On the inconsistency of pollinator species traits for predicting either response to agricultural intensification or functional contribution

The response and effect trait framework, if supported empirically, would provide for powerful and general predictions about how biodiversity loss will lead to loss in ecosystem function. This framework proposes that species traits will explain how different species respond to disturbance (i.e. response traits) as well as their contribution to ecosystem function (i.e. effect traits). However, predictive response and effect traits remain elusive for most systems. Here, we present detailed data on crop pollination services provided by native, wild bees to explore the role of six commonly used species traits in determining how crop pollination is affected by increasing agricultural intensification. Analyses were conducted in parallel for three crop systems (watermelon, cranberry, and blueberry) located within the same geographical region (mid-Atlantic USA). Bee species traits did not strongly predict species’ response to agricultural intensification, and the few traits that were weakly predictive were not consistent across crops. Similarly, no trait predicted species’ overall functional contribution in any of the three crop systems, although body size was a good predictor of per capita efficiency in two systems. So far, most studies looking for response or effect traits in pollination systems have found weak and often contradicting links. Overall we were unable to make generalizable predictions regarding species responses to land-use change and its effect on the delivery of ecosystem services. Pollinator traits may be useful for understanding ecological processes in some systems, but thus far the promise of traits-based ecology has yet to be fulfilled for pollination ecology.

intensification. Analyses were conducted in parallel for three crop systems 23 (watermelon, cranberry, and blueberry) located within the same geographical 24 region (mid-Atlantic USA). Bee species traits did not strongly predict species' 25 response to agricultural intensification, and the few traits that were weakly 26 predictive were not consistent across crops. Similarly, no trait predicted 27 species' overall functional contribution in any of the three crop systems, 28 although body size was a good predictor of per capita efficiency in two 29 systems. So far, most studies looking for response or effect traits in pollination 30 systems have found weak and often contradicting links. Overall we were 31 unable to make generalizable predictions regarding species responses to 32 land-use change and its effect on the delivery of ecosystem services. 33 Pollinator traits may be useful for understanding ecological processes in some 34 systems, but thus far the promise of traits-based ecology has yet to be fulfilled 35 for pollination ecology. Identifying which traits govern species responses to particular threats 55 ('response traits') would provide the first step for predicting future species 56 loss. Furthermore, the magnitude by which ecosystem function declines when 57 a species is lost depends on that species' functional contribution. This, too, is 58 likely to be mediated by the species' traits ('effect traits'). Therefore, the 59 relationship between response and effect traits will mediate the magnitude of 60 the impact of human disturbance on ecosystem services (Schleuning et al. 61 2015). For example, if the same species traits that are associated with high 62 function are also most sensitive to disturbance, ecosystem function would be 63 predicted to decline rapidly (Larsen et al. 2005). 64 However, for the response-effect trait framework to be useful, it is first 65 necessary to identify response and effect traits that are both explanatory and Here, we seek to identify response and effect traits for wild bee species  Table A1). In addition, we also measured percent of open 150 natural/semi-natural habitat, which although it accounts for only a small 151 proportion of the total land cover (Supplementary Table A1 respectively (Greenleaf et al. 2007). 157

Pollinator function 158
To estimate the pollination services provided per bee species, we measured 159 two variables in the field, flower visitation frequency and per visit efficiency. As 160 variation in visitation frequency may be a function of land use at individual 161 farms, we use species abundances for each species at the site with its highest 162 abundance for each crop. Hence, we assess visitation frequency at its 163 maximum, which represents the optimal visitation frequency for each species. 164 To measure the pollination efficiency, we quantified single-visit pollen 165 deposition by presenting virgin flowers to individual bees foraging on the 166 target crop. After visitation, we counted the number of pollen grains deposited 167 per flower visit (watermelon) or the number of pollen tetrads with pollen tubes 168 per flower visit (cranberry and blueberry). Because species identification in the 169 field is not possible for most bees and net collecting immediately after visits is 170 generally not possible, for the measurement of pollination efficiency we 171 grouped bees in species groups. Each group consisted of between one and 172 27 species, with the median number of species per group being 4 species 173 (Supplementary Table A2). Control flowers were left bagged until the end of 174 the field day, and contained few pollen grains (watermelon mean = 3 grains, N 175 = 40 stigmas; cranberry mean = 0 tetrads, N = 82 stigmas; blueberry mean = 176 2 tetrads, N= 734 stigmas). We used mean number of pollen grains deposited 177 by a single visit group and assigned that value to each of the species in the

Species traits 181
Bee species vary in a number of traits that are associated with their response 182 to land-use change (Williams et al. 2010). Moreover, these traits will likely 183 affect the pollinator contribution to function, either by modifying its abundance 184 or because they are related to its per capita effectiveness. We obtained 185 detailed natural history data on 6 traits for the 90 bee species in our study: a) 186 sociality (solitary, facultative social, eusocial), b) nesting placement (hole, 187 cavity, stem, wood, ground), c) brood parasite (yes, no), d) body size, e) diet 188 breadth (level of specialization) and f) tongue length. 189 We obtained the trait data as follows. Species sociality level, nesting behavior 190 and brood parasite status were extracted from the literature ( Bartomeus et al. 191 2013a). Body size (estimated from intertegular span, IT; Cane 1987) was 192 measured in the lab using collected specimens that had been identified to the 193 species level by professional taxonomists. Multiple specimens were measured 194 per species (mean = 6.6 specimens ± 3 S.E.) and the mean across the 195 measured specimens was used as the value for the species. Bee body size 196 also correlates strongly with foraging distance (Greenleaf et al. 2007), and 197 thus is ecologically related to mobility. Tongue length was measured in the lab 198 for 7.7 ± 1.2 SE specimens per species, and the mean across the measured 199 specimens is used. For the 40 specimens for which we cannot obtain a 200 tongue measure, we estimated tongue length from the species' body size and 201 phylogeny using an allometric equation (Cariveau et al. 2016 2014). Nine species had fewer than 20 records and we were unable to 212 estimate diet breadth in the manner described above. Five of these species 213 are known to be specialized and we simulated the diet breadth index of 20 214 individuals visiting the known host plants. The four other species are known to 215 be generalists and we therefore used the mean diet breadth of its genus. 216 These four species were extremely rare (< 5 records each) in our analyzed 217

dataset. 218
To calculate diet breadth for each bee species, we considered the number of 219 plants species as well as the phylogenetic breadth that the bees fed upon by 220 using a rarefied phylogenetic diversity index (Nipperess and Matsen 2013). To 221 determine phylogenetic distances among plants, we first constructed a 222 general phylogenetic tree using the PHYLOMATIC "megatree" (version 223 predictors. The best model based on AICc was selected. When differences 261 between the best models were less than 2 we selected the simpler model. 262 The analysis for efficiency was done at the species group level (see above: 263 pollination function section). To obtain traits at the species-group level, we 264 calculated the mean values over species belonging to the same group, 265 weighted by the species mean abundance within the group. For categorical 266 variables we chose the dominant level, again weighted by species 267 abundance. This way, we assure that while species within a functional group 268 are selected to be functionally similar, the average traits used reflects species 269

composition. 270
All residuals were visually inspected to validate model assumptions. All 271 statistical analyses were performed in R (version 3.0.3, <www.r-project.org>). 272

Results 273
Response traits: Overall, we did not find a strong correlation between any  (Table A4). 307

Discussion: 308
Identifying traits that characterize which species are more sensitive to land-309 use change or those that are functionally important is complex. We found 310 some evidence for response and effect traits but they differed among crop 311 species as well as landscape variable used. Therefore, while some traits may 312 be important in some contexts, no traits were generalizable enough to be 313 used to predict how land-use change will influence the delivery of pollination 314 services across these systems. Further, the relationships identified were 315 weak. This does not negate the importance of traits for understanding which 316 mechanisms underlie species responses to land-use change or pollination 317 effectiveness, but it does suggest that traits commonly used for wild bees 318 might not be suitable for predicting which species will decline or how land-use 319 change will influence the delivery of ecosystem services. In fact, the trait-320 based literature in general is characterized by weak and/or idiosyncratic 321 relationships between traits and either species responses and functional 322 effects (Tables 1 and 2). 323 Being able to identify strong response traits would be a key tool for 324 understanding extinction risk, and an asset for conservation planning. 325 However, characterizing extinction risk based on traits is challenging. Despite 326 some generalities that emerge across taxa, with rare species, big species, 327 specialists, and higher trophic levels being in general more sensitive to  Work specifically on native bees has found that traits such as specialization, 331 body size, and sociality may predict responses to land use (Table 1;