Discussion

For understorey plants in temperate regions, leafing out and flowering before canopy closure should be important for fitness (Ida and Kudo 2008, Augspurger 2008). Because the timing of these two events are often developmentally and temporally correlated, selection on either event cannot be accurately estimated independently (Lande and Arnold 1983, Kelly 1992, Diggle 1999). We showed that early leaf-out was consistently associated with early flowering in a population of L. vernus , and that phenotypic selection was acting on both traits. The direction of direct selection on leaf-out day differed among years, whereas direct selection consistently favoured earlier flowering. Selection on leaf-out day acted indirectly, via flowering time, in one of the three years. Selection favoured shorter development times in two of three years. There was no evidence for correlational selection on leaf-out and first flowering day. Taken together, our results show that flowering phenology is correlated with vegetative phenology during spring, and that selection on vegetative phenology can affect selection on flowering time. Still, in our study selection favoured early onset of flowering in all three years study, also when accounting for indirect selection via leaf-out day.
Leaf-out and first flowering day were significantly positively correlated in all three years, implying that flowering time is constrained by the timing of leaf-out in L. vernus , i.e. individuals must start their vegetative development early in order to flower early (cf. Diggle 1999, Sola and Ehrlén 2007). At the same time, there was also considerable independent variation in the two traits. Some of this variation might be attributed to differences in shoot architecture, in terms of the placement of the first inflorescence on the shoot relative to the first leaf (Diggle 1999, Sola and Ehrlén 2007). The moderately strong association between timing leaf-out and flowering initiation found in this study is in accordance with the results for other herb species (e.g. Kelly 1992, Dahlgren et al. 2007), and suggests that selection can act independently on each trait, as well as on relative timing.
We found direct phenotypic selection on leaf-out day in two of three years. Interestingly, selection was in opposite directions in these two years. Such among-year differences in the direction of selection might be related changes in trait means due to plastic responses to inter-annual variation in spring temperature. In our study, selection favoured earlier leaf-out in the year when development in spring was on average latest, but favoured later leaf-out in the year when average spring development was fastest. This pattern could be the result of that early leaf-out, relative to the population mean, implies a larger risk in years when temperatures during early spring are higher and development on average starts earlier, and that the benefits of an early development are larger in years when development on average starts later. Little is known about the agents of selection on leaf-out time in plants, but for plants where shoot development starts in early spring it is likely that weather conditions, e.g. in terms of the timing of snowmelt or late frosts events, constitute important agents of selection (cf. Inouye 2008, Augspurger 2013). In the alpine shrub Salix herbacea , selection favoured intermediate leaf-out time in sites with late snowmelt, and early leaf-out in sites with early snowmelt (Sedlacek et al. 2015), suggesting that variation in the direction of selection was mediated by the local climate. Late spring frosts likely often mediate selection for later spring development, as suggested by a study reporting that late frost events in spring primarily damaged plants in later developmental stages (Augspurger 2013). Selection on leaf-out could also be mediated by seasonal variation in light availability, and individuals that leaf-out early in spring before canopy closure are likely to have a fitness advantage due to a longer period of resource acquisition. Leafing out before canopy closure has been shown to be important for the growth and survival of understorey tree saplings (Augspurger 2008). Early spring phenology might also infer costs in terms of increased herbivory (Roy et al. 2004, Sedlacek et al. 2015). In L. vernus , grazing is likely to be the most intense early in the season when there are fewer alternative food sources, and it is possible that also variation in grazing intensity among years contributed to the observed variation in selection on vegetative phenology in our study (7% of individuals were grazed in 2013, 26% in 2014 and 38% in 2015).
The consistent selection for early flowering found in this study is in accordance with the results of previous studies with L. vernus , as well as with several other species (Harder and Johnson 2009, Munguía-Rosas et al. 2011, Ehrlén and Valdés 2020). In our study, this pattern persisted also after taking variation in vegetative phenology into account. In addition, the results were similar for analyses that relativized fitness and standardized phenology traits within vs. across years (Table 1, Appendix S5). Selection for early flowering in L. vernus has been found to be mediated by warm April temperatures (Ehrlén and Valdés 2020). In many understorey species, early flowering is likely advantageous because light availability. and possibly pollinator activity, decrease rapidly as the canopy develops (Bertin and Sholes 1993, Ida and Kudo 2008, McKinney and Goodell 2010). This advantage is likely to be particularly large under warm spring conditions. Taken together, selection on both leaf-out and first flowering day in L. vernus is likely driven by multiple biotic and abiotic factors that vary in intensity and relative importance among years.
Although it has been suggested that selection on flowering time can be mediated by correlated life-history traits, for example via correlations between flowering initiation and flowering duration or between flowering time and emergence time (Rathcke and Lacey 1985, Austen et al. 2017), we are not aware of any previous study simultaneously estimating indirect selection on flowering time and leaf-out day. In our study, there was indirect selection for early leaf-out via start of flowering in 2014, suggesting that early flowering initiation was a main benefit of early leaf-out in that year. We found no statistical support for indirect selection on flowering time acting via leaf-out day, and overall our results suggest that observed consistent selection for early flowering in L. vernus is not driven by indirect selection via the timing of leaf-out. Still, selection for later leaf-out might to some extent have counteracted and weakened selection for early flowering in one of our study years; selection for early flowering in 2015 was less than half the size of selection in the two other years. In that year, selection for early flowering was not detectable without accounting for effects of leaf-out day. This suggests that accounting for differences in vegetative phenology can affect the ability to correctly estimate selection on flowering time. However, rather than overestimating the frequency of selection for early flowering, which we might have expected given the previous literature (e.g. Rathcke and Lacey 1985, Austen et al. 2017), neglecting variation in leaf-out time in our analyses would have led to underestimation of the strength of selection for early flowering.
In our study, phenotypic selection favoured short development times between leaf-out and first flowering day in two of three study years. We are unaware of any previous studies estimating selection on the time period between leaf-out and flowering. However, it has been hypothesised that rapid development to reproduction should be favoured since it decreases the likelihood of damage before reproduction (Williams 1966, Post et al. 2008). Still, the fitness effects of a short development time between leaf-out day and first flowering day are difficult to separate from the effects of early flowering, and a short development time might simply be favoured because it allows for early reproduction. It is also possible that selection for short development time is the result of that the optimal leaf-out time is close to the optimal timing of flowering initiation, and that the fitness benefits of a short development time reflect independent benefits of leafing out and flowering during a particular period.
While we found selection on the relative timing of leaf-out and first flowering day in terms of development time, we found no evidence of correlational selection on the two traits. That is, the benefits of a shorter development time were independent of the timing of leaf-out and flowering initiation. Estimates of correlational selection are overall rare (Kingsolver et al. 2001), and for plants, we are aware of only one study reporting significant correlational selection on combinations of phenological traits (flowering time and fruit maturation, Kelly 1992). To better understand how correlations among seasonal events affect the selection on a focal phenological trait, we therefore need studies estimating selection on combinations of seasonal events, as well as correlations and indirect selection for such events.
Our analyses of phenotypic selection provide important information about how selection can act on correlated life history traits. However, it is important to remember that our findings were based on female reproductive success and not on lifetime fitness, and it is possible that selection acting via male fitness differs from selection via female fitness. However, we consider it unlikely that including also male fitness would substantially alter our estimates of selection on phenological traits. Pollen is transferred between male and female parts of flowers that have a similar phenology in L. vernus , and it is likely that donating and recipient plants on average have a similar phenology. It is also true that trade-offs between current and future reproduction implies that some of the advantages of early flowering observed in this study might be offset by reduced fitness in subsequent years. Still, such costs of reproduction appear to be relatively modest in L. vernus , and the probability of flowering is higher in individuals that flowered in the previous year than individuals that were non-reproductive (Ehrlén and Van Groenendael 2001). Lastly, to predict evolutionary responses to observed selection, it is essential to know to what extent observed trait variation correspond to genetically based variation vs. plasticity. A previous study with L. vernusshowed that much of the variation in timing of flowering was plastic (Fogelström and Ehrlén 2019), suggesting that observed phenotypic selection might not correspond to genotypic selection and lead to evolutionary responses.