation of per-
ennial ryegrass to mean climate. This new study investigated
the relationships between within-population allele diversity at
loci strongly associated with adaptive traits and both mean cli-
mate and inter-annual climatic variability at sites of origin of
populations. It focused on traits that were already evidenced
as of primary importance in the climate adaptation of peren-
nial ryegrass, i.e. traits related to vegetative growth season-
ality and reproductive phenology. The conceptual framework
of functional ecology was used to analyse the results in terms
of the relative impacts of mean and uctuating climatic condi-
tions on the adaptive diversity within populations.
MATERIALS ANDMETHODS
Plant material
This study used 385 gene bank accessions sampling the diver-
sity of 385 natural populations of perennial ryegrass. These
accessions were provided by gene banks of European coun-
tries and the USDA and were chosen as to represent the intra-
specic diversity of the species across Europe and the Near
East. Information about collection sites and sampling dates is
reported in Supplementary Data Table S1. Collections were
undertaken by scientists and plant breeders from a number of
agronomic research institutes in Europe and from the USDA
between 1960 and 2013. The sampling protocol in a collec-
tion site was usually the same as the one described in Charmet
etal. (1990) for populations collected in France. At the period
of seed maturity, a balanced amount of seeds was collected
from at least 50 plants across a homogeneous area of 100–
1000m2. Seeds were afterwards bulked (a single seed lot per
collection site) and stored in cold rooms of the gene bank of
the research institute that undertook collection. Perennial rye-
grass is a self-incompatible cross-pollinated species and it is
therefore expected that seeds sampled at a given collection site
are from a panmictic population. Seed lots were regenerated
and increased once soon after collection and then once or twice
again (at intervals of round 15 years) for the oldest accessions.
Regeneration was performed in eld facilities of research insti-
tutes following a standard protocol for grassland grasses agreed
in the frame of the ECPGR network promoting collaboration
between gene banks of European agronomic research institutes
(http://www.ecpgr.cgiar.org). Each seed regeneration step typ-
ically uses 50–100 (up to 200)plants (according to gene bank)
that are grown in isolation to avoid pollination from external
sources and are expected to intercross in panmixia. Seeds are
harvested on each plant and a balanced seed bulk is made for
each population. For the needs of the genotyping implemented
in this study, 300 seeds were drawn from the most recent bulk
Downloaded from https://academic.oup.com/aob/article/128/3/357/6265263 by guest on 08 May 2023

Keep etal. — Local uctuating and wide-range selection in a perennialgrass 359
of seeds available for each of the selected natural populations
in order to grow 300 seedlings for a pooled DNA extraction.
Another seed sample was drawn for each population to sow it
as replicated micro-swards to be used for in-eld phenotyping.
Genotyping
Genotyping was performed at the population level by
estimating the allele frequencies of nuclear genome SNP
markers through sequencing DNA pools extracted from bulked
leaf material of 300 individuals per population, according to
Byrne etal. (2013). Two methods were used to sequence a re-
duced, yet consistent, fraction of the genome: (1) genotyping
by sequencing (GBS) as described by Elshire et al. (2011)
in order to provide genome-wide coverage, and (2) highly
multiplexed amplicon sequencing (Hi-Plex), developed by
Floodlight Genomics (Knoxville, TN, USA), to target candi-
date genes putatively involved in the determinism of various
traits (Veeckman etal., 2019). We performed GBS with the PstI
restriction enzyme using population DNA pools following a
protocol similar to Byrne etal. (2013). The sequencing and bio-
informatics protocols are described in detail in Blanco-Pastor
etal. (2019), who used the same genetic material and the same
GBS assay. For the Hi-Plex sequencing, primers were designed
with Primer3 (Untergasser etal., 2012) in candidate genes using
previous knowledge of their sequence polymorphisms (Barre
etal., 2014; Veeckman etal., 2019). Aset of 185 amplicons
of 80–140bp was used (Keep etal., 2020). Population alterna-
tive allele frequencies of GBS and Hi-Plex bi-allelic SNPs were
estimated using the Bayesian approach of the SNAPE-pooled
script from Raineri et al. (2012). Only SNP markers with minor
allele frequency >5% in at least ten populations were retained.
Eventually, variant calling delivered population alternative
allele frequencies for 189 781 SNP markers distributed over
10 335 scaffolds of the perennial ryegrass reference genome
sequence (Byrne etal., 2015), including 524 SNP markers in
42 candidate gene loci from the Hi-Plex sequencing. The raw
genetic data (sequenced tags) are available in the NCBI Short
Read Archive (SRA) database through accession SRP136600.
The alternative allele frequencies per SNP locus and population
are available in Keep etal. (2020).
Experimental design and protocol for in-eld phenotyping
The 385 natural populations of perennial ryegrass were sown
in experimental gardens at three locations: Poel Island (PO) in
Germany (53.990N, 11.468E) on 8 April 2015, Melle (ME) in
Belgium (50.976N, 3.780E) on 2 October 2015 and Lusignan
(LU) in France (46.402N, 0.082E) on 9 April 2015. In each loca-
tion, each population was sown in three 1-m2 micro-swards with
2, 4 or 6g of seed according to whether the previously checked
germination rate of the seed lot was >80%, between 80 and
60%, or <60%, respectively. The seed density commonly used
to sow dense perennial ryegrass meadows for forage usage is
2gm−2 seeds of good germination quality (>80%). Population
micro-swards were arranged in three complete blocks in each
location. The trials were monitored until the end of 2017 in PO
and ME and until the end of 2019 in LU. Micro-swards were
cut (all aerial biomass >7cm above ground surface) regularly
to simulate the common defoliation regime of meadows used
for green forage production. Cutting dates (dd/mm/yy) were
16/06/15, 06/08/15, 04/09/15, 12/10/15, 04/03/16, 01/06/16,
13/07/16, 31/08/16, 26/10/16, 10/03/17, 07/06/17, 19/07/17,
01/09/17 and 13/10/17 at PO, 13/05/16, 08/07/16, 29/08/16,
13/10/16, 18/04/17, 31/05/17, 13/07/17, 24/08/17 and 04/10/17
at ME and 30/06/15, 03/09/15, 30/10/15, 04/02/16, 08/06/16,
26/07/16, 01/02/17, 13/06/17, 07/09/17, 07/06/18 and 27/08/18
at LU. Anti-dicotyledon herbicide was applied once in 2015 in
each location and once again in each subsequent year if ne-
cessary. In each location, nitrogen fertilization was applied
with 60kgNha−1 2months after sowing and after each aerial
biomass cut, and with 80kgNha−1 after winters 2015–16 and
2016–17 (LU, ME, PO) and winters 2017–18 and 2018–19
(LU) at the start of spring growth. The weather conditions ex-
perienced at each trial location for each season of each year are
detailed in Keep etal. (2020).
Phenotypictraits
Traits related to seasonal vegetative growth potential and re-
productive phenology have been pointed out as important for
plant species tness (Chuine, 2010; da Silveira Pontes et al.,
2015) and climate adaptation (for perennial ryegrass; Blanco-
Pastor etal., 2020; Keep etal., 2021). Such traits were recorded
overall at the level of 1-m2 micro-swards (i.e. without scoring
or measuring individual plants within micro-swards) and are
described below.
Vegetative growth in non-stressful conditions
The rst three traits described here were based on measure-
ments of micro-sward canopy heights. Canopy height meas-
urement is indeed commonly used as a proxy of grass sward
standing biomass, with a correlation between canopy height
and biomass nearing 0.8 (Viljanen etal., 2018; Borra-Serrano
etal., 2019).
Spring canopy height. The 2016 spring at ME was considered
favourable for growth as soil water content remained well above
wilting point and no frost day occurred during this period.
Spring canopy height (SPH) was the estimated canopy height
at 500 degree-days (base temperature of 0°C) measured at ME
in 2016 after the beginning of spring growth, which was the
date when daily minimum temperature and incident shortwave
global radiation did not fall again below 0°C and 60Wm−2, re-
spectively (Laboisse etal., 2018). At the chosen thermal time,
vegetative growth was largely in progress for all populations
but spike emergence had yet to occur. This canopy height was
considered to indicate the vegetative spring growth potential
of populations. When climatic conditions and soil nutrient re-
sources are relatively non-limiting for growth, competitive cap-
acity for light and space in the canopy, provided by early and
strong spring vegetative growth, is expected to be a major adap-
tive feature (Aerts, 1999; Craine and Dybzinski, 2013; Keep
etal., 2021).
Downloaded from https://academic.oup.com/aob/article/128/3/357/6265263 by guest on 08 May 2023

Keep etal. — Local uctuating and wide-range selection in a perennial grass
360
Summer canopy height. The 2016 summer at ME was fairly
wet, with an average soil water content >50% of eld capacity,
and relatively cool, with a daily mean temperature of 18.6°C
and an average daily maximum temperature of 23.7°C. Summer
canopy height (SMH) was the measure of the canopy height of
micro-swards just before the cut undertaken in late summer of
2016 at ME after a 2-month growth period. This canopy height
was considered to be typical of growth during a summer period
with little water and thermal stresses and to indicate the growth
potential of populations in summer. Low summer growth po-
tential is a known adaptation to high summer temperatures,
notably for perennial grasses (Balfourier and Charmet, 1991b;
Volaire etal., 2014; Bristiel et al., 2018, Blanco-Pastor etal.,
2020; Keep etal., 2021).
Yearly cumulated canopy height. The year 2016 at ME was
deemed relevant for the evaluation of biomass production poten-
tial in favourable conditions as no signicant water or thermal
stress occurred from spring to late autumn. The canopy height
measurements preceding each cut were summed over the year
2016 and this sum was used to indicate the annual aerial bio-
mass production potential of populations. High annual growth
potential indicates high resource acquisition capacity, which
is considered to be an adaptation to productive environments
(Grace, 1993, Keep etal., 2021).
Winter growth score. The 2018–2019 winter at LU was the
least stressful winter period encountered in the three trial sites
over the duration of the experiments, with a daily mean tem-
perature of 6.6°C and only 12 frost days. Winter growth score
(WGS) was visually scored on a scale of 1 (poorest growth) to
9 (strongest growth) at the end of this winter period at LU and
was considered to be a relevant indicator of winter growth po-
tential during a mild winter. High winter growth potential has
been recognized as adaptive in mild winter climate conditions
with relatively frequent summer drought for perennial ryegrass
and cocksfoot (Cooper, 1964, Blanco-Pastor etal., 2020; Keep
etal., 2021).
Reproductive phenology
Aftermath heading. After the cut of the rst spring wave of
elongated fertile stems, the intensity of subsequently recurring
fertile stem elongation was visually scored from 1 (no fertile
stem) to 9 (100 % plants with fertile stems). This score was
recorded in 2016 at LU. High aftermath heading (AHD) is typ-
ically found in Mediterranean populations of perennial ryegrass
and reects high investment in sexual reproduction (Balfourier
and Charmet, 1991b; Barre et al., 2018). High investment in
sexual reproduction agrees with a dehydration escape strategy
relevant to drought adaptation, which may, however, have
a negative counterpart in terms of plant persistency (Volaire,
2018; Blanco-Pastor etal., 2020; Keep etal., 2021).
Heading date. After a vernalization period, heading date
(HDT) is the date in spring when at least 20 spikes are arising
at the top of tiller sheaths in a micro-sward. This date was
converted into growing degree-days with a base temperature
of 0 °C counting from the start of vegetative spring growth.
Heading date was the adjusted mean of population heading
dates in 2016 and 2017 at LU and PO. In perennial ryegrass,
early spike emergence is partly physiologically and genetically
correlated to early and strong spring vegetative growth, which
enables competitive capacity for light and space in the canopy
(Thiele et al., 2009; Barre et al., 2018; Blanco-Pastor et al.,
2020; Keep etal., 2021).
Heading in rst year. This is a visual score indicating the
density of fertile (spike-bearing) stems elongated during the
year of sowing (i.e. without vernalization) on a scale of 1 (no
fertile stem) to 9 (100% plants with fertile stems). Heading in
rst year (HFY) was the adjusted mean of population scores
recorded in 2015 at LU and PO. High rst-year heading results
from low vernalization requirements and has been found in per-
ennial ryegrass populations from areas with mild winter and
high annual rainfall (Charmet etal., 1990; Keep etal., 2021).
More details about scoring or measurement of the preceding
traits can be found in Supplementary Data Methods S1. The
mean values of populations for the different traits are displayed
in Supplementary Data Table S2.
Climatic variables
We used a set of climatic variables inspired by the ETCCDI
variables (http://etccdi.pacicclimate.org/list_27_indices.shtml).
We set ne-resolution grids (0.05° longitude and latitude)
over Europe and the surroundings of norms and inter-annual
standard deviations of these variables for the 1989–2010 period
using EURO4M-MESAN and EUMETSAT CM SAF data
(Supplementary Data Methods S2). Values of norms and standard
deviations at sites of origin of the studied populations were set as
the values of grid cells containing these sites and are reported in
Supplementary Data Table S3.
Data analyses
Preliminary analyses of variance. An ANOVA model was ap-
plied to each phenotypic trait to assess the signicance of the
population effect and to compute adjusted means of populations
over replicates within trial location and in some cases over trial
locations (for more details see Supplementary Data Methods
S1). ANOVAs were performed using the functions ‘lm’ and
‘Anova’ of the R (R Core Team, 2018) ‘car’ library. Pearson
correlations between adjusted means of traits were calculated
using the R ‘cor’ function.
Association between phenotypic traits and genomic markers.
The ‘GWAS’ function of the R ‘rrBLUP’ package (Endelman,
2011) was used to implement GWAS in order to assess the
effect of each SNP locus on each phenotypic trait. This func-
tion uses the following mixed linearmodel:
y=μ+Xβ+Zu+e
In this model, y is the n length vector of values of a phenotypic
trait for n populations (adjusted means from ANOVA models),
µ is the intercept vector, X is the n length vector of the alterna-
tive allele frequencies of a given SNP for the n populations, β
is the SNP xed effect, Z is an incidence matrix, u is a vector
of random effects with var(u)=σ2
a×A [σ2
a being the additive
genetic variance estimated by restricted maximum likelihood
Downloaded from https://academic.oup.com/aob/article/128/3/357/6265263 by guest on 08 May 2023

Keep etal. — Local uctuating and wide-range selection in a perennialgrass 361
(REML) and A the genomic relationship matrix] and e is the
vector of residuals with σ2
e, also estimated by REML. The gen-
etic relationship matrix A was computed using the formula
given by Feng et al. (2020). The genetic structure was impli-
citly taken into account in the mixed linear models since kin-
ship (genetic relatedness between populations) encompasses
ne and broad genetic structure. P-values were adjusted to
q-values using a procedure to control the false discovery rate
(Benjamini and Hochberg, 1995).
Estimates of within-population genetic diversity
The allele diversity of a genotyped SNP locus within a given
population was estimated as HeSNP=2 × p × q, where p and
q are the alternative and reference allele frequencies, respect-
ively, within the population. Perennial ryegrass being an obli-
gate outbreeding species, HeSNP could be considered to be an
estimate of the within-population frequency of heterozygotes at
the genotyped locus (heterozygosity).
To estimate the average diversity of SNP loci signicantly
associated with a phenotypic trait in a given population, we
computed a within-population trait-associated diversity cri-
terion (HeA) as the average of HeSNP values over a set of non-
redundant SNP loci found to be associated with this trait by
GWAS. To dene this set, we selected SNPs whose q-value was
≤0.05 and for which allele frequency information was available
for at least 75 % of populations. Then, to avoid redundancy,
pairs of SNP markers whose alternative allele frequency cor-
relation was >0.4 were identied and only the SNP found to be
most signicantly associated with the trait by GWAS was kept.
HeA was not computed for populations for which the alterna-
tive allele frequency information was missing for at least one of
the SNPs of the dened set.
Establishing representative climatic indicators
A principal component analysis (PCA) was performed on norms
of climatic variables and another one on their inter-annual standard
deviations. PCAs were performed using the ‘PCA’ function from
the R ‘FactoMineR’ package with the scale variables option set to
true. The rst four principal components of the PCAs on norms
and on inter-annual standard deviations were used as climate in-
dicators reporting the main spatial variations across Europe of the
mean climate and of its inter-annual variability, respectively. The
rst four principal components of the PCA on climate norms are
hereafter referred to as meanPC1 to meanPC4 (mean climate indi-
cators) and those of the PCA on inter-annual standard deviations as
stdPC1 to stdPC4 (climate variability indicators).
Modelling within-population trait-associated genetic diversity
using climatic indicators
The following procedure was implemented to identify
a limited number of mean climate indicators (meanPC1 to
meanPC4) and climate variability indicators (stdPC1 to stdPC4)
that best predicted the within-population trait-associated
diversity criterion (HeA) of each phenotypic trait. We imple-
mented the multivariate linear model of the ‘cv.glmnet’ function
from the ‘glmnet’ R package, in which HeA was the dependant
variable and the climate indicators were the potential explana-
tory variables. Quadratic effects of climate indicators were also
included as potential explanatory variables in order to report
non-linear relationships. The elastic net mixing parameter was
set to 1 (lasso penalty) and 10-fold cross validation was applied
with mean squared error (MSE) minimization as the criterion
for model evaluation. The explanatory variables retained were
those from the model with the highest possible value of λ (pen-
alty multiplier) and such that the MSE was within one standard
MSE from the MSE of the best tested model. This procedure
was repeated 1000 times and the explanatory variables nally
kept were those retained in the selected model for at least 500 of
the iterations. This implementation of the ‘cv.glmnet’ function
aimed to test a large amount of cross-validation sets in order
to identify stable explanatory variables. Then, for each pheno-
typic trait, a linear regression predicting the HeA criterion from
the selected climate indicators was implemented using the ‘lm’
R function. Astepwise option (‘step’ R function with option
direction set to ‘both’) was included to remove redundancy in
climate indicators according to the Bayesian information cri-
terion (BIC).
RESULTS
Representative climate indicators
The percentages of inertia taken up by the rst four principal
components of the PCA of climatic norms (mean climate in-
dicators) were 35, 25, 16 and 4, respectively. Components
meanPC1, meanPC2, meanPC3 and meanPC4 were most cor-
related to norms of climatic variables related to maximum daily
temperature in summer, cumulated precipitation throughout
the year, daily temperature range throughout the year and
daily minimum temperature in spring, respectively (Table 1,
Supplementary Data Table S4).
The percentages of inertia taken up by the rst four principal
components of the PCA of inter-annual standard deviations of
climatic variables (climate variability indicators) were 25, 18,
8 and 5, respectively. Components stdPC1, stdPC2, stdPC3 and
stdPC4 were most correlated to the standard deviation across
years of climatic variables related to cumulated precipitation
throughout autumn and winter periods, daily maximum tem-
perature and evapotranspiration in summer, number of frost
days in spring and number of frost days in winter, respectively
(Table 1, Supplementary Data Table S4).
GWAS and estimation of within-population trait-associated
diversity
Alternative allele frequency information was missing for
7% of genotyped SNP loci per population on average with a
maximum of 37%. Using all genotyped SNPs, the average cor-
relation between allele frequencies of pairs of SNPs whose dis-
tance was ≤10000, 1000, 100 and 10 bp equalled 0.19, 0.22,
0.26 and 0.32, respectively. When using only pairs of SNPs
Downloaded from https://academic.oup.com/aob/article/128/3/357/6265263 by guest on 08 May 2023

Keep etal. — Local uctuating and wide-range selection in a perennial grass
362
from a scaffold in which signicant SNPs were detected by
GWAS, it equalled 0.3, 0.35, 0.42 and 0.48, respectively.
The number of SNP loci detected by GWAS that were used
to estimate the within-population trait-associated diversity cri-
terion (HeA) varied from 2 for yearly cumulated canopy height
to 17 for aftermath heading. The number of populations for
which HeA could be computed varied from 125 for summer
canopy height to 316 for yearly cumulated canopy height
(Table 2).
Association of trait means and trait-associated diversity criteria
with climate indicators
The values of correlations between population trait means and
mean climate indicators are displayed in Table 3 and the values
of correlations of within-population trait-associated diversity cri-
teria (HeA) with mean climate indicators and climate variability
indicators are displayed in Table 4. Regarding the correlations
between population trait means and mean climate indicators,
aftermath heading, heading in rst year, summer canopy height
and winter growth were most correlated to meanPC1 and heading
date and spring canopy height to meanPC2, and yearly cumu-
lated canopy height was most correlated to meanPC3. Regarding
the correlations between HeA criteria and mean climate indica-
tors, the HeA criteria of aftermath heading, heading in rst year
and summer canopy height were most correlated to meanPC1
and those of heading date, spring canopy height and yearly cu-
mulated canopy height to meanPC2, and that of winter growth
score was most correlated to meanPC3. Regarding the correl-
ations between HeA criteria and climate variability indicators,
the HeA criteria of heading in rst year, spring canopy height and
yearly cumulated canopy height were most correlated to stdPC1
and those of summer canopy height and aftermath heading to
stdPC2, that of winter growth score was most correlated to both
stdPC1 and stdPC2, and that of heading date to stdPC3. Figure 1
displays the spatial distribution of the HeA criterion of heading
in rst year across Europe and shows that this criterion has its
lowest values in north-eastern Europe, where inter-annual pre-
cipitation variability is relativelylow.
Neutral evolutionary forces (migration, drift, mutation) may
generate a spatial distribution of neutral diversity that results
in a large number of signicant correlations between allele
frequency (or diversity) at neutral SNP loci and climatic indi-
cators. Taking into account that most genotyped SNP loci are
expected to be neutral, a relatively high correlation between a
population trait mean (alternatively an HeA criterion) and a cli-
matic indicator was assumed to reveal natural selection only
if the absolute value of this correlation was higher than the
highest absolute values of correlations found between the allele
frequency (alternatively diversity) at most genotyped SNP loci
and this climatic indicator. All population trait means (except
that of spring canopy height) were better correlated (in abso-
lute value) to at least one climate mean indicator than the allele
frequency at 90% of genotyped SNPs (Table 3). The HeA cri-
terion of each trait was better correlated (in absolute value) to
at least one climate mean indicator and one climate variability
indicator than the allele diversity (HeSNP) at 90% of genotyped
SNPs (Table 4).
T . Computation of the within-population trait-associated
diversity criterion (HeA) for the studied traits. ‘Number of SNPs’
is the number of uncorrelated SNP loci found to be associated with
a trait by GWAS and used to compute the HeA criterion of this
trait. ‘Number of populations’ is the number of populations for
which the HeA criterion of a trait could be computed (HeA could
not be computed for a population if the alternative allele frequency
information was missing for at least one SNP locus involved in the
computation)
Trait Abbreviation Number
of SNPs
Number of
populations
Aftermath heading AHD 17 148
Heading date HDT 16 150
Heading in rst year HFY 11 183
Spring canopy height SPH 7 217
Summer canopy height SMH 9 125
Winter growth score WGS 9 131
Yearly cumulated canopy
height
YCH 2 316
T . Mean climate indicators and variability climate indicators. (A) Mean climate indicators (meanPC1 to meanPC4) are the rst
four principal components of a PCA of norms of climatic variables at sites of origin of populations, and (B) climate variability indica-
tors (stdPC1 to stdPC4) are the rst four principal components of a PCA of inter-annual standard deviations of climatic variables. ‘%
inertia’ is the percentage of inertia explained by the principal component corresponding to a climate indicator. ‘Main climatic variable’
indicates the climatic variable for which the inter-annual norm (alternatively standard deviation) is most correlated to a given mean
climate indicator (alternatively climate variability indicator)
(A) Mean climate indicator
meanPC1 meanPC2 meanPC3 meanPC4
% inertia 35 25 16 4
Main climatic variable
(norm)
Summer maximum
daily temperature
Cumulated precipitation throughout
the year
Annual daily
temperature range
Spring daily minimum
temperature
(B) Climate variability indicator
stdPC1 stdPC2 stdPC3 stdPC4
% inertia 25 18 8 5
Main climatic variable
(standard deviation)
Cumulated precipitation
autumn–winter
Summer daily maximum temperature
and evapotranspiration
Number of frost
days in spring
Number of frost days
in winter
Downloaded from https://academic.oup.com/aob/article/128/3/357/6265263 by guest on 08 May 2023

Keep etal. — Local uctuating and wide-range selection in a perennialgrass 363
Prediction of within-population trait-associated diversity criteria
(HeA) from mean climate and climate variability indicators
Models predicting HeA criteria of studied traits from mean
climate and climate variability indicators are described in
Table 5. The coefcient of determination of models varied
from 0.09 for HeA of spring canopy height to 0.38 for HeA
of winter growth score. According to traits, the set of pre-
dictors included one or two climate variability indicators and
from zero to two mean climate indicators. Scatterplots be-
tween HeA criteria and the most signicant climate indica-
tors predicting them are displayed in Fig. 2. The correlation
between an HeA criterion and the most signicant climate
indicator predicting it varied from 0.31 (between HeA of
spring canopy height and stdPC1) to 0.54 (between HeA of
heading in rst year and stdPC1). The most signicant ex-
planatory variable of the HeA criterion of aftermath heading,
summer canopy height and yearly cumulated canopy height
was a mean climate indicator; nevertheless, at least one cli-
mate variability indicator was also found among the explana-
tory variables of each of these HeA criteria (Table 5). For
heading in the rst year, spring canopy height, heading date
and winter growth score, the most signicant explanatory
variable of the HeA criterion was a climate variability indi-
cator; nevertheless, at least one mean climate indicator was
also found among the explanatory variables of the HeA cri-
terion of heading date and heading in the rst year (Table 5).
The mean climate indicator meanPC1 was a predictor of the
HeA criterion of aftermath heading, heading in rst year and
summer canopy height, and meanPC2 was a predictor of the
T . (A) Correlations between population trait means and mean climate indicators at sites of origin of populations. Trait abbrevi-
ations are explained in Table 2.(B) 90, 95, 99 and 100% quantiles of the absolute values of the correlations between population alter-
native allele frequencies at the 189781 genotyped SNP loci and mean climate indicators. Since most genotyped SNP loci are expected
to be neutral, the quantile values are expected to report the distribution of correlations of alternative allele frequencies at neutral loci
with mean climate indicators
Mean climate indicator
Trait/quantile meanPC1 meanPC2 meanPC3 meanPC4
(A) AHD 0.47 −0.11 0.44 0.1
HDT −0.12 −0.34 −0.23 0.05
HFY 0.27 0.19 0.11 0.06
SPH 0.14 0.23 0.13 0.09
SMH −0.27 0.21 −0.27 −0.01
WGS 0.52 −0.06 0.44 0.06
YCH −0.12 0.24 −0.24 0.06
(B) 90% 0.24 0.26 0.23 0.18
95% 0.29 0.31 0.28 0.22
99% 0.37 0.4 0.36 0.28
100% 0.58 0.56 0.54 0.43
T . (A) Correlations of within-population trait-associated diversity criterion (HeA) with mean climate and climate variability in-
dicators at sites of origin of populations. Trait abbreviations are explained in Table 2.(B) 90, 95, 99 and 100% quantiles of the absolute
values of the correlations between allele diversity (HeSNP) at the 189781 genotyped SNP loci and the mean climate and climate vari-
ability indicators. Since most genotyped SNP loci are expected to be neutral, the quantile values are expected to report the distribution
of correlations of allele diversity at neutral loci with mean climate indicators
Climate indicator
Mean climate indicator Climate variability indicator
HeA/quantile meanPC1 meanPC2 meanPC3 meanPC4 stdPC1 stdPC2 stdPC3 stdPC4
(A) HeA_AHD 0.38 0.08 0.34 0 0.24 0.37 −0.13 0.04
HeA_HDT −0.19 0.2 −0.05 −0.06 0.06 −0.14 0.31 0.21
HeA_HFY 0.46 0.29 0.17 −0.13 0.54 0.18 −0.15 −0.1
HeA_SPH 0.11 0.27 0.07 0 0.31 0.01 0.12 0.12
HeA_SMH 0.39 −0.04 0.34 0.01 0.22 0.44 −0.26 −0.23
HeA_WGS 0.33 0.25 0.41 0.02 0.41 0.44 −0.01 −0.03
HeA_YCH 0.08 0.48 −0.06 −0.24 0.47 −0.14 0.17 0.11
(B) 90% 0.23 0.24 0.22 0.18 0.28 0.26 0.16 0.18
95% 0.27 0.29 0.27 0.21 0.34 0.31 0.2 0.21
99% 0.35 0.37 0.35 0.27 0.45 0.4 0.27 0.28
100% 0.56 0.57 0.54 0.43 0.68 0.63 0.44 0.41
For a genotyped SNP locus and a population, HeSNP=2× p× q, where p and q are the alternative and reference allele frequencies, respectively, within the
population.
Downloaded from https://academic.oup.com/aob/article/128/3/357/6265263 by guest on 08 May 2023

Keep etal. — Local uctuating and wide-range selection in a perennial grass
364
HeA criterion of heading date and yearly cumulated canopy
height. The climate variability indicator stdPC1 was a pre-
dictor (positive effect) of the HeA criterion of heading in rst
year, spring canopy height, yearly cumulated canopy height
and winter growth score; stdPC2 was a predictor (positive
effect) of the HeA criterion of aftermath heading, summer
canopy height and winter growth score; and stdPC3 was a
predictor (positive effect) of the HeA of headingdate.
T . Linear models that best predict the within-population trait-associated diversity criterion (HeA) of studied traits from mean and
climate variability indicators. The climate indicators included in each model are displayed as well as their estimated effect in the model
(Estimate) and the P-value of their t-test (Pr>|t|). The coefcient of determination (R2) of the model is also displayed. Trait abbreviations
are explained in Table 2
HeA criterion Regression model Model R2
HeA_AHD Intercept meanPC1 stdPC2 0.22
Estimate 1.2e−01 7.0e−04 7.8e−04
Pr>|t| 1.5e−112 9.4e−05 1.8e−04
HeA_HDT Intercept stdPC3 meanPC2 0.15
Estimate 1.5e−01 1.7e−03 −6.1e−05
Pr>|t| 2.1e−99 1.4e−04 1.4e−03
HeA_HFY Intercept stdPC1 meanPC1 0.34
Estimate 7.0e−02 1.5e−03 9.3e−04
Pr>|t| 8.9e−76 2.4e−08 1.4e−04
HeA_SPH Intercept stdPC1 0.09
Estimate 7.7e−02 1.2e−03
Pr>|t| 2.2e−79 1.0e−05
HeA_SMH Intercept meanPC1 stdPC2 meanPC120.28
Estimate 7.9e−02 5.8e−04 8.9e−04 2.4e−05
Pr>|t| 1.1e−61 1.5e−03 2.2e−03 2.4e−02
HeA_WGS Intercept stdPC2 stdPC1 0.38
Estimate 1.2e−01 1.6e−03 1.3e−03
Pr>|t| 9.8e−98 8.0e−10 5.3e−09
HeA_YCH Intercept meanPC2 stdPC1 0.25
Estimate 8.2e−02 1.9e−03 1.6e−03
Pr>|t| 1.0e−75 1.5e−03 5.3e−03
10°O
60°N
50°N
0.005–0.039
range
HeA_HFY
0.039–0.073
0.073–0.107
0.107–0.141
0.141–0.175
0.175–0.209
40°N
0 500 1000 km
0° 10°E 20°E 30°E 40°E
N
F. . Spatial distribution across Europe of the within-population trait-associated diversity criterion of heading in rst year (i.e. heading without vernalization
requirements) (HeA_HFY) computed for 183 natural populations of perennial ryegrass. The pairing of dot colours and HeA_HFY ranges is given in the inset.
Downloaded from https://academic.oup.com/aob/article/128/3/357/6265263 by guest on 08 May 2023

Keep etal. — Local uctuating and wide-range selection in a perennialgrass 365
−20 −10 0
meanPC1
10 20
0.08
0.10
HeA_AHDHeA_SMHHeA_YCH
HeA_SPH
HeA_WGS
0.12
0.14
0.16 ABC
D
G
EF
r = 0.38
Regression line
Partial R2 = 0.09
Model R2 = 0.22
r = 0.31
Regression line
Partial R2 = 0.09
Model R2 = 0.09
r = 0.54
Regression line
Partial R 2 = 0.13
Model R2 = 0.34
r = 0.39
Regression line
Partial R2 = 0.09
Model R2 = 0.28
r = 0.48
Regression line
Partial R2 = 0.02
Model R2 = 0.25
−15 −10 −5 0
stdPC3
5
0.08
0.10
0.12
0.14
0.16
0.18
0.20
HeA_HDT
10 15 −10 01
02
0
0
0.05
HeA_HFY
0.10
0.15
0.20
stdPC1
−30 −20 −10 0102030 −10 01020 −15 −10 −5 0510 15
meanPC1 stdPC1 stdPC2
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0
0.05
0.10
0.15
0.05
0.10
0.15
0.20
−20 −10 0102030
0
0.05
0.10
0.15
0.20
0.25
meanPC2
r = 0.31
Regression line
Partial R2 = 0.09
Model R2 = 0.09
r = 0.44
Regression line
Partial R2 = 0.22
Model R2 = 0.38
F. . Within-population trait-associated diversity criterion (HeA) of each studied trait (ordinate) plotted against the most signicant climate indicator (abscissa)
in the best multiple regression model predicting this criterion. (A) Aftermath heading (AHD), (B) heading date (HDT), (C) heading in rst year (HFY), (D)
summer canopy height (SMH), (E) spring canopy height (SPH), (F) winter growth score (WGS) and (G) yearly cumulated canopy height (YCH). meanPCi and
stdPCi are the ith principal components of the PCAs of norms of climatic variables and of inter-annual standard deviations of climatic variables, respectively. The
Pearson correlation coefcient (r) between the HeA criterion and the climate indicator is displayed as well as the percentage of variance explained by the climate
indicator (partial R2 estimated by subtracting the coefcient of determination of the complete model without the targeted climate indicator from the one of com-
plete model including the targeted climate indicator). The coefcient of determination of the complete model (with all retained climate indicators) is also displayed
(model R2). The red diamonds represent the regression line or curve (polynomial of degree 2)of the within-population trait-associated diversity criterion (HeA) on
the climate indicator when all other climate indicators in the multiple regression model are set to their mean values.
Downloaded from https://academic.oup.com/aob/article/128/3/357/6265263 by guest on 08 May 2023

Keep etal. — Local uctuating and wide-range selection in a perennial grass
366
DISCUSSION
Oligogenic adaptation
Application of GWAS to mean phenotypes and allele frequen-
cies of populations has already been successfully implemented
to discover QTLs, notably in perennial ryegrass (Byrne etal.,
2013; Fè etal., 2015; Ashraf etal., 2016; Keep etal., 2020).
This approach has been made possible by next-generation
sequencing (NGS) technologies, which enable the sequencing
of pools of DNA from a number of individuals per popula-
tion, thus obtaining allele frequencies at numerous SNP loci
(Byrne et al., 2013). Keep et al. (2020) showed that GWAS
accounting for kinship was able to detect major QTLs when
applied to mean phenotypes and allele frequencies of a set of
natural populations of perennial ryegrass. With the GWAS sig-
nicance level threshold we applied, no more than 17 uncor-
related SNP markers were found to be associated with a given
phenotypic trait. These signicant markers were assumed to be
in linkage disequilibrium with as many different QTLs. The
slower linkage disequilibrium decay in scaffolds containing
SNPs detected by GWAS compared with that observed on
average across all scaffolds suggests that genetic hitchhiking
and selective sweep may have been at play in the neighbour-
hood of adaptive QTLs (Fay and Wu, 2000; Storz, 2005). The
small number of SNPs found signicantly associated with any
trait could be expected under the assumption that these traits
are likely of some importance for climatic adaptation. This as-
sumption is supported by the correlations between population
trait means and mean climate indicators reported in Table 3
which are consistent with adaptive trends already known in per-
ennial ryegrass (see Material and methods, Phenotypic traits).
It has been pointed out that, under directional selection, only a
few large-effect adaptive QTLs can present patterns of allelic
differentiation that contrast with the genome-wide stochastic
background (Storz, 2005). Thus, GWAS based on natural di-
versity which correct for genome-wide genetic relatedness
can be expected to detect loci that are in linkage disequilib-
rium with large-effect adaptive QTLs. The theory of adaptation
based on oligogenic variation (Bell, 2010) states that adaptation
to a changing environment is primarily led by change in al-
lele frequency at a limited number of large-effect adaptive loci,
despite the strong likelihood that quantitative traits are affected
by many loci. It can thus be expected that benecial alleles of
large-effect QTLs are those that can experience the fastest in-
crease in frequency under environmental selection pressure and
thus are likely to spearhead species adaptation. The population
frequencies of such alleles are thus also the most likely to be af-
fected by uctuating directional selection (Bell, 2010). On the
other hand, small-effect alleles that are weakly selected are ex-
pected to increase in frequency more slowly and are vulnerable
to disappearance through drift (Rieseberg etal., 2004).
Inuence of local mean climatic conditions and inter-annual
climate variability on within-population allele diversity at
adaptiveloci
The most signicant predictor of within-population trait-
associated diversity (HeA) was a mean climate indicator for
three traits (aftermath heading, summer canopy height, yearly
cumulated canopy height), whereas it was a climate variability
indicator for the four remaining traits (heading date, heading in
rst year, spring canopy height, winter growth score) (Table 5).
The HeA criterion of a putatively adaptive trait can be primarily
correlated to a mean climate gradient if the range of phenotypic
values (and of allele diversity of underlying QTLs) that allows
for sufcient adaptation is wider at one end of the gradient than
at the other end. Meanwhile, climatic conditions at a given site
can uctuate over different time scales, i.e. inter-seasonal vari-
ations, relatively stochastic inter-annual variations and climate
change over longer time spans. The intensity and direction of
selection imposed by climatic constraints can thus uctuate
accordingly over time. Given that selection is expected to pri-
marily act on oligogenic diversity, which enables rapid adaptive
evolution, uctuating selection over time may lead to uctuating
allele frequencies at main adaptive QTLs (Bell, 2010). Thus,
the probability of xation of adaptive alleles may be inversely
related to the level of temporal variability of the climatic con-
straint. This can explain why a climate variability indicator
was the most signicant predictor of within-population trait-
associated diversity for some of the studied putatively adaptive
traits (heading date, heading in rst year, spring canopy height,
winter growth score) and was also one of its signicant pre-
dictors for the three other traits (aftermath heading, summer
canopy height, yearly cumulated canopy height).
Ecological concepts could be called upon to explain the re-
lationships between HeA trait-associated diversity criteria and
climate variability indicators (Tables 3 and 4). The negative cor-
relation of summer canopy height with meanPC1 (maximum
daily temperature in summer) and the positive correlation of its
within-population trait-associated diversity criterion with both
meanPC1 and stdPC2 (standard deviation across years of daily
maximum temperature in summer) suggest that uctuating se-
lection driven by ecological trade-offs has been at play. In areas
where summer maximum temperatures are high on average but
variable across years, individuals with a limited summer growth
potential are likely best adapted to summers during which heat
waves occur (protection of sensitive vegetative tissues), whereas
other individuals with greater summer growth potential are best
adapted to cooler summers due to their greater ability to inter-
cept light and occupy canopy space. However, the impact of
heat waves on the tness of the second kind of individuals may
be limited if seed production is sufciently advanced before
the stress period. Conversely, populations from areas where
summers are consistently cool enough for substantial growth
can include only light-competitive plants with great summer
growth potential. A similar result was reported in a study of
inter-specic diversity in a Californian grassland, in which the
smaller species gained a competitive advantage during heat
wave periods (Hallett etal., 2019). The positive correlation of
aftermath heading with meanPC1 and the remarkably positive
correlation of its HeA criterion with both meanPC1 and stdPC2
can similarly be explained by uctuating selection. During par-
ticularly hot summers when vegetative growth is near null, re-
current production of seeds favours tness with the counterpart
of reduced persistency. On the other hand, during cool sum-
mers, plants that produce relatively few reproductive tillers
can accumulate resources that contribute to the preservation of
Downloaded from https://academic.oup.com/aob/article/128/3/357/6265263 by guest on 08 May 2023

Keep etal. — Local uctuating and wide-range selection in a perennialgrass 367
vegetative tillers and thus give competitive advantage and better
tness. The high allele diversity at isozyme loci found in per-
ennial ryegrass populations from southern Europe (with likely
high meanPC1) by Balfourier etal. (1998) could also possibly
be explained by uctuating selection mechanisms.
The within-population trait-associated diversity of vegetative
growth in winter (WGS), spring (SPH) and throughout the year
[yearly cumulated canopy height (YCH)] in non-stressful con-
ditions was positively correlated to stdPC1 (standard deviation
across years of spring and autumn precipitation) (Tables 3 and 4).
When water availability is not limiting for canopy growth, fea-
tures that improve competitive ability for light capture, such
as early and strong growth (e.g. substantial winter growth and
high spring growth) may increase the ability to gain canopy
space (Craine and Dybzinski, 2013). On the other hand, when
water availability is limiting for canopy growth, features pro-
viding greater tolerance to water decit and greater competitive
ability to extract water can improve adaptation at the expense of
aerial vegetative growth (Grace, 1990, 1993).
Variability of ability to ower without vernalization require-
ments (HFY), of earliness of spike emergence (HDT), of AHD
and of vegetative growth seasonality and potential (WGS, YCH,
SPH, SMH) provides means to tune the timing of sexual repro-
duction, seed germination, juvenile development and vegetative
organ growth in order to optimize phenology and size differ-
ences among competitors. Such strategies can be assigned to
size-mediated priority effects, or early-arriver advantages, and
play an essential role in the outcome of competitive interactions
(Rasmussen etal., 2014; Rudolf and McCrory, 2018; Rudolf,
2019). Chase (2010) stated that local biodiversity should be
large in environments with high production potential, such as
areas with high annual rainfall, because of stochastic commu-
nity assembly and interactions between competitors. Our results
appear to corroborate this assumption since HeA was positively
correlated to meanPC2 (cumulated precipitation throughout
the year) for heading in rst year, spring canopy height, winter
growth score and yearly cumulated canopy height. Comparable
results were reported by an investigation of relationships be-
tween rice diversity and environmental factors in a Chinese
province (Cheng-yun, 2010). Moreover, uctuating climatic
conditions, such as variable inter-annual precipitation, can dif-
ferently affect the demographic rates of various competitors and
thus can lead to variable seasonal timing of their interactions
and thus to uctuation of competition pressures (Chesson,
1994; Hallett etal., 2019). This can result in uctuating selec-
tion for phenological traits and for timing of vegetative growth
and thus lead to local conservation of causal genetic diversity of
these features (Domínguez and Dirzo, 1995; Bell, 2010). Such
mechanisms could, for instance, contribute to the signicant
positive contribution of stdPC1 (spring and autumn precipita-
tion variability) in the prediction of HeA for heading in the rst
year, spring canopy height, yearly cumulated canopy height and
winter growthscore.
Perennial ryegrass was shown to originate from the
Mediterranean and Balkan areas (Blanco-Pastor etal., 2019).
It has likely experienced more extreme and uctuating climate
conditions in these areas than in others due to its longer time
of presence. Fluctuating directional selection has thus likely
occurred there in this species at long and short time scales
(Bell, 2010). Blanco-Pastor etal. (2019) also pointed out that
perennial ryegrass diversity from these areas experienced intro-
gressions from more northern regions in the late stages of the
species’ expansion. Both long-term uctuating selection and
introgression events may have contributed to the relatively high
within-population trait-associated diversity found in popula-
tions from southern Europe (where meanPC1 is highest) for
aftermath heading, heading in rst year and summer canopy
height. Other studies of perennial ryegrass natural populations
from various areas also found that within-population gen-
etic (isozyme) diversity (Balfourier et al., 1998) and regional
phenotypic variability (Casler, 1995) were substantially high
around the MediterraneanBasin.
Local natural populations of grassland species are threatened
by climate change and will likely need to evolve quickly so as
to remain adapted to their environment (Henkin et al., 2010;
Schierenbeck, 2017). Substantial adaptive genetic diversity
present in these populations could improve their chances of fu-
ture adaptation. The capacity of a grassland species population
to adapt to the changing climate may notably depend on the
level of past inter-annual stochastic local climatic variability.
The results presented here corroborate the hypothesis that such
variability induces uctuating directional selection which con-
tributes to the maintenance of a reserve of adaptive diversity
(Herben etal., 2003).
SUPPLEMENTARYDATA
Supplementary data are available online at https://academic.
oup.com/aob and consist of the following. Table S1: list of the
perennial ryegrass natural populations in the study, including
information on collection sites and gene banks maintaining
seed samples of these populations. Table S2: trait means of
perennial ryegrass natural populations. Table S3: values of
climatic variables at sites of origin of perennial ryegrass nat-
ural populations. Table S4: correlations between climatic
variables and the rst four principal components of PCAs per-
formed on the norms and on standard deviations of climatic
variables. Methods S1: description of traits characterizing
populations, computation of population trait means and trait
summary statistics. Methods S2: description of climatic vari-
ables characterizing sites of origin of natural populations of
perennial ryegrass.
FUNDING
This work was supported by grants awarded to the project
GrassLandscape (2014 FACCE-JPI ERA-NET+ call Climate
Smart Agriculture) from the European Community (grant agree-
ment number 618105), the Agence Nationale de la Recherche
(ANR) and the Institut National de la Recherche Agronomique
(metaprogramme ACCAF) in France, the Biotechnology and
Biological Sciences Research Council (BBSRC) in the UK and
the Bundesantalt für Landwirtschaft und Ernährung (BLE) in
Germany, and by a grant awarded to T.K.from the French ad-
ministrative region Nouvelle-Aquitaine.
Downloaded from https://academic.oup.com/aob/article/128/3/357/6265263 by guest on 08 May 2023

Keep etal. — Local uctuating and wide-range selection in a perennial grass
368
ACKNOWLEDGMENTS
The authors thank the curators of the gene banks that provided per-
ennial ryegrass seed samples for the needs of the project and staff
from European agronomic research institutes who contributed to
in situ collections in 2015. Perennial ryegrass is one of the plant
species covered under the Multilateral System of the International
Treaty on Plant Genetic Resources for Food and Agriculture. All
genetic materials used in this study were made available to the
authors after signature of a Standard Material Transfer Agreement
(SMTA) by the provider and the recipient. Implementation and
signature of an SMTA provides compliance with the provisions
of the Nagoya Protocol for parties wishing to provide and receive
genetic material under the Multilateral System. The authors thank
the technical staff of IBERS, ILVO, INRAE and IPK involved in
the project. Climate data were processed by Milka Radojevik and
Christian Page (CECI, Université de Toulouse, CNRS, CERFACS,
https://cerfacs.fr) from EURO4M-MESAN and EUMETSAT CM
SAF grids. The authors declare no conict of interest. T.K. with
support from J.P.S. conceived the reported investigation and
wrote the manuscript. S.R.and J.L.B.-P. contributed to the main
conceptual ideas and to the interpretation of the results. F.S.and
R.V.recorded trait data. J.P.S., T.R., P.B., K.J.D., M.H., I.L., H.M.,
I.R-R. and E.W.were involved in planning and supervising the
project. T.L., J.L.B-.P., T.R.and T.K.processed the experimental
data. All authors provided critical feedback and helped shape the
research and manuscript.
LITERATURECITED
AertsR. 1999. Interspecic competition in natural plant communities: mech-
anisms, trade-offs and plant-soil feedbacks. Journal of Experimental
Botany 50: 29–37.
AshrafBH, ByrneS, FéD, etal. 2016. Estimating genomic heritabilities at the
level of family-pool samples of perennial ryegrass using genotyping-by-
sequencing. Theoretical and Applied Genetics 129: 45–52.
BalfourierF, CharmetG. 1991a. Relationships between agronomic charac-
ters and ecogeographical factors in a collection of French perennial rye-
grass populations. Agronomie 11: 645–657.
BalfourierF, CharmetG. 1991b. Spaced plant evaluation of Mediterranean
germplasm collections of perennial ryegrass. Euphytica 57: 57–66.
BalfourierF, CharmetG, RavelC. 1998. Genetic differentiation within and
between natural populations of perennial and annual ryegrass (Lolium
perenne and L.rigidum). Heredity 81: 100–110.
BarghiN, HermissonJ, SchlöttererC. 2020. Polygenic adaptation: a unifying
framework to understand positive selection. Nature Reviews Genetics 21:
769–781.
BarreP, ChupeauG, BeguierV, etal. 2014. Développement d’un outil pour
la sélection assistée par marqueurs chez le ray-grass anglais. Innovations
Agronomiques 35: 151–159.
BarreP, RuttinkT, MuylleH, etal. 2018. Natural diversity in vegetative and
reproductive investments of perennial ryegrass is shaped by the climate at
the place of origin. Grass and Forage Science 73: 193–205.
BellG. 2010. Fluctuating selection: the perpetual renewal of adaptation in vari-
able environments. Philosophical Transactions of the Royal Society of
London. Series B, Biological Sciences 365: 87–97.
BenjaminiY, HochbergY. 1995. Controlling the false discovery rate: a prac-
tical and powerful approach to multiple testing. Journal of the Royal
Statistical Society: Series B (Methodological) 57: 289–300.
BessegaC, PomettiC, EwensM, SaidmanBO, VilardiJC. 2015. Evidences
of local adaptation in quantitative traits in Prosopis alba (Leguminosae).
Genetica 143: 31–44.
Blanco-Pastor JL, Manel S, Barre P, et al. 2019. Pleistocene climate
changes, and not agricultural spread, accounts for range expansion and
admixture in the dominant grassland species Lolium perenne L. Journal of
Biogeography 46: 1451–1465.
Blanco-PastorJL, BarreP, KeepT, etal. 2020. Canonical correlations reveal
adaptive loci and phenotypic responses to climate in perennial ryegrass.
Molecular Ecology Resources 21: 849–870.
Borra-SerranoI, DeSwaefT, MuylleH, etal. 2019. Canopy height measure-
ments and non-destructive biomass estimation of Lolium perenne swards
using UAV imagery. Grass and Forage Science 74: 356–369.
BristielP, GillespieL, ØstremL, BalachowskiJ, ViolleC, VolaireF. 2018.
Experimental evaluation of the robustness of the growth–stress toler-
ance trade-off within the perennial grass Dactylis glomerata. Functional
Ecology 32: 1944–1958.
BurdonJJ. 1980. Intra-specic diversity in a natural population of Trifolium
repens. Journal of Ecology 68: 717–735.
Busoms S, Paajanen P, Marburger S, et al. 2018. Fluctuating selection
on migrant adaptive sodium transporter alleles in coastal Arabidopsis
thaliana. Proceedings of the National Academy of Sciences of the USA
115: E12443–E12452.
Byrne S, Czaban A, Studer B, Panitz F, Bendixen C, Asp T. 2013.
Genome wide allele frequency ngerprints (GWAFFs) of populations via
genotyping by sequencing. PLoS ONE 8: e57438.
ByrneSL, NagyI, PfeiferM, etal. 2015. A synteny-based draft genome se-
quence of the forage grass Lolium perenne. Plant Journal 84: 816–826.
CarnicerJ, Brotons L, Stefanescu C, Peñuelas J. 2012. Biogeography of
species richness gradients: linking adaptive traits, demography and diver-
sication. Biological Reviews 87: 457–479.
CaslerMD. 1995. Patterns of variation in a collection of perennial ryegrass
accessions. Crop Science 35: 1169–1177.
Castède S, Campoy JA, García JQ, et al. 2014. Genetic determinism of
phenological traits highly affected by climate change in Prunus avium:
owering date dissected into chilling and heat requirements. New
Phytologist 202: 703–715.
CharmetG, BalfourierF, BionA. 1990. Agronomic evaluation of a col-
lection of French perennial ryegrass populations: multivariate classi-
cation using genotype × environment interactions. Agronomie 10:
807–823.
ChaseJM. 2010. Stochastic community assembly causes higher biodiversity in
more productive environments. Science 328: 1388–1391.
Cheng-yunL. 2010. Correlation between geographic patterns of cultivar di-
versity of rice (Oryza sativa L.) and environmental factors, local culture in
Yunnan. Journal of Yunnan Agricultural University 25: 451–457.
Chesson P. 1994. Multispecies competition in variable environments.
Theoretical Population Biology 45: 227–276.
ChuineI. 2010. Why does phenology drive species distribution? Philosophical
Transactions of the Royal Society of London. Series B, Biological Sciences
365: 3149–3160.
ClausenJ, KeckDD, HieseyWM. 1940. Experimental studies on the nature
of species. I.Effect of varied environments on western North American
plants. Washington, DC: Carnegie Institution of Washington Publication,
452.
CooperJP. 1964. Climatic variation in forage grasses. I.Leaf development
in climatic races of Lolium and Dactylis. Journal of Applied Ecology
1: 45.
CraineJM, Dybzinski R. 2013. Mechanisms of plant competition for nutri-
ents, water and light. Functional Ecology 27: 833–840.
DomínguezCA, DirzoR. 1995. Rainfall and owering synchrony in a trop-
ical shrub: variable selection on the owering time of Erythroxylum
havanense. Evolutionary Ecology 9: 204–216.
DostálP, FischerM, PratiD. 2016. Phenotypic plasticity is a negative, though
weak, predictor of the commonness of 105 grassland species. Global
Ecology and Biogeography 25: 464–474.
ElshireRJ, GlaubitzJC, SunQ, etal. 2011. A robust, simple genotyping-
by-sequencing (GBS) approach for high diversity species. PLoS ONE 6:
e19379.
EndelmanJB. 2011. Ridge regression and other kernels for genomic selection
with R package rrBLUP. Plant Genome 4: 250–255.
Fay JC, Wu CI. 2000. Hitchhiking under positive Darwinian selection.
Genetics 155: 1405–1413.
FèD, CericolaF, ByrneS, etal. 2015. Genomic dissection and prediction of
heading date in perennial ryegrass. BMC Genomics 16: 921.
FengJ, ChenC, CaiY, etal. 2020. Partitioning heritability analyses unveil the
genetic architecture of human brain multidimensional functional connect-
ivity patterns. Human Brain Mapping 41: 3305–3317.
GraceJB. 1993. The effects of habitat productivity on competition intensity.
Trends in Ecology & Evolution 8: 229–230.
Downloaded from https://academic.oup.com/aob/article/128/3/357/6265263 by guest on 08 May 2023

Keep etal. — Local uctuating and wide-range selection in a perennialgrass 369
Hallett LM, Shoemaker LG, White CT, Suding KN. 2019. Rainfall vari-
ability maintains grass-forb species coexistence. Ecology Letters 22:
1658–1667.
HancockAM, BrachiB, FaureN, etal. 2011. Adaptation to climate across
the Arabidopsis thaliana genome. Science 334: 83–86.
HenkinZ, PerevolotskyA, SternbergM. 2010. Vulnerability of Mediterranean
grasslands to climate change: what can we learn from a long-term experi-
ment? In: Porqueddu C, Ríos S, eds. The contributions of grasslands to the
conservation of Mediterranean biodiversity (Options Méditerranéennes:
Série A. Séminaires Méditerranéens; num. 92). Zaragoza: CIHEAM/
CIBIO/FAO/SEEP, 167–74..
Herben T, Krahulec F, Hadincová V, Pecháčková S, WildováR. 2003.
Year-to-year variation in plant competition in a mountain grassland.
Journal of Ecology 91: 103–113.
HillCB, LiC. 2016. Genetic architecture of owering phenology in cereals and
opportunities for crop improvement. Frontiers in Plant Science 7: 1906.
KaweckiTJ, Ebert D. 2004. Conceptual issues in local adaptation. Ecology
Letters 7: 1225–1241.
Keep T, Sampoux JP, Blanco-Pastor JL, et al. 2020. High-throughput
genome-wide genotyping to optimize the use of natural genetic resources
in the grassland species perennial ryegrass (Lolium perenne L.). G3 Genes,
Genomes, Genetics 10: 3347–3364.
KeepT, SampouxJ-P, BarreP, etal. 2021. To grow or survive: which are the
strategies of a perennial grass to face severe seasonal stress? Functional
Ecology 35: 1145–1158.
LaboisseS, CombesD, Escobar-GutierrezA, Hurlus J. 2018. Spatial dis-
tribution of simulated turfgrass photosynthesis in football stadium pitch.
In: 6th International Symposium on Plant Growth Modeling, Simulation,
Visualization and Applications (PMA), Hefei, China, 111–114, 4–8
November 2018. doi:10.1109/PMA.2018.8611615.
Linhart YB, Grant MC. 1996. Evolutionary signicance of local genetic
differentiation in plants. Annual Review of Ecology and Systematics 27:
237–277.
MacelM, Lawson CS, MortimerSR, etal. 2007. Climate vs. soil factors in
local adaptation of two common plant species. Ecology 88: 424–433.
R Core Team. 2018. R: a language and environment for statistical computing.
Vienna: R Foundation for Statistical Computing. https://www.R-project.
com.
Raineri E, Ferretti L, Esteve-Codina A, Nevado B, Heath S, Pérez-
Enciso M. 2012. SNP calling by sequencing pooled samples. BMC
Bioinformatics 13: 239.
RasmussenNL, VanAllenBG, RudolfVH. 2014. Linking phenological shifts
to species interactions through size-mediated priority effects. Journal of
Animal Ecology 83: 1206–1215.
RiesebergLH, ChurchSA, MorjanCL. 2004. Integration of populations and
differentiation of species. New Phytologist 161: 59–69.
RosenzweigML. 1987. Habitat selection as a source of biological diversity.
Evolutionary Ecology 1: 315–330.
RudolfVHW. 2019. The role of seasonal timing and phenological shifts for
species coexistence. Ecology Letters 22: 1324–1338.
Rudolf VHW, McCrory S. 2018. Resource limitation alters effects of
phenological shifts on inter-specic competition. Oecologia 188:
515–523.
SchierenbeckKA. 2017. Population-level genetic variation and climate change
in a biodiversity hotspot. Annals of Botany 119: 215–228.
Scopece G, Juillet N, Lexer C, Cozzolino S. 2017. Fluctuating selection
across years and phenotypic variation in food-deceptive orchids. PeerJ 5:
e3704.
daSilveira PontesL, MaireV, Schellberg J, LouaultF. 2015. Grass strat-
egies and grassland community responses to environmental drivers: a re-
view. Agronomy for Sustainable Development 35: 1297–1318.
SoberE, WilsonDS. 2011. Adaptation and natural selection revisited. Journal
of Evolutionary Biology 24: 462–468.
StorzJF. 2005. Using genome scans of DNA polymorphism to infer adaptive
population divergence. Molecular Ecology 14: 671–688.
ThieleJ, JørgensenRB, HauserTP. 2009. Flowering does not decrease vege-
tative competitiveness of Lolium perenne. Basic and Applied Ecology 10:
340–348.
VanTienderen PH. 1990. Morphological variation in Plantago lanceolata:
limits of plasticity. Evolutionary Trends in Plants 4: 35–43.
UntergasserA, CutcutacheI, KoressaarT, et al. 2012. Primer3– new cap-
abilities and interfaces. Nucleic Acids Research 40: e115.
VeeckmanE, VanGlabekeS, Haegeman A, etal. 2019. Overcoming chal-
lenges in variant calling: exploring sequence diversity in candidate genes
for plant development in perennial ryegrass (Lolium perenne). DNA
Research 26: 1–12.
ViljanenN, HonkavaaraE, NäsiR, HakalaT, NiemeläinenO, KaivosojaJ.
2018. A novel machine learning method for estimating biomass of grass
swards using a photogrammetric canopy height model, images and vege-
tation indices captured by a drone. Agriculture 8: 70.
Volaire F. 2018. A unified framework for plant drought adaptive strat-
egies: across scales and disciplines. Global Change Biology 24:
2929–2938.
VolaireF, BarkaouiK, NortonM. 2014. Designing resilient and sustainable
grasslands for a drier future: adaptive strategies, functional traits and bi-
otic interactions. European Journal of Agronomy 52: 81–89.
WeylPSR, CoetzeeJA. 2016. Morphological variations in southern African
populations of Myriophyllum spicatum: phenotypic plasticity or local
adaptation? South African Journal of Botany 103: 241–246.
WoodwardFI. 1987. Climate and plant distribution. Cambridge: Cambridge
University Press, 188.
ZahnLM. 2018. Fluctuating selection in nature. Science 361: 463.
Downloaded from https://academic.oup.com/aob/article/128/3/357/6265263 by guest on 08 May 2023

Downloaded from https://academic.oup.com/aob/article/128/3/357/6265263 by guest on 08 May 2023