Relative influence of relatedness, conspecific density and microhabitat on seedling survival and growth of an animal-dispersed Neotropical palm, Oenocarpus bataua
Abstract
We report on a 6-year, field-based experiment in north-western Ecuador, evaluating the relative importance of relatedness, conspecific density and microhabitat on the growth and survival of seedlings of an ecologically and economically important palm, Oenocarpus bataua. We planted O. bataua seedlings in high-density seed deposition sites (leks) of the long-wattled umbrellabird (Cephalopterus penduliger) and randomly selected control sites. We analysed seedling performance in relation to conspecific relatedness, conspecific density and microhabitat. Among the parameters we measured, light availability was the most important determinant for survival and growth. Higher relatedness of neighbouring experimental seedlings and higher conspecific seedling densities were both associated with reduced growth, but did not influence survival, even in umbrellabird leks. These findings are consistent with a survival advantage of umbrellabird dispersal into lek sites (i.e. directed dispersal) that counterbalances negative density-dependent processes expected to occur at these high-density deposition sites. This study highlights the importance of a multi-tiered approach to understanding palm demography and underlines the importance of seed disperser behaviour in determining seedling fate.
Introduction
Identifying the factors that regulate seedling demography improves our understanding of the mechanistic processes that govern patterns of species and genetic diversity in plants (Schupp & Fuentes, 1995; Rodriguez-Perez & Traveset, 2007). Animal-mediated seed dispersal (zoochory) creates the template upon which post-dispersal demographic processes act for many plant species (Nathan & Muller-Landau, 2000; Wang & Smith, 2002) and is often non-random in terms of the microhabitat and conspecific neighbourhood where seeds are deposited (Schupp, Jordano & Maria, 2010). The microhabitat (e.g. light, soil, hydrology) and density of nearby conspecifics at seed deposition sites are known important drivers of seedling demography (Janzen, 1970; Augspurger, 1984; Queenborough et al., 2007; Comita et al., 2010; Mangan et al., 2010; Alvarez-Loayza & Terborgh, 2011; Andersen, Turner & Dalling, 2014). In contrast, the ways in which variability in the degree of genetic relatedness within groups of seedlings impacts survival and growth remain unclear (Garcia & Grivet, 2011) and few studies have attempted to simultaneously gauge the relative influence of microhabitat, conspecific density and relatedness for seedling survival. For this reason, more information on demographic dynamics at animal seed deposition sites in relation to these factors would be useful for improving our mechanistic understanding of observed distributions of plant species and community dynamics at the local scale.
Most frugivorous animals use certain sites repeatedly (e.g. latrines, resting sites or display areas) and generate high densities of dispersed seeds at these sites (i.e. ‘contagious’ or ‘destination-based’ seed dispersal; Howe, 1989; Russo & Augspurger, 2004; Karubian & Durães, 2009). Howe & Smallwood (1982) predicted that seed dispersal might be adaptive for plants if these deposition sites are favourable for establishment (directed dispersal hypothesis). In a classic example, directed dispersal into high-light environments such as forest gaps by bellbirds promoted seed survival (Wenny & Levey, 1998). However, if directed dispersal also leads to higher densities of seeds and seedlings, survival may often be lower at these high-density sites due to higher levels of negative density-dependent (NDD) mortality (Harms et al., 2000; Terborgh, 2012; Bagchi et al., 2014). Although there is convincing evidence that NDD mortality operates close to source trees, the relative importance of this process at sites away from source trees is unclear (Comita et al., 2014). Resolving the relative strength of NDD mortality at repeatedly used dispersal sites vs. the potential benefits of directed dispersal is therefore essential for understanding how zoochory influences plant demography (Spiegel & Nathan, 2010).
The genetic composition of frugivore-dispersed seeds at these dispersal sites appears to vary widely across systems, with potentially important consequences for survival and growth trajectories. Some dispersal agents or behaviours create pools of dispersed seeds and seedlings that originate from only one or a few source plants, leading to genetic bottlenecks at these deposition sites (e.g. Grivet, Smouse & Sork, 2005; Karubian et al., 2015), whereas others generate pools of dispersed seeds and seedlings that originate from multiple sources and are quite diverse genetically (e.g. Jordano et al., 2007; Garcia & Grivet, 2011; Scofield et al., 2012). There are three main ecological mechanisms by which genetic relatedness is thought to influence plant performance: susceptibility to attack, kin selection and niche partitioning (File, Murphy & Dudley, 2012). The susceptibility to attack hypothesis suggests that genetic homogeneity among neighbours may increase the probability of attack by herbivores or pathogens. The kin selection hypothesis posits that related plants cooperate better than do groups of strangers and, as a result, that groups of related individuals should enjoy a performance advantage. In contrast, the niche-partitioning hypothesis predicts that related plants overlap to a greater degree in resource use, increasing competition and reducing performance. Many studies to date from natural systems support the niche partitioning perspective (e.g. Milla et al., 2009; Cheplick & Kane, 2010). There is experimental evidence that relatedness among spatially proximate seedlings may also mediate survival (Liu et al., 2015), but little is known about how relatedness, microhabitat and NDD processes may interact to influence seedling demography in the context of contagious seed dispersal.
In north-western Ecuador, male long-wattled umbrellabirds (Cephalopterus penduliger; hereafter umbrellabirds) disperse seeds of preferred fruit species in high densities beneath traditionally used display perches in lek sites, an example of contagious seed dispersal (Karubian et al., 2012). Somewhat surprisingly given strong support for NDD processes from other systems (above), high conspecific densities of seeds and seedlings in umbrellabird leks do not appear to be associated with any detectable survival cost at these sites relative to lower-density ‘control’ sites outside leks (Karubian et al., 2012). Male umbrellabirds forage at multiple trees surrounding the lek and then deposit these seeds at the lek, leading to high maternal seed source diversity in seed pools at leks relative to randomly dispersed seed pools at control sites (Karubian et al., 2010). Karubian et al. (2010) proposed that this high genetic variability may promote seed and seedling survival in umbrellabird leks despite high densities, but this proposition remains untested. Alternatively, it may be the case that microhabitat varies between leks vs. control areas in ways that compensate for any NDD processes that may occur in leks, but again this possibility has not been explored. Resolving the linkages between umbrellabird dispersal, microhabitat and conspecific neighbourhood environment and seedling performance therefore provides a useful context in which to improve our understanding of the mechanistic processes underlying plant demography and resulting patterns of diversity.
In the present study, we report on the results of a field-based experiment designed to assess the relative importance of microhabitat, conspecific density and relatedness on survival and growth of seedlings of the Neotropical palm Oenocarpus bataua Mart. We use the term ‘seedling’ throughout the article to refer to small juvenile plants with or without the cotyledon. Our study design compares growth and survival of seedlings experimentally planted in umbrellabird leks or in control areas outside umbrellabird leks. Based on previous work on palms and other groups, we predicted increased survival and growth in relation to increased light and reduced conspecific density. We also predicted that increased relatedness to neighbouring conspecific seedlings might be associated with reduced survival probability and growth, consistent with niche partitioning. Based on our previous work in this system (above), we predicted increased or equivalent survival and performance in leks relative to control sites, reasoning that more favourable values of light and relatedness in lek sites may outweigh the costs associated with higher densities. Our findings indicate that light availability drives observed differences in O. bataua seedling survival and growth and that relatedness among neighbouring individuals and conspecific density both influence growth, but not survival. Thus, despite higher levels of conspecific density in lek sites, rates of survival are comparable to that of controls, suggesting the existence of some compensatory advantage that balances expected NDD processes in the lek and promotes seedling survival.
Methods
Study area and organisms
The study took place in the 3500-ha Bilsa Biological Station (BBS; 79°45′W, 0°22′N, 330–730 m a.s.l.), located in the Mache-Chindul Mountains in the province of Esmeraldas in north-western Ecuador. BBS lies in the southern portion of the Chocó biogeographical zone, a region characterized by exceptionally high diversity in flora (Gentry, 1992), birds (Carrasco et al., 2013), amphibians (Ortega-Andrade et al., 2010; Jongsma et al., 2014) and other groups.
Our study organism, the canopy palm O. bataua, is widely distributed throughout Neotropical rain forest (ter Steege et al., 2013). It is a long-lived, slow-growing monoecious species that produces inflorescences of thousands of small flowers, and is effectively outcrossed (Ottewell et al., 2012). Fertilized flowers develop large-seeded, lipid-rich fruits (up to 2000 fruits per infructescence) that are consumed and dispersed by a range of large-bodied vertebrates, including birds, primates and humans. Seeds require 1–2 months to germinate and germination rates in nursery conditions are high (> 95%) regardless of whether the aril is removed (J. Karubian, unpubl. data). Oenocarpus bataua is common at BBS, where density is approximately two adults per hectare (Ottewell et al., 2012).
Umbrellabirds are large-bodied (c. 550 g) frugivorous birds endemic to the Chocó and are considered by the IUCN as vulnerable to extinction (IUCN, 2015). At BBS, umbrellabirds remove a significant proportion of ripe O. bataua fruits by ingesting fruits whole and regurgitating the seeds intact 1–4 h after ingestion (J. Karubian, unpubl. data; Karubian et al., 2012). When not foraging, male umbrellabirds aggregate in lek breeding sites that are active year-round; we have monitored seven leks in BBS since 2005, all of which are included in the current study. Most adult males hold fixed territories at a single lek, although some ‘floater’ males move between multiple leks (Karubian et al., 2012; Karubian & Durães, 2014). Males disperse > 50% of the seeds they ingest into leks (Karubian et al., 2012), yielding significantly higher densities of seeds and seedlings of O. bataua and other preferred fruit species at lek sites than at randomly selected control sites outside the lek. Umbrellabirds also generate diverse seed pools representing multiple maternal source trees at leks (Karubian et al., 2010; Scofield et al., 2012). Correlational analyses detected no evidence of elevated mortality associated with high seed densities within the lek and the probability of transition from seed to seedling did not differ between leks and control sites (Karubian et al., 2012).
Experimental design and field methods
We obtained the seedlings used in the current study as part of an earlier investigation of pollen flow in O. bataua (Ottewell et al., 2012). Mature O. bataua fruits were collected directly from infructescences on adult trees in BBS from May to June 2008 and allowed to germinate individually in 0.5-L plastic bags; soil was collected from multiple locations in BBS and thoroughly mixed before filling individual bags. Germination and subsequent growth took place in a common garden nursery located at forest edge, in dappled sunlight. In September 2008, when seedlings were 3–4 months old, they were transplanted in the field. At the time of planting, all seedlings measured 30–60 cm in total height, had two or three leaves, and were within 1 month of age of each other; our experiment was designed to control for any differences in these measures that may have existed at the time of planting (below, Fig. 1). Seedlings were planted in the forest by removing c. 1.0 L of soil with a hand shovel, carefully removing the seedling from the plastic bag in which it was germinated so as to keep most of the soil intact in the bag and placing the seedling and soil in the hole with some of the local soil. At the time of planting, each seedling was marked with an aluminium tag inscribed with a unique field number. There was no evidence of a ‘transplant effect’ (i.e. increased mortality in year 1 associated with transplant) and results were qualitatively similar when excluding the first year of data from our analyses.
Our study design was established to partition the relative importance of location (i.e. leks vs. ‘control’ sites outside the lek), relatedness (i.e. related vs. unrelated to surrounding experimental seedlings), conspecific density (a continuous measure of naturally occurring individuals surrounding experimental seedlings) and microhabitat on survival and growth of O. bataua seedlings. The experimental design employed a total of 560 seedlings sourced from 30 adult O. bataua trees (N = 16 seedlings from each of 25 adult trees; N = 32 seedlings from each of the remaining five adult trees). All adult source trees were located in a 1-km2 area in BBS. We placed seedlings in each of seven active umbrellabird leks, each of which was c. 1 ha, and seven ‘control’ sites. Control sites were obtained by randomly selecting x, y coordinates within the polygon bounded by the seven focal leks; mean pairwise distance between each control plot and the nearest lek was 620 m. All experimental sites were located in contiguous forest that contained a complex matrix of primary, selectively logged and secondary habitat types. In each of these 14 sites, we set up five experimental seedling plots in a star design, with a central plot and four distal plots to the north, south, east and west (Fig. 1). The distance between the central plot and each of the four distal plots was 15 m. Each plot contained eight seedlings in total: four related seedlings and four unrelated seedlings (below). In each plot, related and unrelated seedlings were planted in discrete 0.5-m2 areas; each individual seedling was separated from its nearest neighbour by 0.5 m and the related vs. unrelated subplots were separated from each other by 1.0 m (these subplots were not treated as independent units). We used eight seedlings from each of five source trees in each of our 14 locations, for a total of 40 seedlings per location × 14 locations = 560 seedlings. Seedling survival and growth were monitored via censuses at 9- to 12-month intervals following planting from September 2008 to July 2014, for a total of six census points. We recorded whether each seedling was alive or dead during each census; if alive, we recorded total height from base to tip of the longest leaf and the total leaf number.
Predictor variables
For the variable ‘location’, we assigned each of our 14 sites as either an umbrellabird lek or control. For the variable ‘related’, we distinguished between related seedlings we knew to be either full- or half-sibs (i.e. because they were collected as fruit from the same maternal tree; above) vs. unrelated seedlings that were initially collected from different adult trees. This classification scheme does not consider the identity of the pollen source for each seed, so it is possible that two seedlings that we treated as unrelated because they were collected from different source trees actually shared the same sire. However, because this study population is known to be highly outcrossed and to exhibit relatively high pollen source diversity among progeny arrays (Ottewell et al., 2012), this binary metric of relatedness probably reflects a meaningful difference in degree of relatedness among seedlings.
Conspecific density was recorded at each of the five experimental plots per site, yielding 70 data points in total. To measure conspecific density, we recorded the number of O. bataua seedlings and juveniles within the entire area of a 5-m-radius circle around the centre of each paired plot in 2008. We also counted the number of adults occurring within a 5-m-radius circle around each plot, a distance that we considered to be appropriate because NDD effects are most pronounced in close proximity to conspecific adults (Janzen, 1970).
Environmental data were also recorded at each of our 70 experimental plots. We used data collected in 2008 (i.e. when the seedlings were planted) for statistical analysis presented below because initial conditions seemed most likely to influence survival (see 3); we also replicated all analyses using 2014 environmental data and recovered qualitatively similar results to those presented below. Canopy cover was measured using a concave spherical densiometer (Forestry Suppliers no. 43888) from the centre of each plot. The number of cells (96 total) containing a majority of canopy were scored and multiplied by 1.04 to arrive at a score potentially ranging from 0 (i.e. completely open) to 100 (i.e. completely closed canopy). Independent measures were taken in each of the four cardinal directions and then averaged for a single score for each plot. The number of large trees, which we used as another measure of canopy cover as well as forest structure, was calculated by counting the number of trees with a diameter at breast height (dbh) > 50 cm within a 10-m radius of each plot. Canopy height was estimated visually after training with a digital range finder. Slope (a variable we did not include in most analyses; see below) was measured in degrees using a clinometer pointed across the steepest transect of the paired plot, and elevation was recorded at each paired plot using a handheld GPS units with error of ± 10 m a.s.l. We were not able to measure soil chemistry or hydrology, two environmental factors that may be important for palm recruitment (Eiserhardt et al., 2011).
Statistical analyses
We modelled the survival and growth of O. bataua seedlings using generalized linear mixed models (Bolker et al., 2009). Because we were interested in the relative effects of location (i.e. lek vs. control), relatedness, conspecific density and microhabitat, we included the following variables as fixed effects in our initial models: location (binary: planted in lek = 1, planted in control sites = 0), related (binary: planted with related seedlings = 1, or not = 0), canopy cover, whether there was at least one tree > 50 cm dbh within a 10-m radius (binary), canopy height, slope, elevation, whether there was at least one adult O. bataua in a 5-m radius (binary), and the number of O. bataua seedlings and juveniles in a 5-m radius. We included plot nested within site (N = 70) and site (N = 14) as random effects to account for the dependence between individuals planted in the same plot and to account for unmeasured environmental variables at each plot and site. We also used maternal tree identity (N = 30) as a random effect to account for potential unmeasured genetic characteristics among half- or full-siblings that share a common mother. We modelled survival (whether a seedling survived from 2008 to 2014) with a binomial error structure, height in 2014 with a Gaussian error structure (a standard linear mixed model) and total number of leaves in 2014 with a Poisson error structure (Bolker et al., 2009). We then tested for differences in microhabitat and conspecific neighbourhood characteristics between lek and control sites using plots as sampling units (N = 70) and site as a random effect (N = 14). For each microhabitat and conspecific neighbourhood characteristic, we ran a separate model with lek vs. non-lek as a binary predictor and the characteristic as the response variable, for a total of six models. The number of seedlings and juveniles within a 5-m radius was modelled using Poisson error structure, and trees > 50 cm dbh and number of adults within 5 m were modelled with a binomial error structure. All other habitat and conspecific neighbourhood characteristics were modelled using a standard linear mixed model with a Gaussian error structure.
Models were fitted in the R package lme4 (Bates et al., 2015). We assessed model residuals for deviations from model assumptions (Bolker et al., 2009). A moderately strong correlation between slope and the number of seedlings and juveniles within 5 m (r = −0.45, t = −4.20, df = 68, P < 0.001) yielded a variance inflation factor (VIF) greater than the VIF = 2 threshold suggested by Zuur, Ieno & Elphick (2010), beyond which point colinearity between predictor variables can lead to low power in detecting weak ecological effects. The most effective way to address this issue is to remove one of the colinear variables from the models (Zuur et al., 2010). As one of the core goals of this study was to assess the effects of O. bataua densities on seedling growth and survival, we elected to remove slope from analyses and retain conspecific seed and seedling density as a predictor variable. No other predictor variables had a VIF > 2 for any model.
Continuous covariates were mean centred and standardized by dividing by two standard deviations (SD) prior to analyses to aid in model convergence, to allow direct comparison of regression coefficients of binary and continuous covariates and to standardize effect sizes, which can be used as a point of comparison in future studies (Gelman, 2008; Schielzeth, 2010). After scaling, regression coefficients can be interpreted as the difference in log odds of survival (survival model), difference in height (height model) or, when exponentiated, the difference in number of leaves (leaf growth model) for a 2-SD increase in a predictor variable. Because covariates were mean-centred, intercept terms represent model expectations when all covariates are held to their average value. Information on the mean, SD and range of continuous covariates is provided in Table 1. We declared fixed effects statistically significant if their 95% confidence interval did not overlap 0. Confidence intervals were generated using parametric bootstrapping (N = 999). To assess the biological significance of significant effects detected in this model, we calculated how model predictions would vary in response to changes in values of predictor variables.
| Survival model | Growth models (height and no. of leaves) | |||||||
|---|---|---|---|---|---|---|---|---|
| Mean | SD | Min. | Max. | Mean | SD | Min. | Max. | |
| Lek vs. non-lek | 0.5 | 0.5 | 0 | 1 | 0.62 | 0.49 | 0 | 1 |
| Related vs. non-related | 0.5 | 0.5 | 0 | 1 | 0.51 | 0.5 | 0 | 1 |
| Elevation (m) | 517.21 | 49.28 | 407 | 640 | 521.79 | 55.99 | 407 | 640 |
| Canopy height (m) | 26.26 | 7.42 | 9 | 43 | 24.85 | 7.02 | 15 | 42 |
| Canopy cover (%) | 93.18 | 3.98 | 78.94 | 98.44 | 91.96 | 4.82 | 78.94 | 98.44 |
| Adult within 5 m | 0.06 | 0.23 | 0 | 1 | 0.06 | 0.24 | 0 | 1 |
| Trees with dbh > 50 cm | 0.3 | 0.46 | 0 | 1 | 0.17 | 0.38 | 0 | 1 |
| Seedling and juveniles within 5 m | 9.07 | 17.12 | 0 | 78 | 8.39 | 16.67 | 0 | 78 |
- Presented are the mean, standard deviation (SD) and minimum and maximum values. Note that the number of plots between the survival (N = 70) and growth models (N = 53) differs because there were 17 plots in which no seedlings survived to the end of the study period.
Results
More than half of all experimental seedlings died during the first 3 years of the study, with one-third (33.5%) of the original 560 individuals surviving to our final (6 year) census point in 2014 (Fig. 2A). Likewise, plant height increased rapidly during the first 3 years, but then levelled off (Fig. 2B); average (± SD) height for seedlings that survived until 2014 was 1.22 ± 0.46 m (range: 0.34–2.42 m). In contrast, the average number of leaves per seedling did not vary strongly during the study; the average number of leaves (± SD) for seedlings that survived until 2014 was 3.26 ± 1.26 (range: 1–6) (Fig. 2C).
Lek vs. control
The number of seedlings and juveniles within 5 m of experimental seedlings was almost twice as high in lek sites compared with control sites (mean ± SD; lek: 11.8 ± 15.0, control: 6.3 ± 18.6); adult density did not vary (Table 2). Despite this dramatic difference in conspecific density, survival, height and number of leaves of experimental seedlings were not significantly different in lek sites compared with control sites (Table 3, Fig. 3). Leks did not differ from control sites in any measured habitat characteristic, including light availability (Table 2).
| Intercept | Est. | SE | 2.5% | 97.5% | |
|---|---|---|---|---|---|
| Elevation (m) | 502.94 | 28.540 | 26.64 | −21.967 | 81.526 |
| Canopy height (m) | 27.318 | −2.118 | 2.794 | −7.434 | 3.026 |
| Canopy cover (%) | 93.884 | −1.402 | 1.571 | −4.512 | 1.835 |
| Adults within 5 m | −4.109 | 1.193 | 1.417 | −60.370 | 35.975 |
| Trees > 50 cm dbh | −0.561 | −0.727 | 0.617 | −2.131 | 0.436 |
| Seedling and juveniles within 5 m | −1.182 | 2.77 | 1.23 | 0.840 | 5.547 |
- Each row represents a separate model, with lek vs. non-lek as a binary predictor and the microhabitat and conspecific neighbourhood characteristic in the first column as the response variable. Seedlings and juveniles within 5 m was modelled using a Poisson error structure. Adults within 5 m and trees > 50 cm dbh was modelled with a binomial error structure. All other characteristics were modelled using a standard linear mixed model with a Gaussian error structure. The intercept represents the estimated value in control sites. Note that estimates for adults within 5 m and trees > 50 cm dbh are on the logit scale, whereas the estimate for seedling and juveniles within 5 m is on the log scale. Est = estimated coefficient, which represents the estimated difference between control sites and lek sites; SE = standard error of the difference between control sites and lek sites; 2.5% is the lower confidence interval estimate and 97.5% is the upper confidence interval estimate calculated from parametric bootstraps (N = 999) of the difference between control sites and lek sites; statistically significant results where the confidence interval does not cross 0 are indicated in bold type.
| Survival | Height | No. of leaves | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Est. | SE | 2.5% | 97.5% | Est. | SE | 2.5% | 97.5% | Est. | SE | 2.5% | 97.5% | |
| Intercept | −0.962 | 0.354 | −1.655 | −0.28 | 1.153 | 0.056 | 1.033 | 1.254 | 1.162 | 0.041 | 1.064 | 1.227 |
| Lek vs. non-lek | 0.634 | 0.706 | −0.745 | 2.154 | −0.13 | 0.106 | −0.358 | 0.075 | −0.116 | 0.099 | −0.315 | 0.08 |
| Related vs. non-related | 0.042 | 0.223 | −0.404 | 0.478 | −0.086 | 0.043 | −0.178 | −0.003 | −0.039 | 0.082 | −0.203 | 0.13 |
| Elevation | −0.003 | 0.68 | −1.401 | 1.399 | 0.188 | 0.122 | −0.058 | 0.449 | −0.1 | 0.101 | −0.307 | 0.092 |
| Canopy height | 0.082 | 0.441 | −0.78 | 1.017 | −0.017 | 0.109 | −0.266 | 0.205 | 0.1 | 0.096 | −0.103 | 0.284 |
| Canopy cover | −1.444 | 0.452 | −2.503 | −0.607 | −0.366 | 0.112 | −0.584 | −0.142 | −0.275 | 0.086 | −0.435 | −0.102 |
| Adult within 5 m | 0.04 | 0.378 | −0.766 | 0.798 | 0.04 | 0.096 | −0.163 | 0.225 | −0.034 | 0.089 | −0.22 | 0.135 |
| Tree > 50 cm dbh | −1.347 | 0.42 | −2.211 | −0.557 | −0.136 | 0.092 | −0.32 | 0.042 | −0.219 | 0.096 | −0.428 | −0.048 |
| Seedling + juvenile density | 0.315 | 0.449 | −0.956 | 1.281 | −0.171 | 0.092 | −0.357 | 0.016 | −0.221 | 0.099 | −0.423 | −0.041 |
- Est = estimated coefficient, SE = standard error, 2.5% is the lower confidence interval estimate and 97.5% is the upper confidence interval estimate calculated from parametric bootstraps (N = 999); statistically significant results where the confidence interval does not cross 0 are indicated in bold type. Covariates were mean-centred and scaled by dividing by 2 SD prior to analysis (see Table 1, for information on mean values and SD of covariates).
Relatedness
Seedlings with related neighbours were significantly smaller than seedlings surrounded by unrelated neighbours (Table 3, Fig. 3). There was no relationship between survival or number of leaves and our measure of relatedness (Table 3, Fig. 3).
Conspecific density
The number of conspecific seedlings and juveniles within 5 m of experimental seedlings was associated with fewer leaves at the final census point, but not with a decrease in survival or height (Table 3, Fig. 4). The number of adults within 5 m was not associated with reduced survival or growth (Table 3, Fig. 3). The relationship between conspecific density and growth in terms of number of leaves was robust to removing outlier plots with > 40 seedlings and juveniles (Supporting Information, Table S1).
Microhabitat
Canopy cover, our primary measure of light availability, was significantly associated with decreased seedling survival, height and number of leaves (Table 3, Fig. 4). Survival and total number of leaves, but not height, were also lower in areas that had higher densities of trees > 50 cm dbh (Table 3, Fig. 3); this pattern was robust to outliers (see Supporting Information, Table S2). Elevation and canopy height did not have a significant effect on survival or growth (Table 3).
Discussion
In this experimental study we assessed the relative importance of microhabitat, conspecific density and relatedness on survival and growth of seedlings of the widespread Neotropical palm O. bataua. As with many studies on palms and other rainforest plants, light levels explained most of the variance in seedling performance: increased survival and growth were associated with decreased canopy cover and fewer large trees. Corroborating correlative results of an earlier study (Karubian et al., 2012), there was no evidence for reduced seedling survival in lek sites of a key dispersal agent, the long-wattled umbrellabird, despite a markedly higher density of conspecific seedlings and juveniles in these sites. In addition to high seed and seedling density, umbrellabird leks are also characterized by high maternal seed and seedling source diversity (Karubian et al., 2010; Scofield et al., 2012). Our experiment demonstrated that although relatedness of neighbouring experimental seedlings had no discernible impact on survival, it did negatively affect plant height. These findings shed light on the relative importance of various processes that underlie seedling survival and performance, provide insights into the degree to which NDD mechanisms may operate away from source trees and underline the importance that behavioural ecology of animal seed dispersal agents can have on plant demography.
There is overwhelming support for NDD mortality being a strong driver of seedling demography near source trees, but less is known about how these processes operate at high-density sites away from source trees (Comita et al., 2014). This represents a considerable knowledge gap because most if not all animal dispersal agents produce non-random deposition patterns characterized by high seed and seedling densities at certain locations away from source trees (Karubian & Durães, 2009; Cortes & Uriarte, 2013). Other studies assessing NDD away from source trees report mixed results: high-density sites away from source trees are associated with relatively high survival in some systems (Munoz Lazo et al., 2011; Sica, Bravo & Giombini, 2014) and relatively low survival in others (Kitamura et al., 2004; Russo & Augspurger, 2004), suggesting that patterns may be context specific. Our earlier work on the O. bataua–umbrellabird study system used a correlational approach to infer that there was no reduction in the probability of transition from seed to seedling in umbrellabird leks, despite higher densities at these sites (Karubian et al., 2012). The current project confirms higher seedling densities in leks and shows that there is no measurable impact on seedling survival or height at these sites. We did observe a significant, negative relationship between density and leaf number, but question the biological significance of this finding for three reasons. First, because leaf number does not increase with age (Fig. 2C), the utility of this trait as an index of plant performance is unclear. Second, the effect was relatively weak: a large increase (33.34) in the number of seedlings and juveniles within 5 m of a plot yields a predicted reduction of 0.80 leaves, or 24% of the average number of leaves per individual (Table 3). Third, because slope is correlated with seedling and juvenile density in this system (see 2), it may be that slope is responsible for the decrease in growth rates, rather than seedling and juvenile density per se; the current study was not able to disentangle these two effects due to correlations between these variables. Given that different selection pressures play out at the seed vs. seedling stage (e.g. Fricke, Tewksbury & Rogers, 2014), we may have obtained different results if we sowed seeds rather than germinated seedlings in this study, suggesting one potential avenue for future research.
Why did we fail to document reduced seedling survival in lek sites, as might be expected given higher densities? We can rule out a beneficial effect of gut passage by umbrellabirds (e.g. Fricke et al., 2013), because seeds used in this study did not pass through umbrellabird guts. An alternative possibility is that microhabitat differences between leks and control areas may somehow compensate for any negative effects of higher densities in leks by providing favourable conditions for seedlings at these sites. The strongest predictor of survival and growth among the variables we measured was access to light. The biological significance of these effects appears to be substantial in this system. For example, increasing canopy cover from 85 to 100% led to a predicted 37.6% decrease in seedling survival, and a 9.64% increase in canopy cover led to a predicted height reduction of 36.6 cm, or 30%, of an average plant. Similarly, addition of at least one large tree within 5 m of seedlings led to a predicted 25.2% decrease in survival; the effect on seedling height was also negative although the relationship was not significant. We note that the effect of canopy coverage on seedling height may not be independent from survival; that is, because taller seedlings are expected to have higher survival rates (e.g. Queenborough et al., 2007), shorter seedlings are probably those that die. Leaf number also responded negatively to increased canopy cover (a 9.65% increase in cover led to a predicted decrease of 0.759 leaves, or 23.2%) and presence of a large tree (predicted decrease of 0.803 leaves, or 24.6%), although the uncertainty around how to interpret this variable is this system (above) draws into question the biological significance of these leaf number findings.
Light as a limiting factor for seedling growth is a common pattern among rainforest plants because the floor of closed-canopy forests often receives only a small fraction (< 2%) of available photosynthetic energy (Augspurger, 1984; Chazdon et al., 1996; Queenborough et al., 2007). As ubiquitous components of tropical forests, it is perhaps not surprising that most rainforest palms appear to follow this pattern (e.g. Svenning, 2002; dos Santos et al., 2012). However, there was no significant difference between leks and control sites in light levels, suggesting that other factors not measured in this study could contribute in substantive ways to survival of seedlings in lek and control sites, despite higher densities in leks. In particular, and as is the case in many other palms (Eiserhardt et al., 2011; Andersen et al., 2014), it may be that soil characteristics play an important role in determining survival and growth. We might expect soil characteristics to vary between leks and control sites because umbrellabird leks are frequently located in a distinctive topographical area (i.e. ridge tops, our unpubl. data ). It is also possible that umbrellabirds enhance soil conditions for seedlings via regular defecation at these traditionally used display sites. A future study measuring soil characteristics (including hydrology) would help to provide resolution among these alternatives.
Alternatively, it may be that lower levels of relatedness among naturally occurring (i.e. non-experimental) seedlings in lek sites might have acted to weaken the impact of NDD processes on experimental seedlings in leks relative to those planted in control sites. Currently, little is known about how degree of intraspecific relatedness impacts growth and survival of tropical plants in natural contexts and what the mechanisms driving these outcomes may be. We hypothesized three ways in which degree of relatedness, either among our experimental seedlings or via effects of non-experimental seedlings on our experimental seedlings, might influence the outcome of our experiment: (1) increased susceptibility to attack among closely related neighbouring individuals might have a negative effect; (2) kin selection among closely related neighbouring individuals might lead to a competitive advantage; and (3) increased competition via greater niche overlap among closely related individuals might have a negative effect. There was no difference in plant survival or growth in leks and outside leks, suggesting that any (unmeasured) differences that may have existed in the relatedness of non-experimental seedlings in these two contexts did not have a detectable impact on experimental seedlings.
We found, however, that relatedness had a weak but statistically significant negative impact on seedling growth (but not survival). Seedlings were on average 0.09 m taller when surrounded by unrelated neighbours vs. related neighbours, which corresponds to a c. 7% difference relative to average plant height. We consider this difference to be biologically significant given the intense competition among seedlings for access to light, such that relatively modest differences in height may therefore influence survival (e.g. Queenborough et al., 2007). These findings provide tentative support for the idea that competition that limits resources among more closely related individuals (in this case experimental seedlings planted in close proximity) may impair performance, corroborating similar results from other recent studies (e.g. Milla et al., 2009; Cheplick & Kane, 2010). However, more work is needed to reach confident conclusions on this question. Future studies might move beyond the relatively crude, but clear-cut, categorical measure of relatedness we employed in the current study (i.e. sibs and half-sibs collected from the same infructenscences vs. individuals collected from different trees) to assess how continuous measures of genetic diversity (e.g. heterozygosity, inbreeding, allelic richness) or genetic uniqueness relative to other conspecifics may influence seed and seedling performance (e.g. Liu et al., 2015).
The current study provides two, distinct pieces of information that are consistent with the idea that lek-breeding male umbrellabirds provide directed dispersal that promotes survival among seedlings of O. bataua. First, we experimentally confirm that O. bataua recruitment in leks is equivalent to control sites, despite higher densities of seeds and seedlings in these sites, suggesting that some factor may enhance seedling survival at these sites (i.e. to counter-balance expected NDD effects). Second, lower relatedness was associated with increased plant height, and umbrellabirds are known to produce highly diverse seed pools in leks, suggesting a potential additional advantage to dispersal by umbrellabirds. Like many large-bodied frugivores, umbrellabirds are at risk of extinction and the consequences of disruption to the dispersal mutualism between umbrellabirds and O. bataua remain unknown. It seems likely that umbrellabird extinction might lead to reduced recruitment and increased genetic relatedness among O. bataua seedlings. Consistent with this prediction, Browne, Ottewell & Karubian (2015) found higher genetic relatedness among O. bataua seedlings in forest fragments where umbrellabirds are locally extinct compared with continuous forest with healthy populations of umbrellabirds, citing the lack of umbrellabirds as a potential cause. Understanding the possible consequences of defaunation for plant population dynamics remains a topic of global interest and importance; linking frugivore behaviour with processes driving plant recruitment will be a critical step in achieving this goal.
Acknowledgements
We gratefully acknowledge the support of FCAT (Fundación para la Conservación de los Andes Tropicales), the Jatun Sacha Foundation, local residents and landowners of the Mache-Chindul reserve and the Ecuadorian Ministry of the Environment. Valuable assistance and support was provided by C. Aulestia, J. Bermingham, F. Castillo, J. B. Cox, A. Ruohomäki and M. Gonzalez. This project was supported by the Conservation, Food & Health Foundation, Disney Conservation Fund, National Science Foundation (EAGER no. 1548548, DDIG no. 1501514 and a Graduate Research Fellowship to L.B.), Tulane University and the United States Fish & Wildlife Service (NMBCA no. 5605). All research was conducted with approval of the Ecuadorian Ministry of the Environment (Permit 010-2015-IC-FLO-FAU- DPE-MA).
