ORIGINAL PAPER

Reward and non-reward learning of flower coloursin the butterfly Byasa alcinous (Lepidoptera: Papilionidae)

Ikuo Kandori & Takafumi Yamaki

Received: 17 January 2012 /Revised: 13 July 2012 /Accepted: 14 July 2012 /Published online: 1 August 2012# Springer-Verlag 2012

Abstract Learning plays an important role in food acquisitionfor a wide range of insects. To increase their foraging efficien-cy, flower-visiting insects may learn to associate floral cueswith the presence (so-called reward learning) or the absence(so-called non-reward learning) of a reward. Reward learningwhilst foraging for flowers has been demonstrated in manyinsect taxa, whilst non-reward learning in flower-visitinginsects has been demonstrated only in honeybees, bumblebeesand hawkmoths. This study examined both reward and non-reward learning abilities in the butterfly Byasa alcinous whilstforaging among artificial flowers of different colours. Thisbutterfly showed both types of learning, although butterfliesof both sexes learned faster via reward learning. In addition,females learned via reward learning faster than males. To thebest of our knowledge, these are the first empirical data on thelearning speed of both reward and non-reward learning ininsects. We discuss the adaptive significance of a lower learn-ing speed for non-reward learning when foraging on flowers.

Keywords Positive associative learning . Appetitivelearning . Negative associative learning . Aversive learning .

Aversion learning . Habituation

Introduction

Avariety of insects rely extensively on learning for all major lifeactivities, including feeding, predator avoidance, aggregation,

social interaction and sexual behaviour (Dukas 2008). Learningis a fundamental mechanism for adjusting behaviour to envi-ronmental change (Stephens 1993; Dunlap and Stephens 2009).Although some aspects of learning are costly (e.g. Mery andKawecki 2003, 2004; Burger et al. 2008), learning generallyincreases fitness (Dukas and Bernays 2000; Dukas and Duan2000; Raine and Chittka 2008). Many studies have shown thatinsects can develop a positive association between visual and/orolfactory cues and resources such as nectar or oviposition sites(e.g. Shafir 1996; Cnaani et al. 2006; Dukas 1999; Dukas andBernays 2000; Weiss and Papaj 2003). This is often called‘positive (associative) learning’ (e.g. Papaj et al. 1994; Costaet al. 2010; Rodrigues et al. 2010), ‘reward learning’ (e.g.Menzel 1999; Thum et al. 2007) or ‘appetitive learning’ (e.g.Vergoz et al. 2007; Rodrigues et al. 2010). In addition, insects ina range of taxa can learn to associate visual and olfactory cueswith negative or aversive stimuli such as salt, shock, and toxinsand to avoid or to escape from those cues (e.g. Berenbaum andMiliczky 1984; Tully and Quinn 1985; Lee and Bernays 1990;Matsumoto and Mizunami 2002). This is often called ‘aversionlearning’ (e.g. Bernays 1993; Bowdish and Bultman 1993;Blackiston et al. 2008) or ‘aversive learning’ (e.g. Vergoz etal. 2007; Rodrigues et al. 2010). Learning to associate biolog-ically relevant cues with either positive or negative stimuli canincrease an insect’s fitness (e.g. Dukas and Bernays 2000).Some insects can also learn to associate the cues with theabsence of rewards or unrewarding stimuli and to avoid thosecues (e.g. Papaj et al. 1994). This type of learning is sometimescalled ‘non-reward learning’ (e.g. Dicke et al. 2011). Whilst inaversive learning insects learn to avoid cues associated withnegative or minus stimuli, in non-reward learning, they learn toavoid cues associated with absent or zero stimuli. Non-rewardlearning can also increase an insect’s fitness because unsuccess-ful foraging may be costly in terms of energy and time expen-diture. However, very few studies have demonstrated non-reward learning in insects.

Communicated by: Sven Thatje

I. Kandori (*) : T. YamakiLaboratory of Entomology, Faculty of Agriculture,Kinki University,Naka-machi,Nara 631-8505, Japane-mail: kandori@nara.kindai.ac.jp

Naturwissenschaften (2012) 99:705–713DOI 10.1007/s00114-012-0952-y

Among pollinator insects, positive or reward learningwhilst foraging on flowers has been investigated in many taxa,including bees (e.g. Menzel 1985, 1993; Dukas and Real1991), wasps (e.g. Shafir 1996; Sato and Takasu 2000; Takasuet al. 2007), blowflies (Fukushi 1989), butterflies (e.g. Swihartand Swihart 1970; Lewis 1986; Kinoshita et al. 1999; Kandoriet al. 2009) and moths (e.g. Kelber 1996, 2002; Cunninghamet al. 1998, 2004). Meanwhile, few studies have exploredaversive or non-reward learning, although many studies in-cluded aversive or non-reward conditioning procedures in theexperiments. For example, when insects are trained to twotypes of cue—each associated with rewarding and unreward-ing or deterrent stimuli—during the conditioning procedure(so-called differential conditioning), the choice of the reward-ing cue often increases, which inevitably decreases the choiceof the unrewarding or deterrent cue (apparent avoidance; e.g.Masters 1991; Gigord et al. 2002; Chittka et al. 2003; Dyerand Chittka 2004; Adler and Irwin 2005; Gegear et al. 2007;Internicola et al. 2007, 2008; Rodrigues et al. 2010). However,this phenomenon alone does not demonstrate aversive or non-reward learning, and it can be explained by positive learningalone. To our knowledge, non-reward learning or aversivelearning of pollinator insects has been demonstrated only inhoneybees (Srinivasan et al. 1994; Horridge 2007; Vergoz etal. 2007), bumblebees (Simonds and Plowright 2004) andhawkmoths (Kelber 1996; Blackiston et al. 2008). It is notknown whether butterflies possess non-reward or aversivelearning abilities. In nature, pollinators may experience unre-warding and rewarding flowers more frequently than flowerswith aversive stimuli, such as shock, salt or toxins. Conse-quently, they may have a greater chance to learn via non-reward learning than through aversive learning.

Dukas and Real (1993) suggested that under certain exper-imental conditions, bumblebees learn and remember only therewarding flowers they visit, even if they experience bothrewarding and unrewarding flowers (Dukas and Real 1993).This suggests that because of their limited abilities to processand store information (Waser 1986), insects preferentiallyremember more useful information about rewarding flowersthan less useful information about unrewarding flowers, mak-ing it more difficult and slower to learn via non-reward learn-ing than reward learning. However, there has been noexperimental comparison of the learning speeds of these twotypes of learning within pollinator species.

In this study, we examined both the reward and non-reward learning abilities of the butterfly Byasa alcinous(Klug) whilst foraging for artificial flowers of differentcolours. We specifically addressed two questions: Does thisbutterfly possess both reward and non-reward learning abil-ities? If so, by which type of learning do the butterflies learnfaster? To the best of our knowledge, this is the first empir-ical comparison of reward and non-reward learning speedsin insects.

Materials and methods

Experimental preparation

Larvae and pupae of B. alcinous were obtained in Kyota-nabe, Kyoto, Japan, from June to July 2003 and fromSeptember to October 2005. Approximately 10 larvae werereared together on fresh Aristolochia debilis Sieb. et Zucc.leaves in 450-ml transparent plastic cups at 25 °C and alight/dark (L/D) cycle of 14 h:10 h in a growth chamber inthe laboratory. After eclosion, each adult was numbered onthe hind wings with an oil-soluble marker and kept withoutfood at 15 °C and 12-h:12-h L/D until the experiments werestarted. All butterflies were of similar age (1–3 days) at thestart of the experiments.

Experimental location

We examined the foraging behaviour of butterflies in anindoor incubation room (2.7×1.8×2.0 m) on the Nara cam-pus of Kinki University (see Kandori et al. 2009 for details).In the incubation room, light was provided from the ceilingby 10 fluorescent tubes (Truelite EX-VS, 40 W; ELC, Phil-adelphia, PA, USA). The walls inside the room were cov-ered with a black nylon net, except for the ceiling, and roomtemperature was controlled at 25 °C.

Artificial flowers

Artificial flowers made from disks of coloured paper (5 cmin diameter) were used as the stimuli (Daiei Training colour200, Tokyo, Japan). Rewarding flowers were made from adisk of coloured paper that had a 5-mm-wide hole at thecentre with an attached Eppendorf tube (1.5 ml) containing a10 % sucrose solution. Unrewarding flowers consisted onlyof a disk of coloured paper with a hole at the centre. Allflowers were set on a green plastic circular frame (30 cm indiameter). The number of flowers and their colours differedaccording to the experiment (see below). The plastic framewas elevated 60–70 cm above the floor from the centre ofthe experimental room and exposed to the butterflies.

General experimental rules

We defined a ‘visit’ as a positive response when a butterflylanded and extended its proboscis towards the colouredpaper. In all behavioural experiments, 10–40 butterflieswere released at the same time and allowed to visit theartificial flowers. When a butterfly visited any flower for aset number of times (i.e. five times, except during thetraining session for the second part of experiment 2, inwhich each butterfly was allowed to visit only once; seebelow), we temporarily removed that butterfly until the end

706 Naturwissenschaften (2012) 99:705–713

of the experiment, part or session. Recent studies havereported that bumblebees can copy the flower choice ofexperienced foragers (Leadbeater and Chittka 2005; Wordenand Papaj 2005). However, in butterflies, such behaviourhas not been reported. A preliminary experiment revealedthat naive Peris rapae butterflies chose an artificial flowerwith and without another individual on it nearly equally (I.Kandori, unpublished data). This means that the presence ofanother conspecific on a flower does not act as either anattractant or a deterrent stimulus for P. rapae. We assumethat this is also the case for B. alcinous and that multipleindividuals in the same place would not have a significanteffect on their flower choices.

Experiment 1: Innate colour choices among 12 colours

To determine which colours should be used in the colour-learning experiment, innate colour choices were investigat-ed by allowing naive butterflies to visit 12 artificial flowerscoloured red, red-purple, purple, blue, green, yellow-green,yellow, orange, brown, light blue, white and pink. Thepreference of each butterfly was recorded over five visits.The experiment was conducted in July 2003. During theexperiment, the order of 12 artificial flowers on the framewas changed randomly every 20 min so that flower positionwould not become a factor.

Experiment 2: Reward learning

This experiment examined reward colour learning and itslearning speed. Based on the results of experiment 1, redand orange were used as the two artificial flower colours(see “Results”). Colour traits, which were measured using areflectance spectrophotometer (High Sensitivity Spectro-photometer USB2000, Ocean Optics Inc., Dunedin, FL,USA), are shown in Fig. 1. The experiment consisted oftwo parts (Fig. 2). In the first part, we examined the innatecolour choice for two colours in naive butterflies. We

presented eight flowers (four flowers of each colour) thatwe set alternatively on the frame without rewards, and eachindividual butterfly was allowed to visit the flowers fivetimes. The frame was rotated 45° every 20 min so thatflower position would not be a factor. All individuals werethen separated into two groups: those that visited red moreoften and those that visited orange more often. In the secondpart of the experiment, we examined flower colour learningthrough a reward training session, followed by a test ses-sion. In the reward training session, butterflies from each ofthe two groups were trained to feed on the colour less visitedduring the first part; that is, four flowers of the less visitedcolour were set with rewards, and each individual wasallowed to visit only one of the four flowers, drink thereward and spontaneously leave the flower. Butterflies thatdid not visit a flower spontaneously were placed on theflower manually. These individuals usually learned to feedspontaneously on the flower within two training sessions.The test session was similar to the first part of this experi-ment in that we set eight flowers (four of each of the twocolours) without rewards, and the colour choice of eachindividual was recorded over five visits. The two sessionswere conducted alternately every day and repeated fivetimes (days) each. Therefore, the second part of this exper-iment was conducted for 10 days. Each training or testsession was conducted daily between 1000 and 1400 hours;that is, the interval between training and testing for eachindividual was approximately 1 day (20–28 h), and the inter-training interval was approximately 2 days. We rejectedindividual butterflies that did not finish the task within4 h. Approximately half of the individuals completed all ofthe tasks (six tests, including the innate colour choice test,with five training sessions between each test). Only thesewere used in the statistical analysis (8 and 12 females and8 and 7 males that were trained to red and orange, respec-tively). The experiment was conducted from July to August2003.

Experiment 3: Non-reward learning

This experiment examined whether a butterfly utilises non-reward colour learning. Assuming that non-reward learning ismore difficult (see “Introduction”), we designed an experi-ment to directly detect any effects of non-reward learning.

The experiment consisted of two parts (Fig. 2). In the firstpart, we examined the innate colour choice between red andorange in naive butterflies using the same method describedin the first part of experiment 2. Subsequently, we choseonly individuals that visited red three times or more out offive visits to effectively detect possible non-reward learningfor red. In the second part, we trained these butterflies tounrewarding red and then retested their choice between redand orange. This part was conducted on the day after

0102030405060708090

100

300 400 500 600 700

red

orange

Ref

lect

ance

(%

)

Wavelength (nm)

Fig. 1 Spectral reflectance of the artificial flowers of different colourused in the reward and non-reward learning experiments (experiments2, 3 and 4). Relative spectral reflectance is normalised to a whitestandard (magnesium oxide)

Naturwissenschaften (2012) 99:705–713 707

completion of the first part. It consisted of two sessions: anon-reward training session in the morning (1000–1200 hours) followed by a test session in the afternoon(1200–1600 hours). Therefore, the inter-training intervalfor each individual was between a few seconds and 2 h,and the interval between the last training and testing wasbetween a few minutes and 6 h. In the non-reward trainingsession, four red flowers were set without rewards andbutterflies were allowed to visit them freely. The test sessionwas similar to the first part of this experiment in that we seteight flowers (four red and four orange) without rewards,and the colour choice of each individual was recorded overfive visits. Then, we compared the percentage choice of thebutterflies for red between before and after non-rewardtraining to red. In addition to the group with unrewardedexperience on red, we examined a control group that had nounrewarded experience on red (Fig. 2). The experimentalprocedure for the control group was the same as the groupwith unrewarded experience, except that the control groupwas not exposed to any flower colours in the non-rewardtraining session in the second part of the experiment. Onlythose that finished all tasks were used in the statisticalanalysis, i.e. 27 females and 24 males for the group withunrewarded experience and 26 females and 25 males for thecontrol group. The experiment was conducted in October2005.

Experiment 4: Speed of non-reward learning

We examined whether the avoidance of the unrewardingcolour increased with the amount of non-reward training tothat colour and estimated the speed of increase (i.e. non-reward learning speed). To compare the speeds of rewardand non-reward learning, the experimental procedure for thetwo learning experiments (i.e. experiments 2 and 4) must beidentical, except whether they were trained to flowers withor without reward. If we were to apply the protocol ofexperiment 2 to experiment 4, the design of the latter wouldbe a repetition of one non-reward training in the trainingsession followed by five choices without rewards in the testsession. However, it would not allow us to detect the preciseeffect of non-reward learning and its speed. This is because

five choices without rewards in the test session equals fivenon-reward trainings, which masks the effect of one non-reward training in the training session. Therefore, we usedthe same experimental procedure as that used for the groupwith unrewarded experience in experiment 3, except that wecounted the number of visits (trainings) to unrewarding redflowers by each individual butterfly during the non-rewardtraining session in the second part of the experiment (Fig. 2).Then, the count data were used to investigate whether thepercentage choice for red in the test session decreases whenthe individual butterfly visits unrewarding red more timesduring the training session. If so, the rate of decrease (i.e. thespeed of non-reward learning) was compared with that ofreward learning. The experiment was conducted only forfemales in November 2005. Different individuals were usedin each of experiments 1–4.

Statistical analysis

In experiment 2, to compare the reward learning speedsbetween the two sexes or colours, we used a general non-linear learning model:

P ¼ 1� 1� P0ð Þe�aN ;

where P is the proportion of butterfly visits to a rewardingflower colour, a is the learning speed, and N is the numberof trainings on a rewarding flower colour. P0 refers to P atN00, or the innate colour preference. This model illustratesthe general form of the learning curve: The rate of change inlearning is initially higher and diminishes gradually as theindividual is trained on more flowers. The proportion ofvisits to a rewarding flower colour approaches 1 asymptot-ically. This model describes the butterfly learning seen inour previous experiments (Kandori et al. 2009). This learn-ing curve was fitted to six data points, representing six tests,for each individual butterfly to estimate P0 and a. Then, weused a fixed effects analysis of variance [ANOVA; generallinear model (GLM) with type III sums of squares] to testfor effects on learning speed (log+1-transformed), with sex,colour and their interactions as independent factors. In ex-periment 3, we used the Wilcoxon signed-rank test to

Red: 10% sucroseExperiment 2: 1 visit 5 timesReward learning Orange Red Orange: 10% sucrose Orange Red Total 10 days

1 visitRed: no rewards

Experiment 3: Free visits: numbersNon-reward learning not counted

OnceTotal 1 day

Experiment 4: Red: no rewardsSpeed of Free visits: numbersnon-reward learning counted individually

Red Red-training

No rewards No rewards5 visits

Red Control None5 visits

(5 choices) (5 choices)

Orange Red-training

RedOrange-training

Red Red-training

1st part Innatelymore visited

colour

Groupassignment

2nd partRepetition ofthe 2nd partInnate choice test Training Test

Fig. 2 Schematic diagramshowing the sequence of testingand training in experiments 2, 3and 4. In the second part ofexperiments 3 and 4, we usedonly individuals that innatelychose red more often, and weallowed them to visit non-reward red flowers freely for2 h in the training session

708 Naturwissenschaften (2012) 99:705–713

compare the proportion of visits to red before and after thenon-reward training session for the group with unrewardedexperience and the control group. To compare the learningspeeds between sexes or groups, we used the same nonlinearlearning model described above. Here, all parameters are thesame as above, except that P is the proportion of butterflyvisits to an opposite colour (i.e. orange) than the non-rewardtraining colour (i.e. red) and N consists of only 0 and 1,which represent the tests before and after the non-rewardtraining session, respectively. This learning curve was fittedto two data points for each individual butterfly to estimateP0 and a. Then, we used a fixed effect ANOVA (GLM withtype III sums of squares) to test for effects on learning speed(log+1-transformed), with sex, group (whether butterflieshad unrewarded experience or not) and their interactions asindependent factors. In experiment 4, linear regression anal-ysis was performed to determine whether the percentagechoice for red decreased with increasing unrewarded expe-rience on red flowers. To analyse the difference in learnedperformance between reward and non-reward learning, thepreference rate (the percentage choice) of the rewardingcolour in females in experiment 2 and the avoidance rate(the percentage choice of the opposite colour) of the unre-warding colour in females in experiment 4 were comparedfor the same number of trainings using the Mann–WhitneyU test. IBM SPSS statistics 20 (IBM SPSS 2011) was usedfor all statistical analyses.

Results

Innate colour choices

From the 12 possible colours, naive females chose red (23 %),followed by purple (21 %), orange (20 %) and blue (9 %) in atotal of 160 visits. Naive males chose purple (36 %), followedby red (22 %), orange (14 %) and blue (11 %) in a total of 110visits. Since the percentage choice of the two colours wassimilar within each sex, we selected red and orange as thetwo flower colours used in subsequent experiments.

Reward learning

When they were trained to rewarding colours, both femalesand males of B. alcinous exhibited typical learning curves;that is, as the number of times an individual was rewarded on aparticular flower colour increased, the rate of selection of thatcolour increased (Fig. 3).When the two training colour groupswere combined, the mean proportion of visits to a trainedcolour was always higher for females compared to males afterone to five trainings (Fig. 3). The ANOVA for learning speedindicated that there was a significant effect of sex and a non-significant effect of colour and their interaction (Table 1). This

suggests that females learn a rewarding colour faster thanmales and that the learning speeds are not different betweenthe two training colours for each sex.

Non-reward learning

Butterflies in the group with unrewarded experience on redsignificantly reduced their proportion of visits to red in bothsexes (26–30 % reduction; Table 2). In contrast, butterflies inthe control group that had no unrewarded experience on red didnot significantly reduce their proportion of visits to red (Ta-ble 2). The ANOVA for learning speed showed that there was asignificant effect between groups (Table 3). These results sug-gest that the butterflies learned to avoid the unrewarding colour.However, contrary to reward learning, there was not a signifi-cant effect of sex on the speed of non-reward learning (Table 3).

Comparison of learning speed between rewardand non-reward learning

The more times that female butterflies experienced unre-warding red, the less frequently they visited red (regression:y0−0.087x+0.700, r200.715, n034, F080.39, P<0.0001;Fig. 4). This demonstrates non-reward learning inB. alcinous.Although the starting points of the two types of learning curve

0.0

0.5

1.0

0 1 2 3 4 5

Number of trainings to a rewarding colour

Prop

ortio

n of

vis

its to

the

trai

ned

colo

ur

Fig. 3 Effect of training to a rewarding colour for females (closedcircles) and males (open circles) in B. alcinous. Because flower colourdid not affect learning speed (Table 1), the two groups with differenttraining colours (red and orange) were combined within each sex.There were 20 females and 15 males. Values represent the means±standard deviation

Table 1 ANOVA of reward learning speed

Source df MS F P

Sex 1 0.109 4.850 0.035

Colour 1 0.008 0.366 0.550

Sex×colour 1 0.007 0.324 0.573

Error 31 0.023

Sex refers to male and female effects. Colour refers to the effect of thetwo flower colours (red and orange) that were used in reward trainingfor the butterfly

Naturwissenschaften (2012) 99:705–713 709

were similar, there was a relatively rapid increase in thepreference rate via reward learning and a more gradual in-crease in the avoidance rate via non-reward learning (Fig. 5).Performance was significantly higher for reward learning thannon-reward learning after three to five training sessions(Fig. 5). This indicates that B. alcinous learns via rewardlearning faster than it does via non-reward learning.

Discussion

We found that the butterfly B. alcinous learns to associateflower colours not only with the presence of nectar but alsowith the absence of nectar; that is, B. alcinous uses bothreward and non-reward colour learning whilst foraging onflowers. The speed of reward learning differed by sex, butnot by flower colour (Table 1 and Fig. 3), consistent withour previous study (Kandori et al. 2009).

To explain the faster reward learning in female than in malebutterflies, Kroutov et al. (1999) hypothesised that the com-plexity of female behaviour, such as locating oviposition sitesand determining host-plant suitability, may enhance theirlearning behaviour. Alternatively, we hypothesised that fe-male butterflies require more nectar than their male counter-parts to produce and oviposit eggs; thus, they must forage

more efficiently, which has selected for enhanced learningability (Kandori et al. 2009). In contrast, the speed of non-reward learning did not differ by sex in the present study(Tables 2 and 3). However, further investigations should becarried out to confirm this result. Experiment 3 was designedonly to determine whether non-reward colour learning oc-curred, not to investigate the speed of non-reward learning.

Many studies have shown that learning speed differs byflower colour (e.g. Fukushi 1989; Weiss 1997; Kinoshita etal. 1999). This may occur partly because the learning speedon a certain colour positively correlates with an innatepreference for that colour (Dukas and Real 1991; Weiss1997). Our studies failed to detect a difference in the learn-ing speed between flower colours, perhaps because thebutterflies innately preferred both colours used in the dual-choice test (red and orange in this study).

Table 2 Proportion of visits to red before and after the non-rewardtraining session in which butterflies had an unrewarded experience onred and butterflies had no unrewarded experience

n % visits to red (mean ± SD) Pa

Before After

Butterflies with unrewarded experience on red

Females 27 69.6±12.9 40.0±17.5 <0.0001

Males 24 67.5±11.5 41.7±17.6 <0.0001

Butterflies with no unrewarded experience

Females 26 69.2±12.9 62.3±15.3 NS

Males 25 64.8±15.6 56.8±19.7 NS

NS not significantaWilcoxon signed-rank test

Table 3 ANOVA of non-reward learning speed

Source df MS F P

Sex 1 0.002 0.128 0.722

Group 1 0.595 34.522 0.000

Sex×group 1 0.005 0.277 0.600

Error 98 0.017

Sex refers to male and female effects. Group refers to the effect of thetwo groups (butterflies with and without an unrewarded experience onred)

0

0.5

1

0 1 2 3 4 5 6 7

17

2

1

4

3 43Pr

opor

tion

of v

isits

to r

ed

Number of trainings to unrewarded red

Fig. 4 Effect of training to unrewarding red artificial flowers on visitsto that colour in B. alcinous females. The numbers in the figure refer tothe number of individuals. Values represent the means±standarddeviation

0

0.5

1

0 1 2 3 4 5

**** ***

Number of trainings

Pref

eren

ce/a

void

ance

rat

e

Fig. 5 Performance comparison of reward learning (circles) and non-reward learning (squares) in B. alcinous females. Mean values aretaken from Figs. 2 and 3. The y-axis shows the preference rate for arewarding colour and the avoidance rate of an unrewarding colour forreward and non-reward learning, respectively. Asterisks in the figureindicate that the proportion of avoidance in non-reward learning dif-fered significantly from the proportion of preference in reward learningfor the same number of trainings (Mann–Whitney U test: **P<0.01;***P<0.001). Note that the inter-training interval for reward learningwas approximately 2 days, whilst that of non-reward learning wasbetween a few seconds and 2 h

710 Naturwissenschaften (2012) 99:705–713

Asymmetric confusion may also affect the learning speedof new colours. Blackiston et al. (2011) found that monarchbutterflies (Danaus plexippus) trained to red also visitedorange, but not vice versa (Blackiston et al. 2011). Theyexplained this phenomenon from the viewpoint of stimulusbrightness; that is, butterflies trained to the dimmer colouralways made mistakes on the brighter colour, whereas thereverse was not true. However, their explanation may not bethe case for B. alcinous (this study) and Idea leuconoe(Kandori et al. 2009), in which both butterflies showedneither different learning speeds nor asymmetric confusionbetween red and orange. We think that asymmetric confu-sion occurs between two colours of different innate prefer-ence rather than different stimulus brightness. From thisperspective, monarch butterflies showed asymmetric confu-sion because they innately prefer orange to red, whilst B.alcinous and I. leuconoe did not show asymmetric confu-sion because they innately prefer both red and orange.

We also found that B. alcinous learned faster via rewardlearning than via non-reward learning. Our results suggestthat one rewarded experience on a certain flower colour mayhave a more positive learning effect on a subsequent flowerchoice than the negative learning effect of even four unre-warded experiences on the same colour: In experiment 2,most butterflies ultimately approached 100 % preference forthe rewarding colour, i.e. they chose the rewarding colourfive times in the test session, which means that followingone rewarded experience, they continued to choose the sameflower colour despite four unrewarded experiences. Thedifferences in speed between the two types of learning canbe considered adaptive for two reasons. First, even if acertain plant species in the field produces nectar that butter-flies can utilise, a plant may have more unrewarding flowersthan rewarding flowers due to nectar consumption by otherflower visitors. In this case, to achieve a learned preferencefor that plant species, the increase in preference with onerewarded experience on a flower should be larger than thedecrease in preference with one unrewarded experience onthe same plant species. Second, because of limited abilitiesto process and store information (Waser 1986), insects mayremember more useful information about rewarding flowersin preference to less useful information about unrewardingflowers. Thus, non-reward learning is more difficult andtakes longer than reward learning.

We could not accurately compare the learning speedbetween the two types of learning because the experimentaldesigns differed somewhat between experiments 2 and 4,e.g. the interval between trainings was 2 days in experiment2, but <2 h in experiment 4. However, since learned associ-ation may weaken over time (e.g. Fukushi 1989; Kelber1996), if we had trained butterflies successively in experi-ment 4 using the same interval as in experiment 2, the non-reward learning speed might have been even lower, making

the difference in the speed of the learning types even larger.Alternatively, we may not have detected non-reward learn-ing in butterflies. This is because, in addition to the diffi-culty in establishing non-reward learning, butterflies mayhave a shorter memory for non-reward learning than forreward learning. According to Matsumoto and Mizunami(2002), in the cricket Gryllus bimaculatus, appetitive con-ditioning leads to long-lasting memory retention with nosignificant decay 4 days after training, whereas retentionafter aversive conditioning disappears 1 day after training(Matsumoto and Mizunami 2002). Camponotus fellah antsmay also have difficulty in establishing non-reward or aver-sive learning or in retaining their memories because they didnot show repellence to an odour paired with quinine, butonly showed preference to an odour paired with sugar, afterdifferential conditioning in a Y-maze (Josens et al. 2009).

At this time, it is unclear whether non-reward learningreflects negative associative learning or a specific type ofhabituation to an initially preferred colour. To our knowl-edge, only four studies have demonstrated the non-rewardlearning of flower-visiting insects (a decrease in preferencefor the unrewarded flower colour or pattern). However, noneof these studies clarified whether this was due to negativeassociative learning or habituation. It has been proposed tobe negative associative learning (Srinivasan et al. 1994;Kelber 1996; Horridge 2007) as well as habituation(Simonds and Plowright 2004). It is very difficult to designexperiments to address this issue. In our opinion, habitua-tion to floral visual and olfactory cues may be rare because itwould interfere with positive associative learning thatenhances constant visits to the rewarding flower type.Therefore, the observed non-reward learning likely reflectednegative associative learning.

We conclude that naive pollinators in nature can findrewarding flower species more efficiently via both rewardand non-reward learning; that is, insects may initially visit acertain flower by innate preference. If this flower is reward-ing, they quickly increase their constancy to that flowerspecies through reward learning. If the flower species isunrewarding and common, frequent visits to that flowerspecies enhance non-reward learning to avoid that flowerspecies. They can identify new rewarding flower specieswith a reduced loss of energy and time if they avoid foragingon such abundant, innately attractive but unrewarding flow-er species. However, if the flower species is unrewardingand uncommon, the speed of non-reward learning is so slowthat they may locate a new rewarding flower species beforethey learn to avoid the unrewarding flower species.

Acknowledgments We sincerely thank Drs. T. Sugimoto, Y.Sakuratani, E. Yano and D. R. Papaj for their valuable advice.We also thank H. Nakai, H. Narita and Y. Kinoshita for assistancewith the preliminary experiments. All experiments complied withthe current laws of Japan.

Naturwissenschaften (2012) 99:705–713 711

References

Adler LS, Irwin RE (2005) Ecological costs and benefits of defenses innectar. Ecology 86:2968–2978

Berenbaum MR, Miliczky E (1984) Mantids and milkweed bugs:efficacy of aposematic coloration against invertebrate predators.Am Midl Nat 111:64–68

Bernays EA (1993) Aversion learning and feeding. In: Papaj DR,Lewis AC (eds) Insect learning: ecological and evolutionaryperspectives. Chapman & Hall, New York, pp 1–17

Blackiston DJ, Casey ES, Weiss MR (2008) Retention of memory throughmetamorphosis: can a moth remember what it learned as a caterpillar?PLoS One 3:e1736. doi:e173610.1371/journal.pone.0001736

Blackiston D, Briscoe AD, Weiss MR (2011) Color vision andlearning in the monarch butterfly, Danaus plexippus(Nymphalidae). J Exp Biol 214:509–520. doi:10.1242/jeb.048728

Bowdish TI, Bultman TL (1993) Visual cues used by mantids inlearning aversion to aposematically colored prey. Am Midl Nat129:215–222

Burger JMS, Kolss M, Pont J, Kawecki TJ (2008) Learningability and longevity: a symmetrical evolutionary trade-offin Drosophila. Evolution 62:1294–1304. doi:10.1111/j.1558-5646.2008.00376.x

Chittka L, Dyer AG, Bock F, Dornhaus A (2003) Bees trade offforaging speed for accuracy. Nature 424:388. doi:10.1038/424388a

Cnaani J, Thomson JD, Papaj DR (2006) Flower choice and learning inforaging bumblebees: effects of variation in nectar volume andconcentration. Ethology 112:278–285

Costa A, Ricard I, Davison AC, Turlings TCJ (2010) Effects ofrewarding and unrewarding experiences on the response to host-induced plant odors of the generalist parasitoid Cotesia margin-iventris (Hymenoptera: Braconidae). J Insect Behav 23:303–318.doi:10.1007/s10905-010-9215-y

Cunningham JP, West SA, Wright DJ (1998) Learning in the nectarforaging behaviour of Helicoverpa armigera. Ecol Entomol23:363–369

Cunningham JP, Moore CJ, Zalucki MP, West SA (2004) Learning,odour preference and flower foraging in moths. J Exp Biol207:87–94. doi:10.1242/Jeb.00733

Dicke U, Heidorn A, Roth G (2011) Aversive and non-reward learningin the fire-bellied toad using familiar and unfamiliar prey stimuli.Current Zoology 57:709–716

Dukas R (1999) Ecological relevance of associative learning in fruit flylarvae. Behav Ecol Sociobiol 45:195–200

Dukas R (2008) Evolutionary biology of insect learning. Annu RevEntomol 53:145–160. doi:10.1146/annurev.ento.53.103106.093343

Dukas R, Bernays EA (2000) Learning improves growth rate in grass-hoppers. Proc Natl Acad Sci USA 97:2637–2640

Dukas R, Duan JJ (2000) Potential fitness consequences of asso-ciative learning in a parasitoid wasp. Behav Ecol 11:536–543. doi:10.1093/beheco/11.5.536

Dukas R, Real LA (1991) Learning foraging tasks by bees: a compar-ison between social and solitary species. Anim Behav 42:269–276

Dukas R, Real LA (1993) Learning constraints and floral choicebehavior in bumble bees. Anim Behav 46:637–644

Dunlap AS, Stephens DW (2009) Components of change in the evo-lution of learning and unlearned preference. Proc R Soc B276:3201–3208. doi:10.1098/rspb.2009.0602

Dyer AG, Chittka L (2004) Fine colour discrimination requires differ-ential conditioning in bumblebees. Naturwissenschaften 91:224–227. doi:10.1007/s00114-004-0508-x

Fukushi T (1989) Learning and discrimination of colored papers in thewalking blowfly, Lucilia cuprina. J Comp Physiol, A 166:57–64

Gegear RJ, Manson JS, Thomson JD (2007) Ecological context influ-ences pollinator deterrence by alkaloids in floral nectar. Ecol Lett10:375–382. doi:10.1111/j.1461-0248.2007.01027.x

Gigord LDB, Macnair MR, Stritesky M, Smithson A (2002) Thepotential for floral mimicry in rewardless orchids: an experimentalstudy. Proc R Soc B 269:1389–1395. doi:10.1098/rspb.2002.2018

Horridge A (2007) The preferences of the honeybee (Apis mellifera)for different visual cues during the learning process. J InsectPhysiol 53:877–889. doi:10.1016/j.jinsphys.2006.12.002

IBM SPSS (2011) IBM SPSS statistics 20. IBM Corp., New YorkInternicola AI, Page PA, Bernasconi G, Gigord LDB (2007) Competi-

tion for pollinator visitation between deceptive and rewardingartificial inflorescences: an experimental test of the effects offloral colour similarity and spatial mingling. Funct Ecol 21:864–872. doi:10.1111/j.1365-2435.2007.01303.x

Internicola AI, Bernasconi G, Gigord LDB (2008) Should food-deceptive species flower before or after rewarding species? Anexperimental test of pollinator visitation behaviour under contrast-ing phenologies. J Evol Biol 21:1358–1365. doi:10.1111/j.1420-9101.2008.01565.x

Josens R, Eschbach C, Giurfa M (2009) Differential conditioning andlong-term olfactory memory in individual Camponotus fellahants. J Exp Biol 212:1904–1911. doi:10.1242/jeb.030080

Kandori I, Yamaki T, Okuyama S, Sakamoto N, Yokoi T (2009)Interspecific and intersexual learning rate differences in fourbutterfly species. J Exp Biol 212:3810–3816. doi:10.1242/jeb.032870

Kelber A (1996) Colour learning in the hawkmoth Macroglossumstellatarum. J Exp Biol 199:1127–1131

Kelber A (2002) Pattern discrimination in a hawkmoth: innate prefer-ences, learning performance and ecology. Proc R Soc B269:2573–2577. doi:10.1098/rspb.2002.2201

Kinoshita M, Shimada N, Arikawa K (1999) Colour vision of theforaging swallowtail butterfly Papilio xuthus. J Exp Biol202:95–102

Kroutov V, Mayer MS, Emmel TC (1999) Olfactory conditioning ofthe butterfly Agraulis vanillae (L.) (Lepidoptera, Nymphalidae) tofloral but not host-plant odors. J Insect Behav 12:833–843

Leadbeater E, Chittka L (2005) A new mode of information transfer inforaging bumblebees? Curr Biol 15:R447–R448

Lee JC, Bernays EA (1990) Food tastes and toxic effects: associativelearning by the polyphagous grasshopper Schistocerca americana(Drury) (Orthoptera: Acrididae). Anim Behav 39:163–173

Lewis AC (1986) Memory constraints and flower choice in Pierisrapae. Science 232:863–865

Masters AR (1991) Dual role of pyrrolizidine alkaloids in nectar. JChem Ecol 17:195–205

Matsumoto Y, Mizunami M (2002) Temporal determinants of long-term retention of olfactory memory in the cricket Gryllus bima-culatus. J Exp Biol 205:1429–1437

Menzel R (1985) Learning in honeybees in an ecological and behav-ioral context. In: Hölldobler B, Lindauer M (eds) Experimentalbehavioral ecology. Fischer, Stuttgart, pp 55–74

Menzel R (1993) Associative learning in honey bees. Apidologie24:157–168

Menzel R (1999) Memory dynamics in the honeybee. J Comp Physiol,A 185:323–340. doi:10.1007/s003590050392

Mery F, Kawecki TJ (2003) A fitness cost of learning ability inDrosophila melanogaster. Proc R Soc B 270:2465–2469.doi:10.1098/rspb.2003.2548

Mery F, Kawecki TJ (2004) An operating cost of learning inDrosophila melanogaster. Anim Behav 68:589–598

Papaj DR, Snellen H, Swaans K, Vet LEM (1994) Unrewardingexperiences and their effect on foraging in the parasitic waspLeptopilina heterotoma (Hymenoptera: Eucoilidae). J InsectBehav 7:465–481

712 Naturwissenschaften (2012) 99:705–713

Raine NE, Chittka L (2008) The correlation of learning speed andnatural foraging success in bumble bees. Proc R Soc B 275:803–808. doi:10.1098/rspb.2007.1652

Rodrigues D, Goodner BW, Weiss MR (2010) Reversal learning andrisk-averse foraging behavior in the monarch butterfly, Danausplexippus (Lepidoptera: Nymphalidae). Ethology 116:270–280.doi:10.1111/j.1439-0310.2009.01737.x

Sato M, Takasu K (2000) Food odor learning by both sexes of thepupal parasitoid Pimpla alboannulatus Uchida (Hymenoptera:Ichneumonidae). J Insect Behav 13:263–272

Shafir S (1996) Color discrimination conditioning of a wasp, Polybiaoccidentalis (Hymenoptera: Vespidae). Biotropica 28:243–251

Simonds V, Plowright CMS (2004) How do bumblebees first findflowers? Unlearned approach responses and habituation. AnimBehav 67:379–386. doi:10.1016/j.anbehav.2003.03.020

Srinivasan MV, Zhang SW, Witney K (1994) Visual discrimination ofpattern orientation in honeybees: performance and implication forcortical processing. Philos Trans R Soc B 343:199–210

Stephens DW (1993) Learning and behavioral ecology: incompleteinformation and environmental predictability. In: Papaj DR, LewisAC (eds) Insect learning: ecological and evolutionary perspectives.Chapman & Hall, New York, pp 195–218

Swihart CA, Swihart SL (1970) Color selection and learned feedingpreferences in the butterfly, Heliconius charitonius Linn. AnimBehav 18:60–64

Takasu K, Rains GC, Lewis WJ (2007) Comparison of detection abilityof learned odors between males and females in the larval parasit-oid Microplitis croceipes. Entomol Exp Appl 122:247–251.doi:10.1111/j.1570-7458.2006.00511.x

Thum AS, Jenett A, Ito K, Heisenberg M, Tanimoto H (2007)Multiple memory traces for olfactory reward learning in Dro-sophila. J Neurosci 27:11132–11138. doi:10.1523/jneurosci.2712-07.2007

Tully T, Quinn WG (1985) Classical conditioning and retention innormal and mutant Drosophila melanogaster. J Comp Physiol,A 157:263–277

Vergoz V, Roussel E, Sandoz JC, Giurfa M (2007) Aversivelearning in honeybees revealed by the olfactory conditioningof the sting extension reflex. PLoS One 2:e288. doi:e28810.1371/journal.pone.0000288

Waser NM (1986) Flower constancy - definition, cause, and measure-ment. Am Nat 127:593–603

Weiss MR (1997) Innate colour preferences and flexible colourlearning in the pipevine swallowtail. Anim Behav 53:1043–1052

Weiss MR, Papaj DR (2003) Colour learning in two behaviouralcontexts: how much can a butterfly keep in mind? Anim Behav65:425–434. doi:10.1006/anbe.2003.2084

Worden BD, Papaj DR (2005) Flower choice copying in bumblebees.Biol Lett 1:504–507. doi:10.1098/rsbl.2005.0368

Naturwissenschaften (2012) 99:705–713 713