Download Title: Evolutionary dynamics of the leaf phenological cycle in an oak

1
Title: Evolutionary dynamics of the leaf phenological cycle in an oak metapopulation along
2
an elevation gradient
3
4
Running title: Multivariate evolution of leaf phenology
5
6
Authors:
7
Cyril Firmat1,2*,
8
Sylvain Delzon1,2,
9
Jean-Marc Louvet1,2,
10
Julien Parmentier3,
11
Antoine Kremer1,2
12
13
Adresses:
14
1
INRA, UMR 1202 BIOGECO, F-33610 Cestas, France;
15
2
Univ. Bordeaux, UMR 1202 BIOGECO, F-33405 Talence, France;
16
3
INRA – UE 0393, Unité Expérimentale Arboricole, Centre de Recherche Bordeaux-Aquitaine,
17
Toulenne, France.
18
*Contact: [email protected] - Phone: (+33) (0) 5 57 12 28 31
19
20
Data archival location: Phenological data will be deposited on Dryad after acceptance of the
21
manuscript. Population genetic data are available from the Quercus portal website
22
(https://w3.pierroton.inra.fr/QuercusPortal/)
23
1
24
Abstract
25
Predictions suggest that climate change will radically alter the selective pressures on
26
phenological traits. Long-lived species, such as trees, will be particularly affected, as they will
27
have only one or a few generations to track a moving fitness optimum. The traits describing the
28
annual life cycle of trees generally have a high evolvabilities, but nothing is known about the
29
strength of their genetic correlations. Strong correlations can impose strong evolutionary
30
constraints, potentially hampering the adaptation of multivariate phenological phenotypes. The
31
evolutionary, genetic and environmental components of leaf unfolding and leaf senescence
32
timings are evaluated within an oak metapopulation along an elevation gradient. Then,
33
phenological population divergence is compared to a neutral expectation based on microsatellite
34
markers. This approach allows us (1) to evaluate the influence genetic correlation on the pattern
35
of multivariate local adaptation to elevation and (2) to estimate the trait(s) most likely exposed to
36
past selective pressures induced by the colder climate experienced at high elevation. The genetic
37
correlation was found positive but very weak, indicating that genetic constraints did not influence
38
the local adaptation pattern as expressed on leaf phenology. Both spring and fall phenology were
39
involved in local adaptation, although leaf unfolding was likely the most exposed trait to climate-
40
induced selection. Despite the strong selective pressures experienced, a high genetic variance was
41
preserved albeit reduced with elevation. We conclude that the evolutionary potential of leaf
42
phenology is probably not the most critical aspect for short term population survival under a
43
changing climate.
44
45
Keywords: common garden, extreme environments, genetic constraints, G-matrix, local
46
adaptation.
2
47
48
INTRODUCTION
The capacity of populations to cope with changing climatic conditions depends on their
49
ability to track a fitness optimum moving on the phenotypic space. This involves adaptive
50
plasticity (environmentally induced phenotypic change) and non-neutral genetic variation
51
(microevolution) (Nussey et al., 2007; Lande, 2009). The contribution of microevolution to short-
52
term responses to contemporary climate change remains a matter of debate (Gienapp et al., 2008;
53
Merilä, 2012; Franks et al., 2014; Teplitsky & Millien, 2014). By contrast, climate has long been
54
recognized to be a major, ubiquitous driver of evolutionary dynamics over longer time scales
55
(Hunt & Roy, 2006; Erwin, 2009). However, long-term adaptive responses to a climatic trend
56
require a substantial amount of genetic variation for traits targeted by selection to be maintained
57
over time, to sustain the response to selection depleting initial genetic variation. The existence of
58
abundant additive genetic variation for traits important for fitness is particularly crucial for the
59
survival of tree populations (Kremer et al., 2014). As sessile, perennial organisms, trees typically
60
have slow migration rates, hampering the optimal spatial tracking of their climatic niche (Aitken
61
et al., 2008; Lindner et al., 2010). Furthermore, given the long generation time of these
62
organisms, they will need to adapt to changes in climatic conditions over only one or few
63
generations. The capacity for evolutionary change is therefore mostly dependent on standing
64
genetic variation in the populations. Detailed quantitative genetics investigations of tree
65
populations are therefore required to predict their fate in the context of climate change.
66
Such investigations must include: (1) measurement of the loss of fitness due to climatic
67
change (i.e., the strength of directional and stabilizing selection on traits of adaptive significance)
68
(Lande & Arnold, 1983) and (2) evaluation of the distribution of genetic variation for traits
69
subject to climatic selection. It is difficult to assess loss of fitness in species mating over long
3
70
distances, such as trees. This task will therefore require the use of molecular marker information
71
within a spatially explicit context for estimation of the variance of fecundity (e.g., Klein et al.,
72
2011). By contrast, the distribution of genetic variation for those fitness-related traits is
73
commonly assessed in open-pollinated progeny tests using common garden experiments, in
74
which genetic variance can be estimated together with genetic differentiation between
75
populations (reviewed in Alberto et al., 2013a). In situations in which multivariate genetic
76
variability can be compared to the genetic divergence between populations spread along a
77
climatic gradient, indirect inferences become possible, concerning the ways in which climate-
78
related selective processes (first task above) have interacted with the genetic architecture of traits
79
(second task) to shape the observed evolutionary trajectories. Such inferences are based on the
80
theory of multivariate genetic constraints first formalized by Lande (1979) as 𝛥𝑧̅ = 𝐆𝛽. i.e., the
81
magnitude of mean phenotypic change 𝛥𝑧̅ of a set of traits zi exposed to a multivariate selection
82
gradient 𝛽 depends on G, the matrix of the additive genetic variance and covariance of the traits.
83
Genetic covariances between traits typically result from pleiotropy or linkage disequilibrium. The
84
ways in which these processes influence evolution and adaptation is thus accounted for by the G
85
matrix, and there is growing evidence for a major influence of G on the trajectories of phenotypic
86
evolution (Hansen & Houle, 2008; Bolstad et al., 2014; Teplitsky et al., 2014b).
87
When considering the response to selection of a focal trait zi, ignoring its genetic
88
covariation with other traits may lead to misleading conclusions. If zi is correlated with other
89
traits subject to stabilizing selection, part of its evolutionary potential is captured by stabilizing
90
selection in other dimensions of the phenotypic space (Hansen, 2003). In this case, the adaptive
91
potential of populations facing climate change can be overestimated (Etterson & Shaw, 2001;
92
Duputié et al., 2012). Conversely, if the directional selection vector induced by climate change
4
93
aligns with the main axes of genetic variation in the multivariate phenotypic space, selection
94
response is accelerated (e.g., Schluter, 1996; Hansen & Voje, 2011), underlining the limits of
95
univariate approaches focused on a key phenotypic trait.
96
Phenological traits — the dates or durations of events of the organism’s life cycle — are
97
typically expected to be exposed to selection induced by climate change (Rehfeldt et al., 1999;
98
Forrest & Miller-Rushing, 2010; Gienapp et al., 2014).The phenological cycle of the leaves of
99
deciduous temperate tree species can be described on the basis of two timing traits and one
100
duration trait. The two timing traits are the date of leaf unfolding (LU) and the date of leaf
101
senescence (LS). The duration of the canopy (CD), the duration trait, is determined from the two
102
timing traits as follows: CD = LS – LU. Genetic differentiation between populations in several
103
tree species has been documented for various phenological measurements (review in: Alberto et
104
al., 2013a).
105
Spring phenology (here LU) is known to be crucially sensitive to climate change and has
106
therefore been extensively investigated (Aitken et al., 2008; Alberto et al., 2013a).
107
Comparatively, fall phenology has been largely neglected (Gallinat et al., 2015), but several
108
recent studies have highlighted its importance for deciduous tree populations in a global warming
109
context (Gill et al., 2015; Xie et al., 2015). The significant consequences for ecosystem
110
productivity, carbon cycling, competition and food web of shifts in plant phenology (Richardson
111
et al., 2012) appeal detailed explorations of their modalities. Investigations of the evolutionary
112
relevance of the various biological aspects related to climatic adaptation will require us to
113
position the phenological cycle within a quantitative genetic framework. We provide here a
114
quantitative genetic strategy for exploring the interplay between the two timing traits (LU and LS)
5
115
and the duration trait (GS) and probable consequences in terms of their expected evolutionary
116
changes.
117
Leaf phenological traits in the spring and the fall have been shown to be positively
118
correlated from year to year, at the ecosystem scale (Keenan & Richardson, 2015). However, it
119
remains unclear whether this correlation has a genetic basis. This is a key question for long-lived
120
tree populations having to rapidly respond to changes in climate. Indeed, genetic correlations
121
between phenological traits may influence the adaptation process (Forrest & Miller-Rushing,
122
2010), although this aspect has not yet been investigated.
123
The different biological implications of the three traits described above may go some way
124
to explaining why they are usually analyzed separately (e.g. Savolainen et al., 2004; Alberto et
125
al., 2011). For example, in oaks, leaf unfolding takes place at the same time as flowering and
126
pollen emission. Thus, assortative mating through long-distance pollen flows would be expected
127
to interact with local adaptation in the evolution of this trait (Soularue & Kremer, 2012; 2014).
128
No such interaction would be expected for leaf senescence date. The timing of leaf senescence in
129
the fall is influenced by photoperiod (Richardson et al., 2006; Delpierre et al., 2009; Vitasse et
130
al., 2011), whereas the principal environmental factor controlling leaf unfolding date is
131
temperature-dependent (Menzel & Fabian, 1999; Chmielewski & Rötzer, 2001; Vitasse et al.,
132
2009c). However, the two “timing” traits are together exposed to a trade-off between maximizing
133
canopy duration (earlier flushing and later senescence) and the avoidance of late-spring and
134
early-fall frost damage to the canopy (i.e., favoring later flushing and earlier senescence). Canopy
135
frost damage is associated with high fitness costs (for a review: Vitasse et al., 2014), including
136
physical damage to soft tissues in the spring and leaf abscission before the total resorption of
137
nitrogen content by hard tissues in the fall (Norby et al., 2003).
6
138
We focused on sessile oak (Quercus petraea (Matt.) Liebl.) populations spread along a
139
replicated elevation gradient in two valleys in the Pyrenees (France). The distribution of these
140
populations encompasses the entire elevation range for the species in the Pyrenees (0 – 1600 m),
141
and previous studies have documented the phenotypic and genetic variation of phenological traits
142
with elevation (Vitasse et al., 2009a; Alberto et al., 2011). This study system therefore provided
143
us with an ideal biological context for investigating multivariate genetic adaptation of the
144
phenological cycle in populations faced with extreme climatic conditions.
145
146
147
We aimed to perform a detailed and integrative quantitative genetic analysis of the leaf
phenological cycle to address the following questions:
(1) Which phenological trait is most likely to drive the evolutionary dynamics of the
148
phenological cycle: leaf unfolding, senescence or duration, or a combination of these
149
traits?
150
(2) What evolutionary scenario shaped the current distribution of genetic and evolutionary
151
variation along the elevation gradient? In particular, were genetic constraints
152
responsible for the multivariate pattern of population divergence?
153
Answering these core questions requires the estimation of the selection gradients acting
154
on the components of the phenological cycle, which is a very challenging task in long-lived
155
species, such as trees. However, we show here that assessments of the G-matrix for phenological
156
components and their current divergence in populations along elevation gradients can be used to
157
answer the two questions posed above.
158
7
159
MATERIALS AND METHODS
160
Natural populations and the common garden experiment
161
This study was based on a combination of in situ and common-garden designs for
162
estimations of the components of evolutionary (among populations), genetic and environmental
163
variation in oak populations spread along a replicated elevation gradient on the northern side of
164
the Pyrenees. These designs are briefly described below and further details about the study sites
165
and the measurement protocols for assessing spring and fall leaf phenological traits can be
166
obtained from Alberto et al. (2010; 2011; 2013b), Vitasse et al. (2009a; 2009c) and are detailed
167
in Appendix 1.
168
Spring and fall phenological traits were monitored in 10 populations from two Pyrenean
169
valleys along an elevation gradient extending from 131 to 1630 m above sea level, every year
170
(except 2008 and 2013), from the spring of 2005 to the fall of 2015. This elevation gradient
171
encompasses the entire elevation distribution of Q. petraea in the Pyrenees.
172
Open-pollinated acorns from 152 in situ mother trees were germinated in greenhouse and
173
transplanted in 2008 at a common garden settled at sea level (Toulenne research station,
174
Southeast France), providing favorable conditions for vegetative development (warm
175
temperatures beginning early in spring and no frost until late fall). There were a mean of 15
176
mother trees per population (range: 7 - 33) and the mean number of offspring per tree was 23.0
177
(range: 1 - 123). The common garden was flooded in 2010. Thus, from 2009 to 2015, these
178
sample sizes decreased slightly, to 12.1 (range: 2 - 33) and 10.7 (range: 1 - 88), respectively, due
179
to mortality.
180
181
Leaf unfolding date was assessed over nine (2005-2007, 2009-2012 and 2014-2015) years
in situ and over seven (2009-2015) years in the common garden. Leaf senescence date was
8
182
evaluated over seven (2005-2007 and 2009-2012) years in situ and five years (2009 and 2012-
183
2015) in the common garden. Thus, canopy duration and covariation between traits were based
184
on estimates for seven years in situ and five years for the common garden, resulting in robust
185
estimates of the genetic values for each individual included in the dataset. In total, the databased
186
gathers 15659 entries (sets of measurements per individuals per year).
187
188
Methods
189
Quantitative genetic dissection of phenological cycles
190
The evolutionary dynamics of the components of a tree apical bud phenological cycle can
191
be modeled by assuming that the duration of bud extension (or canopy duration, CD) is a
192
composite trait dependent on the timing of bud burst (here, leaf unfolding date, LU) and the
193
timing of bud set (here, leaf senescence, LS). A phenological cycle is therefore composed by
194
traits measured on two different scales types: an interval scale for “timing” traits (i.e. no natural
195
zero point for LU and LS) and a ratio scale for canopy duration (i.e. a natural, biologically
196
meaningful, zero point exists for CD), which render an integrative analysis of this set of traits not
197
straightforward because of spurious relationships between its components as CD = LS – LU, e.g.
198
it is easy to show that the least-square regression slope of CD on LU, bCD,LU = bLS,LU – 1.
199
We are dealing here with a deciduous canopy, but this model can describe any organ
200
replaced during the life cycle of an organism, such as the cuticle of insects or the teeth of
201
vertebrates. For any individual phenotype i, we have CDi = LSi – LUi, and the expected mean
202
genetic change in each trait between generations can be expressed as follows:
203
̅̅̅̅ = Δ𝐿𝑆
̅̅̅ − Δ𝐿𝑈
̅̅̅̅ [1]
Δ𝐶𝐷
9
204
Assuming that the two timing traits are not subject to stabilizing selection (directional
205
selection is the dominant process) and that G remains roughly constant over time, Lande’s (1979)
206
equation gives:
207
{
̅̅̅ = 𝐺𝐿𝑆 𝛽𝐿𝑆 + 𝐺𝐿𝑆,𝐿𝑈 𝛽𝐿𝑈
Δ𝐿𝑆
[2],
̅̅̅̅ = 𝐺𝐿𝑈 𝛽𝐿𝑈 + 𝐺𝐿𝑆,𝐿𝑈 𝛽𝐿𝑆
Δ𝐿𝑈
208
with the elements of the G-matrix GLU, GLS and GLS,LU the genetic variances of LU, of LS and the
209
genetic covariance between them, respectively, and βLU and βLS the directional selection
210
gradients exerted on each trait. Replacing [2] in [1] yields the general formula:
211
212
̅̅̅̅ = 𝐺𝐿𝑆 𝛽𝐿𝑆 − 𝐺𝐿𝑈 𝛽𝐿𝑈 + 𝐺𝐿𝑆,𝐿𝑈 (𝛽𝐿𝑈 − 𝛽𝐿𝑆 ) [3]
Δ𝐶𝐷
Equation [3] provides a general prediction of the expected genetic change in CD,
213
depending on the selection gradient of LU and LS and their genetic variances and covariances.
214
̅̅̅̅ by considering different strengths of selection
We will now explore different outcomes of Δ𝐶𝐷
215
on LU and LS, and different genetic variance and covariance patterns. As the breeder’s equation,
216
the Lande’s (1979) equation, predicts change from one generation to the other, but applied to a
217
larger time scale (microevolution) under certain scenarios, it predicts a proportional relationship
218
between the G-matrix and the evolutionary variance and covariance matrix of synchronous
219
differentiated populations (Ackermann & Cheverud, 2002; Bolstad et al., 2014). If we consider a
220
simple case in which there is no selection on LS, the change in CD can be simplified as follows:
221
222
̅̅̅̅ = (𝐺𝐿𝑆,𝐿𝑈 − 𝐺𝐿𝑈 ) 𝛽𝐿𝑈 [4]
Δ𝐶𝐷
When selection delays the timing of structure formation, as in the case of early spring
223
frost damage, 𝛽𝐿𝑈 > 0. In such cases, assuming an absence of genetic covariance between traits
224
̅̅̅̅ decreases by 𝐺𝐿𝑈 𝛽𝐿𝑈 in each generation.
and positive selection on LS (Fig. 1, case 1, A-B), 𝐶𝐷
10
225
̅̅̅̅ ≥ 0) would require the term (𝐺𝐿𝑆,𝐿𝑈 − 𝐺𝐿𝑈 ) to be
Evolutionary stasis or an increase in CD (Δ𝐶𝐷
226
positive or zero. Thus, under a scenario of univariate directional selection for delayed structure
227
formation, as observed in our system (see Results), responding to selection on LU while
228
compensating for the decrease in CD would require the G-matrix to satisfy the following
229
condition: 𝒃𝐚 =
230
regression of senescence on the formation timing trait (Fig. 1, case 2, C). For a phenological shift
231
of the cycle resulting mostly from selection on LU, as observed in the Pyrenean sessile oak
232
(Alberto et al. 2011, see Results section), ba ≥ 1 therefore corresponds to an adaptive variance-
233
covariance genetic pattern for the avoidance of late spring frost while maximizing yearly carbon
234
intake, approximated here by canopy duration. A positive slope of the genetic regression line (0 <
235
ba < 1) would result in a partial compensation of the decrease in CD through a correlated
236
response to positive selection on LU only. Alternatively, the decrease in CD might also be
237
compensated in the absence of genetic correlation, but this would require correlated selection on
238
LS (Fig. 1, case 3, C). Correlated changes in the two “timing” traits may therefore result from
239
either a correlated response to selection on LU or direct selection on LS.
240
𝐺𝐿𝑆,𝐿𝑈
𝐺𝐿𝑈
≥ 1 , where ba is the slope of the ordinary least squares additive genetic
We observed a decrease in the mean breeding value of canopy duration with increasing
241
̅̅̅̅ < 0 (Appendix 2, Vitasse et al., 2009c). If we assume that 𝛽𝐿𝑈 and 𝛽𝐿𝑆 are
elevation, i.e. Δ𝐶𝐷
242
̅̅̅̅ decreases during the course of evolution (as in Fig. 1, case 2),
both non-null and that Δ𝐶𝐷
243
rearranging equation [3] gives:
244
245
246
𝛽𝐿𝑈
𝛽𝐿𝑆
>
𝐺𝐿𝑆 − 𝐺𝐿𝑈,𝐿𝑆
𝐺𝐿𝑈 − 𝐺𝐿𝑈,𝐿𝑆
= 𝜃 [5]
The observed evolutionary shortening of canopy duration thus requires a proportional
relationship between the ratio of the trait-specific cumulative selection gradients and θ, a ratio of
11
247
two linear combinations of G-matrix elements (note that the sign “>” above can be reversed if
248
CD increases, or may be replaced by “=” in a case of evolutionary stasis). Thus, assuming that
249
the sign for genetic change in CD over an evolutionary time span and the estimated ancestral G-
250
matrix remain roughly constant, equation [5] can be used to determine the minimal relative
251
strength of selection that has acted cumulatively on the two “timing” traits in term of selection
252
gradients (i.e. the increase in fitness per unit of change in the trait). This makes it possible to
253
determine the difference in climate-induced selection between the two traits (in this case: 𝛽𝐿𝑈 >
254
𝜃𝛽𝐿𝑆 ). By applying Lande’s (1979) equation to a biological cycle, this simple model shows how
255
the covariance between “timing” traits and the relative strength of selection on these traits can
256
interact to govern the evolutionary dynamics of canopy duration in trees.
257
Our analysis of this model and its simplified cases suggested that the G-matrix and the
258
observed evolutionary changes in CD, LU and LS provided clues to the gradient of selection
259
acting on the component of the phenological cycle. The structure of the G-matrix is compared
260
with population divergence to infer the selection scenario driving evolution. This approach based
261
on the estimation of θ is therefore analogous to that used in studies investigating multivariate
262
genetic constraints (e.g. Schluter, 1996; Hansen & Houle, 2008; Agrawal & Stinchcombe, 2009).
263
Thus, interpreting θ in this context implies similar assumptions: the estimated G-matrix is
264
representative of the ancestral G-matrix, it has remained roughly constant, the observed
265
divergence among synchronous populations is insightful in term of the dominant evolutionary
266
processes involved through time (it was not possible to assess evolutionary changes over
267
successive generations in this long-lived species) and this divergence is driven by a contribution
268
of selective forces and stochastic factors. Here, the structural statistical properties of the
269
phenological cycle data made it possible to determine the minimal value of θ, i.e. a gross estimate
12
270
of the relative strength of the cumulative selection gradients exerted on the timing traits during
271
population divergence along the elevation gradient .θ = 1 implies that bivariate divergence is
272
unlikely to have been influenced by a selective pressures stronger on one phenological trait than
273
on the other.
274
275
276
Quantitative genetic and evolutionary estimates
Alberto et al. (2011) found that genetic variance for leaf unfolding date was much smaller
277
for populations at high elevations (more than 1000 m above sea level) than for populations at
278
lower elevations. Consequently, following the same partition (i.e. three sites per valley for low
279
elevations and two sites per valley for high elevations), all the analyses were replicated for low-
280
and high-elevation populations separately. Due to the low number of in situ individuals (mothers)
281
within each population (see Appendix 1) sample size is not sufficient to obtain reliable
282
population specific estimates of genetic (co)variances. Pooling populations in the analyses was
283
therefore necessary.
284
An integrative statistical model combining information from both in situ and common
285
garden designs was fitted. The variance of phenological traits in the system was simultaneously
286
partitioned at the among and within population levels with the following ‘stratified’ mixed-effect
287
model (where uppercase letters denote fixed-effects and lower case letters denote random-
288
effects):
289
zjkmno = Dj + (Dj)Bk + (Dj)(Bp)km + (Dj)pm + ajkmn + djkmn + εjkmno [6]
290
where for any trait z, D is the experimental design fixed effect (i.e. in situ [j = 1] or in common
291
garden [j = 2]), B is the fixed-block effect (in common garden), p is the population random effect
13
292
terms estimated separately within each design (with a Block × population interaction in common
293
garden). The model is adjusted so that among population (co)variances were estimated separately
294
within each experiment: in situ and in common garden. a is the breeding value of the individual
295
n. d is the non-genetic individual effect and ε the within-individual variation (i.e. between-year
296
replicates o). The d term therefore corresponds to the permanent environmental effect (Lynch &
297
Walsh, 1998; Kruuk, 2004), grouping among-year measurements replicated on the same
298
individual to estimate inter-individual non genetic effect (e.g. Coltman et al., 2003; Bolstad et al.,
299
2014). Consequently, the ultimate residuals quantify within individual (co)variance, and the year
300
effect should not be entered as a fixed explanatory factor in the model. Among-year within-
301
individual variation is quantified as a variance component in the ultimate residual term ε, i.e. the
302
variation level (below the non-genetic among-individual term d) estimating the influence of
303
among year climatic variation on the traits at the individual level. The variance component a
304
estimates the genetic variance in an ‘animal model’ framework (Kruuk, 2004), i.e. the within
305
population variance explained by the relatedness among individuals. Applied to a simple open-
306
pollinated offspring design for which parental phenotypic values are available, an animal model
307
makes it possible to combine information from measurements recorded both in situ (parents) and
308
in the common garden (half-sibs) to better estimate the trait’s genetic variance 𝜎𝑎2 . Univariate
309
analyses were carried out for the three phenological traits. A bivariate variance model was
310
estimated for LU and LS. All along the analysis, a constant slope of reaction norms for the traits
311
is assumed. This assumption is supported by preliminary analyses showing a clearly negligible
312
genetic variance for these slopes (Soularue J.-P., Firmat C., Ronce O., Caignard T., Delzon S.,
313
Kremer A., unpublished results).
314
315
For priors of the Bayesian model, we used zero-mean normal distributions with large
variances (108) for the fixed effects, half-Cauchy distributions with scale parameter 30 (“weakly
14
316
informative prior distribution”, Gelman, 2006) for the variance components, and inverse-Wishart
317
distributions for the residuals with a matrix parameter V corresponding to a crude guess
318
estimated from the phenotypic variance-covariance matrix, and a parameter n set to 0.002 and
319
1.002 for the uni- and bi-variate case, respectively (Hadfield, 2010). Parameter estimates were
320
not sensitive to change in the priors. The model was run for 52,000 iterations, including a burn-in
321
of 2000 iterations and a thinning interval of 50 iterations. Autocorrelation levels in the Markov
322
chain-Monte Carlo (MCMC) resampling were consistently below 0.10.
323
At the population level, the model was built to decompose (co)variances separately in situ
324
2
2
and in common garden conditions, 𝜎𝑝−𝑖𝑛
𝑠𝑖𝑡𝑢 and 𝜎𝑝−𝑐𝑜𝑚𝑚𝑜𝑛 , respectively. The former
325
encompass both genetic and environmental sources of differentiation while the second only
326
includes genetic differentiation. Change in the among population relationship between LU and LS
327
was estimated from least square regression slope of LS on LU: 𝑏𝑝 = 𝜎𝑝 (𝐿𝑈, 𝐿𝑆)/𝜎𝑝2 (𝐿𝑈). The
328
separate decomposition allows to measure the discrepancy between the bivariate genetic and
329
phenotypic population differentiation (caused by large scale environmental effects acting in situ)
330
with the contrast 𝒃𝒑−𝒄𝒐𝒎𝒎𝒐𝒏 − 𝒃𝒑−𝒊𝒏 𝒔𝒊𝒕𝒖 .
331
Using 𝜎𝑎2 , the average component of within-population genetic variation, heritability was
332
estimated as ℎ2 = 𝜎𝑎2 /(𝜎𝑎2 + 𝜎𝑑2 ), with the average between-tree difference 𝜎𝑑2 (the permanent
333
environmental effect) as the unique non-genetic term (see equation [6]). This is equivalent to the
334
average of heritabilities calculated separately for each year. We also estimated an heritabily
335
accounting for the among-year within-individual plastic variation in the denominator 𝜎𝜀2 , i.e.
336
ℎ𝜀2 = 𝜎𝑎2 /(𝜎𝑎2 + 𝜎𝑑2 + 𝜎𝜀2 ). Evolvability (Houle, 1992) was also estimated for canopy duration as
337
̅̅̅̅ 2 ,
𝑒 = 100 × 𝜎𝑎2 𝐺𝑆 /𝐶𝐷
338
interval scale, so mean-scaled measurements such as evolvability are meaningless for these traits
with ̅̅̅̅
𝐶𝐷² the squared mean of CD. The two timing traits, however, are on an
15
339
(Hansen et al., 2011). We calculated θ as in equation [4], from the G-matrix obtained with a
340
bivariate version of the model (equation [6]).
341
342
343
FST-QST comparisons
The divergence patterns of the various traits were compared, by considering the QST
344
parameter (for a comprehensive review: Leinonen et al., 2013) calculated as: 𝑄ST =
345
2
2
𝜎𝑝−𝑐𝑜𝑚𝑚𝑜𝑛
/(𝜎𝑝−𝑐𝑜𝑚𝑚𝑜𝑛
+ 2𝜎𝑎2 ), where 𝜎𝑝2 is the between-population genetic variance
346
component and 𝜎𝑎2 is the mean within-population genetic variance, both estimated from the
347
model in equation [6]. Ovaskainen et al. (2011) suggested a multivariate FST-QST comparison
348
method. Here, we needed to analyze CD separately to avoid spurious correlations and LU and LS
349
were only weakly correlated genetically (genetic R2 < 0.05, see Results). We therefore carried out
350
classical univariate Bayesian estimations, the results of which can be interpreted intuitively in
351
this case. The full distribution of MCMC samples was included in each comparison, to account
352
for uncertainty in the QST estimates. We used the FST values estimated from the 16 microsatellite
353
markers of Alberto et al. (2010) as the baseline for the evaluation of a stochastic scenario of trait
354
divergence. Two of these markers with differentiation levels suggestive of deviation from
355
neutrality (i.e. high Fst values, Alberto et al., 2010) were not discarded to provide a conservative
356
test of a hypothesis of directional selection on phenological traits (QST > FST). Following the
357
recommendation of Whitlock (2008), the QST distribution was compared to the extreme tail of the
358
FST distribution: a 95% upper bound threshold value was computed from the mean FST assuming
359
a χ2 distribution with (npopulations – 1) degrees of freedom, i.e. the Lewontin-Krakauer distribution
360
(Lewontin & Krakauer, 1973; Whitlock, 2008). Locus-specific FST values were also estimated to
361
compare QST estimates to the actual range of the genetic differentiation within the genome. All
16
362
FST estimates were replicated for low-elevation, high-elevation and all populations, with the R
363
hierfstat package (Goudet, 2005). QST computed from few populations are generally poorly
364
estimated (O'Hara & Merilä, 2005; Whitlock, 2008). This drawback was kept in mind for the
365
interpretation of the estimates based on population subsets.
366
367
RESULTS
368
Within-population (co)variations
369
Each phenological trait displayed considerable variation (Table 1), but the genetic
370
variance of LS and CD was at least twice that for LU. For each trait, most of the variation in
371
averaged individual phenotypic value was explained by genetic effects, as indicated by the
372
heritability estimates close to one for each trait. The genetic variance was markedly lower at high
373
elevation (Table 1, Fig. 2): 58% lower for LU, 71% lower for LS, and 45% lower for CD.
374
Consistently, the evolvability of CD was 0.05% for low-elevation populations, but fell to 0.03%
375
in high-elevation populations, although this decrease is not significant. A very large proportion of
376
the phenotypic variance of the traits (LU: 91%, LS: 87%, CD: 94%) was intra-individual. This
377
finding reflects the strong plastic variation observed for each of the traits studied, but their
378
individual phenotypic means were nevertheless largely explained by genetic effects (0.95 < h2 < 1
379
for all traits at the scale of the gradient, Table 1). Accounting for the high within individual
380
variance in the denominator (h2ε in Table 1), heritability dropped between 6% and 13% but
381
always remained distinct from zero. The environmental variance was similar between high- and
382
low-elevation populations for LU, but was markedly higher at low elevations for LS and CD
383
(Table 1).
17
384
The estimated G-matrices for LU and LS suggested that there was no close genetic
385
relationship between these traits (Fig. 2, Table 2). For all sets of populations, the estimated
386
genetic R2 never exceeded 5%, indicating that almost all of the variance for a particular trait (i.e.
387
at least 95%) remained available when the other trait was subject to stabilizing selection. The
388
slope of the genetic regression of LS on LU was therefore shallow (Fig. 2, Table 2), indicating
389
that there should not be a strong correlated response to selection on one trait due to directional
390
selection on the other. However, the genetic relationship between these traits, although very weak
391
(R² =0.02), was positive and above zero for the full set of populations and for those at low-
392
elevation sites, indicating a weak genetic relationship between phenological traits. With the
393
marked decrease in the variance of each trait in high-elevation populations, their genetic
394
covariance reached zero (Fig. 2 C, Table 2).
395
Regression slopes were estimated for each bivariate variance component, but the error
396
term of the model was the only term yielding a non-zero estimated slope in each case (Table 2).
397
This indicates that the two traits display a positive environmental covariance, but that this
398
covariance accounts for only a small proportion of the variance in the correlated trait (LS) at this
399
level also, with all R2 values below 3% in any case. The total phenotypic correlation within
400
populations including the within individual term was also positive and very low (R2 = 0.023
401
[0.000, 0.065]). The estimate of the θ parameter, approximating the ratio of the cumulative
402
selection gradients during population divergence given the G-matrix for the two traits (Eq. [4])
403
was systematically greater than one, although the credibility interval overlapped one in high-
404
elevation populations (Table 2). As the population mean genetic value for canopy duration
405
decreased with increasing elevation of the provenance (Vitasse et al., 2011, Appendix S2), θ can
406
be interpreted as βLU / βLS > θ, i.e. the minimum estimate of the ratio of the cumulative selection
18
407
gradients for each trait. The estimated values of θ ranged from 2.87 at low elevation to 1.86 at
408
high elevation (where the credibility interval overlaps unity), with a global estimate of 2.51.
409
Thus, to generate the observed pattern, the cumulative selection on LU was about twice as strong
410
as that on LS.
411
412
Evolutionary (co)variation
413
All traits displayed considerable between-population differentiation both in situ and in
414
common garden conditions with credibility intervals clearly distinct from zero (Table 1). Only
415
common garden estimates for among-population variance in canopy duration at low and high
416
elevation were not distinct from zero. Generally, variation in common garden is markedly
417
reduced compared to variation in situ (Table 1, Fig. 3), highlighting a strong environmental effect
418
for population phenotypic divergence (Table 1, Fig. 3).
419
A strongly negative relationship between LU and LS was observed for in situ conditions
420
(Fig. 3), while a positive relationship was observed on the much narrower range of populations
421
grown in common-garden (Fig. 3). Although the latter relationship was not distinct from zero
422
(regression slope bp-common = 0.45 [-0.29, 1.07], R2 = 0.22), the contrast between the two slope was
423
significant (bp-common - bp-in situ = 0.74 [0.01, 1.42], Table 2). Finally, the trajectory of bivariate
424
populations divergence of their averaged breeding values (common garden condition) was steeper
425
than the within population genetic least-square regression slope for these traits (ba = 0.22 [0.04,
426
0.43], Fig. 3, Table 2).
427
19
428
429
FST-QST comparisons
The FST values were about 2-3% (Fig. 4) and were reliably estimated, as indicated by the
430
narrow bootstrap-based confidence interval obtained for each set of populations: FST = 0.026
431
(95% Credibility interval based on 500 bootstrap replicates: 0.021, 0.031) for all populations, FST
432
= 0.022 (0.017, 0.027) for populations at low elevation and FST = 0.026 (0.021, 0.033) for
433
populations at high elevation. The 95% upper bound (extreme tail) of the FST distribution
434
assuming a Lewontin-Krakauer distribution were: 0.047, 0.046, and 0.055, for all, low and high
435
elevation populations, respectively.
436
There was evidence for a strong pattern of non-neutral differentiation for LU, with high
437
QST estimates, ranging from 38 to 64%, with the lower limit of the 95% credibility interval
438
always exceeding the extreme tail of the neutral (FST) expectation for each population subset
439
(Fig. 4 A-C, Table 1). A similar pattern, though less clear, was found for LS, with QST estimates
440
ranging from 35 to 62% (Fig. 4 D-F, Table 1). The 95% CIs of QST always excluded the extreme
441
tail of the FST distribution. A different pattern was obtained for CD (Fig.4 G-I) with QST ranging
442
from 9 to 26%. In this case, the upper tail of the FST distribution was included within 95% CIs of
443
QST for the low and high populations subsets. However, for the full set of population, the 95%
444
CIs of QST excluded the neutral expectation supporting directional selection on this trait.
445
The between-population variance component was systematically slightly lower for CD
446
than for the other traits and the credibility interval included zero (Table 1). These lower QST
447
values were, therefore, not merely a consequence of the particularly high level of within-
448
population genetic variance estimated for this trait. They also resulted from weaker between-
449
population differentiation for CD than for the other traits. Thus, on the basis of QST -FST
20
450
comparisons, scenarios involving strong directional selection across the elevation gradient seem
451
less likely for CD than for the two timing traits, LU and LS.
452
453
454
DISCUSSION
Phenological traits are known to be extremely plastic in trees (e.g. Vitasse et al., 2010),
455
phenotypic values can therefore fluctuate strongly from year to year affecting the estimate of
456
genetic parameters. Here, each estimate was based on at least five measurements per tree,
457
replicated across years, providing a unique opportunity to obtain more realistic averaged breeding
458
and phenotypic values, together with their population averages. The overall dissection of the
459
phenological cycle into timing and duration traits and the assessment of these traits in situ and in
460
a common garden over a number of years made it possible to identify the target traits subject to
461
natural selection and driving the evolutionary dynamics of the phenological cycle. Finally, this
462
breakdown also made it possible to propose an evolutionary scenario leading to the current
463
divergence of populations along elevation gradients.
464
465
466
Targets of selection
The genetic divergence of quantitative traits (QST) markedly exceeded neutral
467
expectations (FST) at all elevations for both timing traits, leaf unfolding (LU) and leaf senescence
468
(LS) dates. For the canopy duration (CD), a less clear picture emerged from the analyses as the
469
QST estimated within each population subsets (low and high elevations) were not significantly
470
distinct from the FST. This discrepancy might be due to the lack of statistical power for QST
471
estimates relying on a low number of populations (O'Hara & Merilä, 2005; Whitlock, 2008).
21
472
However, it cannot be excluded that this trait was probably exposed to a lower level of directional
473
selection than the other two as a pattern of QST >> FST was found for the same population subset
474
for LU and LS, and as a lower QST was found for CD than for LU and LS when the full set of
475
populations was analyzed. A lower level of differentiation for canopy duration could be due to a
476
partial compensation for the two traits, with a displacement in time of the canopy duration (Case
477
2 in Fig. 1). Thus, a positive offset in the response of LU to selection for an advanced flushing
478
date could have been partly translated into a positive response in LS (positive evolutionary
479
covariation in common garden), leading to a more stabilizing selection-like pattern for CD and
480
then buffering its reduction. This scenario fits with the weak but positive relationship found
481
between the mean genetic values of the two traits across populations. Recently, a clear QST >>
482
FST pattern was documented in Populus angustifolia for autumn phenology and not for other
483
phenological traits (Evans et al., 2016). Thus depending on the species, timing traits (either LU
484
or LS) appear to be preferentially targets of selection rather than the duration (CD).
485
The estimation of θ (equation [5]) provides a crude but independent clue of the relative
486
strength of past directional selection that impacted the two timing traits of the cycle. Although the
487
interpretation of θ relies on several assumptions as does microevolutionary interpretations of the
488
influence of G (see Methods), the θ values confirm the conclusions supported by the QST >> FST
489
results. θ estimates were around two for all populations sets and statistically above unity, except
490
in high elevation populations suggesting that past selection gradients were stronger for LU than
491
for LS. This estimate should however be interpreted with caution as it assumes negligible effect
492
of stabilizing selection, and also cumulative effects over several generations. It should be
493
complemented in the future by instantaneous assessments based on the monitoring of
494
phenological traits and reproductive success in situ (Klein et al., 2011).
22
495
In summary, both timing traits have been exposed to directional selection associated to
496
local adaption to elevation along the gradient, and selection might have been stronger on LU.
497
This has partly been translated into the shallow but positive relationship between the average LU
498
and LS breeding values of populations. There are complementary interpretations to the abiotic or
499
biotic drivers of directional selection of LU in oaks. Evolutionary changes of LU may result
500
from the avoidance to early frosts (Dantec et al., 2015). Response to biotic pressures may also
501
trigger changes of LU, as shifts in leaf unfolding date have been shown to enable plants to avoid
502
insect herbivory (Crawley & Akhteruzzaman, 1988; Tikkanen & Julkunen-Tiitto, 2003;
503
Wesołowskia & Rowińskib, 2008). Finally, pathogens as powdery mildew that expands during
504
the oak flushing period are likely to alter the timing of leaf unfolding (Desprez-Loustau et al.,
505
2010; Dantec et al., 2015).
506
Finally the higher selection gradient for LU in comparison to LS can also be interpreted in
507
terms of comparative fitness costs due to directional selection for LU vs. LS. Canopy duration is
508
correlated with yearly carbon intake and, thus, growth, a crucial trait for juvenile tree fitness in a
509
context of harsh competition for light. Growth has been shown to decrease substantially with
510
increasing elevation of provenance of origin in a similar experimental design (Vitasse et al.,
511
2009a), consistent with the decrease in CD with increasing elevation reported here. This slower
512
growth of high-elevation populations, with a shorter CD in optimal common garden conditions, is
513
consistent with a fitness cost of CD reduction. Consequently, the timing of leaf senescence may
514
have evolved to compensate, at least in part, for the delaying of leaf flushing in spring. If flushing
515
occurs too early, it can be very costly in term of fitness, due to frost damage. The cost of
516
senescence occurring “too late” is likely smaller. Early fall frosts may however one day lost in
517
spring induce leaf abscission and hamper the completion of nitrogen and carbon resorption
23
518
(Norby et al., 2003). However, a “risk-taking” genotype inducing a long canopy duration at high
519
elevation might have an advantage in terms of carbon intake during the most favorable years (late
520
first fall frost), compensating for incomplete nutrient resorption in the least favorable years (early
521
fall frost). The problem is actually more complex, because of differences in photoperiod and
522
photosynthetic capacity between the spring and the fall. Thus, one day of canopy duration in the
523
fall cannot necessarily compensate for one day lost in spring (see Vitasse et al., 2010; Bauerle et
524
al., 2012 and references therein), which can be put in relation with the insights for stronger
525
selection on spring phenology found in this study.
526
527
528
Patterns of population divergence
Although the positive evolutionary relationship between the two traits was weakly
529
supported statistically due to the narrow range of variation in common garden conditions, LU
530
and LS both exhibit strong among population differentiation. This pattern suggests that local
531
adaptation to elevation involves both spring and fall phenology, and is therefore dependent on the
532
full phenological cycle of the apical buds. In Q. petraea, Vitasse et al. (2009a) found that the
533
timings of leaf unfolding and senescence were both negatively correlated with the temperature in
534
the provenance of origin, suggesting that temperature variation may drive the genetic
535
codivergence of these traits (see Appendix S2 for an examination of the relation with elevation).
536
From the quasi-absence of genetic relationship between these traits, the hypothesis according to
537
which evolutionary variation in one trait (e.g. LS) was generated by directional selection on the
538
other trait (e.g. LU, as in Fig. 1, Scenario 2) can be definitely ruled out: correlated response is not
539
expected from such a weak genetic correlation.
24
540
Even with a much tighter genetic relationship, Lande’s (1979) equation predicts that a
541
one-day offset in LU due to selection on this trait alone would imply an expected correlated
542
response of only 0.22 days in LS. An evolutionary slope being twice as great (bLU,LS = 0.45, see
543
Fig. 3) is therefore incompatible with a strict correlated response scenario for LS. As a result,
544
some level of divergent selection of LS has likely contributed to the evolution of the phenological
545
cycle to cope with the harsh conditions at high elevation. Although genetic correlations are often
546
expected to slow the rate of adaptation and range expansion (e.g. Etterson & Shaw, 2001;
547
Duputié et al., 2012), this does not appear to be the case for leaf phenology in sessile oak in the
548
Pyrenees. Among the evolutionary scenarios displayed in Fig. 1, scenario 3 is the most likely,
549
where positive directional selection on both timing traits drives the response of each trait without
550
any genetically correlated response It involves population divergence driven by elevation related
551
to (positive) selection acting on both traits. Overall, our results are the first to corroborate the
552
long-standing assumption that spring phenology is the primary trait affected by climate-mediated
553
selection. Nevertheless, the analogous pattern – yet probably reduced – of divergence found for
554
LS suggests that greater attention should be paid to the role of fall phenology in the long-term
555
adaptation of trees to a changing climate (e.g. Gill et al., 2015; Xie et al., 2015).
556
557
558
Inversion of the among-population correlation
A surprising finding of this study was that the sign of the among populations correlation
559
between the two timing traits (LU and LS) in common garden was positive and clearly opposed to
560
the negative correlation estimated among in situ populations. The common garden experiment
561
revealed that the strong negative correlation is entirely driven by environmental effects, i.e. the
562
large scale in situ climatic conditions related to elevation. In situ, a date of leaf unfolding delayed
25
563
by one day should imply an expected change of leaf senescence date of -0.30 day. In common
564
garden, with a date of leaf unfolding delayed by one day, a change of leaf senescence date of
565
+0.45 day is expected. This difference is substantial as indicated by the positive contrast between
566
the two slopes (Table 2). Population covariation in common garden conditions is assumed to be
567
of purely genetic origin. Taking it as reference, this implies that at the scale of the elevational
568
gradient, one or several environmental factors act on the two traits to generate this strong
569
negative covariation. This suggests that the reaction norms of LU and LS at large scale are of
570
opposite sign (Pélabon et al., 2013 for a detailled graphical model). However, when the
571
environmental correlation between the traits was measured at the scale of the individual tree from
572
the replicated inter-annual measurements (by the ultimate error term of the mixed-model, with a
573
permanent environmental term, see Methods), a positive correlation is found and indicates that
574
reaction norms at a small scale (inter-annual variation within site) are of the same sign. This
575
positive relationship could have some positive effect on tree fitness by allowing a compensation
576
in fall of a growing season shortened by a cold temperature in spring. This study therefore shed
577
light on two distinct patterns of environmental covariation between phenological traits.
578
In cold years, later start of the growing season in spring could thus be plastically
579
compensated by a later end in fall, but this does not hold at the scale of the elevation gradient
580
which include harsh climatic conditions: the negative covariance is the expression of macro-
581
environmental constraints which translate in a shortened growing season at high elevation. At this
582
macro-environmental level, plasticity could be of a “passive”, non-adaptive nature (Westneat et
583
al., 2015), because mostly driven by purely physical processes varying along the gradient such as
584
spring temperature or leaf abscission due to early frosts in the fall.
26
585
In agreement with our results, a reciprocal transplant experiment performed within the
586
same system (elevation range: 0-1200 m), showed a plastic response for a delayed LU date and a
587
shorter CD at high elevation (Vitasse et al., 2010). Thus, macro-environmental factors are
588
suspected to drive this negative correlation between traits and hide a weak but positive genetic
589
relationship between these traits at the scale of the gradient. In term of phenological trait’s
590
coupling, local genetic adaptation was likely opposed to macro-environmental effects on canopy
591
duration.
592
593
594
Evolution of G: causes and consequences
A spectacular erosion of about 50% of the genetic variance along the elevation gradient
595
was observed for LU (Alberto et al., 2011). Here, we document a similar pattern for LS and CD.
596
The evolutionary potential of the entire phenological cycle therefore decreased with adaptation to
597
high elevation. Evolvability, a standardized measurement of evolutionary potential (Hansen et al.,
598
2011), can be calculated for CD only, and was found to be 40% smaller at high than at low
599
elevation. However, it appears unlikely that this large decrease in evolvability would slow local
600
adaptation to the point of jeopardizing population survival. First, the evolvability of CD, even at
601
high elevation (e = 0.03%), remained within the range of evolvabilities capable of producing a
602
substantial change in the trait mean over a limited number of generations under average selection
603
intensity (Hansen et al., 2011). Furthermore, the quasi absence of genetic correlation between
604
traits results in almost all (here, more than 95%) the variation of one trait being available to
605
respond to selection if the other trait is fixed by stabilizing selection. In addition, one may recall
606
that substantial amount of genetic variation can be maintained by long distance gene flow in trees
27
607
(Kremer et al., 2012; 2014). Thus, gene flow may be responsible for the maintenance of
608
substantial genetic variation even at higher elevation.
609
Finally, correlated selection is thought to align the orientation of the G-matrix with the
610
main features of the adaptive landscape, thus influencing the pattern of genetic variation (Arnold
611
et al., 2001). This hypothesis is supported by evidence from various simulation approaches
612
(Arnold et al., 2008; Pavlicev et al., 2010; Jones et al., 2014). However, it remains unclear how
613
much the orientation of G can be altered or maintained by selection in natura (e.g. Roff &
614
Fairbairn, 2012; Firmat et al., 2014; Teplitsky et al., 2014a), as we are lacking empirical data to
615
support this hypothesis. In our study system, as discussed earlier, adaptation to high elevation
616
likely resulted from positive directional selection on the two timing traits, and, theoretically, G
617
would therefore be predicted to be aligned with the direction of bivariate divergence (i.e. positive
618
genetic correlation would appear). However, despite the marked erosion of genetic variances, the
619
between-trait genetic relationship did not evolve to match closely the main vector of population
620
divergence. Instead, genetic covariation tended to vanish with the decrease in variances. This
621
suggests that the orientation of G may be robust to the influence of strong directional selection.
622
Nevertheless, the abundant genetic variation documented in all directions here may not have had
623
a strong enough limiting effect to lead to selection on the orientation of G. Finally, it is worth
624
noticing that canopy duration, as a compound trait, can evolve only through changes in LU and/or
625
LS. Its capacity to respond to selection is given by: 𝜎𝑎2𝐶𝐷 = 𝜎𝑎2𝐿𝑈 + 𝜎𝑎2𝐿𝑆 − 2𝜎𝑎 𝐿𝑈,𝐿𝑆 . The
626
evolutionary potential of CD is decreased by a strong positive genetic covariance between the
627
traits (𝜎𝑎 𝐿𝑈,𝐿𝑆 > 0). Thus, as CD is a trait determining growth rate and then fitness (see above), it
628
cannot be excluded that selection favors a genetic decoupling between LU and LS. For example,
629
in a global warming context, the optimum for leaf unfolding date is expected to be advanced (i.e.,
28
630
negative selection differential. For Q. petraea see Vitasse et al., 2009b) and the optimum for leaf
631
senescence delayed (i.e., positive selection differential). The weak positive genetic correlation
632
found between these traits is not expected to impose constraints on evolutionary response to
633
negative correlative selection induced by warming conditions. LU and LS dates can therefore
634
evolve independently so that canopy duration and timing traits values can be rapidly optimized
635
with regards of the local climatic conditions.
636
637
638
Concluding remarks and summary
We describe here the first integrated evolutionary quantitative genetic dissection of a full
639
leaf phenological cycle of the yearly tree growth. The analysis of phenological cycles is
640
technically challenging because they are composed of different timing and duration traits related
641
by spurious correlations. We quantified the relative importance of the components of the
642
phenological cycle for local adaptation to a steep and ample elevational gradient in sessile oak.
643
Our results confirm that the timing of leaf unfolding date is a highly critical trait for local
644
adaptation but that leaf senescence date is also clearly involved, but probably to a lesser extent.
645
Thus, adaptation to a broad range of climatic conditions probably involves correlative selection
646
on leaf senescence timing to maximize canopy duration in extreme environments. We found that
647
the two types of phenological traits considered (spring and fall traits) displayed only very weak
648
genetic integration and were, therefore, free to evolve independently of each other. Because of
649
the weak genetic integration of phenological traits and their abundant genetic and plastic
650
variances, our results suggest that phenological evolution will not be a limiting component of
651
short term adaptation of sessile oak to a warming climate.
29
652
653
Acknowledgments
654
This research was supported by the ANR grant #13-ADAP-0006-03 MECC (Mechanisms of
655
Adaptation to Climate Change) and by the EU ERC project #FP7-339728 TREEPEACE (The
656
pace of Microevolution in Trees). We thank the staff of the UMR BIOGECO and the INRA
657
experimental units of Pierroton and Toulenne for conducting the phenological monitoring in the
658
common garden and natural populations. We have no conflict of interest to declare.
659
660
REFERENCES
661
Ackermann, R.R. & Cheverud, J.M. 2002. Discerning evolutionary processes in patterns of
662
663
664
665
tamarin (genus Saguinus) craniofacial variation. Am. J. Phys. Anthropol. 117: 260-271.
Agrawal, A.F. & Stinchcombe, J.R. 2009. How much do genetic covariances alter the rate of
adaptation? Proc. Roy Soc. Lond B 276: 1183-1191.
Aitken, S.N., Yeaman, S., Holliday, J.A., Wang, T. & Curtis-McLane, S. 2008. Adaptation,
666
migration or extirpation: climate change outcomes for tree populations. Evol. Appl. 1: 95-
667
111.
668
Alberto, F., Niort, J., Derory, J., Lepais, O., Vitalis, R., Galop, D., et al. 2010. Population
669
differentiation of sessile oak at the altitudinal front of migration in the French Pyrenees.
670
Mol. Ecol. 19: 2626-2639.
671
Alberto, F., Bouffier, L., Louvet, J.M., Lamy, J.B., Delzon, S. & Kremer, A. 2011. Adaptive
672
responses for seed and leaf phenology in natural populations of sessile oak along an
673
altitudinal gradient. J. Evol. Biol. 24: 1442-1454.
30
674
Alberto, F.J., Aitken, S.N., Alía, R., González-Martínez, S.C., Hänninen, H., Kremer, A., et al.
675
2013a. Potential for evolutionary responses to climate change – evidence from tree
676
populations. Glob. Change Biol. 19: 1645-1661.
677
Alberto, F.J., Derory, J., Boury, C., Frigerio, J.-M., Zimmermann, N.E. & Kremer, A. 2013b.
678
Imprints of natural selection along environmental gradients in phenology-related genes of
679
Quercus petraea. Genetics 195: 495-512.
680
681
682
683
684
Arnold, S.J., Pfrender, M.E. & Jones, A.G. 2001. The adaptive landscape as a conceptual bridge
between micro- and macroevolution. Genetica 112: 9-32.
Arnold, S.J., Bürger, R., Hohenlohe, P.A., Ajie, B.C. & Jones, A.G. 2008. Understanding the
evolution and the stability of the G-matrix. Evolution 62: 2451-2461.
Bauerle, W.L., Oren, R., Way, D.A., Qian, S.S., Stoy, P.C., Thornton, P.E., et al. 2012.
685
Photoperiodic regulation of the seasonal pattern of photosynthetic capacity and the
686
implications for carbon cycling. Proc. Natl. Acad. Sci. USA 109: 8612-8617.
687
Bolstad, G.H., Hansen, T.F., Pélabon, C., Falahati-Anbaran, M., Pérez-Barrales, R. &
688
Armbruster, W.S. 2014. Genetic constraints predict evolutionary divergence in
689
Dalechampia blossoms. Philos. Trans. R. Soc. Lond. B Biol. Sci. 369.
690
691
692
693
694
695
Chmielewski, F.-M. & Rötzer, T. 2001. Response of tree phenology to climate change across
Europe. Agr. Forest. Meteorol. 108: 101-112.
Coltman, D.W., O'Donoghue, P., Jorgenson, J.T., Hogg, J.T., Strobeck, C. & Festa-Bianchet, M.
2003. Undesirable evolutionary consequences of trophy hunting. Nature 426: 655-658.
Crawley, M.J. & Akhteruzzaman, M. 1988. Individual variation in the phenology of oak trees and
its consequences for herbivorous insects. Funct. Ecol. 2: 409-415.
31
696
Dantec, C.F., Ducasse, H., Capdevielle, X., Fabreguettes, O., Delzon, S. & Desprez-Loustau, M.-
697
L. 2015. Escape of spring frost and disease through phenological variations in oak
698
populations along elevation gradients. J. Ecol. 103: 1044-1056.
699
Delpierre, N., Dufrêne, E., Soudani, K., Ulrich, E., Cecchini, S., Boé, J., et al. 2009. Modelling
700
interannual and spatial variability of leaf senescence for three deciduous tree species in
701
France. Agr. Forest. Meteorol. 149: 938-948.
702
Desprez-Loustau, M.L., Vitasse, Y., Delzon, S., Capdevielle, X., MarÇAis, B. & Kremer, A.
703
2010. Are plant pathogen populations adapted for encounter with their host? A case study
704
of phenological synchrony between oak and an obligate fungal parasite along an
705
altitudinal gradient. J. Evol. Biol. 23: 87-97.
706
Duputié, A., Massol, F., Chuine, I., Kirkpatrick, M. & Ronce, O. 2012. How do genetic
707
correlations affect species range shifts in a changing environment? Ecol. Lett. 15: 251-
708
259.
709
Erwin, D.H. 2009. Climate as a driver of evolutionary change. Curr. Biol. 19: R575-R583.
710
Etterson, J.R. & Shaw, R.G. 2001. Constraint to Adaptive Evolution in Response to Global
711
Warming. Science 294: 151-154.
712
Evans, L.M., Kaluthota, S., Pearce, D.W., Allan, G.J., Floate, K., Rood, S.B., et al. 2016. Bud
713
phenology and growth are subject to divergent selection across a latitudinal gradient in
714
Populus angustifolia and impact adaptation across the distributional range and associated
715
arthropods. Ecol Evol 6: 4565-4581.
716
Firmat, C., Lozano-Fernández, I., Agustí, J., Bolstad, G.H., Cuenca-Bescós, G., Hansen, T.F., et
717
al. 2014. Walk the line: 600000 years of molar evolution constrained by allometry in the
718
fossil rodent Mimomys savini. Philos. Trans. R. Soc. Lond. B 369.
32
719
720
721
722
723
724
725
726
727
728
729
730
731
Forrest, J. & Miller-Rushing, A.J. 2010. Toward a synthetic understanding of the role of
phenology in ecology and evolution. Phil. Trans Roy. Soc. B. Biol Sci. 365: 3101-3112.
Franks, S.J., Weber, J.J. & Aitken, S.N. 2014. Evolutionary and plastic responses to climate
change in terrestrial plant populations. Evol. Appl. 7: 123-139.
Gallinat, A.S., Primack, R.B. & Wagner, D.L. 2015. Autumn, the neglected season in climate
change research. Trends Ecol. Evol. 30: 169-176.
Gelman, A. 2006. Prior distributions for variance parameters in hierarchical models(Comment on
an Article by Browne and Draper). Bayesian Anal. 1: 515-533.
Gienapp, P., Teplitsky, C., Alho, J.S., Mills, J.A. & MerilÄ, J. 2008. Climate change and
evolution: disentangling environmental and genetic responses. Mol. Ecol. 17: 167-178.
Gienapp, P., Reed, T.E. & Visser, M.E. 2014. Why climate change will invariably alter selection
pressures on phenology. Proc. Roy. Soc. Lond. B. Biol. Sci. 281.
Gill, A.L., Gallinat, A.S., Sanders-DeMott, R., Rigden, A.J., Short Gianotti, D.J., Mantooth, J.A.,
732
et al. 2015. Changes in autumn senescence in northern hemisphere deciduous trees: a
733
meta-analysis of autumn phenology studies. Ann. Bot. 116: 875-888.
734
735
736
737
738
739
740
741
Goudet, J. 2005. hierfstat, a package for r to compute and test hierarchical F-statistics. Mol. Ecol.
Notes 5: 184-186.
Hadfield, J.D. 2010. MCMC methods for multi-response generalized linear mixed models: The
MCMCglmm R Package. J. Stat. Soft. 33: 1-22.
Hansen, T.F. 2003. Is modularity necessary for evolvability? Remarks on the relationship
between pleiotropy and evolvability. Biosystems 69: 83-94.
Hansen, T.F. & Houle, D. 2008. Measuring and comparing evolvability and constraint in
multivariate characters. J. Evol Biol. 21: 1201-1219.
33
742
743
744
Hansen, T.F., Pelabon, C. & Houle, D. 2011. Heritability is not evolvability. Evol. Biol. 38: 258277.
Hansen, T.F. & Voje, K.L. 2011. Deviation from the line of least resistance does not exclude
745
genetic constraints: a comment on Berner et al. (2010). Evolution 65: 1821-1822.
746
Houle, D. 1992. Comparing evolvability and variability of quantitative traits. Genetics 130: 195-
747
748
749
750
751
752
204.
Hunt, G. & Roy, K. 2006. Climate change, body size evolution, and Cope's Rule in deep-sea
ostracodes. Proc. Natl. Acad. Sci. USA 103: 1347-1352.
Jones, A.G., Bürger, R. & Arnold, S.J. 2014. Epistasis and natural selection shape the mutational
architecture of complex traits. Nature comm. 5: 3709-3709.
Keenan, T.F. & Richardson, A.D. 2015. The timing of autumn senescence is affected by the
753
timing of spring phenology: implications for predictive models. Glob. Change Biol. 21:
754
2634-2641.
755
Klein, E.K., Carpentier, F.H. & Oddou-Muratorio, S. 2011. Estimating the variance of male
756
fecundity from genotypes of progeny arrays: evaluation of the Bayesian forward
757
approach. Meth. Ecol. Evol. 2: 349-361.
758
Kremer, A., Ronce, O., Robledo-Arnuncio, J.J., Guillaume, F., Bohrer, G., Nathan, R., et al.
759
2012. Long-distance gene flow and adaptation of forest trees to rapid climate change.
760
Ecol. Lett. 15: 378-392.
761
762
763
764
Kremer, A., Potts, B.M. & Delzon, S. 2014. Genetic divergence in forest trees: understanding the
consequences of climate change. Funct. Ecol. 28: 22-36.
Kruuk, L.E.B. 2004. Estimating genetic parameters in natural populations using the "animal
model". Phil. Trans. R. Soc. Lond. B Biol. Sci. 359: 873-890.
34
765
766
767
768
769
770
771
Lande, R. 1979. Quantitative genetic-analysis of multivariate evolution, applied to brain - body
size allometry. Evolution 33: 402-416.
Lande, R. & Arnold, S.J. 1983. The measurement of selection on correlated characters. Evolution
37: 1210-1226.
Lande, R. 2009. Adaptation to an extraordinary environment by evolution of phenotypic
plasticity and genetic assimilation. J. Evol. Biol. 22: 1435-1446.
Leinonen, T., McCairns, R.J.S., O'Hara, R.B. & Merila, J. 2013. QST-FST comparisons:
772
evolutionary and ecological insights from genomic heterogeneity. Nat. Rev. Genet. 14:
773
179-190.
774
775
776
Lewontin, R.C. & Krakauer, J. 1973. Distribution of gene frequency as a test of theory of
selective neutrality of polymorphisms. Genetics 74: 175-195.
Lindner, M., Maroschek, M., Netherer, S., Kremer, A., Barbati, A., Garcia-Gonzalo, J., et al.
777
2010. Climate change impacts, adaptive capacity, and vulnerability of European forest
778
ecosystems. Forest Ecol. Manag. 259: 698-709.
779
780
Lynch, M. & Walsh, B. 1998. Genetics and analysis of quantitative traits. Sinauer Associates
Inc., Sunderland.
781
Menzel, A. & Fabian, P. 1999. Growing season extended in Europe. Nature 397: 659-659.
782
Merilä, J. 2012. Evolution in response to climate change: In pursuit of the missing evidence.
783
784
Bioessays 34: 811-818.
Norby, R.J., Hartz-Rubin, J.S. & Verbrugge, M.J. 2003. Phenological responses in maple to
785
experimental atmospheric warming and CO2 enrichment. Glob. Change Biol. 9: 1792-
786
1801.
787
788
Nussey, D.H., Wilson, A.J. & Brommer, J.E. 2007. The evolutionary ecology of individual
phenotypic plasticity in wild populations. J. Evol. Biol. 20: 831-844.
35
789
790
791
O'Hara, R.B. & Merilä, J. 2005. Bias and precision in QST estimates: problems and some
solutions. Genetics 171: 1331-1339.
Ovaskainen, O., Karhunen, M., Zheng, C., Arias, J.M.C. & Merila, J. 2011. A new method to
792
uncover signatures of divergent and stabilizing selection in quantitative traits. Genetics
793
189: 621-U729.
794
Pavlicev, M., Cheverud, J.M. & Wagner, G.P. 2010. Evolution of adaptive phenotypic variation
795
patterns by direct selection for evolvability. Proc. Roy. Soc. B 278: 1903-1912.
796
Pélabon, C., Osler, N.C., Diekmann, M. & Graae, B.J. 2013. Decoupled phenotypic variation
797
between floral and vegetative traits: distinguishing between developmental and
798
environmental correlations. Ann. Bot. 111: 935-944.
799
Rehfeldt, G.E., Ying, C.C., Spittlehouse, D.L. & Hamilton, D.A. 1999. Genetic responses to
800
climate in Pinus contorta: niche breadth, climate change, and reforestation. Ecol. Monogr.
801
69: 375-407.
802
803
804
Richardson, A.D., Bailey, A.S., Denny, E.G., Martin, C.W. & O'Keefe, J. 2006. Phenology of a
northern hardwood forest canopy. Glob. Change Biol. 12: 1174-1188.
Richardson, A.D., Anderson, R.S., Arain, M.A., Barr, A.G., Bohrer, G., Chen, G., et al. 2012.
805
Terrestrial biosphere models need better representation of vegetation phenology: results
806
from the North American Carbon Program Site Synthesis. Glob. Change Biol. 18: 566-
807
584.
808
809
Roff, D.A. & Fairbairn, D.J. 2012. A test of the hypothesis that correlational selection generates
genetic correlations. Evolution 66: 2953-2960.
810
Savolainen, O., Bokma, F., Garcı́a-Gil, R., Komulainen, P. & Repo, T. 2004. Genetic variation in
811
cessation of growth and frost hardiness and consequences for adaptation of Pinus
812
sylvestris to climatic changes. For. Ecol. Manag. 197: 79-89.
36
813
814
815
816
817
818
819
820
Schluter, D. 1996. Adaptive radiation along genetic lines of least resistance. Evolution 50: 17661774.
Soularue, J.P. & Kremer, A. 2012. Assortative mating and gene flow generate clinal phenological
variation in trees. BMC Evol. Biol. 12 79
Soularue, J.P. & Kremer, A. 2014. Evolutionary responses of tree phenology to the combined
effects of assortative mating, gene flow and divergent selection. Heredity 113: 485-494.
Teplitsky, C. & Millien, V. 2014. Climate warming and Bergmann's rule through time: is there
any evidence? Evol. Appl. 7: 156-168.
821
Teplitsky, C., Robinson, M.R. & Merilä, J. 2014a. Evolutionary potential and constraints in wild
822
populations. In: Quantitative Genetics in wild populations (A. Charmantier, D. Garant &
823
L. E. B. Kruuk, eds). Oxford University Press.
824
Teplitsky, C., Tarka, M., Møller, A.P., Nakagawa, S., Balbontín, J., Burke, T.A., et al. 2014b.
825
Assessing multivariate constraints to evolution across ten long-term avian studies. PLoS
826
ONE 9: e90444.
827
Tikkanen, O.-P. & Julkunen-Tiitto, R. 2003. Phenological variation as protection against
828
defoliating insects: the case of Quercus robur and Operophtera brumata. Oecologia 136:
829
244-251.
830
Vitasse, Y., Delzon, S., Bresson, C., Michalet, R. & Kremer, A. 2009a. Altitudinal differentiation
831
in growth and phenology among populations of temperate-zone tree species growing in a
832
common garden. Can. J. For. Res. 39: 1259-1269.
833
Vitasse, Y., Delzon, S., Dufrene, E., Pontailler, J.-Y., Louvet, J.-M., Kremer, A., et al. 2009b.
834
Leaf phenology sensitivity to temperature in European trees: Do within-species
835
populations exhibit similar responses? Agr. Forest Meteorol. 149: 735-744.
37
836
Vitasse, Y., Porté, A., Kremer, A., Michalet, R. & Delzon, S. 2009c. Responses of canopy
837
duration to temperature changes in four temperate tree species: relative contributions of
838
spring and autumn leaf phenology. Oecologia 161: 187-198.
839
Vitasse, Y., Bresson, C.C., Kremer, A., Michalet, R. & Delzon, S. 2010. Quantifying
840
phenological plasticity to temperature in two temperate tree species. Funct. Ecol. 24:
841
1211-1218.
842
Vitasse, Y., Francois, C., Delpierre, N., Dufrene, E., Kremer, A., Chuine, I., et al. 2011.
843
Assessing the effects of climate change on the phenology of European temperate trees.
844
Agr. Forest. Meteorol. 151: 969-980.
845
846
847
848
849
850
Vitasse, Y., Lenz, A. & Körner, C. 2014. The interaction between freezing tolerance and
phenology in temperate deciduous trees. Front. Plant. Sci. 5: 541.
Wesołowskia, T. & Rowińskib, P. 2008. Late leaf development in pedunculate oak (Quercus
robur): An antiherbivore defence? Scan. J. For. Res. 23: 386-394.
Westneat, D.F., Wright, J. & Dingemanse, N.J. 2015. The biology hidden inside residual withinindividual phenotypic variation. Biol. Rev. 90: 729-743.
851
Whitlock, M.C. 2008. Evolutionary inference from QST. Mol. Ecol. 17: 1885-1896.
852
Xie, Y., Wang, X. & Silander, J.A. 2015. Deciduous forest responses to temperature,
853
precipitation, and drought imply complex climate change impacts. Proc. Natl. Acad. Sci.
854
USA 112: 13585-13590.
855
856
857
38
858
Supporting information
859
Additional Supporting Information may be found online in the supporting information tab for this
860
article:
861
Appendix S1. Supplementary Method section: Details on the study sites and the Toulenne
862
common garden design – Methods for the assessment of phenological traits.
863
Appendix S2. Supplementary Results section: Population divergence of phenological traits as a
864
function of site elevation.
865
39
866
867
868
TABLES
869
870
Table 1. Univariate variance component analysis applied to the three phenological traits of
871
the cycle and derived statistics
Traits
Leaf unfolding timing (LU)
Leaf senescence timing (LS)
Canopy duration (CD)
Components
All populations
Low elevation
High elevation
All populations
Low elevation
High elevation
All populations
Low elevation
𝝈𝟐𝒑−𝑰𝒏 𝒔𝒊𝒕𝒖 (day2)
481.81
212.8
282.7
70.10
40.62
66.97
843.91
537.44
376.94
(158.19, 1352.49)
(36.14, 1072.88)
(19.74, 2546.45)
(17.69, 205.85)
(0.12, 221.1)
(1.46, 973.04)
(275.63, 2336.86)
(112.12, 2449.74)
(29.67, 8450.06)
𝝈𝟐𝒑−𝑪𝒐𝒎𝒎𝒐𝒏 (day2)
𝝈𝟐𝒂 (day2)
𝝈𝟐𝒅 (day2)
𝝈𝟐𝜀 (day2)
QST
𝒉𝟐
𝒉𝟐𝜺
High elevation
9.01
8.07
9.76
10.85
16.67
14.17
8.95
5.79
1.92
(3.37, 22.23)
(1.21, 34.65)
(1.17, 159.68)
(3.09, 32.12)
(0.15, 141.04)
(0.55, 121.18)
(1.54, 24.88)
(0.00, 34.48)
(0.00, 47.81)
4.35
6.64
2.75
9.32
15.67
4.48
12.9
17.28
9.48
(3.78, 4.88)
(5.48, 7.72)
(2.21, 3.35)
(6.93, 11.91)
(10.56, 21.33)
(2.6, 6.86)
(9.67, 16.02)
(11.15, 23.6)
(6.71, 13.01)
0.01
0.03
0.02
0.17
0.50
0.21
0.21
0.63
0.21
(0.00, 0.09)
(0.00, 0.27)
(0.00, 0.14)
(0.00, 1.43)
(0.00, 3.61)
(0.00, 1.63)
(0.00, 1.61)
(0.00, 4.32)
(0.00, 1.55)
44.21
44.42
44.00
129.68
147.79
117.53
150.48
172.67
136.16
(43.25, 45.13)
(42.97, 46.12)
(42.92, 45.31)
(125.31, 133.6)
(140.47, 155.59)
(112.58, 122.2)
(146, 155.57)
(164.38, 183.1)
(130.00, 141.33)
0.51
0.38
0.64
0.37
0.35
0.62
0.26
0.14
0.09
(0.30, 0.74)
(0.12, 0.74)
(0.32, 1)
(0.14, 0.64)
(0.09, 0.86)
(0.23, 0.98)
(0.07, 0.51)
(0.00, 0.51)
(0.00, 0.72)
1
1
0.99
0.98
0.97
0.95
0.98
0.97
0.98
(0.98, 1.00)
(0.96, 1.00)
(0.95, 1.00)
(0.86, 1.00)
(0.78, 1.00)
(0.69, 1.00)
(0.89, 1.00)
(0.78, 1.00)
(0.85, 1.00)
0.09
0.13
0.06
0.07
0.09
0.04
0.08
0.09
0.06
(0.08, 0.10)
(0.11, 0.15)
(0.05, 0.07)
(0.05, 0.08)
(0.06, 0.13)
(0.02, 0.06)
(0.06, 0.10)
(0.06, 0.12)
(0.05, 0.09)
-
-
-
-
-
-
e (%)
0.04
0.05
0.03
(0.03, 0.05)
(0.03, 0.07)
(0.02, 0.04)
The variance components p, a, d and ε quantify variation at the between-population (in situ and in common garden conditions), genetic,
individual non-genetic and within-individual (environmental) levels, respectively, see the Methods section (equation [6]). The 95%
credibility intervals are indicated in brackets. See the Methods section for a description of QST, 𝒉𝟐 , 𝒉𝟐𝜺 and evolvability (e) estimations.
872
873
40
874
Table 2. Bivariate variance component analysis applied to the timing of leaf unfolding (LU)
875
and leaf senescence (LS), and derived statistics. For each component the first line indicates
876
the least squares regression slope b and the second line the R2 of the relation
Components
All populations
Low elevation
High elevation
𝒃𝒑−𝑰𝒏 𝒔𝒊𝒕𝒖*
-0.30 (-0.48, -0.09)
0.64 (0.12, 0.97)
0.45 (-0.29, 1.07)
0.22 (0, 0.64)
0.74 (0.01, 1.42)
-0.29 (-0.84, 0.08)
0.41 (0, 0.89)
0.65 (-0.61, 1.85)
0.27 (0, 0.81)
0.96 (-0.35, 2.07)
-0.11 (-1, 0.71)
0.18 (0, 0.79)
0.47 (-1.36, 2.58)
0.28 (0, 0.85)
0.65 (-1.25, 2.98)
0.22 (0.04, 0.43)
0.02 (0, 0.07)
0.03 (-20.93, 20.11)
0.07 (0, 0.46)
0.25 (0.21, 0.29)
0.02 (0.02, 0.03)
2.51 (1.48, 3.58)
0.33 (0.08, 0.55)
0.05 (0, 0.12)
-0.03 (-18.17, 27.29)
0.16 (0, 0.75)
0.22 (0.16, 0.29)
0.01 (0.01, 0.02)
2.87 (1.36, 4.74)
0.02 (-0.24, 0.33)
0.01 (0, 0.05)
0.12 (-19.03, 20.76)
0.16 (0, 0.78)
0.27 (0.22, 0.32)
0.03 (0.02, 0.04)
1.86 (0.96, 3.11)
𝒃𝒑−𝑪𝒐𝒎𝒎𝒐𝒏
Contrast 𝒃𝒑 †
𝒃𝒂
𝒃𝒅
𝒃𝜺
𝜽‡
* Each slope estimate corresponds to 𝒃 = 𝜎(𝐿𝑈, 𝐿𝑆)/𝜎 2 (𝐿𝑈), with the indices p, a, d and ε denoting the regression
slopes estimated at the between-population (for in situ and common garden experimental designs), genetic, individual
non-genetic and within-individual levels, respectively, see the Methods section (equation [6]). 95% credibility intervals
for b and R2 are indicated in brackets.
† Contrast bp is the difference: 𝒃𝒑−𝑪𝒐𝒎𝒎𝒐𝒏 𝒈𝒂𝒓𝒅𝒆𝒏 − 𝒃𝒑−𝑰𝒏 𝒔𝒊𝒕𝒖.
‡ θ is the expected ratio βLU / βLS required for the evolutionary stasis of canopy duration given the estimated G-matrix
parameters (see equation [5]). θ ≈ 1 implies that bivariate divergence is unlikely to have been influenced by a selective
pressures stronger on one phenological trait than on the other.
877
878
879
41
880
FIGURES
881
882
Fig. 1. Schematized scenarios of the evolution of a phenological cycle under selection for
883
delayed leaf unfolding (e.g. colonization at high elevation). Case 1. No genetic relationship
884
between the dates of leaf unfolding (LU) and leaf senescence (LS) (ba ≈ 0) and no positive
885
selection on leaf senescence. Then, (A) leaf unfolding timing responds to selection and (B) leaf
886
̅̅̅̅. Case 2.
senescence timing remains constant, resulting in a canopy duration shortened by Δ𝐿𝑈
887
Due to a positive genetic relationship between LU and LS (ba > 0), and no selection on LS, a
888
shortened canopy duration caused by early LU is (partially) compensated by an offset in LS of
889
GLU,LS βLU (C). Case 3. Similar conditions in the absence of a genetic relationship (ba ≈ 0): only
890
correlated selection on LS is expected to compensate for the decrease in canopy duration by GLS
891
βLS (C).
892
893
42
894
Fig. 2. Patterns of genetic variance and covariance for the two timing traits: leaf unfolding and
895
(LU) and leaf senescence (LS) dates estimated for different sets of populations: all the
896
populations (A), and populations at low (< 1000 m, B) and high (> 1000 m, C) elevations. Left
897
panels: Bivariate representation of the estimated G-matrices (black ellipse) and the corresponding
898
posterior distribution (gray ellipses). The ellipses represent the distribution of the bivariate
899
breeding-values centered on zero. Right panels: the same information represented as
900
(co)variances estimates (black dots) and their 95% posterior credibility intervals (gray segments).
901
902
43
903
Fig. 3. Bivariate population divergence between the timing of leaf unfolding (LU) and the timing
904
of leaf senescence (LS). Within population genetic regression slope (dashed line) is given for
905
comparison (ba = 0.22, R² = 0.02). A. This figure illustrates the range of variation and the
906
substantial difference between the phenotypic (in situ, negative: bp = -0.30, R² = 0.63) and the
907
genetic (common garden, positive: bp = 0.45, R² = 0.22) slopes for among population bivariate
908
divergence (see Table 2). B. Focus on the population (co)variation in common garden. The
909
darkness of the dots is proportional to the elevation the original sampling sites (range: 131-1630
910
m). Population estimates are the Best Linear Unbiased Predictors (BLUB) for populations
911
estimated by the mixed-effect model. DOY: day of the year (number of days since January 1st).
912
913
914
44
915
Fig. 4. QST-FST comparisons for each phenological trait, leaf unfolding (A-C), and leaf
916
senescence (D-F) timings, and canopy duration (G-I). The analysis is replicated for all the
917
populations (upper row) and for populations at low (< 1000 m, median row) and high (> 1000 m,
918
lower row) elevation. The continuous vertical red line and its surrounding orange area
919
respectively represent the mean FST value for the corresponding set of populations and its range
920
of variation over the 16 microsatellite loci. The dotted vertical red line represents the 95% upper
921
bound of the confidence interval of a Lewontin-Krakauer distribution estimated from the same
922
loci. The histogram shows the Bayesian posterior distribution (1000 samples) for QST. The
923
vertical gray bar is the median QST and the shaded area is the 95% credibility interval obtained
924
from this posterior distribution (see Table 1).
925
45