Introduction

Climate change is shifting the distribution of species1 and altering ecological assemblages, as not all taxa respond to warming by shifting their range, and species that move do so at different rates2. This reorganization of ecological communities can lead to some biotic interactions being lost, when interacting species no longer co-occur in space, and can also lead to new interactions among previously separated taxa3. In some instances, these climatic impacts on species interactions can outweigh or even reverse the direct effects of climate4,5.

Climate-mediated disruptions in biotic interactions linked to species redistributions are already impacting the health of ecosystems and humans, as well as our economy6. Documented impacts range from increased impacts of pests and diseases to unpredictable fisheries and crop yields6,7. A better mechanistic understanding of how climate change alters species interactions is therefore key to understand and predict future ecosystem function.

Changes in species assemblages are manifesting at particularly rapid rates in marine ecosystems1. Species redistributions are causing a global ‘tropicalisation’ of marine communities, as species from the tropics become increasingly abundant in higher latitudes8. Plant–herbivore interactions are particularly important in the ocean9 and have been significantly impacted by climate change and tropicalisation10,11. Around the world, range-expanding tropical and subtropical herbivores are overgrazing canopy seaweeds and removing their recruits, causing regime shifts towards systems dominated by low biomass turf algae or urchin barrens8,10,12. In particular, in recent years the range expansion of tropical and warm-temperate herbivorous fish has been linked to the deforestation of seaweed forests around the world, including in Japan13, the eastern Mediterranean12 and the eastern and western coastlines of Australia10,11.

The deforestation of temperate reefs by range-shifting herbivorous fishes may be facilitated by functional differences in seaweed-herbivore interactions between tropical and temperate systems8,14. In particular, herbivorous fishes in the tropics are much more diverse taxonomically and functionally than in temperate reefs15. The tropics are also characterized by higher rates of fish herbivory than temperate systems16. These patterns have been linked to differences in the palatability of seaweeds, with some evidence that tropical seaweeds may be less palatable and better defended chemically than temperate taxa17. Further, nitrogen content, a critically important nutrient for herbivores, is generally higher in temperate plants and macrophytes than in tropical species18 and may therefore further stimulate consumption. These patterns have led scientists to hypothesize that the range expansion of tropical herbivorous fish into higher latitude reefs may be facilitated by the presence of more nitrogen-rich and weakly defended seaweeds in temperate reefs14,19.

However, most mechanistic palatability assays focusing on identifying specific nutritional traits that mediate feeding have focused on invertebrates14,17 which are functionally different from fish in terms of their mobility and their ability to produce the cellulase necessary to digest brown algae20. Gaining a better understanding of how the palatability of brown seaweeds varies among species and what factors influence fish herbivory are particularly important in the context of continued range expansions by tropical herbivorous fish, as these seaweeds include the most important habitat-formers in temperate reefs (e.g., Laminariales and Fucales) and fish are the taxa that are shifting their distribution polewards at the fastest rate21.

Here, we experimentally test the hypothesis that temperate brown seaweeds are preferentially consumed by fish in a tropicalised coastline because they are more nutritious and less chemically defended than tropical brown seaweeds. The palatability of co-occurring temperate and tropical seaweeds was measured in the field in southeastern Japan. This region is a global hotspot for environmental change in response to ocean warming where tropical fishes are now dominant22. Temperate kelp forests have been decreasing in these reefs since the late 1990s23,24, while coral cover and diversity have increased extensively25,26. Further, a rapid shift from temperate Sargassum species (e.g., S. micracanthum and S. yamamotoi) to tropical Sargassum species (e.g., S. ilicifolium and S. carpophyllum) has been occurring since the late 1980s23. Although the direct effect of rising water temperatures on seaweed physiology is considered an important factor for long-term changes in seaweed vegetation in this region (seaweed bed decline and algal species changes)23,27, it is still unclear how indirect effects such as increases in fish herbivory may be involved in these changes. Serisawa et al.27 reported fish damage on kelp blades and bladelets in Tosa Bay. Further, herbivorous fish damage to canopy-forming seaweed (Laminariales and Fucales) populations has also been reported from several regions in Japan since the late 1990s28,29,30,31,32. Since similar coastal environment changes and increases in herbivorous fish abundance/activity are also occurring in other temperate locations of the world11,33, assessing whether and how herbivorous fish preferentially consume temperate seaweeds can contribute to understanding the potential mechanisms facilitating the decline of kelp and the range expansion of tropical seaweed species.

To assess whether herbivorous fish are selectively feeding on temperate seaweeds and whether the physiology and nutritional value of seaweeds are related to this feeding preference, we used a combination of multiple-choice filmed feeding experiments and chemical analyses of co-occurring temperate and tropical canopy-forming brown seaweeds. This included the dominant kelp in the study region, Ecklonia cava, a perennial species characterized by high biomass throughout the year, as well as dominant temperate and tropical Sargassum species, which are ephemeral and have relatively short periods of high vegetation cover during the winter–early summer34. The feeding experiments were performed during summer, and subsequent assays with the kelp Ecklonia cava (the only canopy species present year-round in the region) were performed on a monthly basis to characterize changes in kelp consumption throughout the year35.

The specific questions we asked are: (i) Are herbivorous fish abundance and biomass higher at tropicalized locations than at kelp-remained locations? (ii) Are temperate brown seaweeds consumed preferentially by herbivorous fishes? (iii) Are fish feeding preferences mediated by seaweed nutritional or chemical defence traits? (iv) How does consumption of the dominant perennial kelp Ecklonia cava vary temporally?.

Materials and methods

Study locations and study species

This study was performed in Tosa Bay in Kochi prefecture, southeastern Japan (33°N, 133°E). Fish and benthic surveys were performed at two coral-dominated (Tei and Yokonami) and two nearby kelp-dominated locations (Usa and Tanoura; Fig. S1). Usa and Tanoura are the only remaining kelp forests in the region. Tei and Yokonami are known as tropicalised locations. Extensive kelp beds (180 ha) existed at Tei23, but corals have rapidly developed since the disappearance of kelp in the early 2000s8. At Yokonami, Sargassum beds were present until the 1980s, but corals have now developed22. In situ feeding experiments were performed at the coral dominated Tei and Yokonami only, where herbivorous fish biomass was higher (see Results). Surveys were performed in 2014 during summer (July), which represents the biomass peak of all Sargassum species34. Sargassum species are ephemeral and disappear from mid summer onwards34. An additional bioassay experiment was set up in October 2014 to measure consumption rates of the perennial kelp species Ecklonia cava in the absence of Sargassum species. This was followed by monthly bioassay deployments as described below. The water temperature in central Tosa Bay varies seasonally between 15–16 °C (February) and 28–29 °C (August)22,34.

We used six canopy-forming brown algae in our feeding experiments, three of tropical distribution (Sargassum ilicifolium, S. carpophyllum, and S. alternato-pinnatum) and three of temperate distribution (Ecklonia cava, S. micracanthum, and S. patens). All species are abundant canopy-forming algae in Tosa Bay23. Their depth distribution is between 1 and 10 m. Seaweed palatability may vary between populations in regional and latitudinal scales17 probably because seawater nutrient conditions may affect palatability36. This study aimed to compare the differences in palatability among macroalgae species therefore all seaweeds were collected from the same location (Pacific Ocean side of the Yokonami Peninsula, central Tosa Bay). Although the epiphytes are considered important food resources for browsing parrotfishes37, we did not remove epiphytes in this study because (1) we targeted not only parrotfish but also all herbivorous fishes, and (2) we aimed to clarify the selectivity of herbivorous fishes for each macroalga species under natural conditions. We avoided using macroalgae with high epiphytic load in the experiments. The experimental seaweeds were collected on the day or a day before the experiment. The latter were maintained in a large, free-flowing seawater tank (2 t tank at the Usa Marine Biological Institute, Kochi University) until the experiment day and then transported to the experimental sites using lidded buckets with aeration under appropriate cool condition.

All methods were carried out in accordance with the UNSW Animal Research Ethics Procedure. An Animal Ethics permit was not necessary in this instance as fish were filmed in their natural environment and there was no direct interaction or manipulation of fish.

Characterisation of herbivorous fish and benthic communities

The herbivorous fish and benthic community composition of each reef were quantified in July 2014 using n = 5 transects haphazardly deployed between 9 am and 3 pm and separated by at least 10 m. Herbivorous fish assemblages were surveyed by the same expert (Yohei Nakamura) using underwater visual surveys (25 m long, 5 m wide) by identifying all individual fish to species. Total length of each fish was estimated to the nearest cm (roving fishes) and 0.5 cm (resident fishes). Length estimates of individual fish were converted to biomass using the allometric length–weight conversion W = a * TLb, where W = weight in grams, TL = total length and parameters a and b are constants obtained from Fishbase (www.fishbase.se). The total abundance of herbivores per transect was contrasted among habitat type (coral and kelp; fixed factors) and location (coral: Tei, Yokonami; kelp: Usa, and Tanoura; fixed factors nested in habitat—these were considered fixed as the kelp sites represent the only kelp-dominated sites in the region) using a negative binomial generalised linear model. The negative binomial distribution was used as it provided the best data fit. The total biomass of herbivorous fish per transect (log transformed) was contrasted among the same covariates using a linear model. Models were run using the manyglm and manylm functions in the R package mvabund38, with statistical inferences from likelihood ratio tests using the anova function and with Monte Carlo resampling for abundance models and residual resampling for biomass models, with 1,000 iterations. Model assumptions (here and for and all mvabund models hereafter) were checked as per Wang et al.40 using mean–variance versus fitted plots (manyglm, manylm) and mean–variance relationship plots (manylm). Transformations were made where the mean–variance assumption was not met.

Benthic assemblages were characterized using photo-quadrats (50 cm × 50 cm) taken every meter (25 per transect) by a diver swimming at least 5 m behind the diver undertaking the fish counts. Three points were randomly overlaid on each image using CPCe (National Coral Reef Institute, v4.1), and the taxon or substratum underneath each point was identified, giving 75 points per transect. Each point was classified using the following eight categories: kelp, other macroalga (Sargassum and macroalgae), table coral, other live coral (all live coral besides table coral), turf and calcareous algae, dead coral (all dead corals), rocky reef and bare (bare rock and sand). The benthic community matrix was contrasted among habitat type (coral, kelp; fixed factors) and location (coral: Tei, Yokonami; kelp: Usa, Tanoura; fixed factors nested in habitat) using a multivariate linear model and the manylm function in R. Statistical inferences were carried out using likelihood ratio testing with the anova function and residual resampling with 1000 iterations.

Feeding preference experiments

We set up the feeding experiments in shallow reef habitat (1–3 m), in patches dominated by a thin layer of epilithic algal matrix (EAM) cover sensu Wilson et al.38 within the coral-dominated locations (Tei and Yokonami). For each replicate assay, we tied individuals of the six seaweeds at 15 cm intervals to a 1 m rope that was weighed down at both ends. We aimed to keep initial biomass values as similar as possible between species within each replicate assay, however Ecklonia cava was always heavier than all other species and Sargassum micracanthum always lighter. The order in which the six seaweeds were attached was haphazard and varied between replicates. Each replicate assay included a paired control assay that was protected from herbivores by a 3 mm mesh cage (120 × 30 × 30 cm) to control for any biomass change not resulting from herbivory (e.g., handling losses and algal detachment resulting from water movement). At each location and time, we deployed between 8 and 15 replicate assays over a span of 2–3 days: a total of 9 assays were deployed at Tei in July and 15 in October, and 7 at Yokonami in July and 15 in October 2014. To maximize independence among replicates, we separated individual replicate assays by at least 5 m and positioned them in a different location each day. In October 2014, an additional set of feeding assays were performed using only Ecklonia cava, as this was the only species still remaining at that time of the year. The water temperature during the experimental period was about 26 °C in July and about 24 °C in October.

We deployed assays over intervals of 3 h, between 10:00 a.m. and 5:00 p.m. because herbivorous fishes are generally diurnal in the wild39. We pat dried the macroalgae to remove excess water and measured fresh weight to the nearest 0.1 g before and after deployment. For each species, we calculated the biomass lost by subtracting the proportion of alga lost from the caged controls (initial g − final g/ initial g) from the proportion of biomass lost from the uncaged treatments uncaged–caged;40.

Video recording was started immediately after the assay was deployed. We filmed the deployed feeding assays for 2–4 h (mean ± SE = 202 ± 0.7 min) each day to quantify the seaweed consumption patterns of individual fishes. On each filmed replicate, video footage was analysed. We counted the total number of bites taken by individual fish on either the experimental seaweed species tethered to the rope or on the EAM surrounding the assays. Forays where rapid consecutive bites by an individual fish took place without discernable pauses were conservatively classed as a single bite, as it was not possible to accurately measure the number of bites.

Biomass loss was contrasted among seaweeds of tropical/ temperate origin (fixed factor) and among algal species (fixed factor nested within tropical/ temperate origin) as well as between locations (fixed factor) using a linear model with manylm. Statistical inferences were carried out with likelihood ratio tests using residual resampling with 1,000 iterations.

The number of bites taken by the four most common herbivorous fish species observed (Calotomus japonicus, Kyphosus vaigiensis, Naso unicornis and Siganus fuscescens) was also contrasted among tropical/ temperate origin of algae, among algal species and between locations using negative binomial generalised linear models with manyglm. Using the anova function, statistical inferences were carried out with likelihood ratio tests using Monte Carlo resampling with 1000 iterations.

Temporal patterns in kelp consumption

Consumption of the kelp Ecklonia cava in October 2014 when offered as a single-species assay appeared to be higher than in July, when E. cava was offered alongside other seaweed species (Fig. 2). To quantitatively determine whether kelp consumption varies throughout the year, additional feeding experiments were conducted monthly throughout the year of 2018 at Yokonami. Six E. cava kelp individuals were deployed each month (mean ± SE = 50.6 ± 2.6 g), three as treatment and three as control with herbivore exclusion cages (2 mm mesh, 20 × 20 × 20 cm), except October (n = 2 for treatment and control, respectively). The kelps used in the experiment were in good condition (not wilted), with few epiphytes in all months. Reproductive sori were observed on kelp blades from September to December at Usa, where the experimental kelps were collected. Each replicate assay was haphazardly placed at least 20 m apart from each other. An underwater video camera was deployed in each of the three treatment kelps, and the kelps and cameras were retrieved two hours after the start of the experiment. The percentage of kelp removed was calculated from the wet weight of the seaweed before and after the deployment. The results were compared between the treatments (treatment vs. control; fixed factor) and the sampling months (January–December; fixed factor) by two-way ANOVA using natural log transformed data. Model diagnostics (residuals plots and Durbin–Watson test) were run using the DHARMa package41. No temporal autocorrelation was detected.

Seaweed nutritional traits

The nutritional traits (carbon, nitrogen and phenolics) of the six seaweeds were compared. Individual thalli of each species (n = 5) were freeze-dried and ground. Carbon and nitrogen content were analysed using a CHN Elemental Analyser (TruSpec®Micro Series, Michigan) at the Mark Wainwright Analytical Centre at UNSW Sydney. Phenolic content was analysed colorimetrically using the Folin-Ciocalteu’s method using gallic acid as a standard. Phenolics were extracted from ground seaweed (4–4.5 mg) using methanol–water (1 ml; 1:1). Samples were shaken at room temperature for five minutes at 30 Hz using a Tissuelyser (Qiagen, WestSussex, UK) and incubated for 24 h at 4 °C prior to colorimetric analysis using the Folin-Ciocalteu reagent and 20% sodium carbonate solution. Absorbance was read at 765 nm with a fluorescent platereader (SynergyTM, Vermont). Total phenols were expressed as gallic acid equivalents (GAE % dry weight).

The total nitrogen content, C:N ratio, and total phenols (log transformed) were contrasted between tropical and temperate seaweeds (fixed factor) and among algal species (fixed factor nested within tropical/ temperate) with linear models using manylm. We used likelihood ratio testing and residual resampling with 10,000 iterations for statistical inferences.

To test whether any of the measured nutritional traits (N, C:N ratio and phenolics) is related to consumption of seaweed species by the dominant fish species observed biting (Calotomus japonicus), we performed a multiple regression with the lm function in R. The total number of C. japonicus bites for each assay was the response variable with the scaled covariates of the mean nitrogen, C:N ratio, and phenolics of each seaweed species as the predictor variables.

Results

Characterisation of herbivorous fish and benthic communities

Total herbivorous fish abundance differed between coral and kelp dominated habitats (df = 1, deviance = 5.41, P = 0.010) and among locations (df = 4, deviance = 11.16, P = 0.005; Fig. 1a). Total herbivorous fish biomass was higher at coral-dominated habitats in contrast to kelp habitats (F1,18 = 6.505, P = 0.015; Fig. 1b) and differed among locations (F6,16 = 1.163, P = 0.001026) driven by very low biomass of herbivores at Usa (Fig. 1b).

Figure 1
figure 1

Total abundance of herbivorous fish (a), and total biomass of herbivorous fish (b) observed across kelp-dominated ‘temperate’ locations (in blue) and coral-dominated ‘tropicalised’ locations (in red). Bars are mean (± SE) and dots are the raw observations per transect.

Full size image

In the kelp habitat, Prionurus scalprum and Girella punctata dominated the herbivorous fish communities in terms of abundance (47% and 40% of the total herbivorous fish across the two kelp sites, respectively). However, their body sizes were mostly small (5–15 cm TL and 5–10 cm TL, respectively). In the tropicalized coral-dominated habitats, P. scalprum dominated the herbivorous fish communities in terms of abundance (71% of the total herbivorous fish across the two coral sites). The herbivorous fish community in the coral-dominated habitats was characterised by the high abundance of large size individuals of P. scalprum and C. japonicus (15–30 cm TL and 20–40 cm TL, respectively), as well as the presence of tropical surgeonfishes (Acanthurus nigrofuscus, A. dussumieri, A. lineatus, and Naso unicornis) and tropical parrotfish (Scarus ghobban).

The overall benthic community composition differed between kelp and coral dominated habitats (F1,18 = 7266.00, P = 0.002) and between locations (F6,16 = 46.00, P = 0.002). Temperate locations (Tanoura and Usa) were dominated by the kelp, Ecklonia cava, which had the highest percent cover at Usa and was completely absent from Tei and Yokonami (Fig. S2). Tropicalized locations (Tei and Yokonami) were dominated by live table coral, which was more common at Yokonami and completely absent from Tanoura and Usa (Fig. S2).

Feeding preferences by herbivorous fish

We observed five herbivorous fish species taking a total of 7689 bites on the feeding trials: Calotomus japonicus, Kyphosus vaigiensis, Naso unicornis, Siganus fuscescens and Prionurus scalprum. However, P. scalprum was only observed biting in four videos and was subsequently excluded from analyses. We observed the turtle Caretta caretta attempting to feed on Ecklonia cava once, but it failed to take a proper bite. There was no feeding observed in three replicate camera deployments in Tei; this data was excluded from the analysis as it provides no information on relative feeding preferences. A total of nine caged controls (three in Tei and six in Yokonami in July) were disturbed during the experiment allowing herbivores to consume the algae. The values given to these controls were the average initial and end biomass of undisturbed controls from that location. Final sample sizes in July were n = 14 for Tei, n = 13 for Yokonami; in October, n = 15 for Tei, n = 15 for Yokonami.

Biomass loss was higher at Yokonami in contrast to Tei (F1,178 = 48.10, P = 0.002; Fig. 2) and was lower for temperate seaweeds (F1,177 = 4.49, P = 0.034). Algae biomass loss differed among seaweed species (F6,171 = 15.98, P < 0.001) but this was not consistent between locations (seaweed species and location interaction; F6,166 = 6.01, P < 0.001; Fig. 2).

Figure 2
figure 2

The proportional biomass loss from each seaweed species during feeding trials in (a) Tei and (b) Yokonami. Bars are the mean (± SE) algal biomass lost in treatments minus the algal biomass most in paired controls for each algal species and dots are the raw data.

Full size image

The number of bites taken by C. japonicus was not influenced by whether an alga was of tropical or temperate origin (df = 1, deviance = 1.70, P = 0.193; Fig. 3a). However, Calotomus japonicus was observed feeding more often at Yokonami for four algae species (location and algal species interaction; df = 6, deviance = 14.28, P = 0.031; Fig. 3a). The kelp E. cava during October experienced the highest number of bites by C. japonicus at both locations (Fig. 3a). Bites by K. vaigiensis were not influenced by the tropical or temperate origin of seaweeds (df = 1, deviance = 0.02, P = 0.885; Fig. 3b). K. vaigiensis was observed feeding more often at Tei (df = 1, deviance = 14.40, P < 0.001; Fig. 3b), driven mostly by the large number of bites taken from Sargassum patens in one of the videos (interaction between location and species; df = 7, deviance = 0.001, P = 0.015; Fig. 3b). Bites by N. unicornis were also not influenced by the orgin (tropical or temperate) of seaweeds (df = 1, deviance = 0.0001, P = 0.998; Fig. 3c) and were marginally more frequent at Yokonami (df = 1, deviance = 4.00, P = 0.051; Fig. 3c). S. fuscescens also fed equally on tropical/ temperate seaweeds (df = 1, deviance = 1.89, P = 0.171; Fig. 3d), and fed more often at Tei, except for in October where more bites were taken from E. cava in Yokonami (location and species interaction; df = 6, deviance = 12.14, P = 0.002; Fig. 3d).

Figure 3
figure 3

Number of bites taken per hour from each alga for the four (a) Calotomus japonicus, (b) Kyphosus vaigiensis, (c) Naso unicornis, (d) Siganus fuscescens on the two experimental sites: Tei (left side) and Yokonami (right). Bars are mean (± SE) bites per hour and dots are raw data. Bars sharing a letter did not differ among algae of the same distribution. Note truncated y-axis in (b) and (d).

Full size image

Temporal patterns in kelp consumption

We identified clear changes in kelp herbivory across the annual cycle in our assays, with the proportion of kelp biomass remaining after deployment being significantly lower in the herbivore-exposed treatments than in the controls, but only from September to December (treatment and month interaction; F11,46, P < 0.001; Fig. 4). Kelp herbivory was most intense in October (Fig. 4). Video analysis showed that all feeding was by Calotomus japonicus. In the months when there was no herbivory on kelp (January-July), some C. japonicus were also captured on video footage, but they were not observed feeding on the kelp bioassays.

Figure 4
figure 4

Monthly changes in the mean percentage remaining of the kelp biomass at Yokonami in 2018. The number of replicates is n = 3 for all months and for each treatment except October, where n = 2. Vertical bars indicate standard error. The broken line indicates the monthly change in seawater temperature during the experiment in each month at the study site.

Full size image

Seaweed nutritional traits

Nitrogen content differed between algae with tropical and algae with temperate distributions (F1,33 = 10.63, P = 0.003; Fig. 5a), with tropical algae generally displaying lower levels of nitrogen. Nitrogen content also differed among algal species (F6,28 = 18.02, P < 0.001; Fig. 5a). The C:N ratio differed among seaweed species (F6,28 = 13.66, P < 0.001) regardless of whether the species has a tropical or temperate distribution (F1,33 = 1.42, P = 0.236; Fig. 5b). The total phenolics differed among algal species (F6,28 = 23.36, P < 0.001) and were higher in temperate species in contrast to tropical species (F1,33 = 41.76, P < 0.001; Fig. 5c).

Figure 5
figure 5

Nitrogen (a), C:N ratio (b) and total phenols (c) from each algal species in July and from Ecklonia cava in October. Bars that share a letter did not differ in Tukey’s post hoc tests contrasting algae from tropical and temperate distributions.

Full size image

We found a significant negative relationship between the number of bites by C. japonicus and increasing nitrogen content (r2 = 0.099, F1,98 = 10.924, P = 0.001; Fig. S3a), no relationship with C:N ratio (r2 = 0.029, F1,98 = 1.040, P = 0.310; Fig. S3b), and phenolics (r2 = 0.079, F1,98 = 0.05, P = 0.831; Fig. S3c).

Discussion

In this study, we found no evidence that temperate brown seaweeds are more palatable to herbivorous fish than tropical species, as originally hypothesised. Although nitrogen content was slightly higher in temperate species, none of the herbivorous fish observed feeding displayed a preference for temperate species, and overall tropical species declined in biomass at higher rates. Consumption of the declining kelp Ecklonia cava was low when this species was offered alongside tropical and temperate Sargassum species, which were generally preferred by all herbivorous fish. Kelp consumption, however, varied over a yearly cycle and was highest during the late summer and the autumn, when Sargassum species are absent due to natural senescence. High consumption rates on E. cava in the autumn coincide with the reproductive season for this kelp42, and may contribute to the long-term historical decline of this kelp in Tosa Bay23.

Kelp consumption was dominated by the parrotfish Calomotus japonicus, one of the most common endemic herbivorous fish in southern Japan. This parrotfish has also been identified as a key consumer of other kelp species in warming reefs elsewhere in Japan43. Since the highest rates of kelp consumption in the autumn coincide with the reproductive season of E. cava, it is possible that C. japonicus may be targeting reproductive tissue in plants. The herbivorous labrid Odax pullus is also known to target reproductive receptacles on brown algae44. However, although high consumption rates on Ecklonia cava in the autumn coincided with the highest nitrogen content recorded and nitrogen accumulation is associated with sorus formation45,46, we found an overall negative relationship between nitrogen content and C. japonicus consumption rates in our multiple choice assay.

Another possible explanation for the observed changes in Ecklonia cava consumption by C. japonicus may be linked to kelp surface-associated epibionts. Recently, Clements et al.37 hypothesized that browsing parrotfish target protein-rich epiphytes and microbes (e.g. cyanobacteria) on macrophytes. Some studies show that the surface microbiome of kelp changes seasonally and with ocean temperature47, and it is therefore possible that C. japonicus respond to those kind of changes. The broad preference by multiple herbivorous fish towards Sargassum could also be related to the epibionts. Indeed, several studies have suggested that herbivorous fish in the tropics known to consume Sargassum may be targeting epiphytes living on the seaweed rather than the seaweed itself48,49. The Sargassum species used in the feeding assays have a more complex structure at small scales (i.e. a greater number of small lateral thalli) than the kelp Ecklonia cava, which has a flatter, more laminar structure. The greater structural complexity of Sargassum species may create more suitable microniches for the accumulation of epiphytes and microbes, which may lead to greater consumption of Sargasssum by herbivorous fish.

Overall, seaweed biomass loss was slightly higher for tropical taxa. Tropical species also displayed lower levels of phenolic compounds, which are considered as deterrents for some herbivores50 although not all species are affected51. In this study, none of the individual herbivorous fish showed a preference for phenolic-poor tropical taxa. Other nutritional and structural components not measured here may also have influenced feeding preferences, as traits such as lipids and a broad range of secondary metabolites and structural materials have been previously linked to herbivore feeding patterns52,53,54.

Consumption of brown canopy-forming seaweeds was dominated by just four herbivorous fish. This finding is consistent with multiple studies from coral reefs, where a small number of species drive consumption patterns of seaweeds such as Sargassum in the Indo-Pacific55,56. Further, three of the species identified as the main consumers here, the drummer Kyphosus vaigiensis, the surgeonfish Naso unicornis and the rabbitfish Siganus fuscescens, are among the species most frequenly observed feeding on brown seaweeds in coral reefs55,56. S. fuscescens, and to a lesser degree K. vaigiensis, have also been identified as key consumers of temperate kelp in tropicalised regions of Australia where kelp is declining and the abundance of tropical and subtropical herbivores is increasing11,57,58. Further, two congeneric Siganus species morphologically similar to S. fuscescens are the main consumers responsible for overgrazing canopy forming seaweeds in the Mediterranean12, thus confirming the importance of a small number of taxa in driving climate-mediated phase shifts.

The tropicalization of coastal environments can occur heterogeneously in space and time within the same region13. In some locations, the changes may occur rapidly, while in others they may be slower. If herbivorous fishes are partly responsible for tropicalization, herbivores are expected to be fewer in the kelp-remaining locations than in the tropicalized locations. Our study supported this assumption. The water temperature in the locations where kelp remains (Usa and Tanoura) is about one degree Celsius cooler than in other areas throughout the year, probably because there is a river nearby in each location, which may provide suitable conditions for kelp survival and slow down the invasion rate and growth of herbivorous fish populations under ocean warming. The higher biomass of herbivores in tropicalised reefs was driven by the abundance of large browsing herbivores such as Calotomus japonicus and by large grazers such as the surgeonfish Prionurus scalprum. Although P. scalprum were generally not observed consuming large quantities of seaweeds, they were observed biting surrounding turf and epilithic algae, which may contain recruits of Sargassum and kelp. This species may thus play a role in keeping reefs in coral-domianted states by preventing the reestablishment of seaweeds through cropping10,11. A similar ecological role may also be played by other grazers/scrapers, such as tropical surgeonfishes (Acanthurus spp.) and parrotfish (Scarus ghobban) that occurred abundantly in coral-dominated locations. The increase in tropical herbivorous fishes in addition to native temperate herbivorous fishes is thought to increase the diversity of herbivory functions that promote the decline and prevention of the recovery of seaweed beds, as seen in Western Australia59. Since the loss of kelp beds in Tosa Bay is linked to high water temperature stress during the summer months23, our results suggest a scenario where high herbivory during the kelp reproductive and recruitment seasons exacerbate this decline.

To conclude, our findings suggest that the shift from temperate to tropical Sargassum species observed in Tosa Bay is probably not related to changes in fish herbivory, and may instead be caused by direct physiological effects from ocean warming on the seaweeds. In contrast, the dramatic historical decline in Ecklonia cava kelp forests observed in Tosa Bay23 is likely to be not only due to direct physiological effects on the kelp27, but may also be at least partly driven by high feeding pressure on adult kelp during times of the year when other seaweeds are not abundant and when the kelp itself is reproductive. This information contributes to our understanding of shifting trophic interactions in warming ecosystems and can inform coastal conservation measures such as seaweed restoration efforts in response to the decline of seaweed beds associated with ocean warming.