Departures from isotropy: the kinematics of a larval snail in response to food

Many benthic marine invertebrates reproduce through a planktonic larval phase. The dispersal and ultimate settlement of the larvae are influenced by both environmental and behavioral factors (Cowen and Sponaugle, 2009; Pineda and Reyns, 2018). Individual larvae can exhibit a wide variety of behaviors that are controlled by both their morphology and environmental cues (Young and Vazquez, 1995; Chan, 2012). The behavior of larvae at the scale of the individual can affect their ability to find and identify a suitable habitat, settle and survive (Young and Chia, 1984; Koehl, 2007), whereas at the scale of the population, larval behavior affects dispersal patterns (Metaxas, 2001). Larval behavior is not expected to be isotropic, as environmental and ecological motivations differ in each direction in the ocean. For example, upward and downward motion are associated with settlement, and horizontal motion is associated with dispersal. However, the influences of foraging and feeding on the anisotropy of larval behavior are not obvious.

Many invertebrate larvae are ciliated, beating their cilia both to swim and to feed, resulting in potential interactions and trade-offs between these two actions (Emlet et al., 1985). Larvae have been described as ‘good eaters, but poor swimmers’ owing to compromises in larval form, as increases with feeding efficiency are often met with decreases in propulsion and swimming efficiency (Strathmann and Grünbaum, 2006). Observations of the flow induced by bat star larvae indicate that some ciliated larvae may switch between feeding and swimming modes to address this trade-off, resulting in slower swimming while feeding (Gilpin et al., 2017). This suggests that the free-swimming behavior of larvae may be influenced by food availability, and that larvae may swim slower while feeding. However, these observations were of confined larvae, and therefore open questions remain regarding how their results translate to free-swimming larvae (von Dassow et al., 2017).

Molluscan veliger larvae have been the subject of many studies of larval behavior; they have opposed bands of cilia which they use to swim and feed (Pernet, 2018). Observations of veligers show they can mediate their food intake and reject food particles (Gallager, 1988). Although the exact mechanisms by which they decouple feeding and swimming remain elusive, their ability to separate the two functions suggests the larvae may be able to potentially minimize any trade-offs (Strathmann et al., 2019). In this study, we specifically consider the veliger larvae of the gastropod Crepidula fornicata as a model system. Although ciliary beating can control the strength of swimming and feeding currents, C. fornicata are of interest because their swimming speed does not directly correlate with their ciliary beating (Chan et al., 2013), but rather with velum extension and position. This suggests that swimming–feeding trade-offs in ciliated veligers may provide new insights when compared with other organisms.

Another defining characteristic of veliger behavior is that it is inherently anisotropic, as gravity passively induces a stable preferential orientation. Their heavy shells induce an ‘upright’ posture with their shell oriented downwards, and their velum oriented upwards (Kessler, 1986; Fuchs et al., 2015); veligers produce propulsion in the direction of their velum, thus they are biased toward upward swimming. In addition, they are negatively buoyant and must therefore exert a minimum swimming propulsion to keep from settling downward. In this way, downward transport can be due to passive sinking, whereas maintaining a constant vertical position necessitates active swimming. Horizontal movement relative to the fluid environment must be due to active swimming. Therefore, if a larva is swimming upward at a constant rate, and it reorients horizontally, the vertical component of the larva's motion will decrease or reverse and the larva will sink unless it increases its overall swimming speed. This suggests that any increase in relative horizontal motion may be met with changes in vertical motion.

Larval behavior is also affected by the anisotropy of the ocean environment. Light and nutrient availability are often stratified, with food being patchy and potentially distributed in thin horizontal layers. Ocean currents are depth-dependent as well. Therefore, larvae exhibit inherently anisotropic behavior, a topic that has been of general interest to zooplankton behavioral ecology (e.g. Yen 1988; Mahjoub et al. 2011; Schuech and Menden-Deuer 2014). In terms of transport, vertical swimming alters a larva's position in the water column, which can control horizontal transport in sheared oceanic or tidal currents (Shanks, 1995; McManus and Woodson, 2012). Vertical swimming occurs during ontogenetic or daily vertical migrations. Downward swimming reduces dispersal of larvae, as they are exposed to slower flow in the benthic boundary layer (Shanks, 2009). Horizontal swimming of small larvae is not necessarily strong enough to affect net transport when compared with oceanic currents (Porch, 1998), but at slow flow speeds in lagoons or near benthic substrata, horizontal swimming can affect settlement location (e.g. Butman et al., 1988; Bingham and Young, 1991; Maciejewski et al., 2019). Horizontal swimming is associated with feeding in thin layers (Menden-Deuer and Grünbaum, 2006) and food patches (Metaxas and Young, 1998; Sameoto and Metaxas, 2008). Other behaviors, such as diving or swimming in helices, may be used for feeding, predator avoidance or in response to physical or chemical cues (Fuchs et al., 2004; Fuchs et al., 2018; Wheeler et al., 2015, 2017; Maciejewski et al., 2019).

Owing to anisotropic environmental influences in the ocean, morphological constraints, and the intrinsic link between swimming and feeding in ciliated larvae, larval behavior is expected to exhibit varying levels of anisotropy. We propose measuring this anisotropy under controlled cues to investigate the response of veliger larvae to food. In this study, we specifically evaluated the behavior of larvae of the common slipper limpet, C. fornicata. This species has been studied as a model system for developmental and larval biology (Henry et al., 2010) and has previously been observed to exhibit highly variable swimming speeds and behaviors (Chan et al., 2013). We exposed fed and starved C. fornicata larvae to still-water flasks with and without food, in order to quantify and describe differences in larval behavior that may be related to feeding–swimming trade-offs. We analyzed the distributions of their kinematics, and employed behavioral anisotropy as an indicator to characterize behavioral response. We expected that the anisotropy of their behavior would change in the presence of food, with larvae exhibiting more horizontal motion while feeding. An increase in horizontal motion is expected to occur with a decrease in vertical motion. Any change to vertical motion is expected to change the vertical distributions of larvae as well. Finally, we expected that any food-mediated response in behavior would be more prevalent in starved larvae.

Adult Crepidulafornicata (Linnaeus 1758) were collected in Vineyard Sound, MA, USA. Stacks of individuals were collected from 30 m depth and were transported to Woods Hole Oceanographic Institution. Adults spawned on 2 July 2018 after spending approximately 1 week in captivity, and larvae were collected within 24 h by reverse filtration with 150 µm mesh.

Larval C. fornicata were cultured in clean 2 liter glass jars, which were filled to 1.5 liters with 0.2- µm filtered seawater. Every day, larvae were transferred to clean jars and fed with a Tahitian strain of Isochrysis galbana, resulting in a concentration of 1.8×10⁵ cells ml⁻¹. This concentration was also maintained as the food source in the swimming experiments. After a week in culture, larvae were split into fed and starved treatments, with fed larvae continuing to be fed as normal and starved larvae receiving no food for 4 days prior to the experiment. The experiments were conducted 9 days post hatch. The diameters of the larvae were measured under a dissecting microscope. The fed and starved larvae measured on average 522±9 and 508±6 µm, respectively.

To isolate the larval response to the presence of food, we conducted experiments that manipulated only food availability and history of starvation. Many other external variables can affect larval behavior, such as light, temperature, salinity, predators and flow cues. We have excluded those variables in our experiment, yet cue hierarchies undoubtedly exist and should be investigated further.

Behavioral experiments were conducted for fed (Fed) and previously starved (St) larvae, either in the presence of food (F) or without food (0) in filtered seawater, resulting in four distinct treatments. The food levels in the F treatments were at a satiating concentration level, identical to the level used in culturing the larvae. For each treatment, five replicate experiments were conducted. Each replicate contained observations of approximately nine to 16 individual larvae, resulting in concentrations of 0.18–0.32 larvae ml⁻¹. For each treatment, the total number of larvae observed across all five replicates was 72, 66, 70 and 62 individuals in the St-0, St-F, Fed-0 and Fed-F treatments, respectively. The experiments were run over long enough times that the larvae could swim the length of the flask multiple times in the observation period, resulting in converged kinematic distributions. The sequence of replicate flasks and environmental treatments was randomized using a random number generator.

All larval swimming observations were conducted in a temperature-controlled chamber at 20°C in the dark to eliminate any thermal or phototactic influences on larval behavior. Each experimental trial was conducted in a new 65 ml flat-sided flask; the flasks measured 4.5 cm wide, 2.3 cm deep and 6.1 cm tall. The flask was filled to a depth of 5 cm with filtered seawater (no food) or well-mixed I. galbana in filtered seawater (food). Crepidulafornicata larvae were concentrated into a small volume of water and introduced to the experimental flasks on a small piece of 100 µm mesh, which was inserted into the flask with forceps. The flask was illuminated from behind with a diffuse near-infrared LED array (Olymstore, 12 V, 2A, 850 nm), and larval swimming behaviors were recorded over 5 min at 30 frames s⁻¹ with a CMOS camera (Basler acA2040-90umNIR). The recorded field of view focused on the center plane of the flask, and showed a vertical cross-section of the flask that included the full vertical and horizontal extent of the flask.

Video observations of larval behavior were recorded as a series of .tiff images, which were used for particle tracking. Larval centroids were identified in the images with MATLAB, and the x- and z-positions and horizontal and vertical (u,w) velocities of each larva in each frame were evaluated with particle-tracking algorithms based on Kelley and Ouellette (2011), which use a predictive algorithm to link trajectories and a Gaussian smoothing and differentiation kernel to calculate velocities. The angular velocity ω of each larva in each frame was calculated based on the rate of change of the swimming direction. Example trajectories and raw video footage for each treatment are shown in Fig. S1 and Movies 1 and 2.

The particle-tracking velocities calculated for each larva represent the resultant velocities of the larvae, based on the combined propulsive swimming force, gravitational force and hydrodynamic forces such as drag and added mass. Therefore, the direction of the larval motion is not necessarily the direction of their swimming propulsion alone (see Fuchs et al., 2013). Swimming indicates an active response, whereas larvae sink passively; therefore, observed larval speed is not necessarily proportional to larval swimming strength. Because of the small Reynolds number of C. fornicata larvae and the still water in which they were observed, we assume the hydrodynamic forces are much smaller than the forces owing to gravity and propulsion. Therefore, we assume all horizontal behavior is due solely to active swimming, and vertical behavior is due to swimming plus a constant gravitational force that is a function of the larval size and density.

From the larval velocity measurements, we also assessed the direction of larval motion. Direction was calculated as the arctangent of the instantaneous velocity ratio w/u. The directional behavior of the larvae was assessed by the comparison of the probability density functions (PDFs) of their movement direction for each treatment. To isolate strictly horizontal and vertical behaviors, we calculated the fractions F_horiz and F_vert of direction observations within 15 deg of (0 deg, 180 deg) and (90 deg, 270 deg), respectively. Confidence intervals were quantified by bootstrapping the data (Efron, 1979).

After the larvae were introduced to the flask, there was an adjustment period of approximately 10 s, during which most individuals went to the bottom of the flask. This time period was excluded from our analysis. In many instances, larvae that encountered the upper free surface of the water repeatedly dipped below it and re-approached it (as seen in Fig. S1D). Larvae at the free surface and on the bottom of the flask were also difficult to track owing to imaging interference based on refraction of light by the clear flask. Therefore, trajectory segments that were within 1 cm of the top or bottom of the flask were excluded from the kinematic analyses.

Many studies of larval behavior focus on either distinct behaviors (diving, helices) or point-statistics such as average swimming speed (Meyer et al., 2018; Maciejewski et al., 2019). However, these statistics lack complexity, which can be important for accurately reproducing the observed distributions of larvae (e.g. Daigle and Metaxas, 2012). Many larval behaviors fall along a continuum, so instead of classifying individual trajectories, we used aggregate statistical approaches to classify behavioral differences. We treated each instantaneous value of horizontal velocity, vertical velocity, angular velocity and speed as a single data point and constructed PDFs of these parameters. This resulted in over 10,000 data points for each treatment, pooled across all replicates.

The vertical distribution of larvae was assessed by calculating the vertical centroid of the observed trajectories. For each of the five replicates for each treatment, the average vertical position was calculated for all tracks over the last minute of the observational period, resulting in five data points per treatment. All larval trajectory data were used; larvae that disappeared from the field of view, either by approaching the top of the flask or the bottom, were assumed to stay at their respective vertical location until they reappeared. Note that no velocity data were available for the larvae after they had disappeared from the field of view. The validity of this approach was confirmed by visually inspecting the raw images. To test for significant differences in larval position in the water column, we used a two-way crossed ANOVA with fed/starved and with/without food as factors, plus the interaction. A Tukey test was used for pairwise post hoc tests.

Velocity data for larvae in each treatment were analyzed in aggregate. Individuals had a range of velocities, which were characterized by calculating PDFs for velocity values in each treatment. For each velocity distribution, we calculated the root mean square (RMS) to characterize the overall magnitude and spread of velocity behavior. We do not report mean directional velocity values, as the horizontal velocity mean values were approximately zero, and the vertical behavior of the larvae was constrained by the flask. The RMS statistics were bounded by 95% confidence intervals (CI), which were calculated by bootstrapping the data (Efron, 1979).

We compared the velocity data with an ideal model of isotropic behavior in order to quantify the degree of anisotropy. The model is based in a model of diffusion (Okubo, 1994); however, owing to the confined nature of the experiments, total diffusion (dispersal) of the larvae could not be directly quantified, as the larvae eventually encountered the boundaries of the flask. A diffusion model assumes the larvae disperse based on a diffusion coefficient D≈U²τ, where U is a characteristic velocity scale and τ is the correlation time scale of the larval motion, which can be thought of as a turning timescale (Porch, 1998; Visser and Kiørboe, 2006). Instead of approximating D, we report velocity kinematics directly. We assumed that isotropic diffusion of the larvae would result in normally distributed velocity distributions, leading to larvae that diffuse in all directions uniformly. From normally distributed velocities, it follows that the distribution of larval absolute speed would follow a Maxwell–Boltzmann distribution, which we employed to characterize total larval speed.

It follows that the expected value of V is ⁠, and thus only σ is needed to characterize the magnitude of the distribution. To compare speeds between treatments, the observed V PDFs were fit to Eqn 2 with σ as the fitting parameter. The curve-fitting was conducted by minimizing the mean squared error weighted by Eqn 2. This process was repeated recursively until the value of σ converged for each treatment.

where Σ_if(i)=Σ_if₀(i)=1. Note that this formulation of χ² is independent of sample size, and thus allows for direct comparison between distributions with variable sample sizes. For each distribution, we used the same binning to calculate Eqn 3. Uncertainty of χ² is assessed by bootstrapping the data to yield 95% CI. For identical distributions, f=f₀ and χ²=0; the larger the value of χ², the larger the difference between the distributions.

The observed velocity (u,w) PDFs for each treatment are plotted in Fig. 1. The horizontal velocity u distributions (Fig. 1A) follow symmetric, normal distributions much more closely than those of the vertical velocity w (Fig. 1B), where the latter display more off-center peaks and asymmetry. We quantitatively assessed the degree of normalcy in the distributions with the χ² statistic, reported in Table 1. Across all treatments, χ² of u is lower than for w, which indicates that larval swimming velocities in the vertical direction deviate more strongly from a normal distribution than in the horizontal. The lowest χ² values for and occur for the St-0 treatment, indicating that the St-0 treatment exhibited larvae with velocity behavior that best agrees with a normal distribution.

In comparing the velocity distributions across treatments, the horizontal velocity distributions follow a clear pattern, whereas the distributions of the vertical velocity are more complicated. There is a higher degree of spread, and a larger range of velocities for the horizontal velocity u of the Fed treatment larvae when compared with the starved (St) treatment, as well as with the presence of food (F) in both treatments. With respect to the vertical velocity w distributions, the Fed treatments also show more spread and a larger range of velocities when compared with the St treatments. Also, the St-F treatment exhibits a larger range of both upward (positive) and downward (negative) vertical velocities when compared with the St-0 treatment. Similar faster vertical velocities are observed for the Fed-F treatment compared with the Fed-0 treatment, but only for the downward velocities. The upward vertical velocities show similar ranges of values for both Fed treatments, with a peak of fast upward velocities in the Fed-F treatment. Owing to the large sample sizes of each data set, all calculated distributions are significantly different from each other and the normal distribution with P-values <0.001 (Pearson's chi-squared test).

Absolute observed two-dimensional speed V PDFs for each treatment are shown in Fig. 2A. To further characterize and compare the curves, best fits to the data for the two-dimensional Maxwell–Boltzmann distribution are shown in Fig. 2B, with the fitting parameters reported in Table 2. The fitted distribution parameter σ characterizes the overall speed of the larvae for each treatment. Comparing σ between treatments allows us to quantify the response of the larvae to the presence of food in both feeding history cases. In response to food, an increase in total larval speed is seen with similar magnitude for both starved and fed larvae, Δσ=0.0078 and 0.0068 cm s^–1, respectively, suggesting that the increase in larval speed owing to the presence of food is not dependent on whether the larvae were previously starved. The data also show that the starved larvae on average move at slower speeds than the fed larvae.

The horizontal and vertical larval velocities are further assessed by calculation of the RMS values, plotted in Fig. 3A. The St-0 larvae have the lowest u_RMS and w_RMS values out of all treatments, which agrees with the observation that they had the lowest absolute speeds. The St-F larvae have higher u_RMS and w_RMS relative to the St-0 treatment, so the presence of food increases larval speed in both the horizontal and vertical directions for the starved larvae. The increase in horizontal speeds must be a direct result of increased horizontal swimming, but the increased vertical speed could be due to both higher vertical swimming speeds, as well as faster passive sinking. The fed larvae show similar trends in the horizontal velocities: u_RMS for the Fed-F treatment is larger than u_RMS of the Fed-0 treatment. However, there is no difference between w_RMS owing to the presence of food for the fed larvae. Therefore, we see that the presence of food increases the magnitude of observed horizontal velocity for both fed and starved larvae, but only increases the observed vertical velocity for the starved larvae.

In order to assess anisotropy of larval motion, we compared horizontal and vertical RMS velocities for each treatment. The ratio u_RMS/w_RMS is shown in Fig. 3B. A ratio close to 1 indicates isotropic behavior. By this definition, the St-0 treatment has the most isotropic larval behavior, while larvae in the St-F, Fed-0 and Fed-F treatments have more anisotropic behaviors, with the ratio u_RMS/w_RMS≈0.5 for the St-F, Fed-0 and Fed-F larvae treatments. Therefore, the larvae are moving approximately twice as fast in the vertical as in the horizontal direction. In all cases, the ratio u_RMS/w_RMS<1, indicating that all of the larvae observed on average move faster vertically than horizontally.

Another measure of larval behavior is their vertical distribution in the water column. This is assessed by comparing the mean vertical position of larvae ⁠, which varies by treatment, as shown in Fig. 4. Higher values indicate that larvae are higher in the water column. Starved larvae without food remained near the bottom of the flask, but larvae were significantly higher in the water column when fed (ANOVA, d.f.=1, F=5.96, P=0.02) and when in the presence of food (ANOVA, d.f.=1, F=16.3, P<0.001). There was no significant interaction of fed/starved or with/without food (ANOVA, d.f.=1, F=0.49, P=0.49). Pairwise post hoc tests revealed significant differences between the treatments St-0 and St-F (P=0.01) and St-F and Fed-0 (P=0.001). These findings show that when food was present, larvae were found higher in the water column, and that this relationship was strongest for the starved larvae.

The PDFs of larval movement direction are plotted for each treatment in Fig. 5A. In both Fed treatments (Fed-0 and Fed-F), larvae are observed traveling downward the majority of the time. Larvae in the St-0 and St-F treatments have a higher fraction of observed horizontal motion, with the largest horizontal peak in the St-F treatment.

Directional behavior is further quantified in Fig. 5B, where the exact fractions of observations of horizontal and vertical movement (F_horiz, F_vert) are reported for each treatment. In terms of horizontal movement, while the presence of food does increase F_horiz in both the starved and fed larvae, the largest increase is seen in the starved larvae, where the St-F treatment has the highest value of F_horiz of all treatments. The fraction of vertical movement is also shown for reference, which is also seen to increase from the St-0 to St-F treatments, with no change seen between the Fed-0 and Fed-F treatments. These observations indicate that, especially for the starved larvae, the presence of food increases the frequency of horizontal motion, but in all treatments, the fraction of horizontal motion remains less than that of vertical motion.

Isotropy of the directional PDFs is assessed by comparing the distributions to a uniform distribution, quantified through χ², reported in Table 3. A lower χ² value indicates higher isotropy. Therefore, the St-0 treatment has the most isotropic directional behavior, with almost equal probability of larvae swimming in every direction. The χ² values show that the presence of food increases larval directional anisotropy in all treatments. Furthermore, starved larvae have more isotropic behavior when compared with the fed larvae treatments.

To further assess the larval vertical behavior, we divided the observations in each treatment into upward and downward moving. We compared absolute speed V and angular velocity ω for each case in each treatment (Figs 6 and 7). Larvae in the St-0 treatment had almost identical speed and angular velocity distributions when moving upward and downward (Figs 6A and 7A), indicating a strong isotropy in larval vertical behavior. This symmetry breaks down, however, in the presence of food: the St-F speed distributions differ between upward- and downward-moving larvae (Fig. 6B). This result shows that the presence of food can increase behavioral anisotropy.

For both starved and fed larvae in the presence of food, there is a higher peak in the PDF at lower speeds for downward moving relative to that of upward moving. This peak indicates that larvae are more often moving slowly going down than when going up. In all treatments except St-0, there are strong differences in the shapes of the speed PDFs between upward and downward movement. We also calculated and plotted mean absolute horizontal velocities and vertical velocities for the upward- and downward-moving larvae for each treatment in Fig. 6C,F. In all but the St-0 treatment, the upward mean horizontal velocities are higher than the downward velocities, with the difference being the largest in the Fed-F treatment (see Fig. 6C). The differences in the mean absolute vertical velocities are less dramatic, with higher speeds downward in the St-F and Fed-0 treatments, and approximately equal upward and downward speeds in the St-0 and Fed-F treatments (see Fig. 6F). These differences are small, do not show a clear trend, and do not provide as much information as the whole PDFs.

Angular velocity PDFs are nearly identical for upward- and downward-moving larvae in the St-0 and St-F treatments, indicating strong angular isotropy for these larvae. Overall, differences in angular velocity between treatments are small, as shown from total ω_RMS in Fig. 7F. However, differences in angular velocity ω are apparent between upward- and downward-moving larvae in the Fed-0 and Fed-F treatments, as shown in Fig. 7D,E; the tails of the distributions for ω are higher for upward-moving larvae than for downward-moving larvae in these treatments. The ω_RMS values differ between upward- and downward-moving larvae in the Fed-0 and Fed-F treatments, reinforcing this result (Fig. 7C). These results show that fed larvae turn more as they swim up, which could be related to feeding or exploratory behavior. It is also likely related to the fact that downward motion is more passive than upward motion.

We assessed the swimming behaviors of C.fornicata larvae in response to food by measuring the deviations from a model that assumes isotropic kinematics, normally distributed directional velocities, and a uniform directional distribution. The larvae were observed in still-water flasks either with or without food, and were either previously fed or starved. Our results revealed anisotropic larval behavior across treatments, as expected. However, we did observe largely isotropic behaviors in the St-0 treatment (starved larvae without food), indicated by: (i) normally distributed horizontal and vertical velocities, (ii) similar magnitude of horizontal and vertical velocities, (iii) similar distributions of upward and downward speeds, and (iv) the most uniform velocity direction distribution of all treatments. Therefore, we found that an isotropic diffusion model of behavior is a good approximation for the observed starved C. fornicata larvae without food. In other words, in the absence of food, previously starved larvae tend to move in all directions with equal probability and similar speed.

However, in the field, larvae are rarely – if ever – completely starved and without food. Our results showed increased behavioral anisotropy for larvae that were fed and in the presence of food. Food sources in the environment are patchy, both horizontally and vertically (Daro, 1988). Because of this environmental variability, larvae may constantly readjust their behaviors based on their immediate surroundings, thus influencing dispersal and settlement. Our findings further indicate that the anisotropy of the larval behavior is not static, and may be another useful parameter in quantifying behavioral response.

The observed increase in both magnitude and frequency of horizontal swimming in response to food may be motivated by the fact that phytoplankton are often present in thin horizontal layers in the environment (Durham and Stocker, 2012). An increase in horizontal swimming, and a corresponding decrease in vertical movement, would increase larval residence times and aggregation at a specified depth. Changes in the kinematics of plankton in response to thin layer cues have been observed in laboratory settings. Aggregation behavior was observed in heterotrophic dinoflagellates in still water, where an increase in both horizontal motion and swimming speed, and a decrease in vertical velocity, were observed in the presence of a thin layer food cue (Menden-Deuer and Grünbaum, 2006). Avoidance behavior has also been observed in copepods when presented with a thin layer of chemical toxins from harmful algae in the laboratory (True et al., 2018). In our observations, while horizontal velocity magnitude (Fig. 3A) and frequency of horizontal movement (Fig. 5B) increased in the presence of food for all larvae, increases in vertical swimming magnitude and frequency were also observed in the presence of food for the starved larvae. Therefore, we found that starved larvae met their increased horizontal motion with an increase in their vertical motion, and aggregation was not observed. However, we note that the experiments of Menden-Deuer and Grünbaum (2006) explicitly used phytoplankton in a thin layer, whereas we used a well-mixed concentration. It remains to be seen whether C. fornicata would exhibit aggregation behavior when encountering a thin layer of food, and we recommend future laboratory experiments to explore this hypothesis.

Overall speed of the larvae also increased in the presence of food for both the previously fed and starved larvae. In this case, speed includes both passive motion owing to sinking and active motion owing to swimming. However, the swimming effort can be partially isolated. We found the average upward vertical velocity of the larvae increased with the presence of food in all treatments (see Fig. 6F); this is direct evidence of increased swimming effort in response to food. We also saw increases of horizontal velocity magnitudes in the presence of food in both fed and starved treatments, another direct result of increased horizontal swimming effort (see Fig. 3A). The largest increase in horizontal velocity magnitude was seen for the larvae traveling upward (Fig. 6C). These observations show that larvae increased both their vertical and horizontal swimming efforts in the presence of food, thus indicating a direct link between foraging and increased swimming speed. Although there was no direct observation of feeding in this study, the measurements of larval velocity in the presence of food suggest that the larvae can feed without any reduction of swimming speed.

The amount of vertical larval motion also changed across treatments (see Fig. 5B). Compared with the treatments without food, the starved larvae in food were observed moving vertically more often, and the fed larvae in food were observed moving vertically an equal amount. This does not agree with the hypothesis that vertical swimming is an appropriate foraging strategy to encounter horizontal thin layers of food in the ocean environment. The starved larvae without food would theoretically be more desperate to encounter food and swim vertically, yet it was the fed larvae without food that exhibited the highest fraction of vertical motion. The fed larvae exhibited an effective foraging strategy in the absence of food by increasing their swimming anisotropy with a high frequency of vertical motion and a low frequency of horizontal motion. This discrepancy between the fed and starved larvae observations may be explained by an energy-conserving strategy of depth-keeping for the starved larvae in the absence of food.

Our observations further demonstrate differences in the shape of the distributions of horizontal and vertical velocity. Across all treatments, the horizontal velocity distributions of the larvae are more isotropic and normally distributed than the vertical velocities. This suggests that larvae may have more diffusive-type motion in the horizontal direction than in the vertical direction, which agrees with observations of marine protists in the laboratory (Schuech and Menden-Deuer, 2014). In addition, it illustrates fundamental differences in vertical and horizontal behavior, as expected. Vertical migration controls how larvae disperse in sheared ocean currents, and spending more time at the surface when feeding might mean larvae are exposed to faster currents that could affect their dispersal distance. Horizontal motion of larvae is more likely to influence how larvae interact with their local environment, for example when settling. Side-to-side motion when swimming in helices may help larvae locate settlement cues (Maciejewski et al., 2019) and can influence dispersal over small spatial scales in calm waters (Bingham and Young, 1991).

Changes in the vertical distribution of larvae were also observed in response to food, with the strongest response for the starved larvae. Overall, larvae were observed higher in the flasks in the presence of food in both treatments. This is consistent with previous work that found changes in vertical distribution as a response to food for other larval species (Metaxas and Young, 1998; Arellano et al., 2012). The vertical distribution calculations in the present study included larvae at the bottom of the flasks that were not swimming. Thus, not only did the presence of food alter how the observed larvae swam, it affected their swimming propensity. Larvae that remain near the seafloor tend to disperse less (Shanks, 2009). In contrast, foraging behavior can reduce larval dispersal distances by an order of magnitude relative to distances predicted by passive transport (Woodson and McManus, 2007).

Our observations correspond with the anatomical swimming structure of larval C. fornicata. Specifically, the distinction between behaviors in the vertical and horizontal dimensions are most likely due in part to anatomical constraints on veliger swimming. Crepidulafornicata larvae and other veligers have a heavy shell; this shell provides a torque that keeps the larvae oriented with the velum pointed upward (Chan et al., 2013). This torque biases the larvae to produce propulsion upward most often when swimming. Also, because larval C. fornicata are negatively buoyant, they must actively swim upward, but can passively sink downward. We hypothesized that the difference in active and passive behaviors would create an anisotropy in vertical behaviors. As expected, the upward- and downward-moving behaviors of larvae showed differences across most treatments. Specifically, angular velocity and horizontal speed of the upward-moving larvae were larger than or equal to those of the downward-moving larvae (Figs 6C and 7C), illustrating more active behavior when swimming upward.

The shell of C. fornicata grows as the larvae mature (Pechenik, 1984), so swimming anisotropy may also be a function of age. In a previous study, C. fornicata larvae swam fastest at 6 days post-hatching, with older larvae swimming more slowly and exhibiting more settlement behavior (Chan et al., 2013). Swimming speed in C. fornicata larvae depends most heavily on velum size (Chan et al., 2013). Velum size can be influenced by food availability during growth, as the velum of C. fornicata has been observed to grow faster under reduced food concentrations (Klinzing and Pechenik, 2000). In this work, we found similar responses to food regardless of feeding history. Specifically, a similar increase in larval speed magnitude for both starved and fed larvae (0.0078 and 0.0068 cm s^–1, respectively) was observed, suggesting that starvation history is not as important as the intrinsic behavioral response to food availability.

In many studies, observations of larval swimming behavior are presented as average values, either averaged across individuals or over time. Our results reveal that averaged values, which lack information about the range of observed behaviors, do not properly represent larval behavior. For example, the mean horizontal velocities for larvae in each treatment are zero, but the RMS values are non-zero and show a clear trend. In a vertically constricted experiment, average vertical velocities also are not as representative of behavior as vertical distributions. In addition, averaging inherently filters the data, removing information. Behavior of C. fornicata is unsteady, so describing it with average values does not capture the full range of observed behaviors. Because individual behaviors are often difficult to classify, we recommend using distributions of velocity as well as anisotropy indicators to continually assess responses to environmental cues.

As accuracy and resolution in hydrodynamic simulations improve, it is important to include realistic models of behavior (Metaxas and Saunders, 2009). Therefore, we must continually reassess how well existing models capture real larval behaviors and where they may fall short. Behavior is intrinsically linked to dispersal, so future modeling studies should include more refined models of larval behavior in the field, including increased swimming speed, increased horizontal swimming and higher vertical position in the water column of C. fornicata when food sources are encountered.

We thank Jan Pechenik (Tufts University) for assistance in culturing Crepidula fornicata larvae and Anthony Pires (Dickinson College) for helpful discussions over the course of this work. We also thank the anonymous reviewers for their valuable comments, which have greatly improved the paper.