The degradation and fragmentation of tropical forests, along with their conversion to production landscapes, is driving widespread defaunation (Canale et al., 2012; Dirzo et al., 2014) with large frugivorous birds particularly affected (Bovo et al., 2018; Bregman et al., 2014). This loss of species can disrupt seed-dispersal networks (da Silva & Tabarelli, 2000), thereby limiting forest regeneration (Galetti et al., 2013; Gardner et al., 2019). Defaunation alters assemblages of frugivores in disturbed forests, with few species occurring in adjacent non-forest landscapes (Bregman et al., 2016). This potentially impedes recolonisation by tree species that have become extirpated, reducing ecosystem resilience and raising the cost of reforestation programmes (Chazdon & Uriarte, 2016). Few studies, however, have assessed how habitat degradation affects potential seed transfer from intact tropical forests to adjacent cleared areas by examining frugivore communities.

One possibility is that many seed-dispersers associated with intact forests rarely if ever visit non-forest habitats, such that species transferring seeds into matrix habitats are primarily those associated with degraded forests and forest edges. If this is the case, tree species recruitment will be adversely affected because these frugivorous birds are a non-random subset in terms of key ecological traits, such as beak or gape size (Bovo et al., 2018). In particular, the absence of species with larger beaks and wider gapes may constrain seed transfer because the ability of avian frugivores to disperse seeds is limited by gape width, which places an upper physiological constraint on the size of seeds they can ingest whole 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 and hence which tree species can be dis persed (Burns, 2013 ; Wheelwright, 1985). This traitmatching between avian frugivores and their food plants offers a useful tool for assessing the structure and function of mutualistic interaction networks (Dehling et al., 2014; McFadden et al., 2022) and their response to environmental change (Schleuning et al., 2020).

In accordance with the trait-matching hypothesis, the local extinction of large-gaped birds has been shown to impair dispersal of larger seeds (da Silva & Tabarelli, 2000; Galetti et al., 2013). These local extinctions of key seed dispersers may lead to declining plant populations, particularly for large-seeded tree species, resulting in changes to recruitment and composition (Sethi & Howe, 2009). Movement constraints on seed-dispersers may also cause the rewiring of interaction networks between seed dispersers and their food plants, with unpredictable outcomes driven by species-specific responses to different land covers (Habel et al., 2019; Rehm et al., 2017). In effect, the type of seeds dispersed between different land cover types will be governed by the traits of frugivores that can persist in or routinely disperse through both forest and non-forest landscapes. In addition, this process of seed transfer may involve an extra step from intact to degraded forest and then to matrix simply because intact forests are usually separated from the matrix by an intervening buffer of degraded or secondary forests (Mayhew et al., 2019; Nunes et al., 2022). To determine the effects of disturbance on seed dispersal, two key questions must be answered. First, given that the potential to disperse different seeds is dependent on species traits, it is important to examine the functional diversity (FD) of the bird species assemblage. FD metrics use traits to quantify the potential roles played by species in a given community (Cadotte et al., 2011; Petchey & Gaston, 2002). One way to examine FD is with a volumebased measure (Mason et al., 2005) with morphological traits providing a representation of ecological niche space (Pigot et al., 2020; Tobias et al. 2020). If functionally unique species are lost, the functional volume is reduced, indicating the likely impairment of seed dispersal services for certain plant species. If, however, the species lost are not functionally unique, 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 this can still result in reduced functional redundancy, i.e., where different species perform similar roles, species loss does not necessarily reduce FD but nonetheless leads to an increase in future risk by leaving seed dispersal service reliant on fewer species (Biggs et al., 2020). Second, it is important to understand how frugivores use the modified landscape. If certain species or traits are restricted to a single land cover type, seed dispersal between different land covers will also be restricted, as birds are not able to act as “mobile links'' (Sekercioglu, 2006), which will in turn reduce the potential for forest regeneration. In this study, we aim to investigate how land-use change and habitat degradation affects the potential for natural seed dispersal by assessing changes in key traits and FD in assemblages of fruit-eating birds. To investigate the potential functional impacts of defaunation at a regional scale, we compare the species assemblages predicted by geographical range maps to those recorded by intensive field surveys using the distribution of individual traits and measures of functional volumes (which account for trait combinations). Previous research shows local declines of larger-beaked seed-dispersers in tro pical forests (Galetti et al. 2013 ; Dirzo et al., 2014; Pérez-Méndez et al., 2016; Bovo et al. 2018) because of a range of factors, including slower reproductive output, larger spatial requirements and sensitivity to hunting. We therefore predict that assemblages recorded by recent surveys will have fewer large species than predicted by range maps, contributing to a reduction in FD.

To examine different seed-dispersal scenarios, we subdivide land cover into three broad classes – intact forest, degraded forest and the agricultural/silvicultural matrix. Previous work has identified high levels of species turnover across land cover types in the same assemblages (Hatfield et al., 2020; Moura et al., 2013; Solar et al., 2015) and elsewhere (González-Varo et al., 2017), suggesting that direct transfer of seeds between very different habitats is reduced. Because of this and the pattern of extirpation of larger species we predict that movements between intact forest and the matrix are potentially limited to a reduced shared species pool representing a narrower range of traits, whereas there is a 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 higher potential for seed dispersal between intact and degraded forests, or between degraded forests and matrix (Figure 1).

We analysed avifaunal data collected from the state of São Paulo (Brazilian Atlantic Forest; 23ºS 45ºW) and two areas in the state of Pará (Brazilian Amazonia): Santarém (03ºS 55ºW) and Paragominas (03ºS 48ºW). These three survey sites are each in a different biogeographic province and differ in their disturbance histories, providing a useful comparison. All datasets used three 15-minute point counts spaced along a transect to sample bird assemblages but differed slightly in overall study design (owing to different habitat extent and configuration). Amazonian sampling was conducted in 2010–2011 at 352 transects (300 m, 75 m radius points 150 m apart, surveyed twice; Supplementary Methods; Lees et al., 2012, 2013). Atlantic Forest sampling was conducted in 2015–2017 at 147 transects (150 m, 25 m radius points 75 m apart, surveyed four times; Supplementary

We estimated FD using the convex hull volume of the functional space occupied by an assemblage (Villéger et al., 2008). The functional space was constructed in three dimensions using axes based on the total range of traits across all avian frugivores in each region independently (combining the species lists from transects and geographical range maps).

Functional axes were generated using a two-step PCA process designed to partition the effects of body size from morphological trait variation associated with foraging and locomotion (Trisos et al., 2014). Specifically, two PCAs were initially conducted – a PCA based on foraging traits (gape size, beak length along the culmen, beak length to the anterior edge of the nares, beak width and beak depth) and a second based on locomotion traits (tarsus length, tail length, wing length and HWI). The first axis from both the foraging and locomotion trait PCAs were then combined in a second PCA that reflects variation in body size. This provided the three axes to construct the functional space, one from the second step PCA reflecting variation in body size, the second axis from the foraging PCA reflecting variation in trophic niche and the second axis from the locomotion PCA reflecting variation in locomotion (Figure S2). FD calculations were conducted in the R package ‘betapart’ (Baselga et al., 2021).

This analysis of FD allowed us to compare species assemblages predicted by range maps and those recorded by surveys. The list derived from range maps provides an approximation of the expected species assemblage. The proportion of the functional volume of the expected species assemblage (inferred from range maps) that was retained (shared) and lost by the observed assemblage (inferred from site transects) was calculated. The percentage retained was then also compared to that retained by randomly generated assemblages (with richness held constant; Supplementary Methods). This allowed us to examine whether the species losses estimated from the survey data reduced FD retention to 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 a greater extent than a trait-independent process of loss whereby species are lost at random with respect to their traits as opposed to a trait-based filter (e.g., loss of the largest species). All estimates were conducted for the full avian frugivore assemblage and then for a restricted sample of forest species.

When assessing the potential for seed transfer among land cover types we chose to focus primarily on frugivore gape width as this provides the best measure of seed dispersal limitation with regards to size. The land cover classifications provided in the original studies (Hatfield et al., 2020; Moura et al., 2013) were aggregated into three categories: intact forest, degraded forest and matrix (Table S1 and S2). Intact forest is used as a relative term as these landscapes are still likely affected by historic and ongoing disturbance. First, we compiled frugivore lists for each land cover type (each region independently). Next, we produced lists of species common to both intact forest and degraded forest then, intact forest and the matrix. We also produced a list of species common to both degraded forest and the matrix. This allowed us to calculate the percentages of species in common and to evaluate possible seed dispersal pathways in the context of gape width, with the overlap lists corresponding to the potential steps in Figure 1. Species overlaps were also calculated when sampling effort across land cover types was fixed (Supplementary Methods). To include estimates of dispersal ability, we also repeated the overlap comparison with HWI and with body mass (AVONET; as an indication of body size). The proportion of functional diversity not retained by the species shared between land cover types was also estimated by applying the above methods (2.3) to lists of species occurring in multiple habitat combinations (Supplementary Methods). For both gape width and FD, we compared the observed species overlaps to random selections to test for trait-based filters (such as an increased likelihood of extirpation for species with large body size; Supplementary Methods). 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258

In addition to the percentage overlaps for each region, overall figures were calculated by combining the species by land cover lists from the three regions, then calculating overlaps. This approach differs from averaging values across regions as some species are found in multiple regions and each region has a different species richness. Land cover associations may also differ between regions as sensitivity to forest degradation is not uniform across a species’ geographic range (Moura et al., 2016; Orme et al., 2019). This means that acrossregion pooling may not accurately reflect the associations of individual species in a particular region. On the other hand, many of these species have few detections even with the large sampling effort considered, so the pooled figures represent an increased sampling effort for species found in multiple regions.

Field surveys of frugivorous birds detected 73 species in Santarém, compared with 104 predicted by geographical range maps for the region. Similarly, surveys detected 76 species compared to 87 species predicted in Paragominas, and 46 species compared to 81 species predicted in São Paulo. The species assemblages recorded by field surveys in both Amazonian study regions suggested a loss of ≤ 6% of FD compared to original levels of FD calculated for the same regions based on range maps. The equivalent reduction in FD for São Paulo was much higher, with an estimated decline of 28% compared with expected FD calculated from range maps. Similar results were found when the assemblage was limited to only forest species (Table 1 and Figures 2, S3-4). When comparing observed assemblages to those randomly generated from the species pool, we found no strong evidence of losses being clustered in relation to traits (Table 1).

Considering individual traits (Table S4–S6; Figure S5–S7), the Santarém survey assemblage had lower maximum values for beak length and HWI compared to that predicted from range maps. The mean (± sd) gape size was similar for the field survey (15 mm ± 7.8 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 mm) and range map (14.7 mm ± 7.7 mm) assemblages. For Paragominas, only HWI differed in maximum values, with a lower maximum in the field survey assemblage. The mean gape size was similar for the field survey (14.7 mm ± 7.0 mm) and range map (14.5 mm ± 6.8 mm) assemblages. In São Paulo, observed maximum values for tarsus length, wing length and HWI were lower, and most traits showed reduced maximum values for the forest species assemblage comparisons. For this region there was a slight reduction in the mean gape size when comparing the field survey (11.6 mm ± 6.8 mm) to the range map (12.5 mm ± 6.9 mm) assemblage.

Across study regions, we found that 57% of species found in intact forest were not detected in the matrix, and are therefore highly unlikely to disperse seeds between the two (Figure 1a). Regionally the number of species found in intact forest but not in the matrix was 28 (62% of intact forest assemblage) in Santarém, 30 (79%) in Paragominas and 12 (46%) in São Paulo. In all three regions, only a few species present in intact forest were absent from nearby degraded forest (Figure 1b; 10% of frugivore species richness in intact forest across all sites): 1 species (2%) in Santarém; 4 species (11%) in Paragominas; 6 species (23%) in São Paulo. Roughly half (47% overall) of frugivore species were found in degraded forest but not in the nearby matrix (Figure 1c): 38 species (54% of the degraded forest assemblage) in Santarém; 48 species (70%) in Paragominas; 12 species (32%) in São

Our results suggest that seed dispersal between intact tropical forest and matrix landscapes will proceed through a combination of direct and stepwise processes (scenarios a–c; Figure 1). Given the high species turnover observed across land cover gradients, we conclude that relatively few frugivore species are involved in direct seed dispersal from remaining patches of intact forest into the surrounding matrix, with these events likely to be rare. Moreover, we can infer that direct dispersal is mainly restricted to smaller seeds, because a relatively low number of large-gaped frugivorous bird species are shared between both habitats. This finding implies that large-seeded tree species are more likely to require a stepwise process, potentially delaying natural forest regeneration by decades if the distance to intact forest is too large. 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335

The number of avian frugivores shared between more structurally similar habitats (that is, shared between intact and degraded forests; or between degraded forests and the matrix) is relatively high. Additional seed dispersal between intact forest and matrix may therefore follow a stepwise pathway through degraded intermediaries, such as secondary forests. This stepwise dispersal pattern is much less efficient, however, as it requires seeds deposited in intermediate habitats to develop into seed-bearing trees before they can be transferred to the surrounding matrix. These results also indicate that transport of larger seeds by birds is reliant on a relatively small number of large-ga ped species (Burns, 2013 ; Wheelwright, 1985), providing an important insight into likely challenges for natural regeneration in human modified tropical forest landscapes.

Reliance on few large-gaped species as potential dispersal vectors between intact forest and matrix is clearest for the sites in São Paulo and Paragominas but can still be seen to a lesser extent in Santarém (Figure 3). In São Paulo and Paragominas, it is not only that there are few large-gaped species shared between intact forest and matrix, very few large-gaped species are recorded in the matrix at all (Figure S11). These findings align with previous studies reporting that relatively few bird species use both intact and cleared land-cover types (Rehm et al., 2017) and that these ecologically flexible species with the greatest potential to act as mobile links for seed dispersal tend to be smaller bodied with smaller gapes (Pizo & dos Santos, 2011).

In our Amazonian landscapes, the dispersal of very large seeds between intact forest and the cleared matrix relies on species like Pteroglossus aracari and other toucans (Ramphastidae). In the Atlantic Forest landscapes, birds with larger gape sizes detected in both intact forest and cleared matrix included toucans (Ramphastos dicolorus) and cotingas (Pyroderus scutatus). While the importance of large-gaped frugivorous birds for the recruitment of large seeds is well known (Beltrán & Howe, 2020; Reid et al., 2021), relianc e on very few species reduces resilience to future perturbations, particularly if population size 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 is also low. Our results show that, especially for São Paulo, most large-gaped species are either rare visitors to matrix areas or missing from these landscapes altogether. All our surveys detected fewer bird species than predicted by range maps, resulting in slightly lower estimates of FD than anticipated from geographical ranges. In contrast with some previous studies (e.g., Bovo et al, 2018), we found no evidence that larger-gaped species were more likely to drop out from current assemblages. Nonetheless, their loss has a disproportionate impact on FD because redundancy declines towards the extremes of the trait distribution. The absence of one large-gaped species therefore represents a greater decline in ecological function and resilience than the loss of a different species more centrally placed on the gape-size gradient, where redundancy is high (Ali et al. 2023) .

It is worth emphasising that distribution maps can provide misleading information about historical assemblages. Mapping exercises tend to overestimate species richness at finer scales (Hurlbert & Jetz, 2007), and are unlikely to account for habitat specialisation (J etz et al., 2008). In addition, field surveys are subject to imperfect detection of species with cryptic behaviours or spatiotemporally patchy distributions. Even with intensive sampling effort, the diversity of rare or inconspicuous species in tropical bird communities makes detection challenging (Banks-Leite et al., 2014; Robinson et al., 2018). Nonetheless, the number of species overlooked by our field surveys is likely to be very low, particularly for large-gaped species which tend to be highly conspicuous when present. Almost all the losses we detect are therefore likely to represent true losses of function or declines in functional redundancy (Figure 2). In addition, even the frugivorous species detected in our study landscapes may be greatly reduced in population size by habitat loss and hunting, meaning that the volume and distance of seed dispersal may be substantially reduced in comparison with historical landscapes. 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387

Our analyses focus exclusively on specialist avian frugivores, potentially overlooking the contribution of bird species for which fruit is not a major part of their diet, such as omnivores (Carlo & Morales, 2016; Rehm et al., 2017) or invertivores that supplement their diet with some fruit (Camargo et al., 2020). Migratory tyrant-flycatchers, for instance, may consume large quantities of berries in degraded Amazonian and Atlantic forests during the nonbreeding season, with potentially large impact on early successional stages (Athiê & Dias 2016). Other taxa such as fish, reptiles and mammals can disperse seeds but again may perform different roles in the dispersal system (de la Peña-Domene et al., 2014; Donatti et al., 2011). Moreover, they face similar pressures from habitat loss and direct exploitation in human-modified landscapes (Costa-Pereira & Galetti, 2015). Although birds cannot provide a complete picture of seed dispersal on land cover gradients, they appear to play a major role, especially for later successional species (de la Peña-Domene et al. 2014). We have argued that patterns of species occurrence across land cover gradients can provide insights into the likely transfer of seeds between habitat types (Sekercioglu, 2006). This is based on the reasoning that the locations in our study represent adjacent portions of a gradient and previous evidence suggests that individual birds will move between adjacent land cover types (Mayhew et al. 2019) with widespread spillover effects (Hatfield et al. 2020). However, this type of information is sparse and species may vary in how often they move between land cover classes. Our approach also makes assumptions about the food plants of frugivorous birds, despite relatively limited knowledge of diets and trophic interactions, particularly in the context of tropical forest birds.

To address these issues, further work is required to clarify rates of movement between land covers by tracking individual birds, and to provide more detailed information about avian diets through techniques such as metabarcoding (Hoenig et al., 2022). This improvement in dietary information would allow for a more comprehensive treatment of species, including abundant generalists, for which fruit is not the main component of their diet but which may nonetheless function as important seed dispersers for particular plant species (Pizo, 2004). 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412

Nonetheless, the addition of these species is unlikely to alter our main conclusions because generalist species tend to be more important in degraded forests and are unlikely to provide dispersal services for larger seeds.

Along with more detail on movements and diet, we also require robust data on abundance. Without a better grasp on the number of individual birds involved, it is challenging to evaluate the implications of presence-absence data, as functions are often effectively lost even when particular frugivore species are still present but reduced to such low numbers that they are essentially not delivering any effective seed dispersal function. Future analyses should also encompass the role of seed-dispersing mammals such as primates and bats, which may play a vital role in mediating regeneration potential of human-modified tropical forests (Stoner et al., 2007).

With tropical deforestation continuing at a rapid pace, maintaining the potential for forest regeneration is a critical concern. Logged and disturbed rainforests contain tree species with lower average wood density and smaller seeds (Bello et al., 2015; Hawes et al., 2020) while cleared areas have greatly reduced value for biodiversity. The relevance of strategies promoting rapid and affordable forest regeneration is brought into sharp focus by policy targets, including the Brazilian government’s Nationally Determined Contributions under the Paris Agreement, which represent a binding commitment to achieve carbon neutrality and mitigate climate change impacts, partly through forest restoration. For example, the state of Pará has committed to restore c.5 Mha of Amazonian rainforest by 2030 (Plano Estadual Amazônia Agora, 2020). Such targets can be achieved through direct interventions such as tree-planting programmes. However, the passive restoration of tropical forests via natural processes has a number of benefits over active restoration, including increased diversity (Chazdon & Guariguata, 2016) and greater potential to track climate change (Fricke et al., 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 2022) . In addition, natural regeneration is the only financially viable solution over extensive degraded areas because it can be delivered at minimal cost (Chazdon & Guariguata, 2016). Where forest restoration is a key objective, our results suggest that dispersal from natural seed sources would require close proximity to intact native forest, supporting earlier work (e.g., Cardaso da Silva et al., 1996). Cleared areas further away from native forests may require alternative approaches, such as applied nucleation and nurse trees, to help encourage seed dispersers to use more of the landscape (Pizo & dos Santos, 2011; Freeman et al., 2021). The likelihood of recruitment will often be low, however, as many matrix environments provide unsuitable terrain for seed germination and sa pling growth (Reid & Holl, 2013 ). Our results also suggest that degraded forest plays a critical role in supplying seeds and seed-dispersal agents. These disturbed habitats not only hold more biodiversity than fully cleared landscapes, they also support landscape-level resilience by providing a vital source and conduit for seed transfer into the non-forest matrix. Thus, prioritising the conservation and management of substantial areas of logged and secondary forests is not only beneficial in terms of their efficient nutrient cycling and carbon uptake (Malhi et al. 2022), but also vital for promoting the healthy seed-dispersal dynamics required for rapid and economical reforestation programmes.

While secondary and disturbed forests can play a key role, our findings highlight the importance of preserving extensive intact forests to promote seed transfer from large-seeded trees, including commercially important hardwoods and other species with tall stature crucial to achieving the vegetation structure and microhabitat complexity of old growth forests (Hawes et al., 2020). In effect, the potential for forest resilience and regeneration will best be delivered by mosaic landscapes with degraded-forest buffers around core areas of intact forest that are not too distant. This tallies with proposals to focus restoration efforts on landscapes with intermediate forest cover levels and good connectivity with intact forest 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 5 References cause disproportionate reductions of functional diversity and uniqueness. Functional Ecology, 37, 162–175. https://doi.org/10.1111/1365-2435.14201 Athiê, S., & Dias, M. M. (2016). Use of perches and seed dispersal by birds in an abandoned pasture in the Porto Ferreira state park, southeastern Brazil. Brazilian Journal of Biology, 76, 80-92.

Banks-Leite, C., Pardini, R., Boscolo, D., Cassano, C. R., Püttker, T., Barros, C. S., & Barlow, J. (2014). Assessing the utility of statistical adjustments for imperfect detection in tropical conservation science. Journal of Applied Ecology, 51, 849–859. Baselga, A., Orme, D., Villeger, S., De Bortoli, J., Leprieur, F., & Logez, M. (2021). betapart: Partitioning beta-diversity into turnover and nestedness components. R package version 1.5.3. https://cran.r-project.org/package=betapart Bello, C., Galetti, M., Pizo, M. A., Magnago, L. F. S., Rocha, M. F., Lima, R. A. F., Peres, C. A., Ovaskainen, O., & Jordano, P. (2015). Defaunation affects carbon storage in tropical forests. Science Advances, 1, e1501105. https://doi.org/10.1126/sciadv.1501105 Beltrán, L. C., & Howe, H. F. (2020). The frailty of tropical restoration plantings. Restoration Ecology, 28, 16–21. https://doi.org/10.1111/rec.13066 Biggs, C. R., Yeager, L. A., Bolser, D. G., Bonsell, C., Dichiera, A. M., Hou, Z., Keyser, S. R., Khursigara, A. J., Lu, K., Muth, A. F., Negrete Jr., B., & Erisman, B. E. (2020). Does functional redundancy affect ecological stability and resilience? A review and meta-analysis. Ecosphere, 11, e03184. https://doi.org/10.1002/ecs2.3184

BirdLife International. (2022). IUCN Red List for birds. www.birdlife.org 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485