Tropical invertebrate community assembly processes are robust to a gradient of land use intensity

Understanding how community assembly processes drive biodiversity patterns is a central goal of community ecology. While it is generally accepted that ecological communities are assembled by both stochastic and deterministic processes, quantifying their relative importance remains challenging. Few studies have investigated how the relative importance of stochastic and deterministic community assembly processes vary among taxa and along gradients of habitat degradation. Using data on 1,645 arthropod species across seven taxonomic groups in Malaysian Borneo, we quantified the importance of ecological stochasticity and of a suite of community assembly processes across a gradient of logging intensity. The relationship between logging and community assembly varied depending on the specific combination of taxa and stochasticity metric used, but, in general, the processes that govern invertebrate community assembly were remarkably robust to changes in land use intensity.


12
Community assembly processes drive biodiversity patterns, and a key goal in community ecology is to 13 quantify the relative importance of different community assembly processes. Simultaneously, it is well-14 documented that land use changes impact biodiversity (Newbold et al. 2015), but it is much less well-known 15 that they can also affect community assembly mechanisms (Wearn et al. 2019). Currently, therefore, we have 16 a strong awareness of the patterns of biodiversity change that are generated by land use change (Newbold et 17 al. 2015), but little understanding of the extent to which the fundamental community assembly processes that 18 create that change are impacted. Moreover, the specific assembly processes generating ecological 19 communities may differ among taxa due to differences in trait evolution (Weiher et al. 2011), meaning 20 studies examining land use impacts on the assembly processes of mammals (e.g. Wearn et al. 2019) may 21 provide little insight into the impacts on other taxa. Attempts to rely on natural ecological processes to 22 restore biodiversity rely, by definition, on naturally occurring community assembly processes (Palmer et al. 23 1997, Hilderbrand et al. 2005). It is therefore of fundamental importance that we gain a deeper understanding 24 of whether those assembly processes in modified habitats are the same or different to those observed in 25 primary habitats. 26 Community assembly involves a combination of determinism and stochasticity, there has been long-standing 27 debate over their relative influences (Connor and Simberloff 1979 6 be more heterogeneous (high turnover) or homogeneous (low turnover), and comparing these patterns with 93 neutral expectation can indicate the relative importance of stochasticity (Ning et al. 2020). 94 There is a long-standing need to evaluate the relative importance of stochastic and deterministic processes 95 along environmental gradients and among taxa (Weiher et al. 2011). Here, we address that knowledge gap by 96 using a variety of indices to quantify the relative contribution of stochasticity to community assembly for 97 seven invertebrate taxa across a gradient of logging intensity. We also quantify the relative importance of a 98 suite of community assembly processes across the logging gradient. Our data encompass a comprehensive 99 gradient of logging intensity, from areas that have never experienced logging to areas that have been salvage 100 logged. We quantified community assembly for a range of invertebrate taxa including three groups of 101 Coleoptera (beetles), along with Formicidae (ants), Lepidoptera (moths), Orthoptera and Araneae (spiders). 102 Together, these taxonomic groups encompass a range of feeding guilds and are of immense ecological 103 importance (Barlow and Woiwod 1989, Didham et al. 1998, Grimaldi et al. 2005 Birkhofer 104 2017, Oumarou Ngoute et al. 2020). We use our data to test two hypotheses: (1) stochastic assembly will 105 decrease in importance as logging intensity increases, as logged forest should have stronger environmental 106 filtering; and (2) stochastic assembly will have a lower relative importance for trophic specialists, compared 107 to trophic generalists, because trophic specialists are expected to be more strongly assembled by selective 108 environmental filtering. Finally, we investigate whether the relative importance of a suite of different 109 community assembly processes varies across a gradient of logging intensity for different invertebrate taxa. 110

112
The study sites were located within the Stability of Altered Forest Ecosystems (SAFE) project (4° 38′ N to 4° 113 46′ N, 116° 57′ to 117° 42′ E), a large-scale ecological experiment encompassing a gradient of land use 114 intensities in the lowland tropical forests of Sabah, Malaysian Borneo (Ewers et al. 2011). We used data 115 from 14 out of the 17 experimental sampling blocks at SAFE (Ewers et al. 2011), excluding three blocks 116 located in oil palm plantation. Ten sampling blocks were located in twice-logged forests and four were 117 located in protected areas. Two of these protected area blocks were in the Maliau Basin Conservation area 7 and have never experienced logging, while the other two had experienced light logging through both legal 119 and illegal processes. Each sampling block comprised a set of 4 -43 sampling sites (mean = 19) and covered 120 a spatial area of 4 -229 ha (mean = 56). We grouped invertebrate samples collected within each block which 121 we considered as one local community for analysis. The aggregation of all local communities across all 122 sampling blocks was considered to represent the metacommunity.  We combined community composition data collected from seven invertebrate taxa: three groups of beetles, 133 plus ants, moths, spiders and Orthoptera. Different groups had different sample sizes and not all groups were 134 sampled in all 14 sampling blocks (Table S1, Figure S1). Beetles were sampled between 2011 and 2013 135 using combination pitfall-malaise traps in all 14 sampling blocks. Three different groups of beetles were 136 sampled: Curculionoidea (weevils), Staphylinidae (rove beetles) and Scarabaeoidea (scarabs) (Sharp et al. 137 2018, 2019). Because of differences in their feeding guilds, each group was considered a separate taxon and 138 was analysed separately: weevils are predominantly herbivorous; most scarabs in our dataset are dung-139 feeders; and rove beetles can belong to several feeding guilds. Beetles were identified primarily to 140 morphospecies, except some scarabs which were identified to species. Ants were sampled between 141 There are different ways for communities to express stochasticity, so there is value in assessing multiple 154 metrics of stochasticity on the same communities (Vellend and Agrawal 2010). We quantified stochasticity 155 using three null-model based mathematical frameworks that summarise stochasticity both at the community 156 level and at the level of individual species. All three stochasticity metrics were calculated from a separate 157 community composition (site × species) matrix for each of the seven taxa. We calculated stochasticity 158 metrics for all sites in the composition matrix, and grouped sites together by sampling block to calculate the 159 mean of each stochasticity metric for each sampling block. metacommunities for the null expectation, the total number of species was fixed as observed, and within 168 each local community, individuals were drawn at random from the metacommunity with probabilities 169 proportional to their regional occurrence frequencies. Second, to test the effect of different assumptions in the null modelling framework, we also used the 171 modified stochasticity ratio (MST) which transforms NST under the assumption that observed community 9 similarity is equal to the mean of the null expected community similarity under stochastic assembly. MST is 173 also normalised from 0 to 1 with values closer to 1 indicating a higher importance of ecological stochasticity. 174 The NST and MST metrics assess ecological stochasticity at the community level, whereas our third metric, 175 the neutral taxa percentage (NTP) assesses stochasticity at the level of individual species (Burns et al. 2016). 176 NTP is the proportion of species with occurrence frequencies that could be predicted by Sloan's neutral 177 model (Sloan et al. 2006(Sloan et al. , 2007. This model assumes that assembly is driven solely by chance and dispersal. comparisons between all sites. We then grouped sites by sampling block and calculated the mean relative 205 importance of each process for each taxa x sampling block combination. 206 The iCAMP framework divides taxa into groups ('bins') based on their phylogenetic relationships, then 207 identifies the dominant community assembly process in each bin. Phylogenies were not available for the 208 taxa included in this study, so we used taxonomy trees as proxies for phylogenies. Taxa were divided into 209 bins based on taxonomic identity (Table S2, Figure S3), and bins with fewer than the minimum number of 210 taxa (n = 9) were merged into the bin to which they were most closely related (Ning et al. 2020). The number 211 of randomisations used for the null model analysis was 500. 212 To calculate the relative importance of community assembly processes, iCAMP uses null model analysis of 213 phylogenetic diversity (beta net relatedness index, βNRI) and taxonomic β-diversity (modified Raup-Crick 214 metric, RC) to identify the process governing each bin. According to the framework developed by Ning et al. We also analysed species richness trends among taxa and across logging gradients, and compared these 233 trends to those of the stochasticity metrics to see whether the metrics were affected by species richness. 234 We calculated the mean of each stochasticity metric (NST, MST and NTP) and of each community assembly 235 process (heterogeneous selection, homogeneous selection, dispersal limitation, homogenising dispersal and 236 drift) for each taxa x sampling block combination. We estimated the 95 % quantiles of these means using 237 bootstrapping; we sampled the values used to calculate the mean 1000 times with replacement, then took the 238 5 th and 95 th quantiles of this distribution. 239 To compare the relative importance of stochasticity among taxa we used one-sample t-tests to compare the 240 overall unweighted mean NST, MST and NTP across all taxa to 0.5, which we used as a boundary point 241 separating stochastic (>0.5) from deterministic (<0.5) community assembly. We also used ANOVA to test 242 for differences in NST, MST and NTP among taxa. To test the hypothesis that trophic generalists (ants and 243 rove beetles) would be more stochastic than trophic specialists (moths, Orthoptera, spiders, scarabs and 244 weevils), we used beta regression with a single categorical predictor that describes whether the taxon is 245 considered a trophic generalist or specialist. We fitted three separate beta regression models, each with a 246 stochasticity metric as the response variable. We broadly categorised each taxonomic group into trophic 247 generalists or specialists based on trophic level. Ants were considered generalists because their diets can Spiders were all predatory (Russell-Smith and Stork 1995), scarabs were primarily coprophagous (Sharp et 253 al. 2018) and moths were herbivorous as larvae and nectarivorous as adults (Romeis et al. 2005), so we 254 classified these groups as trophic specialists for our analysis. 255 To investigate the effect of logging on the importance of each stochasticity metric we used beta regression, 256 with the mean ACD of each block as the predictor. To see how this relationship varies among taxa, we fitted 257 another beta regression model for each stochasticity metric with taxa, ACD and their interaction as predictors 258 of stochasticity. A similar analysis was conducted to investigate the effect of logging on the relative 259 importance of each community assembly process, but Dirichlet regression was used instead of beta 260 regression as Dirichlet regression is appropriate for proportions with more than two categories (Douma and 261

Weedon 2019). 262
To gain a taxa-independent metric reflecting the overall change in the relative importance of the five 263 community assembly processes across the logging gradient, we combined the slopes for all taxa to give a 264 weighted summary mean of the slopes for each process, using a method similar to a fixed effects meta-

276
Across all datasets, our analysis included 32,294 individuals belonging to 1,645 species or morphospecies 277 (Table S1). In general, community similarity between sampling blocks was low. Each of the three beetle 278 taxa, as well as moths, showed a decrease in community similarity compared to old growth forest as logging 279 intensity increased, whereas the remaining three taxa (ants, Orthoptera and spiders) did not ( Figure S2). Community assembly processes 315 Dispersal limitation was the dominant community assembly process when all taxa were combined and 316 weighted by sample size (overall relative importance = 64 %, 95 % CI = 26 -97). Dispersal limitation was 317 the dominant process for four out of the seven taxonomic groups (rove beetles, scarabs, weevils and ants), 318 with a relative importance ranging from 94 to 97 % (95 % quantiles = 88 -99). Drift was the dominant 319 process for spiders, Orthoptera and moths; 89 % of spider community assembly (95 % quantiles = 78 -96) 320 was estimated to be underpinned by drift, while this was estimated to be 75 % for Orthoptera (95 % quantiles 321 = 55 -87) and 67 % for moths (95 % quantiles = 54 -78) (Figure 2A; Table S4). 322 When comparing the relative importance of community assembly processes among sampling blocks, each of 323 which had a different level of logging intensity, the relative importance of the five community assembly 324 processes was not significantly affected by logging intensity for most taxa ( Figure 2B, Table S5). When all 325 groups of taxa were combined, the weighted mean slopes for each process were not significantly different 326 from zero. The strongest effects were a decrease in the importance of dispersal limitation as logging intensity 327

344
While the relative importance of stochasticity varied among taxa and metrics, in general, the balance 345 between stochasticity and determinism appeared robust to a gradient of land use intensity. At a finer 346 resolution, dispersal limitation, inferred from community patterns, was the dominant assembly process 347 overall. Two out of the three stochasticity metrics (NST and MST) were not significantly different between 348 trophic generalists and trophic specialists. Only the NTP metric supported our hypothesis that trophic 349 generalists would be more stochastic than trophic specialists. In general, for all taxa except moths, land use 350 change had little impact on the relative importance of a suite of community assembly processes. Together, 351 this suggests that, while logging has profoundly negative impacts on biodiversity, it tends to have little 352 impact on the main community assembly drivers for most invertebrate taxa studied. 353 Overall, there was at best a very weak effect of land use intensity on the role of ecological stochasticity in 354 structuring insect communities. The direction of the relationship between logging and stochasticity varied 355 among taxa, but these relationships were not statistically significant (Figure 1). The relative importance of 356 different community assembly mechanisms was also not significantly affected by logging intensity for six 357 out of seven taxonomic groups ( Figure 2B). Moths, however, were the exception, showing changes in 358 community assembly processes ( Figure 2B) and turnover in species composition ( Figure S2) with logging. 359 For the six taxa that showed no change in community assembly with logging, we might expect to find little 360 evidence of a change in the assembly processes governing invertebrate communities if those communities do 361 not exhibit turnover in species composition across the logging gradient. For three of the taxa (ants, 362 Orthoptera and spiders), this assumption holds true: we found no evidence of changing taxonomic identities 363 across the logging gradient ( Figure S2) which aligns well with a lack of change in the assembly processes 364 governing those taxa ( Figure 2B). However, all three beetle taxa (rove beetles, scarabs and weevils) did 365 exhibit significant turnover in taxonomic identity as logging intensity increased ( Figure S2), which is 366 consistent with previous studies (Hamer et al. 2003, Cleary et al. 2007, Sharp et al. 2019). Yet these taxa did 367 not exhibit significant changes in community assembly metrics across the logging gradient, suggesting that 368 the species turnover was generated by the same ecological processes, regardless of logging intensity. This 369 leads to the general conclusion that, regardless of whether taxonomic identities change, community assembly 370 processes remain robust to changes in above-ground carbon density. 371 Since the community assembly metrics used here are based on taxonomic and phylogenetic diversity, they 372 could be influenced by changes in species richness, which can change systematically among taxa and along 373 logging gradients (Burivalova et al. 2014). However, we analysed changes in species richness and showed 374 that they are unlikely to underpin our results, as patterns of species richness (Figures S3, S4) differed from 375 those of stochasticity metrics and community assembly processes (Figures 1, 2). This suggests that the trends 376 in stochasticity and community assembly are independent of any trends in species richness. 377 One possible explanation for why community assembly processes appear to be strongly conserved across the 378 gradient of logging intensity is that our data were collected after logging had taken place. While our study 379 landscape encompasses a very wide range of historic logging intensity, the time delay between the logging 380 event itself and our description of the invertebrate communities means any transitory impacts of logging on 381 community assembly processes would not have been detectable. Mahayani et al. (2020) showed that 382 phylogenetic diversity and community structure of tree communities had recovered 10 years after a single 383 logging cycle in Bornean tropical forest. Therefore, it could be possible that the ecological communities we 384 sampled might have recovered their basic, pre-logging structures so that the community assembly 385 mechanisms in logged forests now resemble those of unlogged forests. 386 In general, the processes that govern community assembly were robust to logging in our study. However, 387 some relationships were significant for certain combinations of taxa and process. Interestingly, moths 388 showed a decrease in the importance of heterogeneous selection, and an increase in the importance of 389 homogeneous selection and homogenising dispersal, with increasing logging intensity. Together, these trends 390 suggest that logging could be driving the biotic homogenisation of moth assemblages. 391 When inferring assembly processes from community patterns, dispersal limitation was the most important 392 driver of community assembly for ants and beetles, whereas drift was the main driver of assembly for 393 spiders, Orthoptera and moths ( Figure 2A). Spiders, Orthoptera and moths were sampled at fewer sites than 394 ants and beetles (Table S1), so we cannot definitively rule out the possibility that this result may represent a 395 sampling effect. We do note, however, that the scarab community had a lower total number of individuals 396 and a higher number of species than the Orthoptera, suggesting that an undersampling-driven effect should 397 have exerted a greater impact on them than on the Orthoptera. Further, when we conducted a sensitivity 398 analysis by grouping sites within the same block, drift was still the dominant community assembly process 399 for spiders, moths and Orthoptera (Text S1). 400 The overall importance of dispersal limitation as the dominant community assembly process, especially for 401 ants and beetles highlights the importance of maintaining and, if necessary, restoring landscape connectivity 402 in logged forests (Wearn et al. 2019). We emphasise that dispersal limitation was not directly measured in 403 this study but was inferred from community patterns. Dispersal limitation can result in high spatial turnover 404 in community composition due to low levels of exchange of organisms among local communities (Stegen et 405 al. 2013). The slight increase in dispersal limitation we detected as logging intensity increased, though not 406 significant, is in a direction that is consistent with previous studies suggesting that animal communities in 407 logged tropical forests may experience lower levels of dispersal compared to primary forests (Stratford and 408 This study has quantified community assembly mechanisms across a gradient of logging intensity for seven 410 groups of invertebrate taxa in Bornean rainforests. The effect of logging on stochasticity, and on different 411 community assembly processes, varied among the different taxa and different metrics of stochasticity, but 412 painted a general picture in which the dominant community assembly mechanisms are not impacted by 413 logging disturbance. Although logging did not alter the balance between stochastic and deterministic 414 community assembly processes for most taxa, we emphasise that logging, and in particular severe logging, 415 profoundly reduces species richness and changes community composition (Thorn et al. 2018). The 416 robustness of invertebrate communities to logging disturbance in our study suggests that knowledge of 417 primary community assembly can be useful in planning the restoration of modified communities as, for six 418 out of seven taxonomic groups, there were not significant changes in assembly processes despite changes in 419 land use intensity. 420