The Amazon harbors one of the most diverse tree floras on earth, and most species depend on mutualists for pollination and seed dispersal. This makes them susceptible to reproductive decline in fragmented forest because many of these mutualists suffer area-related extinction in fragments. It remains unknown, however, whether this highly biodiverse tree flora will reproduce and ultimately persist in fragmented forest. We conducted a two-year study of seed-fall in an experimentallyfragmented, highly-diverse Central Amazonian forest. We determined the effect of fragment size (1 ha, 10 ha, 100 ha and continuous forest control) on the density, species richness, functional diversity and functional composition of seeds separated into two datasets: dispersed tree seeds, and undispersed tree seeds. Our results show a 3x reduction in the density of undispersed, non-pioneer tree seeds in fragments of all sizes, indicating reduced seed production of the non-pioneer tree community. The density of dispersed tree seeds was 6x reduced in fragments of all sizes, while species richness was 6x reduced in 1 ha fragments and 3x reduced in 10 ha and 100 ha fragments compared to intact forest. This provides evidence of reduced community-wide seed dispersal, which became more pronounced with declining fragment size. The functional diversity (FRic) of dispersed tree seeds was 9.6x reduced in small fragments, and significant shifts in the functional composition for 8 of the 10 reproductive and ecological traits studied were identified, suggesting compromised ecosystem functioning. These functional compositional shifts provide evidence for disrupted mutualistic processes in fragments, which include loss of pollination by bees, especially small eusocial (meliponid) bees, and loss of dispersal by primates and large birds, which reduced the frequency of large-seeded tree species. Fragments also lost rare and mature-forest species, and collectively these changes suggest that future tree communities in fragmented Amazonian landscapes will retain a taxonomically- and functionallyimpoverished species pool with a biased functional composition unless efforts are undertaken to conserve dispersal by large frugivores and pollination by meliponid bees.

Experimental design

Plot (1 ha) establishment.— We established a total of 11 one hectare plots in the centers of fragments of different sizes (1 ha, 10 ha, and 100 ha) and intact forest controls. These plots were located at three cattle ranches (Dimona, Porto Alegre and Esteio), and two areas of continuous forest (Km 37 and Km 41). Three plots were established in three different 1 ha fragments at two ranches (Dimona and Porto Alegre), two plots were established in the centers of 10 ha fragments located at two different ranches (Esteio, Porto Alegre), two plots were established in the centers of 100 ha fragments located at two different ranches (Dimona, Porto Alegre), and four were located in continuous forest controls, two at each of the two different sites (Km 37 and Km 41). We therefore had a fully-replicated experimental design with respect to fragment size at the plot level. The number of plots we established in each fragment size was constrained by the number of isolated fragments that are available at the BDFFP; we utilized all the 100 ha fragments, two of the three 10 ha fragments, and all the 1 ha fragments available.

Seed trap establishment.— Seed traps were part of a broader experimental framework which included 10 m2 subplots established to document seedlings; each had an associated seed trap. For plots located in 1 ha fragments, a transect was extended perpendicular to the forest edge into the forest fragment, while for plots located in the centers of the 10 ha and 100 ha fragments and continuous forest, a transect was extended perpendicular from one randomly located plot boundary. In each plot, we placed eight 10 m2 seedling subplots (71 cm x 14.2 m) at 4 distances (13 m, 29 m, 52 m, and 81 m) from the fragment edge / plot boundary with two replicates at each of these distances (Fig. 1). Subplot placement at each distance was located randomly, with the constraint that the second replicate must be located at least 20 m from the first (so that canopies of different trees would be over each subplot, therefore making each a true replicate with regards to seed dispersal), and neither could be closer than 20 m from the side of the plot. One 0.5 m2 seed trap (71 cm x 71 cm, elevated 0.5 m above the ground) was located one meter from one randomly-chosen side of each of these seedling subplots. In total, 88 seed traps were utilized.

Data collection.—We visited each seed trap at approximately monthly census intervals for two years (2005 – 2006; 2007 – 2008). At each visit, we counted all seeds and determined their origin as either undispersed or dispersed. Each seed encountered was photographed, measured, and identified to species level, or given a morphospecies designation. For all analyses, we used the total number of seeds per seed trap, summed over all census intervals during the two years. More details on plot and seed trap placement, methodology to determine whether tree seeds were dispersed or undispersed and taxonomic identification are given in Appendix S2.

Seed rain functional classification.— All tree species were functionally classified, using ten traits describing key reproductive and ecological characteristics (Appendix S1). Reproductive traits included (i) reproductive strategy, (ii) pollination vector, (iii) bee-pollination vector (a subset of ii which delineates five bee guilds), (iv) seed size category, (v) seed dispersal syndrome, (vi) primary dispersal vector, and (vii) secondary dispersal vector. Ecological traits included (viii) successional status, (ix) adult abundance in intact forest (rarity), and (x) maximum adult tree height. For each of the 10 functional traits, the assignment of trait state for each species was based on data collected from seed traps (trait iv), BDFFP phytodemographic data (W. F. Laurance pers. comm.) (trait ix), and for all other traits we used published information, detailed in Appendix S3: Tables S1–S7. More details on functional classification are found in Appendix S2 and Appendix S3.

Functional diversity (FD) indices.— We used three complementary functional diversity (FD) indices, which together describe the distribution of species and their abundances within functional trait space. These indices include: 1) functional richness (FRic), the volume of functional trait space occupied by species in a community; 2) functional evenness (FEve), the evenness of species abundance distribution within functional trait space; and 3) functional divergence (FDiv), how far species abundances diverge from the center of functional trait space (Villéger et al. 2008). To compute these three FD indices, we used the dbFD function in the FD package in R (Laliberté and Legendre 2010). We pooled the data for the 8 seed traps found in each plot, and calculated these indices per plot. More details on functional diversity computation are found in Appendices S2 and S4.