Introduction

The destruction of tropical forests constitutes a massive extinction crisis in some of the world’s most biodiverse regions (Giam 2017; Keenan et al. 2015; Wright and Muller-Landau 2006). In the neotropics, deforestation and forest fragmentation are driven primarily by the demand for cattle pasture (De Sy et al. 2019). In Colombia, considered one of the world’s 12 “megadiverse” countries, this is no exception, but also driven by illegal mining, oil extraction, land appropriation, illicit crops, and expansion of urban centers (Armenteras et al. 2006; Etter et al. 2008; Romero et al. 2009). Despite being one of the few neotropical countries with a national restoration plan, Colombia is inarguably far from reaching a breakeven point as its annual deforestation rate is nearly 200,000 ha compared to approximately 5000 ha in regeneration (Ceccon and Martínez-Garza 2016; DANE and IDEAM 2021; MinAmbiente 2015). Reaching that breakeven point requires a multifaceted approach to restoration actions that is specific to land use histories, tailored to local socioeconomic needs, and well-informed by ecology and natural history.

Ecological restoration is a solutions-based approach to land degradation and forest fragmentation that in combination with the sciences of conservation and sustainability can protect and restore biodiversity while fostering increased local socioeconomic resilience (Gann et al. 2019). Within the ecological restoration toolbox there are two approaches for intervention: passive (also referred to as natural regeneration), which specifically involves stopping the source of disturbance, and active (also referred to as assisted regeneration), which involves the implementation of actions to reestablish or increase the abundance of organisms, species, or populations. Choosing between the approaches requires an evaluation of the potential for natural regeneration (Chazdon and Guariguata 2016). Within the active intervention toolbox, one common approach is planting trees to ameliorate seed dispersal and seedling establishment limitations following designs that enhance landscape connectivity (de la Peña-Domene and Martínez-Garza 2018; Rey Benayas et al. 2008). Crucial to the creation of these plantings is information on seed germination requirements of native tree species. With this study, we contribute ecological knowledge on 11 species native to the Colombian Amazon that have potential for restoration, and we evaluate how seed structural traits predict seed responses to pre-germination and light treatments as well as seedling performance.

Seeds of tropical species exhibit an extraordinary array of shapes, dormancy mechanisms, and longevities, which influence their effective propagation (Vázquez-Yanes and Orozco-Segovia 1993). This great variation in form and function is due to the predominance of biotic-dispersal of tropical tree seeds along with the history of coevolutionary forces that have shaped distinctive plant–animal relationships (Howe and Smallwood 1982). The variety of niches that plant species can fill throughout ecological succession is also a relevant factor (Rüger et al. 2020). Generally speaking, biotically dispersed seeds exhibit enhanced germination after gut passage, often as a consequence of the mechanical and/or chemical abrasion of the seed coat (Izhaki and Safriel 1990; Traveset 1998; Traveset et al. 2001), although some simply and only from the removal of the pulp (Jordaan et al. 2011). To mimic the beneficial effects of gut passage on germination, a range of mechanical, chemical, and hormonal pre-germination treatments have been developed with variable success and with substantial interspecific variation (Mousavi et al. 2011). Part of this variation is explained by how seed dormancy and germination are often also regulated by proteins and hormones in a seed’s light signaling pathway, which induce germination when access to light meets sufficient thresholds, again with great interspecific variation (Yang et al. 2020). Light-sensitive seeds generally are small and produced by species that fill niches with abundant light, such as those present in canopy gaps (Dalling et al. 2005), but small seeds that respond only to mechanical or chemical treatments have also been reported (Acuña and Garwood 1987). Moreover, some small-seeded species also occur in late-successional niches with low light (de Souza and Válio 2001). Since generalizations on the seed germination requirements of tropical species are seldom flawless, this presents a significant challenge for restoration projects that require tree planting.

Having accurate predictors of germination requirements and seedling performance would improve the likelihood of success during the early stages of restoration, and these predictors could be functional traits that are part of a species’ strategy for growth, reproduction, and survival (Violle et al. 2007). For example, for planted seedlings in a restoration experiment in Veracruz, México, leaf dry mass content was a good predictor of seedling survival in pastures (Martínez-Garza et al. 2013). For germination and seedling performance, potential seed functional traits include but are not limited to seed size, seed mass, and seed coat thickness. Small-seeded species tend to have higher germination percentages and rates and greater RGRs than larger-seeded species (Murali 1997; Poorter and Rose 2005), but here too are there exceptions to the rule (Deb and Sundriyal 2017). As for seed coats, a thick seed coat is better adapted for passage through the digestive tract than a thin one. (Traveset 1998). Thus, thick-coated species are more likely to benefit from pre-germination treatments than thin-coated seeds. Larger seeds tend to also have thicker seed coats and are less dependent on light for germination than small seeds (Milberg et al. 2000; Pearson et al. 2002).

Here, we measured six seed functional traits and evaluated their potential as predictors of germination, germination requirements, and RGRs of 11 species. In addition, we grouped these species by their functional traits using cluster analysis to determine which if any generalizations can be made. Seeds were given pre-germination treatments (hot water immersion, sandpaper abrasion, or control) and then incubated in full, partial, or no light. We predicted that functional groups would vary in their response to pre-germination and light treatments, with small-seeded functional groups responding the best to increased access to light, and large-seeded functional groups responding the best to pre-germination treatments. We also predicted that germination metrics (germination percentage, velocity of germination, and germination vigor) and RGRs would correlate negatively with seed size.

Methods

Study species

The 11 species selected for this study are all native to the humid tropical forests department of Caquetá, Colombia, which is part of the Andean-Amazon piedmont region. Species were selected based on socioecological value and sufficient seed availability and include: Miconia minutiflora Bonpl. (Melastomataceae), Miconia tomentosa Rich. (Melastomataceae), Vismia baccifera L. (Hypericaceae), Piper aduncum L. (Piperaceae), Piptocoma discolor Kunth (Asteraceae), Ochroma pyramidale Cav. ex Lam. (Malvaceae), Apeiba membranacea Spruce ex Benth. (Tiliaceae), Cecropia engleriana Snethl. (Urticaceae), Ormosia nobilis var. santaremnensis Tul. (Fabaceae), Cedrelinga cateniformis Ducke (Fabaceae), and Minquartia guianensis Aubl. (Oleaceae).

Experimental design

Seeds used in this study were collected from fruiting trees growing in a set of chronosequence plots in Caquetá, Colombia established in 2017 by the Amazonian Scientific Institute (SINCHI). For each species, a total of 810 seeds were collected and divided up into 27 aluminum germination trays (30 seeds per tray) except for M. guianensis (15 seeds per tray) and C. cateniformis (6 seeds per tray) due to seed scarcity. The germination experiment was set up as a 3 × 3 factorial design using two independent variables (light and pre-germination treatments) with three levels each creating nine variable combinations. Each variable combination had three replicates (9 × 3 = 27 trays per species). For pre-germination, seeds were either placed in a hot water bath at 80 °C for 1 min, abraded using sandpaper to wear off the seed coat, or left without pre-germination treatment (control). However, abrasion by sandpaper was only possible for four species (A. membranacea, O. pyramidale, O. nobilis, and C. engleriana) due to seeds being too small, too thin a seed coat, or having a porous seed coat. Seeds were slightly buried in peat substrate and placed in an incubator (Climacell 707 EVO) set to 86% relative humidity, a photoperiod with 12 h of light per day, and alternating temperatures of 25 °C, 30 °C, and 33 °C every eight hours, with the higher temperatures occurring during light hours. Light treatments were created from seed tray placement in the incubator; trays closest to LED lights (36 W) on the highest level belonged to the full light treatment while trays farthest from the light on the bottom level belonged to the no light treatment, and those in between belonged to the partial light treatment. Seeds were watered with distilled water every 2 days.

As functional traits, we measured length (mm), width (mm), height (mm), wet mass (mg), dry mass (mg), and seed coat thickness (mm) from an additional 216 seeds. For small-seeded species (M. minutiflora, M tomentosa, V. baccifera, P. aduncum, P. discolor, and C. engleriana), seeds were weighed together by tray and divided by the number of seeds. Seed coat thickness was measured using a handheld digimatic QuantuMike Mitutoyo micrometer with 0.001 mm resolution. Prior to measurement, seeds were cut longitudinally and had their endosperm and embryo extracted to isolate the seed coat. A seventh trait, seed moisture content, was determined by weighing seeds before and after they were dried in a Riossa® oven at 130 °C for 17 h in, but the precision of measurements for small-seeded species was limited, so moisture content was excluded from analyses.

Data collection

For each species, we calculated germination percentage (PG%) which corresponds to the percent of germinated seeds, coefficient of velocity of germination (CVG) which corresponds to the average number of seeds that germinated within a 20-day period, and germination vigor (GV) which corresponds to the percent successful establishment of seedlings. Seedlings were harvested after 15 days of growth (15 plants per species, five from each light treatment) to measure their final seedling length, stem length, root length, and stem diameter (all mm). We then calculate relative growth rates (RGRs) for each of these measurements by dividing the measured values by 15. For reasons unknown, RGR data for M. tomentosa were missing at the time for data analysis. Since plants were haphazardly selected for harvest, we do not analyze the effect of pre-germination or light treatment on RGRs.

Statistical analysis

To group the 11 species into functional groups we used a hierarchal cluster analysis with Ward’s method on a Gower’s similarity index of the six seed structural traits (Gower 1971). Three functional groups were determined, largely by their seed size, and named as such: “Small-Seeded Functional Group”, “Medium-Seeded Functional Group”, and “Large-Seeded Functional Group”, or SSFG, MSFG, and LSFG, respectively (Fig. 1). To compare functional groups by their seed functional traits. we used Kruskal–Wallis and Conover post-hoc tests; non-parametric tests were used here as none of the seed functional traits showed homoscedasticity. We also used a Principal Component Analysis (PCA) to visualize the partitioning of the three functional groups by their seed structural traits. To evaluate the effects and interactions of light intensity and pre-germination treatments on the germination metrics (PG%, CVG, and GV), we used linear and generalized models for each response variable. For PG%, we used a linear model. For CVG and VM, we used generalized models using the gamma family (logit link for CVG, inverse for GV) where we added a constant of 0.05 to all values as the gamma distribution does not accept non-positive numbers. When interactions were significant, we created new models to evaluate effects by factor level. p-values obtained by installing the package ‘lmerTest’ (Bates et al. 2015).

Fig. 1
figure 1

Hierarchal clustered dendrogram built using method on a Gower’s similarity index of the 11 species’ seed structural traits. Three functional groups were determined largely by traits of seed size and named thusly

Full size image

To determine if the seed functional traits influenced germination and relative growth rates, we used principal component regression (PCR) with linear models. We chose PCR as a method to deal with the high collinearity between the seed structural traits (VIF < 10). Since they come from separate datasets, tray germination metrics were randomly matched to individual seeds and their corresponding trait values for each species to evaluate the relationship between traits and germination. For RGRs, because seedlings were not harvested from each tray, there was no direct correspondence between a seed tray and RGRs; so we calculated averages for each principal component and RGRs by species and light treatment to obtain a single dataframe on the same scale. Pearson’s correlations were calculated to evaluate how the response variables correlated with the principal components. All statistical tests were carried out using R (R Core Team 2021).

Results

Seed functional traits

The three functional groups differed significantly in their seed structural traits, supporting the separation into three groups by the hierarchal cluster analysis (Kruskal–Wallis: p < 0.001; Fig. 2). Post hoc Conover tests showed statistically significant differences all around (p < 0.05) except for seed height (mm) between MSFG and LSFG (t = 0.715, p = 0.475) and a marginally significant difference in seed coat thickness between SSFG and MSFG (t = 1.69, p = 0.93). The large-seeded functional group represented by a single species, M. guianensis, showed the highest values for all six structural traits, followed by MSFG and lastly by SSFG with the smallest values.

Fig. 2
figure 2

Mean structural traits by functional group (SS = Small-Seeded Functional Group, MS = Medium-Seeded Functional Group, LS = Large-Seeded Functional Group). Error bars represent standard error. Letters above error bars represent statistically significant differences from Conover post-hoc tests (p < 0.05). Letters above error bars paired with an interpunct (2F) represent marginally significant differences (0.05 < p < 0.1)

Full size image

Germination experiment

We found a significant interaction between functional groups and pre-germination treatment on all three response variables (PG%: F2,219 = 2.142, p = 0.096; VG: χ2 = 27.278, p < 0.001; GV: χ2 = 24.007, p < 0.001; Fig. 3) and no significant interactions between functional group and light treatment (PG%: F2,219 = 1.187, p = 0.318; VG: χ2 = 2.155, p = 0.707; GV: χ2 = 3.792, p = 0.435).

Fig. 3
figure 3

Mean germination percentage, germination velocity, and germination vigor by functional group and pre-germination treatment. The large-seeded functional group lacked an abrasion treatment. Error bars represent standard error. Letters above bars represent statistically significant differences (p < 0.05). Letters with interpuncts (·) represent marginally significant differences (0.05 <  p < 0.1)

Full size image

For germination percentage, pre-germination treatment had a significant effect for the MSFG (F2,36 = 9.734, p = 0.001) and marginally significant effect in SSFG (F2,162 = 2.406, p = 0.093). Within ES, abrasion showed a significantly higher germination percentage than hot water (t = 1.99, p = 0.048) or control groups (t = 1.984, p = 0.049). Within MSFG, the control group showed a lower germination percentage than the ones that received abrasion (t = 4.176, p < 0.001) and hot water treatments (t = 2.93, p = 0.005). Abrasion showed a marginally higher germination percentage than the hot water treatment (t = 1.784, p = 0.082). Evaluated by pre-germination treatment, within the control group the SSFG group had a higher germination percentage (vs MSFG: t = 4.293, p < 0.001; vs LSFG: t = 2.237, p = 0.028). MSFG and LSFG did not differ in germination percentage (t = 0.617, p = 0.028; Online Resource 1a). Light treatments also differed significantly in germination percentage (F2,219 = 6.299, p = 0.002). Germination percentage was lowest when there was no light (vs partial: t = 2.751, p = 0.006; vs full: t = 3.103, p = 0.002) but did not differ between partial and full light treatments (t = 0.352, p = 0.725; Fig. 4A).

Fig. 4
figure 4

Mean a germination percentage, b germination velocity, and c germination vigor by light treatment. Functional groups did not differ in their germination response to light treatments and are, therefore, pooled together (ANCOVAs: p < 0.1). Error bars represent standard error. Letters above bars represent statistically significant differences (p < 0.05)

Full size image

For the velocity of germination, pre-germination treatment had significant effects in MSFG (χ2 = 85.733, p < 0.001) and LSFG (χ2 = 9.046, p = 0.003) and marginally significant effects in SSFG (χ2 = 4.936, p = 0.085). Within ES, abrasion had higher germination velocity than control (t = 2.115, p = 0.036) but neither differed from the hot water treatment (t = 1.43, p = 0.155 and t = 0.927, p = 0.355, respectively). Within MSFG, abrasion and hot water showed no difference in germination velocity (t = 0.221, p = 0.826) but both had higher values than the control (t = 8.596, p < 0.001 and t = 10.798, p < 0.001, respectively). Within LS, the control showed a higher germination velocity (x̄ = 0.74) than the hot water group (x̄ = 0.33; t = 3.040, p = 0.008). The large-seeded functional group did not include an abrasion pre-germination treatment. Evaluated by pre-germination treatment, velocity of germination in control and abrasion treatments was lower between SSFG and MSFG (t = 3.04, p = 0.03; t = 4.591, p = 0.039, respectively; Online Resource 1b). Light treatments did not have a significant effect on velocity of germination (χ2 = 0.839, p = 0.657; Fig. 4B).

For the germination vigor, pre-germination treatment had significant effects in MSFG (χ2 = 85.015, p < 0.001) and LSFG (χ2 = 11.087, p = 0.003) and marginally significant effects in SSFG (χ2 = 4.696, p = 0.096). Within ES, abrasion had marginally higher germination vigor than hot water (t = 1.864, p = 0.064) and control groups (t = 1.837, p = 0.068). Abrasion and hot water did not differ (t = 0.46, p = 0.646). Within MSFG, hot water resulted in the highest germination vigor (vs abrasion: t = 2.744, p = 0.009; vs control: t = 12.056, p < 0.001) followed by abrasion (vs control: 7.1, p < 0.001). Within LS, the control showed higher germination vigor than the hot water group (t = 3.45, p = 0.003). Evaluated by pre-germination treatment, germination vigor within the abrasion treatment was marginally higher in SSFG over MSFG (t = 4.591, p = 0.067; Online Resource 1c). Light treatments did not have a significant effect on germination vigor (χ2 = 1.158, p = 0.56; Fig. 4C).

Species-specific responses on germination percentage

Within SS, species differed in how germination percentage was affected by pre-germination and light treatments (F9,114 = 4.979, p < 0.001; F14,114 = 3.801, p < 0.001). Within MSFG, species differed in how they responded to pre-germination (F2,30 = 16.426, p < 0.001) but not to light treatments (F2,30 = 1.418, p = 0.258). A species-specific list of the conditions that elicited the highest germination percentage for each species is provided in Table 1. A full breakdown of species-specific treatment differences and pairwise comparisons within treatments is provided in Online Resource 3.

Table 1 Species-specific conditions that elicited the highest germination percentage for each species. Species within functional groups differed in their response to pre-germination and light treatments
Full size table

Principal component analysis

In the PCA, PC1 accounted for 74.24% of the variation and placed SSFG and LSFG at opposite sides and MSFG in the middle (Fig. 5). The highest eigenvectors in PC1 were length (− 0.46), width (− 0.44), height (− 0.41), wet mass (− 0.42), and dry mass (− 0.42). PC2 explained an additional 16.5% of the variation with seed coat thickness having by far the highest eigenvector (0.922). The MSFG functional group exhibited the greatest range in PC2 largely due to the difference in seed coat thickness between the two MSFG species, C. cateniformis and O. nobilis. Similarly, LSFG exhibited the least range in PC2 due to inappreciable variance (se = 0) in the seed coat thickness of the sole species that represents it, M. guianensis.

Fig. 5
figure 5

Principal Component Analysis for the three functional groups created using six seed structural traits (seed length, seed width, seed height, wet mass, dry mass, and seed coat thickness). Points represent mean values from seeds belonging to 11 species organized into three functional groups. Ellipses around functional groups represent 95% confidence intervals. Distance between points in the multidimensional space represents trait similarity

Full size image

Principal component regression on germination and relative growth rates

For the PCR, we used as predictor variable the first components of the PCA, which accounted for most of the cumulative variation (74.24%). For interpretation and based on the eigenvector values (Table 2), PC1 is understood to represent the seed structural traits of seed length, seed width, seed height, wet seed mass, and dry seed mass. PC2 largely represented seed coat thickness and accounted for an additional 16.5% of the variation. However, to obtain the predictor value of seed coat thickness by itself, and because seed coat thickness is not as correlated with other seed functional traits, we used the seed coat variable itself as a predictor. To prevent pseudoreplication due to negligible variance in seed coat thickness values within species we split the dataset by pre-germination treatment, calculated average values by species, and applied Pearson’s correlations.

Table 2 Eigenvalues, cumulative percent variation, and eigenvectors of the six principal components (PCs) for the six seed structural traits of the three functional plant illustrated in Fig. 2
Full size table

For germination percentage, PC1 showed a significant but weak and negative correlation (t = 3.148, p = 0.002, r = − 0.221; Fig. 6A). Light and pre-germination treatments accounted for 0.17% and 7.76% of the residual variance, respectively. For germination velocity, PC1 showed a significant but weak and negative correlation (t = 2.714, p = 0.007, r = − 0.192; Fig. 6B). Light and pre-germination treatments accounted for 0% and 7.16% of the residual variance, respectively. For germination vigor, PC1 showed a marginally significant, negative, and weak correlation (t = 1.912, p = 0.057, r = − 0.134; Fig. 6C). Light and pre-germination treatments accounted for 0% and 3.07% of the residual variance, respectively. Seed coat thickness did not correlate significantly with the germination metrics (PG%: t = 0.853, p = 0.416, r = − 0.274; CVG: t = 0.642, p = 0.537, r = − 0.209; vs. GV: t = 0.441, p = 0.67, r = − 0.145). Split by pre-germination treatment, there were no statistically significant or marginally significant correlations (p < 0.1).

Fig. 6
figure 6

Mean a germination percentage, b germination velocity, and c germination vigor by principal component 1 of the seed structural trait PCA. Points represent values from 216 germination trays. Solid lines indicate statistically significant correlations (p < 0.05). Double-dashed lines indicate marginally significant correlations (0.05 <  p < 0.1)

Full size image

PC1 showed a moderately strong and positive correlation with RGRTotal Length (t = 6.167, p < 0.001, r = 0.594; Fig. 7A), RGRStem (t = 2.902, p = 0.009, r = 0.537; Fig. 7B), and a weak but still positive correlation with RGRRoot (t = 2.927, p = 0.008, r = 0.477; Fig. 7C). PC1 did not correlate significantly with RGRDiameter (t = 1.662, p = 0.111, r = 0.311; Fig. 7D). Seed coat thickness did not correlate significantly or marginally significantly with any of the RGRs (vs. total length: t = 0.536, p = 0.620, r = 2.59; vs. stem: t = 1.721, p = 0.160, r = 0.652; vs. root: t = 1.286, p = 0.268, r = 0.541); vs. diameter: t = 1.26, p = 0.277, r = 0.532).

Fig. 7
figure 7

Relative growth rates for a total seedling length, b stem length, c root length, and d stem diameter by principal component 1 of the seed structural trait PCA. Points represent values from seedlings harvested after the germination experiment. Solid lines indicate statistically significant correlations (p < 0.05). Dashed lines indicate non-significant correlations (p < 0.1)

Full size image

Discussion

The 11 tropical tree species included in this study clustered into three groups, driven chiefly by highly correlated variables of seed size and mass. Despite grouping, species within functional groups differed in how they responded to pre-germination and light treatments. The multivariate trait axis of seed size and seed mass measurements (PC1) proved to be strong predictors of germination metrics and relative growth rates.

Seed functional traits and pre-sowing treatments

Strong correlations among our measured seed functional traits are not surprising—seeds tend to be heavier with an increase in size, and large seeds tend to have thicker coats than small seeds (Pearson et al. 2002). However, as previous studies have extensively documented, seed size and its correlated variables are not sufficient to predict the best pre-germination treatment because seed dormancy can occur for a variety of reasons across seeds of similar size (Ashraf and Foolad 2005; Mousavi et al. 2011; Zalamea et al. 2018). The only generalization backed by previous studies is that small-seeded species are more dependent on light for germination than large-seeded species (Milberg et al. 2000). In our study, functional groups did not differ in their germination response to light treatments. However, when split by functional groups, the SSFG group did show lower germination percentage in the absence of light while MSFG and LSFG did not (Online Resource 2). Not all SSFG species responded to the light treatment, suggesting exceptions to the rule, but in the case of C. engleriana, it was the access to light that permitted the abrasion and hot water treatments to influence germination percentage (Online Resource 3). Vismia baccifera and O. pyramidale also showed higher germination percentage when seeds had access to partial or full light. Lack of response to light by other small-seeded species reflects either the absence of the proper pre-germination treatments or lack of photosensitivity.

Responses to pre-germination treatment in other species varied greatly. For ES, abrasion yielded higher germination metrics than the control, but our conclusions on abrasion are limited by the minority of species that received this treatment—only C. engleriana actually showed a significant difference between abrasion and control, and only when seeds were under partial or full light. Hot water and control treatments showed similar germination percentage, but split by species, hot water yielded higher germination percentage for O. pyramidale and a lower germination percentage for M. minutiflora. The congeneric M. tomentosa showed similar germination percentages between hot water and control groups, underscoring the challenges associated with predicting dormancy-breaking mechanisms from functional seed traits. In MSFG it is clear that pre-germination treatments increase germination, but the two species differed starkly in which method actually had an effect (hot water for C. cateniformis and abrasion for O. nobilis). In LSFG, germination of seeds treated with hot water did not differ much from the control group, but since LSFG in our study is represented by a single species, we cannot extrapolate beyond M. guianensis. Similarly, as MSFG is represented by two species, we are limited in how their responses to pre-germination and light treatments can be extrapolated. Future studies that attempt to use functional traits to predict germination requirements should select traits based on the different kinds of dormancies that are present in seeds.

Predicting germination and relative growth rates

The principal component regression on germination metrics showed that germination decreased with increasing seed size and mass, supporting our hypothesis. This result is supported by previous studies showing small seeds germinating faster than large seeds (Murali 1997). High germination percentages and rates are an adaptation often seen in gap-specialist and early-successional species (Denslow 1980). Since small-seeded species also reach reproductive maturity earlier and have higher annual seed output than large-seeded species (Moles et al. 2004), small seeds tend to characterize the seed bank in canopy gaps (Dalling et al. 1998). The tradeoff with seed size becomes apparent during seedling establishment when seedlings of small-seeded species exhibit a higher mortality rate than larger-sized species (Moles and Westoby 2004). In our study, all small-sized species are or belong to genera that predominate with early-successionals, but not all small-sized species are gap-specialists, early-successionals, or light-demanding species (Dalling et al. 2005). Thus, while seed functional traits of size and mass present a good model for predicting germination, exceptions to the model should be anticipated.

The principal component regression on RGRs showed that RGRs increased with increasing seed size and mass (except for in stem diameter), contradicting our hypothesis. A meta-analysis of 23 studies conducted by Poorter and Rose (2005) found that RGRs and seed mass correlated negatively, emphasizing the shade-tolerance and slow-growth strategy of large-seeded species. However, the authors point out that part of the slow-growth strategy of large-seeded species rests on their hypogeal germination, where cotyledons remain belowground (Ibarra-Manríquez et al. 2001). In our study, the species with largest seed size (M. guianensis) had cryptocotylar epigeal germination with reserve storage cotyledons, which while not photosynthetic still differs from other large-seeded species. Other studies have also reported a fast-growth rate for M. guianensis in its other life stages (Hunter 1991; Nebel 2000). Minquartia guianensis might be an exception to the rule described by Poorter and Rose (2005).

Experimental limitations and errors

Our study is limited to 11 tree species from one of the most biodiverse regions in the country and the world (Romero et al. 2009), so extrapolations from our results should be done with care. Further details should be taken into consideration: (i) conclusions on the effect of manually abrading the seed coat with sandpaper are limited to the species this treatment was applied to, namely: Apeiba membranacea, O. pyramidale, O. nobilis, and C. engleriana. Future studies should attempt physical scarification using heated needles, small cement mixers filled with rocks, or other suggested methods (Ashraf and Foolad 2005; Rudolf and Owston, 2003). (ii) Despite the use of a micrometer, the lack of intraspecies variance in seed coat thickness suggests that this tool was not used to its most accurate potential. We refer to Pearson et al. (2002) for more accurate reports of seed coat thickness for some of our study species. (iii) Measurements of seed mass for small-seeded species are estimates obtained by measuring multiple seeds at once on a mass balance and dividing by seed abundance—due to the lack of accuracy, we did not use seed moisture content as one of our functional seed traits despite its potential as predictor of seed and seedling performance. (iv) Seeds were not tested for viability prior to germination trials. (v) The selection of seedlings for harvest, while balanced by light treatment (5 each), was not randomly conducted, preventing us from evaluating the effect of any of the pre-sowing treatments on RGRs. (vi) RGRs were calculated after 15 days of growth and therefore only represent early seedling growth.

Potential for restoration

Species selection for restoration purposes should take into consideration: (i) seed availability and available pre-sowing treatments, (ii) species abilities to ameliorate dispersal and establishment limitations, and (iii) socioeconomic value or ecosystem services. From our study, we can make some recommendations for restoration in the Andean-Amazon piedmont region of Colombia.

Cecropia engleriana (SSFG) is a medium-sized tree species (15 m tall) that reaches reproductive maturity early. Cecropia trees attract a wide variety of generalist dispersal agents, so including them in restoration greatly alleviate dispersal limitation early on (Howe 2016, 2017; Martínez-Garza et al. 2013). Cecropia spp. do not provide much canopy cover and thus are unlikely to shelter recruiting seedlings from desiccation, so we recommend planting it in conjunction with fast-growing tree species that yield dense and wide canopies. For germination, we recommend plenty of light and a pre-germination treatment (hot water or abrasion).

Ochroma pyramidale (SSFG) is a fast-growing, evergreen canopy tree species (30 m tall). As a wind-dispersed species, this species’ fruit does not provide sustenance for dispersers, but its flowers are consumed by members of the Psittacidae family (Lee et al. 2014). In addition, its fast growth and broad canopy shelter recruiting seedlings and prevent the establishment of recalcitrant understory layers by invasive species (Douterlungne et al. 2013). Wood harvested from O. pyramidale has a myriad of commercial uses in Colombia due to its light weight, resilience, and floatability (Fuentes and Arango 2005). We recommend planting in conjunction with Cecropia to provide shelter for recruiting seedlings. Seeds of this species germinate well in full light after a hot water treatment.

Apeiba membranacea (SSFG) is a fast-growing, medium-sized tree species (15 m tall). Saplings of A. membranacea are hardy, capable of establishing successfully in soil disturbed by petroleum extractions (Villacís et al. 2016). The fruits of A. membranacea and related species are consumed and dispersed by monkeys, highlighting its potential for reducing dispersal limitation by larger, arboreal animals (Stevenson 2005; Wehncke et al. 2003). Use for lumber from A. membranacea is limited due to its low wood density but is still used for boards and general carpentry (Fuentes and Arango 2005; Montero et al. 2016). Seeds are harvested and used as brilliantine and hair tonic due to their oil content (Salazar and Soihet 2001). This species germinates fairly well regardless of pre-sowing treatment, a pattern also found by Pearson et al. (2002). In fact, its seeds are capable of germinating within the fruit capsule (Montero et al. 2016). We recommend planting to attract a greater diversity of dispersal agents and maintain a mid-canopy layer.

Cedrelinga cateniformis (MSFG) is a slow-growing, canopy tree species (50 m tall). Like A. membranacea, this species’ saplings establish and grow well in soil disturbed by petroleum extractions (Villacís et al. 2016). As a legume, C. cateniformis can help restore soil nutrients through its symbiosis with nitrogen-fixing, Gram-negative Rhizobium bacteria (Phillips 1980). Cedrelinga cateniformis is used for general construction, plywood, boxes, and more (Fuentes and Arango 2005). Indigenous people of the Colombian Amazon use the wood to make canoes and the bark to extract camphor, a terpenoid with medicinal and pest deterrent properties. Although wind-dispersed, its seeds are consumed by spider monkeys (Saimiri spp.) and Psittacids (Montero et al. 2016). We recommend the use of a hot water treatment to germinate this species’ seeds and planting it in conjunction with fast-growing species.

Miconia minutiflora (SSFG) is a shrub and small tree species (15 m tall) that reaches reproductive maturity early. This species is known for its “big bang flowering” where shrubs flower synchronically for 2–3 days attracting a myriad and overabundance of pollinating insect species (Mori and Pipoly 1984). Its fruits are consumed and seeds are dispersed by a variety of small birds and mammals (Messeder et al. 2021). Miconia minutiflora has been used traditionally as a medicinal plant, with a recent study showing how its leaf methanol extract has anti-inflammatory and antinociceptive properties (Gatis-Carrazzoni et al. 2019). We recommend germinating without pre-germination treatment and planting to attract dispersers and pollinators. Sautu et al. (2006) recommend simply separating seeds from the fleshy fruit under water.

Miconia tomentosa (SSFG) is a shrub and small tree species (13 m tall) that reaches reproductive maturity early. Fruits of M. tomentosa are consumed and presumably dispersed by the critically endangered Caquetá Titi Monkey (Plecturocebus caquetensis) (Acero-Murcia et al. 2018). The fruit of M. tomentosa is considered an unconventional food plant (UFP) in Caquetá, which describes native, underutilized or not cultivated species with potential as a food resource (Cárdenas et al. 2012). We recommend germinating with plenty of light and planting it to provide food for the Caquetá Titi monkey and other native species.

Minquartia guianensis (LSFG) is a canopy tree species (30 m tall) with unusually fast growth (Nebel 2000). Faster growth rates are also reported in saplings and young trees relative to a pioneer tree species (Hunter 1991). In our study, seedlings of this species also had the fastest growth. The fruit of M. guianensis is dispersed by monkeys (Alouatta spp. and Aletes spp.), pacas (Cuniculus paca), and agoutis (Dasyprocta punctata), highlighting its potential for reducing dispersal limitation by mammals (Hunter 1991; Julliot, 1997). The fruit of M. guianensis is considered an UFP (Cárdenas et al. 2012). Timber from this species has a long history of use in construction and the tree canopy is described as evergreen, full, and bushy that protects seedlings from desiccation (Hunter 1991). In our study M. guianensis showed higher germination velocity and vigor in the control treatment compared to seeds given a hot water treatment, but the difference between the two is small. According to Flores (1994), as long as humidity is high and stable, seeds will germinate in both shade and direct sunlight.

Ormosia nobilis (MSFG) is a canopy tree species (30 m tall). The Ormosia genus has seeds with a bright red coat that deceives avian dispersers into ingesting them, mistaking them for a fleshy fruit (Foster and Delay 1998). Some propose that rather than deception there is mutualism, whereby large-bodied birds purposefully consume Ormosia seeds as gizzard grid substitutes (Peres and van Roosmalen, 1996). Seeds are also used by the Emberá-Katío indigenous group of Caquetá for handicrafts (Frausin et al. 2008). As a legume, O. nobilis can help restore soil nutrients through its symbiosis with N-fixing bacteria (Phillips 1980). We recommend germinating using abrasion and planting this species to attract seed-dispersing fauna and to nitrify soils.

Piper aduncum (SSFG) is a shrubby tree species (7 m tall) with fast-growth that reaches reproductive maturity early. Seeds of this species germinate well regardless of pre-sowing treatment, which likely facilitates its successful invasion of the forests of Papua New Guinea (Hartemink 2010). Piper aduncum is dispersed by a variety of bird and bat species (Hartemink 2010). Essential oils extracted from P. aduncum have antifungal properties and impressive repellent activity against mosquitos (Aedes aegypti) (Guerrini et al. 2009; Misni et al. 2008). While its RGR was low in our experiment, more long-term experiments show 0.33 cm d−1 growth rates for 2-month-old saplings (Susanto et al. 2018). We recommend planting this species to attract dispersers early on.

Piptocoma discolor (SSFG) is a fast-growing canopy tree species. This species showed the lowest germination percentages and was largely unresponsive to pre-sowing treatments. However, a study by Mendoza et al. (2005) showed a 95% germination without any pre-germination treatment within 30 days. Furthermore, a study by Stimm et al. (2008) found that P. discolor seedlings take almost a quarter of a year to emerge. It is likely that seeds in our study were viable but required more time for their germination. In Ecuador there is a strong market for the timber from this species to make pallets and crates for fruit transport and could be potentially developed in Caquetá (Erazo et al. 2013). We recommend germinating P. discolor seeds with the expectation that they will take a long time to germinate and planting it to secure a full canopy.

Vismia baccifera (SSFG) is a small tree and shrub species (14 m tall) that reaches reproductive maturity early. Its fruit is dispersed by spider monkeys and frugivorous bats (Chapman 1989; Kelm et al. 2008). Vismia baccifera is used in the Amazon as a medicinal plant and recent studies have unraveled how its leaf extract can be used to treat tumor growths (Trepiana et al. 2018). Its lumber is also used for general carpentry and construction (Bonilla Céspedes and Calderón Falla 2016). We recommend germinating V. baccifera with plenty of light and planting it as a mid-canopy species.

Conclusion

Seed size and seed mass are good indicators of germination performance and seedling growth, but the positive correlation with RGR is likely due to the unusual seedling growth of M. guianensis as large-seeded species typically show slow-growth rates (Poorter and Rose 2005). The tradeoff in RGR and germination observed between M. guianensis and the smaller-seeded species is possibly explained by differences in the resources stored in the seeds and the type of germination (hypogeal vs epigeal). Stakeholders in restoration using these species should consider these tradeoffs during seeding and seedling growth stages of restoration. Pre-sowing treatment effects differed greatly among species with similar seed size and seed coat thickness, indicating that for the purpose of restoration, the application of pre-sowing treatments to seeds should not be determined based on these traits.

Author contribution statement

MNNG, CHR, and LLRF contributed to the study conception and design. Material preparation, data collection, and data analysis were conducted by MNNG for her Master’s thesis. LCBL revised the data analysis for publication and drafted the manuscript. LLRF and CHR provided comments on previous versions of the manuscript. All the authors read and approved the final manuscript.