Spatio‐temporal turnover and drivers of bentho‐demersal community and food web structure in a high‐latitude marine ecosystem

Assess the spatial and temporal turnover of bentho‐demersal marine fauna by integrating ecological metrics at the community and food web levels and evaluate their main environmental and anthropogenic drivers.


| INTRODUC TI ON
Species distributions are constrained by environmental conditions and biotic interactions, shaping biogeographic patterns at the macroecological scale (Chase & Leibold, 2003;Soberón, 2007). In the case of strong environmental control on species distribution, species that co-exist within an ecological assemblage are likely to share similar ecological traits that enable them to survive under local environmental conditions (Keddy, 1992;Violle et al., 2007). This has notably been shown in the marine realm, where the trait composition of fish communities is spatially structured along environmental gradients, such as salinity and depth gradient (Beukhof et al., 2019;Mouillot et al., 2007;Pecuchet et al., 2016). The spatial and temporal variation, that is, turnover, in community taxonomic and trait composition is well studied across realms and taxa (e.g. Blowes et al., 2019;Magurran et al., 2015;Pecuchet et al., 2018;Soininen et al., 2018), and several studies have also investigated the topological turnover in food web properties resulting from changes in taxonomic composition (Albouy et al., 2019;Kortsch et al., 2021Kortsch et al., , 2019Olivier et al., 2019) and/or feeding preferences (e.g. rewiring [Olesen et al., 2011]). Whereas spatial turnover patterns of community and food web have been analysed separately, their joint study is largely unexplored. Yet, such a joint study could inform how community structure and functions influence food web properties across multiple-stressor gradients (Frelat et al., 2022;Tylianakis & Morris, 2017).
A food web represents the feeding interactions between organisms co-occurring in a specific location, and are as such arising from species distribution, and local community composition. With their myriad of interactions, food webs are inherently complex. This complexity can be summarized into analytical metrics, that is, food web properties, which are emerging from the configurations and strengths of feeding interactions between species (Delmas et al., 2019). Feeding interactions are largely mediated by the traits of a predator (consumer) and its prey (resource) .
The trait composition of a community is thus likely to influence the properties of the trophic network emerging from the feeding interactions within this community. As the trait composition of a community is varying along environmental gradients, it is thus expected that food web structure will also co-vary along these gradients (Pellissier et al., 2018). For example, in the marine environment, feeding interactions are highly structured by the predator and prey body size Laigle et al., 2018;Pecuchet et al., 2020), and the body size structure within ecological communities can vary along environmental gradients (e.g. fish size along depth gradients ). Analysing co-variation between community trait composition and the properties of ecological networks is thus a key step towards understandings the impact of functional trait diversity on food web properties and functions (Gravel et al., 2016).
The Barents Sea is a highly productive arcto-boreal ecosystem that sustains ecologically and commercially important fish and invertebrate populations (Haug et al., 2017). The Barents Sea benthodemersal fish and invertebrate communities are spatially structured and characterized by a biogeographic divide between boreal species, which are distributed in the southwest Atlantic-dominated waters, and Arctic species, which are distributed in the north and northeast Arctic-dominated waters Johannesen et al., 2017;Jørgensen et al., 2015). Food web properties in the Barents Sea are also structured along environmental gradients of sea temperature and ice-cover . The Barents Sea ecosystem has been strongly affected by fishing activities and more recently by other human activities such as oil and gas production and shipping (ICES, 2019). The ecosystem is changing fast, warming at a rate exceeding the global average, and with a diminishing sea ice extent (Smedsrud et al., 2013), and an associated decrease in salinity and stratification (Lind et al., 2018). These environmental alterations have led to a restructuring in species biomass and distribution, with a northward shift of both boreal and Arctic communities and increased biomass of boreal species in the historically Arctic part of the Barents Sea Ingvaldsen et al., 2021).
This rapid borealization may have profound consequences on the functioning and vulnerability of the Barents Sea ecosystem, as the boreal species share different traits and life-history strategies than their Arctic counterparts. For example, Arctic fish communities are characterized by the dominance of small benthivorous species while the boreal, northward-expending, species are characterized by large body size and piscivorous diet (Frainer et al., 2017). Due to their generalist diet, these species might affect the structure and vulnerability of the Arctic food web by increasing the connectivity between the benthic and pelagic compartments (Kortsch et al., 2015;Pecuchet et al., 2020). Community and food web spatial patterns in marine ecosystems, including the Barents Sea, have not yet been studied in an integrative way, but are studied in isolation using different spatial scales and focusing on different compartments and taxonomic resolutions. Yet, an integrative comparison of spatial patterns in community-and food web-level properties is needed to assess their co-variation and to promote an understanding of how food web structure might emerge from the composition at the community level. In addition, it is still unclear whether it is the same set of environmental and anthropogenic pressures that structure communityand food web-level properties.
In this study, we combine the information of a trait dataset and an extensive food web with bottom-trawl catch data from an ecosystem survey to calculate biomass-weighted community metrics and unweighted food web properties. We assess spatial and temporal changes in the community and food web properties of the bentho-demersal compartment of the Barents Sea (fish and K E Y W O R D S arcto-boreal ecosystem, Barents Sea, fish-benthos communities, fishing pressure, functional diversity patterns, marine food web, spatial turnover macrobenthic-invertebrates) and relate them to variations in environmental and fishing pressures. Specifically, we ask: (1) what are the spatial patterns and co-variation of the bentho-demersal community and food web properties, (2) are these patterns related to spatial patterns in the environment and trawling intensity, (3) have the community and food web properties changed in the Barents Sea during the last decade (2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017). Based on prior knowledge of the Barents Sea ecosystem and its recent dynamics (e.g. Eriksen et al., 2021;Fossheim et al., 2015;Frainer et al., 2017;Johannesen et al., 2017;Jørgensen et al., 2015;Kortsch et al., 2019), we expect that the communities and food web metrics are spatially structured along known environmental gradients in the Barents Sea, especially along the south-north gradient in temperature and sea ice-cover ( Figure 1). Spatial patterns of fishing are influenced by diverse social, economic, environmental and ecological factors (van Putten et al., 2012); here, we expect that trawling activities, which target specific bentho-demersal communities composed of commercial fish and invertebrate species, will, in turn, varies along this south-north ecological and sea ice-cover (i.e. accessibility) gradients. Temporally, we expect to see the most changes in the communities and food webs in the Arctic part of the Barents Sea, that is, we expect different temporal trends between boreal and Arctic ecosystems and that these changes are mainly due to changes in species biomass rather than species occurrence (i.e. larger changes in biomass-weighted metrics than unweighted food web metrics).  (Eriksen et al., 2018). Sampling was performed in August-September for the period 2009-2017. Each year demersal fish and megabenthos were sampled at spatially predetermined stations (~270) positioned along a regular grid (~65 km between stations) using Campelen 1800 bottom trawl at a mean towing speed of 3 knots during 15 mn equivalent to a towing distance of ~1.4 km. The mean vertical trawl opening is 25 m and the mesh size is 80 mm (stretched) in the front and 22 mm at the codend, allowing the capture and retention of small fish and megabenthos from the seabed. The catchability of each megabenthos species by the scientific bottom trawl will vary depending on the species length and body shape; we expect body sizes larger than the mesh size (22 mm) to be better retrieved inside the trawl. We, therefore, consider the sampling of megabenthos species to be semi-quantitative. This semi-quantitative approach allows us to compare values in space and time, and it has proven valuable in long-term monitoring (Degen et al., 2016;Jørgensen et al., 2022Jørgensen et al., , 2019. All individuals of fish and invertebrates caught in the trawl hauls were sorted and identified to the lowest taxonomic level possible, their wet biomass is weighted, and abundance was counted and standardized by unit swept area. Not all individuals were identified at the species level, but for simplicity, we will hereafter refer to these different entities, both species and higher taxonomic grouping, as taxa.

| Bentho-demersal communities
We divided the Barents Sea on an equal-area grid composed of 60 hexagonal cells of about 23,300 km 2 and the cell centres about 165 km apart. The grid was computed using the dggridR package (Barnes et al., 2018). To characterize the spatial patterns in bentho-demersal community composition, we did, for each grid cell and all years combined, sample-based effort standardization by assembling bentho-demersal communities from a random draw of F I G U R E 1 Location of the Barents Sea and (a) bottom temperature at the time of the Barents Sea ecosystem survey (August-September) and (b) number of days within a year with sea ice-cover averaged for the study period 2009-2017 10 megabenthos and demersal fish sampling stations. This effort standardization permits the elimination of possible bias due to differences in sampling effort between grid cells. We reiterated this random draw 100 times, and for each iteration taxa biomass and abundance were calculated as the sum of their biomass and abundance density (kg or number/haul) across the 10 randomly selected stations. This resulted in 100 bentho-demersal communities assembled for each grid cell using the same sampling effort (10 stations).
Finally, all bentho-demersal communities were filtered to only retain the taxa that are documented in the meta food web (metaweb, Planque et al., 2014, see below).

| Food web
The bentho-demersal food web was derived from an extensive meta food web (metaweb) documenting 2461 feeding links between 239 taxa that inhabit the Barents Sea (Planque et al., 2014), ranging from plankton to benthic invertebrates, fish, sea birds and marine mammals. A taxon can refer to various levels of taxonomic resolution, from species (e.g. Gadus morhua) to class (e.g. Hydrozoa), with this set of one or more species sharing the same resources and predators (Yodzis & Winemiller, 1999). The feeding links were documented from peer-reviewed literature, reports, personal communication with experts or inferred from taxonomically similar species and spanned the period from 1927 to 2012 (Planque et al., 2014). In addition, we used information on fish diet from the 2015 Norwegian-Russian "Year of the stomach" (Eriksen et al., 2021) to update and improve the bentho-demersal food web. Here, we are only studying the taxa that were recorded in the ecosystem survey bottom trawls; we thus did a subset of the metaweb to keep only the fish and the invertebrates sampled in the trawls (see species list in Table S1). For appropriate trophic level estimation, we retained the basal groups' phytoplankton and detritus, and the group zooplankton in the food web, and considered them present across the entire Barents Sea (i.e. present in all grid cells). The final bentho-demersal food web used in this study had 111 taxa and 1082 feeding links ( Figure S1). On average per grid cell, 99.3% (94.9%-100%) of megabenthos and 99.5% (98.7%-100%) of demersal fish individuals sampled were included in the food web ( Figure S2), albeit sometimes specified at a higher taxonomic level within the food web.

| Community and food web metrics
To investigate spatial patterns in community and food web composition, we calculated for each grid cell 18 community and food web metrics based on the assembled 100 bentho-demersal communities. These metrics can be calculated based on changes in taxonomic composition (presence/absence), taxa biomass, prey biomass in stomach contents (feeding preferences) or energy fluxes (Dunne et al., 2004;Gauzens et al., 2019;Kortsch et al., 2021). Each metric type can reveal different facets of community and food web spatial and temporal changes, and it is thus important to use multitype metrics to obtain a more comprehensive assessment (Kortsch et al., 2021;Olivier et al., 2019). The calculated food web properties were the number of taxa, number of feeding links, feeding link density, connectance, modularity, nestedness and average food chain length ( Table 1). These metrics were based on taxa presence-only (unweighted). Modularity and food chain length were calculated using the R package igraph (Csardi & Nepusz, 2006) and nestedness the R package vegan (Oksanen, 2017). To complement the food web properties, we calculated metrics to characterize the composition of the bentho-demersal communities: the total biomass, the biomass evenness, the biomass ratio between fish and megabenthos, the biomass-weighted mean and variability of trophic level, omnivory, generality and body size. Trophic level represents the feeding position of a taxon in the food web, here it goes from 1 (primary producers) to 4.2 (Greenland shark Somniosus microcephalus). Generality represents a taxon' number of documented preys in the metaweb, here, going from one (e.g. strict detritivorous) to 85 for Atlantic cod Gadus morhua. Omnivory indicates the extent to which a taxon feed on multiple trophic levels, here it goes from zero (e.g. strict detritivorous) to 0.83 for starfishes Ophiura sp. Body size, obtained from (Pecuchet et al., 2020), represents the size in centimetres of a taxon's longest body axis and was log-transformed due to right-skewed distribution. The values of trophic level, omnivory and generality were calculated for each taxon using the Barents Sea bentho-demersal metaweb and were thus kept constant through space and years (i.e. each taxon is assigned one value based on the metaweb). The trophic level and omnivory were calculated using the R package NetIndices (Kones et al., 2009). The food web-and community-level metrics were calculated using the same input data, that is, by using the 100 bentho-demersal communities assembled per grid cell that were filtered to only retain the taxa present in the metaweb. Biomass was log-transformed before calculating the biomass-weighted mean and variability indices due to right-skewed biomass distribution, and by doing so to avoid the indices only depicting spatial turnover of dominant and widely distributed fish species (such as Atlantic cod and haddock Melanogrammus aeglefinus). In addition, the bottom trawl gear has a higher catchability for fish than megabenthos due to the height-width of the trawl allowing not only fish from the seabed, but also semi-pelagic to enter the trawl, log-transforming the biomass permits to increase the importance of megabenthos species in the calculations of the biomass-weighted metrics. The ratio of fish-tomegabenthos biomass was calculated using absolute fish and megabenthos biomass, and the obtained ratio was then log-transformed to approximate symmetrical data distribution. Spatial maps of the ecological metrics can be found in (Figures S3-S5).

| Barents Sea environment
We selected 18 variables to characterize the Barents Sea environmental conditions and demersal fishing pressure. These variables were further split into three variable types that characterize the mean annual conditions, the intra-annual variability and the withingrid cell spatial variability; these variable types allow to describe, respectively, the mean conditions, the seasonality and the habitat heterogeneity experienced by the bentho-demersal fauna. This grouping was done to test whether different community and food web metrics tend to respond to different types of pressure variables. Notably, we expect that the biomass-weighted mean trait

Biomass-weighted mean and variability of traits
The mean and the variability of trophic level, generality, omnivory, and body size trait values in a community, weighted by species log-transformed biomass (Sonnier et al., 2010). The mean gives indication on the averaged trait values in a community, whereas the variability complements the mean indicator by characterizing the dispersion of the trait values within a community Total biomass Total biomass in the bentho-demersal communities Fish/Megabenthos biomass The log-transformed ratio between the biomass of fish and the biomass of megabenthos in the communities Biomass evenness Measure the evenness in the biomass distribution between species in the community. High values point towards communities with dominating species lower values communities with comparatively well spread biomass between species. Calculated using the Simpson's evenness index Food web metrics (unweighted) Number of taxa (TR) The total number of taxa in the food web, that is, taxa richness. Note that the values of this metric are much lower than the actual number of species because many species are grouped at higher taxonomic level (e.g. sponges, bryozoans, hydroids)

Number of links (N links )
The total number of feeding interactions in the food web Link density (LD) The average number of feeding interactions per taxa, calculated as N links /TR (Dunne et al., 2002) Connectance The proportion of realized feeding links in comparison with all the potential feeding links, and calculated as N links /TR 2 . It characterizes network complexity and is linked to the robustness of food webs to perturbations (Dunne et al., 2002) Modularity Measure whether the food web can be partitioned into distinct groups of interacting species, or modules. Low modularity characterizes food webs that are not organized around distinct groups, whereas high modularity characterizes food web that are highly compartmented (Stouffer & Bascompte, 2011) Nestedness Measure how much the feeding interactions are nested in the food web, for example, a food web is highly nested when specialist consumer diets are subset of generalist diets (Ulrich et al., 2009).

High nestedness indicates low trophic complementarity
Average food web chain length On average, the minimum number of feeding links between two taxa randomly picked from the food web. It influences community structure and ecosystem functions and is linked to food web robustness to perturbations (Post, 2002) the standard deviation across the study period. The within-grid cell spatial variability was calculated by computing the spatial standard deviation per month and year and then averaging the standard deviation across the 3 years. Mean depth was calculated based on the bottom-trawl sampling depth and thus match the biological sample locations. Depth variability per grid cell was calculated as the standard deviation of the sampling depth. Ice-cover (number of days with >15% daily sea ice concentration) was extracted for each year at the sampling station from the National Snow and Ice Data Center (Cavalieri et al., 1996), and annual ice-cover was proxied by the average total number of days within a year with ice-cover ( Figure 1).  (Davies et al., 2004) and grouped sediment types into seven categories, among which only three were retained in the analysis as they covered a proportion of at least 10% on average across all grid cells, that is, "mixed sediments," "mud, clay and sandy mud" and "sand, muddy sand." We kept only fishing activities from vessels categorized as "trawlers" (including both bottom-and mid-water) and "dredge fishing." To calculate the annual mean, we first summed the fishing hours per year of all selected vessels for each hexagonal grid cell used in this study (see Figure S6) and then we averaged the fishing hours across the 3 years. To calculate the intra-annual variability of trawling, we first summed the trawling hours per month and year, then computed the standard deviation of the monthly values per year and then averaged the standard deviation across the 3 years. The within-grid cell spatial variability in the trawling was calculated by first summing the fishing hours per year per 1-km 2 , then computing the spatial standard deviation per year and finally averaging the standard deviation across the 3 years. The variables "number of days with ice-cover" and "trawling intensity" were log-transformed due to right-skewed distribution.

| Statistical analysis
2.5.1 | Cluster analysis of community and food web metrics To describe the spatial patterns of the bentho-demersal community and food web properties, we clustered the different grid cells to identify groups of grid cells with similar community and food web properties. We clustered the grid cells based on their ecological properties using an archetype analysis (AA, (Cutler & Breiman, 1994). AA focuses on identifying "freak" observations rather than the mean or median observations of a cluster. Here, it identifies grid cells where the bentho-demersal community and food web properties are differing the most. The 18 ecological metrics were normalized (i.e. to a mean of 0 and a variance of 1) before analysis to ensure equal weights in the AA. We tested a range of fixed number of archetypes k = 1, 2, 3, … 10, and the residual sum of squares (RSS) of 99 iterations was calculated for each k using the package "archetypes" in R (Eugster & Leisch, 2011). We used the "elbow criterion" to select the optimal number of clusters permitting minimizing the RSS while minimizing the number of archetypes ( Figure S7). This is done visually by assessing the number corresponding to a substantial drop in the RSS. Each grid cell is then characterized as a combination of k archetypes instead of being assigned to a specific archetype, that is, each grid cell is characterized by k proportions based on the proximity of the values of the grid cell ecological metrics to the values of the archetypes. Each archetype was exemplified by the grid cell, which resembled the most the archetype (proportion close to 1). We represented the food web at each of these archetype grid cells by displaying taxon biomass, feeding links and taxon groups (fish or megabenthos). For simplicity, we only displayed the taxa that were occurring in more than half of the 100 assembled communities (present in more than 50 communities).

| Response to the environment redundancy analysis
To investigate potential spatial co-variations within, and between, the ecological metrics and the environmental and anthropogenic pressures, we performed a redundancy analysis (RDA). The RDA is a multivariate method that has been shown to give a good overview of community trait response to environmental gradients (Kleyer et al., 2012). The RDA permits assessing linear relationships between two matrices, here between the response matrix of 18 community and food web metrics and the explanatory matrix of 12 environmental and trawling intensity variables. We used a hierarchical partitioning to estimate the importance of each explanatory variable and group of variables (spatial, seasonal, average) and how well they explained the spatial variability of the ecological metrics. Hierarchical partitioning calculates the variable importance from all subset models, leading to an unordered assessment of importance (Lai et al., 2022). All the analyses were performed with the R packages "vegan" and "rdacca.

| RE SULTS
Following the archetype analysis on the ecological metrics, we selected four archetypes as a compromise between the amount of variability explained and complexity (elbow criterion, Figure S7).
The clusters associated with the archetypes separate the Barents (c-f) Representation of the four food webs in each archetype grid cell, only the taxa present in at least 50% of the 100 assembled communities per grid cell are represented. Each dot corresponds to a taxon and the taxon name is labelled using the first three letters of their taxonomic classification, if classified at the species level the first three letters correspond to the genus name and the following three letters to the species name. The dots and labels are coloured orange for megabenthos, blue for fishes and grey for detritus, phytoplankton, and zooplankton. The grey lines represent the documented feeding links between taxa. The taxa are placed on the y-axis based on their trophic level and on the x-axis semi-randomly (grouping species within three clusters based on their shared feeding interactions and an added random value). The size of the taxa labels represents their relative biomass in the bentho-demersal community (small to large represents low to high biomass, respectively). The species with the highest biomass in the archetype food webs are: ACT_G_SP = Actinaria sp., ANA_DEN = Anarhichas denticulatus, BOR_SAI=Boreogadus saida,

| DISCUSS ION
Biogeographical patterns in marine fauna are seldom investigated by integrating across organizational levels and organism groups.
Here, for the first time, we integrated spatial patterns in the community composition and food web properties of fish and benthic invertebrates in the Barents Sea and assessed their relations to environmental and fishing pressures. This integration permitted us to obtain a more holistic understanding of how environmental conditions can constrain community composition, which in its turn can influence food web structure. For example, the southwestnortheast gradient in sea bottom temperature and sea ice-cover shapes the community composition with higher megabenthos biomass and smaller body size in the northeast, which results in a shorter food chain and higher modularity of the food web. We found that the annual mean in the environmental pressures, and notably sea bottom and surface temperature and depth, explained most of the spatial variability in the ecological metrics. However, spatial and temporal variability also co-varied with some ecological metrics. Notably, the variability in body size and omnivory within the community was higher in areas with high spatial heterogeneity in surface temperature. In these areas, located in the central Barents Sea, the Arctic and Atlantic water masses mix. This reveals that more heterogenous habitats can have higher functional diversity by diversifying the ecological niches available for species.
The common study of bentho-demersal community and food web metrics allowed us to partition the Barents Sea into four sub-regions with similar ecological characteristics, that is, southwest, south-east, The Barents Sea is characterized by a strong environmental southwest-northeast gradient linked to warm Atlantic water F I G U R E 5 Redundancy analysis showing (a) loadings of the bentho-demersal community and food web metrics (grey crosses and labelled in dark for community-and grey for food web metrics), and the set of environmental variables (grey arrows and italics blue labels) along the third and fourth principal axis of the RDA. The spatial pattern of the (b) third and (c) fourth principal axis of variation F I G U R E 6 Temporal changes of selected unweighted food web (a, b) and biomass-weighted community (c,d) metrics and environmental variables (e-g) for each archetype cluster. Only five grid cells per clusters, including the archetypes, were used to calculate the ecological metrics (grid cells with intense colours in h). (a-d) the bold coloured lines represent the metric' median value. The dark shaded areas represent the 50% credible interval and the light shaded areas the 90% credible interval. The environmental variables (e-g) were calculated as the average value of each cluster's five grid cells. Note that in 2014 no ecological metrics data is available for the two western clusters (red and purple) and that in 2012 the sampling for megabenthos was irregular in the two eastern clusters (green and blue) due to the use of a different gear. The years 2012 and 2016 are highlighted by a vertical dashed line as these years were characterized by extreme environmental conditions in the Barents Sea, with very warm sea bottom temperature and very low ice-cover, as identified in (Husson et al., 2022;Mohamed et al., 2022) entering the southwestern Barents Sea and bringing boreal conditions in the southern areas and to opposite cold Arctic water entering the northeastern Barents Sea and bringing Arctic conditions in the northern areas. Our results showed a similar biogeographical pattern for the bentho-demersal communities: the southwestern area is dominated by boreal species, whereas the northeastern area is dominated by Arctic species with the addition of seasonal migrant boreal species, such as Atlantic cod or capelin Mallotus villosus, which migrate northward when the ice retracts (Olsen et al., 2010). In the central area where Atlantic and Arctic water mix, both Arctic and boreal species co-occur. This turnover in taxonomic composition was also documented for demersal fish (Johannesen et al., 2012), benthos (Jørgensen et al., 2015) and their spatial co-variation (Johannesen et al., 2017). The secondary production of Barents Sea megabenthos also follows this spatial pattern, with the highest production in the northeast seasonally ice-covered areas and lowest in the permanently ice-free southwestern areas (Degen et al., 2016). In this study, as expected, the main spatial axis of the variability of the community traits and food web followed also a southwest-northeast gradient. Fish biomass and diversity were, in general, higher in the southwestern communities, whereas benthos biomass and diversity were higher in the northern communities. Community body size, trophic level, omnivory and the food chain length decreased along the southwest-northeast axis. This gradient displays the Arctic-boreal biogeographic divide between communities in the southwestern Barents Sea dominated by boreal species, such as Atlantic cod, haddock, saithe and the large Geodia sponge grounds, and the northeast communities dominated by Arctic species, such as polar cod, snow crab, poriferas and Arctic brittle stars. Compared with the Arctic fish species, the boreal fish species are generally larger, more generalist and have a higher trophic level , which is reflected in these community mean trait values and which also appear to affect the food web structure (co-variation with modularity and food chain length).
Marine food webs are largely structured by taxa traits such as body size Pecuchet et al., 2020), the differing trait composition in the Arctic and boreal communities, along with taxa richness, seem to impact the topology of the benthodemersal food web, with higher modularity in the northeastern Arctic food web and higher food chain length in the southwest- robustness and invasibility) of food webs and are thereby valuable indicators for ecosystem approaches to management and conservation that account for systemic risk (Landi et al., 2018;Levin & Lubchenco, 2008).
The spatial patterns in trait variability within the community permitted distinguished bentho-demersal communities in the northwest, central and south-east Barents Sea to communities north and south of this diagonal. The communities in this area had higher variability in the body size and omnivory values. In addition, the number of taxa and the average number of feeding links per taxa (link density) were, generally, also higher in this cluster. Within this central area of the Barents Sea, the warmer and more saline Atlantic waters mix with the fresher and colder Arctic waters (Oziel et al., 2016). The biological communities are dominated by boreal and Arctic species south and north of this area, respectively .
These two communities with different body size and feeding traits overlap in the mix-waters of the central Barents Sea, which is reflected in the increased trait variability observed in these central communities. This shows the importance of studying not only average community trait but also the variability in the trait composition within the community when assessing spatial variation in community composition. Studying the variability can help inform where two ecotypes meet, here boreal and Arctic communities, and can be used as an early-warning signal of ecosystem change (Frainer et al., 2021).
Connectance and nestedness were the highest in the shallow south-eastern communities and co-varied to a certain extent with variability in generality. The combination of higher connectance and higher nestedness implies that there are more realized feeding interactions in these communities and that these interactions tend to overlap across species. The high nestedness and variability in generality indicate that both generalist and specialist species co-occur in these communities and that the diet of the specialist species is a subset of the diet of the generalist species. High nestedness suggests that several taxa share a similar trophic niche, and thus high trophic redundancy in the system.
High trophic redundancy is expected to lead to a more resilient food web as the redundancy buffers the ecosystem against species loss caused by external perturbations (Sanders et al., 2018).
However, in the case of the south-eastern Barents Sea, the high nestedness is associated with low taxonomic richness and biomass evenness which results in high trophic redundancy but low trophic niche partitioning, that is, low trophic complementarity with few trophic groups (Poisot et al., 2013). Networks with low trophic complementarity might result in non-optimal resource use and can impact ecosystem functions such as consumer biomass production (Poisot et al., 2013). This observed high nestedness and low trophic complementarity might be due to differing environmental conditions linked to less varied topographic and oceanographic conditions in the area, shallow (less 100 m) and sand-dominated sediments, which could limit bentho-demersal community to species with a preference for this habitat, but might also be due to the high biomass of the invasive red king crab in these areas (Anisimova et al., 2005;Falk-Petersen et al., 2011;Oug et al., 2018).
The spread of non-native species in the marine realm is one of the major drivers of ecosystem change with impacts on community dynamics and ecosystem function and services (Molnar et al., 2008;Walsh et al., 2016).  (Anisimova et al., 2005). The generalist invasive snow crab and red king crab have top-down effects on the benthic communities, as they reduce the biomass of large longlived species which benefit opportunistic short-lived species (Oug et al., 2018;Pedersen et al., 2018). While it is evident that high snow crab and red king crab biomasses lead to high turnover and reduced biodiversity of benthic species, the implications for ecosystem structure and functioning are still uncertain. To gain a better understanding of their present and future ecosystem impacts as they keep expanding in the Barents Sea, further monitoring and impact assessment of these invasive species is needed (Jørgensen et al., 2015).  (Heath et al., 2014) or altering benthic invertebrates communities through seabed abrasion McLaverty et al., 2021;Tillin et al., 2006).
Here, we found that trawling intensity co-varied spatially with multiple ecological metrics. The trawling intensity was higher in areas where the bentho-demersal communities were characterized by higher community-weighted mean trophic level and size and where the length of the food chain was longer. Yet, this covariation does not imply causation, as disentangling the effects of environment and fisheries on community and ecosystem structure is a difficult endeavour. Indeed, fishing activities are not randomly distributed but mostly take place in the areas that are accessible (e.g. no ice-cover) and where (and when) specific demersal communities prevail, notably where the biomass of commercial demersal species is the highest and are as such also linked to species seasonal migrations. These species often share similar traits, that is, in the Barents Sea the most fished demersal species are often large predators such as Atlantic cod and haddock. Moreover, fishing pressure and the environment can have cumulative impacts on ecosystems (Coll et al., 2016). Yet, there is large evidence that bottom trawling decrease the overall benthos abundance, biomass and species richness (Hiddink et al., 2020;Sciberras et al., 2018). In the Barents Sea, bottom trawling has been linked to a large decrease in total zoobenthos biomass, by as much as 70% (Denisenko, 2001;Jørgensen et al., 2016). low ice-cover (Husson et al., 2022;Mohamed et al., 2022). Several fish populations in the Barents Sea reacted to these extreme years by changing their spatial distribution (Husson et al., 2022).
These extreme climatic events might have also impacted the recruitment and survival of early stages of fish and invertebrates.
For example, the biomass of 0-group polar cod was very low in 2016 and was nearly absent of its common distribution area in the south-eastern Barents Sea (Eriksen et al., 2020). Here, we show that these displacements during extreme climatic events appear to also affect the community traits and food web properties of the bentho-demersal communities. Although it must be noted that in 2012, the sampling for megabenthos was irregular due to the use of a different gear on-board the Russian vessel, which might in part influence the 2012 peaks observed in the eastern clusters. The biomass-weighted metrics did not appear to change more than the unweighted food web metrics, with no directional turnover in species and biomass composition in the investigated areas over the last decade, whereas on a longer time-scale large ecological changes have been observed across organism groups (Denisenko, 2001;Eriksen et al., 2017;Fossheim et al., 2015;Frainer et al., 2017;Leonard & Øien, 2020). Here, the food web metrics were calculated based on the presence/absence of feeding interactions; this unweighted approach does not allow observations of potential changes in the strength of the interactions, whereas it has been shown that considering food web fluxes can give complementary information on the dynamics of an ecosystem (Kortsch et al., 2021).
The ecosystem-based management (EBM) aims to sustainably manage natural resources and biodiversity and minimize impacts on ecosystems, conserving biodiversity and ecosystem functions and services, while balancing social and economic targets. To move towards, EBM requires assessing the states of ecosystem components, their past and future dynamics, and their response to cumulative environmental and anthropogenic pressures. In the marine environment, these pressures include notably climate change, with both long-term trends and increased variability, fisheries and the spread of invasive species (Jørgensen et al., 2019). These pressures can have impacts at the population levels, to community, food web and ecosystem levels. It is, therefore, important to study their impacts across organizational levels. Here, we analysed bentho-demersal communities and food webs turnover along a biogeographical (Arctic-boreal) and pressures spatial gradients. This allowed us to delineate areas where the characteristics of bentho-demersal communities and food webs are similar, and which pressure, or combination of pressures, might have the most impact in these areas (e.g. snow crab and loss of ice-cover in the northeast Barents Sea). Such assessment of ecological indicators and their pressures is needed to prioritize conservation efforts. The Barents Sea is warming at a rate exceeding the global average, and the ice-cover extent is diminishing (Ingvaldsen et al., 2021), which may cause the expansion of invasive and boreal species and the fisheries that target them into these fragile ecosystems (Fauchald et al., 2021). These concerns caused the Norwegian Government to tentative close the high north areas of the Barents Sea for trawling .
This illustrates the importance of taking management measures that mitigate the cumulative impacts of environmental and anthropogenic pressures on the Arctic communities and food webs (e.g. marine protected areas, fishing gears restrictions) and continuous monitoring of ecosystem states. This study focused on the benthic compartment and assessed mainly feeding traits and interactions; future analyses should include other compartments (e.g., pelagic) and ecological indicators and pressures to study cumulative impacts on ecosystem functioning and services.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The food web, species traits, and ecological metrics datasets as well as the Rcode to reproduce the figures are available on github.com/