Leaf phenology of thirteen African origins of baobab (Adansonia digitata (L.)) as influenced by daylength and water availability

Abstract Water availability is often described as one of the main drivers of phenology of tropical dry forests but experimental studies to identify the cues triggering phenological changes are few. In a greenhouse trial, we exposed seedlings of thirteen origins, seven from West and six from East Africa, respectively, of Adansonia digitata L.(baobab) to a well‐watered control treatment and a water withholding treatment in combination with exposure to three different daylengths (11.5, 12.0, and 12.5 hr). Responses were measured in terms of meristematic activity, number of leaves, and height growth followed over 6.5 months. Multi‐level mixed‐effects statistical models were used to analyze how environmental and inter‐population variables affected phenological behavior. Flushing was influenced by the daylength with the lowest degree of meristematic activity observed in the short daylength treatment. Daylength also influenced the number of leaves through an interaction with the water regime. The water regime influenced all variables through interactions with the origins. Seedlings subjected to water stress had higher meristematic activity, but initially lower numbers of leaves than continuously watered plants. Height growth in continuously watered plants was fastest or at par with water‐stressed plants, depending on the origin. Origins from West Africa tended to have higher meristematic activity and their phenology was found to be less influenced by water withholding than East African origins. There were no signs of significant differences between origins in their response to photoperiod. In conclusion, baobab seedlings show opportunistic behavior, setting leaves when water is available, but larger activity is found when days are long. We discuss the results in terms of triggering factors for baobab phenology and adaptation to specific environmental conditions at the site of origins.


| INTRODUC TI ON
Phenology (derived from Greek words phaino meaning to bring light and logos meaning study) is generally described as the analysis of the timing of life cycle phases or activities in plants and animals in relation to their expected potential drivers (Lieth, 1974). For trees, the annual timing of leaf flush and flowering is usually controlled by environmental cues, such as spectral composition of the light, length of the photoperiod, temperature, and water availability (Beck, Heim, & Hansen, 2004;Oquist & Huner, 2003).
In temperate and boreal ecosystems, many phenological phases are a function of temperature (Badeck et al., 2004), explaining why plants respond to the rising temperature and changing climate in general. For example, in many forests of the world (Wolkovich et al., 2012), the earlier leafing dates of trees and shrubs are interpreted as a response of the organisms to the phenomenon of the rising temperature (Polgar & Primack, 2011;Polgar, Gallinat, & Primack, 2014). The time from leaf appearance to leaf senescence defines the length of photosynthetic activity. Therefore, an early flushing and a late leaf fall extends the length of photosynthetic activity, affecting water, carbon and nutrient cycling (Panchen et al., 2015), increasing the plant productivity, the carbon dioxide uptake, and the annual net ecosystem production (Richardson et al., 2010;Wu, Gough, Chen, & Gonsamo, 2013).
The seasonality of tropical tree phenology, especially in dry areas, appears to be mainly adapted to the periodicity, duration, and intensity of drought (Justiniano & Fredericksen, 2000;Singh & Kushwaha, 2006;Wallace & Painter, 2002). The impact of water availability is not yet fully understood, partly because there are several tree types that are adapted to seasonal drought in different ways (Borchert & Rivera, 2001). While deciduous species shed their leaves during the early dry season and flush around the onset of the rainy season (Borchert, 1994), evergreen species are characterized by a more irregular leaf turnover. Stem succulent trees shed their leaves during the early dry season without being water stressed (Borchert & Rivera, 2001). For most species, the photosynthetic activity is confined to the rainy season suggesting that it is mainly controlled by the water availability (Borchert, 1994). However, water availability appears not to be the main environmental trigger of leaf flushing in woody species in the central Sahel as in fact most Sahelian species produce new leaves before the beginning of the first rains (Seghieri, Carreau, et al., 2012;Seghieri, Do, Devineau, & Fournier, 2012). In southern Africa, a similar greening before the onset of the rainy season was observed by satellite (Ryan, Williams, Grace, Woollen, & Lehmann, 2016;Tian et al., 2018). This suggests that leafing in the dry tropical trees may be triggered by other cues. For example, the study on tropical stem succulent trees conducted by Borchert and Rivera (2001) indicated that the synchronous bud break and dormancy could be induced by a change in the length of photoperiod. Still, it is not clear how phenology in seasonally dry tropical forests is influenced by the environmental conditions (Giraldo & Holbrook, 2011). It can be hypothesized that for species with widespread distribution, trees from different origins are likely to have adapted to the specific environmental conditions of the local region, resulting in different reactions to environmental stimuli, and in different timing of their phenological events. Such differences have been observed previously for temperate tree species (Aitken, Yeaman, Holliday, Wang, & Curtis-McLane, 2008;Benito Garzón, Alía, Robson, & Zavala, 2011).
Common garden studies of African tree species indicate that performance in terms of growth vary between origins and in some cases show clinal variation with parameters as for example rainfall (Bayala, Ouédraogo, & Ong, 2008;Larwanou & Reij, 2011). To the best knowledge of the authors, no papers have addressed genetic variation in phenology of different origins of tree species in dry tropical Africa although there are examples where different populations have been observed in situ for example Vihotogbé, Berg, Bongers, Sinsin, and Sosef (2014). Korbo et al. (2012) studied the seasonal development in productivity of leaves for West and East African origins of baobab in a common garden test in Mali, arguing for evolution of a geographical race or ecotype of Western African populations of baobab. Information on cues in phenology may be extracted from long-term records, but long time series of phenological data for Africa are particularly rare and almost absent for the region of West Africa (Seghieri, Do, et al., 2012). That such time series are rare is hardly surprising given the challenge of spatial scales and the lack of historical records necessary for assessing changes over time (Herrmann, Sall, & Sy, 2014). Similarly, although the development of satellite-based detection of vegetation phenology has increased the availability of both long-term satellite (Zhang, Friedl, Tan, Goldberg, & Yu, 2012) and ground-based NDVI (Normalized Difference Vegetation Index) measurements (Hmimina et al., 2013) also in the dry tropical Africa, this approach has not yet been applied in the region at the appropriate spatial scale required to identify and single out the phenology of individual woody species. Besides, this method necessarily confounds genetic and environmental effects, leaving the extent of phenological adaptation between different populations unresolved. An alternative approach that we apply in this paper is to subject seedlings of the same woody species but from 13 origins to different experimental treatments to reveal the nature of the triggering factors. This paper investigates how leaf phenology of baobab is affected by the combined effects of drought and day length, using Africa's iconic baobab tree (Adansonia digitata L.) as a case study.
Baobabs are drought resistant species and several studies have examined water use during the dry season for both adult trees Fenner, 1980) and seedlings (Bouda, Bayala, Jensen, Markussen, & Raebild, 2014;Van den Bilcke, De Smedt, Simbo, & Samson, 2013). In this study, we include plants from a large part of the distribution area from West to East of Sub-Saharan Africa (Figure 1) to analyze whether the populations show signs of variation in the leafing phenology, and whether such variation can be related to the climate at the site of origin.

| Study species
Adansonia digitata L. (baobab) is a stem succulent and one of the best known tree species in semi-arid Africa. The species is found in the Sudano-Zambezian region with 200-800 mm annual rainfall, but also in areas with higher rainfall (up to 1,400 mm) (Wickens, 1982;  (Wickens, 1982). The dry season coincides with the short days both north and south of the equator. Before the onset of rain, stem water supports the first flush of leaves .
The leaves of the baobab are simple to digitate. Trees begin each season by producing simple leaves that are soon (after 1 week) followed by 2-3-foliolate leaves, ending up with leaves that are 5-7(-9)-foliolate (Wickens, 1982).

| Selection of the origins
The seeds of Adansonia digitata L. selected for this trial were collected at 13 sites (origins) in 7 African countries (Benin, Burkina Faso, Malawi, Mali, Sudan, Tanzania, and Togo (Figure 1)) by project partners around 2005 and subsequently stored at a temperature of 4°C. The seeds were collected from 15 mother trees at each origin and kept separately for each mother tree. The origins span a range of environmental conditions characterized by annual precipitation from 650 to 1,050 mm, and maximum daylengths at the origins varied from 12.2 to 12.9 hr (Table 1).

| Seed pre-treatment and germination
A germination test prior to the establishment of the trial showed that seeds had a germination rate down to 40%. Therefore, 30 seeds were selected per mother tree (in order to obtain ~12 seedlings) and placed in 0.4 L plastic cups, soaked in boiled water and left to cool overnight (Sacande, Rønne, & Sanon, 2006). F I G U R E 1 Distribution of Adansonia digitata in Africa (gray dots) according to Wickens (1982). The origins selected for the experiment are marked with red dots, and the background colors indicate the mean annual precipitation, based on the annual average of CHIRPS (Climate Hazards Group InfraRed Precipitation with Station data) precipitation [2000][2001][2002][2003][2004][2005][2006][2007][2008][2009][2010][2011][2012][2013][2014][2015] In June 2013, the seeds were sown in a greenhouse at 3 mm depth in trays (60 × 35 × 10 cm) that could host around 200 seeds (among 6 and 7 mother trees). The soil consisted of 10% sand mixed with 90% peat soil with added NPK and micronutrients, wetted with warm water. The seedlings attained an average height of almost 10 cm after 1 month and were transplanted to 10 cm diameter plastic pots with drainage holes. The same mixture of soil and sand was used for the transplanting.

| Trial design
The study was carried out in a tropical greenhouse at Frederiksberg Campus, University of Copenhagen. Plants were arranged on six tables in two rooms (three tables in each). Starting from February 2014 (see below), the three tables in each room were subjected to three different daylengths. On each table, two water regimes were applied (see below), and within each water regime on all tables, origins were represented by 10 individuals from 10 mother trees, distributed randomly. Thus, each of the six tables contained 13 origins ×10 plants ×2 water regimes. The total number of plants was 1,560 for the whole experiment.

| Environmental conditions and treatments
The seedlings were sprayed once with Movento Spirotetramat-Bayer CropScience Limited (5 ml/10 L) in September 2013, at the beginning of the experiment, to prevent attacks from aphids.
Afterward, biological control was applied every two-three weeks with predators of aphids, thrips, and mites.
Dataloggers (Onset HOBO Ware Pro, Bourne, MA) were used to measure temperature and monitor daylength, while the relative humidity was logged by the Priva climate control system (Priva E-Measuring Box, Ontario, Canada). The measures were taken at 5 min intervals throughout the period of experiment.
The air temperature was set to a minimum of 18°C and went up to 34°C max during the day and 29°C max during the night. The relative humidity varied between 19% and 100% in both rooms.
Until October 2013, all plants were watered once per week.
Starting from October 25 (day of year [doy] 298), two water regimes were applied. The plots of control plants were irrigated once per week (called "constant"), while the plots of the water-stressed plants were irrigated every three weeks (called "water stressed"). TA B L E 1 Origins of A. digitata used in the experiment. Annual precipitation, length of dry period, mean annual temperatureand the altitude elaborated represent estimates from New loc-clim-1.10, while daylight information was obtained from the UNL (University of Nebrasca-Lincoln) Astronomy-Daylight Hours Explorer

| Data collection
Measurements were taken in the months of December, January- during the last assessment in June. The meristematic activity was registered in the last four assessments. A plant was coded "1" for meristematic activity if any of its buds showed signs of growth and "0" if no buds were growing. Counting of the number of leaves took into account the leaf differentiation described by (Wickens, 1982), and leaves were counted only when they were sufficiently developed to allow a visual identification of their morphology.
From the first assessment, the plants differentiated into simple and 2-3-4-5-foliolate leaves. The number of each type of leaf was counted on each plant at all assessments except the last, where only the total leaf number was counted.

| Statistical analysis
Statistical analysis was performed using R with the nlme, MASS, where logit (Prob(activity for individual (i) )) is the probability of presence/absence of meristematic activity, coded as 1 or 0. Fixed effects α Wr(i),daylength(i),Origin(i),doy(i) represent the set of regression coefficients for each treatment and their interactions as stated above.

B(plant i )represent the random variation between plants. To generate
p-values for each model coefficient, we used repeated measures of ANOVA. Because budburst was not assessed in the first two assessments we redefined the data set to exclude these assessments from the analysis.
For number of leaves and height growth, we used a linear mixedeffects model (Dai et al., 2004) with nested random effects (Laird & Ware, 1982), applying interactions among the four explanatory variables used to model the meristematic activity, as described above, including the effect of room (2 levels).
A Gaussian spatial residual correlation with a nugget effect work if all data are >0 (Box & Cox, 1964). Because the height was not measured at every assessment, the corresponding missing values were removed from the analysis. Visual inspection of residual plots did not reveal any obvious deviations from homoscedasticity or normality. p-values were obtained by Chi-square tests of the effects in the chosen model, and adjusted means for each effect were estimated using the lsmeans package (Lenth, 2015).
Finally, we used simple graphical analyses to investigate links between origin, climate, and performance.

| Meristematic activity
A general increase in the fraction of plants having meristematic activity was observed from doy 35 to the end of the experiment, with the most noticeable increase between day 84 and day 126 and in some origins a decrease from day 126 to day 164 (Figure 2a).
Random effects explained a considerable part of the variation, of which the largest part was due to variation between plants (Table 2).
Hence, even within the origins of the same treatments, a large variability in the response was observed.
(1) logit Prob activity for individual (i) There was a significant interaction between water regime, origin, and day of year, suggesting that the origins were affected differently by the water regime over time (Table 2). Exposure to water stress leads to a slightly larger probability of meristematic activity across all origins (χ 2 (1) = 23.63, p < 0.0001, Figure 2a). Origins (distributed as in Figure 1) from East Africa tended to have lower meristematic activity than those originating from West Africa. For plants that had been exposed to water stress, the origins from East and West Africa also followed a different temporal development, as plants from West Africa tended to have a higher level of meristematic activity at the onset of recording day 35 ( Figure 2a). The origin from Togo (Lomé), which represents a relatively moist origin and is the only origin in the trial with two rainy seasons, had the highest level of meristematic activity both for plants exposed to water stress and for plants kept under constant water supply.

The significant interaction between daylength and day of year
suggests that the three levels of daylength influenced the meristematic activity of the plants differently (Table 2). In the long day treatment (12.5 hr), the probability of meristematic activity constantly increased with day of year. In the medium daylength (12.0 hr) treatment, meristematic activity initially increased to stay at a relatively high level, while in the short-day treatment, meristematic activity always stayed at a low level, although it increased until day 126 after which it decreased (Figure 3). The interaction between origins and daylengths was not significant (χ 2 (72) = 74.36, p = 0.41), suggesting that the triggering effect of daylength may be similar on all origins.

| Height growth
The interaction between water regime, origin, and day of year was highly significant for height, indicating that growth of the plants in relationship to the two water treatments was different among the origins (Table 2). Under the steady watering in the control plants, height showed an increasing trend for most origins (Figure 2b). For the plants subjected to water stress, height growth was more modest, absent, or even negative due to death of parts of the stem.
The fastest height growth was found in the Lomè origin, and in general, the western origins were growing faster than the eastern origins under constant water supply (Figure 2b). We found no other evidence of geographical trends in growth.
Daylengths had only a limited effect through the interaction between water regime, daylength, and room (not shown), and the lack of interaction between daylength and origin suggested no difference between origins in the effect of daylength on growth (p = 0.748).

| Number of leaves
The baobab seedlings always had more single leaves than compound leaves, although the proportions varied (Figure 4).
Simple leaves also showed the largest changes. Across all treatments, the number of compound leaves tended to be largest at the assessments at doy 85 and 126. Analyzing the total number of leaves per plant, the selected model showed a significant interaction between water regime, daylength, and day of year (Table 2)

| Effects of water stress and daylengths
Baobab seedlings show a complex phenology and to some extent seem to apply an opportunistic strategy, where leaf phenology is influenced both by daylength and water availability. Plants with a F I G U R E 2 Least square means and 95% confidence intervals for the interactions between water regime, origin, and day for three observed response variables: A percentage of plants showing meristematic activity, B height growth, and C number of leaves. The lines show the model responses to water availability (dotted line: water stressed; solid line: control); Within each panel, the upper row represents origins from West Africa, while the lower row represents East Africa. Within each row, origins are organized from West to East continuous water supply tended to have larger numbers of leaves than water-stressed plants, not only during the water stress treatment but also when watering was resumed for all plants. It is wellknown how some deep-rooted and stem succulent tropical species use stored water to maintain their leaves under drought conditions (Borchert, Rivera, & Hagnauer, 2002), and baobab seedlings have  (Wickens & Lowe, 2008). Likewise, Korbo et al. (2012) showed that baobab seedlings would produce leaves when watered regularly during the cold and dry season in a study in Mali. Yet in the same study, leaf production was low during the cold short-day period and increased markedly during the warm and dry months of March, April, and May to culminate during the wet season. This indicates that baobab leaf phenology is affected by water availability in combination with temperature or daylength, but that the leaf area is regulated to be maximal during the rainy season.
The sensitivity of phenology to water availability vary between dry zone species. For example, (Elliott, Baker, & Borchert, 2006) showed that Dipterocarps in dry seasonal forests in Thailand were evergreen on low, moist sites and deciduous on high dry sites. On the other hand, (Bate & Franklin, 2015) showed that irrigation had no impact on leaf phenology in Brachychiton megaphyllus, a shrub from the northern Australian forests. With its phenology affected TA B L E 2 Results from statistical analyses for the three variables

Random effects Estimated variance Proportion of total variation
Meristematic activity by both water availability and daylength, Baobab seems to belong to species that are more sensitive to changes in water availability.
Unfortunately, our setup does not allow to determine whether water availability is a triggering factor for leaf phenology, or it is merely a resource controlling the magnitudes of responses to other cues.
Numerous coincidences between budbreak and the spring equinox have been documented for example (Elliott et al., 2006;Lawton & Akpan, 1968;Rivera et al.., 2002). Although such coincidences indicate that length of the photoperiod may be important, there is a risk that other, uninvestigated factors correlate with daylength, and therefore such claims need to be substantiated by experiments. Unfortunately, such experiments are rare for tropical tree species, although Borchert and Rivera (2001) subjected various tropical stem succulent trees to increased daylength and found that an increase in daylength of 30 min was sufficient to trigger phenological changes. In the current study, analysis of the meristematic activity confirms Borchert and Rivera (2001)'s finding that shoot growth seems to be favored by long daylengths.
Moreover, daylength (through the interaction with water availability) had a significant impact on the number of leaves (Figure 4), even though the changes over time revealed a complex pattern.  (Table 1) than at higher latitudes, the seasonal differences thus appear to be sufficient to trigger phenological events including budbreak. Interestingly, Borchert et al. (2015) suggested that the cue for phenology could be an increase in insolation (defined as the total amount of solar radiation received) rather than in daylength, based on an analysis of a wide range of species of the American continents. We estimated insolation in the different daylength treatments (data not shown) by using data from an outdoor PAR (photosynthetically active radiation) sensor and measurements of the strength of the supplementary light. This revealed only modest differences between daylength treatments, as the accumulated insolation in the 11.5 hr treatment was approximately 94% of insolation in the 12.5 hr treatment (estimated from the start of the daylength treatments to the last records at doy 164). In addition, in the 11.5 hr treatment meristematic activity was decreasing at a time where the insolation was increasing strongly. This would suggest that differences among treatments are due to differences in daylength rather than to changes in insolation.

| Differences between populations
Populations of baobab had different phenology in response to water availability as evidenced by the significant interactions between origin, water availability, and day of year. However, the observed difference between eastern and western populations in the meristematic activity immediately after release from stress ( Figure 2a) represents the only clear geographical signal in the data that we could identify.
A possible interpretation is that the East African populations have a deeper dormancy and are less responsive to water once stress is relieved. The difference between groups is supported by earlier observations of differences in growth and stress response between western and eastern populations (De Smedt et al., 2012;Korbo et al., 2012) and also by studies of chloroplast and ITS (Internal Transcribed Spacers) haplotypes, showing a major division between western and eastern/southern populations (Cron et al., 2016;Pock Tsy et al., 2009).
Within the sampled dry zone origins, rainy seasons north and south of the equator are essentially mirrored, corresponding to the summer in both hemispheres. For example, in the Sudanese location, the rainy season stretches from July to September, whereas in the tropical and sub-tropical origins south of the equator, the rainy season is approximately from December to March. This may explain the large similarities in phenology between northern and southern origins. However, it is conspicuous that the origin from Lomé, which is the only population with a bi-modal rainy season, had high meristematic activity, a large number of leaves and fast height growth. Due to the annual movements of the ITCZ (intertropical convergence zone), the rainy season in the coastal location of Lomé-Togo is distributed from the end of April to July with a second and shorter rainy season in September and October, whereas for the inland sites only one rainy season from July to September is observed (WMO 2015).
With its origin in a climate with less pronounced seasons, it could be a selective advantage to be more responsive to water availability. Similarly, Morogoro rural, the wettest origin in the East African group (Figure 1), had high levels of meristematic activity and number of leaves compared to other origins within the western group, confirming the opportunistic strategy of the species.
As the populations did not interact significantly with daylength, there is no evidence that they respond differently to the daylengths.
This suggests that they are either triggered by a common signal (e.g. daylengths ≥12 hr), or that the changes in daylength triggering differences are smaller than the 30 min increments applied in our treatments. More detailed studies are required to resolve which of the two hypotheses are correct.

| CON CLUS I ON S AND OUTLOOK
Baobab phenology is affected by both water stress and daylength.
Drought causes a loss of leaves, and when the water returns there is an increase in meristematic activity that continues if daylength is long (12 hr or more). The daylength hence triggers increased numbers of leaves but this effect was found to have less effect under humid conditions. The seedlings reacted differently depending on their origin, but only in their reaction to water availability. The question is whether or not the phenology of other, non-succulent species from dry seasonal areas will be controlled by the same cues. Several functional types have been identified with respect to leaf phenology, and it could be interesting to examine deciduous species with various leafing periods, semi-evergreen species, and evergreen species (Borchert & Rivera, 2001;Seghieri, Do, et al., 2012). Examples of such species from the study area include Acacia Senegal (L.) which has a deciduous leaf habit but occasionally flushes leaves before the onset of the rainy season, and Parkia biglobosa (Jacq.) G.Don which is almost evergreen, but sheds leaves, is flowering and has leaf flush in the middle of the dry season. This seems especially relevant in times of climatic change, where species whose phenology is controlled by daylength may make them less likely to respond to changes in water availability.

ACK N OWLED G M ENTS
The authors gratefully acknowledge partner institutions in Africa for their assistance in collection of seeds. We wish to thank gardeners at the Frederiksberg campus greenhouses for taking care of plants, and technicians and student assistants for their help in assessing the trial. Finally, we thank the two anonymous reviewers for their constructive suggestions to improve the paper. The study was financed through a University of Copenhagen scholarship to LD.

AUTH O R CO NTR I B UTI O N S
LMD and AR conceived the ideas; LMD and AR conducted the greenhouse trial and collected the data; LMD and BM analyzed the data; LMD led the writing with assistance from AR and RF.

DATA ACCE SS I B I LIT Y
The data associated with this publication are deposited at UCPH ERDA data repository and can be accessed at https://dx.doi.