www.sciencemag.org/content/349/6244/177/suppl/DC1

Supplementary Materials for

Climate change impacts on bumblebees converge across continents Jeremy T. Kerr,* Alana Pindar, Paul Galpern, Laurence Packer,

Simon G. Potts, Stuart M. Roberts, Pierre Rasmont, Oliver Schweiger, Sheila R. Colla, Leif L. Richardson, David L. Wagner,

Lawrence F. Gall, Derek S. Sikes, Alberto Pantoja

*Corresponding author. E-mail: jkerr@uottawa.ca

Published 10 July 2015, Science 349, 177 (2015) DOI: 10.1126/science.aaa7031

This PDF file includes:

Materials and Methods Supplementary Text Supplementary Acknowledgments Figs. S1 to S4 Tables S1 to S3

Materials and Methods Bumblebee observations

Observations of 214 bumblebee species from 1901 to 2010, inclusively, were assembled from North America and Europe. These observations included 754,127 georeferenced, dated records, including 429,625 European records and 324,502 North American records (14, 15). Many records excluded georeferencing information. After their removal, the analysis dataset included 423,411 records for 31 North American and 36 European species distributed across 369,239 unique location-years (i.e. spatiotemporally unique observations; Fig. S1). The median species-period sample size was 536 location-years (Tables S1, S2).

To produce this database from a candidate set of records, potentially unreliable

records were first flagged and removed. Criteria for reliability were completeness of species, locality, and sampling year information, and agreement between record georeferencing and stated country of origin. In addition, records located in the ocean less than 2500m from a high-resolution coastline were assumed to be coastal observations with spatially-imprecise georeferences, and were reassigned to the nearest point on land. About 6% of the records obtained from GBIF lacked latitude-longitude coordinates for collection localities. For these records, we obtained georeferencing data from the digital gazetteer, GeoNames (http://geonames.org; Creative Commons Attribution 3.0 License). Among these records, we retained those situated near populated places for which reliable geographic coordinates were available.

Duplicate collection records of a species for a given location-year (about 40% of

database) were removed to better reflect species occurrence rather than sampling or population density. The reduced set of records was sorted into four time periods: a basal period prior to most anthropogenic climate change (28) (1901-1974) and three subsequent periods during which the human impact on climate in Europe and North America is readily detectable (1975-1986, 1987-1998 and 1999-2010). We retained species with at least 100 spatially unique records in the basal period and at least 30 unique records in each subsequent period (Tables S1, S2). In most cases, species-period sample sizes were large: only Bombus sporadicus in the 1987-1998 had fewer records but was retained because it was otherwise well sampled in each time period. Climate change and geographical response variables

We expected to find the geographical position of these species shifting north given warming over recent decades (29). Species’ thermal limits should remain constant if the boundaries of their ranges shifted to track changing temperatures through time (30).

North American temperature data (°C) were available at 5 arcminute resolution and

derived from meteorological station data observed annually across the continent since 1901 (31). Temperature surfaces were obtained using ANUSPLIN (32). European temperature data (°C) were provided by the Climate Research Unit (CRU 1.2 (33) and 3.21 (34)) and downsampled to 5 arcminutes from coarser resolution (from 10 arcminutes

2

for CRU 1.2; and 30 arcminutes for CRU 3.21) data using an adiabatic lapse rule (35) (see equation 2.89) where the temperature was corrected by the difference in elevation in each raster cell from a digital elevation model at either 10 or 30 arcminutes resolution (CRU 1.2) and a digital elevation model at 5 arcminutes resolution (ETOPO5; http://www.ngdc.noaa.gov/mgg/global/etopo5.HTML). Raster datasets were further processed to extract annual maximum and minimum temperature values (°C) from high-resolution annual observations.

We measured shifts in bumblebee species' distributions in geographical space and

with respect to species’ thermal limits based on means of 5, 10, and 20 observations, respectively, from species’ latitudinal and thermal extremes through four time periods (baseline: 1901-1974 and three subsequent, equal-length climate change impact periods, 1975-1986, 1987-1998, and 1999-2010). Latitude is measured as distance (in kilometers) from the equator. While some climatic change in the baseline period reflect human activities, global climate change accelerated significantly after that time (36). Temperature observations approximate each species’ realized thermal niche, which is the range of temperatures where species can persist after accounting for biological interactions (37). Shifts were calculated as the difference in species’ thermal limits (warm or cool) and latitudinal limits (north or south) between each time period and the baseline (1901-1974) period. Variation in latitudinal range limits and thermal limits could have reflected differences in numbers of records (5, 10, or 20) used to calculate those limits or differences in sampling effort per time period for bumblebee species, but no such effects were detected in exploratory statistical models. Per-time-period sampling intensity was included as a covariate in models testing for shifts along latitudinal and thermal boundaries for species through time. We tested whether trends differed among continents using continent (Europe or North America) as a categorical variable in models.

We expected that species would shift to higher elevations in response to warming

where topography permits (38, 39). We measured the mean elevation of all species observations in each time period from digital elevation models at 30 arcsecond resolution from the Shuttle Radar Topography Mission (http://worldgrids.org/doku.php?id=wiki:demsre3) and used these to measure shifts in the elevation of species’ ranges relative to their mean geographical position. We evaluated potential elevational shifts against mean latitudes of species observations using the modeling procedure described below. Phylogenetic and statistical analyses

Widespread habitat conversions in Europe and North America (e.g. of forest clearance for agriculture) predate recent time periods examined in this study and could influence species persistence (40) and capacity to track shifting climatic conditions. Such habitat limitations, which could include both habitat loss and fragmentation effects (41), could cause range losses from southern areas or along species’ warm thermal limits. Habitat limitations could also prevent species from colonizing unoccupied areas beyond their previous geographical distributions. Similarly, changing agricultural land use

3

practices around the use of pesticides (especially neonicotinoid pesticides) have been implicated in bumblebee decline in many areas and could similarly alter bumblebee species’ responses to changing climatic conditions at their latitudinal or thermal limits.

We measured the potential impacts of both land use change and pesticide use as

alternative explanations to climate change-related shifts by testing for these effects in statistical analyses. Land use change measurements were available through time from 1900 to 2005 across both Europe and North America at 5’ resolution (from the HYDE 3.1) (42). These data are derived from combinations of historical agricultural measurements, recent satellite remote sensing-based land cover measures, and long term data and models of human population. To test if change in southern latitudinal or warm thermal limits resulted from changing pesticide use rather than observed climate changes, we focused on species with distributions in the United States and measured the mean annual species exposure to pesticides (149 of the most common pesticides in aggregate) and to the neonicotinoid, Imidacloprid. We used data on annual area of application per county from 1992, the first year of spatially detailed pesticide data availability, to 2009 (43). Imidacloprid is the most extensively used neonicotinoid pesticide in the United States and the only one used widely in this area by at least the latter part of the 1987-1998 time period.

We constructed ordinary least squares (OLS) regression models for each of the

change metrics for bumblebee species against the baseline observations (Table 1) using R 3.1.0 (44). Of primary interest in these regressions was the change in the response variable (i.e. shifts in the northern or southern latitudinal range limits per species, the warm or cool thermal limits per species, or mean elevation) over time, relative to baseline observations per species. We tested three models, all containing the independent variable (e.g. latitudinal position of species’ northern range limits), two containing a categorical continent term, and a third with an interaction term to test for cross-continental consistency in bumblebee responses. The results of the best model, selected using AIC, is plotted with its 95% confidence intervals (Fig. 1, 2). Where the model with the lowest AIC contained the continent or interaction effect, we colored the regression fits and confidence bands associated with each continent differently. Where a significant continent or interaction effect was noted (P < 0.05), we colored the regression fits and confidence bands associated with each continent differently. Otherwise, models presented with a single fit and confidence band demonstrate that thermal, latitudinal, or elevation trends across Europe and North America were consistent in that time period. Measurements of land use change observed in areas from which bumblebee measurements were derived were also included in candidate regression models. Land use changes were included as full interactions with the continent term, enabling detection of differing (or consistent) impacts between Europe and North America. Measures of pesticide and neonicotinoid use were included as covariates in models for species in the United States, where impacts could contribute to range losses from southern latitudinal or warm thermal limits, respectively. We constructed these models for both 1987-1998 and 1999-2010 periods (Table S3).

4

There is strong evidence that species with greater shared evolutionary history will have similar biological traits, such as thermal tolerances (45). If so, there could be a phylogenetic signal in species’ responses to climate change. While species’ latitudinal range limits and elevation are not traits upon which selection can operate, such limits may integrate a host of traits that could respond to selection. We tested whether bumblebee species' latitudinal and thermal limits were related to their degree of shared evolutionary history (46). The phylogeny for bumblebees is based on DNA sequences from 1 mitochondrial and 4 nuclear markers for 218 species (17) and was extracted using TreeSnatcher (47). We report phylogenetic least squares (PGLS) models here (e.g. Table 1) to explicitly account for phylogenetic effects relative to other predictors. Analogous ordinary least squares regression models (selected using AIC) yield nearly identical coefficients for predictor variables. Within this PGLS framework, we evaluated three models of trait evolution: Brownian motion, a random walk model (48), Ornstein-Uhlenbeck, a random walk model where traits return to an optimum (49), Pagel’s lambda, a random walk model where the covariance among traits (such as minimum temperature at which the species has been observed) are multiplied by lambda, a measure reflecting the degree of species' phylogenetic divergence and, independent, a model that assumes no phylogenetic signal (50) - using functions provided by the nlme (51) and ape (52) packages for R 3.1.0 (44). As part of the fitting the Pagel model, lambda was also estimated and reported (53, 54) (Table 1, Table S3). We calculated intercept-free models for Europe and North America using the final AIC-selected PGLS model to measure continent-specific intercepts and associated, correct standard errors for elevation shifts (55).

Supplementary Text Additional pesticide results

Neonicotinoids can harm bumblebees at low doses (20), but our results do not appear to reflect such influences. AIC-selected phylogenetic least squares models excluded both total pesticide and neonicotinoid applications, respectively (Table S1). These analyses test for effects along the latitudinal and thermal limits of species’ ranges. Bumblebees did not track their cool thermal limits even in the 1975-1986 period that wholly predates neonicotinoid use in the US (which were available after 1991, Fig. S2). Although we excluded bumblebee species that were historically rare and did not measure potential pesticide impacts on those species’ thermal and range limits, we included others that were historically very common but have recently declined to near-extinction (e.g. Bombus affinis). Additional land use change results

Extensive land use conversions to different forms of agriculture across Europe and North America predate bumblebee declines by decades to centuries in some regions of Europe. Land use changes between the baseline period (1901-1974) and subsequent climate transition periods (1975-1986, 1987-1998, and 1999-2010) are not significant in PGLS models. Despite such profound land use history differences between continents, the timing and rate of range losses and retraction from historically-known, warm thermal

5

limits share similar timing and coincide with the advent of recent, anthropogenic warming. Failures of range expansion along northern geographical and cool thermal limits conceivably reflect complex, locality-specific land use patterns that prevent species from colonizing warming landscapes. Qualitatively, such effects are improbable because failures to colonize these areas are consistent across latitudes within and across continents, including areas with very limited present-day (or historical) land use change and almost non-existence agricultural land uses (e.g. substantial areas of Canada) as well as those where land uses are both intensive and extensive (e.g. much of western Europe, central North America). However, changes in cropland and pasture extent explain no variation in bumblebee species’ northern range extent through time, nor in their observed failure to track warming along species’ cool thermal limits (Table 1).

Supplementary Acknowledgments We are grateful to data contributors from North America: Bee Biology and

Systematics Lab, USDA-ARS, Utah State University; John Ascher, National University of Singapore and American Museum of Natural History, New York, USA; Doug Yanega, University of California, Riverside (NSF-DBI #0956388 and #0956340), California, USA; Illinois Natural History Survey, Illinois, USA; Packer Lab Research Collection, York University, Canada; Canadian National Collection, Agriculture and Agri-Food Canada; Canada; Peabody Museum, Yale University; Sam Droege, USGS Patuxent Wildlife Research Center, USA; Boulder Museum of Natural History, University of Colorado, Colorado, USA. From Europe: Status and Trends of European Pollinators (STEP) Collaborative Project (grant 244090, www.STEP-project.net); Bees, Wasps and Ants Recording Society; BDFGM Banque de Données Fauniques (P. Rasmont & E. Haubruge); BWARS (UK, S.P.M. Roberts); SSIC (Sweden, B. Cederberg); Austria (J. Neumayer); EISN (Netherland, M. Reemer); CSCF (Suisse, Y. Gonseth); Poland (T. Pawlikowski); NBDC (Eire, U. FitzPatrick); FMNH (Finland, J. Paukkunen); Czech Republic (J. Straka, L. Dvorak); France (G. Mahé); NSIC (Norway, F. Odegaard); UK (S.P.M. Roberts); Italy (A. Manino); Spain (L. Castro) Global Biodiversity Information Facility (GBIF), http://gbif.org for records from North America and Europe.

6

Fig. S1. Distribution of species unique sampling locations from (A) North America and (B) Europe. Observations were used to analyze changes in latitudinal and thermal limits of species and their mean elevations through time.

7

Fig. S2 Climate change responses of 67 bumblebee species across full latitudinal and thermal limits in Europe and North America. For each measurement, the y-axis shows differences in the latitude of species’ range limits (A – northern, C – southern) or thermal limits (B – cool, C – warm), respectively, by the second time period (1975-1986) relative to baseline conditions (1901-1974). Each point represents the mean of five observations at the latitudinal or thermal limits for one bumblebee species (green circles are for Europe and pink for North America). Confidence bands (95%) of regression models are shown. Null expectations (dashed lines) are for no temporal change in latitudinal limits or thermal limits.

8

Fig. S3 Climate change responses of 67 bumblebee species across full latitudinal and thermal limits in Europe and North America. For each measurement, the y-axis shows differences in the latitude of species’ range limits (A – northern, C – southern) or thermal limits (B – cool, C – warm), respectively, by the third time period (1987-1998) relative to baseline conditions (1901-1974). Each point represents the mean of five observations at the latitudinal or thermal limits for one bumblebee species (green circles are for Europe and pink for North America). Confidence bands (95%) of regression models are shown. Null expectations (dashed lines) are for no temporal change in latitudinal limits or thermal limits.

9

A

B

Fig. S4 Change in elevation of 67 bumblebee species through interim time periods (1975-1986, panel A, and 1987-1998, panel B) relative to species mean latitudes. Elevations are calculated using mean elevations across species observations. The confidence intervals (95%) of regression slopes are shown.

10

Table S1. Summary of 31 North American bumblebee (Bombus) species in the analysis dataset. Species are sorted by increasing mean latitude (i.e. distance north of the equator), including numbers of spatially unique observations per period.

N

Bumblebee species

Mean distance from equator (km)

Mean elevation (m)

1901-1974

1975-1986

1987-1998

1999-2010

fraternus 3973 354 368 43 110 109 pensylvanicus 4111 427 5265 955 552 885 morrisoni 4217 1743 1006 188 104 382 vandykei 4322 1137 216 65 30 212 vosnesenskii 4326 883 4311 581 288 712 impatiens 4511 203 4157 817 766 2187 auricomus 4555 340 509 54 152 390 bimaculatus 4594 235 1338 445 495 924 huntii 4612 1660 1543 282 123 587 griseocollis 4624 375 1388 301 378 1499 fervidus 4665 675 3590 584 485 1079 affinis 4688 269 1859 584 329 74 appositus 4721 1881 783 140 62 207 citrinus 4775 201 465 139 118 442 bifarius 4789 1756 3916 697 245 957 melanopygus 4798 834 1630 274 186 754 nevadensis 4862 1461 606 112 41 245 vagans 4866 321 1701 595 466 736 insularis 4916 1452 1051 149 67 430 rufocinctus 4931 1175 1358 197 106 608 centralis 4972 1624 1336 231 98 535 perplexus 4999 295 779 317 150 533 terricola 5081 339 2680 723 447 396 ternarius 5085 319 937 253 208 549 occidentalis 5153 978 4442 673 365 574 flavifrons 5155 1464 1449 198 107 629 borealis 5246 343 708 79 77 153 bohemicus 5273 261 530 208 132 80 mixtus 5312 986 1150 205 66 796 sylvicola 5651 1681 671 119 36 285 frigidus 6356 659 797 114 34 241

Totals: 52539 10322 6823 18190

11

Table S2. Summary of 36 European bumblebee (Bombus) species in the analysis dataset. Species are sorted by increasing mean latitude (i.e. distance north of the equator), including numbers of spatially unique observations per period.

N

Bumblebee species

Mean distance from equator (km)

Mean elevation (m)

1901-1974

1975-1986

1987-1998

1999-2010

sicheli 4899 1933 220 108 93 360 mesomelas 4904 1768 352 194 159 483 mucidus 4931 1832 199 83 56 164 pyrenaeus 4980 1926 333 112 81 369 mendax 4981 2097 170 64 49 144 ruderatus 5406 309 1591 570 263 426 pomorum 5476 432 499 61 34 39 humilis 5505 506 1709 702 626 1575 wurflenii 5547 1234 1108 268 212 412 ruderarius 5654 477 2398 1418 793 2185 vestalis 5726 115 850 1016 1640 3311 barbutellus 5756 221 975 663 306 438 veteranus 5771 164 1246 170 103 124 rupestris 5783 351 820 598 488 1834 sylvarum 5826 276 2300 818 481 2655 campestris 5834 160 1311 1169 723 1680 terrestris 5839 153 3777 4511 4278 15810 lapidarius 5851 167 3778 3626 5441 15201 quadricolor 5869 726 108 119 31 299 hortorum 5882 223 3699 3441 3199 8889 sylvestris 5884 251 1259 1436 1191 2485 pascuorum 5897 172 8176 6909 9145 24329 soroeensis 5902 614 1748 446 478 3660 pratorum 5911 206 3901 4255 4702 11557 muscorum 5930 131 1646 602 546 1191 magnus 5938 339 293 151 177 486 subterraneus 5941 317 1225 229 100 1282 cryptarum 5954 469 234 180 168 538 monticola 5971 710 435 704 564 2401 lucorum 5975 240 4325 4887 5636 15782 hypnorum 6028 177 2511 917 860 5908 bohemicus 6051 282 1627 1588 1388 3621 norvegicus 6096 281 227 219 66 422 distinguendus 6234 94 1287 181 76 947 jonellus 6280 274 1500 823 887 3677

12

N

Bumblebee species

Mean distance from equator (km)

Mean elevation (m)

1901-1974

1975-1986

1987-1998

1999-2010

sporadicus 7073 326 175 108 19 264

Totals: 58012 43346 45059

13

Table S3. Phylogenetic generalized least squares models including including US-only pesticide and land use change effect on bumblebee species warm thermal and southern latitudinal limits. Changes in latitude, thermal or elevation variables observed by the time period listed for each species (relative to the 1901-1974 baseline period) are regressed against predictors listed on the left. Models reported in each column were selected using Akaike’s Information Criterion (AIC), which can include statistically non-significant variables. Sample sizes in each time period were also tested but were neither significant nor included by AIC. Variable coefficients are given with standard errors in parentheses. A dashed line indicates that this variable was not part of the AIC-selected model. OLS regression summary statistics (adjusted R2) are provided to enable comparison with PGLS results.

Southern latitudinal limit Warm thermal limit Predictors 1987-1998 1999-2010 1987-1998 1999-2010

Intercept -463.09 (753.8) 147.0 (429.3) -8.37 (2.34) -6.14 (2.08) Latitudinal or thermal limit (1901-1974)

0.22 (0.21) 0.04 (0.11) 0.3 (0.15) 0.18 (0.11)

Covariates Δ Crop land (1999-2010) - - - -

Δ Pasture (1999-2010) - - - -

Models of trait evolution AIC (Independent) 432.7 405.6 143.8 132.7

AIC (Brownian motion) 446.6 419.4 152.1 138.9

AIC (Ornstein-Uhlenbeck) 431.8 405.4 -- 127.8

AIC (Pagel) 434.6 402.3 144.7 134.7 Pagel's λ 0.16 0.76 -0.12 0.01

Equivalent OLS regression summary statistics Adjusted R2 0.081 0.027 0.071 0.00

14