A guest blog post by Postdoc Dr. May Dixon on her recent paper.
To what extent is animal learning shaped by selection from the challenges of foraging? One hypothesis poses that animals should learn about food differently depending on whether their food is spatially stable. Animals with spatially predictable food, like scatter-hoarders that refind caches, should learn and remember spatial cues more than animals with spatially unpredictable foods, like predators, which should instead rely more on learning “feature” cues like associated shapes or sounds (to help them find similar prey in the future). This idea has been supported by work comparing birds that do or don’t scatter-hoard: scatter-hoarding birds rely more on spatial cues than closely related non-hoarding birds.
In bats, this hypothesis is called niche specific cognitive strategies, which was posed by Stich and Winter in 2006. They observed that nectivorous Glossophaga soricina, which return repeatedly to the same flower in the jungle, easily learn to associate food with a spatial location (right or left). But, these bats had a much harder time learning to associate food with a shape. They proposed that phyllostomid bats should vary in how well they learn spatial cues depending on the spatial stability of their food , with nectivores on one end of the spectrum, frugivores in the middle (using location to say, return to a profitable tree, but feature cues to find specific fruit), and predatory bats on the other end (learning prey feature cues, but not attending much to locations, because their prey can move about).
Gerry and I were both independently fascinated by this question. Gerry ran an experiment between his Masters and PhD comparing how a frugivore and a nectivore learned different kinds of cues. He found that both relied heavily on spatial cues. This result didn’t clearly support or refute the niche specific cognitive strategies hypothesis because both bat species were predicted to rely on spatial cues, but he didn’t detect a clear difference between the species.
During my PhD I tried to compare bats with different foraging ecologies with different experimental designs but found this to be extremely difficult. For example, I first tried to compare how quickly different species could learn spatial, sound, or odor associations in a Y-maze for a food reward. The problem was that the bats mostly weren’t interested in the food, but instead just tried to escape the mesh maze I had trapped them in. I switched to using escape itself as a reward, but there were other problems. For one, how could I study use of foraging cues when the bats weren’t even foraging? The maze experimental design was also completely unnatural. How does crawling up a maze compare to flying through the jungle to find katydids? Picking the bats up and moving them back to the start box between every trial was surely stressful, what if it changed the way they learned? I felt I wouldn’t be convinced of the results even if I found the hypothesized differences.
Gerry was a postdoctoral scientist in the same lab at Panama at that point, and at some point I complained to him about my frustrations. He suggested using the same experimental design he did previously: a ‘cue-dissociation study’. For testing this kind of hypothesis he stated, “I don’t know why anyone does anything else”. In this experimental design you first train subjects to visit a rewarded feeder with a compound cue: e.g. a unique spatial location, odor, and sound. Then, in the key test, you separate these cues, so now the subject has to choose between them. Their choices tell you which cue(s) they associated with food.
I liked that this free-flying test was much more naturalistic than the other experimental designs I had tried. So that’s what we did! In collaboration with Gerry and my two PhD advisors, Mike Ryan and Rachel Page, we remodeled the cue-dissociation experiment to work with new bat species. This time, we decided to compare a frugivore (Artibeus jamaicensis) with a species that was predicted to use feature cues extensively: the predatory Lophostoma silvicolum. L. silvicolum is an expert katydid hunter and finds them by eavesdropping on their mating calls. We compared how they learned locations (positions of feeders in a flight cage), odors (super concentrated candy oils), and sounds (transposed cell-phone ringtones).
Some really excellent people helped me run this experiment. Scranton interns Tate Ackerman and Dylan Valente helped me throughout much of the project, catching bats, helping me keep them alive, and scoring videos. The then bat lab manager Vanessa Pérez Pinzón and later my wonderful field techs Amanda Savage and Dineilys Aparicio also scored a ton of videos.
What did we find? In short, rather than finding that the predatory L. silvicolum relied relatively more on the sound cue as predicted, we didn’t find any differences between the two types of bats. Both species flew to the spatial cue more than the other cues in tests. Surprisingly (to me) neither bat showed any evidence of learning the sound or odor cues: they didn’t choose these feeders significantly more than the control feeders.
What should we make of this? While this result doesn’t support the niche specific cognitive strategies hypothesis, we don’t think it is strong evidence against it, either, because any differences in cue learning between these two bats could be overshadowed by an overwhelming preference for spatial cues in this test.
To understand this point, imagine a test of how a human and a dog might relocate a rotisserie chicken they remember. The human might use primarily spatial memory (“I left it on the kitchen table”), followed by vision (“I see it on the table”), followed by smell (“I can smell it as I get closer”). Clearly, dogs use their sense of smell to find food more than humans. However, we might not detect the different use of smell between dog and human if both species overwhelmingly used spatial memory in our test. If we put the smell of chicken on the left wall of the kitchen and a visual image of the chicken on the right wall of the kitchen, both the dog and human might still go look for it on the kitchen table.
Spatial cues can easily outcompete other cues in this type of test, even for bats that might not normally use them extensively while foraging. This was especially striking in the case of the predatory L. silvicolum. From previous work with these bats, I knew that L. silvicolum is very attentive to sounds and can quickly learn to fly to novel sounds for food. Bats would often pancake full-speed into a speaker if I played even a half second of a preferred sound. It was fascinating that this bat did not apparently use the sounds at all in this test in favor of the location cues. We also noticed that there are more examples of animals preferring spatial cues than feature cues in the cue-dissociation literature.
Why might spatial cues be so salient? We have a few ideas: First, ‘spatial cues’ are not really a ‘type’ of cue because locations are really a collection of many cues that can be accessed with different senses, and so locations may be more salient than any single type of cue. Second, cue selection likely always depends on salience and context: it may have been much easier for the bats in our study to discriminate between the four locations than the four sounds or smells. Since use of cues for learning is context-dependent, cue-selection studies may be most informative when they are comparative: comparing differences in responses between groups. Third, it’s possible that our very premise was wrong. Maybe nectivores and frugivores don’t rely more on spatial cues to forage compared to predatory bats. Predatory bats could use spatial learning extensively to return to profitable foraging patches. Fourth, both the frugivore and the predator in our study live and fly in the dense, cluttered jungle (see image below). They likely both have to use spatial learning extensively just to return to their roosts every night. If these bats rely on spatial cues in other contexts, they may easily generalize to using spatial cues while foraging.
Many different follow-up experiments could shed light on these possibilities! For example, experimenters could attempt to uncover differences between bat species by biasing the experiment against spatial cues- for example by making many more spatial options (1:10) than sound or smell options (1:4), or by making the spatial cues unreliable. GPS tracking studies could compare how often bats of each species return to previously visited locations while foraging. It would also be interesting to test if bats that fly in cluttered space rely more on spatial cues than those that fly in the open.
This experiment once again affirmed to us that cognition is complex, and simple stories are usually not true. As always in science “each step that we make in the more intimate knowledge of nature leads us to the entrance of new labyrinths.”
I have a lot of fond and frustrating memories from designing and running this experiment, but my strongest impression of this time was my own predatory challenge of keeping the Lophostoma fed. These bats preferred wild-caught katydids. So, I (and many amazing friends and helpers) spent hours every night out in the fields and woods around Gamboa, Panama, butterfly net in hand, stalking katydids by their mating calls and our headlamp lights. I did a lot of sleepy meta-cognitive pondering (what cues am I using to find this katydid?). I also met a lot of other animals in the tall grass and saw the wrong side of a lot of sunrises.
Carter Lab 