adjacent to a patch of 36 flowerless plants
(from six different species; see supplemen-
tary materials for details). The colony and
plants were replaced every 3 weeks, and we
assessed plant damage and bee activity on
weekdays (except when bees were inactive
owing to weather). Bumble bee workers
damaged plants in our experimental patches
throughout this phase of the experiment
(Fig.4A);furthermore,aspredicted,weob-
served a strong decline in damage by the end
of April, at which point local floral resourceswere relatively abundant ( 26 , 27 )(fig.S2).
These results thus complement the findings
from our laboratory studies showing that dam-
aging behavior is influenced by pollen avail-
ability. As we only placed one microcolony at
a time during this study, some temporal varia-
tion in damage might alternatively be explained
by colony effects, although our laboratory exper-
iments revealed minimal intercolony variation
in damaging behavior (Fig. 3B and table S2).
The second phase of our 2018 outdoor ex-
periment started on 4 June 2018 and continued
through 20 July. This phase replicated the de-
sign of the first, except that we placed 100 plants
in flower adjacent to the focal patch of flower-
less plants (~1 m from the hive) to ensure the
availability of local floral resources (see sup-
plementary materials for details). The inten-
sity of damage inflicted during this phase was
far lower than that observed during phase 1 (Fig.
4A), which again is consistent with the pre-
diction that abundant floral resources reduce
damaging behavior. Furthermore, in addi-
tion toB. terrestris,workersfromatleasttwo
other bumble bee species damaged flower-
less plants in our experimental patch during
phase 2. Specifically, fourB. lapidariusand
threeB. lucorumworkers were observed in-
flicting leaf damage in a manner similar to
B. terrestris(table S3). These observations
also confirm that damaging behavior is not
exclusive to commercial bumble bee hives butPashalidouet al.,Science 368 , 881–884 (2020) 22 May 2020 2of4
Fig. 2. Effects of bee damage on flowering time.
Flowering time ofSolanum lycopersicumandBrassica
nigraplants subjected to three treatments: (i) bee
damage, (ii) mechanical damage, and (iii) no damage
(control). Bars represent least squares means ±
confidence intervals. Data were converted to a binary
response variable (flowering or not flowering, for each
plant on each day), and treatment effects were analyzed
using generalized additive models (GAMs), avoiding
assumptions of linearity. Interactions between“time
since damage”and plant were modeled by smooth
functions (table S1). (A)S. lycopersicum: All treatments
had highly significant effects on flowering time (bee
damage:P< 0.001, estimate = 13.697; mechanical
damage:P< 0.001, estimate = 9.131; control:P< 0.001,
estimate =−24.279). The interaction between time and
plant (smooth term) was also significant:P<0.001,
estimated degrees of freedom (edf) = 0.974, Chi square
(Chi.sq) = 39.49. GAM explained 77% of deviance, with
coefficient of determination (R^2 )=0.794over4800
observations.n= 20 plants per treatment. (B)B. nigra:All
treatments had highly significant effects on flowering
time (bee damage:P< 0.001, estimate = 2.955;
mechanical damage:P< 0.001, estimate = 2.601;
control:P<0.001,estimate=−5.747). The interaction
between time and plant (smooth term) was also
significant:P< 0.001, edf = 0.994, Chi.sq = 181.9.
GAM explained 48% of the deviance, withR^2 = 0.532 over
1200 observations.n= 10 plants per treatment.253035404550556065707580Bee damage Mechanical damage Control
TreatmentTime since damage (days)510152025303540Bee damage Mechanical damage Control
TreatmentTime since damage (days)A S. lycopersicum B B. nigra
>Fig. 1.Bombus terrestrisworkers damagingSolanum melongenaleaves.(A) Sequential images of a
worker penetrating a leaf with its proboscis (taken over ~3 s). (B) A worker cutting into a leaf with its
mandibles. (C) Characteristic bee-inflicted damage.
RESEARCH | REPORT