Fish communities associated with cold-water corals vary with depth and substratum type

Understanding the processes that drive the distribution patterns of organisms and the scales over which these processes operate are vital when considering the effective management of species with high commercial or conservation value. In the deep sea, the importance of scleractinian cold-water corals (CWCs) to fish has been the focus of several studies but their role remains unclear. We propose this may be due to the confounding effects of multiple drivers operating over multiple spatial scales. The aims of this study were to investigate the role of CWCs in shaping fish community structure and individual species-habitat associations across four spatial scales in the NE Atlantic ranging from “regions” (separated by > 500 km) to “ substratum types ” (contiguous). Demersal fish and substratum types were quantified from three regions: Logachev Mounds, Rockall Bank and Hebrides Terrace Seamount (HTS). PERMANOVA analyses showed significant differences in community composition between all regions which were most likely caused by differences in depths. Within regions, significant variation in community composition was recorded at scales of c. 20 – 3500 m. CWCs supported significantly different fish communities to non-CWC substrata at Rockall Bank, Logachev and the HTS. Single-species analyses using generalised linear mixed models showed that Sebastes sp. was strongly associated with CWCs at Rockall Bank and that Neocyttus helgae was more likely to occur in CWCs at the HTS. Depth had a significant effect on several other fish species. The results of this study suggest that the importance of CWCs to fish is species-specific and depends on the broader spatial context in which the substratum is found. The precautionary approach would be to assume that CWCs are important for associated fish, but must acknowledge that CWCs in different depths will not provide redundancy or replication within spatially-managed conservation networks.


Introduction
Understanding how fish are distributed across marine landscapes is vital in establishing effective management strategies for their conservation and sustainable use. This is particularly true where management is to be largely based on spatially explicit management tools (e.g. Marine Protected Areas (MPAs); FAO, 2007). The deep sea is one such environment, with management measures increasingly targeted towards identifying and CWCs on the overall fish community, though it was suggested that some taxa may use CWCs preferentially at different life stages. In the NW Atlantic, Auster (2005) found that coral substrata in the Gulf of Maine were functionally indistinguishable from substrata created by other large epifauna and did not support a distinct fish assemblage. Baker et al. (2012) examined fish abundance and community composition in three canyons in the Grand Banks region, but failed to find any association between fish abundance or community composition and CWCs, instead citing depth as the major influence. Stone (2006) noted that apparent associations could arise because certain fish and "habitat-forming" fauna share a preference for similar substrata leading to covariance which may be difficult to separate. The studies considered here include a diverse range of methodologies and taxa and cover a wide geographic range, but when taken together suggest that the distributions of fish within CWC areas may be influenced by a range of processes operating across multiple scales of organisation.
The importance of scale in ecological studies is well known (e.g. Levin, 1992;Chave, 2013).  Ross and Quattrini (2009) determined that faunal associations at the Blake Plateau were driven primarily by depth and habitat structure over regional scales (700 km), though the nature of these relationships varied between sites. At fine scales, Quattrini et al. (2012) determined that other habitat characteristics were important to distributions of fish at the Blake Plateau, and their importance was specific to particular fish species. Linking fine-scale variability in habitat diversity and habitat-use patterns to broader scales that are appropriate for management use is likely to be important in understanding the high variability observed in fish associations with CWCs to date. However, the influence of multiple spatial scales has not yet been examined within a single study, which may lead to difficulties in extrapolating from one study to another due to differences in methodologies and temporal variation.
The aims of the present study were to examine the importance of CWCs in shaping the distribution patterns of demersal fish populations and communities and to determine how they may be influenced by the scale at which the analysis is conducted. The aims are addressed using opportunistically-collected ROV video footage from the NE Atlantic collected over four nested spatial scales and the data are used to provide recommendations for future management of deep-sea fish.

Study Sites
The distributions of fish were studied in three regions of the NE Atlantic ( Figure 1

Materials and Methods
Data on the demersal fish were collated from opportunistically-collected high-definition video footage captured during research cruise JC073 (Roberts, 2013) using an Insite Mini Zeus camera mounted on the ROV Holland I. In total, 17 ROV transects provided 27 hours 7 minutes of useable video footage covering a total linear distance of 17.9 km ( Figure 2, Table   1). Additional metadata for each transect are provided in Supplement A.
The study area was subdivided according to four nested spatial scales. "Regions" were the broadest scale (c. 175 km -540 km), and contained a number of "reefs" (5.5 km -49.5 km).
Reefs in turn contained a number of "transects" (containing footage from one ROV survey dive; 20 m -3400 m) and each transect contained contiguous "substratum patches" (hereafter simply referred to as "patches"). These categories should be considered approximations of spatial scale, as they varied between regions. Notably, the HTS did not contain "reefs" and the distances between transects were greater than in other regions (15. Each transect was initially reviewed by one of two observers to identify the locations of fish fauna and changes in substrate type. Transects were assigned to an observer at random and analysed in a randomised order. Footage was only analysed when the ROV was moving over the seafloor at an approximately steady speed and direction, and when the camera was fully zoomed-out and stable. Footage was excluded where poor visibility prevented detection of the fish fauna, and from periods when the ROV was stationary, moving erratically, or was engaged in other activities. Only transects containing more than five minutes of useable footage were processed. All useable footage was then reviewed and transects divided into discrete patches. The start and end times of each patch were recorded. Each transect was reviewed a second time and the fish fauna were counted and identified to the highest possible taxonomic resolution based on morphological and behavioural characteristics, following Hureau (1996). Individuals that could not be formally identified to species but that were morphologically distinct from the other taxa were classified as distinct morphotypes (e.g. "Macrouridae sp. 1"). Individuals that could not be identified were classed as "indeterminate species" and excluded from analysis. The time at which each fish was first observed was recorded. Finally, all substratum classifications and species identities were reviewed to remove observer bias.
Time, depth and position of the ROV over the seabed were recorded at two-second intervals using a USBL navigation sensor. The locations and lengths of each patch were calculated by cross-referencing their start and end times to the USBL data. Degrees latitude and longitude were converted to UTM (Northing and Easting) and combined with the depth measurements to describe the ROV's position in metres using an x, y, z grid system. Outliers were manually removed from the 3D position data and the remaining data smoothed using moving averages (N = 10 data points). Any small sections of data which remained erratic (i.e. where the distance travelled was unfeasibly high) were removed and substituted with mean data from neighbouring patches. Estimates of mean depth (m), Northing, Easting and survey speed (m min -1 ) and length (m) were calculated for each patch. The mean gradient ("slope") of the seafloor was estimated for each patch by dividing the depth range by the horizontal distance

Data Analysis
Fish community structure was analysed using non-metric multivariate comparisons of community composition within and between sample groups using patches as the sampling units. Since the three regions were spatially distinct from each other ( Figure 2) and did not always have the same nested structure (i.e. the HTS did not contain "reefs"), all analyses were conducted in two stages. The first examined the broad-scale effects of "region" on community structure, and the second stage examined the finer-scale variation within each region separately.
Multivariate analyses were conducted using PRIMER 6 software with PERMANOVA (Clarke and Gorley, 2006). Multivariate results were considered significant at p < 0.05.
Samples that contained no fish were excluded as they would strongly bias the results. Fish counts were standardised by patch length to control for differences in survey effort between different patches and produce an estimate of relative abundance (N m -1 ). While this approach does not account for fine-scale spatial autocorrelation between neighbouring patches, it will nonetheless allow us to examine general patterns of fish associations with CWCs over the total study area. The relative abundances were then multiplied by 1000 for ease of presentation and analysis. Scaling in this manner has no effect on the analytical outputs.
However, the abundances should not be extrapolated beyond the spatial limits of the present study as they may not be accurate over broader spatial scales. Finally, the data were squareroot transformed prior to analysis. Bray-Curtis similarity matrices were generated to analyse the relative abundance data, and Euclidian distance matrices for the environmental data. Six outliers, each containing a single individual from a unique species, were identified using nonmetric Multi-Dimensional Scaling (nMDS) and removed to avoid biasing the results.
PERMutational ANalysis Of VAriance (PERMANOVA; Anderson, 2001) was used to test the effects of substratum type and environmental variables on community composition. The effects of "region" were tested separately from the environmental variables, because depth and location covaried with region. For analyses within each region, substratum type was included as a fixed effect nested within "transect" (random effect), which was nested within "reef" (random effect) as appropriate. Environmental data were included as covariates.
Latitude and longitude were excluded from analyses conducted within regions, because they were not considered to be biologically meaningful at these spatial scales. In all cases, sequential (type I) sums of squares were used as appropriate for nested data with covariates, and environmental terms were included before substratum terms. Models were permuted 9999 times under a reduced model. Backwards model selection was used to produce the fitted model from the saturated model. Pairwise comparisons were used to identify where significant differences occurred between factor levels, using Monte-Carlo sampling if the number of unique permutations was too small to allow calculation of p-values by permutation. Where significant differences were identified, PERMDISP analysis (Anderson, 2006) was used to determine whether these differences could have been caused by differences in the multivariate dispersion of points rather than their location. SIMilarity PERcentages (SIMPER) analysis was used to determine which species contributed most to any significant results.
Within each region, differences in environmental variables between substrata were tested using linear models (LM) in R software (Version 3.1.0, R Core Team, 2014). The effects of substratum type and the other environmental variables were tested on the patch occupancy (PO; a binary response) and raw counts (N) of the dominant fish taxa using Generalised Linear Mixed Models (GLMMs). All samples were included in these analyses, including those that contained no fish. "Transect" was included as a random effect and "substratum type", "survey speed", "slope" (log e transformed), "reef" and "depth" were included as fixed effects as appropriate (Equation 1). "Reef" could not be included as a random effect as it contained too few levels to produce valid results (Bolker et al., 2009). "Patch length" was included as an offset term. Fish counts were modelled using packages "glmmadmb"

Results
Analysis of the useable footage revealed a total of 1949 identifiable fish (plus 80 indeterminate individuals) from 57 taxa ( The environmental characteristics of the three regions showed significant differences. All regions were spatially distinct from each other and occurred at different depths ( Table 1)

Community data
A total of 839 fish were identified from 16 taxa ( Table 2) Gadidae sp. 1 occurred in transitional substrata, though both were present in non-coral substrata. Differences between transects and reefs appeared to be driven primarily by differences in the relative abundances of common taxa, rather than by a different in species composition.

Individual Species Trends
Three species accounted for over 80% of the total fish at Rockall Bank: Gadidae sp. 1 Hard ground was significantly steeper, and soft sediment flatter, than other substrata (LM: F = 2.5, DF = 4, p < 0.05).

Community data
A total of 483 fish were identified from 25 taxa ( Table 2). Substratum type (PERMANOVA: Pseudo-F = 1.45, DF = 9, p < 0.05; Figure 5c) and depth (PERMANOVA: Pseudo-F = 5.62, p < 0.01) were found to significantly affect community composition. Significant differences in multivariate dispersion were detected between substratum types (PERMDISP: F = 3.68, p < 0.03) and significant variation was detected between transects (PERMANOVA: Pseudo-F = 8.51, DF = 2, p = 0.0001). SIMPER analysis suggested that differences between substrata were driven by a greater relative abundance of Neocyttus helgae (Oreosomatidae) over transitional than non-coral substrata, and fewer Lepidion eques over soft sediment than other substrata. The macrourids Coryphaenoides rupestris and C. labiatus appeared to be more abundant over soft sediments than other substrata. Indeterminate macrourids, halosaurs and anguillids also contributed to the differences between substratum types.
No variables had significant effects on the relative abundance of Neocyttus helgae, although it was significantly more likely to occur over transitional substrata than other types (GLMM (which included some gravid individuals) was closely associated with CWC reef substrata.
Links between Sebastes spp. and CWCs have been previously reported from the NE Atlantic Interpreting apparent species-habitat associations must be done with care. It is generally assumed that individuals will select a particular habitat to maximise their success (e.g. Orians  (Auster, 2005). Consequently, any assessment of a species' true preference will require detailed study and experimentation on the organism in question over appropriate temporal and spatial scales. This will prove challenging in the deep sea, but would be beneficial in developing a more complete understanding of the ecological importance of CWCs to deep-sea fish.
While a more detailed understanding of the importance of CWCs to fish may be desirable from a scientific perspective, any such studies are unlikely to produce results for some time.
From a management perspective, it may be more prudent in the short-term to adopt a precautionary approach to the spatial management of deep-water fish, and the results from the present study suggest some possible considerations. While the constraints of the survey methodology mean that the results of the present study should be interpreted cautiously, if MPAs are intended to protect a representative subset of the wider fish community, then our data suggest that they should take account of both broad-and fine-scale spatial drivers of community structure. This would require accounting for the broad-scale effects of depth by selecting an appropriate depth range and then accounting for fine-scale variability within that range by including a sufficient range of substratum types over the spatial scales at which community composition was observed to vary (in this case, at "intermediate" scales of c. 5 -

Conclusions
Our findings suggest that CWCs support different fish assemblages to non-CWC substrata, but that the precise composition of that assemblage is modified by the broader spatial context, including the effects of depth or the composition of the regional species pool for example.
Understanding how different drivers interact to affect the fish fauna across multiple spatial and temporal scales would allow a far greater understanding of the importance of CWCs to different fish and how this may be tied to their life-history traits. The maintenance of natural fish assemblages is nonetheless a valid conservation aim. The precautionary approach would be to assume that CWCs are important areas for the associated fish, and that this should be considered when designing future MPAs. For fish assemblages to be fully protected, MPAs will be needed that encompass both broad-and fine-scale variability by covering a suitable depth range and variety of substrata, including CWC and non-CWC areas. For those species which appear to associate strongly with CWCs (e.g. gravid Sebastes sp. 1 at Rockall Bank), it would be prudent to assume that such areas provide "essential habitats" and to manage them accordingly.