The present study was conducted in an open fish cage located at Igueste de San Andrés (NE Tenerife, coordinates 28°31’30”N 16°09’20”W). In the study area, two fish leases (ca. 10000 m² surface area) were situated >300 m off the coast and both leases consisted of ten 20-m-diameter cages (Fig. 1). Species cultured in the studied fish cages were gilt-head sea bream (Sparus aurata) (200 t) and sea bass (Dicentratus labrax) (150 t) from 2004 to 2007. From 2007 to 2009, 450 t were cultured in the studied fish leases (250 t of S. aurata (250 t) and 200 t of D. labrax). Fishes were fed both by hand and automatically using commercial, pelleted and extruded diets with an average nutrient content (on a dry weight basis) of 49% protein, 18.5% fat, 22% carbohydrate, 7.9% nitrogen and 1.08% phosphorus.

Each fish cage had a diameter of 20 m and reached a depth of 10 m. Fish cages were installed on seabeds at a depth of 25-28 m and their temperature throughout the year ranged from 18°C in January to 22°C in September. Currents had a mean value of 7-8 cm s^–1 in the water column (unpubl. data). There are no extensive seagrass meadows in the study area, only small patches of the seagrasses Cymodocea nodosa and Halophila decipiens. Rocky substrates are scarce in the subtidal but dominate in the intertidal of the study bay.

Total counts around cages were considered inappropriate as independent replicates due to obvious circling of cages by fishes. Instead, we conducted 5-min rapid visual counts (Kingsford and Battershill 1998) using SCUBA beginning at the centre of each sampling location and proceeding through the adjoining water mass. A total of six visual counts were conducted each time at each sampling location. Each count covered a volume of ca. 11250 m³ (15 m wide × 15 m deep × 50 m long) at each location. Visibility was high, varying from 20 to 50 m on days when counts were performed and sea surface remained constant at all locations during the sampling campaign. Each count was made with two divers. The first diver concentrated on estimating the abundance of the dominant species, which were counted in groups of 1, 2-5, 6-10, 11-30, 31-50, 51-100, 101-200, 201-500 and >500 to minimize error (Harmelin-Vivien et al. 1985). The second diver followed slightly behind the first and specifically looked for less obvious cryptic species and smaller individuals that may have been missed by the first diver. Counts of both divers were summed.

The monitoring programme was conducted over six years, from December 2004 to September 2009 during ten sampling campaigns (1 December 2004 [1st], April 2005 [2nd], August 2005 [3rd], December 2006 [4th], May 2007 [5th], December 2007 [6th], May 2008 [7th], December 2008 [8th], April 2009 [9th] and September 2009 [10th] at each sampling location). Five locations were sampled throughout the study period (2004-2009), two located beneath the fish cages (“Impact”, 25-28 m depth) and three over 500 m away that were not influenced by the cages (“control”, 20-30 m depth) (Fig. 1). All sampling sites (impact and controls) were located on sandy bare seabeds with similar grain-size characteristics.

Non-parametric multivariate techniques were used to compare fish assemblages from impact and control locations, as well as, sampling times throughout the study period. All multivariate analyses were performed using the PRIMER statistical package (Clarke and Gorley 2006). Data were fourth-root transformed to weight the contributions of scarce species in the similarity coefficient (Clarke 1993). Similarity matrices were calculated using the Bray-Curtis similarity coefficient (Clarke and Warwick 1994). Non-metric MDS (n-MDS) was used as the ordination method to represent differences in fish composition between sampling groups (impact and control) throughout the study period. Taxa that had more influence on similarities of both location groups (impact and control) and dissimilarities between these groups of assemblages were calculated using the similarity percentages (SIMPER) procedure (Clarke 1993) to identify species responsible for the differences between groups (impact vs control). Permutational multivariate analysis of variance (PERMANOVA) was applied to test for differences in fish assemblage structure, considering Year (2004-2009) and Impact (impact and control) as fixed factors.

Additionally, for analysing the temporal trend of bogue over the six-year period (2004-2009), bogue abundance was fitted by generalized additive models (GAM). This term includes any GLM estimated by quadratically penalized (possibly quasi-) likelihood maximization, using the MGCV package routines of R (www.r-project.org) (Wood 2011). The GAM regression technique consists of fitting smooth additive functions for each covariate included in the model structure. The smooth functions are linear combinations of a finite number of basis cubic spline functions, with the smoothness of the function estimated by minimizing the generalized cross-validation criterion (Wood 2011) that balances the goodness-of-fit and the smoothness of the functions. To avoid overfitting, we constrained the number of basic functions to at most three. We regressed each campaign variations of bogue abundances throughout the study period (2004-2009).

A total of 66672 fishes were observed during the study period (2004-2009) at the five sampling locations (impact and control). Bogue (Boops boops) was the dominant species (93.89%), followed by black seabream (Spondyliosoma cantharus; 1.85%) and white kingfish (Pseudocaranx dentex, 0.44%). The remaining species (36 taxa) were very scarce (<300 individuals). Species richness was higher at impact (13.2±1.06 taxa, mean±SE) than at control locations (8.22±1.5 taxa) (Table 1). The most common species in Impact locations were bogue (B. boops), black seabream (S. cantharus), white kingfish (P. dentex) and sharpnose puffer (Canthigaster capistratus). At control locations they were bogue (B. boops) and black seabream (S. cantharus). Differences between impact and control locations were more pronounced in terms of abundances since fishes were concentrated around fish cages (6207.6±1816.4 ind 11250 m^–3, mean±SE), approximately 13 times more than at control locations (456.9±257.8 ind 11250 m^–3) (Table 1). Bogue was present only at control locations of three campaigns throughout the study period, ranging in dominance (contribution to the overall fish abundance) from 50.84% in December 2007 to 96.52% in May 2008. In the impact group, bogue was present in all sampling campaigns, with the exception of May 2007. Bogue dominance varied from 69.65% in December 2004 to 98.97% in April 2009 (Table 1). Shannon diversity showed the opposite pattern of fish abundances, with higher values at control (1.13±0.2) than at impact locations (0.54±0.13) (Table 1) because of the dominance of few species. Shannon diversity in the control group ranged from 0.30±0.11 in April 2009 to 2.09±1.18 in December 2008, whereas in the impact group it ranged from 0.16±0.09 in May 2008 to 1.17±0.67 in April 2005 (Table 1).

SIMPER analysis showed higher similarity at impact locations (32.57%) than at control locations (7.64%). A total of seven fish species were the most important contributors to the impact group similarity (Boops boops, Canthigaster capistrata, Bothus podas, Trachinus draco, Sphoeroides marmoratus, Dasyatis pastinaca and Taeniura grabata), though most of the contribution to the assemblage homogeneity was provided by the bogue (55.43%). This species overwhelmingly dominated the overall fish abundance and thus changed the whole assemblage structure. The similarity in the control group was mainly due to Trachinus draco, which was responsible for 27.60% of homogeneity, followed by Sphoeroides marmoratus (20.75%).

Bogue was the most important species for explaining the dissimilarity between the two groups (91.43% dissimilarity), with a contribution of 30.89%. The remaining fish were scarcely responsible for the dissimilarity, with low percentages per species (<10%). Bogue dominated fish assemblages at control locations (70% of abundance), but the percentage of dominance was lower than at impact sites (96% of abundance) and other species (e.g. black seabream (Spondyliosoma cantharus)) contributed significantly to the overall fish structure (Table 1).

Similarity in caged stations within each year ranged from 5.25% (2004) to 69.68% (2006). Bogue contributed consistently throughout the study period at caged stations from 48.58% (2007) to 97.85% (2008), with a mean contribution of 78.25%. The abundance variations of bogue were the main factors responsible for dissimilarities at caged stations among years, varying from 53.31% (2005-2006) to 96.89% (2004-2007).

Highly significant differences were obtained in the Year×Impact interaction (Table 2). Fish assemblage structure was significantly different between locations beneath fish cages and controls (Impact Pseudo-F=96.246, p=0.001), but these differences showed interannual variations (Year×Impact, Pseudo F=33.783, p=0.001) (Table 2, Fig. 2). Temporal variability was more pronounced at control locations (5 different groups) due to low fish abundances and aggregated distribution. Impact locations were included in one group, except in May 2007 (Fig. 2).

The analysis of the temporal trend of the bogue population by GAM showed an increasing pattern over the study period (Fig. 3), especially due to the high abundances of bogue in the last sampling years (2007-2009).