- 1.The Janzen–Connell hypothesis (JC) is one potential mechanism to explain the maintenance of high alpha diversity of tree species in tropical forests, operating through differential pressure by natural enemies.
- 2.We proposed that this differing pressure could arise from the richness of damage types due to natural enemies (RDNE). Following a community compensatory trend (CCT), we hypothesized greater RDNE on common species than on rare species.
- 3.We evaluated this novel interpretation of the JC by assessing damage patterns on leaves, as a proxy for natural enemy species in 44 tree species. We first evaluated which abiotic and biotic factors affect RDNE. Then, we tested whether increasing RDNE leads to an increasing amount of foliar damage.
- 4.We found that RDNE
- Was affected by biotic environments: RDNE increased with mean seedling species abundance. RDNE was higher on species occurring near more closely related neighbours.
- Was not impacted by abiotic factors. Yet, seedlings of shade-tolerant species hosted a higher RDNE than seedlings of shade-intolerant species.
- Was positively correlated with amount of foliar damage at the species level.
- 5.Finally, we tested whether RDNE increased seedling mortality risk. We found that
- Foliar damage, species abundance and RDNE2 increased mortality risk.
- Richness of damage types due to natural enemies linearly decreased mortality risk more strongly than RDNE2 increased it.
- Seedling age decreased mortality risk.
- 6.Synthesis. The richness of damage types due to natural enemies increased with abundance of the host species, suggesting an important role of enemy diversity in the maintenance of tree diversity. Supporting a novel interpretation of the Janzen–Connell hypothesis, we found a greater mortality risk with increasing RDNE2, but not with increasing RDNE. There was a stronger negative linear effect of RDNE on mortality risk. Rare species with low RDNE as well as species with very high RDNE suffered greater mortality than species hosting intermediate RDNE, reinforcing the complexity of the effect of multiple enemies on prey.
A central question in forest community ecology is: How can hundreds of different species of trees coexist in a single hectare? A multitude of hypotheses have been proposed to explain species coexistence, but the question remains unresolved (Chesson 2000). Among the most prominent of explanations of tree species coexistence is the Janzen–Connell hypothesis (Janzen 1970; Connell 1971) (JC), which proposes distance- and density-dependent natural enemies have strong host specializations that favours heterospecific tree replacement. An extension of JC is that natural enemies disproportionately attack seeds and seedlings of common species [i.e. a community compensatory trend or CCT per Connell, Tracey & Webb (1984)], offering a stabilizing mechanism for species coexistence (Chesson 2000). We propose a novel interpretation of the JC and CCT that does not depend on host specificity of the natural enemies; the advantage of being rare could arise from lower richness of damage types due to natural enemies (RDNE) compared with the RDNE on common species. This rare species advantage arises because higher RDNE results in a higher mortality risk for seedlings.
Numerous studies have documented patterns of distance- and density-dependent seedling mortality (Clark & Clark 1984; Kobe & Vriesendorp 2011), but few have identified particular agents causing such patterns (Gilbert & Webb 2007; McCarthy-Neumann & Kobe 2008). The few studies that have explicitly incorporated natural enemies have focused on the amount of damage, which generally increases with increasing conspecific density (Augspurger & Kelly 1984; Schuldt et al. 2010), but relationships with distance are less well resolved (Augspurger 1984; Norghauer et al. 2010). Most studies testing the JC have focused on mortality pattern (see review by Clark & Clark 1984), with some studies supporting negative density-dependent mortality (NDD) at the individual level (Augspurger 1983; Clark & Clark 1984; but see Silander 1978; Boucher 1981). Similarly, at the species level, some studies are consistent with a CCT (Connell, Tracey & Webb 1984; Webb & Peart 1999; within shade-tolerant species Comita & Hubbell 2009), while others show the opposite pattern (Klironomos 2002; Comita et al. 2010; Mangan et al. 2010). The inconsistency in results concerning CCT may originate from complex below-ground and above-ground NDD interactions with each other and with the plants (Wardle & Bardgett 2010). In this study, we focus on above-ground effects via foliar natural enemies.
Strong host preference has been proposed as a necessary condition for natural enemies to constrain the populations of common species (Janzen 1970). However, natural enemies in the tropics target a wide range of hosts (Zhou & Hyde 2001; Novotny et al. 2010): a tropical rain forest tree species can host more than 200 herbivore species, among which approximately only 25% are host specific (Novotny et al. 2010). Low specialization rates lead some studies to conclude that natural enemies could not be responsible for density-dependent mortality (Hamilton, Basset & Tanese 2010). However, specialization is not a necessary assumption of a CCT if more common species accumulate a greater diversity of natural enemies, which in turn increases seedling mortality risk.
Closely related species could provide equally suitable hosts for specialized natural enemies and thus could share natural enemies (Gilbert & Webb 2007). Coley (1983) is frequently cited as showing evidence that tree life-history traits influence susceptibility to natural enemies (also see McCarthy-Neumann & Kobe 2008; Kobe & Vriesendorp 2011), but this study also suggests the importance of phylogenetic distance on survival. Similarly, other studies have confirmed the importance of looking beyond conspecific and heterospecific species in understanding negative density dependence (Strauss, Webb & Salamin 2006). Phylogeny may be successful at predicting functional traits that are involved in plant–enemy interactions, but abiotic conditions can heavily influence those traits, which in turn will influence natural enemies (Fine, Mesones & Coley 2004). Thus, abiotic factors have to be considered when looking at plant natural enemies.
Little research, however, has been conducted in the tropics to understand the effects of abiotic environmental conditions on natural enemies even though the importance of the disease triangle [Stevens (1960): a suitable host, environmental conditions, and presence of the natural enemy] is well established in temperate systems. Some studies suggest that damage decreases at high irradiance levels (Augspurger 1984), while other studies found that an increase in irradiance can increase damage by natural enemies (Alvarez-Loayza et al. 2008). The effect of abiotic factors may vary with traits of the host species; for example, drought may increase vulnerability to natural enemies in drought-intolerant, but not drought-tolerant species (Fine, Mesones & Coley 2004). Similarly, fast-growing pioneer species are thought to invest significantly in growth and little in defences, while slow-growing shade-tolerant species are thought to invest more in defences (Coley, Bryant & Chapin 1985). Therefore, it is important to consider natural enemies in the context of abiotic and biotic conditions.
In this study, we rely on 10 years of demographic data to test the hypothesis that common seedling species host a higher RDNE relative to rare species, which in turn may lead to a demographic disadvantage. Because of the large number of potential natural enemies, foliar damage patterns were used as a proxy for natural enemy species (Carvalho et al. 2011). Specifically, we ask the following questions:
- How do biotic factors, namely seedling abundance, distance to large conspecific trees and neighbourhood phylogenetic relatedness, influence RDNE in seedlings?
- We hypothesize that a higher RDNE, (i) is hosted by common species, (ii) occurs when a species has lower average distances to conspecific trees and (iii) occurs when a species on average has lower phylogenetic distance with neighbouring individuals;
- How do abiotic factors (namely light availability, soil fertility and moisture conditions) together with species life histories influence damage patterns?
- We hypothesize a higher RDNE with (i) decreasing species shade tolerance and (ii) with increasing average soil fertility and moisture conditions experienced by the species;
- How does RDNE influence foliar damage at the species level?
- (f) We hypothesize that a higher RDNE would lead to an increase in foliar damage.
- How does RDNE affect seedling mortality risk at the individual level?
- We hypothesize that (g) higher foliar damage, RDNE and species abundance would increase seedling mortality risk and that (h) mortality risk would decrease with seedling age.
Materials and methods
This project was conducted at La Selva Biological Station in the Atlantic lowlands of Costa Rica (10º26′ N, 84º00′ W). The core data set consists of five 1 × 200 m belt transects, which were established across a soil gradient representative of major soil types: three transects on residual soils and one each on older and recent alluvial soils. All newly germinating and surviving woody seedlings have been surveyed in 1000 1m2 plots systematically every 6 weeks since 2000 (Kobe & Vriesendorp 2011). Liana and small palm seedlings were excluded here because trees were not mapped. All trees (> 5 cm diameter) within 20 m of the seedling quadrats were mapped and identified.
From 13 to 22 January 2010, we took photos of every leaf on all seedlings within each of the five transects (a total of 869 seedlings of 113 species). A scaled, waterproof white sheet of paper was used as background. Based on the photographs, damage patterns were visually assessed for each seedling and differentiated using the main following criteria: (i) position of the damage (e.g. edge or middle of leaf, proximity to principal vein), (ii) shape of the damage (rounded, linear cut, irregular), (iii) size of the damage (< 1 mm, < 1 cm, > 1 cm), (iv) colour (especially relevant to disease and pathogens), and (v) other defining characteristics (cut through veins, penetration through leaf or superficial grazing) (see Key to damages and Fig. S1).
The use of damage-type diversity as a proxy for enemy diversity is common in palaeoecology (Wilf et al. 2005; Currano et al. 2008) and has been demonstrated to offer a ‘robust relationship’ between damage type and enemy diversity (Carvalho et al. 2011). This proxy works better with specialist enemies (Bernays & Chapman 1994); it may underestimate true enemy diversity (Dr. Robert Marquis, personal communication) and provides a conservative estimate of the diversity of ecological associations (Wilf & Labandeira 1999; Labandeira, Johnson & Wilf 2001; Wilf et al. 2005). We separately analysed damage types for each plant species and did not make any inferences regarding total diversity of enemies in the community.
Categorical Foliar Damage
During each census from May 2008 to December 2010, damage on each seedling was evaluated and reported as a categorical variable representing the percentage of leaf damage (0–10% of the leaf is damaged, 11–20%, …, to 100% where the entire leaf was damaged). For each seedling that was observed between May 2008 and December 2010 (n = 1911, 213 species), we calculated the average amount of damage per day observed between May 2008 and December 2010.
To investigate whether biotic factors influence the number of damage patterns, a proxy for natural enemy richness, we measured a variety of biotic factors (see Table S1). Seedling abundance for a species was assessed as the mean density of conspecific seedlings per m2 during the sampling period. To examine whether past seedling species composition influenced natural enemy richness, we calculated and tested whether seedling abundances over the past cumulative 1, 2, 3, … 9, 10 years were correlated with RDNE.
Abundance of tree species was assessed as the average number of conspecific trees within a radius of 20 m around each seedling of the species, a typical neighbourhood distance as estimated by Kobe & Vriesendorp (2011). The sum of tree basal area was used as a second proxy for species abundance at the tree stage. Distance from parent tree was assessed as the average distance from the nearest conspecific tree.
Using seedling mortality models [Kobe & Vriesendorp (2011) and unpublished data], we calculated mortality under 1% irradiance to represent species shade tolerance. Previous studies stressed the importance of quantifying the ages of leaves when assessing damage because of an accumulation effect (Lowman 1984). To take this into account, we used the life span of each individual seedling (in days) to calculate the average life span of the seedling of each species, recognizing that seedling life span is a coarse approximation of leaf life span.
Using a phylogenetic tree constructed in Phylocom (Webb, Ackerly & Kembel 2008), we calculated a matrix of phylogenetic distances between seedlings of all species present in the transects using the package Picante for R. We used an angiosperm base tree available with the newest version of Phylocom that gives ages for the nodes of the tree according to Wikstrom, Savolainen & Chase (2001). Subsequently, we calculated the average neighbourhood phylogenetic distance for each species known as the mean pair-wise phylogenetic distance method; for each individual i of a given studied species s, we averaged the phylogenetic distances between the individual and all the seedlings j present in the same one m2 quadrat. Then, we averaged the distances among individuals of the species s.
One hundred and fifty-three individuals from 50 different morphospecies, representing 12.5% of the total number of seedlings in the transects, were removed from the analysis because they were identified to the genus level solely. As for seedling abundance, we calculated the neighbour phylogenetic indices over different time periods: from the current time of the census to the past 10 years.
To estimate effects of abiotic factors on RDNE, we measured diverse environmental variables (see Table S1). We used previous measurements of canopy openness and soil resource availability (N, C, P and base cations) (Holste, Kobe & Vriesendorp 2011) to calculate the mean resource conditions experienced by each species. Soil moisture was measured every 10 m and was linearly interpolated to estimate the soil moisture for each seedling plot. Finally, we measured atmospheric humidity at 1 m height along the transects 5 times during April 2010.
We used R-project computing software (http://www.r-project.org) to perform all data analyses. To allow comparison of diversity estimates drawn from different sample sizes, for each species we developed rarefaction curves by calculating the cumulative number of different damage patterns when increasing the number of seedlings sampled using all possible combinations of sampling order.
We then fit a Michaelis–Menten function to the cumulative damage pattern for each species similarly to the extrapolation method to estimate total species richness used by EstimateS (Colwell & Coddington 1994):
where Ndi,si is the cumulative number of damage patterns found on seedling species i for si seedlings; Ndmax,i is the asymptote of the curve and represents the expected total number of natural enemies hosted by tree species i; Ki is a constant, which corresponds to the number of seedlings sampled to get to half the number of natural enemies hosted by the tree species (half of Ndmax,i) (see Fig. S2). Models were fit for each species through minimization of sum of squared errors.
To answer our two-first questions and test the factors that influence RDNE across host species (the ‘community model’), we tested for relationships between Ndmax,i and potential factors (see Table S1) using linear and nonlinear regressions. This analysis was realized for a subset of 30 species because of missing values for some of the factors. The best models were selected per Akaike information criterion (AIC). To estimate parameters, the model was then fit to all the species with values for factors selected by AIC in the first round of modelling. The set of best models was evaluated by AIC and r2. The robustness of the best community model was assessed using a leave-one-out cross-validation; the model was fitted using N-1 species, and the Nth species is then predicted. This was repeated 29 times in order to get an estimate of RDNE for each species. To examine within-species variation in natural enemy species load, we used a similar regression approach for all species with >20 seedlings. The number of natural enemy species found on individual seedlings was modelled as a linear regression using the same abiotic and biotic factors as the community-level model, but measured at the individual level rather than mean values for a species.
To answer our third question and test whether a higher RDNE leads to a disadvantage, we first performed a linear regression between RDNE and amount of leaf damage experienced in 2008 at the species level. To validate the categorical damage observations, pictures of each leaf were taken in January 2010 to calculate damaged leaf area using WinFolia. Photo estimates of leaf damage were then compared with the categorical observation (r2 = 0.7).
Finally, we performed a mortality analysis to test whether seedling mortality risk increases with amount of damage, seedling age, seedling species abundance and RDNE at the individual level. We analysed mortality using a hierarchical Bayesian approach (eqn 3):
where damagei is the categorical damage assessment on individual seedlings from 2008 through 2010, Statusis represents the status of individual i of species s the last time the seedling was censused (dead = 1, live = 0), agei represents the age of the seedling i at its last recorded status, abundances is the seedling species abundance (calculated as the average species abundance over the last 10 years) and RDNEs is the richness of damage due to natural enemies for the species s. We used RDNE2 in order to account for the fact that even a low diversity of efficient enemies could result in similar seedling mortality as a high diversity of less capable enemies. For each individual, we calculated the amount of damage per day across the damage census period of the individual using the categorical damage data set. Finally, μs represents the species-specific mean mortality risk that arises from a normal distribution centred on mu with a precision τ. This model was fitted using JAGS (Plummer 2005) statistical software to run three Markov chain Monte Carlo (MCMC). Convergence was assessed using R-hat (Brooks & Gelman 1997). The significance of the parameters was evaluated using the credible intervals. Finally, in order to assess our mortality model, we used 4/5 of the data to predict the remaining 1/5.
Ndmax within Tree Species
For each of the 44 species with ≥ 3 seedlings (see Table S2 and Fig. S3), the cumulative number of natural enemy species increased less than proportionately with increasing numbers of seedlings sampled. No tree species reached asymptotic levels of RDNE. Nevertheless, the Michaelis–Menten function provided good fits to the data and provided estimates of both the maximum number of natural enemies (Ndmax) and the number of seedlings for half Ndmax (k) (Gotelli & Colwell 2001). The average fit of the model, using the square of the Pearson correlation, was r2 = 0.63 (range: 0.333–0.94, > 0.50 for 34/44 species).
We also tested for correlations between RDNE and biotic and abiotic factors at the individual seedling level for the five species with adequate sample sizes (N > 20). Among the 5 × 13 linear regressions, only four were significant, which is the experiment-wise expectation for significant results by random chance when the significance level is 0.05 for individual comparisons (see Table S3). Thus, we do not further discuss within-species results.
Across Host Species, Ndmax was Affected Solely by Biotic Factors and Light
We tested linear and nonlinear effects on RDNE by past host species seedling abundance and phylogenetic neighbour index (see Table S5, S6 and Fig. S4), and other abiotic and biotic factors. We performed the analysis with and without Pentaclethra macroloba, which had a very high abundance (the most common tree species at La Selva), but a low foliar RDNE. Removing P. macroloba did not impact on the significance of the relations, yet it decreases the fit. Leverage analysis indicated that P. macroloba was an outlier; therefore, we excluded it for the rest of the analysis. Excluding P. macroloba, we worked with 32 species (see Fig. 1 and Table S4). Under the best supported model, RDNE increased linearly with conspecific seedling abundance over the past 5 years (AIC = 245, Table 1) and survivorship under 1% full sun (i.e. shade tolerance), and RDNE increased as seedlings of a species tended to be near closely related individuals (as manifested by a negative relationship between RDNE and current neighbourhood phylogenetic index) (Fig. 2 and Table 1).
|Model||Formula||Estimates (SD)||AIC||r2 (%)|
|Power||a*Abundance^b||–||38.02 (2.43)||0.21 (0.06)||–||258||35|
|Linear||i + a*Abundance||27.96 (2.55)||6.16 (1.31)||–||–||249||42|
|Power||i + a*Abundance^b||22.80 (4.99)||13.14 *5.87)||0.61 (0.22)||–||249||47|
|Linear||i + a*Abundance + b*Shade_Tolerance + c*Phylogenetic_Index||47.49 (8.07)||5.75 (1.26)||−79.01 (28.79)||−0.09 (0.05)||246||55|
|Power||i*Abundance^a + b*Shade_Tolerance + c*Phylogenetic_Index||29.27 (5.82)||0.33 (0.10)||ns||0.09 (0.04)||257||39|
|Michaelis||(i*Abudance)/(a+Abundance) + b*Shade_Tolerance+ c*Phylogenetic_Index||ns|
Richness of damage types due to natural enemies was less sensitive to the neighbourhood phylogenetic index, a relationship that was marginally significant (−0.09, P = 0.05). Sum of tree basal area, a proxy for tree abundance at the community level, was significantly positively correlated with RDNE, but was no longer significant when other factors were added to the model. None of the interactions and the other abiotic and biotic factors were significantly associated with the load of natural enemies hosted by different species. Seedling abundance explained the greatest portion of the variability in Ndmax (0.417), followed by phylogenetic index (0.0649) and then shade tolerance (0.0581).
The data also supported models with nonlinear increases in Ndmax with abundance (AIC = 249, r2 = 0.042–0.047, Table 1). As expected, host abundance positively affected RDNE. However, the nonlinear model was not supported once shade tolerance and the phylogenetic index were included.
The community model was simple; it made use of solely three factors and aggregated individual seedling variation to the species level. Yet, the models explained 55% of the variance in Ndmax across species. The model was also robust with respect to the leave-one-out cross-validation method; leave-one-out estimates were correlated with observed RDNE (linear model: r2 = 0.39) (see Fig. S5).
Sample Size Effect
The relationship between RDNE and species abundance could arise as an artefact of RDNE increasing with seedling sample size because we sampled more individuals of common species than rare species and sample size and long-term species abundance were positively correlated (r2 = 0.45). In order to verify that our estimates of RDNE were not artefacts of sample size, we randomly subsampled with replacement three individuals (the lowest sample size) for each species in 1000 simulations. We then fit the Michaelis–Menten function to each species (consisting of three seedlings each) for each of the 1000 simulated sub-data sets to obtain 1000 sets of estimated RDNE for the focal species. This method allowed us to calculate the 95% CI for the RDNE while correcting for sample size. For each of the 1000 simulated data sets, we fit the same linear community-level model of RDNE vs. explanatory factors. RDNE significantly increases with species abundance over the past 5 years and shade tolerance for each of the 1000 simulations. However, the phylogenetic index significantly decreased with increasing Ndmax for only 29% of the simulations, and it was marginally significant for 71% of the simulations.
Overall, the significant relationship between Ndmax vs. species abundance over the last 5 years and shade tolerance was robust to small sample sizes. We conclude that the consistency of our results to 1000 bootstraps support the validity of using extrapolated estimates of RDNE to compare variation across species.
RDNE, Foliar Damage and Seedling Mortality
At the species level, higher RDNE was positively correlated with the amount of damage experienced on average by a species in 2008 (0.0184 ± 0.00226, P < 0.001 and Fig. 3). The linear model successfully predicted the average amount of damage experienced by species in 2009 (r2 = 0.64, P < 0.001). At the individual level, probability of mortality increased with amount of damage per day, decreasing seedling age across species and increasing species abundance (Fig. 4 and Table 2). RDNE was also a significant factor to explain seedling mortality with the positive relationship between RDNE squared (RDNE2) and mortality, but a negative relationship between RDNE and mortality (Fig. 5 and Table 2). The survival model successfully explained seedling survival; when confronting the model to a cross-validation analysis, the model correctly predicted 60% of seedling survival.
Here, we extend the Janzen–Connell hypothesis to RDNE. Foliar RDNE increased with species abundance. We also found that species whose seedlings tended to occur near closely related seedlings had the highest RDNE, consistent with our second hypothesis. However, contrary to expectation, shade-tolerant species hosted more natural enemy species than shade-intolerant species. At the species level, RDNE was positively correlated with amount of foliar damage, which increases seedling mortality risk. At the individual level, seedling mortality was increased with increasing species abundance, foliar damage and squared RDNE. These results support that higher RDNE leads to a disadvantage for very common species.
Maximum Number of Natural Enemies Species was Influenced by Biotic Factors
Consistent with the first hypothesis, species that were more common at the seedling stage for the past 5 years hosted a higher RDNE. This result is consistent with a CCT, which predicts natural enemies should preferentially target seedlings of common tree species (Connell, Tracey & Webb 1984). The strong difference in the number of natural enemy damage types hosted by common and rare species, and the effect of damage on mortality (Eichhorn et al. 2010) reinforces the potentially important role of natural enemies in creating and maintaining diversity (Janzen 1970; Fine, Mesones & Coley 2004). However, P. macroloba follows an opposite trend with a high abundance but a low RDNE that is consistent with results from Comita et al. (2010) and Mangan et al. (2010). The divergence of P. macroloba from the general species relationship between natural enemy richness and species abundance could explain its dominance (Hartshorn 1983) at La Selva. Its dominance has been interpreted to result from its high tolerance of infertile soil and a weak dry season (Hartshorn 1972), but our results suggest that its low number of foliar natural enemy species also could contribute to its high abundance. Leaves of P. macroloba seedlings are composite, with a high edge area ratio. Edges are less nutritious (Scriber & Slansky 1981) and are often avoided by herbivores. The leaves also close at night, offering fewer surfaces for herbivores, which are especially active at night (Elton 1973). The form and diurnal changes in P. macroloba leaves could explain its low RDNE.
Contemporary local seedling density (observed in January) failed to predict the maximum number of natural enemy damages. In contrast, the cumulative abundances of the host species over the past 1–10 years were significant predictors of Ndmax with host abundance over the past 5 years as the best predictor. A lag in RDNE with species abundance suggests a longer-term assemblage of the community may be driving RDNE. Although we cannot identify why a lag between abundance and RDNE occurs, the lag does provide an opportunity for seedlings to become established after a mast event, ensuring the continued presence of the species in the community (Janzen 1974; Silvertown 1980).
According to JC, large conspecific trees should act as a source for natural enemies. Thus, we were expecting to find a higher RDNE when the number of large conspecific trees and the average distance from the nearest conspecific tree were higher (Augspurger 1984). Unexpectedly, neither were important factors in predicting RDNE (contrary to hypothesis 2). The lack of an effect was unlikely due to the lack of variation in mean distances among species, as distance ranged from 3 to more than 40 m. Wide intraspecific variation in distances to conspecific trees may have obscured species differences in distance effects. Thus, despite the potential importance of distance from conspecific trees at the individual level (Janzen 1980), this factor may be less informative at the species level, due to high intraspecific variation.
Ontogenetic shifts in leaf traits (Ishida, Yazaki & Hoe 2005) between seedlings and trees also could lead to different ensembles of foliar enemies and a decoupling of tree presence/distance with seedling dynamics. In general, seedling leaves are thinner, have lower N and photosynthetic rates and are less tough than tree leaves (Ishida, Yazaki & Hoe 2005). Since leaf traits between seedlings and trees are different, foliar enemies may also be different, consistent with the finding that both seedling and tree density can influence negative density dependence in seedlings (Kobe & Vriesendorp 2011).
Consistent with our third hypothesis, species that tended to occur with seedlings of more closely related species had higher RDNE. This finding is consistent with recent studies that proposed to expand the Janzen–Connell hypothesis beyond the notion of conspecific and heterospecific (Webb, Gilbert & Donoghue 2006; Gilbert & Webb 2007).
Maximum Number of Natural Enemies Species was Inversely Related to Shade Tolerance
Shade tolerance was negatively, weakly correlated with RDNE (r2 = 0.0581), contrary to expectations based on the correlation between defences against enemies and shade tolerance (Augspurger & Kelly 1984; McCarthy-Neumann & Kobe 2008; Kobe & Vriesendorp 2011). Shade-tolerant species have tougher leaves (high leaf mass per area, LMA) and low nutrient content, making them less palatable for natural enemies (Coley, Bryant & Chapin 1985; Kitajima & Poorter 2010). One potential explanation of this counter-intuitive result is that shade-tolerant species may be better able to tolerate damage. Shade-intolerant species, in contrast, may drop their leaves soon after experiencing damage. Nevertheless, other studies have shown that damage increases with shade (Augspurger 1984). If natural enemies prefer shaded environments, they may preferentially feed on the plant species that are more available in the shade. Shade-tolerant species are also characterized by a long leaf life span (Kitajima & Poorter 2010), allowing more time for foliar natural enemies to accumulate. However, damage occurs principally as leaves are expanding (Lowman 1985), making an effect of leaf life span less likely.
No Effects of Soil Resources on RDNE
We were expecting to find a higher RDNE when on average a species occurs at higher total N content in the soil, since soil N often results in higher foliar N, which is preferred by herbivores (Mattson 1980). However, we found no effects of soil resources. The total N content in soil may not be a good predictor of the N content in leaves. Moreover, the type of N compound influences the behaviour of herbivores; insoluble N compounds are not nutritious for insects (Scriber & Slansky 1981). Finally, N availability influences the level of plant defences (Coley, Bryant & Chapin 1985). Therefore, if a species experiences different N availability, this may create different levels of defences that on average mask the effect of N on RDNE.
Seedling Mortality Risk Increases with Foliar Damage and RDNE Squared
As hypothesized, we found that common species suffer greater mortality than rare species consistent with a community compensatory trend. We hypothesized damage from natural enemy to be responsible for such a pattern. Indeed, we found that at the species level high RDNE leads to a greater amount of leaf damage. RDNE may not necessarily correlate with the amount of damage, since a single enemy species can be extremely efficient at targeting its plant host (Bottrell, Barbosa & Gould 1997).
Nevertheless, at the individual level a higher amount of damage also led to a higher risk of mortality (consistent with Eichhorn et al. 2010). Yet, RDNE did not show a positive linear effect on mortality, but rather it shows a parabolic relationship with seedling mortality suggesting that both species with a low and high RDNE are at high mortality risk. The parabolic relationship can explain both why common species with high RDNE do not become over-dominant and why rare species with low RDNE remain rare. From an evolutionary perspective, it has been suggested that an increase in number of enemies could be associated with both an increase and a decrease in defence (Poitrineau, Brown & Hochberg 2003). From an ecological point of view, one could hypothesize that the addition of enemies might decrease their efficiency through interspecific competition or negative effects on plant defences until reaching a threshold after which the extra addition of enemies might enhance enemy efficiency (Sih, Englund & Wooster 1998). Therefore, complex interactions between enemy efficiency or plant defences and enemy diversity might explain our finding. The linear negative effect of RDNE on mortality risk was stronger than the positive effect of squared RDNE, suggesting that despite a lower diversity of damage, rare species might experience greater mortality risk than common species. This is consistent with a recent study that found stronger negative conspecific effect in rare species compared with common species (Comita et al. 2010). Our study did not take into account below-ground community that hosts additional enemies as well as mycorrhizal fungi. Mycorrhizal fungi can offer protection against natural enemies (Newsham, Fitter & Watkinson 1995); thus, it can complexify the relation between enemy efficiency/plant defences and enemy diversity. Seedling age also was negatively correlated with mortality risk consistent with the theory that seedlings may become more and more tolerant to damage (Del-Val & Crawley 2005). There may not always be a relationship between damage and survival because of tolerance to damage (Beck 1965; Gehring, Cobb & Whitham 1997; Parker & Gilbert 2007). However, our results support the hypothesis that natural enemies preferentially target common species and disadvantage them by increasing mortality risk that is consistent with a CCT (significant positive effects of RDNE2 and abundance factors). Species abundance was a significant predictor of seedling mortality when RDNE and foliar damage were also taken into account, suggesting that foliar natural enemies and foliar damage are not the only agents responsible for a community compensatory trend.
We acknowledge that using damage-type diversity as a proxy for natural enemy diversity is not perfect and underestimates the ‘true’ diversity of enemies. Nevertheless, this approach can reveal important differences between host species in the diversity of ecological associations (e.g. Carvalho et al. 2011).
Variation in seedling and leaf ages may introduce variability in the species-specific relationships between RDNE and sampled seedlings. Seedlings varied in age from 40 to 3579 days. Older leaves may accumulate more damage. However, since 80% of the damage accumulates during leaf expansion (Coley 1983), the artefacts of leaf age should be minimal. We did not explicitly include leaf traits in the community model of RDNE. Inclusion of shade tolerance in our analysis may have accounted for some variation in leaf traits due to covariance in shade tolerance and leaf traits (e.g. Kitajima & Poorter 2010). Leaf mass/area (Currano et al. 2008; Kitajima & Poorter 2010), trichomes (Levin 1973) and tannins and other leaf compounds (Coley 1983) all guard against herbivory.
These findings add a new facet to the Janzen–Connell hypothesis by demonstrating that common tropical seedlings host a greater number of natural enemy species than rare seedlings, which in turn leads to a higher amount of foliar damage and a greater mortality risk. However, the relationship between diversity of damages and mortality risk is parabolic with a strong linear negative effect, suggesting that plants hosting a high RDNE might have a lower mortality risk than plants hosting a low RDNE. Future research should incorporate below-ground community as well as enemy efficiency and plant defences data to understand further the complex effect of RDNE on plant mortality. The unforeseen finding that shade tolerance increased with the number of natural enemies hosted raises several new questions. For example, could this pattern be a consequence of a longer leaf life span for shade-tolerant species? Or co-occurrence of both natural enemies and shade-tolerant seedlings in the shade? Moreover, these findings reinforce the importance of the diversity of natural enemies in regulating tropical tree seedling populations.
We are grateful to NSF (DEB 0075472, 0640904, 0743609) for funding a large part of this project. We also thank the Organization for Tropical Studies (OTS), which lent further financial support through a post-course grant.
We would like to thank Ademar Hurtado Flores and Ralph Garcia Robleto who are responsible for the long-term census being conducted at La Selva.
This paper would not have been possible without Patricia Alvarez-Loayza during the conception of the idea and without Corine Vriesendrop who provided long-term seedling abundance data. Furthermore, the paper would not be what it is without the useful comments of Rachael Eaton, Maria Uriarte, Timothy Paine, Danae Rozendaal, Andrea Maguire, Ellen Holste, Kristen Nolting and David Minor. We are grateful for the useful help received from Andrew Finley and Nathan Swenson. We thank Jenny Borhman whose talented English expression helped us improve this article. Finally, we thank Susan Letcher for her invaluable assistance with phylogenetic classifications.