Study sites

The study was conducted at the Isthmus of Panama, which exhibits a pronounced rainfall gradient from 1600 mm/year at the Pacific Coast to over 3000 mm/year at the Atlantic coast across a distance of only 65 km [4, 8]. The length of the dry season, which typically starts in January and ends in May, correlates negatively with annual rainfall [8]. Mean annual temperature is 25°C with little variation across the gradient or throughout the year.

The experiment was conducted in two forests about 50 km apart: a drier semi- deciduous forest located in the national park Camino de Cruces (9° 2'N, 79°35'W, 2000 mm annual rainfall; modeled based on BIOCLIM data, [37]), and a wetter evergreen lowland forest in the national park San Lorenzo (9°16'N, 79°58'W, 3200 mm annual rainfall). Both sites are located in the Tropical Moist Forest Life Zone [38, 39]. However, rainfall and moisture regime, as well as species composition vary greatly. Dry season length is approximately 150 and 120 days, and rainfall in the driest quarter of the year 530 mm and 800 mm [8, 38], respectively. Soil water potentials in the upper soil layer of the dry site reach values well below -2 MPa in the dry season, but remain high throughout the year in the wet site (Engelbrecht, unpublished data). Both forest sites were mature secondary forest located on sedimentary bedrock. Only about 10% of the species in the areas overlap [3, 38]. In the following we refer to these sites as “dry” and “wet”, respectively. Permits for working in the national parks were granted by the “Autoridad Nacional del Ambiente (ANAM).

Experimental design

At each forest site, 60 paired plots (90 cm x 90 cm) were established, with each pair including a pesticide treatment (fungal pathogen and insect herbivore exclosure, see below) and a control plot (2 sites x 2 treatments x 30 plots). Seeds of 15 species with “wet origin” and 11 species with “dry origin” (S1 Table, for definitions see below) were sown into each plot, with one seed of each species in each plot. Germination, seedling survival in the dry and the wet season, and growth were followed over one year.

Seedling plots

The plots were set-up in the forest understory avoiding any gaps, with pairs separated at least 70 m from each other, spanning an overall area of about 300 ha in San Lorenzo (wet site) and 150 ha in Cruces (dry site). The control and exclosure plots were separated by at least 2 m. Where relevant, the control plots were set-up uphill from the exclosure plots to prevent runoff from the treatment to the control (three times in San Lorenzo and two times in Cruces). To allow access of insect herbivores but prevent seed or seedling removal by rodents or other mammals, which were not the focus of this study, all plots were caged with 2.5 x 2.5 cm wire mesh.

Study species and plant material

Study species were selected to include common species with strong association to the dry or the wet side of the isthmus. We focused on shade-tolerant species, since they represent about 80% of the species in these forests [40, 41]. Species with small seeds (< 0.5 cm length) were excluded to facilitate their manipulation and localization in the field. Potential study species were selected based on their occurrence in 50 1 ha plots spanning the rainfall gradient [42] and/or their abundance in a wet and a dry forest plot (Sherman, 6 ha and Cocoli, 4 ha, respectively, see [42]). Species with predominantly wet Caribbean side occurrence, that did not occur on the dry Pacific side of the Isthmus, or that had at least double the abundance in the wet than the dry side plots were classified as “wet origin species”, whereas species occurring predominantly on the dry Pacific side, that did not occur on the wet Caribbean side of the Isthmus, or that had at least double the abundance in the dry than the wet side plots were classified as “dry origin species” (S1 Table).

Seeds were collected in mature secondary forests across the Isthmus within their respective natural home range in the national parks San Lorenzo, Soberania, Chagres and Camino de Cruces during the dry season and beginning of the wet season 2012 (March to mid-May). Ripe seeds were collected from a minimum of three mother trees per species by directly harvesting from the tree, or from freshly fallen fruits. Damaged seeds were removed after visual inspection for damage by predators or pathogens. Final selection of the study species was based on the availability of enough undamaged seeds, resulting in 15 “wet” and 11 “dry” origin species.

Pest exclosure treatment

To exclude fungal pathogens and insect herbivores (summarized as pests) a combination of a fungicide and an insecticide (summarized as pesticides) was applied monthly to the treatment plots. Actara (active ingredient: Thiamethoxam), a systemic broad-spectrum insecticide, and Diligent (active ingredients: Methalaxyl and Chlorothalonil), a systemic broad-spectrum fungicide with protectant properties effective against true fungi as well as oomycetes were used. According to the specification of the manufacturer, Actara was used in a solution of 0.5 g/l water, and Diligent in a solution of 5 g/l. Each exclosure plot was sprayed with a mixture of 40 ml of the insecticide and 40 ml of the fungicide solution. The control plots were sprayed with the same amount of rainwater, to ensure that results were not biased by additional water availability in the treatment. Studies using similar pesticide treatments have discarded negative influences on non-target organisms, including the plants themselves [43, 44]. Seeds in the exclosure plots were additionally pre-treated with the broad-spectrum insecticide Brigadier (active ingredient: Bifenthrin), and the fungicide Diligent (see above) to avoid seed predation. The insecticide was used undiluted with 50 ml/kg seeds. The seeds were briefly soaked in both solutions. Seeds of the control plots were soaked in rainwater.

Seed sowing

120 seeds per species (2 sites x 2 treatments x 30 plots, with one seed per species per plot, totaling 3600 seeds) were sown at the end of the dry season/beginning of the wet season (starting in March 2012). To ensure high germination rates in the typically recalcitrant seeds and to mimic natural seeding periods, seeds were sown as soon as possible after collection (maximum two days later), and distributed evenly between exclosure and control plot and wet and dry site (i.e. not all seeds of one species were planted at the same time). Seeds were planted on the mineral soil under the leaf litter, in a 15 x 15 cm grid, with species assigned randomly to the positions. Leaf litter was disturbed as little as possible to ensure natural microhabitat conditions in the plots. To prevent washing away and to facilitate relocation, seeds were fixed to the ground with wooden toothpicks and positions marked.

Seed germination, survival and growth

Seed germination and seedling survival were monitored between March 2012 and April 2013, i.e. during the transition between initial dry and wet season, a wet season and a second dry season. Rainfall during the study period did not differ substantially from the long-term average, except for almost the double amount of rainfall in November and December 2012 [45]. During the first 3.5 months, the time of highest germination, censuses were conducted biweekly to ensure that all germinating seeds were recorded; radicle emergence was counted as germination. Thereafter, censuses were conducted at monthly intervals for seedling survival, based on aboveground living biomass, and for occasional further germination. Overall growth was assessed at the last census based on seedling height, measured from the ground to the highest meristem.

From the census data we quantified six performance parameters: (1) overall establishment success (proportion of remaining seedlings at the end of the experiment relative to the original number of seeds sown; covers the period from March 2012 to April 2013); (2) overall growth (height of the seedlings at the end of the experiment in April 2013); (3) germination (proportion of seeds that germinated until the end of the experiment, relative to the original number of seeds sown); (4) survival during the wet season (proportion of seedlings that survived until December 2012, relative to the number of germinated seeds); and (5) survival during the dry season (proportion of seedlings that survived until April 2013, relative to the number of seedlings present at the start of the dry season).

Soil moisture and light

We recorded gravimetric soil moisture at each census, and averaged over the dry and wet seasons, respectively. For light availability, canopy openness was assessed once during the dry and wet season, respectively (Fig 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Gravimetric soil moisture (A) and canopy openness (B) in the dry and wet site across seasons.

Colors indicate the end of the dry season 2012 (red), wet season 2012 (blue) and dry season 2013 (dark red). Included are results of an ANOVA for effects of site, season and site x season interactions. Different letters represent significant differences at the 0.05 level in a Tukey post-hoc test. Presented are means (thick horizontal lines), 95% CI (thin lines), and raw data (points).

https://doi.org/10.1371/journal.pone.0143955.g002

For soil moisture, three random soil cores 15 cm deep were taken within 1 m from each plot pair; fresh and dry weight (after drying to constant weight at 105°C) were determined and percent gravimetric water content was calculated based on dry weight. We assessed canopy openness (percent open sky) from hemispherical photographs taken 1 m over each plot during the dry (April 2012) and during the wet season (October 2012), using a Nikon Coolpix P5000 camera with a Fisheye Converter FC-E8. Photographs were analyzed with the program Gap Light Analyzer v2.

Gravimetric soil moisture and canopy openness varied significantly across sites and between seasons: gravimetric moisture was lower in the dry than in the wet forest and lower in the dry than in the wet season (Fig 2a). Conversely, canopy openness was higher in the dry than in the wet forest, and higher in the dry than in the wet season (Fig 2b).

Statistical analyses

Our main aim was to assess the effect of site (as a categorization of moisture, see below), species origin, pest exclosure and light on probabilities of germination and survival, as well as on growth.

We initially tested for correlations between the explanatory variables (S2 Table). Soil moisture correlated strongly between seasons as well as with the factor site (r ≥ 0.8, see S2 Table). Therefore, soil moisture and site could not be maintained together in the same model. Models treating soil moisture as continuous variable and models with the factor site (wet/dry, see below) yielded qualitatively the same results. We present results of the models with the factor site, because we were interested in responses to the large-scale rainfall gradient rather than small-scale responses within sites, and because using the factor site better reflected the experimental setup with a separate “dry” and “wet” site.

The performance parameters we analyzed were: (1) establishment success, (2) growth, (3) germination, (4) survival during the wet season and (5) survival during the dry season (Table 1). To account for heteroscedasticity and non-normality of the residuals, growth data was log-transformed using the natural logarithm.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Effects of site, species origin, pest exclosure and light on performance parameters of tropical tree seedlings.

https://doi.org/10.1371/journal.pone.0143955.t001

We initially assessed species effects on performance by fitting a separate model for each performance parameter (Generalized linear mixed effects with binomial distribution (GLMM), or linear mixed effects (LMM) for growth, see S1 Fig). Species was used as fixed effect factor in each model. Random intercepts were plot-pairs and plots, with plot nested in plot-pairs.

To assess the effects of site, origin, treatment and light availability on performance, one model per performance parameter was fitted (GLMM or LMM, respectively). For every model, fixed effect factors were site (dry/wet), origin of the species (dry/wet), treatment (pest exclosure/control) and the average light availability (canopy openness in %) for the period analyzed (dry season, wet season, annual mean of dry and wet season, respectively). We also included the triple interaction term site x origin x treatment (which includes the pairwise interactions site x origin, site x treatment and origin x treatment), and the interaction term origin x light availability (Table 1). Random intercepts were species, plot-pair and plot. We nested species in plots and plots in plot-pairs. Single term deletion of non-significant terms was used for model selection. We removed sequentially first all interaction terms and then all explanatory variables that led to a model with a lower Akaike Information Criteria (AIC).

To test our specific expectations (Fig 1), we conducted eight planned comparisons using least squares means [46] with Tukey correction as post-hoc tests (Table 2). To assess if seed germination and seedling survival varied across sites or with species origin under the natural condition of the habitat, we assessed under control conditions the effect of origin of the species within the dry and the wet site, respectively, and the effect of site on wet origin and dry origin species, respectively (i.e. four contrasts, Table 2a and 2b). To assess to what extent germination and survival were affected by pests, we assessed the pest exclosure effect in each site (wet/dry) and in species with different origin (wet/dry, i.e. four contrasts, Table 2c and 2d). Tukey post-hoc comparisons, means and standard errors in tables and figures are from the least squares table [46].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Planned comparisons of (A) effects of origins and sites under natural (control) conditions, and (B) effects of pest exclosure on performance parameters.

https://doi.org/10.1371/journal.pone.0143955.t002

All statistical tests were done using R 3.0.2 [47] with the packages lme4 1.0.5 [48], lsmeans 2.00–5 [46] and LMERConvenienceFunctions 2.5 [49].

Seedling establishment

At the end of the experiment after one year, 22.2% of the seedlings out of all seeds that were sown survived (799 out of 3600). We expected species to have a disadvantage outside their home range exhibiting lower overall establishment success, with dry origin species being limited by pest pressure and/or low light availability in the wet site, and wet origin species being limited by drought in the dry site. Contrary to our expectation, for control seedlings there were no significant differences in the establishment success of the two origins within sites (Table 2a, Fig 3a), nor across sites (Table 2b, Fig 3a). Thus, under natural conditions, dry origin and wet origin species did not differ in their probability to establish in both sites. Pest exclosure significantly enhanced seedling establishment, and the strength of the effect depended on site and origin (significant treatment x site x origin interaction, Table 1, Fig 3a): the pest exclosure enhanced establishment success of wet origin species on both sites (Table 2c and 2d), but of dry origin species only in the dry site (Table 2c). Thus, contrary to our expectations, pest exclosure had no positive effect on dry origin species in the wet site (Table 2d). Light had no overall significant effect on the establishment success over the whole experimental period. There was a marginally significant origin x light interaction; however, opposite to our expectations, wet origin species profited more from higher light availability than dry origin species (Table 1, Fig 3b). None of the other performance parameters exhibited a significant light x origin interaction (Table 1).

Growth

On average, after one year species had grown to a height of 14.1 cm (3.5–62.5 cm). We found a significant site x origin interaction (Table 1): Dry origin species grew significantly less than wet origin species in the wet site (Fig 3c). Even in the dry site dry origin species grew less than wet origin species, although the difference was less pronounced than in the wet site (Table 2a, Fig 3c). Dry origin species grew less in the wet site than in their dry home range, while wet origin species showed no difference in growth performance across sites (Table 2b, Fig 3c). Independent of species origin, overall growth was lower in the wet site compared to the dry site (Table 1). Dry origin species had an overall lower growth than wet origin species independent of site (Table 1). Pest exclosure did not affect growth, neither alone nor in interaction with origin nor site, indicating that pests did not limit growth in our experiment (Table 1, Table 2c and 2d). Growth was also not affected by the variation of light availability encountered in this study (Table 1, Fig 3d).

Germination

Out of the seeds sown in the experiment, 38% germinated (1384 of 3600). Germination ranged from 0% to 94% among species (S1 Fig). Within sites, no differences were found between origins under natural conditions (Table 2a, Fig 4a). Contrary to our expectations, dry origin species showed a marginally significant trend (p = 0.06) to germinate better in the wet than in the dry site under natural control conditions, while wet origin species did not show any differences between sites (Table 2b). As expected, pest exclosure benefited dry origin species more than wet origin species, but opposite to our expectations, only in their home range (Table 2c and 2d, Fig 4a). Wet origin species showed no higher germination rates when pests were excluded, neither in the dry nor in the wet site (Table 2c and 2d). Light availability did not affect germination, neither for dry nor for wet origin species (Table 1, Fig 4b).

Seedling survival during the wet season

In the wet season, 72% of the germinated seeds survived (994 of 1384). Although the interaction site x origin x treatment was marginally significant (Table 1), trends did not conform to our expectations (Fig 4c). Wet origin species did not show higher survival rates than the dry origin species in any of the sites (Table 2a, Fig 4c), although we had expected them to show lower mortality during the wet season than dry origin species. Also opposite to our expectations, there was a trend for wet origin species to perform better in the dry site, while dry origin species showed no difference between sites (Table 2b). Pest exclosure benefited wet origin species both in the dry and the wet site, and dry origin species only in the dry site (Table 2c and 2d). Wet season survival of the seedlings increased with increasing light availability, but there were no differences between the two origins (Table 1, Fig 4d).

Seedling survival during the dry season

The subsequent dry season was survived by 80% of the seedlings (799 of 994). There was a significant site x origin interaction in the dry season, with seedling survival in both sites dependent on their origin (Table 1, Fig 4e). Under natural control conditions in the dry site, wet origin species had a significantly lower survival than dry origin species (Table 2a). Wet origin species also had a lower survival in the dry than in the wet site (Table 2b). Survival in the exclosure followed the same pattern (Fig 4e), indicating that drought rather than pests led to the lower survival of wet origin species in the dry site. In contrast, dry origin species under control conditions showed no differences in survival between sites (Table 2b). Pest exclosure increased survival of wet origin species both in the dry and the wet site (Table 2c and 2d). For dry origin species, the exclosure effect was only marginally significant in the dry site (Table 2c). During the dry season, the triple interaction site x origin x treatment was not significant (Table 1). Light availability had no significant effects in the dry season (Table 1, Fig 4f), although the difference in light availability between the sites was highest during the dry season (Fig 2b), probably due to leaf shedding.

Exclusion of wet forest species from dry sites

Our results indicate that our focal wet forest species were less drought-tolerant than the dry forest species, and that drought limited their survival in the dry site during the dry season, as we had expected: Their dry season survival in the dry forest was much lower compared to dry forest species (Fig 4e, Table 2a), and compared to the wet forest site (Fig 4e, Table 2b). We can rule out that these effects were due to pest pressure, because seedlings in controls and exclosures followed the same pattern (Table 2c and 2d, Fig 4e), or that they were due to light, because light availability had no effect on seedling survival (Table 1, Fig 4f). Therefore, drought was directly responsible for reducing survival in the dry site. These results confirm previous studies in tropical forests worldwide which show that seedlings of wet forest species (or occurring in wet sites) are less drought-tolerant, i.e. more susceptible to drought, than seedlings of species occurring in dry forests exposed to a strong dry season [6, 8–10, 50].

Despite their lower drought tolerance, after one year wet forest species in the dry site did not perform poorer than dry forest species neither in terms of overall establishment nor growth (Fig 3a and 3c, Table 2a and 2b). Slightly, but non-significantly higher wet season survival in wet compared to dry origin species in the dry site (Fig 4c, Table 2a), may have counterbalanced their lower survival during the dry season, resulting in no overall difference in establishment success during our study period (Table 2a).

The strength of the dry season (i.e. the duration and the water deficit reached) varies considerably across years, and consequently, dry season seedling performance also varies [50]. Pronounced seedling mortality, especially of drought sensitive species, occurs predominantly in particularly dry years [50], while the dry season in our study period was well within the long-term average [45]. Rather than contradicting the physiological tolerance hypothesis, our results, together with the previous studies, thus underline the importance of strong and repeated dry seasons, such as those occurring during El Niño Southern Oscillation (ENSO) events, for exclusion of wet origin species from dry forests.

Exclusion of dry forest species from wet sites

While the mechanisms underlying distribution limits of wet forest species to dry sites are quite well understood, the mechanisms excluding dry origin species from wet forests are not yet resolved. Contrary to our expectations, we found no indication of either high pest damage or low shade tolerance limiting the performance of dry origin species in the wet site.

The role of early life stages for species distributions

In this study, we focused on the initial life stages of germination and early seedling establishment, since these stages are considered the most vulnerable in the face of biotic and abiotic stressors [33, 34] and may thus be critical in shaping species distribution patterns across tropical rainfall gradients. However, germination (i.e. radicle emergence) patterns did not reflect the occurrence patterns of the species (Fig 4a, Table 2a and 2b), indicating that species partitioning along the rainfall gradient did not occur at this stage. We found support that differential dry season survival (in dry sites) and differential growth (in wet sites) during early life stages contribute to shaping tree distribution patterns across tropical rainfall gradients (see above). However, effects were weak and not sufficient to lead within the initial year to a clear home advantage of the species on the dry or wet side, respectively. Across a topographic moisture gradient, processes within one year after emergence were also insufficient to explain habitat preferences of adult plants [33]. This strongly suggests that longer time spans reaching into later life stages, and repeated and pronounced dry seasons are important for filtering tree distribution patterns.

The importance of later life stages and longer time periods for shaping distribution patterns is supported by local scale studies: If habitat associations of adults are shaped by failure to germinate or to establish, older juveniles and adults should exhibit the same habitat associations. However, most species have different associations at seedling and late life stages [56].