Enhanced long-term memory and increased mushroom body plasticity in Heliconius butterflies

Summary Heliconius butterflies exhibit expanded mushroom bodies, a key brain region for learning and memory in insects, and a novel foraging strategy unique among Lepidoptera – traplining for pollen. We tested visual long-term memory across six Heliconius and outgroup Heliconiini species. Heliconius species exhibited greater fidelity to learned colors after eight days without reinforcement, with further evidence of recall at 13 days. We also measured the plastic response of the mushroom body calyces over this time period, finding substantial post-eclosion expansion and synaptic pruning in the calyx of Heliconius erato, but not in the outgroup Heliconiini Dryas iulia. In Heliconius erato, visual associative learning experience specifically was associated with a greater retention of synapses and recall accuracy was positively correlated with synapse number. These results suggest that increases in the size of specific brain regions and changes in their plastic response to experience may coevolve to support novel behaviors.


Heliconius butterflies have enhanced visual long-term memory compared to outgroups
Heliconius recall learned color associations up to 13 days without reinforcement Mushroom bodies of H. erato show greater developmental plasticity than Dryas iulia In H. erato, learning experience is linked to mushroom body synapse number Figure 1.Mushroom body expansion and pollen feeding in Heliconius (A) Distribution of pollen feeding in Heliconiini butterflies with dated phylogeny adapted from Couto et al. (2023).Long-term memory assays were conducted using the bolded species, while brains were sampled from species also underlined.intimately linked to mushroom body expansion in Heliconius butterflies.Indeed, initial comparative data have shown that Heliconius erato outperform the Heliconiini Dryas iulia in their ability to learn and recall complex visual cues and maintain long-term memories of learned color associations. 25ere, we extend previous data by testing visual long-term memory across additional species to build a dataset of three Heliconius (Heliconius erato, Heliconius melpomene and Heliconius hecale) and three outgroup Heliconiini (Dryas iulia, Agraulis vanillae and Dryadula phaetusa).We use this to test for consistent, superior memory fidelity across the Heliconius genus, providing considerably greater basis for generalization of effects.We subsequently investigate the neural basis of this behavioral difference, examining calyx volume, synapse densityinferred from an active zone marker-and the number of Kenyon cells in H. erato and D. iulia that participated in the long-term memory assay, in conjunction with age-matched controls, and freshly eclosed butterflies.We quantify neural plasticity in these individuals, and test for neural correlates of learning experience.

Heliconius show consistently superior visual long-term memory relative to outgroup Heliconiini
We trained butterflies to associate a sugar-protein reward with either yellow or purple feeders over four days and then tested their recall accuracy after 16 h (STAR methods).Heliconius exhibited slightly higher accuracy over non-Heliconius individuals in an initial recall test (Figure 2; Table S3; z ratio = À2.240,p = 0.048), but all six species successfully learned the color-food association (Figure 2; Table S4).To test the stability of these learnt associations over a longer period of time, butterflies were then deprived of the learned color cues and fed solely on white feeders for eight days.After this period, we conducted subsequent preference trials using purple and yellow feeders with Heliconius individuals exhibiting significantly greater fidelity to the previously learned colors (Figure 2; Table S5; z ratio = À5.807,p < 0.0001).Furthermore, while all species exhibited a decline in accuracy over the eight days between the initial trained test and the long-term memory test (Figure 2; Table S4), this drop-off was significantly higher for non-Heliconius individuals than in Heliconius (c 2 = 5.309, d.f.= 1, p = 0.0212).We further assayed the recall abilities of H. melpomene, H. hecale, A. vanilla, and D. phaetusa after an additional four days deprived of the learned color cues (a total of 13 days without reinforcement), again finding higher accuracy in Heliconius individuals (Figure 2; z ratio = À3.731,d.f.= inf, p < 0.001).At this point, the color preferences of A. vanillae and D. phaetusa were not different from random (Table S6), suggesting a loss of the learnt association.In contrast, H. melpomene maintained their learned preference during this period, while H. hecale also exhibited a marginally nonsignificant bias for the learned color (Table S6).To our knowledge, this period of 13 days represents the longest example of the persistence of a learned association without reinforcement in an insect, ahead of an 11-day period tested in the honeybee. 52mportantly, our results extend previous data showing that Heliconius erato has more stable long term memories than Dryas iulia. 25By including additional Heliconius species and outgroup species, we sampled major clades within Heliconius and across the Heliconiini, to confirm consistent shifts between Heliconius and other Heliconiini.These data strongly suggest that Heliconius as whole possess an enhanced visual long-term memory relative to other Heliconiini.This difference is consistent with mushroom body expansion in Heliconius being associated with an improved visual long-term memory, further supported by the dominant role of increased visual processing in  53 Heliconius mushroom body expansion. 25This behavioral change may have been driven by the cognitive demands of traplining for pollen in the context of increased individual longevities. 41The role of the mushroom bodies in the formation and maintenance of olfactory long-term memories is well established. 32,48,54However, the present results add to evidence that for at least certain groups of insects, including many Hymenoptera, 36 the mushroom bodies also play a significant role in visual long-term memory. 21,22In contrast to the significant clade effects we find in our long term memory assays, comparisons across a similar sample of Heliconiini species did not find Heliconius to be superior at learning shape cues 55 or reversal learning of color cues. 56When coupled with our present results, these findings suggest that mushroom body expansion in Heliconius has not led to an overall improvement in general cognition, but rather enhancement in specific, ecologically relevant cognitive tasks.Mushroom bodies of Dryas iulia and Heliconius erato vary in their response to learning experience We further explored the neural underpinnings of this cognitive shift in Heliconius at the cellular and synaptic level by using immunohistochemistry to measure several traits in the mushroom body calyx of Heliconius erato and Dryas iulia that had completed the long-term memory assay (Learning group).These individuals were compared with age-matched individuals from a ''non-learning'' environment (Control group) and freshly eclosed butterflies (Day 0 group).The ''non-learning'' group experienced the same experimental set up as the learning group, but both colors were equally reinforced and punished, preventing a conditioned preference (Figure S3; Table S7).Overall, when comparing Day 0 butterflies to either the Control or Learning groups, the mushroom bodies of Heliconius erato showed considerably more plasticity than D. iulia (Figure 3; Tables S10-S12).In Heliconius erato, the Learning and Control group individuals had a lower synapse density and fewer total synapses in the calyx, but increased calyx volumes, relative to Day 0 individuals (Figures 3A-3C; Table S11).In contrast, for D. iulia, although similar trends were observed, neither synapse density, calyx volume nor synapse number varied significantly between groups (Figure 3; Table S11).Age-associated increases in calyx volume, and decreases in synapse density, similar to those we observe in H. erato have also been described in a number of Hymenoptera including bumblebees, 17,57 honeybees, 34,58 ants 35,38 and paper wasps. 59,60The decrease in synapse number with age observed in H. erato also parallels observations in honeybees 34,58,61 and desert ants. 35This ''pruning'' of synapses, which is also present in vertebrates, is a recognized method of refining neural connectivity involving the selective elimination of axonal branches and increasing the strength of remaining synaptic connections. 62otably, individual H. erato in the Learning group had a significantly higher number of synapses in the calyx than the Control group individuals (Figure 3C; Table S11), which have similar synapse counts to D. iulia despite having much larger calyces (Figure 3; Table S11).This suggests a higher maintenance of synapses in the Learning group in H. erato and more extensive synaptic pruning in the Control group.The experience of the Learning and Control groups was identical, except for the reinforcement of color cues during the four-day training period, a relatively modest environmental difference.These differences persisted for eight days after exposure to colored feeders, suggesting that learning-associated synaptic connections in the calyx are being maintained for considerable amounts of time after exposure.Increased density or number of mushroom body synapses has previously been linked to visual 17 and olfactory 32,33 learning and long-term memory in Hymenoptera, and these results extend those findings to visual learning and memory in Heliconius.Interestingly, in Drosophila, synaptic reorganization of the mushroom body appears to be essential for only certain associative learning tasks, being necessary for aversive, but not appetitive, associative odor learning. 63eliconius erato did not show an increase in calyx volume in response to associative color learning specifically (Figure 3B), suggesting learning and memory of the color cues was achieved solely through synaptic reorganization.Contrasting with our findings, visual learning has been linked with calyx growth in honeybees 17 and the butterfly Pieris rapae. 18In honeybees 32 and leaf-cutting ants, 33 however, olfactory learning has been linked to increased synapse density without expansion of the calyx.In addition, we extend previous reports of an absence of adult neurogenesis in Heliconiini, 64 by showing that learning experience does not promote an increase in Kenyon cell number (Figure 3; Table S11), as observed in some hemimetabolous insects. 65,66Differences in synapse count between Heliconius erato groups therefore appear to be a result of changes in the number of synapses per Kenyon cell (Table S12).Day 0 individuals had significantly more synapses per Kenyon cell than Control individuals, but not the Learning group (Table S12).Together with the lack of neurogenesis in honeybees, 67 the present findings suggest that, for some insect groups, learnt associations are supported primarily through plasticity in synapse strength and number without adult neurogenesis, despite its importance in some vertebrate brain regions. 68call accuracy in Heliconius erato, but not Dryas iulia, is associated with increased synapse density and number in the mushroom body calyx We further tested for specific neural correlates of within-species variation in recall performance in both the initial recall (16 h removed from the color cues) and long-term recall tests (eight days removed).For Heliconius erato, performance in the initial recall test was positively correlated with calyx synapse density (c 2 = 5.473, d.f.= 1, p = 0.0193) and number (c 2 = 9.199, d.f.= 1, p = 0.0024), and the ratio of synapses to Kenyon cells (c 2 = 12.852, d.f.= 1, p < 0.001) (Figures 4G, 4I, and 4K; Table S13).Together with our prior finding of learning experience being associated with an increased calyx synapse count in Heliconius erato, this strongly suggests that the synaptic connections in the calyx between Kenyon cells and projection neurons from primary visual neuropils are playing a key role in visual learning and memory.This mirrors data from honeybees where a similarly positive correlation is reported between synapse density in the mushroom body collar, an area of the calyx receiving solely visual input, and visual memory. 17Unlike H. erato, initial recall accuracy performance in D. iulia was not correlated with calyx synapse density or number (Figures 4A, 4C, and 4E; Table S13), but was, surprisingly, negatively correlated with calyx volume (Figure 4B; Table S13).This is suggestive of an altered relationship between mushroom plasticity and the consolidation of the learned food-color association between these two species.Neither species showed any correlation between performance in the long-term recall test and measured traits in the calyx (Figure S4).However, when a clear outlier individual (E43) is removed, positive relationships with synapse count (c 2 = 9.263, d.f.= 1, p = 0.002) and the ratio of synapses to Kenyon cells (c 2 = 10.918,d.f.= 1, p = 0.001) were recovered in H. erato (Figure S5).No other data point has this significant an effect on the result (Table S14), while the result is highly stable once it is removed (Table S15).This may suggest the associations detected above also persist to the 8-day recall trial.Regardless, the above results suggest an important role for synaptic reorganization in the mushroom body calyx in the formation of visual associative memories in Heliconius.

Behavioral innovation and the coevolution of mushroom body volume and plasticity
Our data evidence a distinct shift in visual long-term memory ability in Heliconius coinciding with expansion of the mushroom body and changes in its developmental plasticity.These changes in memory stability likely co-occur with the emergence of trapline foraging, which depends on the long-term memory of resource locations and provides the selective advantage for increased memory longevity. 41Comparative neuroanatomical studies have tended to focus on variation in the volume of the brain, or specific brain regions. 9However, our findings indicate that the evolution of neural plasticity is an important axis to consider, and that the size of specific brain structures and their plasticity may coevolve to support behavioral adaptations.Finally, we provide an important step toward revealing the evolutionary changes in neural connectivity, and information processing and storage, that occur in expanded learning and memory circuits.With these results, Heliconius butterflies offer an example of how structural changes in the brain can support a distinct shift in visual memory, providing an important case study toward understanding the evolution of cognition.

Limitations of the study
Eueides, the sister genus to Heliconius, exhibit mushroom bodies intermediate in size between Heliconius and other Heliconiini 25 and would make an excellent inclusion in the present experiment.However, owing to their smaller size and shorter proboscises, Eueides spp.do not engage with the artificial feeders used in this experiment as readily as other Heliconiini genera and consequently tend to not perform well in these types of behavioral assays.We note, however, that Eueides mushroom bodies are intermediate in size on a log scale only, and much smaller in absolute volume compared with Heliconius. 25 In addition, our experiment does include D. phaetusa, which have mushroom bodies of a size comparable to Eueides, 25 and perform in line with other outgroup Heliconiini.We also note that, unlike in Hymenoptera, 69 the subdivisions of the calyx receiving visual and olfactory input are not easily visually discernible in the Heliconiini.We therefore used our synapse density measurements to estimate synapse numbers for the entire calyx rather than regions receiving visual input specifically, assuming a homogeneity of synapse density across the calyx.Given the visual nature of the learning task in this study, synaptic reorganization could be expected to be heightened in visual regions of the calyx, relative to other regions.Therefore, this method of approximation has the potential to dampen the apparent magnitudes of the group differences that we detect.Finally, these results could be strengthened by collecting neural plasticity data for the other four Heliconiini species included in the behavioral assays to confirm whether the differences we detect between Heliconius erato and Dryas iulia are repeated clade-wide.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:
All original code has been deposited at Mendeley Data and is publicly available as of the date of publication.The DOI is listed in the key resources table.Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS
The behavioral assays in this study used six Heliconiini species.

Agraulis vanillae Dryadula Phaetusa Dryas iulia Heliconius erato Heliconius hecale Heliconius melpomene
All experiments used freshly eclosed, adult, captive-reared butterflies with no prior experience.After eclosion, butterflies were kept in a mesh pop-up and transferred to pre-training the next morning.Individuals were, therefore, 1-day-old upon beginning the behavioral assay.For Heliconius erato and Dryas iulia, Learning group individuals were dissected the evening after completing the 8-day recall test and were therefore aged to at least 17 days.Control group individuals were aged-matched to the Learning group.Day-0 individuals were dissected the evening of the day they eclosed.
Butterfly stock populations were established with locally-caught, wild butterflies and maintained at the insectaries at the Smithsonian Tropical Research Institute in Gamboa, Panama.Stock butterflies were kept in 2 3 2 3 3m mesh cages in ambient conditions with natural light.Larvae were reared in mesh pop-ups and were provided with fresh leaves daily.H. erato, Dryas iulia, Dryadula phaetusa and Agraulis vanillae were reared on P. biflora, H. melpomene on P. triloba, and H. hecale on P. vitifolia.Training and testing of butterflies was conducted in 2 3 2 3 3m mesh cages in ambient conditions under natural light.A single Palicourea elata, with all flowers removed, was placed in the rear right corner of these cages as a roosting site.H. erato and Dryas iulia individuals were further dissected for the neuroanatomical investigations.
Sex was randomly distributed for all species.We initially included sex effects in these models, but found that it had a significant effect on calyx volume and Kenyon cells number in H. erato only (Table S9), consistent with previous findings of marginally larger mushroom bodies in female Heliconius, but not outgroup Heliconiini. 25Sex was therefore included as a random effect when testing for differences in calyx volume and Kenyon cell number only using the function glmmTMB from the package glmmTMB v 1.1.2.3 for R. 78 H. erato and Dryas iulia individuals were randomly assigned to the Learning, Control and Day 0 treatment groups.

Long-term memory assay
Long-term memory (LTM) experiments, using color cues, were carried out on captive-reared butterflies between January and April 2019 in Gamboa, Panama.The experiments used two colors, purple and yellow, colors chosen based on previous experiments using H. erato which showed that neither color was particularly attractive. 73The experiments used five-pointed, star-shaped, artificial feeders made from colored foam, 3 cm in diameter, with a centrally placed 0.5 mL Eppendorf tube that could be filled with liquid.
The day after eclosion, individuals were transferred to a pre-training cage, where they were fed solely with white artificial feeders containing a sugar-protein solution (20% sugar, 5% Vertark Critical Care Formula, 75% water, w/v) for two days (from 08:00 to 12:00) to familiarise them with the use of artificial feeders.After pre-training, butterflies were introduced to a testing cage to determine initial feeding preferences between purple and yellow.Testing cages contained 12 purple and 12 yellow feeders arranged randomly in a 4 X 6 grid, with 6.5 cm between feeders on each side.To ensure that butterflies responded exclusively to visual cues, feeders in the testing cages were empty.Preference testing lasted for 4 h from 08:00 to 12:00 and was filmed from above using a GoPro Hero 5 camera mounted on a tripod.Butterflies were individually numbered on their wings for identification using a permanent marker.The film was then reviewed to count the number of feeding attempts per individual on each color, with up to 40 attempts recorded per individual.A feeding attempt was only counted if the butterfly landed on the feeder and probed it with its proboscis.
Butterflies were then trained to associate a food reward with their non-favoured color, based on the results of their initial preference test.For butterflies that initially preferred purple, the training cage contained yellow feeders containing a sugar-protein solution, and purple feeders containing a saturated quinine solution, an aversive stimulus.The opposite arrangement was employed for individuals that initially preferred yellow.This training period lasted for four full days.After training, butterfly preferences were re-tested, following the same protocol as the initial preference test, to verify that individuals had indeed acquired the colour-food association.After the trained preference test, butterflies were placed for eight days in a cage identical to the pre-training cage, containing only white feeders filled with a sugar-protein solution.The deprivation of color stimuli for this period allowed for testing the long-term memory retention of the colour-food association acquired during the training period, and ensured that long-term memory was being tested rather than short-term or mid-term memory. 79A period of eight days was chosen because Heliconius are known to maintain their foraging routes over periods ranging from weeks to months, 45,46,80,81 during which time a pollen resource could be unproductive for several days due to competition or damage, but ultimately rewarding over the long term.Butterflies were then subject to a third preference test to determine if the learned preference was maintained, following the same protocol as the initial preference test.H. melpomene, H. hecale, Dryadula phaetusa and Agraulis vanilla individuals were also subject to an additional extended long-term memory test by placing them in a cage with only white for a further four days before a final preference test.

Kenyon cell and synapse counting and calyx measurement Treatment groups
Neuroanatomical measurements were taken for Dryas iulia and Heliconius erato from three treatment groups -Day 0, Learning and Control.Day 0 butterflies comprised individuals that were dissected the evening of their day of emergence.These butterflies were kept in a small, mesh pop-up cage until dissection and had no foraging or free-flight experience.The Learning Group comprised Dryas iulia and Heliconius erato individuals that participated in the long-term memory experiment and were dissected the day they finished (Learning Group).The Control Group consisted of individuals aged-matched to the Learning group reared in a ''non-learning'' environment.Control individuals were acclimatised to the use of artificial feeders and then tested for naive color preference following the same protocol used for the Learning Group.These butterflies were then introduced to a ''training'' cage for four days, which contained an even number of purple and yellow feeders, presented in random spatial arrangement.Unlike the Learning group, where feeder color was consistently associated with either a food reward or quinine punishment, half of the feeders for each color were filled with the rewarding sugar-protein mixture, and half with the aversive quinine solution.Each color, therefore, was as equally rewarding as punishing.After four days in this environment, color preference was tested again following the protocol established for the long-term memory experiment.The Control butterflies were then introduced to a cage with white feeders for eight days, replicating the long-term memory waiting period of the Learning group.Finally, after eight days feeding on white feeders, Control butterflies were exposed to empty purple and yellow feeders from 08:00 to 12:00 on their final morning, mirroring the last preference test of the Learning Group, before being dissected in the evening.

Brain dissection, fixing and training
All brains were dissected and fixed at the Smithsonian Tropical Research Institute in Gamboa, Panama, following established protocols. 26,82utterflies were decapitated using scissors and the head was submerged under HEPES-buffered saline (HBS; 150 mM NaCl; 5 mM KCL; 5 mM CaCL 2 ; 25 mM sucrose; 10 mM HEPES; pH 7.4).A small aperture was cut into the head cuticle between the eyes to improve permeation of the fixative.The brain was fixed in situ for 16-20 h at room temperature under gentle agitation in zinc-formaldehyde solution (ZnFA; 0.25% [18.4 mM] ZnCl 2 ; 0.788% [135 mM] NaCl; 1.2% [35 mM] sucrose; 1% formaldehyde).After fixing, the whole brain was dissected out under HBS using a scalpel and forceps, placed into 80% methanol, 20% dimethyl sulfoxide (DMSO) under agitation for 2 h and then transferred to 100% methanol for long-term storage at À20 C.
Brains were subject to three immunohistochemical stains. 25Anti-synapsin was used to mark active zones of synapses in the calyx 17,32,61 ; this approach has previously been used as a proxy for synapse density/number and provides comparable results when equivalent estimates of microglomeruli density/number are made when double stained with a post-synaptic marker, anti-phalloidin. 25Due to the need for long-term storage of the brains during fieldwork, we use a protocol that includes methanol, and therefore prohibits successful anti-phalloidin staining. 25We note that variation in staining intensity could be caused by differential up/down-regulation at synapses with different degrees of activity.Our density estimates may therefore partially capture variation in synapse active zone activity as well as variation in number.We also used DAPI to identify nuclei in Kenyon cell bodies, 82,83 along with HRP (anti-horseradish peroxidase) to label neuron membranes to confirm that counted nuclei were neuronal. 84Brains were stained in batches of eight that included individuals from all treatment groups to avoid the possibility of batch effects skewing results.Prior to staining, brains were first rehydrated in a decreasing methanol series (90%, 70%, 50%, 30%, 0% in 0.1 M Tris buffer, pH 7.4) for 10 min each.Because quantification of the Kenyon cells and synapses requires imaging the brain at 633 magnification, it was necessary to section the brains so that the calyx tissue would be within the working distance of the objective lens of the microscope.Brains were embedded in 5% agarose which was cut into a rectangular prism.The agarose block was cut along a corner so that individual slices could be correctly orientated later during mounting.The embedded brain was submerged in 0.1 M Tris buffer and sliced horizontally into 80 mm sections using a Leica VT1000 S vibrating blade microtome.

Figure 2 .
Figure 2. Heliconius show superior visual long-term memory relative to outgroup Heliconiini Long-term memory accuracy in six Heliconiini species in a two-color preference assay.The boxes encompass the two middle quartiles, with central line showing median.Whiskers extend to the furthest data point within 1.5 times the interquartile range.Trained = 16-h recall performance after four days training.LTM = recall performance after eight days deprived of the learned color stimuli by being fed solely on neutral (white) feeders; LTM2 = recall performance after an additional four days on white feeders.n.s.= no statistically significant difference; * = p < 0.05; *** = p < 0.001 with P-values calculated using a z-test and corrected with the Sida ´k correction.Butterfly images from Warren et al. (2023).53

Figure 3 .
Figure 3. Mushroom bodies of Dryas iulia and Heliconius erato vary in their response to learning experience Variation in (A) synapse density, (B) volume, (C) and total synapse number of the mushroom body calyx, and (D) total Kenyon cell number, between three treatment groups, Day 0, Learning and age-matched Controls, in Dryas iulia and Heliconius erato.The boxes encompass the two middle quartiles, with central line showing median.Whiskers extend to the furthest data point within 1.5 times the interquartile range.All values correspond to a single hemisphere of the brain.n.s.= no statistically significant difference; * = p < 0.05; ** = p < 0.01; *** = p < 0.001 with P-values calculated using a t-test and corrected with the Sida ´k correction.

Figure 4 .
Figure 4. Recall accuracy in Heliconius erato, but not Dryas iulia, is associated with increased synapse density and number in the mushroom body calyx (A-L) Relationship between 16-h recall accuracy of a learned food-color association in a two-color preference test and several neuroanatomical measurements in the mushroom body calyx for (A-F) Dryas iulia and (G-L) Heliconius erato.All values correspond to a single hemisphere of the brain.Regression lines, derived from generalized linear mixed models, with standard errors, are shown only when the trait was a significant predictor of performance, assessed using a chi-square test.

TABLE
d RESOURCE AVAILABILITY B Lead contact B Materials availability B Data and code availability d EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS d METHOD DETAILS B Long-term memory assay B Kenyon cell and synapse counting and calyx measurement d QUANTIFICATION AND STATISTICAL ANALYSIS