Similar Trait Structure and Vulnerability in Pelagic Fish Faunas on Two Remote Islands

The link between biodiversity and ecosystem functioning has been the topic of considerable research, but it remains unclear how biodiversity decline is compromising ecosystem functionality, particularly in the pelagic realm. Here, we explore how pelagic sh species diversity relates to functional diversity by sampling two locations, which, on the basis of biogeography, environmental conditions, and human pressures, were expected to host pronounced differences in species composition and abundances, and therefore functionality. Strings of ve drifting mid-water Baited Remote Underwater Video Systems were used to survey pelagic vertebrate diversity and abundance in two isolated oceanic island systems, the Malpelo Fauna and Flora Sanctuary – a large, 25-year-old marine protected area – and an unprotected area in Cape Verde. Functional diversity, which offers insight into a community’s resilience against disturbance, was analysed using six key functional traits of marine shes. Cape Verde showed high overall abundance (Total MaxN 873) and low biomass (3,559 kg), with a predominance of smaller shes. Malpelo showed high biomass (7,839 kg) but lower abundance (Total MaxN 465), with a predominance of large species. Species and functional diversity were marginally different between locations. Multivariate analysis of species relative abundances showed signicant divergence between locations, although community functional traits overlapped strongly, suggesting that both communities share a similar structure and vulnerability. The existence of a common functional ‘backbone’ in diverging species communities across the oceans, under different productivity regimes, and under different protection levels, suggests that although pelagic communities may differ considerably in terms of species composition, this does not translate into a differing functional structure and resilience potential. Whether this vulnerability is a common feature of pelagic communities and how this contrasts with benthic systems warrants further research.


Introduction
The pelagic realm is the world's largest habitat, covering 71% of Earth's surface (1368.10 6 km 2 ), and providing millions of tons of sh biomass annually (Chassot et al. 2010). Pelagic systems have a massive in uence on global nutrient cycling, food production, and climate change, each of which may be in uenced by the abundance and composition of pelagic communities (Duffy & Stachowicz 2006, Sala et al. 2021). Yet, the vast pelagic area with dynamic environmental conditions (Breitburg et al. 2018)) and heterogeneous distribution of animals (Denderen et al. 2018) make it di cult to monitor and understand pelagic faunas (Briscoe et al. 2016), challenging the design of marine protected areas (Sala et al. 2021) and the appropriate management of sheries (Pons et al. 2018).
Pelagic ecosystems hold unique characteristics compared to the benthos, with likely implication for vulnerability and resilience. While pelagic diversity is considered relatively low compared with that of demersal sh communities (Tittensor et al. 2010), pelagic communities may show high local diversity, as a function of biogeography, bathymetry, and productivity (Bouchet et al. 2020). Oceanic islands, banks, and seamounts are notable hotspots of species richness (Letessier et al. 2019), which predictably aggregate sh biomass, including mobile predators, by acting as navigation points and reliable feeding grounds (Hosegood et al. 2019). Because prey density in the open ocean is generally low, predators need to forage over a wide range to ful l their energetic requirements, leading to resource translocation from other habitats and promoting food web connectivity (Huepel et al. 2014).
Ecosystem functioning and resilience are increasingly assessed using trait-based analyses (McLean et al. 2019), an approach that determines which functional roles are being lled and by which species (Mouillot et al. 2013). At the community level, functional traits explain differences of vulnerability between species: for example, predators and large individuals may be preferentially targeted by shing (Mbaru et al. 2020).
In the face of increasing sheries pressure and environmental change, a trait-based approach may capture novel aspects of pelagic community vulnerability missed by classic community-based analyses. However, it remains unclear to which degree different pelagic systems are similarly vulnerable, due in part because of limited pelagic sampling methodologies (Letessier et al. 2017).
Here, we use mid-water Baited Remote Underwater Video Systems (BRUVS, Bouchet & Meeuwig 2015) to assess how pelagic sh communities differ taxonomically and functionally. Our study took place in Malpelo Island and in Cape Verde, two remote tropical islands located in different biogeographical provinces (Kulbicki et al., 2013) which are exposed to contrasting levels of both environmental conditions and human pressures. We hypothesise that species community differences in taxonomy and relative abundance at each location may lead to trait dissimilarity, with implication for resilience and vulnerability.
Our objectives were to 1) describe and contrast species communities on each island using mid-water BRUVS and multivariate analyses, and 2) assess the extent to which any species dissimilarity translates into functional dissimilarity, using multivariate trait-based analyses, and then (3) make general inference concerning pelagic vulnerability.

Study locations
Our study took place at Malpelo Island and Cape Verde (Fig. 1A). Malpelo is a remote oceanic island located in the Tropical Eastern Paci c, approximately 500 km west of Buenaventura, Colombia. The interaction of multiple seasonal currents results in distinct cold and warm water seasons at Malpelo (Bessudo et al. 2011). Sampling took place during the cold season, which occurs between January and April and is characterized by a shallow thermocline -around 15 m depth -decreased visibility due to high primary production caused by upwelling, and an average sea surface temperature of about 25° C (Soler et al. 2013). An area of 8,575 km² surrounding Malpelo is designated as the Malpelo Flora and Substantial IUU shing also contributes to overall sheries landings (Medina et al. 2015). Pelagic target species include tuna, bill sh and mackerel scad, with sharks as important bycatch (Santos et al. 2013).

Sampling protocol
The pelagic community in each location was assessed using drifting mid-water BRUVS (Bouchet & Meeuwig 2015). Each rig consisted of a metal frame with two GoPro cameras in underwater housings mounted on a bar perpendicular to an arm supporting a bait canister lled with 1 kg of crushed sh (tuna and mackerel). The two cameras were intended to be used for stereo measurements, but could not be calibrated in the eld, so only footage from the right-hand camera of each rig was used for analysis. The rigs were suspended from buoys at a depth of 10 m ( We opted for this approach as 200 m separation between rigs is probably not su cient to guarantee independence, certainly not for large sharks, which can cover more than this distance during the 2-hour soak time. In the absence of stereo measurements, the biomass of each species was computed using common lengths and Bayesian length-weight coe cients available from FishBase and scaled by abundance (Table S1). These biomass estimates are speculative since they are based on calculations from FishBase rather than true measurements. The conclusions on biomass patterns from this study are therefore rough estimates. Future studies should include stereo measurements to more accurately assess patterns in sh biomass.
Total abundance and biomass by deployment or site were analysed using PERMANOVA. The differences between species communities at each site were tested with an analysis of similarities (ANOSIM) on Bray-Curtis dissimilarity between all pairs of communities. A similarity percentage (SIMPER) analysis using Bray-Curtis dissimilarity and 1,000 permutations were used to compare species groups by site. Species community abundance and biomass were illustrated using non-metric multidimensional scaling (NMDS).  2014), using custom R scripts by Villéger available online (http://villeger.sebastien.free.fr/Rscripts.html). A dissimilarity matrix quantifying the functional distance between species was computed using Gower's distance, which is able to accommodate categorical traits (Maire et al. 2015). A Principal Coordinates Analysis (PCoA) was performed with this matrix, and the mean squared deviation (mSD) was used to select the best quality functional space. The mSD quanti es the "mean squared deviation between the initial functional distance and the scaled distance in the functional space" and -when using Gower's distance -ranges from 0 to 1, the closest value to zero indicating the most robust functional space (Maire et al. 2015). The four-dimensional space -or the space using the rst four axes of the PCoa -is typically the highest quality space (Maire et al. 2015). Our four-dimensional space had an mSD of 0.0026 (Fig. S1).
Three complementary functional diversity indices were calculated using species position in the multidimensional space: functional richness (FRic), functional evenness (FEve), and functional divergence (FDiv) (Villéger et al. 2008). These metrics are favoured for functional diversity analysis because they can work with multiple traits. FRic measures how much functional space is lled by a species community and accounts only for species position in the multidimensional range. FEVe and FDiv account for species abundance and measure regularity and variance within the occupied space, respectively (Villéger et al. 2008). To further explore functional overlap between communities, functional beta-diversity and its percentage due to species turnover were computed using Jaccard's dissimilarity index (Villéger et al. 2013). All analyses were performed in R statistical software (R Core Team 2018).
In contrast with expectations, communities from the two locations lled the same amount of the functional space (functional richness Malpelo: 0.542, Cape Verde: 0.452), and species abundance were similarly distributed in the lled space (functional divergence: Malpelo: 0.833, Cape Verde 0.913) in both locations. While the volume of the overall space occupied by each location was relatively high, the majority of the species were densely distributed in one area of the space, displaying low functional evenness (0.425 for Cape Verde, and 0.346 for Malpelo). The rst two dimensions showed most species clustered on the right-hand side (Fig. 3C), de ned by mobility, whereas the outliers on the left side of the space had a more restricted range (Fig. S2A). Highly mobile species, clustered into two groups, one characterised by purely pelagic species of very small, and medium size, and with a planktivorous or omnivorous diet and one by large or very large, benthopelagic species living solitary or in small groups, and with either a piscivorous or invertivorous diet. The third and fourth dimensions of the functional space showed some of the same strati cation observed on the rst two dimensions. However, the species which diverged from the main group were planktivores and invertivores which fed on mobile prey (Fig. 3D), with loose grouping according to different daily activity patterns (Fig. S2B).
Functional divergence and functional evenness are indices of the variance and regularity of species' distribution within the functional space, respectively, weighted by abundance (Villéger et al. 2008). A highly divergent and minimally even community is one in which some functional roles are much better represented and insured than others, which leaves points of exposure to disturbance, particularly whenas was the case here -dominant and common species are sensitive (McLean et al. 2019). Two of the most abundant species observed at either location are currently on the IUCN Red List Red, namely yellow n tuna (Near Threatened) and Atlantic horse mackerel (Vulnerable; IUCN 2020). Furthermore, ubiquitous top predators like the silky shark and scalloped hammerhead, essential to ecosystem functioning, are Red Listed as Vulnerable and Critically Endangered, respectively (IUCN 2020).
Overall functional beta-diversity between the locations was 0.6, to which species turnover contributed 90.3%. The remaining proportion arose from distinct trait combinations, often between confamilials like the trigger shes (mobile in Malpelo, and reef-associated in Cape Verde), surgeon shes (large in Malpelo and small in Cape Verde) and jacks (large and piscivorous in Malpelo, and medium and invertivore in Cape Verde). Fishing remains the primary threat to pelagic elasmobranchs and teleosts (Pacoureau et al 2021). Although the bulk of the functional space was similar between locations, the presence of such unique trait combinations suggests that some nuanced differences in sensitivity may still render certain locations more or less resilient (Villéger et al. 2013). Future research should aim to identify pelagic systems which overperform compared to expectations, in order to identify unique resilient traits associated with either positive or negative ecosystem outcomes (Cinner et al., 2016).
Our survey, although limited in spatio-temporal scale, included different biogeographical provinces and covers contrasting ends of environmental and human pressure gradients. It presents evidence that pelagic ecosystems may share a common 'backbone' of functional traits related to mobility and predatory diet. Such a backbone of 21 common traits has already been documented to exist within global reef ecosystems (McLean et al. 2021). Since ecological disturbance is likely to affect species with identical functional traits in similar ways (Mouillot et al. 2013), low trait diversity within pelagic ecosystems may make the pelagic faunas particularly vulnerable to disturbance. We propose that further study speci cally aims to determine whether this is a universal feature of the pelagic realm, and the degree to which it may affect the resilience of mid-water communities. Sampling locations and deployment sites. Sampling locations (A) and BRUVS deployment sites in Malpelo (B) and Cape Verde (C, D). Line segments indicate BRUVS trajectories and are color-coded by string, with each line representing an individual rig. Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.