Olfactory tracking strategies in a neotropical fruit bat

ABSTRACT Many studies have characterized olfactory-tracking behaviors in animals, and it has been proposed that search strategies may be generalizable across a wide range of species. Olfaction is important for fruit- and nectar-feeding bats, but it is uncertain whether existing olfactory search models can predict the strategies of flying mammals that emit echolocation pulses through their nose. Quantitative assessments of how well echolocating bats track and localize odor sources are lacking, so we developed a behavioral assay to characterize the olfactory detection and tracking behavior of crawling northern yellow-shouldered bats (Sturnira parvidens), a common neotropical frugivore. Trained bats were presented with a choice between control and banana-odor-infused solutions in a series of experiments that confirmed that bats are able to locate a reward based on odor cues alone and examined the effect of odor concentration on olfactory search behaviors. Decision distance (the distance from which bats made their change in direction before directly approaching the target) was distinctly bimodal, with an observed peak that coincided with an inflection point in the odor concentration gradient. We observed two main search patterns that are consistent with both serial sampling and learned route-following strategies. These results support the hypothesis that bats can combine klinotaxis with spatial awareness of experimental conditions to locate odor sources, similar to terrestrial mammals. Contrary to existing models, bats did not display prominent head-scanning behaviors during their final approach, which may be due to constraints of nasal-emitted biosonar for orientation.

We refer to this complex behavior as "route-following" to reflect that the animals can learn the spatial arrangements of their environment (natural or experimental) and can deduce the most efficient routes for inspecting multiple likely source coordinates. For example, mice can use airborne gradients to locate odor rewards, but find rewards faster when relying instead on previous experience (Gire et al., 2016). Bats are known to use spatial memory while foraging (Fleming et al., 1977;Thiele and Winter, 2005), and so may also be able to combine olfactory cues with spatial information to locate odor sources.
In this study, we quantitatively analyzed the locomotor patterns and behavioral strategies of a phyllostomid bat (Chiroptera: Phyllostomidae) searching for an attractive odor source while crawling downwards. We chose to focus on the northern yellow-shouldered bat, Sturnira parvidens (Goldman, 1917) ( Figure 1A) because of its diet, wide distribution, and its use of olfaction for social communication (Faulkes et al., 2019;González-Quiñonez et al., 2014). The northern yellow-shouldered bat is a small frugivore (13 -18 g) common to much of Central America (Hernández-Canchola et al., 2020). This species feeds on a variety of fruits, including banana, wild fig (Ficus), and neotropical fruits in the genus Solanum (including S. hazenii, S. angulate, S. americanum and S. torvum;Castro-Luna and Galindo-González, 2012;Fleming et al., 1977). Field observations suggest these bats may first use olfactory cues in flight to identify trees bearing ripe fruit, prompting them to land and crawl along branches where they may rely upon olfaction to find fruit obscured by foliage.
Preliminary behavioral experiments confirmed that Sturnira readily sought out food in an experimental setting without requiring extensive training, and thus could provide a useful model for measuring bat olfactory tracking capabilities and characterizing their locomotor search strategies. First, we established that naïve crawling bats would successfully localize an attractive odor source in the absence of salient biosonar cues. We then analyzed the locomotor search patterns by quantifying trajectories, speeds and head-scanning behaviors throughout the search to provide a comprehensive characterization of their odor localization strategies across experimental conditions.

Materials and Methods:
Field Conditions: We conducted field experiments from April 23 -May 2, 2019 at Lamanai Outpost Lodge, Orange Walk, Belize (17°45′N; 88°39′W). Bats were captured using mist-nets from along forest trails and clearings in the Lamanai Archeological Reserve (within 2 km of the Lodge). On the night of capture, we placed individual bats in the experimental arena for between 1 -2 hours with several pieces of banana in plastic hexagonal weigh boats on the floor of the arena. Only individuals that spontaneously sought out and consumed the banana reward by the end of this trial period were retained for behavioral experiments, resulting in N = 10 male bats. We only used adult male bats in this study to reduce potential confounding factors of sex or age.
Ethical Note: Experiments were carried out under permits from the Belize Forestry Department (permit number FD/WL/1/19 (10)) and were approved by the Texas A&M University Institutional Animal Care and Use Committee (AUP # 2017-0139). Between experiments, bats were housed together in soft mesh cages (60.9 x 60.9 x 91.4 cm) in a dark, quiet location and provided water ad libitum. During the first 24 hours following capture, bats had access to small bowls containing ripe banana at the bottom of the cage. We released all bats at their capture site after a maximum of five days.

Experimental Assays:
We measured olfactory localization behavior in naïve bats using a two-choice olfactory assay and standard operant procedures. The testing arena was a soft mesh cage (37 x 37 x 71 cm) oriented vertically to allow bats to hang and move naturally ( Figure 1B). Pilot behavioral experiments conducted in Belize in 2018 found that bats were more motivated to investigate a possible food reward when allowed to crawl vertically as opposed to crawling horizontally on surface, as this more closely mimics natural hanging and crawling conditions (such as might be seen in a roost). The front face of the cage was made of clear plastic to allow video recording. Experiments took place between 20:00 and 06:00 hours and were video-recorded with a Basler Ace model ac640-um digital video camera connected to a laptop running Basler Video Recording Software (Ahrensburg, Schleswig-Holstein, Germany). Videos were recorded at 30 frames per second and 640 x 480-pixel resolution. We ran all experiments in complete darkness, except for illumination with infrared LED light strips attached to the sides of the arena, to remove any confounding visual cues. At the Journal of Experimental Biology • Accepted manuscript beginning of each trial, bats were placed at the top center of the arena, and stimuli were presented in small plastic bowls (2.5 cm diameter weigh boats), placed at the bottom on opposite sides of the arena. Positively reinforced stimuli (S+) included real banana pieces or a chemical olfactory cue mixed with sugar water. Chemical olfactory stimuli were prepared using food-grade banana baking emulsion, composed of artificial and natural flavors (LorAnn Professional Kitchen, Michigan, USA). We prepared four concentrations of banana solution using serial dilution, adding 1 ml of banana emulsion (or resulting dilution) to 9 ml of 30% (w/w) sugar solution. All dilutions were prepared from the same batch of banana emulsion and sugar solution. Neotropical bats can discriminate between natural and artificial banana odor (Laska, 1990a), but will still readily consume artificial banana (A. Brokaw, personal observations). We chose to use a baking emulsion as an olfactory cue instead of a pure chemical compound (such as isoamyl acetate) to allow bats to safely consume or taste a reward, in order to maintain motivation during the behavioral trials. During the acclimation and initial training period following capture, we presented bats with banana pieces supplemented with 10% banana-sugar solution to ensure that bats associated the artificial olfactory stimulus with the real banana reward. Unreinforced stimuli (S-) were distilled water or an unflavored piece of sponge cut to mimic the shape and texture of a piece of banana.
In preliminary experiments, we placed a condenser microphone (model CM16, Avisoft Bioacoustics, Berlin, Germany) at the base of the arena to record their echolocation behavior during the olfactory searches. We only detected their broadband echolocation calls when the bat was very near to and directly facing the microphone. However, we noted that the nose-leaf twitched every time the bat emitted a pulse, and based on this it was evident that the bats were continuously emitting pulses whenever they were moving. Since we could not reliably record the pulses throughout the arena as the bats moved we did not try to quantify their echolocation beyond confirming that they actively echolocated throughout all trials.
The following experiments were designed to evaluate the olfactory search behaviors used by bats locating an odor cue. The first experiment was designed to ensure that naïve bats would reliably seek out a familiar food reward possessing a strong olfactory cue in the test chamber. In the second set of experiments, we controlled for the possible effects of echolocation during olfactory search by testing if bats could locate the S+ in the absence of salient sonar acoustic cues, by presenting an unscented shape or removing shape cues completely. In the third experiment, we tested the effect of changing odorant concentration on the bat's olfactory localization performance (Table 1). Lastly, we used bat movement trajectories from all experiments to quantitatively describe the behavioral search strategies of crawling bats.

Acclimation and Training:
On the first night after capture, we introduced naïve bats to the arena and gave them up to two hours to explore the cage and find the banana food rewards. Bats were gently repositioned by hand at the top of the arena each time a new piece of banana was added to the dish to acclimate them to being handled and reinforce the goal-seeking behavior. Most, but not all, bats quickly learned the task after one night, allowing experimental trials to begin on the second night. Bats that did not seek out food within the arena on the first night were released at their capture site the following night. To reinforce the behavior each night, each experimental session began by presenting the bats with two banana pieces supplemented with 0.5 ml of 10% banana extract solution, which was done ensure that the bats would associate the extract banana smell with real banana reward even if the bats perceived a difference between extract and real banana smell. We allowed bats to explore the arena until they located and consumed both pieces of banana. We recorded the location where the bat found the first banana. For the following experimental trial, the olfactory stimulus was switched to the opposite side to discourage side bias.

Experimental Animals:
On a given night, in-between trials, bats were held individually in soft, cloth bags in a quiet area. Each night, we randomized the order that individuals were tested. The arena was wiped with 95% ethanol and allowed to dry between trials to reduce confounding odor cues.
Although experiments are presented and analyzed separately, the trials for all three experiments were randomized within and across nights, to avoid potential confounding effects of learning and maximize sampling across limited individuals and time. We aimed to test each individual ten times at each treatment. Trials with a banana reward were arranged to be every 4th or 5th trial, to ensure bats sustained motivation, and so were repeated more than 10 times per individual. The location of S+ was pseudo-randomized for each trial, with its position repeated no more than three consecutive times. We carried out trials under ambient airflow conditions, and temperature and relative humidity were recorded at the start and end of each trial.

Experiment 1: Localization of food reward using odor
During this experiment, bats had the option to choose between a banana reward (S+) and control object (S-). Both choices were placed in plastic weigh boats at the bottom of the arena. We cut ripe bananas into cubes, approximately 1 cm 3 . The control object was a cosmetic sponge cut into the same 1 cm 3 shape as the banana piece. We supplemented the banana reward with 0.1 ml of 10% banana-sugar solution. Both stimuli were prepared and placed in the arena immediately prior to the start of the behavioral trial. We placed an individual bat at the top of the arena to start the trial. Trials lasted until the bat located and consumed the piece of banana, or after a maximum of five minutes had elapsed. If bats did not attempt to feed on the banana after five minutes, then a "no-choice" result was logged.

Experiment 2: Role of acoustic cues during reward localization
The following treatments were designed to isolate olfactory cues from acoustic cues and determine whether or not both sensory modalities (acoustic or odor) were necessary or preferred by the bats during odor localization. In Experiment 2A, we placed 0.5 ml of 10% odor-sugar solution alone (S+) in a plastic weigh boat on one side of the arena, while the other side of the arena held an unscented cosmetic sponge cube cut to resemble a piece of banana placed in 0.5 ml of distilled water (S-). This was designed to test which cue type (odor or acoustic) was more important in the bat's search behaviors. In Experiment 2B, we tested how well bats could localize an odor when there was no salient acoustic cue (cosmetic sponge) by placing 0.5 ml of 10% odor-sugar solution (S+) on one side of the arena, while the other side held 0.5 ml of distilled water (S-). If bats were successfully able to locate the odor cue, this would provide strong evidence for localization using only odor cues. Trials began when we placed a bat at the top of the arena, and continued until the bat touched, grabbed, or licked one of the stimuli. If bats did not select either target after five minutes, then a "no-choice" result was logged.

Experiment 3: Effect of odor strength on localization success
To evaluate whether or not odor concentration influenced localization performance or search strategies, we challenged the bats with four different concentrations of banana odors.
During these experiments, we placed two cosmetic sponge cubes (1 cm 3 ) in plastic dishes on opposite sides of the arena. One of the sponges held 0.1 ml of odor-sugar solution (S+), while the other side held a sponge and 0.1 ml of distilled water (S-). We tested bats with four different odor concentrations: 100% (only banana extract), 10%, 1% and 0.1%. We determined bats made a choice when their nose or mouth touched the sponge or weight boat of one of the stimuli. If bats did not select either target after five minutes, then a "no-choice" result was logged.

Behavioral Scoring and Movement Analysis:
We recorded every trial for all experiments to analyze and reconstruct of the locomotor patterns and pathways used by the searching bats. This information can reveal whether or not the bats consistently used any of the previously defined search strategies seen in other animals (i.e., cast and surge) while tracking odor sources across experimental contexts. We extracted and analyzed bat locomotor patterns and two-dimensional trajectories using Noldus EthoVision XT 13 (Leesberg, Virginia, United States, Figure 1C). The coordinate space was calibrated automatically in EthoVision XT by inputting the real-world height and width of the back of the experimental arena (where bat movement would be measured). The coordinate space was calibrated individually for each video, to account for any movements of either the arena or camera between trials. Bat choices was determined when a bat touched their nose or mouth to one of the stimuli (touching either banana, sponge, or weigh boat) Trials were scored a 'success' when bats correctly chose the side with the S+.
For each trial, we measured or calculated the following: start distance (cm), total distance travelled (cm), average velocity (cm/s), path straightness, decision distance (cm) and path shape ( Table 2). Total distance travelled and average velocity were automatically calculated in EthoVision XT. We manually measured or classified starting distance, decision distance, and path shape from each trial using the integrated tracking view in EthoVision XT. Starting distance and decision distance was calculated in EthoVision XT as the straight-line distance between the bat center point and the odor location at start of the trial (starting distance) and at the time point where the bat made its last change of direction before moving towards its target (decision distance). To investigate if and how bats use head movements during an olfactory localization task, we also analyzed head scanning behavior for successful trials in Experiment 1 and Experiment 3. A head scanning event was counted each time the bat rotated its nose at least 45 degrees off axis to one side or the other and were only observed to occur consistently when the bat was stationary. Actively crawling bats generally kept their nose leaf pointed forward in line with the body axis; during locomotion any changes in head orientation were coordinated with concurrent changes in body orientation and therefore not interpreted as head scanning. We extracted the distance from the odor source at which each head scanning event occurred using EthoVision XT. In addition to counting total number of head scanning events, we also recorded the number of head scans that occurred before or after the bat started moving towards the bottom of the arena (starting distance), and before and after the bat made its final decision (decision distance). Head scanning events at the start or decision distance were counted as occurring 'after' this cutoff.

Journal of Experimental Biology • Accepted manuscript
Only trials where bats remained along the back of the arena until making a choice were used for trajectory analysis and classified into path shapes ( Figure 2). Due to inaccuracies in tracking introduced by three-dimensional motion, trials where bats flew or hovered during the trial, or crawled along the side panels of the arena were excluded from trajectory analysis, although these trials were included in the analyses of bat success rates.
Path shapes were qualitatively classified visually from the detailed tracking view in EthoVision XT 13, and trajectories were defined as one of four categories: top casting, direct, middling, and bottom casting ( Figure 2A, Table 2). Top casting was defined by horizontal movement from the bats' staring position at the top of the arena, in which bats crossed the midline of the arena at least once before making a straight path downwards towards one of the stimuli. In direct strategies, bats moved downward without making horizontal shifts in movement. These paths were either straight downward or had a slightly diagonal shape, depending on the bat's exact starting point. Bottom casting strategies were essentially the inverse of top casting paths, in which bats made a straight movement downwards towards one of the stimuli, but then moved horizontally across the bottom of the arena (crossing the midline at least once) before making a final choice. Paths in this category produce a distinctive "L" shaped pattern. The middling strategy was characterized by general meandering of the path across the arena, in which bats shifted towards the middle of the arena while moving downwards, and then angled diagonally to one of the stimuli between one-third to one-half the way down the arena (vertical distance).

Estimating the Odor Concentration Gradient:
To estimate the distribution of odors in the arena, we recreated the field setup in the lab (College Station, Texas, United States) to measure odor concentrations using a handheld photoionization detector (PID), (PhoCheck Tiger, Ion Science, Royston, United Kingdom).
We placed the same type of plastic weigh boat used in field trials and containing 0.1 ml of 100% banana extract on one side of the olfactory arena, at the same location where the odor stimuli were placed during behavioral trials. We divided the back of olfactory arena into 120 grid spaces, each approximately 5.5 cm 2 . Each grid space was measured at 1 second intervals for 5 seconds, and values were averaged for each space. The PID was set to use isoamyl acetate as a standard and was zeroed in clean air using a carbon filter attachment immediately prior to measurements. Since measuring the entire arena would take longer than the maximum time bats were in the arena, we also took measurements of the horizontal and vertical odor distributions at time point zero (immediately following placement of the odor in the arena) and after five minutes, representing the start and end conditions of each trial.
While the lab environment is expected to be different from field conditions, the purpose was not to recreate the precise olfactory environment bats may have been exposed to, which undoubtedly varied slightly between trials, but rather to provide a general estimate for how odors may be distributed within the arena.

Statistical Analysis:
The percentage of trials that the bat correctly chose the odor stimuli (S+) were taken as a measure of performance in all three experiments. Trials where bats did not select either stimuli (no-choice) were excluded from analysis. Bat performance between treatments was analyzed using generalized linear mixed models (GLMM) with a binomial distribution (using glmer in the 'lme4' package in R, Bates et al., 2015). Bat ID was included as a random effect to account for repeated testing of individuals. We first tested if environmental conditions (temperature and humidity) significantly influenced bat performance. We averaged the temperature and humidity for each trial ([start value + end value]/2) and analyzed their effect using a GLMM, with temperature and humidity as fixed effects. Post-hoc tests for significant variables (P < 0.05) were carried out using Tukey contrasts, adjusted for multiple comparisons (glht in package 'multcomp ', Hothorn et al., 2008). To test if the bats were overall able to discriminate better than chance levels within each treatment, we used an intercept-only binomial GLMM predicting bat performance, accounting for repeated measures. In this type of model, the parameter estimate for the intercept can be interpreted to determine if bats did better than random choice (after Maynard et al., 2019). We used onetailed binomial tests to assess if individual bats performed better than chance (50%) during the two-choice trials.
To explore how bat strategies varied across trials, we tested if bat performance could be predicted by certain behavior patterns (such as movement speed, amount of distance traveled or trajectory shape). Search behavioral parameters were log-transformed where appropriate and histograms inspected for outliers before analysis to meet assumptions of normality. We fitted the data to a GLMM with a binomial distribution pooling trials across all experimental treatments (excluding trials where tracking was unreliable due to bat flight or leaving the back of the arena). Fixed effects included average velocity (cm/s), distance travelled (cm), movement time (s), decision distance (cm), and path shape, with Bat ID as a random effect. To test the significance of each fixed effect as a predictor of bat performance, we used a model simplification approach (Crawley, 2013). No interactions were included in the models due to limited sample size. If a significant effect was detected in the model (P < 0.05), we used a post hoc Tukey contrast adjusted for multiple comparisons to examine any differences.
Behavioral strategies are also likely to be context dependent, and individuals can show plasticity in their strategies. To examine how bats may adjust their search behaviors as the difficulty of the task increases, we isolated the successful bat trials from banana and odor solution (concentrations 100% -0.1%) treatments. We fitted linear mixed models (LMM) with treatment as an explanatory variable and different trajectory measures (average velocity, distance travelled, decision distance) as response variables (using restricted maximum likelihood, lme in package 'nlme ', Pinheiro et al., 2018). Bat ID was included in the model as a random effect to account for repeated testing of individuals.
Finally, we investigated the role of head scanning in bat localization strategies by quantifying head movements during the successful trials when the bats were localizing banana and odor solution treatments. We used a GLMM with a Poisson distribution to test if there was an effect of either treatment or path shape on the frequency of head scanning events, and used a likelihood ratio test to compare a null model to the fitted model separately for each variable. To test if bats changed their head scanning behavior with distance from the odor source, we compared the average number of head scanning events for each bat that occurred before and after the bat made a decision using a paired Wilcoxon sign-ranked test.
All analyses were carried out using R (version 3.5.0, R Core Team, 2018) and RStudio (R Studio Team, 2016).

Results:
We recorded 648 behavioral assay trials across ten individual bats and seven experimental treatments. Bats made a choice (correct or incorrect) in 529 trials. Due to limitations in the field, the number of trials for each treatment for each bat was not equal. The minimum number of trials recorded for a treatment was five and the maximum number of trials for a treatment was 19. All 10 bats were tested across all experimental treatments except for three individuals, who were not exposed to the odor-only treatment.
Average temperature was fairly consistent across all trials (27.9 C  0.04 SE), and did not have a significant effect on bat performance (all trials pooled, binomial GLMM, z = 1.414, P = 0.158). Average relative humidity varied slightly more across trials (70.4%  0.11 SE) and did have a significant effect on bat performance (all trials pooled, binomial GLMM, z = -2.032, P = 0.042). To account for this variation, average relative humidity was included as a random effect in the generalized linear mixed models. There was also no effect of trial order on performance, that is bats were not more successful at localizing odors in later trials than trials early in the experiment (all trials pooled, binomial GLMM, z = -1.491, P = 0.136).

Experiment 1: Localization of food reward using odor
In this experiment, we established whether bats could consistently and successfully locate a rewarded odor. Bats were reliably able to locate the location of a rewarded odor, with eight out of ten individuals performing above chance in a two-choice assay (one-tailed binomial test, P < 0.05, n = 10 bats, 120 trials, 8 -18 trials per bat). For the two bats that did not perform better than chance, they only made a choice during three (Bat 7) and six (Bat 8) out of ten trials, suggesting low motivation and not lack of tracking ability. On average, bats successfully located the odor reward 90.7% ( 6.99 SE) of the time (excluding trials where bats did not make a choice), exhibiting non-random preference for the odor-rewards side (intercept-only binomial GLM, P < 0.01).

Experiment 2: Role of acoustic cues during reward localization
In the first part of this experiment (Experiment 2A), we tested whether bats would localize an attractive odor cue without the appropriately matching echolocation cue. On average, bats performed better than chance at locating the odor-rewarded side, even when there was not an accompanying shape cue (intercept-only binomial GLM, P < 0.01) and successfully chose the odor cue in most of the trials (79.7%  8.24 SE, n = 10 bats, 72 trials, 4 -9 trials per bat). In the second part of the experiment (Experiment 2B), we tested if bats could successfully locate an odor cue when no salient echolocation cues were present, by removing shape cues (i.e. banana piece or cosmetic sponge). Again, bats performed better than chance at locating the odor-rewarded side in both treatments (intercept-only binomial GLM, P < 0.01 for both treatments). The average success rate for bats localizing an odor without a distinctive echolocation target was lowest compared to other experimental treatments (76%  5.72 SE, n = 7 bats, 64 trials, 3 -11 trials per bat). Comparing bat performance across treatments from Experiments 1 and 2, experimental treatment had an effect on localization success across all 10 bats (binomial GLMM, F = 3.5308, df = 2). Bats were more successful at locating the banana reward compared to trials when there were no distinctive echolocation cues available to guide them (z = -2.652, P = 0.0217) ( Figure 3A).
Neither start latency (time at top of the arena before moving downward) or decision distance had an effect on bat performance.

Journal of Experimental Biology • Accepted manuscript
We also tested how decreasing odor concentrations would affect bat olfactory localization performance. Overall, bats performed better than chance when locating 100%, 10% and 1% odor concentrations (intercept-only GLMM, P < 0.01 for all three treatments).
Average percent success decreased with a decrease in concentration and bats had the highest average success rate when localizing the 10% odor solution (79.57%  4.73 SE). Bats were least successful when searching for the 0.1% odor solution, particularly when compared to the 10% (z = 2.838 P = 0.0233, Figure 3B). While four out of ten bats performed better than predicted by chance at locating the 10% concentrations (binomial one-tailed test, P < 0.05), we did not have sufficient power to make conclusions on individual performance due to limited trial sample sizes for most individuals (3 -11 trials per bat, per treatment after 'nochoice' trials were removed).

Behavior and Movement Analysis:
Across all experiments, we analyzed bat movements to quantify and categorize the potential odor localization strategies bats are using to localize an odor source. Only trials where bats crawled along the back of the arena to reach their choice (S+ or S-) were included in this analysis (n = 420 trials, e.g., Movie 1), consisting of 79% of all recorded trials in which bats made a choice (420/529 trials). Of the analyzed trajectories, 53.1% of trajectories were from trials where the odor was presented on the left side of the arena (223/420 trials) and 46.9% were trials where the odor was presented on the right (197/420 trials).
We log-transformed average velocity and total distance travelled to meet assumptions of normality. Inspection of the distribution for decision distance revealed a bimodal distribution ( Figure 4A). When separated between successful and unsuccessful bat trials, there was a peak in number of successful trials where bats made their decision between 25 and 35 cm from the stimulus ( Figure 4B-C). This bimodality was not seen when looking only at unsuccessful trials ( Figure 4B). Neither distance travelled nor average velocity had a significant effect on bat performance, but there was a significant relationship between bat performance and trajectory shape (GLMM, F = 4.067, df = 3).
Looking only at trials where bats successfully located the banana odors, there was a significant difference in the log-distance travelled during the trial, log-average velocity and decision distance across treatments (excluding treatments from Experiment 2, n= 327 trials) (LMM, P < 0.05). Bats travelled a shorter distance when localizing the 10% odor concentration compared to 1% (z = -3.411, P = 0.005) and banana (z = 2.86, P = 0.034) treatments. Bat trajectories were also more direct (as measured by straightness) when
Bats also moved fastest when navigating towards the 10% odor, particularly when compared to the banana treatment (z = -2.753, P = 0.046). Decision distance was variable across treatments, with bats making their final decision closer to the banana stimuli compared to the 100% concentration (z = -2.988, P = 0.0214).
All four locomotor patterns were observed in successful trials across treatments, but bottom casting was significantly more frequent in successful bats than in unsuccessful bats (z = 2.688, P < 0.01) ( Figure 2B). All individuals used each of the four search strategies at least once. To validate our qualitative categorization of search strategy, we compared the total distance traveled, path straightness and decision distance using a repeated-measures ANOVA (with Bat ID as a random factor). Straightness was significantly different between path shapes (F = 21.09, P < 0.001, Figure 2D). Both direct and middling strategies were significantly straighter than either casting strategy (Tukey's pairwise comparison, P < 0.001) but were not significantly different from each other (t = -1.183, P = 0.634). Similarly, straightness of top casting and bottom casting did not differ significantly from each other (t = 1.143, P = 0.660). However, bats did travel significantly further (total distance) when using the top casting strategy compared to all other strategies (P < 0.05 in pairwise comparisons, Figure 2C). The decision distance was not significantly different between top casting and direct strategies (t = -2.041, P = 0.171), but both were significantly farther away from the correct stimuli compared to the other two strategies (pairwise comparison, P < 0.001, Figure   2E). Based on these differences, we conclude that the strategies are qualitatively and quantitatively different from each other. The top casting strategy is characterized by the farthest traveled distance, the furthest decision distance and least straight trajectory compared to the other three strategies. While similar to top casting in straightness and total traveled distance, bottom casting had the closest decision distance of all four strategies. In contrast, the direct strategy was the straightest path observed in these trials and bats made their decision at similar distances compared to top casting. While similar to the direct strategy in straightness and total distance travelled, the decision distance for the middling strategy was closer to the correct stimuli, but not as close as in bottom casting trajectories.
To analyze head scanning behavior, we pooled successful trials from which we were able to obtain high quality reconstructions of their trajectories from Experiment 1 (banana) and Experiment 3 (percent odor concentrations), resulting in a total of 247 trials across 10 individuals (15 -40 trials per individual). We observed 849 total head scanning events across all trials. Most head scanning behavior occurred at distances between 60 and 80 cm from the odor source, i.e. when the bats were at the top of the arena ( Figure 5A). Bats performed significantly more head scans both before starting their downward trajectory towards the odor source (Wilcoxon sign-rank test, V = 53, P = 0.005), and before making their final direction decision ( Figure 5B, Wilcoxon sign-rank test, V = 55, P = 0.001). Neither concentration or path shape had an effect on the total number of observed head scanning events (GLMM likelihood ratio test, P > 0.05).

Estimating the Odor Concentration Gradient:
Odors in the arena were not evenly distributed, but the odor structure in the arena was consistent with a Gaussian distribution with the highest concentrations recorded immediately above and next to the odor stimulus. The odor concentrations declined rapidly with distance from odor in both horizontal and vertical directions ( Figure 4D). After five minutes (the maximum trial time), odor concentrations along the vertical axis stayed either constant or increased, staying higher along the middle of the arena compared to the horizontal odor distribution. Along the horizontal axis, the odor concentration gradient dropped close to 0 (or below detectable levels using the PID) around 30 cm from the odor source, while it did not drop to 0 until between 35 to 55 cm in the vertical direction ( Figure 4E).

Discussion:
Our results suggest that bats use klinotactic olfactory tracking strategies similar to other terrestrial mammals, including humans (Jinn, 2019), mice (Gire et al., 2016;Liu et al., 2020), and rats (Bhattacharyya and Bhalla, 2015). While previous work demonstrated that bats are able to detect and discriminate concentration gradients to localize odors rewards (Laska, 1990a;Laska, 1990b), this is the first study to specifically quantify the locomotor patterns and olfactory search strategies of bats. Similar to previous research demonstrating the importance of olfactory cues in other echolocating and non-echolocating bat species (Hodgkison et al., 2007;Korine and Kalko, 2005;Parolin et al., 2015;Tang et al., 2007;Thies et al., 1998;Von Helversen et al., 2000), northern yellow-shouldered bats were able to localize an odor reward using olfaction under experimental conditions that controlled for echolocation cues. By recording the bats movements in an open-field type behavioral setup (as opposed to a Y-maze or other choice paradigm), we were able to exploit this behavior and quantitatively describe the search routes bats followed while localizing an odor reward. We show that bats were able to find odor sources even when the measured concentration of odors in the air was very low, consistent with previous studies on bat olfactory sensitivity (Laska, 1990b) which reported detection thresholds in the range of approximately 3-15 parts per billion.

Journal of Experimental Biology • Accepted manuscript
Olfactory localization strategies are often multi-modal, with animals integrating olfactory cues with visual, mechanosensory and acoustic inputs (Cardé and Willis, 2008;Gomez-Marin et al., 2010;Vickers, 2000). While bats, including Neotropical leaf-nosed bats use vision as part of their orientation and foraging strategy (Gutierrez et al., 2014), it is unlikely that visual cues provide much detailed information. Most bat-dispersed fruits in the Neotropics do not change color with ripening, opposite of the pattern observed in many birdand primate-dispersed plant species (Kalko et al., 1996;Lomáscolo et al., 2010). Visual cues and acoustic cues are also less reliable against cluttered backgrounds (such as a fruit cluster on a leafy branch), and it has been shown that removal of visual cues does not significantly impact bat foraging success (Korine and Kalko, 2005;Thies et al., 1998).
Like other Neotropical leaf-nosed frugivores, Sturnira produces low intensity, high frequency echolocation calls, with peak frequencies ranging from 65 kHz to 92 kHz (Jennings et al., 2004;Yoh et al., 2020) emitted via the nose. Fruit and nectar feeding bats within the family Phyllostomidae (including Sturnira) are thought to primarily use echolocation for general orientation, as well as final approach and selection of food items (Gonzalez-Terrazas et al., 2016;Kalko and Condon, 1998;Leiser-Miller et al., 2020;Thies et al., 1998). We controlled for potential effects of echo-acoustic information during odor localization in Experiment 2. Bats performed better than expected by chance even when the echolocation cue was paired with the non-rewarded, no odor control (Experiment 2A) and there was no obvious echolocation cue (Experiment 2B) ( Figure 3A). Based on these results, we conclude that acoustic cues did not significantly contribute to the bats ability to discriminate the odorized targets and that that the primary sensory cue bats were using in these assays was olfaction. This is further supported by observations that even when bats chose the wrong side (S-), they did not attempt to consume the control sponge, which would be predicted to have the same acoustic signature at the S+ sponge, while they often bit and tasted the banana-scented sponge.
We observed a peak in decision distance at 25 -35 cm from the odor source for successful attempts across all concentrations. This distance coincided with an inflection in the steepness of the odor gradient which provided optimal conditions for bats to detect spatial differences and orient towards the higher concentrations. At distances where the odor is detectable but the concentration gradient is still shallow, large movements or changes in direction (casting) are more efficient (Catania, 2013). Once the gradient becomes steeper near the source, short movements, head scanning and bilateral inputs may be sufficient to find an odor source (see for example Figure 7 in Catania, 2013;Jinn, 2019). That this distance is also about the same as the observed olfactory decision distance of mice following an odor plume (Liu et al., 2020) suggests this may be a common pattern across mammals.
Our trajectory analysis identified four distinctive search locomotor patterns routinely displayed by all bats within the experimental chamber. (Figure 2A). Since bats are also expected to perform this task in flight at high velocities, we anticipated the possibility of exaggerated or unusual locomotor patterns relative to terrestrial mammals such as dogs or rodents. Contrary to expectations, none of the recorded tracks exhibited the forward zig-zag pattern that characterizes the olfactory tracking trajectories displayed by walking mammals or flying insects (Svensson et al., 2014;Vickers, 2000). The least common pattern, top casting, had broad lateral movements back and forth across the top of the arena that could be characterized as zig-zagging, but these zig-zag motions rarely resulted in net forward motion.
This type of movement at the edges of a concentration gradient are consistent with the model posed by Catania (2013), with large movements and serial sampling helping to provide directional information in shallow gradients. The most commonly observed successful locomotor pattern was the direct strategy, representing a relatively direct search pattern with no major changes in orientation during the track ( Figure 3A). Assuming bats are receiving motivational cues from the top of the arena, then this strategy could be compared to the "aim and shoot" strategy used by some flying insects to locate odor sources (Cardé and Willis, 2008), which does not always result in a successful search , similar to what we observed in our experiments. This downward movement can be paired with serial sampling as observed in other taxa (Catania, 2013;Liu et al., 2020), allowing the animal to more accurately reassess the direction of the odor gradient when they get nearer the odor source (Jinn, 2019;Thesen et al., 1993). This behavioral strategy is also consistent with the middling locomotor pattern we observed, wherein the bats moved down the center of the chamber until they had sufficient directional information within the odor gradient to select the correct direction.
Bats may be able to pair movement with increased active sampling, such as sniffing (Baker et al., 2018;Khan et al., 2012;Vergassola et al., 2007) and simultaneous headscanning (Gomez-Marin et al., 2010;Khan et al., 2012). Sniffing and head scanning improve the efficiency of klinotactic olfactory localization by allowing an organism to maintain its body orientation within an odor plume, while permitting a longer period to sense and integrate the chemical signal (Dusenbery, 1992). Liu et al (2020) proposed that at distances far from the source serial sampling (sniffing) is performed with whole body movements, which may be replaced by increased head scanning as mice approach the odor source. In contrast, the bats in our study performed most of their head scanning movements at the top of the arena before moving towards the odor source ( Figure 5A) and were only observed when bats were stationary. As these bats use echolocation for orientation (Hernández-Canchola et al., 2020) and bats are known to use head movements to keep biosonar beam projections fixated on obstacles and targets (Surlykke et al., 2009), we were not able to separate head movements associated with sniffing from those associated with biosonar emissions. It remains possible that bats process olfactory inputs during passive breathing and echolocating (Eiting et al., 2014;Wachowiak, 2011) but the predominance of biosonar for navigation may preempt the use of head scanning purely for olfactory search. Although more research in this area is needed, this observation represents a key departure from the current synthesis of olfactory search models proposed for mammals (Baker et al., 2018;Catania, 2013;Liu et al., 2020). This, in combination with having narrow nostrils for emitting pulses through the nose suggests that bats may be constrained in their ability to use stereo-olfaction and head scanning during the final approach phase of olfactory searches.
Trial-and-error or route-following strategies could help bats overcome the trade-offs between echolocation and serial sampling. Of the four locomotor patterns observed, bottom casting appeared to be consistent with what has been termed route-following in other animals. This strategy consisted of rapidly approaching one of the targets and coming within several centimeters of S-before sharply changing direction towards the S+, which suggests the bats were following a route with a limited number of known options. Under natural foraging conditions, animals supplement sensory information such as olfactory cues with long-range navigation and cognitive strategies. Studies in rats have demonstrated that under certain circumstances (e.g., small number of targets and known locations), strategies such as route-following are faster and more robust than gradient following or casting (Bhattacharyya and Bhalla, 2015; Gire et al., 2016), particularly as familiarity with the task increases (Gire et al., 2016). In our assay, there were only two possible locations for the odor reward, which with experience shifts the olfactory task from "where" to "which" (Bhattacharyya and Bhalla, 2015). Bats, particularly nectar feeding bats, have been shown to have extraordinary spatial working memories (Henry and Stoner, 2011;Toelch et al., 2008;Winter and Stich, 2005).
Short-tailed fruit bats (Carollia) rely more strongly on spatial memory than sensory cues when foraging in the wild (Fleming et al., 1977) and spatial memory may even overshadow the use of sensory cues such as odors (Carter et al., 2010). Bat flight is also metabolically expensive, so relying on spatial memory and returning to quality foraging locations may be more efficient for foraging fruit bats than following odor plumes, provided they are exploring a known space.

Journal of Experimental Biology • Accepted manuscript
Flying bats are exposed to highly variable olfactory environments when foraging under natural conditions, but they also use olfaction while crawling in roosts or when perched in trees, where their movements are slow and the local olfactory landscape is more stable.
Our results suggest that when bats are restricted to crawling, they displayed olfactory tracking strategies similar to other terrestrial mammals, with only minor constraints arising from echolocation. Future work quantifying how bats navigate towards an odor source while flying would provide more insight into how bats use odors in their natural environment, as well as how use of olfactory sensory cues integrate with other navigational strategies such as echolocation and spatial memory. The straight-line distance from the bats' starting position to the center of S+

Distance travelled (cm)
The total distance a bat crawled in the arena before making contact with either S+ or S-

Average velocity (cm/s)
The velocity of the bat crawling in the arena, averaged over the entire trial time.

Path Straightness
A value between 0 and 1 that indicates how directly the bat moved towards its choice, calculated as the ratio of the bats' starting distance from its choice (either S+ or S-) to the total distance travelled. Values close to 1 indicate a direct route while values close to 0 indicate a more meandering path.

Decision distance (cm)
The straight-line distance from S+ at which the bat made its last change of direction before moving towards its target (either S+ or S-)

Head Scanning Behavior
The number of times a bat performed a lateral movement of the head. A single head scanning event was counted each time the bat rotated its nose at least 45 degrees to one side or the other. Path shape Visual classification of the paths followed while navigating towards the target. Trajectories were classified qualitatively into four categories: A, B, C, and D Top casting Bat moves horizontally along the top of the arena before making direct downward movement to its final choice. Direct Bat executes a direct, downward movement starting from the location where it was located at the start of the trial. Middling Bat initially moves downward but alters trajectory between one-third to one-half of the way down the arena. Bottom casting Bat makes direct movement downwards on one side of the arena, but pauses directly above target and redirects to alternate target, moving horizontally towards its final choice. Paths of this type have a distinctive "L" shape.
1 Movie 1. Search behavior of crawling Sturnira parvidens locating an odor reward. Real time video of a bat navigating the olfactory assay, following initial placement into the arena. Tracking did not begin until after the cage was closed. The search path tracing was generated automatically using EthoVision XT 13. Color represents instantaneous velocity, with dark blue representing slower time points. Trials were stopped when the bat touch the dish or reward with its nose or mouth. Video was filmed at 30 frames per second.