Indirect effects shape macroalgal epifaunal communities

Abstract We tested the response of algal epifauna to the direct effects of predation and the indirect consequences of habitat change due to grazing and nutrient supply through upwelling using an abundant intertidal rhodophyte, Gelidium pristoides. We ran a mid‐shore field experiment at four sites (two upwelling sites interspersed with two non‐upwelling sites) along 450 km of the south coast of South Africa. The experiment was started in June 2014 and ran until June 2015. Four treatments (predator exclusion, grazer exclusion, control, and procedural control) set out in a block design (n = 5) were monitored monthly for algal cover for the first 6 months and every 2 months for the last 6 months. Epifaunal abundance, species composition, algal cover, and algal architectural complexity (measured using fractal geometry) were assessed after 12 months. Predation had no significant effect on epifaunal abundances, while upwelling interacted with treatment. Grazing reduced the architectural complexity of algae, with increased fractal dimensions in the absence of grazers, and also reduced algal cover at all sites, though the latter effect was only significant for upwelling sites. Epifaunal community composition was not significantly affected by the presence of herbivores or predators but differed among sites independently of upwelling; sites were more similar to nearby sites than those farther away. In contrast, total epifaunal abundance was significantly affected by grazing, when normalized to algal cover. Grazing reduced the cover of algae; thus, epifaunal abundances were not affected by the direct top‐down effects of predation but did respond to the indirect effects of grazing on habitat availability and quality. Our results indicate that epifaunal communities can be strongly influenced by the indirect consequences of biotic interactions.

regulated by a combination of bottom-up (resource availability) and top-down (predation) effects, with the balance between the two differing across space and time and across gradients of environmental stress. Importantly, in marine systems a range of bottom-up effects are driven by water movement, including the supply of nutrients, suspended food particles, and larvae (Menge, 2000). For example, large-scale phenomena such as upwelling affect nutrient supply to primary producers (Andrews & Hutchings, 1980;Pitcher et al., 1991), but very small-scale effects such as guano input (Methratta, 2004) or even epifaunal excretion (Probyn & Chapman, 1983;Taylor & Rees, 1998) can also be important at local scales. The effects of bottom-up forcing mechanisms such as upwelling can, however, be strongly context-dependent (Blanchette et al., 2009), with rates of bottom-up supplies influencing top-down species interactions (Menge et al., 2003).
The existence of trophic cascades is an indication that top-down effects can be very strong yet indirect. Examples include the collapse of populations of a top predator, cod, that resulted in high abundances of the planktivorous Sprat, which then hindered the recovery of cod by preying on their larvae (Casini et al., 2009). Another well-known example is the decline of kelp forests in North America due to the increase in sea urchin grazing following reductions in urchin predators (Tegner & Dayton, 1991). In the Aleutian Islands, this decline has further been linked to the top-down control of sea otter population by killer whales, with subsequent cascading positive and negative effects on sea urchins and kelp (Estes et al., 1998). Top-down effects are typically trophic, but we wished to test the possibility that they could lead to indirect, non-trophic effects, specifically through their influence on habitat availability and quality, and how this might interact with the direct top-down effects of predation. To test this, we examined how the abundance and community structure of macroalgal epifauna are affected by the direct top-down effects of predation and the indirect consequences of top-down and bottom-up effects on the availability and quality of macroalgal habitat. We studied the rhodophyte Gelidium pristoides, a commercially harvested macroalga (Anderson et al., 1989) in South Africa that dominates algal cover on the mid-shore. It usually comprises separate tufts and grows to a maximum of about 15 cm in length (Gibbons, 1988). Gelidium pristoides generally experiences high levels of grazing by macrograzers, particularly from a range of limpets such as Siphonaria spp. and Cymbula oculus, as well as various snails (Branch, 1981).
Macrophytes such as Gelidium spp. are important as primary producers and as ecosystem engineers providing habitat for a wide range of organisms (Cattaneo & Kalff, 1980). These include meiofauna (Gibbons & Griffiths, 1986) and high densities of epifaunal invertebrates, mostly polychaetes, small crustaceans, and gastropods. These species use macroalgae for protection from predators (e.g., Kon et al., 2009;Machado et al., 2019), mainly fish in our system (Newcombe & Taylor, 2010), and can feed on the macroalgae directly (Duffy, 1990) or feed on the periphyton they support (Brawley & Fei, 1987;Klumpp et al., 1992). The species composition of invertebrate communities can differ among algal species, and this appears to be related to the physical architecture of the algae including the structure, design, and organization of the algae rather than their species identity Taylor & Cole, 1994).
Growth rates of macroalgae respond directly to increased nutrient availability provided through upwelling or eutrophication (e.gNielsen & Navarrete, 2004;Worm & Lotze, 2006), and in areas with high nutrient input, they can strongly shape intertidal communities. For example, rocky shores in upwelling regions support significantly greater algal cover, abundances of sessile organisms, and biomass of herbivorous limpets per unit area than shores in regions lacking coastal upwelling (Bosman et al., 1987). Intertidal macroalgae have frequently been shown to be controlled by grazing (Duffy & Hay, 2001;Hawkins & Hartnoll, 1983), but the effects of grazing in intertidal systems can result in patchy distributions of macroalgae that can show marked spatial determinism (Diaz & McQuaid, 2011). In the case of epifaunal communities, macroalgae provide a point of attachment and protection from predation during high tide and a refuge from heat and desiccation stress during low tide (Wright et al., 2014). By reducing habitat availability, the top-down effects of grazing on macroalgae could have significant indirect consequences for epifaunal abundance and community structure. Similarly, macroalgal growth rates respond directly to increased nutrient availability so that the direct bottom-up effects of nutrient input on macroalgal growth and abundance are likely to have indirect consequences for the associated epifauna.
In this context, we used manipulative experiments to test the following hypotheses: 1. Predation has direct top-down effects on epifaunal abundances.
2. The effects of predation are stronger where upwelling is less frequent and less intense.
3. Top-down (grazing) and bottom-up (upwelling) effects on macroalgae will have indirect consequences for epifaunal communities.

| ME THODS
This study was undertaken at four moderately exposed rocky shores, separated from each other by 10 s of km along c.500 km of the south coast of South Africa (Figure 1)

| Characterizing upwelling
After subjectively categorizing the study sites as upwelling or nonupwelling, we confirmed this using two complementary approaches.

| In situ temperature data
To gauge the frequency and intensity of upwelling, three tempera-

| Wind data
Wind speed and direction were used to characterize coastal upwelling. We used wind data to calculate an upwelling index, which was used to quantify the duration and intensity of upwelling events at our four study sites. Wind data for the duration of the experiment were collected from four meteorological stations, each within 10 km of one of the four sites (South African Weather Service, 2015).
These were as follows: East London, Port Alfred, Port Elizabeth, and Knysna. Hourly wind speed and direction for each day were used to calculate an upwelling index (UPW) following Bakun (1975)

as:
where ρ a is the air density, C d is the drag coefficient (i.e., ca. 0.0014), v is the mean height-corrected wind speed, � ⃗ v is the alongshore vectorial component (estimated as zonal winds from the study sites), f is the Coriolis frequency (f = 9.9 × 10 −5 rad s −1 ) at middle latitudes, and ρ w is the water density (i.e., ρ w = 1025 kg m −3 ). Positive values represent periods of upwelling, and negative values represent periods of downwelling. The duration of upwelling events was categorized as long (≥6 days), medium (3-6 days), or short (≤3 days). an open roof allowing access to pelagic predators, but not benthic grazers; and (iv) control areas (Co), which had screws marking the four corners of the plot but were otherwise unaltered. Preliminary tests were conducted to infer potential cage artifacts, such as the effects of altered hydrodynamics and/or light availability on algal growth. Observations carried out before the beginning of the experiment confirmed the effectiveness of the treatments. Grazers such as the limpet Siphonaria concinna were observed within the grazer-only cages and control plots but were very rarely observed in the total exclusion or predator-only cages. Cages were regularly checked throughout the experiment and on the few occasions when grazers were found in inappropriate treatments they were removed.

| Experimental design
The experiment ran from June 2014 to June 2015 with sampling each month for the first 6 months and every two months thereafter.
During each sampling event, photographs were taken of algal cover in each plot and a wire brush was used to remove any algae growing on the cages. Because G. pristoides was the only alga present in the plots, there was no problem of algal overgrowth and percentage algal cover within plots could be calculated from the photographs using Coral Point Count (CPCe) and the point intercept method (Kohler & Gill, 2006;Lathlean et al., 2015).
Map showing all the four study sites in the southeast coast of South Africa. Red representing non-upwelling and blue upwelling sites After 12 months, the cages were removed and all G. pristoides and the associated epifauna were collected and stored in 10% formalin before subsequent sorting took place. In the laboratory, algal samples were washed of all organisms before the blotted weight and the dry weight of G. pristoides were measured. Blotted weight was taken immediately after washing with the G. pristoides rolled in tissue paper. Dry weight was measured after drying at 60°C for 48 h. All organisms washed from the algae were stored in 70% ethanol and identified to species level.
The architectural complexity of G. pristoides was assessed using fractal geometry (Mandelbrot, 1983), a mathematical tool that can be used to describe the structure of complex objects in situations where Euclidean descriptors are inappropriate (Seuront, 2010;Sugihara & May, 1990). For each treatment, five individual G. pristoides were photographed from four different angles, each picture being taken at right angles to the preceding one to capture the three-dimensional structure of the algae. The fractal dimension, for rotation of the initial 2D grid of 5° increments from 0 to 45° (Seuront et al., 2004). The potential presence of anisotropy in the structural complexity of G. pristoides was assessed through a comparison of the fractal dimensions D i using an analysis of covariance (Zar, 1999). If the null hypothesis of nonsignificant differences in D i between replicates was rejected, the data from the four replicates were pooled and a common fractal dimension D was used in further analysis after successfully inferring the lack of statistical significance between intercepts (Zar, 1999).

| Data analysis
Data used to distinguish between upwelling and non-upwelling sites fulfilled the prerequisites for parametric analysis (Shapiro-Wilk test and Levene's test) and were analyzed using one-way ANOVA. Data for algal cover similarly fulfilled the prerequisites for parametric analysis and were analyzed using a three-way nested ANOVA to assess the influence of treatment (fixed, four levels), upwelling (fixed, two levels), and site (nested in upwelling, random, four levels) on percentage algal cover. This was done at the start of the experiment (no significant effects of any factor or interaction) and again toward the end of the experiment. Because fractal dimensions D were non-normally distributed (p > .05), multiple comparisons between treatments were conducted using the Kruskal-Wallis test, and a subsequent multiple comparison procedure based on the Tukey test was used to identify distinct groups of measurements (Zar, 1999).

| Upwelling
The two upwelling sites, Port Alfred and Brenton-on-Sea, experienced more persistent and more frequent upwelling events (44 and 39, respectively) than Kidd's Beach and Kini Bay (27 and 13 events, respectively; Figure 2) although intensity, measured as mean upwelling index, was lowest at Brenton-on-Sea (Table 1).

| Algal cover and complexity
Overall algal cover varied with season across all treatments, being greatest in summer and lowest in winter, but changes in cover throughout the course of the experiment differed among the four experimental treatments at all study sites. These differences were most pronounced in July-September (Figure 3). Upwelling regime had no significant effect on cover of G. pristoides after 10 months (i.e., immediately before Brenton-on-Sea was inundated by sand in March) (ANOVA: F1,67 = 0.65, p > .05; Table 2). Cover did, however, differ among treatments, but the nature of such differences differed among sites (Treatment × Site interaction, ANOVA: F3,67 = 4.56, p < .05; Table 2). For example, at all sites, treatments allowing access to grazers (Control and Grazer +) showed lower algal cover than those excluding grazers (Closed and Predator +), suggesting strong grazing pressure, but the difference was only significant at the two upwelling sites (Port Alfred and Brenton-on-Sea) (Figure 4, Tukey HSD). This reflects the fact that grazer numbers are significantly higher at upwelling sites (unpubl. data).
The structural complexity of G. pristoides was consistently de-

| Epifaunal community structure
Each site supported between 13 and 19 species of epifauna, with an overall total of 44 species identified across the four sites. The most abundant taxa were crustaceans (amphipods and isopods), with some polychaetes and gastropods (Table A1).
PERMANOVA of raw data and of data normalized for algal cover showed that neither upwelling nor treatment had a significant influence on the community structure of epifaunal communities, but site did (PERMANOVA, p < .0001 in both cases) (Tables 3 and 4).
The influence of site was partially due to geography, with the SIMPER analysis indicating that sites that were farther apart were generally more dissimilar than sites that were close to one another.

| Epifaunal abundance
Generally, comparisons of treatments produced different results for total abundances and data normalized for algal cover. The key tests for the effects of predation on total epifaunal abundance were the comparison of Control vs Grazer + plots and of Closed vs Predator + plots, with both comparisons being nonsignificant.

| Total abundance
The only significant effect on total epifaunal abundances within plots was site (p = .033;

| Normalized to algal cover
When epifaunal abundances were normalized to unit area of algal cover, the only significant effect was treatment, with significantly lower numbers within treatments where grazers and predators were excluded (Tukey's HSD, p < .05; Table 2). Interestingly, plots that were open to grazers (i.e., Controls and Grazer +) supported higher normalized abundances even though the percentage cover of G. pristoides was found to be negatively affected by the presence of grazers ( Figure 6).
This means that grazing reduced algal cover, but not total epifaunal abundances, resulting in higher densities per unit of algal cover. 2. Grazing-It reduced the cover of G. pristoides (though the effect was significant only at upwelling sites) and its architectural complexity but had no effect on total epifaunal abundances. Together, these effects of grazing resulted in a significant increase in epifaunal densities.

| D ISCUSS I ON
Gelidium pristoides is an economically and ecologically important seaweed; it influences the structure and functioning of epifaunal communities and can be viewed as a keystone species, foundation species, or ecosystem engineer. Gelidium pristoides is responsible for influencing the composition and abundance of other species in the community (Jones et al., 1997;Shelton, 2010) by modifying environmental conditions, relationships between species, and the availability of resources. Ecosystem engineers can be used to assess the likelihood of successful ecosystem restoration (Byers et al., 2006) as they can be manipulated to facilitate the change of a community to a desired state. Studying the factors that affect keystone species and their influence on the environment can provide knowledge on the type of changes necessary for successful ecological restoration and how restoration efforts can be most effectively applied through natural ecosystem engineering.
Algae have frequently been shown to provide protection to epifauna against predation Bueno et al., 2020;Lanham et al., 2020;Ware et al., 2019), but in this system, we found no evidence of direct top-down control of epifaunal numbers by predators. On the contrary, we did find evidence of both top-down (grazing) and bottom-up (nutrient supply) effects on macroalgae, which had indirect effects on epifauna by modifying their habitat.
We compared the strength of these direct and indirect effects by manipulating grazing and predation pressure at sites experiencing either frequent or infrequent upwelling events. We rejected our predictions that predation would have significant effects at all sites and that this effect would be particularly strong where algal growth is not promoted by upwelling. Similarly, we rejected our hypothesis that upwelling would have an indirect effect on epifauna. Lastly, we conclude that grazing has a significant indirect effect on the epifauna when epifaunal densities were normalized to algal cover. To carry out our experiment, we characterized sites a priori as strongly or weakly influenced by upwelling, using this as a proxy for the degree of nutrient supply. We then tested our initial characterization by estimating the intensity, duration, and intensity of upwelling at each site. Many studies have successfully used a rapid drop in water temperature as a measure of the frequency of coastal upwelling (e.g., while data on hourly wind speed and direction have been used to calculate an upwelling index by, allowing the estimation of not just the frequency of upwelling events, but also their duration and intensity. We used both approaches, using sea surface temperature (SST) to calculate the number of upwelling events and wind data (speed and direction) to confirm upwelling frequency and estimate the intensity and duration of events, and distinguishing upwelling events of short, medium, and long duration. The results from both data sets confirmed our a priori classification of shores by upwelling (i.e., Port Alfred and Brenton-on-Sea exhibited more upwelling events and upwelling days), though, unexpectedly, Brenton-on-Sea had markedly lower upwelling indices than the other sites.

| Algal responses
We expected upwelling to affect the results of grazing by altering the balance between algal growth and removal. In this case, upwelling did indeed interact with grazing, but counterintuitively, effects of grazer exclusion were significant only at upwelling sites. Increased nutrient supply due to upwelling is usually associated with enhanced local primary production and algal standing stocks (Xavier et al., 2007), but this was not the case in the present experiment, with upwelling sites having roughly the same percentage algal cover as non-upwelling sites at both the start and end of the experiment. By the end of the experiment, treatments that allowed access to grazers (i.e., Grazer+ and Control) had similar levels of algal cover at both upwelling and non-upwelling sites and this may be attributed to the fact that there were roughly twice as many grazers at upwelling sites (AN unpub. data). This presumably results in greater grazing pressure and cancels out the positive effects of upwelling. In this case, upwelling did indeed interact with grazing, but counterintuitively, the effects of grazer exclusion were significant only at upwelling sites where higher numbers of grazers were more than compensated for (presumably) higher algal growth rates.
Algal cover in all treatments showed strong seasonality at all sites, declining through the winter months (June-August) and increasing through spring and summer. The seasonal pattern was, however, modified where grazers had access and treatments that allowed grazing showed much greater intersite variation during the course of the year. This modification within treatments that allowed access to grazers indicates the overriding of seasonal variation by top-down forcing.
Strong effects of treatment on algal cover (Figure 4) indicated strong grazing effects, with plots that excluded grazers (Closed and Predator+) having significantly greater cover than plots allowing access to grazers (Grazer+ and Controls). The interaction between upwelling and treatment was significant, but this reflected a difference in the intensity of this pattern, not a difference in the pattern. At both types of sites, grazed plots had less algal cover, but the Predator+ vs Grazer+ and Closed vs Control comparisons were nonsignificant at one of the non-upwelling sites and thus the pooled data, leading to an interaction.
Critically, grazing also affected the structural complexity of algae. Because of its complex ramified structure, the habitable volume of a tuft of macroalga is not equivalent to the three-dimensional space in which it resides. Instead, it is defined by the complex spacing between thalli, and its fractal dimension, D, lying between D = 2 and D = 3, is a measure of the degree to which space is filled. The disproportionate increase in surface area with decreasing scales results in more usable space for the associated epifauna (Gee & Warwick, 1994a,1994bGunnarsson, 1992;Lawton, 1986;Morse et al., 1985;Shorrocks et al., 1991).
Algae at the study sites are subject to grazing not only by benthic grazers but also by herbivorous fish (e.g., Götz et al., 2009;Heemstra & Heemstra, 2004)

| Epifauna
The community structure of epifaunal communities, based on species abundances, showed no effects of either upwelling or treatment, but did show a significant effect of site, with all sites differing from one another. This appears to have been a geographic effect as sites that were closer together were more similar. Importantly, while total epifaunal abundance responded only to the effects of site, when the data were normalized for habitat availability there were significant treatment effects. These effects indicated a response of the epifauna to indirect effects on their habitat, but not to the direct effects of predation. The important result here was the contrast between treatment effects on epifaunal abundances when these were normalized to algal cover ( habitat to abundant and diverse epifauna (Reynolds et al., 2014;Williams et al., 2013). Invertebrate grazers relying on macroalgae as food have strong direct effects on algae, including altering not only their abundance but also their structure (Cook et al., 2011;Reynolds et al., 2014). In our experiments, grazing had a signif-

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflict of interest. Writing-review & editing (equal).

DATA AVA I L A B I L I T Y S TAT E M E N T
The data sets generated and/or analyzed during the current study