The recent collapse of predatory sunflower sea stars (Pycnopodia helianthoides) due to sea star wasting disease (SSWD) is hypothesized to have contributed to proliferation of sea urchin barrens and losses of kelp forests on the North American West Coast. We used experiments and a model to test whether restored Pycnopodia populations may help recover kelp forests through their consumption of nutritionally poor purple sea urchins (Strongylocentrotus purpuratus) typical of barrens. Pycnopodia consumed 0.68 S. purpuratus day⁻¹, and our model and sensitivity analysis shows that the magnitude of recent Pycnopodia declines is consistent with urchin proliferation after modest sea urchin recruitment, and even small Pycnopodia recoveries could generally lead to lower densities of sea urchins that are consistent with kelp-urchin coexistence. Pycnopodia seem unable to chemically distinguish starved from fed urchins and indeed have higher predation rates on starved urchins due to shorter handling times. These results highlight the importance of Pycnopodia in regulating purple sea urchin populations and maintaining healthy kelp forests through top-down control. The recovery of this important predator to densities commonly found prior to SSWD, whether through natural means or human-assisted reintroductions, may therefore be a key step in kelp forest restoration at ecologically significant scales.

1) Sea urchin conditioning

Data: gonads.xlsx

Code: Gonad Index Analysis.R

Method: Our interest was in whether Pycnopodia will prey on  S. purpuratus that exist in sea urchin barrens, which have minimal gonad content and low nutritional value to predators. However, S. purpuratus are rare in the San Juan Islands and do not currently form urchin barrens. We therefore used a conditioning protocol to create urchin-barren analog urchins by starving and spawning out urchins and kelp forest analog urchins by feeding the urchins with kelp ad libitum prior to the predation trials, summarized in Supplementary Table 1. We conducted 3 collection dives over July 18-20 at known populations of purple urchin sites near Cattle Pass, Lopez, and San Juan Islands. We collected 300 S. purpuratus from rocky subtidal habitat with healthy bull kelp forests and high flow, in roughly the interquartile range of the sizes we observed in the field; we avoided the largest and smallest of the urchins. The sizes of the urchins used in the experiment ranged from 45–70 mm (57.9 ± 5.3, Mean ± SD). Sea urchins were transferred to the lab after capture and sorted into two shaded long-term holding tanks with an equal number of urchins of roughly the same test diameter size distribution. We randomly assigned each tank to the feeding or starvation treatment. On July 20, 2020, we started to feed the urchins in the feeding treatment tank bull kelp (N. luetkeana) blades, and we did not feed urchins in the starvation treatment tank. On July 30 (10–12 days after collection), we sacrificed 20 new urchins collected from the same field site as our other urchins to assess gonad index from ‘wild’ urchins for comparison with our experimental urchins. 

From August 15-17 (~1 month after collection), we induced spawning in the urchins in the starvation treatment, in order to further diminish gonadal condition. Urchins were injected through the peristomal membrane with 0.05M KCl at 0.02mL of KCl per 1mL urchin volume (e.g. 1mL KCL for a 50mL urchin). In total, 53% of starving urchins were successfully spawned (gamete release observed) and of the urchins that spawned, 27% were female and 73% were male. All injected urchins were allowed to recover in holding tanks for at least 19 days before the start of any sea star experiment, and urchins continued to starve during the recovery period.

Gonadal condition was assessed several times throughout the experiment to measure changes to gonad through time: July 30 (n=13, “Wild”); September 7-10 (n=15 “Pre-Experiment Starved” and n=14 “Pre-Experiment Fed”); October 22-23 (n=9 “Post-Experiment Starved” and n=15 “Post-Experiment Fed”). Whole urchin weight (g), test diameter (mm) and height (mm) were recorded prior to dissection. Urchins were dissected along the equator and all internal tissues were carefully removed. Digesta was removed from the gut, and tissues were blotted for 30 seconds prior to weighing. Gonad wet weight (g) in addition to gut plus gonad weight (g) were recorded. Gonadal index was calculated as 100*[gonad weight/urchin volume in cm3]. Urchin volume was calculated using the formula for an oblate spheroid as [4/3π * radius2 * (0.5*height)]. We examined gonad tissue under a compound microscope to determine sex. As expected, we documented a significant difference in gonadal condition between fed and starved urchins within 40 days and this difference was maintained throughout the experiment days (Supplementary Fig. 2, Supplementary Table 3). To determine if our conditioning treatments were indeed reflective of the body condition of sea urchins in kelp forests and urchin barrens, we compared the gonad indices of our starved and fed treatments to published gonadal index data from the Channel Islands, California during the spawning season (December to March) of 2011-2014. We analyzed the differences in gonad index (see calculation above) between fed, starved, kelp forest, and urchin barren urchins using a generalized linear model (stats package) and follow-up tests (lsmeans package) in R 4.0.0 and RStudio 1.2.5042.


2) Pycnopodia prey choice experiment

Data: Trial 1:  "singleurchin.csv"; Trial 2: "Starved-Fed.csv",

Code: YMaze.Rmd

Method: We performed prey choice experiments on each star to assess whether Pycnopodia could detect chemical cues and exhibit a preference for nutritionally valuable fed sea urchins typical of a kelp forest over starved sea urchins typical of an urchin barren. We ran each of the 24 stars through 2 different y-maze trials, using a large plexiglass y-maze (57cm width x 170cm length). In the choice experiment, all 24 stars had previous experience with purple sea urchins, since the choice experiments were performed following the predation rate experiments. In the two trials, a star was placed at the base of the maze, with 1) one side of the maze having no urchin, the other having several fed or starved urchins; and 2) one side of the maze with fed urchins and the other with starved sea urchins (Supplementary Fig. 3). Trial 1 assessed whether the stars could identify any prey in the y-maze experiment, and trial 2 assessed whether the stars selected fed or starved sea urchins. The run was scored as a ‘choice’ for the prey if the sea star advanced up the y-maze 90cm. Occasionally the stars did not move from the start of the maze, and the run was scored as ‘no choice’ and was not included in the analysis. To minimize side bias, water flow rate in each side of the y-maze was equalized before the start of each run, and the treatments were alternated between left and right sides of the y-maze. Total runs were N=19 and N=21 for trial 1 and 2 respectively. To determine if one choice was more likely than the other, we ran a chi-squared test for each trial, using chisq.test() in the stats package in R

3) Pycnopodia predation rate experiment

Data: SuppPycnoInfo.xlsx, FeedingRate.xlsx, CapturedUrchins.xlsx, FeedingRateTimeSeries.xlsx

Code: No code, performed in JMP

Methods: To quantify Pycnopodia feeding rates on S. purpuratus, we divided the 24 stars randomly into 4 groups and allocated treatments to the stars randomly. We set up predation experiments as 6-7 day trials, buffered by a 24-hour acclimation period after transferring a star to a new location before the start of any trial (see Supplementary Tables 1 and 2). We used 12 replicate aquaria (60cm wide x 90cm long x 30cm deep) with an inflow of seawater on one side and the outflow on the opposite end. We covered the tanks with two large outdoor tents so they received no direct sunlight, but were still exposed to indirect ambient light and with a natural light cycle. To protect against time as a confounding factor on the experiments, all stars were randomly assigned to an experimental schedule. We conducted a total of 4 trials, so that all of the 24 individuals could go through two 6-7-day periods of getting only starved or only fed sea urchins (Supplementary Fig. 4  shows a schematic of the experiment). In between trials, sea star hunger levels were reset by resting in their holding tank for 6 days with a standardized maintenance diet (2 mussels every other day), followed by 5 days of starvation. 

All predation trials were recorded with GoPro HERO5 cameras mounted facing down and directly above each of the 12 aquaria. The GoPros recorded during daylight hours for the entire predation trial. At the start of each trial, the recording would start, and two sea urchins (either both fed or both starved, always from the same feeding treatment) were added to the tank. To ensure that sea urchins were not too big to eat for a given sea star, we gave the larger urchins to the larger sea stars throughout the experiment. The video documented star and urchin movements throughout each day and we used them to measure movement and several interactions between the predator and prey. We checked on the tanks every 12 hours, recorded the status of the urchins (captured or egested) and if an urchin had been captured within that time frame, we would add a new urchin from the same treatment into that tank. Thus, there was always at least one live sea urchin in the tank with each Pycnopodia, ensuring that they could feed ad libitum. At the end of each 6-7 day trial, we moved the stars back to their long-term holding tanks, and drained and scrubbed the aquaria with fresh water to remove chemical cues from the previous trial. 

We calculated predation rates as urchins captured per day, handling times as the hours elapsed between capture and egestion, and ‘rest’ times as the hours elapsed between egestion of one sea urchin and capture of another. Note that we checked the tanks only every 12 hours, so while the rates and time estimates may be somewhat imprecise, our high replication allows us to accurately estimate these values, and the imprecision should not affect our comparisons between treatments. We analyzed Pycnopodia predation rates, handling times, and rest times using a mixed effect linear model in JMP Pro 16 (SAS Institute) fitted using restricted estimated maximum likelihood (REML). We tested the effects of urchin treatment (fed vs. starved) and Pycnopodia source habitat (purple urchins present or absent) and their interaction on the number of urchins captured per day in each tank. We included the trial and the sea star as random variables to control for the non-independence of tanks tested in the same trial and of sea stars being used in successive trials. We also investigated whether sea star and urchin sizes affected predation rate, handing time, and rest time, and found no relationships. This finding does not mean that these metrics are unrelated to sea star or urchin size in nature. Rather, it shows that we successfully matched appropriate urchin sizes to sea star sizes during the trial and that our estimates are not likely biased by the sea urchins being too big for a given sea star to consume nor biased by larger sea stars eating more sea urchins to support their energetic demands. We also tested whether Pycnopodia that were naive to S. purpuratus prey would eat them at different rates, and whether Pycnopodia that have been exposed to S. purpuratus in the wild would be more likely to exert preferences for fed or starved urchins. We took advantage of the fact that some Pycnopodia were collected in areas with and without local populations of S. purpuratus.

4) Quantifying Pycnopodia and urchin movement and behavior

Data:

Summ_TrajUrchByHour Summ_TrajPycByHour Summ_TrajUrchByDay Summ_TrajPycByDay Summ_TrajPycByHour2 Traj_Urch_ByHour Traj_Pyc_ByHour

Code: CoordAnalysis_Dryad.Rmd

Methods: Predation trials were fully video-recorded during daylight hours. Urchin and sea star position over the course of each trial were tracked using DeepLabCut pose estimation software, a Python toolbox that employs deep neural networks to track animal features. We trained two networks on a subset of still video frames (approximately 300 frames each) to recognize the objects of interest: network 1 tracked the corners of the experimental tank and the center of the sea star; network 2 tracked the center, plus top, bottom, left, and right edge of each sea urchin (relative to the orientation of the video frame). We visually overlaid the urchin, seastar, and tank x, y coordinates with the raw video footage in Python to confirm the quality of the networks’ tracking. Additional frames were labeled and networks were retrained as needed to improve tracking performance. We filtered out videos with technical problems or poor lighting conditions, and used the pose estimation software to analyze a total of 298 daily videos, which included 11 stars from Trial 1, 11 stars from Trial 2, 10 stars from Trial 3, and 10 stars from Trial 4. Video data from a 10-hour daytime period (8am to 6pm) when all experimental tanks were well-lit and tracking was optimal was used for a subset of days from each trial (days 2-7 for Trial 1, days 2-6 for Trials 2-4). Timestamped coordinate data from the urchin and sea star networks were merged into a single data set to compare sea star and urchin movement patterns through time and comprised 5,843,701 total coordinates between all time points and species.

To quantify sea urchin and Pycnopodia movement, we analyzed the movement trajectories for each animal using the as.ltraj function in the adehabitatLT package in R. For each time step, we calculated the rate of movement of each animal and the distance between the sea star and each sea urchin. All data were converted from pixels to cm using a ruler in a still frame at the start of each video. We excluded day 7 from experiments since not all trials lasted 7 days, and excluded any speeds > 40cm per minute for Pycnopodia and >12cm per minute for sea urchins since these values were rare and likely due to video tracking errors. To distill the data into relevant and analyzable time frames, we first calculated the average hourly movement of and distance between each animal, then analyzed these hourly data. For sea urchins, we averaged the data for the two urchins in a given tank for each hour since the video tracking frequently switched the numbered designations between individuals (i.e. urchin 1 did not necessarily represent the same urchin for the duration of the video nor trial). We used mixed effects generalized linear models in R (lme4 package ) to test the effects of sea urchin treatment, day of the experiment, and their interaction on the hourly movement of sea urchins, including tank number as random variable to control for repeated measures, and specified a negative binomial distribution. We performed a similar model on Pycnopodia movement, but also included Pycnopodia source habitat (S. purpuratus present or absent) and its interactions with urchin treatment and day of experiment.      

5) Population model

Data: full_runs.csv

Code: functions.R, model_runs.R

Method: To assess the possible impacts of Pycnopodia predation on purple sea urchin density, we used a modeling exercise based on available California and Oregon subtidal community data.  Specifically, we ask how Pycnopodia density and known handling time can alter mean densities of purple urchins in a simple, heuristic model grounded in empirical field data. We did not use a size-structured model in this case as we focus on predation of mid-to-large individuals, which showed little variation in rate of predator-driven mortality, and thus the numeric effects of a size-structured and size-agnostic model are identical under the assumption that Pycnopodia attack rate and handling time of urchins is consistent by size. Importantly, this solution is a conservative estimate, as Pycnopodia likely are capable of consuming more small than mid- and large-sized S. purpuratus per unit time because of reduced handling time. 


The cumulative abundance of urchins across sizes can be represented by the following equation:

Equation 1



where represents urchin density, represents Pycnopodia density, represents recruitment rate of urchins (no. per year entering at <2cm), represents non-Pycnopodia related per-capita mortality rate (per year), and Pycnopodia related mortality (per prey, per year) is some function of and . Pycnopodia density .  If we assume follows a Type II functional response (i.e., consumption rate per predator, per prey, per year at high prey densities is limited by handling time) then the equation expands to: 

  Equation 2where represents the handling time in years for each Pycnopodia per S. purpuratus captured while represents the attack rate per predator, per prey, per year. The functional response asymptotes (at high urchin densities) at urchins per predator per year. While this model has no analytical solution (it is a nonlinear function of ), the expected equilibrium density of urchins is easily solved numerically. The expected density depends on recruitment rates (), non-urchin mortality () as well as Pycnopodia density and consumptive parameters ( and ). Because Pycnopodia move relatively quickly to capture new prey, and prey are often in high aggregated densities, we assume is high relative to , leading to consumption rates near until declines to very low densities. Because recruitment of S. purpuratus can vary dramatically among locations and time periods, we vary recruitment (r) of adult urchins (~2cm) to illustrate how differences in population productivity affect the control that Pycnopodia can exert on urchin population size. 

R and R studio