Distance-responsive natural enemies strongly influence seedling establishment patterns of multiple species in an Amazonian rain forest


Correspondence author. E-mail: vs12@duke.edu


1. In a faunally intact lowland Amazonian rain forest stand, we conducted a long-term multi-species experiment aimed at determining the primary mechanistic basis of seedling establishment patterns. We deployed a total of 1050 experimental seedlings, representing 11 common tree species in mixed compositions and at fixed, highly elevated densities in shaded understorey sites, representing extremes of distance from large conspecific trees. We used mesh exclosures to isolate the effects of distinct classes of natural enemies, and monitored survival for up to 45 months.

2. Final seedling survival of all species pooled represented a 40% increase at sites located far from (‘F’ sites) versus close to (‘N’ sites) large conspecific trees, and median seedling lifetime was 75% longer. These differences between N and F sites were significant for all species pooled, and for five out of 11 (survivorship) and four out of nine (lifetime) individual species examined. Survival analysis based on multiple censuses revealed that a ‘distance effect’ persisted and intensified over time, with the onset of significant distance-related differential mortality differing amongst species.

3. The use of mesh exclosures (<2 mm mesh size) and a factorial experimental design revealed that host-specific organisms <2 mm in size and/or below-ground soil-borne organisms are more strongly distance-responsive and depress seedling establishment primarily in the vicinity of large conspecific adults whereas above-ground organisms >2 mm in size appear to have a negative impact on seedling establishment at all distances.

4. No evidence was found for the effect of intra-cohort resource competition on seedling establishment even though initial density of experimental seedlings at all sites was elevated to c. 25 times the mean natural density of the mixed-species seedling layer in this forest.

5.Synthesis. Our study provides strong, multi-species support for the influence of host-specific distance-responsive natural enemies on seedling establishment, and suggests that negative density-dependent patterns of tree recruitment in tropical rain forests are at least partly produced at early life stages as an outcome of processes described by the classic Janzen–Connell model.


In the 40 years since it was originally proposed, the Janzen–Connell (J–C) model for the spatial organization and maintenance of tree diversity in tropical forests (Janzen 1970; Connell 1971) has greatly stimulated the field of community ecology and generated an impressive volume and variety of research efforts in tropical and temperate ecosystems (reviewed by Clark & Clark 1984; Hammond & Brown 1998; Carson et al. 2008). However, most empirical tests of the J–C model to date contain serious limitations or design flaws that have generated a mixed bag of results and contributed in great part to a current lack of unequivocal acceptance or rejection of the model. A common element of the vast majority of these studies has been a single species, single life stage, focal tree approach, aimed primarily at identifying density-dependent and/or distance-dependent patterns of seed predation (e.g. Janzen 1972; Wright 1983; Burkey 1994, etc.) or post-germination seedling survival (e.g. Augspurger 1983; Howe, Schupp & Westley 1985; Howe 1990; Cintra & Horna 1997; Zuidema & Boot 2002; etc.) in the focal species. Very few of these studies have followed seed or seedling survival for longer than a few months (e.g. Clark & Clark 1984, 1989) and simultaneous multi-species evaluations of interspecific differences in patterns of survival and causes of mortality at a single study site are scant (e.g. Augspurger 1984; Curio et al. 2003; Queenborough et al. 2007). Apart from a small minority of mostly single-species studies (e.g. Augspurger & Kelly 1984; Gilbert, Foster & Hubbell 1994; Packer & Clay 2000; Bagchi, Press & Scholes 2009; Ruibo et al. 2009) the overwhelming majority of previous tests of the J–C model have either not attempted, or have failed to explicitly distinguish between effects of density-versus distance-responsive, and generalist versus host-specific, natural enemies.

Over the past two decades, a number of studies have provided strong evidence for the role of density-dependent processes in structuring tropical rain forest tree communities and maintaining species diversity (Condit, Hubbell & Foster 1994; Wills et al. 1997; Webb & Peart 1999; Harms et al. 2000; Peters 2003; Comita & Hubbell 2009). Although the above-mentioned studies have identified patterns of survival, growth or recruitment that indicate processes regulated by negative density dependence (NDD), they have failed to identify or clearly distinguish between the potential underlying causes of these processes, which can be attributed to two very distinct mechanisms. The classic definition of NDD originated in plant population biology research (Yoda et al. 1963; Antonovics & Levin 1980; Watkinson 1985) and refers strictly to syntrophic (i.e. within a trophic level) competitive interactions between conspecific and/or heterospecific members of a cohort, with typically skewed and negative effects on growth and survival as a result of competition for scarce resources, such as light and nutrients. Density dependence as defined in the J–C model on the other hand, refers explicitly to inter-trophic (i.e. between trophic levels) interactions involving host-specific natural enemies that cause disproportionate levels of mortality of conspecific (compared to heterospecific) propagules located in the vicinity of large reproductive trees. While some of these natural enemies may be distance-responsive and others density-responsive (or even both) in their manner of predation, their influence is inescapably spatially restricted to and focused around the vicinity of large reproductive individuals (Schupp 1992). Thus, even though inter-trophic interactions between plants and their natural enemies, as described by the J–C model and resultant patterns of survival, growth or recruitment, can be considered a form of NDD, they are fundamentally mechanistically different from patterns of NDD produced by syntrophic competitive interactions.

Paine et al. (2008) demonstrated that intra-specific competition for abiotic resources is very weak even within extremely dense, naturally occurring monospecific seedling cohorts of three tropical tree species and Svenning, Fabbro & Wright (2008) used seedling competition experiments on five shade-tolerant species to show that inter-seedling competition did not significantly affect seedling establishment. However, neither study examined the role of inter-trophic interactions on seedling establishment, nor did the study designs permit the examination of the effect of proximity to conspecific adults on seedling establishment patterns. Apart from a few single-species studies (Bell, Freckleton & Lewis 2006; Pigot & Leather 2008), manipulative field experiments that examine the relative importance of syntrophic competitive interactions versus inter-trophic competitive interactions in early recruitment stages are lacking, and the mechanistic basis of NDD remains unclear (Zimmerman, Thompson & Brokaw 2008).

In this study, we use a long-term multi-species experiment to examine the relative influence of syntrophic versus inter-trophic interactions on seedling establishment patterns. We accomplished this by deploying mixed compositions of experimental seedlings at fixed, highly elevated densities in shaded understorey sites situated at appropriate extremes of distance from large conspecific trees. We used mesh exclosures and a factorial experimental design that isolated distinct classes of natural enemies, and monitored survival for up to 45 months. We tested the following predictions: (i) seedlings located within the crown zone of large conspecific trees experience significantly lower overall survival and life span compared to seedlings growing under similar abiotic conditions but located far away from their nearest large conspecific tree; (ii) depressed seedling survival and life span within the crown zone of conspecific adults is caused by multiple classes of host-specific natural enemies; (iii) onset of significant distance-related seedling establishment patterns differs amongst species; (iv) intra-cohort resource competition is a not an important factor in seedling establishment.

Materials and methods

Study site and experimental set up

The research site was Cocha Cashu Biological Station (CCBS, 11°51′ S, 71°19′ W), located in the heart of Manu National Park in Madre de Diós, Perú. The site is well known for its high biological diversity and intact vertebrate community (Gentry 1990; Terborgh, Foster & Núñez 1996). The seedling survival experiment was established in a 4.0-ha (200 × 200 m) long-term forest dynamics plot located in mature floodplain forest. All stems ≥ 10 cm d.b.h. (diameter at breast height) in the plot have been mapped and > 98% have been identified to species. A total of 11 common tree species that represent phylogenetically diverse families and exhibit a range in traits such as adult stature, crown radius, seed size, germination mode and dispersal mode were chosen (Table 1). All species used were relatively large seeded (> 5 mm diameter) and weakly to strongly shade-tolerant as saplings. We specifically did not use species that are known heliophilic gap-colonizing specialists because their germination and growth is far more strongly influenced by light levels than biotic interactions. These species comprise a very small fraction of hyper-diverse tropical forest communities; the vast majority is comprised of shade-tolerant species present in the ‘advance regeneration’ guild (Hubbell 1979; Hubbell et al. 1999).

Table 1.   Life-history traits of species used in seedling survival experiment
  1. 1YI, year of initiation of species in seedling survival experiment.

  2. 2Disperser: BA, bats; PR, primates; W, wind.

  3. 3Germination mode: E, epigeal (foliaceous cotyledons); H, hypogeal (storage cotyledons).

  4. 4Vertical stratum: C, canopy; SC, sub-canopy; UT, understorey treelet (usually <15 m tall).

  5. 5CR, mean crown radius of focal large individuals in the tree plot. Each individual’s CR was the average of the distance from trunk base to tip of the outermost branch in each of the four cardinal directions.

  6. 6Regeneration guild: A, shade-tolerant ‘advance regeneration’ guild; common and persistent as saplings; B, high light levels critical for seedling-to-sapling transition, but not for seed germination and seedling establishment; rare as saplings.

  7. 7N, number of large individuals in tree plot (stems >25 cm d.b.h. for C species, >15 cm d.b.h. for SC species, >10 cm d.b.h. for UT species).

2003Brosimum lactescensMoraceaePRHC10B7
2003Calatola microcarpaIcacinaceaePRESC5A14
2005Clarisia racemosaMoraceaePREC9A5
2003Dipteryx micranthaFabaceaeBAEC11B5
2003Klarobelia candidaAnnonaceaePRHUT3A19
2004Lecointea peruvianaFabaceaeBAEC7A11
2004Leonia glycycarpaViolaceaePRESC3A8
2005Otoba parvifoliaMyristicaceaeBAHSC4A124
2003Pseudomalmea diclinaAnnonaceaePREC5A40
2004Pterocarpus rohriiFabaceaeWEC8B7
2004Spondias mombinAnacardiaceaePRHC10B6

The experiment was initiated in November 2003 with five species. A second iteration was implemented in November 2004 with four additional species, and a final iteration in March 2005 added two more species (Table 1). The procedures detailed below were replicated at each iteration. Sufficient quantities of undamaged, uninfested seeds of the focal species were collected from under multiple fruiting adults that were widely distributed across the forest and located outside the tree plot. Seeds were germinated in a controlled-environment greenhouse facility in individual biodegradable peat pots. The soil used in these pots was collected from multiple random locations in the forest away from a conspecific adult of any of the focal species, and homogenized. At the onset of the rainy season in early November 2003 and November 2004, and toward the latter portion of the rainy season in March 2005 – periods when community-wide natural seedling emergence reaches a peak in the forest (Terborgh 1983) – young seedlings (between 3 and 5 weeks old) were transplanted to predetermined seedling plot locations within the forest stand with their pots and soil-root network undisturbed so as to avoid transplant shock (the success of which was confirmed by > 85% survival of most species at the first census taken 8 months following transplantation).

Seedling plots and main effects

Two main effects were tested, each with two states: ‘Distance’–‘near’ and ‘far’ (N and F) from nearest conspecific adult; and ‘Treatment’–‘exclosure’ and ‘control’ (‘Ex’ and ‘C’).

We predetermined the locations of all experimental seedling plots using a digitized stand map of the entire encompassing 4-ha tree plot. For each species, distance to the nearest large conspecific tree was determined using spatial analysis tools in arcgis 9.1 (ESRI 2005), and areas that satisfied N or F distance criteria were identified for each species. For all the study species, a location that represented an ‘N’ condition for a focal seedling was within 5 m of the base of a large conspecific tree. Locations that represented an ‘F’ condition for a focal seedling were determined using previously estimated species-specific crown radii of large conspecific adults (Table 1): at least two or more crown radii away from the nearest large conspecific adult for species with very large statured adults (e.g. Dipteryx micrantha), and several crown radii from the nearest conspecific adult for species with smaller crown dimensions. Based on the above described criteria, appropriate seedling plot locations were first determined in the digitized stand map, and then found within the actual tree plot. For each of the 11 focal species, experimental seedlings were distributed across a minimum of four spatially disjunct ‘N’ and ‘F’ sites, i.e. the focal conspecific adult tree/s that defined each ‘N’ site for a species was at least three crown radii away from the other ‘N’ sites, and ‘F’ site locations for each species were at least 15 m apart. The eventual location of every seedling plot represented either an N or F condition for each of the seedling species used at that site, i.e. none of the locations were situated at an intermediate distance (> 5 m, < 2 crown radii units) from the nearest large tree for any species. The pairing of seedlings of multiple species within the same seedling plot, some representing the ‘N’ condition and others representing the ‘F’ condition, was specifically designed to determine the prevalence of distance-dependent mortality as well as the host-specificity of potential mortality-inducing natural enemies, by directly contrasting their performance under otherwise identical abiotic conditions.

Each seedling plot measured c. 50 × 20 cm (0.1 m2). Stakes made from the rot-resistant wood of the palm Iriartea deltoidea, were hammered into the four corners of the plot, and an outer frame of coarse mesh (c. 1 cm mesh size) c. 30 cm in height was wrapped around the wooden stakes and anchored to each stake with galvanized wire. Once all the seedlings were transplanted into a plot, the soil within the plot was levelled and covered with leaf litter. Each seedling plot contained a total of 16 seedlings per plot, comprised by a minimum of three or more (usually four) seedlings per species and up to four species per plot. Natural density of the mixed-species seedling layer (10–50 cm tall woody stems) in the same forest has been previously estimated as 6.4 ± 4.8 m−2 (Harms, Powers & Montgomery 2004). Thus, initial seedling density of 160 m−2 in all seedling plots represented c. 25 times the mean natural density of the mixed-species seedling layer. In seedling plots that had < 16 seedlings comprised by any combination of the 11 focal species, the deficit was made up by using seedlings of Ziziphus cinnamomum, a large-seeded species with shade-tolerant, long-lived seedlings. Ziziphus seedlings were used only in seedling plots located farther than a distance of three crown radii (> 25 m) away from the lone conspecific adult tree present in the tree plot.

Only one of the focal species –Clarisia racemosa– is known to produce natural dense monospecific aggregations of seedlings, i.e. ‘seedling carpets’ around reproductive adults, but none of the focal conspecific Clarisia individuals (all > 30 cm d.b.h.) in the tree plot produced a seedling carpet over the duration of the experiment. Therefore, at a larger spatial scale of multiple square metres, densities of experimental seedlings in both N and F conditions were equivalent, and potential confounding effects due to high densities of naturally germinated seedlings around large conspecific adults was not an issue.

The second main effect –‘Treatment’– was designed to clarify the relative importance of distinct classes of natural enemies on seedling survival, based on a design used previously by Lopez & Terborgh (2007). Cages of ‘Ex’ plots were fitted with a hood of commercial Brite™ screening (< 2 mm mesh size) that allowed c. 40 cm of vertical growth space to the seedlings within. The base of the hood was buried in the soil and firmly anchored with metal stakes to deny access to above-ground organisms > 2 mm in size. Brite™ screening was specifically chosen because it attenuates a minimal amount of incident light and thus does not substantially alter the natural amount of understorey light available to seedlings. Cages of ‘C’ plots were also fitted with identical Brite™ screening hoods to simulate the microenvironment of the Ex plots, but had 10 × 20 cm ‘gates’ cut into each of the longer sides of the Brite™ screening to allow access to above-ground organisms > 2 mm in size. A factorial design was used, with paired C and Ex plots established c. 4 m apart at each site. Thus, each individual seedling plot represented either an N or F condition as well as a C or Ex condition for each of the species in the plot. This experimental design provides a means of determining the relative importance of two distinct classes of natural enemies on seedling establishment patterns. Above-ground organisms > 2 mm, especially macroinvertebrates such as Orthopteran adults, Lepidopteran and Coleopteran larvae, etc. are unable to penetrate the 2-mm mesh of the Brite™ screening in Ex plots, and their influence is therefore restricted to C plots. Natural enemies < 2 mm such as microarthropods (e.g. thrips, leaf miners, etc.), above- and below-ground fungal pathogens, as well as soil-borne organisms such as nematodes, have access to seedlings in both C and Ex plots. Both treatments were designed to deny access to large vertebrate herbivores such as white-lipped and collared peccaries and deer, thus excluding potential effects of vertebrate browsing on seedling survival patterns. This decision was based on an extensive review of the published literature, which shows that the majority of observed instances of distance-dependent mortality patterns have involved invertebrates or fungal pathogens, while vertebrate herbivores have seldom been shown to exhibit distance-responsive predation (Terborgh et al. 1993; Hammond & Brown 1998; Wright 2002).

From all three iterations combined, a total of 106 seedling plots (53 each for controls and exclosures) were established, containing a total of 1050 seedlings of the 11 focal species. Seedling plot locations were distributed across the tree plot such that there were at least four or more seedling plots representing each of four possible combinations of main effects (N,C; F,C; N,Ex; F,Ex) for each focal species. In addition, the total numbers of individual seedlings representing each of the four combinations of main effects was comparable within each individual species as well as for all species pooled (see Table S1 in Supporting Information). While eventual sample sizes of seedlings and treatment replicates reflected logistical limitations, a power analysis (Steidl & Thomas 2001) suggested that the sample size of all species pooled, and most individual species (except Clarisia racemosa and Otoba parvifolia), was sufficient to detect a significant result (P < 0.05) for as low as a 10% difference in survival arising from the main effects. Seedling survival was monitored at regular intervals over the course of the experiment, with the final census conducted in July 2007. Corresponding census intervals following transplanting were at: 1, 8, 10, 13, 17, 24, 34 and 45 months for the 2003 iteration; 1, 5, 12, 22 and 33 months for the 2004 iteration; and 1, 8, 18 and 29 months for the 2005 iteration. Thus, the total duration of the experiment for species used in the November 2003, November 2004, and March 2005 iterations was 45, 33 and 29 months, respectively.

Understorey light measurement

All plots were located in closed-canopy shaded understorey in order to minimize variation in understorey light environment at the outset. Areas of the tree plot located within, or in the vicinity of a current or recent tree fall gap were avoided. Following set up of all the seedling plots in each iteration, canopy images were taken with a Nikon FC-E8 Fisheye Converter Lens fitted on a Nikon Coolpix 990 digital camera and positioned 1.5 m above each seedling plot. Images were always taken under uniformly overcast conditions because direct sunlight causes bright reflections on foliage that are difficult to distinguish from sky, and overestimates canopy openness and light availability. Canopy images were processed in hemiview 2.0 canopy analysis software (Delta-T Devices Ltd. 1999) to obtain total site factor (TSF) values, which measure the proportion of diffuse radiation reaching a given location (Brown et al. 2000).


Final survival

The effects of distance from large conspecific trees (‘Distance’), protection from macroinvertebrates (‘Treatment’) and variation in TSF values (‘Light’) on the proportion of seedlings alive at the final census were examined using various forms of nested logistic regression models, with final survival status (dead, alive = 0, 1) as the dependent variable, Distance and Treatment as categorical explanatory variables, and Light as a continuous explanatory variable. Since the pooled data contained seedlings from different iterations of the experiment, year of initiation (Year) was also included as a factor for the pooled data.

The effects of interactions between Distance and Treatment on the proportion of seedlings alive at the final census were explored using multiple tests of equal proportions (Newcombe 1998a,b) on groups defined by their interactions (‘N,C’ vs. ‘F,C’ vs. ‘N,Ex’ vs. ‘F,Ex’), which simultaneously generates confidence intervals for each proportion (‘prop.test’ in r version 2.6.4). Tests were performed with the pooled data from all 11 species, as well as individually for each of the 11 focal species that showed a significant interaction between the main effects.

Seedling lifetime

Seedling lifetime was calculated based on the total number of months a seedling lived from initiation of the experiment until either the final census if it were still alive, or until the census prior to the one in which a seedling was recorded as dead. While this is an admittedly imprecise estimate of seedling lifetime – since a seedling could have died at any point between two censuses – it allows for standardized comparison between treatments. Seedling lifetime values were not normally distributed because they were restricted to a finite set of possible values that corresponded with census intervals. Therefore, a Wilcoxson rank-sum test (Hollander & Wolfe 1973) was used to compare median lifetime values of seedlings exposed to F versus N conditions for individual species as well as all species pooled.

Survival analysis

For individual species, trends in survival over the entire duration of the experiment based on multiple censuses were analysed by computing nonparametric survival functions for N and F plots using the Fleming & Harrington (1984) method, which is most appropriate for the right-censored data used in this study. Survival curves with 95% confidence intervals were plotted using the estimated survival functions, which allowed for visual comparison of survival trends. Cumulative survival functions of seedlings situated in N and F conditions were tested for significant differences using the G-rho family of tests (Harrington & Fleming 1982). Survival functions for seedlings situated in N and F conditions were calculated only for species that had significantly higher final survival in F versus N conditions.

All statistical analyses were performed using the computing environment r version 2.6.4 (R Development Core Team 2008).


Final survival

The proportion of original seedlings alive at the final census ranged from 0.20 to 0.66 for individual species, and was 0.37 for all species pooled (Table S1). Seedling survivorship at F sites (0.42) represented a 40% increase compared to N sites (0.30), and survivorship in exclosures (0.45) represented a 61% increase over controls (0.28) (Fig. 1a). Mean TSF values were comparable between F sites (0.16 ± 0.03) and N sites (0.15 ± 0.03). Distance or Treatment alone had a highly significant effect on the proportion of seedlings alive at the final census whereas Light alone had a marginally significant effect, and Year alone did not have a significant effect (Table 2). Models that already contained either the Distance or Treatment variable were not significantly improved by adding Light as an additional variable. In contrast, addition of either the Distance or Treatment variable to a model that contained only the Light variable caused a highly significant reduction of the residual deviance. Models that contained only the Distance or Treatment variable were significantly improved by adding the other main effect.

Figure 1.

 Proportion of seedlings alive at final census (y-axis) for all species pooled, grouped by (a) main effects, (b) distance × treatment: Error bars are 95% confidence intervals generated from a test of equal proportions. Groups (x-axis) – N: near; F: far; C: control; Ex: exclosure; N,C: ‘near’ controls; N,Ex: ‘near’ exclosures; F,C: ‘far’ controls; F, Ex: ‘far’ exclosures.

Table 2.   Comparison of nested models using logistic regression
ModelResidual deviance (RD)Reduction in RDP-value (α2 test)
Null model1019.98  
Year only1019.40.480.46
Light only1015.14.880.03
Light + Distance1001.1313.980.0001
Light + Treatment1005.0910.020.001
Distance only1004.0915.890.00007
Distance + Light1001.132.960.09
Distance + Treatment993.1710.920.0009
Treatment only1009.2410.750.001
Treatment + Light1005.094.150.04
Treatment + Distance993.1716.070.00006

Overall seedling survival was significantly different across groups defined by their interaction (Table 3, Fig. 1b). Within both controls, and exclosure treatments considered separately, survival of seedlings at F sites was significantly higher than at N sites. Within controls, seedling survivorship at F sites (0.32) represented a 33% increase compared to N sites (0.24), and within exclosures survivorship at F sites (0.53) represented a 47% increase over N sites (0.36). Survival of seedlings in ‘near exclosure’ (N,Ex) plots versus ‘far control plots’ (F,C) was not significantly different. Within both N and F sites considered separately, survival in exclosure plots was significantly higher than in controls.

Table 3.   Results of tests of equal proportions between groups represented by the two main effects and their interactions, for all species pooled
Testα2Significance (P-value)
N,Ex ≠ N,C ≠ F,C ≠ F,Ex62.47< 0.00001
F,C > N,C4.35< 0.05
F,Ex > N,Ex17.77< 0.00001
N,Ex > N,C9.28< 0.001
F,Ex > F,C30.21< 0.00001
N,Ex > F,C0.99> 0.05

Amongst individual species, survival of seedlings at F sites was higher than in at N sites for eight of 11 species (Fig. 2a–h, Table S2) and the difference in survival was significant in five species –Dipteryx, Calatola, Pseudomalmea, Spondias and Pterocarpus (Fig. 2a–e). Survival in exclosures was significantly higher than in controls for five of 11 species –Brosimum, Calatola, Clarisia, Lecointea, Pseudomalmea and comparable for six species –Dipteryx, Klarobelia, Leonia, Otoba, Pterocarpus, Spondias (Fig. 2, Table S2). Of the three species that did not have higher survival at F vs. N sites, Lecointea had significantly higher survival in exclosures versus controls whereas Klarobelia and Leonia did not.

Figure 2.

 Proportion of seedlings alive at final census for individual species (y-axis) in groups defined by the two main effects. Groups (x-axis) – near (N), far (F), control (C) and exclosure (Ex). P-values are for comparisons of F vs. N, and Ex vs. C. Significance: *****P < 0.00001, ***P < 0.001, **P < 0.01, *P < 0.05, *#0.05 ≤ P ≤ 0.08, #P > 0.08.

The interaction between exclosure effect and distance effect was significant in four species (Fig. 3, Table S2). Of these, Calatola and Brosimum had significantly higher survival at F sites only in exclosures, Pterocarpus had significantly higher survival at F sites only in controls, and Dipteryx had significantly higher survival at F sites separately in controls as well as in exclosures. Seven species did not have significantly higher survival at F versus N sites in either controls or exclosures when considered separately.

Figure 3.

 Proportion of seedlings alive at final census for individual species (y-axis) in groups defined by interactions of the two main effects. Groups (x-axis) –‘near controls’ (N,C), ‘far controls’ (F,C), ‘near exclosure’ (N,Ex) and ‘far’ exclosure’ (F,Ex) treatments. P-values are for comparisons of F,Ex vs. N,Ex and F,C vs. N,C. Significance: ****P < 0.0001, **P < 0.01, *P < 0.05, #P > 0.05.

Seedling lifetime

Median seedling lifetime varied greatly amongst species, from 3.5 months (Brosimum) to 44 months (Klarobelia) (Table 4). For all species pooled, median seedling lifetime was 9 months (75%) longer for seedlings at F sites compared to N sites. The maximum difference in median seedling lifetime between N and F sites was 17 months (Calatola). Median seedling lifetime was significantly longer at F vs. N sites for all species pooled together, and four out of nine individual species that were used in the 2003 and 2004 iterations (Wilcoxon rank-sum test, Table 4).

Table 4.   Wilcoxon rank-sum test of difference in median seedling lifetime (months)
SpeciesSeedling lifetime (months)Test statistic and P-value
All plotsFar plotsNear plots
All species pooled162112W = 92769, P = 0.0002
Calatola microcarpa233316W = 803.5, P = 0.002
Dipteryx micrantha162312W = 1293.5, P = 0.002
Pterocarpus rohrii323221W = 44178.5, P = 3.78e-08
Spondias mombin112111W = 3791.5, P = 0.007
Klarobelia candida444444W = 478, P = 0.84
Pseudomalmea diclina232323W = 1613, P = 0.12
Leonia glycycarpa427W = 276.5, P = 0.60
Brosimum lactescens3.507W = 726, P = 0.47
Lecointea peruviana4411W = 772, P = 0.53

Survival analysis

Cumulative survival functions were calculated for five species that had significantly higher final survival at F versus N sites –Dipteryx, Calatola, and Pseudomalmea from the 2003 iteration, and Pterocarpus and Spondias from the 2004 iteration (Fig. 4). Four out of the five species tested had significantly higher cumulative survival functions at F versus N sites, with Pseudomalmea (Fig. 4c) being the only exception. Examination of the seedling survival curves of individual species revealed that significantly different survival rates between F and N sites appeared after widely varying times among species (Fig. 4a–e). In some species, the ‘distance effect’ became significant at a very early stage and persisted (Fig. 4a) or intensified (Fig. 4b) over time, while in others, the distance effect became significant only at the very end of the experiment (Fig. 4d,e).

Figure 4.

 Survival curves (thick lines) with 95% confidence intervals (thin lines) for seedlings in F (solid line) and N (dashed line) conditions. y-axis: Proportion of seedlings survived from start of the experiment until the final census. x-axis: Duration of survival (months). Tick marks correspond to census intervals. Species are: (a) Dipterx micrantha, (b) Calatola microcarpa, (c) Pseudomalmea diclina, (d) Pterocarprus rohrii, (e) Spondias mombin.


Results of this experiment provide strong multi-species support for all four predictions tested. Seedlings located within the crown zone of large conspecific adults experienced significantly lower overall survival and life span compared to seedlings growing under similar abiotic conditions but located far away from their nearest large conspecific tree (prediction 1). ‘Escape in space’ from distance-responsive natural enemies provided a 40% increase in final survival and 75% increase in median lifetime. The use of mesh exclosures and a factorial experimental design with Treatment nested within Distance revealed the different influences of two distinct classes of natural enemies on seedling establishment patterns (prediction 2). While both classes include important natural enemies, it appears that host-specific organisms < 2 mm (including below-ground soil-borne organisms) are more strongly distance-responsive and depress seedling establishment primarily in the vicinity of large conspecific adults whereas above-ground organisms > 2 mm appear to have a negative impact on seedling establishment at all distances. Pooled seedling survival in exclosures was significantly higher than in controls even at far distances, and in the case of Lecointea, irrespective of distance. The vulnerability of young seedlings to larger, more generalist, and less distance-responsive natural enemies also serves to mask the effect of the smaller, more strongly distance-responsive natural enemies on seedling establishment. Pooled seedling survival in controls was 33% higher at F versus N sites (P < 0.05); whereas in exclosures it was 47% higher survival at F versus N sites (P < 0.00001). This could explain why a significant distance effect was seen only in exclosures but not in controls for two out of four individual species that showed a significant interaction between the two main effects. In these species (Calatola and Brosimum), smaller sample sizes and low survival in controls at both N and F sites masked an overall distance effect that was being caused primarily by natural enemies < 2 mm or below-ground soil-borne organisms. Only Pterocarpus displayed a significant distance effect only in controls and not in exclosures, suggesting that its distance-responsive host-specific natural enemies are primarily aboveground organisms > 2 mm.

Three of the five species that had significantly higher seedling survival at F sites –Dipteryx, Pseudomalmea and Pterocarpus– also experience considerable seed predation under fruiting trees (V. Swamy, personal observation), and therefore appear to experience distance-based early recruitment bottlenecks at the seed as well as the seedling stage; this has previously been well-documented for the genus Dipteryx (Clark & Clark 1984; Cintra 1997). In contrast, Calatola and Spondias are known to persist as intact seeds for extended periods of time on the forest floor (Terborgh et al. 1993) but experienced significantly higher seedling survival at F sites, suggesting that distance-based differential mortality in these species is restricted to the post-germination phase of early stage recruitment. Three species –Klarobelia, Leonia and Lecointea– which did not show a significant ‘distance effect’ on seedling establishment, are known to experience high levels of seed predation by Hymenopteran and Coleopteran insects (V. Swamy, personal observation). The early stage recruitment bottleneck in these species therefore appears to be strongest at the seed stage. This does not rule out an additional bottleneck at later recruitment stages, particularly for a species like Klarobelia with very high overall seedling survival at 45 months.

The considerable variation in the onset of significant distance-related survival patterns amongst the focal species (prediction 3) illustrates that a variety of host-specific distance-responsive natural enemies have adapted to a range of plant life-history traits. For Dipteryx, the critical bottleneck had occurred by the first census, with the resulting difference in survival between F and N conditions remaining relatively constant over the duration of the experiment even though further mortality was experienced by seedlings in both conditions (Fig. 4). In contrast, a significant distance effect in Calatola emerged only after 17 months, and became progressively stronger over the remainder of the experiment. Spondias and Pterocarpus showed gradually developing distance effects that became progressively larger over the course of the experiment. Silman (1996) showed that seed size and germination mode alone explained a significant proportion of the variance in overall seedling survival in the same forest, with significantly higher survival in larger-seeded and hypogeally germinating species. However, Silman’s study did not explicitly examine the effect of proximity to large conspecific adults on survival. All of the species used in this experiment were relatively large seeded (> 5 mm diameter), and distance effects were observed for epigeal as well as hypogeal modes of germination. Therefore, while specific life-history traits and combinations thereof may significantly enhance overall seedling survival in some species over others, distance-responsive natural enemies appear to exert a consistent and strong influence on spatial patterns of seedling establishment by creating a strong selective advantage for dispersal away from large conspecific adults.

The above results coupled with the varying experimental duration (29, 33 and 45 months) of the three separate iterations provides some perspective on the minimum experimental duration required to capture the onset of a significant distance effect on survival. Although different species were used for each iteration and the results are therefore not directly comparable, it appears that even 29 months is an insufficiently long experimental duration for strong distance-dependent seedling establishment patterns to develop in some species (Clarisia and Otoba), and may require even longer periods of time to develop in other species (Klarobelia). Studies that follow seedling survival over the course of a single field season or a few months – which comprise the vast majority of previous tests of distance dependence in early stage recruitment – are therefore unlikely to detect significant results in many species, and instead arrive at premature and incorrect conclusions.

No direct evidence was found for the influence of intra-cohort resource competition on seedling establishment (prediction 4) even though experimental seedlings were densely packed together in all plots at c. 25 times the natural density of the mixed-species seedling layer in this forest. Under these conditions, a seedling establishment pattern with a few survivors and several stunted or dead individuals in each seedling plot irrespective of distance to large conspecific trees would have supported resource competition; this was not observed. In contrast, a highly significant distance-dependent mortality pattern emerged even though the relative fractions and species composition of N and F seedlings in each plot varied widely across plots. Although the experiment did not manipulate seedling density, the disproportionately high mortality of N seedlings growing side-by-side with F seedlings at fixed, high densities indicates that natural enemies take precedence over competitive interactions in affecting seedling establishment even under artificially elevated small-scale densities. Thus, NDD in terms of intra- or inter-specific competition for abiotic resources does not appear to be a significant factor in the earliest recruitment stages of species with life-history traits similar to those used in the experiment. Two recent studies (Paine et al. 2008; Svenning, Fabbro & Wright 2008) arrived at the same conclusion but our study goes further by using mixed aggregations of multiple species and simultaneously documenting the effect of distance from conspecific adults on seedling establishment.

At a larger spatial scale of multiple square meters, the density of conspecific experimental seedlings of any given study species was comparably low at both N and F sites, even taking into account naturally germinated seedlings of focal species, as none of these species formed natural dense aggregations of seedlings, i.e. ‘seedling carpets’ around reproductive adults that would inflate conspecific seedling density only at N sites. Under these conditions, the significantly higher mortality of seedlings at N sites compared to F sites suggests that their host-specific natural enemies are primarily influenced by distance from conspecific adults rather than density of conspecific seedlings i.e. they are distance-responsive, and not density-responsive in their mode of predation. Even if the effect of variation in conspecific adult densities at larger spatial scales were considered, areas of highest conspecific adult density at scales relevant to inter-trophic interactions are inescapably situated in close proximity to large conspecific adults (Schupp 1992). In a rare, well-documented natural situation involving the tapir-dispersed tropical palm Maximiliana maripa, distance from conspecific adults was found to be the most important factor in seed predation and seedling establishment, and naturally high seed densities in tapir latrine sites located far from conspecific adult stands did not adversely impact seed survival or seedling establishment (Fragoso 1997; Fragoso, Silvius & Correa 2003). Certain host-specific natural enemies that eliminate dense seed or seedling carpets that form naturally around reproductive adults of some species are likely to be ‘purely’ density-responsive (e.g. Bell, Freckleton & Lewis 2006) but the high-density conditions they are attracted to are non-random and inextricably linked to the crown zones of reproductive adults.

The minimal effect of light availability in the shaded understorey on seedling establishment patterns observed in this study is tempered by the a priori selection of shade-tolerant species and the location of seedlings plots in closed-canopy shaded understorey, where light availability is already greatly reduced and does not vary greatly. However, this does not rule out an important role for light in seedling establishment even for shade-tolerant species, particularly in naturally occurring high light conditions such as tree fall gaps where biotic and abiotic factors can have interactive effects, as shown in previous studies (Cintra & Horna 1997; Alvarez et al. 2008).


Results of this study provide strong, multi-species support for the influence of host-specific distance-responsive natural enemies on post-germination seedling establishment, and suggest that negative density-dependent patterns of tree recruitment in tropical rain forests are at least partly produced at early life stages as an outcome of processes described by the classic Janzen–Connell model. The strength of observed distance-dependent patterns of seedling establishment highlights the importance of seed dispersal for recruitment success. Disproportionately low seedling survivorship in the proximity of large conspecific adults observed in this study concur with later-stage sapling recruitment patterns documented in the same forest where the median distances between saplings and their nearest large conspecific adult are equal to multiple adult crown-widths for the majority of common species (Terborgh et al. 2002). Thus, undispersed seeds, represented in this study by seedlings situated in close proximity to reproductive adult trees, appear to make a minimal contribution to later stages of recruitment. The influences of distance-dependent mortality and seed dispersal on community composition are further demonstrated by the drastically altered recruitment patterns observed in an ‘empty’ forest devoid of large mammalian and avian dispersers (Terborgh et al. 2008).

Our results highlight the limitations of meta-analyses that combine multiple single-species studies conducted at different sites. Hyatt et al. (2003) combined data from studies conducted in tropical and temperate forests: studies that focused exclusively either on seed or seedling survival, and studies in which the focal species were shrubs or herbs. In addition, the designs and durations of individual studies were highly variable, and many had serious limitations in their capacity to detect distance dependence. The enormous variation in individual results resulting from the aforementioned issues produced an overall insignificant result, which – in light of the results presented here – is not surprising and illustrates the advantages of a multi-species comparative study at one site.

This study also underscores the importance of an experimental approach focusing on the earliest stages of recruitment in order to fully understand tree recruitment processes in tropical forests (Freckleton & Lewis 2006). Complex biological interactions at the earliest stages of recruitment are completely overlooked in observational studies that only consider stems ≥ 1 cm in diameter (e.g. Hubbell & Foster 1986; Hubbell, Condit & Foster 1990; Condit et al. 1992, Condit, Hubbell & Foster 1994; Hubbell et al. 1999; Condit et al. 2000 and numerous others). Species that may appear to be ecological equivalents as adults or even as saplings are often drastically different at early life-history stages, which could account for differences in distributions and abundances observed at later life-history stages (Grubb 1977). Ultimately, the continued documentation of processes and patterns occurring at the earliest stages of recruitment, using well-designed experiments when required, will provide a more complete understanding of the mechanisms that allow for species coexistence and the stability of hyperdiverse tropical forests.


We thank Jorge Arnaiz, Julian Huarancasi, Ricardo, Florencio, Darwin Osorio, Christina Supples, Michelle Hu and Chris Martin for help with fieldwork, and Simon Lunagomez and Micheal Lavine for advice on statistical analyses. Financial support to V.S. included multiple Duke University Graduate School Research Travel Grants and Mellon Research Travel Grants from the Center for Latin American and Caribbean Studies; the Lewis and Clark Fund for Exploration and Field Research from the American Philosophical Society, a Francis Bossuyt Fellowship from Organization for Tropical Studies and a Sigma Xi Grants-in-Aid of Research Award. We are grateful to the Instituto Nacional de Recursos Naturales (INRENA), Peru for providing authorization to conduct field research in Manu National Park.