Chemical cues from a predatory fish (Parapercis colias) suppress feeding rates of the New Zealand sea urchin (Evechinus chloroticus)

Abstract

Changes in sea urchin behavior following detection of chemical cues from predatory fishes may influence key ecological dynamics but have rarely been experimentally quantified. Here, we measured feeding rates on a habitat-forming macroalgae by two size classes of the New Zealand sea urchin (Evechinus chloroticus) exposed to either ambient seawater or seawater carrying excretions from a predatory fish (blue cod; Parapercis colias). We created a Bayesian model that combined uncertainty in kelp growth rates and probability of urchin feeding to generate robust estimates of fish-exposure effects on multiple feeding metrics. We then compared our results to re-analyzed data characterizing behavioral responses of E. chloroticus to lobster cues (Jasus edwardsii). Larger urchins (6–8 cm test diameter) consumed ~ 40% less kelp in both predator treatments, exhibiting indistinguishable responses to blue cod and lobster despite being less susceptible to fish predation. Responses of smaller urchins (3–5 cm test diameter) to both predators were equivocal, though were more consistent with reduced feeding in the presence of lobster. Here we provide novel evidence that fish cues can suppress urchin feeding rates, even in the absence of urchin alarm cues, and discuss our findings in the context of the specificity of predator cue detection-reception pathways and possible mechanisms for risk-induced reductions in urchin feeding.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-23130-8.

Introduction

Consumption of sea urchins by predatory fishes can be an important driver of urchin population dynamics, in turn influencing ecosystem state by modulating herbivory on habitat-forming macroalgae^1–3. Predator–prey interactions among urchins and fishes are often strongly size-structured, with urchin spine canopies acting as exceptional deterrents against smaller predators lacking a sufficient gape width, dentition, or specialized feeding behavior^4–6. Increasingly, abundances of large-bodied fishes with the greatest capacity to consume urchins have declined following fisheries exploitation or environmental degradation, creating settings where urchins may overlap with predatory fish species but experience reduced risk of lethal interactions^7–9. Where urchins are less frequently preyed upon, defensive behaviors in response to the perception of risk have the potential to more strongly influence patterns of herbivory than top-down regulation of urchin density^10,11. Furthermore, incorporating quantitative estimates of both consumptive and non-consumptive effects of predators on urchins can yield improved predictions of kelp forest composition and recovery rates following disturbance^12–14. However, experimental evidence in urchins of predation-risk effects induced by the presence of fishes is limited compared to records of behavioral responses to kairomones (hereafter, risk or predator cues) excreted by invertebrate predators (e.g., decapods^15,16; sea stars^17,18; gastropods¹⁹), due in part to the comparative difficulty of manipulating fish in laboratory and field settings¹⁸.

A number of studies have reported associations between predatory fish populations and urchin behavior in the field, especially an increase in urchin sheltering behavior and crypticity when fish are abundant, often in protected areas^1,20,21. Although observational field-based datasets provide useful details regarding reef-scale patterns of urchin behavior, they generally cannot be consulted to isolate the direct response of urchins to detection of predatory fishes due to confounding interactions with risk cues produced by other predator taxa or injured conspecifics^18,22,23. Compared to observational work, only a few studies have specifically assessed predation-risk effects of fish on urchins using controlled laboratory or field-based experiments. For example, Cowen²⁴ described increases in the proportion of exposed Mesocentrotus franciscanus individuals following removal of the echinivorous California sheephead (Semicossyphus pulcher) from experimental plots. Similarly, Hagen and Mann²⁵ reported that the green sea urchin Strongylocentrotus droebachiensis formed smaller aggregations in the presence of the Atlantic wolffish Anarhichas lupus. Despite providing convincing experimental evidence for a behavioral component of interactions between predatory fishes and urchins in field conditions, in both examples the specific effect of directly-produced fish cues on urchin distribution cannot be disentangled from coinciding changes in urchin density, predation frequency, or exposure to conspecific alarm cues.

The best described antipredator behaviors of urchins that were strictly induced by exposure to non-feeding predatory fishes have been changes in patterns of urchin movement. For example, Rodriguez and Ojeda²⁶ measured a size-dependent fleeing response of the black urchin Tetrapygus niger in the presence of the Chilean sandperch Pinguipes chilensis, while Scheibling and Hamm²⁷ observed that S. droebachiensis individuals moved away from cues produced directly by cunner (Tautogolabrus adspersus). Furthermore, Hagen et al.²⁸ measured both strong freezing and fleeing reactions of S. droebachiensis to wolffish secretions, but only those containing byproducts of digested urchin material. Direct measurements of urchin responses to fish-produced cues (especially in the absence of alarm cues or predation-driven changes in urchin density) are therefore sparse, dependent on species pairing, and capture a narrow subset of potential behaviors^29,30. A plausible reaction by urchins following exposure to predatory fish would be a reduction in feeding rates, a common antipredator behavior usually attributed to increased prey vigilance or avoidance of cue-contaminated food^16–19. To our knowledge, predator-induced changes in urchin feeding rates have only been assessed in the presence of invertebrates. In some settings, slower rates of drift kelp consumption may allow detrital production to keep up with urchin demand, stalling or preventing a switch to feeding on intact fronds that precipitates kelp forest loss or inhibits regeneration^31–33. Measuring the relative extent by which risk cues produced by various predators may suppress urchin feeding rates, including fish cues, could facilitate ongoing efforts to incorporate behavioral components into models of ecosystem dynamics¹⁴ across a broader range of ecological or management scenarios than is currently possible.

In Aotearoa New Zealand, interactions between the New Zealand sea urchin (kina, Evechinus chloroticus) and various predators, including fishes, have been repeatedly demonstrated as a key driver of ecosystem dynamics in some contexts^21,34,35. Behavioral responses of E. chloroticus to predator or alarm cues can drive sheltering behavior and an avoidance of exposed grazing, promoting survival even in areas with relatively high predator biomass, though predator-specific cues that induce cryptic behavior have not been identified^36,37. The primary urchin-eating fish in the northern New Zealand region are snapper (tāmure, Pagurus auratus), though distribution of the species does not extend far into the colder waters bounding Te Waipounamu the South Island^38,39. Where snapper are not abundant, among the most prominent fish species known to consume E. chloroticus is the blue cod (rāwaru, Parapercis colias). Although sometimes described as a primary predator of urchins in parts of southern New Zealand^35,40, surveys of blue cod stomach contents suggest that urchins constitute a relatively small proportion of overall diet^41–43. Blue cod are targeted by an expansive and valuable fishery that can result in local depletion^43,44, in particular of large individuals that are more capable of urchin predation (Supplementary Fig. S1). Besides the occasional documentation of consumption, interactions between blue cod and urchins are generally uncharacterized and may well be comprised in large part by predation-risk effects on urchin behavior – especially where large fish have been removed.

In the present study, we performed a laboratory experiment designed to quantify the effects of predator cues produced by blue cod on the feeding rates of two size classes of E. chloroticus: cryptic subadults (30–50 mm test diameter, TD) and mature adults that are generally exposed, but small enough to remain vulnerable to predation (60–80 mm TD). Using a Bayesian statistical approach to facilitate cross-study comparison, we then compared behavioral responses to blue cod to re-analyzed predation-risk effects originally reported in Curtis and Wing⁴⁵, which described interactions between E. chloroticus and an invertebrate predator, the red rock lobster (kōura, Jasus edwardsii⁴⁵; hereafter, “the lobster experiment”) .

Results

Blue cod experiment

Urchins consumed kelp in 82% of assays during the blue cod experiment (131/160), demonstrating a general willingness to feed in the testing environment. Urchins that did not consume kelp were evenly represented in both predator treatments (predator: 14, ambient: 15), reflected in strongly overlapping posterior distributions of p_t within size classes (the likelihood of feeding in each predator treatment; Table 1). For calculations of feeding rates among urchins that did consume kelp, our model generated reasonable estimates of proportional kelp growth with 99% of imputed values falling within the growth range observed in controls (mean ± SD: + 2.7% ± 1.8%; range: − 3.6% to + 10.2%; Fig. 1). Among consumers, the posterior distributions for control-adjusted feeding rates by ambient and fish-exposed urchins were virtually identical for small individuals but differentiated in the large size class, featuring 90% credible intervals (CIs) that excluded median estimates from the alternate predator treatment (Table 1). Comparing median estimates of the combined consumption metric (probability of eating x feeding rate), large urchins consumed ~ 1% less kelp relative to body mass (~ 40% proportional reduction) when exposed to blue cod cues, with 90% CIs for predator effects that excluded “no-effect” thresholds (Tables 1, 2, Fig. 2). For an average urchin in the large size class, these modeled predator effects correspond to a median reduction in per-capita kelp consumption of ~ 0.5 g/day in the presence of blue cod (90% CI 0.1–1.05 g/day; Fig. 3). Conversely, the same comparisons for small urchins were equivocal, with 90% CIs encompassing “no-effect” thresholds and posterior distributions indicating meaningful probabilities of either suppressed (34% of posteriors) or heightened feeding (66% of posteriors) in predator-exposed individuals (Table 2, Fig. 2).

Table 1.

Median values and 90% credible intervals (CIs) for posterior estimates describing urchin feeding behavior in both blue cod and lobster⁴⁵ experiments. Feeding is presented using two metrics and their product: probability of feeding p and mean mass-specific feeding rate m, expressed as % urchin mass.

Blue cod				Lobster
Size class	Predator	Median	90% CI	Size class	Predator	Median	90% CI
Probability of feeding (p)				Probability of feeding (p)
Small	Ambient	0.75	0.64, 0.86	Small	Ambient	1	0.97, 1
Small	Cod	0.83	0.73, 0.92	Small	Lobster	0.80	0.63, 0.96
Large	Ambient	0.88	0.80, 0.96	Large	Ambient	1	0.98, 1
Large	Cod	0.83	0.73, 0.92	Large	Lobster	0.89	0.79, 0.98
Consumer feeding rate (m)				Consumer feeding rate (m)
Small	Ambient	2.91	1.83, 4.19	Small	Ambient	6.71	4.75, 8.71
Small	Cod	3.01	1.93, 4.27	Small	Lobster	6.05	4.16, 7.91
Large	Ambient	2.37	1.68, 3.20	Large	Ambient	5.32	4.11, 6.51
Large	Cod	1.44	1.00, 1.95	Large	Lobster	3.99	3.05, 4.93
Combined consumption (p*m)				Combined consumption (p*m)
Small	Ambient	2.18	1.3, 3.18	Small	Ambient	6.63	4.65, 8.65
Small	Cod	2.48	1.54, 3.56	Small	Lobster	4.72	2.91, 6.53
Large	Ambient	2.07	1.44, 2.84	Large	Ambient	5.28	4.07, 6.48
Large	Cod	1.19	0.79, 1.63	Large	Lobster	3.52	2.59, 4.44

Fig. 1.

Distributions of observed (Tukey box plots; n = 109 total observations, from 11 to 17 per trial) and imputed (mirrored density plots) proportional changes in unconsumed kelp mass (kelp_out/kelp_in) for the 8 trials in the blue cod experiment. Imputed distributions are derived from 600,000 estimates of kelp growth used to calculate consumption for each consumer (n = 131) and show consistency with observed values over the course of the experiment.

Table 2.

Median values and 90% credible intervals (CIs) for effect sizes describing differences in urchin feeding rates (“combined consumption” in Table 1; % urchin mass) between ambient treatments and in the presence of a potential predator (blue cod or lobster⁴⁵). Effect sizes are presented in terms of raw and proportional differences between treatments, as well as percent difference of feeding rates in predator-exposed urchins relative to ambient conditions.

Effect	Size class	Blue cod		Lobster
		Median	90% CI	Median	90% CI
Predator—Ambient (Raw)	Small	0.30	− 1.01, 1.62	− 1.89	− 4.22, 0.47
	Large	− 0.88	− 1.68, − 0.15	− 1.75	− 3.22, − 0.28
Predator/Ambient (Proportional)	Small	1.14	0.58, 1.79	0.71	0.42, 1.04
	Large	0.57	0.34, 0.83	0.67	0.45, 0.91
% Difference from ambient	Small	+ 14%	− 42%, + 79%	− 29%	− 58%, + 4%
	Large	− 43%	− 66%, − 17%	− 33%	− 55%, − 9%

Fig. 2.

Density distributions of 600,000 posterior estimates for combined mass-specific consumption metrics (m_t x p_t) in (a) the blue cod experiment and (b) the lobster experiment, expressed as % urchin mass. Lower overlaps in density distributions indicate stronger effects of predator exposure on feeding rates. Treatment effects are visualized for (c) the blue cod experiment and (d) the lobster experiment as posterior density distributions of proportional predator effects (combined consumption posteriors from predator/ambient treatments), including red bars indicating 90% credible interval bounds and a dashed line for a value of 1, which signifies equivalence in feeding behavior between treatments. Predator effects were determined to be substantially negative for large urchins in both experiments, while small urchins exhibited less clear responses to predator cues (Table 1).

Fig. 3.

Median posterior estimates (points) of mass-specific (top: a, c) and per-capita (bottom: b, d) feeding rates in each urchin size class, including non-consumers (set to 0), from the blue cod (a, b) and lobster (c, d) experiments. Distributions are summarized using per-treatment medians and 90% credible intervals (CIs) of the combined consumption metric (m_t x p_t). Relative overlaps of 90% CIs demonstrate reduced feeding rates following predator exposure in large urchins, but no clear or consistent treatment effect in small urchins (see Table 1).

Using any metric besides feeding rates, urchin activity did not systematically differ in response to predator exposure in an obvious or strong manner. There was a weak trend of reduced change in daily position (relative to body size) in small urchins exposed to blue cod, but little to no statistical evidence for any association with predator treatment in either size class (Small: − 3.13 body lengths, p = 0.13; Large: + 1.06 body lengths, p = 0.88; Supplementary Table S3). The vast majority of urchins interacted with kelp at least once during the experiment and ~ 75% of urchins were always observed interacting with kelp (Fig. 4), with almost no variation between predator treatments. Urchins were only recorded under the provided shelter in 15% of observations, with comparable frequencies of shelter use in ambient and predator treatments (Fig. 4). Urchins were most commonly observed at the surface near the inflow or the floor below the outflow, with similar intermediate rates of occurrence elsewhere in the feeding arena (though generally avoiding exposed positions on the tank walls). Although there were some subtle differences in the proportion of urchins occupying specific locations in the feeding arena between predator treatments, these are derived from a relatively small number of observations and do not suggest a consistent response to blue cod predator cues (Fig. 4).

Fig. 4.

Observed number of interactions in each urchin size class with (a) kelp and (b) shelter during daily observation in all trials of the blue cod experiment. An x-axis value of 0 indicates no observed interaction of an individual urchin with kelp or shelter, while an x-axis value of 3 indicates an individual urchin was always observed in contact with kelp or under the provided shelter. The maximum y-axis value is 40 (number of urchins/size class). (c) Total number of observations of urchins in each size class at each tank position in trials 2–8 of the blue cod experiment. Tank positions were characterized in three vertical positions (bottom, middle, top) and three horizontal positions (back/outflow, middle, front/inflow). The maximum possible number of observations is 105 (35 urchins/size class × 3 observations). Despite some small variation, these data indicate similar patterns in ability to locate food, sheltering rates, and within-tank positioning by urchins in either ambient or blue cod treatment groups.

Lobster experiment re-analysis

Re-analysis of data from the lobster experiment using a Bayesian framework reaffirmed the primary conclusions reported in Curtis and Wing⁴⁵, but also provided output that facilitated direct comparison to the effects of blue cod on urchin feeding rates. Within the context of the lobster experiment, urchins consumed kelp in ~ 90% of trials (55/61) with all of the non-consumers in the predator treatment. As opposed to the blue cod experiment, posterior distributions of p_t for the lobster experiment featured non-overlapping 90% CIs within each size class, indicating a lower likelihood of feeding in the presence of lobster cues (Table 1). Using the combined consumption metric, lobster presence resulted in a modest but detectable decrease in mass-specific feeding rates of large urchins (median difference: 1.75% urchin mass or -34% proportional feeding; Tables 1, 2), corresponding to an average reduction in per-capita feeding of 0.75 g kelp/day (90% CI 0.12–1.43 g kelp/day; Fig. 3). In small urchins, posterior distributions for the combined consumption metric were subtly distinguished between lobster and ambient treatments relative to the large size class (Table 1, Fig. 2). Although 90% CIs of treatment effects for small urchins included “no-effect” thresholds, 91% of posterior estimates were consistent with suppressed feeding in the predator treatment relative to ambient conditions (median difference: − 29% proportional feeding; Table 2, Fig. 2). Collectively, these results constitute weak statistical evidence for an effect of lobster cues on feeding rates of small urchins that may have been obscured by limited sample sizes (Table 2).

Comparing estimates of urchin feeding behavior between the blue cod and lobster experiments, all posterior distributions of p_t featured strongly overlapping 90% CIs except the ambient treatments in the lobster experiment, where all urchins consumed kelp. The “raw” effects of lobster presence on feeding rates in large urchins were approximately two-fold larger than observed for blue cod (Table 2, Fig. 3). However, feeding rates in the lobster experiment were higher overall than during the blue cod experiment, even among predator-exposed urchins (Table 1). Providing a more directly comparable metric, the 90% CIs for proportional effects of predator treatment on feeding rates in both urchin size classes were almost completely overlapping between experiments (Table 2). Therefore, our results do not provide strong evidence for different reactions of tested urchins to blue cod and lobster, given uncertainty.

Discussion

Here we provide the first direct evidence that fish-produced predator cues can suppress the feeding rates of a temperate sea urchin in the absence of injured or digested conspecifics. Defensive behaviors of fish-exposed E. chloroticus did not inversely scale with presumed size-based vulnerability to predation^9,21, rather were only clearly detected at test diameters exceeding 6 cm. Behavioral responses of large urchins to blue cod were moderate, and proportionally very similar to those measured in the presence of red rock lobster (Table 2, Fig. 2). Our estimate for mature E. chloroticus of a 35–40% proportional reduction in kelp consumption following exposure to predators is consistent with effect sizes repeatedly measured in another short-spined urchin, Strongylocentrotus purpuratus^16,17,46, and close to the median value of 45% reported in a meta-analysis of indirect predator effects on feeding rates of prey¹⁰. However, decreases in urchin feeding rates in the presence of predators can be much stronger than observed in our experiments, in some cases exceeding 80%^15,18,19, although such dramatic reactions may primarily occur when shelter is unavailable in feeding trials⁴⁶.

Our effect sizes are likely close to the maximum expected response of E. chloroticus (in terms of feeding rates) to excretions from blue cod or lobsters in the absence of alarm cues, given that we tested urchins in circumstances shown to enhance antipredator behaviors: using well-fed individuals, a relatively short assay, and high concentrations of continuously-provided predator cues^17,28,47. However, the kelp offered to urchins (Macrocystis pyrifera) is readily consumed by E. chloroticus relative to other algae⁴⁸ and thus might have constituted a high-quality reward that enticed urchins to feed despite the detection of risk, dampening reactions to predator presence^11,47. Whether dietary preferences modify the strength of antipredator behaviors (or vice versa) has not been examined in urchins but is an intriguing topic for further research, especially for E. chloroticus given that M. pyrifera is patchily distributed throughout coastal New Zealand and expressions of risk-exposed feeding may differ on more common forage (e.g., Ecklonia radiata).

Exposure to either lobster or blue cod suppressed feeding rates of large urchins to a similar degree (Table 2, Figs. 2, 3). The only difference in the response of large urchins to lobster cues was a distinguished effect of predator presence on the probability of kelp consumption, which did not vary between ambient and blue cod treatments (Table 1). Regardless, the estimated difference in the likelihood of feeding by lobster-exposed large urchins was qualitatively minor (median: -11%) and driven by only three observations. Our results therefore do not provide compelling evidence that lobster excretions cause mature E. chloroticus to entirely avoid kelp consumption to a greater extent than cues from blue cod, instead suggesting that reductions in feeding rates following detection of both predators may be primarily dictated by responses of active consumers. The broad similarity of antipredator behaviors exhibited by large urchins in both experiments is perhaps surprising as lobsters can capably prey on urchins > 6 cm TD, while consumption of urchins in this size class by the largest blue cod is plausible but probably infrequent due to gape limitation^9,49 (Supplementary Fig. S1). However, animal responses to perceived risk do not always scale in proportion to the likelihood of predation, and can be strong even in the presence of species that represent little or no threat²⁹.

Conversely, our results provide weak evidence that detection of lobsters may more strongly depress feeding rates of small urchins than cues from blue cod. Accounting for uncertainty, we cannot rule out the possibility that small urchins modestly reduced consumption of kelp in the presence of blue cod (Lower CI₉₀ − 42%, Upper CI₉₀ + 79%; Table 2). However, our median estimate was essentially no behavioral response to fish exposure by small urchins, with only 34% of estimated effect sizes consistent with reduced feeding rates in the blue cod treatment (vs. 91% in the presence of lobsters; Fig. 2). Although we cautiously interpret our results to suggest a stronger response of small urchins to lobster-produced cues than excretions from blue cod, more conclusive assessment of relative antipredator behaviors would require additional experimentation—ideally using greater sample sizes over a longer trial period to allow for more extensive feeding and reduce uncertainty. Still, even our median estimate of the effect of lobster exposure on small urchins would only correspond to an average reduction in per-capita feeding rates of 0.2 g kelp/day, or ~ 20% of the response exhibited by large urchins (Fig. 3). Therefore, our experiments do not indicate a strong possibility that suppression of per-capita feeding rates in small urchins induced by either lobster or blue cod cues would substantially contribute to the balance of drift kelp consumption in natural settings, except perhaps in populations with very high densities of small individuals and few adults.

Weaker responses of small individuals to predator cues can stem from a lower energy budget and need for more consistent feeding, although other urchin species have shown an energetic capacity to downscale kelp consumption at similar sizes (3–5 cm TD) in comparable experiments^15,16,47. Learning via association of overlapping predator and alarm cues can also explain a lack of predation-risk effects in young or naive animals^30,50,51. Given that we saw an inversion of behavioral response strength relative to size-based vulnerability to blue cod, it is possible that large urchins were not reacting specifically to risk of fish predation but had instead come to perceive blue cod cues as a likely indicator of threat from other predators on healthy reef communities, such as lobsters. Due to their predominantly cryptic lifestyle, small urchins may conversely not have been sufficiently exposed to overlapping predator and alarm cues to develop a specific defensive response to blue cod, especially compared to lobster which occupy more similar habitats in reef crevices and interstices.

Little is known about the compounds that elicit antipredator behaviors in urchins, limiting speculation about whether large urchins in our experiment detected general cue(s) common to two fish-fed predators or could identify distinct threats and yet responded equitably. A differentiated response to the presence of lobsters and blue cod would require excretion of at least one compound that was specific to each predator, as well as urchin receptors that were capable of separately detecting each cue. Instead, both predators may have excreted similar waste compounds or metabolites following digestion of the same diet (fish tissue), communicating the presence of a high-order carnivore to urchins but not necessarily predator identity^30,52. Other urchins have demonstrated a capacity to distinguish among cues from multiple species, or show behavioral responses that defend against foraging techniques of a particular predator, suggesting at least some potential for specificity in cue production-reception pathways^18,25,53. Our estimation of a weakly defined difference in the effects of predator exposure on small E. chloroticus may support production of separate cues by blue cod and lobster, although this inference would be stronger if behaviors were surveyed simultaneously rather than sequentially (and if feeding responses by small urchins to lobster were more clearly characterized). Further experimental work is therefore necessary to clarify the risk-signaling compounds that induced reactions of E. chloroticus in our trials, particularly assays using excretions of coastal piscivores that do not consume urchins to quantify responses to digested fish tissue in the absence of indicators of direct threat.

Lobsters and blue cod do not occur in isolation in the wild, and urchins in natural conditions will be presented with combinations of direct cues from both these species and other fish and invertebrate predators (in addition to alarm cues where urchins are regularly consumed). Byrnes et al.¹⁸ demonstrated that reductions in feeding rates by a community of kelp forest herbivores (including urchins) scaled with diversity of predator cues, though the urchin response was entirely attributed to a predatory sea star. Future research could similarly assess reactions of E. chloroticus to isolated and overlapping excretions from various combinations of urchin predators and other common reef species, including interactions with alarm cues or compounds produced during digestion of urchin tissue^16,22,28. Together with quantifications of size-specific vulnerability to predation, parameterized estimates of E. chloroticus behavior and mortality in the presence of differing risk and chemical landscapes may then be included in models of local kelp forest dynamics, advancing prediction of the effects of overfishing or protection of specific predator taxa for urchin population characteristics and herbivory^14,54.

Previous assessments of mechanisms that underpin defensive reductions in urchin feeding rates have focused on patterns of movement or activity^15,17, although other possibilities include predator-induced changes in encounter rates with food (especially via induction of crypsis^21,36), shielding behavior, or heightened vigilance that shifts focus away from active feeding⁵⁵. Unlike McKay and Heck¹⁵, who observed movement of S. droebachiensis away from a predatory crab in conjunction with reduced feeding rates, we found no consistent shift in positioning of E. chloroticus in the presence of blue cod (Fig. 4), though this discrepancy is likely due to our delivery of predator cues via aquarium line rather than more direct exposure through a perforated barrier. Similar to our finding that E. chloroticus activity did not vary with predator treatment, Whippo et al.¹⁷ did not detect an effect of exposure to a predatory sea star on the frequency of movement or distance traveled in S. purpuratus that could explain a reduction in feeding rates. Critically, our daily observations offered only coarse information regarding urchin activity, and did not encompass nighttime periods when many urchins (including E. chloroticus) are most active and undertake the broadest foraging movements^22,37,56. However regardless of any potential effects of predator exposure on overnight behavior, urchins in both blue cod and ambient treatments were almost always observed outside of the provided shelter and attached to kelp each morning (Fig. 4). Given that detection of risk did not evidently induce crypsis or reduce the propensity of urchins to acquire food, a deflection of attention towards vigilance and away from active feeding stands as the most likely mechanism for observed feeding reductions in the context of our experiments, rather than freezing, fleeing, or avoidance of predator cues.

Escape options are constrained in laboratory-based feeding experiments and our relatively small enclosures likely increased encounter probability with kelp, thereby promoting interaction with food rather than predator avoidance especially over a 62 h period⁵⁷. When both kelp and direct predator cues are present, some urchins will still associate with food sources in lieu of escape^27,58, although the balances of behaviors are influenced by starvation state, predator-urchin pair, and the duration of cue exposure^17,57,59. Specifically in E. chloroticus, alarm cues can induce fleeing at small spatial scales and cryptic behavior regardless of food availability³⁶. In a setting where urchins were allowed to move more freely than during our feeding trials, it therefore remains plausible that detection of predator excretions could similarly affect the scale and success of exposed foraging excursions by E. chloroticus. Cumulative effects of urchin antipredator behaviors on grazing intensity will ultimately depend on the extent to which predator avoidance reduces encounter rates with kelp, as well as how strongly vigilance suppresses feeding rates following kelp acquisition. Further measurements of movement, habitat association, and urchin feeding behavior in the presence of predator-produced, alarm, and combined chemical cues (ideally in a natural setting or larger mesocosm) will help to clarify the dominant response of urchins following exposure to various risk cues, as well as the primary mechanisms for risk-mediated changes in herbivory.

In some field settings, urchin behavior can strongly influence the abundance of habitat-forming macroalgae^31,33 and has been shown to vary based on chemical indicators of predation threat in natural conditions^19,53,59. Even though responses of E. chloroticus to direct cues from predatory fishes or invertebrates would almost certainly be weaker than our estimated effects if tested in field-relevant concentrations, exposure durations, and the presence of other urchins, even subtle changes in animal behavior can have profound effects on consumer-resource dynamics¹⁰. Our experimental design most directly simulates a low-flow environment where urchin density is low, kelp and habitat are available, and predator density is high but consumption is uncommon, perhaps due to a generalized predator diet or mismatches in predator–prey size structures. Including the present research, recent studies in E. chloroticus have now documented size-specific changes in habitat association and feeding rates following exposure to excretions from a fish-fed invertebrate and fish predator, as well as crushed conspecifics^36,37,45. Additional quantification of urchin behavior (including feeding rates) in the presence of interacting risk cues, across a representative range of urchin densities and sizes, and on reefs with various hydrodynamic regimes will help to identify the settings and ecological contexts where predation-risk effects may most strongly influence persistence of kelp forest communities and the biodiversity and productivity that they support.

Materials and methods

Experiment details

To assess behavioral changes of E. chloroticus in the presence of blue cod predator cues, we conducted a series of feeding assays from December 2023-February 2024 at the Portobello Marine Laboratory near Ōtepoti Dunedin, New Zealand. Blue cod used in the experiment were collected using baited traps near Te Awa Koeo Brinn’s Point, 12 km north of the Otago Harbour entrance. Fish were held in two ~ 850 L tanks at an average stocking density of 8 adults (26–34 cm total length, TL) per tank and were fed southern-blue whiting (Micromesistius australis) ad libitum. Holding times for individual fish prior to inclusion in the experiment varied, but always exceeded two weeks to ensure consistent feeding and acclimation to captivity.

Sea urchins used in the experiment were collected from Patea Doubtful Sound (Fiordland) in December 2022 in association with the lobster experiment⁴⁵. During the pre-trial period, ~ 130 urchins (80 of which were used in the blue cod experiment) were held in one of three 150 L tanks supplied by a semi-recirculating seawater system that was isolated from other aquaria. Urchins were fed ad libitum with Macrocystis pyrifera throughout the holding period. Holding tanks were supplied with ambient seawater that was mechanically filtered to 200 µm. Over the course of the experiment, ambient water temperature varied from 14.3–18.8 oC (mean ± SD: 16.5 ± 0.85 oC).

Feeding assays were set up as a split-plot experiment, with each set of measurements (trial) replicated eight times over the study period. The number of trials was assigned before the experiment based on the results of a power analysis parameterized using data from Curtis and Wing⁴⁵ (Supplementary Table S1). Specifically, an experiment comprised of eight trials was estimated to provide ~ 90% power to detect an effect of predator exposure on feeding rates of large (60–80 mm TD) urchins if the response was a similar magnitude as observed in the presence of a lobster (Supplementary Table S2). Conversely, only a strong response (> 80% reduction in feeding rates) of smaller urchins (30–50 mm TD) would be detectable (via parametric mixed effects models) using any practical number of trials and the available laboratory infrastructure, due to low and variable per-capita feeding rates within the size class (Supplementary Tables S1, S2).

Split-plot designs accommodate manipulation of factors at different scales (i.e., predator exposure and urchin size class), yielding two levels of replication (whole-plot and subplot replicates⁶⁰). For each trial, we supplied ten 65 L header tanks (55 × 45x26 cm) with flow-through seawater at 2 L/min (header tank residence time: ~ 32 min), which was further oxygenated via air bubbled through an air stone to offset blue cod respiration. Outflow from the header tank was split through a plastic manifold and gravity fed at 6.5 mL/s into groups of three receptacles: two 17 L feeding arenas (36  × 24 ×  20 cm; ~ 43 min residence time) and a 5 L control tank (25 × 18x10 cm; ~ 13 min residence time). For both header tanks and receptacles, flow rates were regularly monitored and maintained throughout the experiment within ± 10% of target values. Ten groupings of header tanks and receptacles (hereafter clusters) were arranged in a row along a laboratory bench with a common seawater source split into a single inflow pipe per header (Fig. 5).

Fig. 5.

(a) Diagram of layout for one trial in the blue cod experiment, with red tanks indicating the “predator” treatment and blue tanks indicating the “ambient” treatment. (b) Photograph of one cluster, with two 17L feeding arenas and one 5L control tank (Credit: Joseph Curtis).

To assess the behavioral responses of E. chloroticus to the presence of blue cod, we measured urchin feeding rates on M. pyrifera across a range of urchin sizes exposed to either ambient seawater (“ambient” treatment) or seawater flowing through a header tank containing a single blue cod (“predator” treatment). We collected kelp within 48 h of the beginning of each trial from canopy fronds near the laboratory, which we stored in flow-through seawater before processing. To prepare kelp for feeding trials, we removed individual mature blades from attached fronds, scraped off any epiphytic bryozoans, and pre-dried the whole blade using a salad spinner for 30 s. We then sectioned blades into ~ 100 cm² pieces with a minimum width of 6 cm. Following thorough blot drying we weighed individual kelp pieces, which were only included in trials if they weighed 3–6 g. We then randomly aggregated pre-weighed kelp pieces into batches of three, which we accepted as suitable for inclusion in feeding trials if the combined mass totaled 12–15 g. After processing, we placed kelp into each feeding (or control) arena in a random order, with one corner of all three pieces weighted underneath a single 5 × 5 cm stone tile to ensure kelp was accessible to urchins.

Before each trial, we randomly selected 10 urchins from each of two size classes: “small” (30–50 mm TD) and “large” (60–81 mm TD). Urchins from both size classes were then placed in a separate tank for a 96 h starvation period—a duration shown to encourage kelp consumption without stifling responses to predator cues⁴⁷. Given the ad libitum provisioning of kelp for several months pre-trial, urchins would still be considered “fed” compared to individuals characterized as “starved” in other similar experiments^17,47. The ranges for urchin size classes were designated on either side of a 50 mm TD threshold associated with the onset of maturity in E. chloroticus, enhancement in per-capita feeding rates, and adoption of an exposed lifestyle^45,61. Upper and lower bounds were based on the size distribution of available urchins that would allow for suitable, balanced replication within each size class (see Supplementary Material). Following starvation, we randomly assigned individual urchins from each size class to one of the two feeding arenas within each cluster. Using the resulting balanced design, each header flowed into three total receptacles containing either one small urchin, one large urchin, or no urchin (control). Control tanks were used to measure and account for changes in kelp mass in the absence of feeding (generally positive and thus referred to as growth). Feeding rates of the same eighty urchins were measured twice: once in trials 1–4 and once in trials 5–8. Within these two blocks of trials, urchins were selected randomly, with the exception that urchins used in trial 4 were not included in trial 5 to allow a minimum of one week of recovery and ad libitum feeding before re-testing. We measured the diameter (TD) and wet weight (g) of each urchin before placement into feeding arenas underneath an artificial shelter (semi-circular PVC pipe, ~ 12 cm in all dimensions).

Immediately after the addition of urchins to feeding arenas, we transferred a single adult blue cod (26–35 cm TL; 270–670 g) into a random selection of half (5/10) of the header tanks, with the only constraint being that no half of the array (clusters 1–5 or 6–10) contained more than three clusters assigned to the same treatment within a trial. Blue cod were fed approximately eight hours before transfer into header tanks. To reduce handling stress, we avoided reusing any single blue cod in back-to-back trials by alternating between the two fish holding tanks when selecting individuals for experimental use. Each trial lasted ~ 62.5 h (61–63.5 h), with overhead artificial lighting provided on a 9 h:15 h day:night cycle (8AM to 5PM). We recorded urchin position and activity each morning for the trial duration, roughly at 14, 38, and 62 h following initiation, using metrics modified from McKay and Heck¹⁵. Specifically, we recorded position as the nearest of 25 locations that identified all combinations of vertical position (floor, middle, or surface), lateral position (front/inflow, middle, or back/drain), and horizontal position (left or right; Supplementary Material). We also noted use of shelter (covered or exposed), and interaction with kelp (touching or not).

At the conclusion of each trial, we transferred urchins and blue cod back into holding tanks, then removed, weighed (dry mass), and photographed remaining kelp from feeding arenas in a randomized order. We used a benchtop balance with a precision of 0.01 g for all measurements of urchin and kelp mass. To clean the array between trials we thoroughly rinsed all tanks and components (PVC shelter, header air stone, and stone tile) with fresh water, followed by scrubbing using brushes soaked in a 1% bleach solution and a final freshwater rinse. We soaked all feeding arenas, supply lines, and tank components in fresh water for at least 48 h before resetting the array for the following trial.

Methods applied throughout the lobster experiment were very similar, and are described in full in Curtis and Wing⁴⁵. A few key differences that are relevant to comparative analysis or interpretation of the results of both experiments are noted in relevant sections. For both blue cod and lobster experiments⁴⁵, all protocols for collection, holding, handling, treatment, and eventual release of P. colias and J. edwardsii individuals were reviewed and approved by the University of Otago Animal Ethics Committee (Protocol IARMS-AUP-22–105). We confirm that all experiments described in the present report were ethically performed in accordance with the relevant guidelines and regulations. All research involving live animals followed ARRIVE⁶² guidelines and are reported accordingly.

Data analysis

To assess the effect of blue cod predator cues on the feeding rates of E. chloroticus, we developed a Bayesian model that incorporated uncertainty in autogenic growth of M. pyrifera into consumption calculations, then provided posterior estimates of mean consumption rates for each combination of size class and predator treatment. While balanced split-plot designs can usually be analyzed using Bayesian or frequentist methods with similar results⁶⁰, we favored a Bayesian approach over frequentist estimates of effect sizes because it facilitated imputation of variability in kelp growth rates into consumption calculations. Accurate treatment of autogenic growth was important as growth had the potential to be similar to per-capita feeding rates of sampled urchins over the trial period, especially in small individuals, yet could not be directly observed in any instance where urchins consumed kelp. Robust estimates of feeding activity therefore required thoughtful imputation of autogenic growth rates across a distribution of plausible values, which is achievable using Bayesian hierarchical models. For our approach, we divided the data analysis into two distinct modules⁶³: a ‘growth’ module used to assess autogenic changes in kelp mass and a ‘consumption’ module that assessed urchin feeding rates, accounting for unobserved kelp growth using output from the growth module. The application of a Bayesian model to analyze these data also allowed more direct comparisons between the results of the current experiment and (re-analyzed) results from the lobster experiment through comparison of equivalent outputs from each model. We performed all analyses in R v.4.2⁶⁴ linked to JAGS⁶⁵.

Before analysis, we carefully reviewed photographs of all post-trial kelp to look for evidence of consumption by urchins. Instances where kelp remained wholly intact were labelled “unconsumed”, whereas instances featuring unambiguous urchin bite marks were labeled “consumed”. Unconsumed batches of kelp provide information about urchin feeding rates (i.e., that those individuals did not feed), but also function as additional opportunities to measure autogenic kelp growth. We therefore grouped all observations of unconsumed kelp from control tanks and feeding arenas for input into the growth module.

We modeled observations of proportional kelp growth (G = kelp_out/kelp_in) as random samples from a normal distribution using:

1

where μ_g is the mean growth at the conclusion of each trial, ε_w is a temporal random effect of trial number (w), and n_u is the number of observations of unconsumed kelp. A weakly-informative prior distribution was provided for the mean growth parameter μ_g in the form of N(1.02,0.04), with parameters chosen based on pilot measurements of M. pyrifera growth rates in the same laboratory setting. Trial effect ε_w was modeled as normally distributed:

2

Priors for standard deviations (σ_g, σ_w) were converted from scaled gamma distributions defined for respective precision terms (τ_g, τ_w) using . This results in the following distribution for standard deviation terms:

3

where represents a t-distribution limited to positive values with three degrees of freedom and s is a scalar⁶⁶. A scalar of 0.05 was provided as a prior for σ_g and σ_w, yielding a relatively vague prior given the observed amounts of growth.

We fitted the growth model using four Markov chains, with the first 5000 steps discarded during the model adaptation phase. The following 1250 posterior draws of μ_g from each chain (5000 total) were retained as inputs for the consumption module. Before accepting the growth model output, we assessed model convergence for parameter estimates using the Gelman-Rubin statistic and a threshold of R_c < 1.01⁶⁷. We selected 5000 iterations of growth parameters based on the observation that it well exceeded the threshold required for consistent convergence of all parameters (~ 1250 iterations) while limiting the computational workload required to iteratively execute the consumption module⁶³. During model development we determined that growth rates simulated by a model parameterized with the most extreme posterior estimates of all three growth parameters (μ_g, σ_g, ε_w) yielded unrealistic values far outside of the range of observed data or biological plausibility over 62 h. We therefore used the median posterior estimate of ε_w and σ_g² to impute growth estimates in the final consumption model, deriving variation in simulated growth data from the distribution of 5000 posterior estimates of μ_g.

For the consumption module, measurements were split into two components: the probability of feeding and the mass-specific feeding rate (g_kelp/g_urchin) of urchins that were positively identified to have eaten any amount of kelp (“consumers”). We modeled the probability of feeding using a Bernoulli distribution:

4

where u indicates each individual feeding assay (up to n = 160) and t indicates the four treatments (combinations of size class and predator exposure). We formulated a vague prior for p_t as:

5

For the consumption module, we first imputed growth (except where consumption was observed to be 0) using:

6

where G_c is the imputed proportional growth of kelp for each consumer (c), i is the iteration of μ_g generated in the growth module, is the median posterior estimate of the random effect of trial number on growth, and σ_g is the median posterior estimate of standard deviation in autogenic growth rates. Using imputed growth estimates, we then modeled consumption as:

7

8

Here, F is the modeled mass-specific feeding rate for each consumer (c), μ_t is mean mass-specific consumption for each treatment t, and ε_r and ε_h are random error terms associated with trial number and header tank, respectively. Note that ε_r is distinct from ε_w, as we chose to separately model the effect of trial number (mostly a proxy for temperature) on consumption (i.e., urchin metabolism and behavior) and autogenic growth. Additionally, ε_r is implicitly nested within balanced blocks of trials associated with the repeated inclusion of urchins in feeding trials (Trials 1–4: first assay, trials 5–8: second assay)_, and would therefore account for any variation in feeding rates driven by re-testing of individuals. We modeled F as a log-normal distribution based on observations of crude approximations of consumption (kelp_in – kelp_out) within each treatment, which showed clear right skewness. Estimates of F were related to the difference between observed initial kelp mass (i) and final kelp mass (o) for each consumer (c), accounting for imputed autogenic growth (G) and urchin mass (M). A separate standard deviation term for the log-normal distribution was modeled for each size class (s) to allow for potential heterogeneous variation in mass-specific feeding rates. Semi-informative priors were provided for μ_t based on feeding rates of E. chloroticus on M. pyrifera measured in similar size classes⁴⁸ as N(− 3,1), which corresponds to a natural-scale mean of 0.05 g_kelp/g_urchin (± 1 SD: 0.018,0.135). Vague priors were provided for random effect terms (N(0,1)), and σ_s (via τ_s) following a t-distribution with 3 degrees of freedom and a scalar of 0.5.

Finally, treatment effects were directly encoded in the model as follows:

9

10

11

12

where m indicates μ_t transformed to the natural scale (Eq. 13), the first subscript indicates predator treatment (A: ambient, P: predator) and the second subscript indicates urchin size class (S: small, L: large). In plain terms, describes the effect of predator exposure on small urchins, describes the effect of size class on kelp consumption, and describes the effect of predator exposure on large urchins.

We ran 5,000 iterations of the consumption module, parameterized using each of the posterior estimates of μ_g retained from the growth module. To ease computational workload, we divided execution of consumption model iterations into batches of 1000, each performed across 5 computer processors running separate instances of R. For each processor, the first model iteration (using the 1st, …, 4001st posterior estimate of μ_g) was performed across three Markov chains with an adaptation phase of 5000 steps. For subsequent iterations (e.g., using μ_g posterior estimates 2–1000 for the first processor), the final posterior estimates of all output parameters from the preceding iteration were used as initial values. Therefore, a shorter adaptation period was required for subsequent iterations (2000 steps) because starting values were likely closer to the posterior distribution than at the start of the first iteration (which was initialized using a random draw from the prior distribution for each parameter). We retained 40 draws from each consumption model iteration for a total of 600,000 posterior estimates (40 estimates × 5000 iterations/chain × 3 chains). As for the growth module, we assessed convergence for each consumption model output using the Gelman-Rubin statistic and a threshold of R_c < 1.01⁶⁷.

From posterior distributions generated by the model, we estimated the effect of treatment on the likelihood of consumption, mean feeding rate among consumers, and the product of both metrics (“combined consumption metric”, which approximates mean feeding rates of all urchins in a treatment using a value of 0 for non-consumers). Before analysis, we transformed logged posterior estimates of mass-specific feeding rate parameters () to the natural scale using:

13

and the associated output from each posterior draw.

We assessed the distribution of differences between posterior estimates of the combined consumption metric for each predator treatment within each urchin size class, using both raw (ambient – predator) and proportional (ambient/predator) comparisons. Effect strength was determined based on the position of 0 relative to the posterior distribution of raw treatment differences or the position of 1 relative to the posterior difference of proportional treatment effects. Alternatively, for direct comparisons of means or effect sizes we assessed overlap in 90% credible intervals (CIs) and the median posterior estimate for each distribution where such comparisons were clear, e.g. substantial overlap or clear differentiation.

Associations between treatment and urchin position in the feeding arena, interaction with kelp, and use of shelter were visualized and assessed qualitatively. To compare relative activity of urchins during trials, we calculated the total distance of the most direct path between all observed urchin locations and divided by urchin diameter, describing relative changes in daily position in terms of body lengths (Supplementary Fig. S2). We analyzed the relationship between relative activity and treatment using a linear mixed effects model with minimum distance traveled as the independent variable, urchin size class and predator treatments as fixed effects, and trial number and header tank as random effects. Modeling was performed using lme4⁶⁸ and effect sizes were estimated using a Kenward-Rogers approximation to estimate degrees of freedom⁶⁹. For this test especially, but also in presentations of Bayesian outputs, we describe our results using thresholds and language regarding strength of evidence adapted from Muff et al.⁷⁰. Data from the first trial was excluded from analyses of urchin position due to lack of specificity, but provided sufficient detail regarding interaction with kelp and use of shelter for inclusion in those comparisons.

Lobster experiment re-analysis

To compare the effect of blue cod exposure on urchin feeding behavior with the effect of lobster exposure, we re-analyzed data from the lobster experiment using the Bayesian model described above with a few modifications. To maximize comparability of the two datasets, only feeding measurements of urchins with a size range represented in the blue cod experiment were included from the lobster experiment (i.e., data from urchins < 30 mm TD were excluded). Despite being infrequent compared to positive changes in kelp mass among controls, minor degradation of kelp did occur in both experiments in the absence of consumption (blue cod: 6/109, lobster: 4/24). However, kelp degradation in the lobster experiment was less extensive than observed during the blue cod experiment (maximum loss: 0.6% vs. 3.6%). Imputed estimates for proportional kelp growth in both experiments were therefore modelled to include the possibility of mass loss but were truncated at a lower bound of 0.98 (2% loss) for the lobster experiment, compared to the blue cod experiment where a lower growth limit was not explicitly modelled.

Consumption estimates from the lobster experiment approximated a normal distribution (rather than the log-normal distribution modeled in the blue cod experiment), so Eq. (7) from the blue cod consumption module was modified to:

14

with a prior for m_t of N(0.08,0.02) for small urchins and N(0.03, 0.02) for large urchins⁴⁸. Random effect priors (for ε_r and ε_h) were assigned as N(0,0.02), and mean random effect posteriors were added to m_t posteriors before calculation of treatment effects. All other equations and analyses comparing urchin feeding rates were conducted as described for the blue cod experiment, with the only differences being sample sizes (n_SA = 8, n_SP = 14, n_LA = 13, n_LP = 26 vs. n = 40 for all blue cod treatments), number of trials (6), and number of headers (3). Urchin position and activity were not systematically recorded during the lobster experiment in a manner that allowed further analysis or comparison to results from the blue cod experiment.

Author contributions

JC and SW conceived of project. JC primarily organized project logistics, with assistance from SW. JC, PD, and SW contributed equally to study design. JC solely participated in data collection. JC and PD designed and performed statistical analyses. JC authored primary manuscript draft, with substantial editorial contributions from PD and SW.

Data availability

Data and novel code supporting this manuscript are available on Figshare: 10.6084/m9.figshare.28319942.

Declarations

Competing interests

The authors declare no competing interests.