We worked with the database of the French National Agency for Water and Aquatic Environments (Onema), which contains records of standardized electrofishing protocols performed between 1990 and 2009 during low-flow periods (May-October). Electrofishing is considered the most effective non-destructive sampling procedure for describing fish assemblage structure [24]. Sampling protocols were defined depending on river width and depth. Streams were sampled by wading (mostly two-pass removal), while fractional sampling strategies were undertaken in larger rivers. Since the implementation of the EU Water Framework Directive’s surveillance monitoring, protocols follow the recommendations of the European Committee for Standardization [25]. To compare inter-annual densities, however, only surveys performed with the same sampling protocol were selected in the whole data set. Fishes were identified to species level, counted and then released. We worked with the 26 species for which trait data were available (around 90% of the total abundance catch) (Table 1).

We extracted collecting events from Onema’s fish database using two different criteria:

The dataset of human pressures was provided by the European land-cover database CORINE, which classifies landscape units larger than 25 ha into one of 44 classes [39] on the basis of satellite digital images (e.g. SPOT and LANDSAT). We used the 2000 update and considered three alternative groupings of seven habitat classes: (i) The CORINE Land Cover (CLC) yields 5 habitat classes: “Forest”, “Meadow”, “Farming”, “Urban”, and a “Mix” (i.e. a Mix between agricultural and urban habitats) (ii) The EUROWATER (a special variant of CLC for freshwater common to the European scale), yields 6 habitat classes with the addition of the “Intensive Urban” habitat class, and (iii) The ONEMA land use classification (a special variant of CLC and EUROWATER for freshwater common to the national scale) yields 7 habitat classes with the addition of the “Intensive Farming” class. Only the latter two classifications are used to test the reproducibility of our indicator.

Here, we consider land use classification as a gradient of landscape artificialization under human pressures. Land use is a common proxy for human pressures in terrestrial communities [6,22]. The link between land use and human pressures in river has been reviewed at a global scale [40], but also on regional scale in North America [41,42] and Europe [43]. Marzin and al. (2013) showed a clear link between the CORINE Land Cover (CLC) dataset and different pollution and physical modifications at both local and regional scales [43]. If both human pressures are correlated with CLC, water quality parameters are more strongly correlated to land use than physical modifications.

All statistical analyses were performed using R 2.15.1 (R Development Core Team. 2012), and more particularly the ade4 and nlme packages [44,45]. We calculated one index for each kind of data set, thus for a total of 6 indices: 4 functional originality indices, 1 phylogenetic originality index, and 1 specialization index.

The four trait distance matrices can be visualized using trees (Figure 3). Trees based on life-history traits (Figure 3a), functional niche (Figure 3b), and diet-linked morphological traits (Figure 3c) were well balanced in the sense of Blum et al. [51]. These authors defined the balance of a tree as the average balance of its nodes, “assuming that a given node is completely balanced if it splits the sample into two subsamples of equal size”. At the opposite, the tree based on habitat-linked morphological traits (Figure 3d) was highly unbalanced by the European eel (Anguilla anguilla), and, to a lesser extent by the groups common bream (Abramis brama), crucian carp (Carassius sp.), ninespine stickleback (Pungitus pungitius), three spines stickleback (Gasterosteus aculeatus). Using the equal-split metric, we computed four originality indices for each species to evaluate the three functional datasets and the phylogeny (Figure 4). As expected, A. anguilla was characterized by a high originality score for the habitat-linked morphological trait matrix (SOI = 0.81 compared to a mean of 0.24). The two other imbalanced nodes had smaller originalities (Abramis group = 0.46 and Pungitius group = 0. 37). The species specialization index ranked the European eel as the most specialist species and the common bleak, Alburnus alburnus as the most generalist species (Figure 4). At the community level, the habitat-linked morpho-COI was sensitive to the presence of the European eel and to a lesser extent to the presence of the common bream, Abramis brama, and the crucian carp, Carassius sp. We tested the sensitivity of the Species Originality Index (SOI) based on traits to the addition of species in the initial dataset. The life history traits index and the habitat-linked morphological index were strongly correlated (respectively R²=81, R²=86). The diet-linked morphological index and the niche index were less correlated (respectively R² = 65 and R²=68) and thus more sensitive to the addition of fish species in the initial dataset [see the File S1].

All statistical models retained by the AIC, with both datasets, contained the same variables: geographic coordinates, year, and their interaction, except for the Diet-COI where the watershed effect gave a better model (AIC). Because Corsica appeared to be an outlier (Figure 5), we re-ran all analyses excluding data from this area. With Corsica excluded, we found that watershed was a better spatial effect than geographic coordinates (AIC).

Corsica clearly comes out as a hotspot of fish originality and specialization (Figure 5). In contrast, the Seine watershed presented the lowest originality and the most generalist fish communities. The CSI, habitat-linked morpho–COI and LHT–COI presented limited variation among sites and appeared to be ineffective to discriminate sites. In contrast, the diet-linked morpho -COI highlighted a strong originality in all small rivers, especially mountainous streams. Although the temporal effect was always retained in statistical models on the AIC, it was not always significant. However, Niche and both Morpho-COI increased significantly over the last years (Table 3).

Our results showed that all COI, which are based on the same originality metrics, tended to be correlated with each other, while the CSI, based on a different approach and metric, was correlated with a subset of the COI only. All these correlations took into account space and time effect for the statistical independence of data. The habitat-linked morpho –COI was correlated with the LHT – COI (R²= 0.47, F_8,12829 = 1407, P < 0.001), and with the diet-linked morpho –COI (R²= 0.24, F_8,12829 = 518, P < 0.001). Interestingly, the phylogeny approach seemed to capture different proportion of morphological variation in function of the diet or the habitat niche axis [phylo –COI and habitat-linked morpho -COI (R²= 0.53, F_8,12829 = 1842, P < 0.001), and phylo-COI and diet-linked morpho COI (R²= 0.39, F_8,12829 = 1037, P < 0.001)]. Unlike other vertebrates such as birds or mammals (birds: [52]), the fish Life History Traits index presented a weak correlation with the phylo-COI (R²= 0.13, F_8,12829 = 239, P < 0.001). The niche-COI was more strongly correlated with the phylogenetic index (R²= 0.56, F_8,12829 = 2013 10⁴, P < 0.001) than with the morphological indices (Diet: R²= 0.33, F_8,12829 = 798, P < 0.001; and Habitat: R²= 0.50, F_8,12829 = 1596, P < 0.001) even though the latter are assumed to represent niche axes. The niche-COI was also correlated with the LHT –COI (R²= 0.23, F_8,12829 = 475, P < 0.001).

The CSI was strongly correlated with the LHT-COI (R²= 0.74, F_8,12829 = 3.43 10⁴, P < 0.001) and habitat-linked morpho - COI (R²= 0.47, F_8,12829 1397, P < 0.001) but weakly with the niche (R²= 0.11, F_8,12829 208, P < 0.001), and not with the diet-linked morpho (R²= 0.04, F_8,12829 65, P < 0.001) and phylo – COI (R²= 0.08, F_8,12829 149, P < 0.001). It is important to note that the level of specialization measured here is more relevant to describe the Fish Life History Traits component than the habitat niche component.

The sensitivity to human pressures of the six community indices was evaluated with respect to land use data and two variations of CORINE Land Cover (Table 4). All indices correlated with land use (Table 4-6), but with some variation. For example, some indices were sensitive to the different human pressures (farming or urban) represented here by an artificialization gradient (Figure 6). In contrast, the CSI was significantly higher for urban area than for agricultural or natural habitats (Figure 6). We used two variations of CORINE Land Cover (ONEMA and EUROWATER) to get an estimate of the community indices reproducibility in function of the arbitrary habitat classifications [Table 4-6, see the File S2] and only one COI was robust to the effect of habitat classifications: the Niche – COI (Table 4-6). The response of this latter index was also significantly sensitive to the different type of human pressures with a consistent behavior at national and regional scales (Figure 7). Within each watershed or over all watersheds the relationship between human pressures and Niche-COI is negative when it is expected to be negative (e.g. under human pressures like farmland and urban habitat) and reciprocally (e.g. under natural habitat).

There is a growing consensus that functional diversity based on species traits is a better predictor of ecosystem functioning than species number per se [53]. Species richness is currently the most used biodiversity index (and indicator) but it is highly scale-dependent, with local increases that are often accompanied by regional or global decreases and increases in between-site similarity [54]. Particularly, fish species richness tends to increase from upstream to downstream [55]; and the upstream part of many French rivers sustain only a few fish species (< 5 fish species, [56]). Species richness is thus an inadequate surrogate in the context of ecosystem function unlike community traits approaches, which appear more and more relevant in the literature especially to examine ecosystem integrity [16,22,57]. Community-trait indices take into account the species present in the area considered, species being grouped depending on their ecological or phylogenetic affinity. These indices compute the average value for a trait or character, depending on the frequency of species locally present. So, we did not consider intra-specific variability, which sometimes represents a significant proportion of the variability and complex spatio-temporal dynamic [58,59]. Even though specialization community indices seem to give the same message with presence/absence data [60], it is not always the case when a process or a function is measured using functional traits [19]. Moreover one of the most interesting points to use common species is based on the assumption that abundance plays an important role in ecosystem functioning [57,61].

Community Specialization Index (CSI) is a different approach than Community Originality Index (COI) approaches. If CSI is not clear on the underlying mechanisms explaining the precise ecosystem function, its well-known power comes from its holistic habitat approach. The central focus of CSI is not the species feature but its interaction with the environment by the habitat approach. This statement is closer to the Grinnell niche theory approach than the Hutchinson one [62]. And thus, in the CSI approach, the crucial point is the relevance of habitat description, not the species traits data set.

On the other side, with the COI we study the distinctiveness of precise species traits and lineages, and thus we postulated that trait variation among species variation relates to functional differences in the ecosystem, which allows an interpretation in terms of ecosystem function or “services”. The set of traits selected is a crucial step toward this goal, especially if we want the dynamics of the indices to reflect ecosystem function [57]. Here, we worked only with traits having a demonstrated functional role in fish biology. For example, morphological traits such as the mouth position or the length of barbell are linked to the diet and food acquisition [29,35]. Because our originality index is based on distance metrics we verified that it was not species richness dependant of the initial dataset. The life history traits index and the habitat-linked morphological index are less sensitive (R²>80, see S1) than the diet-linked morphological index and the niche index (respectively R² = 65 and 68, see S1). Moreover, some community indices could be especially sensitive to one or a few species with extreme values of originality. In this study, it was the case with the European eel, which disproportionately increased the original community indices based on phylogeny and on habitat-linked morphology. This is the main purpose of this index, to weight unique species. However, because the European eel is classified as critically endangered at the national and global scales by the IUCN, these original indices also met in this particular case, the red list species indicator. Moreover, the European eel is a patrimonial species, there is a strong cultural heritage in France associated with this species for their fishing and cooking and thus for their taste, but also for their unique form and shape. For this last one based on the human vision, an originality index based on morphology could be common avenues for all species to be “objectively” quantified on the arguments develop by naturalists trying to preserve the unique forms and shapes that have been emerging on Earth.

We found that for the river fish communities, mapping each diversity component separately reveals partially congruent patterns between functional or phylogenetic originality, and specialization. All indices highlight Corsica, and to a lesser extent the Channel watersheds, as hotspots of originality and specialization. In contrast, the Seine watershed presents a less original and specialized fish community. For all the other watersheds, the different functional and phylogenetic COI and CSI are not congruent. For example, diet-linked and habitat-linked morphological COI present completely different patterns. The former highlights a strong originality in all small rivers especially mountainous streams, whereas the latter did not present any strong variation pattern at all, except for Corsica Island.

These common patterns suggest that species occurring locally may be derived from regional species pools with similar biogeographical and evolutionary histories [63,64]. Moreover, for a given regional pool, species may respond to environmental gradients in different ways affecting the spatial distribution of the different biodiversity components and generating a spatial mismatch between functional and phylogenetic COI and CSI [64,65]. These results challenge the use of a single component as a surrogate for the others, and stress the need to first understand the different processes underlying each index and second to adopt a more integrative approach for conservation. One option to deal with the different messages given by the functional properties of communities and the resulting set of measurements is to be able to provide a hierarchy of their meanings depending on the context and perspectives or more reasonable to only use common patterns.

A temporal effect has been detected for three original community indices (Table 3). Both morphological and Niche community indices significantly increased over time. This temporal dynamic should result from a global increase of the total species abundance, which has been shown (t= 5,09; p<0.001) and thus, may be the result of a global improvement of the entire river ecosystems. Indeed, global water quality has improved compared to the last century thanks to significant efforts to decrease organic substances [26,66]. Moreover, fish populations in Europe are still in their re-colonizing process since the last glaciations and some species are expected to extend their geographical area [67]. More precisely, each index is more influenced by the population dynamic of a few species presenting a high original value. The European eel has a very high original value especially for both habitat-linked morphological and niche indices, even though this species is declining (t= -6,53; p<0.001) which implies an important increase by other species in compensation. For the niche-COI, the global increase could be mainly linked with the population expansion of the European perch, northern pike, and minnow. The northern pike is very popular with anglers favoring their introduction and thus, may have a positive impact on the population dynamic [26]. In the case of the diet-linked morphological-COI, it could be explained by three increasing species: the three spines stickleback, the common nase, and again the minnow. And in the case of the habitat-linked morphological-COI, the temporal increase could be mainly linked with the increase of two introduced species: the crucian carp and the pumpkinseed sunfish. Introduced species may increase faster than native species due to their rapid spread and their repeated introductions (accidental or deliberate).

If originality and specialization approaches are completely different in their mathematical calculations, they describe complementary components of the functional properties of communities with similar expectations for their roles in ecosystem services. It has been theoretically and experimentally shown that the alteration of biodiversity disrupts ecological functions performed by species assemblages [68], and we know that species niche partitioning is fundamental in ecosystems properties [69]. Thus the more we are losing specialist or original species, the more we are losing irreplaceable functions in the ecosystem [6,70].

The theoretical background underlying the link between COI, CSI, and ecosystem functioning is growing in the literature, but it does not yet mean that both types of indices are relevant as functional biodiversity indicators. Community indicators have to be sensitive to anthropogenic pressures and give a clear and simple message to be technically and practically used by the targeted audience. Thus, the interpretation of indicators has to be as simple as possible and some communication qualities have to be accounting for. An indicator that is well recognized by all biodiversity stakeholders is more likely to be used in the future. According to the Millennium Ecosystem Assessment [17], one of the most important direct drivers of biodiversity loss and ecosystem service changes are land use modifications, including the physical modification of rivers. In this study we build innovative indices for community river fish in order to open the way for a new generation of indicators based on traits or niche linked with ecosystem functions. We compared each of our created community indices with the land use dataset (Figure 6). Interestingly the CSI present a very high score in urban areas. Fish communities are composed of specialist species in this artificialized habitat, probably because the environmental filter is very strong and species need to be specialists of this disturbed area such as the Black bullhead (Ameiurus melas), an invasive species which have a strong SSI. This pattern of urban specialist species in fish seems to be similar to the bird one, with urban specialist species such as pigeons (Colombia liva) or house sparrows (Passer domesticus) [21].

One COI appears particularly relevant to become a functional indicator of river fish communities, the COI based on species niche. This functional COI is sensitive to the different kinds of human disturbances with a simple interpretation: higher disturbance correlates with lower indicator values. Indeed, when we sorted land use on an artificialization gradient from natural to farming to urban areas, we observed a decrease in the niche-COI (Figure 6). Moreover, the three different variations of the CORINE Land cover data set give exactly the same results for the niche-COI, and thus we can be confident in its reproducibility over classification criteria. Finally, we evaluated its sensitivity to spatial scale. Is the pattern observed at the national scale still present at the regional scale? We observed a consistent pattern over the different watersheds, with the exception of the Mediterranean region (Figure 7), which may be due to the sample size. More interestingly, but not useful as an indicator, it may due to a local scale dependence or an eco-regional dependence. Indeed, Mediterranean watershed is very small and without any big river. In addition, all the urban areas are concentrated along the coast in this part of the Mediterranean eco-region. Further investigations need to be done to confirm the context of the use of this potential bio-indicator.

Otherwise, Niche-COI as a functional biodiversity indicator encompasses species indicators like the CBD headline indicator “trend of selected species” because they consider a complete ecological group with their functions and their common dynamics. As a result, they carry more significant ecological information so that expectations and objectives for biodiversity stakeholders can be derived. Originality indices have already been used as indicators in biodiversity conservation contexts. Isaac et al. [71] built one called EDGE (Evolutionary Distinct and Globally Endangered) based on the phylogenetic originality and conservation status of species. At the community level, Mouillot et al. [70] have suggested functional and phylogenetic COI to evaluate conservation action areas. Because interspecific competition is more intense among species sharing common traits due to the limiting similarity principle [72], they expected that under protected areas competition might drive the better colonization or subsistence of the most original species because of niche complementary. We believe that even if one of the goals of this study was to develop functional biodiversity indicators for environmental policy makers, the Niche-COI could also be used at the scale of conservation reserves and may be used by managers of protected areas.

Finally, we have to keep in mind that biodiversity indicators help to prioritize conservation actions to conserve ecological functions and in fine ecosystem “services”. However, evaluation and measure alone are not sufficient in order to stop biodiversity loss, human pressures also must be limited.