Identifying copepod functional groups from species functional traits

We gathered information on the functional traits of the most representative copepod species in the Mediterranean Sea. Our database includes 191 species described by 7 traits encompassing diverse ecological functions: minimal and maximal body length, trophic group, feeding type, spawning strategy, diel vertical migration and vertical habitat. Cluster analysis in the functional trait space revealed that Mediterranean copepods can be separated into groups with distinct ecological roles.

similar traits into functional groups (i.e. sets of species with similar effects on ecosystem functions and/or similar response to environmental conditions; Gitay and Noble, 1997) or to describe functional diversity of zooplankton communities (Barnett et al., 2007;Vogt et al., 2013;Pomerleau et al., 2015). Identifying and describing zooplankton functional groups should then increase our understanding of zooplankton ecological roles in marine ecosystems.
In this study, we developed a trait database for the most commonly sampled and abundant Mediterranean copepod species. Copepods are abundant in marine pelagic ecosystems (e.g. Siokou-Frangou et al., 2010) where they constitute the main trophic link between primary producers and higher trophic levels, such as small pelagic fishes (Costalago et al., 2015). Copepods are also relatively well documented in terms of distribution and biology (Razouls et al., 2005(Razouls et al., -2015. From this trait database, the aim of this study was to estimate how many functional groups could be identified among Mediterranean copepods, to characterize them and to discuss their ecological significance. We considered 191 copepod species (Table I) that are the most representative of the Mediterranean copepod communities, in terms of both abundance and presence (Siokou-Frangou et al., 2010;Mazzocchi et al., 2014). More details on how this list of species has been gathered are available in Supplementary data, Material S1. We used the following traits covering various types and ecological functions , known to be ecologically meaningful (Kiørboe and Sabatini, 1994;Kiørboe, 2011;Kiørboe et al., 2015) and commonly used for zooplankton (Barnett et al., 2007;Barton et al., 2013;Pomerleau et al., 2015): -two morphological traits which relate to many ecological traits: minimum and maximum adult body (cephalothorax) length (mm); -one physiological trait defining the species' trophic group (Carnivore, Omnivore -Carnivore, Omnivore, Omnivore-Herbivore, Omnivore -Detritivore); -one behavioural trait depicting feeding strategy, classified into three classes (Kiørboe, 2011): active ambush feeding, cruise feeding and filter feeding, mixed feeding (for species that can switch between the three strategies); -one life history trait related to reproduction, defining the egg-spawning strategy (broadcast-spawner, sac-spawner); -one behavioural trait related to diel vertical migration (DVM) behaviour, classified into four classes according to the intensity of the observed migration: Non-migrant, Weak migrant (DVM occurs within tens of metres), Strong migrant (over several hundreds of metres), Reverse migrant (for species that migrate deeper at night).
All body sizes were obtained from Razouls et al. (Razouls et al., 2005(Razouls et al., -2015, whereas the other traits were obtained from an extensive literature review (see Supplementary data, Material S4 and Table SII for the  full list of references). Additionally, to discuss the potential role of each functional group in the pelagic ecosystem, the species' preferential depth layer was established (epi-/meso-/bathypelagic). We were able to determine at least 5 of the 7 functional traits for 171 species. The trait database for the 191 copepod species is available as Supplementary In order to identify functional groups, we performed a multiple correspondence analysis (MCA) on the trait matrix. MCA is an ordination method in reduced space for the multivariate analysis of categorical variables (Husson et al., 2010). The computation of the MCA functional space was performed on four traits: class of maximum body length (Size_1: 0.501.80 mm, Size_2: 1.89 -2.85 mm, Size_3: 3.00 -5.70 mm, Size_4: 6.10-11.0 mm), binary trophic group (Carnivore, Omnivore, Herbivore, Detritivore), feeding type and spawning strategy. Indeed, the minimum body length was highly correlated to the maximum body length (R 2 ¼ 0.866, n ¼ 191) and DVM behaviour was not taken into account since it tends to be very plastic for most species, meaning that it is known to vary greatly according to the environmental fluctuations and species' ontogeny (see Pomerleau et al., 2015). A preliminary MCA incorporating these traits showed they had no impact in the definition of functional groups. Species for which the four traits were not fully defined were used as supplementary objects, meaning that they are associated with a group a posteriori, from their informed traits (see Supplementary data, Material S2 for more details on the MCA). The Euclidean distance among the 191 species in the functional space was computed using their coordinates along the four significant axes of the MCA (70.77% of the variance). Hierarchical agglomerative clustering was performed on this distance matrix using a synoptic aggregation method (Ward's; Husson et al., 2010). Depending on the cutting level, two, three or six clusters could be identified (Fig. 1). The first level distinguished species according to the trophic group (carnivore vs. non-carnivore). Among non-carnivore species, the second level separated broadcasters from sacspawners. Then, each of these groups was divided into two subgroups with different size and/or feeding type. Since higher cut levels could not be clearly related to functional traits, six functional groups were retained (Tables I  and II). We will now detail each functional group and discuss their ecological role in the Mediterranean pelagic food web.  Temora stylifera The species are numbered by alphabetical order but gathered by functional groups, as revealed by the hierarchical clustering on the first four axes of the MCA on functional traits (see Fig. 1). The species with bold numbers are the 99 species used in the MCA space calculation.
Group 1 is composed of large carnivores that primarily feed by cruising. There is no unique reproductive strategy, though the species are mainly sac-spawners. Many species have a broad vertical distribution, ranging from the epipelagic to the bathypelagic. Representative genera are Calanoids such as Candacia spp., Haloptilus spp., Heterorhabdus spp., members of the Euchaetidae family and Cyclopoids of the Sapphirina genus. These species are known to prey on smaller copepods, as well as other zooplanktonic taxa, such as doliolids (Takahashi et al., 2013), larvaceans (Ohtsuka and Onbé, 1989) and even fish larvae (Yen, 1987). Group 2 is defined by smaller carnivore species, all active ambush feeders and mainly sac-spawners, belonging to the Corycaeidae family. These are small visual predators that prey on nanozooplankton, nauplii, younger stages of copepodites through active ambush tactics (Landry et al., 1985), on wide depth intervals. Together, Groups 1 and 2 mainly contribute to the top-down control of mesozooplankton, including the other copepod functional groups.
Group 3 gathers a lesser number of species and consists of Calanoids of the genera Centropages and Acartia. They are small omnivorous broadcasters, but phytoplankton can become an important component of their diet. They exhibit mixed feeding strategies, depending on the available food items. Acartia spp. and Centropages spp. are generally restricted to the epipelagic and are affiliated with neritic environments (Siokou-Frangou et al., 2010).
Group 4 is the largest group and comprises almost all filter-feeding species, spanning all size classes, with a clear tendency towards herbivory. The species of this group for which reproductive strategy could be found were mainly broadcasters. This group contains not only small-bodied calanoids that are numerically very important in the Mediterranean epipelagic (Clausocalanus spp., Calocalanus spp., Temora stylifera; Mazzocchi et al., 2014), but also larger calanoids, some of which are strong vertical migrants, such as Calanus helgolandicus, Pleuromamma spp. or Neocalanus spp. (Andersen et al., 2001(Andersen et al., , 2004. The small surface calanoids are the target prey for larval and juvenile pelagic fish (Borme et al., 2013;Costalago et al., 2015), whereas larger calanoids are the preferential prey of Mediterranean mesopelagic fishes (Palma, 1990). Additionally, the strong calanoid migrants might play a differential key role in carbon cycling as they graze upon microalgae in the euphotic zone, and then migrate below the permanent thermocline where they excrete their lipid reserves (lipid pump hypothesis; Jónasdóttir et al., 2015). Therefore, Groups 3 and 4 are crucial for the transfer of energy from photoautotrophs to higher trophic levels, both in neritic (Group 3) and in oceanic environments (Group 4). Also, the latter group might comprise species that play a potentially underestimated role for the carbon flux (Jónasdóttir et al., 2015).
Group 5 consists essentially of Oithona spp. These are small active ambush-feeding omnivores that carry their Fig. 1. Identification of functional groups among the 191 most representative copepod species of the Mediterranean Sea from hierarchical clustering on the first four axes of the MCA based on four functional traits (class of maximum body length, binary trophic group, feeding type and spawning strategy). Depending on the cutting level, two, three or six clusters could be identified. The first level distinguished species according to trophic group (carnivore vs. non-carnivore). Among non-carnivore species, the second level discriminates broadcasters vs. sac-spawners. Then, each of these groups can be divided into two subgroups with different size and/or feeding type. Since higher cut levels could not be clearly related to functional traits, six functional groups were retained (Table II). Copepod species are indicated by numbers (Table I) . eggs. It is difficult to assign a particular ecological function to such a group, since Oithona spp. are a major component of the global ocean's plankton, independently of environmental conditions (Gallienne and Robins, 2001). Feeding and trophic group were unknown for the other species of Group 5. Therefore, they are related to Oithona spp. only because of they are small (,1.8 mm) sacspawners. Group 6 also comprised small sac-spawning omnivores, but these are mainly cruising detritivores (Oncaea spp., Microsetella spp.) or herbivores (Clausocalanus spp.). The former usually exhibit a wide vertical distribution, while the latter are epipelagic. Oncaea spp. and Microsetella spp. are known associates of appendicularian houses (Alldredge, 1972;Steinberg et al., 1994). Several calanoid species of this group (Euchirella spp., Scolecitrichidae) are also deepwater detritivores. Consequently (and as suggested by a higher cut level on dendrogram Fig. 1), Group 6 could be sensibly divided into two subgroups: (i) deep-water dwelling detritivores that actively participate in the recycling of particulate organic matter and (ii) small cruising grazers contributing to epipelagic secondary production.
By focusing on Mediterranean copepods, we were able to gather information on functional traits for 191 species, with at least 6 traits described for 135 species and 7 for 66 species. Using 4 of these traits, functional groups with different ecological roles were described. Although the definition of these groups was robust (similar groups were found using a K-means partitioning method, or accounting for all traits), the main limitation, as for any traitbased approach, remains the scarcity of trait descriptions at the species level. Compiling a trait database for a larger number of zooplankton species remains challenging, but the present initiative demonstrates the usefulness of this endeavour.
The present study also confirmed or revealed tradeoffs among zooplankton functional traits , Kiørboe, 2011. For instance, small carnivores and small omnivores were active ambush feeders, while large The number of species recorded within each trait's class and within each functional group is reported. The groups are based on hierarchical clustering on the first four axes of the MCA based on four functional traits: class of maximum body length, binary trophic group, feeding type and spawning strategy. For information, the type of DVM, the vertical distribution, the average minimum body length (mm) and the average maximum body length (mm) are also indicated (in grey). NA, not available; DVM, diel vertical migration.
carnivores were cruise feeders. This relationship between size and feeding strategy could be explained by the differences in metabolic requirements. Indeed, whereas cruise feeders have to swim actively to encounter their prey, ambush feeders passively encounter them (Kiørboe, 2011), which requires less energy. As metabolic rates scale with body size (Kiørboe and Hirst, 2014), this difference in metabolic requirement could explain their difference in size. We also found that sac-spawners were active ambush feeders or cruise feeders, whereas broadcasters were mainly filter feeders or mixed feeders. This could be due to optimal resource allocation: egg-carrying ambush-feeding copepods have longer lifespans and lower fecundity rates than broadcasting active feeders (Kiørboe and Sabatini, 1994;Kiørboe et al., 2015). We also found that three quarters of the carnivorous species were sac-spawners; hence, these exhibit a higher degree of parental care and avoid predation on their own eggs. Conversely, for epipelagic omnivorous filter feeders, which are heavily preyed upon by other zooplankters and fishes, broadcasting is likely to be favoured to avoid being eaten together with the eggs, as an adaptation to the elevated mortality of ovigerous females (Kiørboe and Sabatini, 1994). Together, these results call for a better understanding of the mechanistic processes that lead to such trade-offs across traits. The functional trait database compiled here can be used to estimate the functional diversity of zooplankton communities (Vogt et al., 2013;Pomerleau et al., 2015) and test whether traits can be related to environmental variables (Barton et al., 2013). Finally, the existence of several functional groups encourages the integration of more diverse planktonic assemblages in ecosystem models.

S U P P L E M E N TA RY DATA
Supplementary data can be found online at http://plankt. oxfordjournals.org.

DATA A RC H I V I N G
The trait database for the 191 copepod species can be downloaded from PANGAEA (http://doi.pangaea.de/ 10.1594/PANGAEA.854331).