According to the requirements of the European Water Framework Directive [1
], the Canadian Aquatic Biomonitoring Network (CABIN; [3
]), and the U.S. Environmental Protection Agency [4
], macroinvertebrates have been commonly used for the bioassessment of anthropogenic disturbances in rivers because they are (i) reliable bioindicators of water and sediment qualities [5
], (ii) efficient and cost-effective biomonitoring tools [7
], and (iii) useful to differentiate reference conditions from impaired sites [9
]. In streams and small rivers, studies showed that the structure of the benthic macroinvertebrate community reflects the impacts of anthropogenic disturbances such as water acidification, organic pollution, metal contamination, and habitat degradation [11
]. In large rivers, changes in the macroinvertebrate community are related to multiple environmental factors [15
], including changes in habitat vegetation [16
], water-level fluctuations [18
], water quality [19
], sediment grain size and contamination [21
], and human disturbances [24
Bioassessment approaches comparing reference and disturbed sites are designed to determine whether poor water or sediment qualities are stressing the macroinvertebrate community beyond the range of natural variation [25
]. However, this is a difficult task due to the complexity of river ecosystems and the interaction of multiple factors which limit the possibility to predict the overall responses of macroinvertebrate assemblages to environmental changes, either natural or anthropogenic. Difficulty to find reference sites in large and complex rivers under significant anthropogenic stressors is another problematic issue that further limits bioassessment. Approaches comparing sites on a disturbance gradient are now more suitable for bioassessment in rivers because they help to establish the relationships between macroinvertebrate community structure and natural environmental conditions and anthropogenic stressors [6
Over the past 20 years, 60% of the biological indicators used to assess ecological quality of rivers were based on macroinvertebrate communities [27
]. A myriad of indices and metrics have been applied in bioassessment approaches to establish the various sensitivities of macroinvertebrates to different types of disturbances. However, the development of most suitable macrobenthic indices and metrics for the bioassessment of the ecological quality status of rivers and lakes is still in progress. The first indices that come to mind are the diversity indices and metrics based on taxon richness, used since the eighties. However, their relevance has been discussed [28
] because taxon richness and diversity indices depend more on geographical, climatic, historical, and ecological factors than on the direct impact of anthropogenic stressors, except in the case of extreme physical or chemical disturbance. Diversity indices and metrics alone are no longer recognized as relevant tools in biological assessment [29
] but they could be included such as multimetric STAR-ICM index in small, lowland rivers in Europe [30
]. Biotic indices combining richness and abundance of sensitive or tolerant taxa were more successful in detecting ecological changes among sites and the effects of anthropogenic stressors in rivers [31
], lakes [32
] and ponds [33
]. Among the indices used to assess stressor-specific disturbances, we can cite: (i) the saprobic index [8
] and the Hilsenhoff index [35
] for organic pollution, (ii) the Index of Community Sensitivity (ICS, [22
]) or the Invertebrate Community Index (ICI, [37
]) for water quality and sediment metal contamination. More complex integrated monitoring based on multimetric procedures has recently been implemented for biological assessment in Europe [12
] and North America [22
]. These procedures allow the selection and aggregation of metric scores in a single index that helps to determine whether action or restoration is needed and simplify management and decision-making. Successful multimetric score procedures have been validated for river pollution surveys. As examples, we can cite: (i) the Biological Monitoring Working Party (BMWP) used in Spain [41
], the UK [12
], Poland [27
], and Canada [42
], (ii) the Index of Biotic Integrity (IBI, [43
]) and the Panel Index [40
] used in the USA, (iii) the Belgian Biotic Index (BBI, [44
]) or the Multimetric Macroinvertebrate Index Flanders (MMIF, [12
]) used in Belgium, and (iv) the Macroinvertebrate-Based Multimetric Index (IBMA) applied in Martinique and Guadeloupe territories [39
]. In Canada, multimetric indices have been developed to assess sensitivity of benthic biota to river flow regimes [45
] and water quality [42
]. However, in Canada, there are still few developments towards a multimetric approach using macroinvertebrates to assess sediment contamination in large rivers compared to those in other countries [15
]. The future needs for the development of sediment bioassessment methods in large rivers include: (i) the selection of relevant macroinvertebrates indices and metrics based on ecological principles underlying metric choice for specific disturbances, (ii) the validation of the potential of indices and metrics to discriminate sites according to a gradient of environmental conditions and disturbances, and (iii) the determination of criteria and management thresholds that indicate environmental degradation and the need for quantitative assessment studies and remediation projects.
The present research focuses on the St. Lawrence River (QC, Canada), one of the most important large rivers in the world draining a watershed area of 1,610,000 km2
and flowing across 1000 km from Lake Ontario to the Gulf of St. Lawrence. Since the 1950s, intensive agriculture, urbanization, and industrialization have caused the contamination of sediments in the St. Lawrence River [48
]. Macroinvertebrates are a critical component of the wetland and sediment food webs in the St. Lawrence River [18
]. Previous studies showed that their distribution, community composition and functional traits vary according to ecological and toxicological factors such as vegetation types (filamentous algal mats, emergent and submerged macrophytes) [16
], water quality and sediment contamination [17
], landscape features and hydrological regime [18
]. However, most of these studies were limited to littoral wetland habitats or to specific approaches based on species assemblages and functional traits. Since macroinvertebrates can be impacted by sediment contaminants, their biomonitoring is required to complete the contamination assessment, to evaluate the ecotoxicological risk, and to determine the remediation needs at sediment-contaminated sites in the St. Lawrence River. This large river is also an essential transportation route in northeastern America, and periodic dredging of sediment is required for the maintenance of the waterway and harbour facility. Because dredged sediments may contain a range of contaminants that could affect benthic organisms at deposit sites, environmental risk assessment and management of these dredging projects are required to determine sediment quality [50
No study has yet compared the potential of multiple indices and metrics to assess specific stressors such as sediment dredging and contamination across the fluvial section of the St. Lawrence River. The objectives of this study are as follows: (i) determine the composition of macroinvertebrate community in sediments of typical habitats across a gradient of disturbance, (ii) select relevant indices and metrics from a panel of macroinvertebrate indices and metrics based on their ecological relevance for the assessment of sediment quality and contamination and on their potential for large river ecotoxicological risk assessment, (iii) investigate whether sensitivity of selected indices and metrics differ across habitats and/or sediment quality classes, and finally, (iv) determine the thresholds for critical contaminants related to significant changes in the most relevant indices and metrics.
3.1. Typology of Macroinvertebrate Communities in the Fluvial Section
The sediments of the fluvial section of the St. Lawrence River supported abundant and diverse macroinvertebrate communities composed of fourteen taxonomic groups (Figure 2
). Overall, the taxa belonged to 45 families and 109 genera (see Table S2
for the full list of taxa, Supplementary Materials
). Macroinvertebrate composition varies among habitat zones and sites. In Lake Saint-Francois, macroinvertebrates were mainly composed of arthropods (Insecta), molluscs (Gastropoda), and crustaceans (Malacostraca). Community composition differed from upstream to downstream and between the north and south shores. Nematoda were found in greater abundance on the south shore than on the north shore while the Oligochaeta were relatively more abundant downstream than upstream, and inversely for the Malacostraca. In Lake Saint-Louis, community composition and dominance patterns were more variable from site to site than in Lake Saint-François. The Nematoda were more common on the north shore, the Gastropoda and Insecta in the bay of the island, and the Oligochaeta, Malacostraca and Bivalvia on the south shore. The Oligochaeta were more frequent downstream and the Nematoda upstream. In Lake Saint-Pierre, macroinvertebrate communities were also relatively diverse. Oligochaetes, Nematodes, Insects, and Bivalves were the most predominant groups. Community composition varied among the north and south shores, and along the longitudinal gradient. On both the north and south shores, communities were dominated by worms (Oligochaeta, Nematoda). However, Insecta and Bivalvia were more common in the north shore and at the upstream sites, while worms were more common at downstream sites. The Montreal Harbour supported the most disturbed community with a low diversity and a predominance of Oligochaeta associated to Insecta (mainly Diptera Chironomidae) and Bivalvia. Despite a large variability in macroinvertebrate composition among sites and habitat zones, the typology suggests a gradient of disturbance.
3.2. Selection of Macroinvertebrate Indices and Metrics
Selection procedures based on ANOVA and correlation analyses allowed us to select 157 indices and metrics (in bold in Table S3
). These included 20 indices of diversity and similarity, 49 metrics of taxa richness, 36 metrics of taxa abundance and trophic guilds, 13 biotic indices based on taxa tolerance, and 39 functional traits (Table 2
). Most of the selected indices and metrics (142) showed significant differences among habitat zones (highlighted in blue in Table S3
). In contrast, only 5 metrics based on taxa abundance showed significant variation among sediment quality classes (ANEM, AHIR, ADIP, AHYD, AGOLD: highlighted in orange in Table S3
), and only 3 metrics based on number of tolerant taxa and the dominance of scrapers, as well as 7 functional traits showed significant variation among both habitat zones and sediment quality classes (highlighted in green in Table S3
The PCA analysis based on the 157 indices and metrics allowed us to eliminate additional 61 indices and metrics showing no significant contribution to spatial patterns in macroinvertebrate communities and sampling sites (Figure S2, Supplementary Materials
). All fluvial lakes sites were grouped together in the center of the ordination plan, except for one extreme site in Lake Saint-Louis located in the lower right quadrant. The most impaired sites of Montreal Harbour were dissociated from those of the fluvial lakes in the upper right quadrant.
Finally, the stepwise RDA procedure comparing pairs of indices and metrics based on six criteria allowed us to eliminate another 143 indices and metrics and to retain only 14 metrics and indices as the most selective and parsimonious choices (Table 3
). We selected the indices and metrics which were recognized as (1) relevant and easily to apply for bioassessment in large rivers, (2) having the potential to distinguish macroinvertebrate communities among habitat zones and/or sediment quality classes based on ANOVA analyses, (3) having the highest contributions in PCA ordination, and (4) showing significant relationships with sediment characteristics and contamination based on correlation analysis, RDA and regression tree analysis. We also gave priority to indices and metrics calculated at the genus level since our previous studies have shown a higher explanatory power at this taxon level [23
] (5) and based on abundance and tolerance (6). Selective choices were made among indices and metrics that were collinear or had similar ecological principles. For instance, the metrics Ita and AOL were collinear (ρ Spearman 0.95, p
= 0.9925) (see also vector projections in Figure S2, Supplementary Materials
). Thus, we retained only the metric AOL based on the abundance of Oligochaeta which was associated with the most impaired sites of the Montreal Harbour as shown with ANOVA and PCA analyses (Table 4
, Figure S2
). For example, we eliminated the metric Ital that was developed for small Italian rivers and judged inappropriate for a large and complex river such as the St. Lawrence River. The comparative selection of pairs of indices and metrics is detailed and presented in Table S4 (Supplementary Materials)
The 14 indices and metrics selected were the most relevant for distinguishing habitat zones in the fluvial section (Figure 3
). Increased diversity indices (SHANG, DM), greater richness in Ephemeroptera, Trichoptera, Coleoptera, Odonata and Bivalvia taxa (NBTETG, NBTCOBG), and higher abundances of Mollusca and Diptera Chironomidae (%Mach, %DipG, %GOLD, AGOLD) were associated primarily with Lake Saint-François sites (upper right and left quadrants), and opposed to certain sites of the Lake Saint-Pierre, Lake Saint-Louis and of the Montreal Harbour (lower left quadrant). In contrast, a higher percentage and abundance of non-insects (%NoIns) and worms such as Oligochaeta and Nematoda (ANEM, AOL, %OliG) were associated with certain impaired sites of the Lake Saint-Pierre, Lake Saint-Louis, and the Montreal Harbour (lower right quadrant).
Some of the final metrics were collinear (Figure 3
) and based on similar ecological principles. For instance, the metrics %GOLD and AGOLD are redundant as well as the metrics %DipG and %Mach. Thus, for a more parsimonious selection, it may be appropriate to retain only those metrics with the highest contribution in the PCA ordination of sites such as AGOLD and %Match. Therefore, the ANEM, AHIR, SHANG, and NBTCOBG metrics could also be eliminated as they have the lowest contributions. The ANOVA analyses indicated which indices and metrics have the highest potential to discriminate habitat zones and sediment quality classes (Table 4
) and complement the selection procedures.
Concerning habitat zones, diversity indices (DM, SHANG) distinguished only the sites of the fluvial lakes from the most impaired sites of the Montreal Harbour. Among metrics based on taxa richness, NBTCOBG also segregated the Montreal Harbour from the fluvial lake sites, while NBTETG segregated the LSF and LSL sites from the LSP and MH sites. The metrics %OliG and P5FD had a better potential than AOL, which distinguished only the two extreme habitat zones (LSF from MH). The metric %OliG segregated the MH impaired sites on one hand and the LSF sites on the other hand, but did not differentiate LSL and LSF sites, and LSP and LSL sites. The metrics %DipG, %Match and %NoIns segregated LSF sites from the sites of the two other fluvial lakes and the Montreal Harbour. The metrics ANEM, AHIR, and AGOLD failed to distinguish habitat zones, and the metric %GOLDG did not separate the less impaired LSF sites from the most impaired MH sites.
When considering the sediment quality classes, 12 final indices and metrics (except AHIR and AGOLD) presented a potential for distinguishing sediment quality classes when considering both categories of the class 3 (Table 4
). Macroinvertebrate indices and metrics with the greatest potential to distinguish sediment quality classes were the diversity indices (DN, SHANG), the number of taxa at the genus level for the Ephemeroptera, and Trichoptera (NBTETG) as well as the total number of taxa of Trichoptera, Coleoptera, Bivalvia and Gastropoda (NBTCOBG). Most of them clearly differentiated the less impaired sites (Class 1) from the most impaired sites (Class 3A and 3B), indicating improved sediment quality. In contrast, the metrics based on the abundance and dominance of Oligochaeta (AOL, %OliG) discriminated the most contaminated sites (Class 3A), indicating lower sediment quality. The other metrics based on the abundance or percentage of tolerant taxa (%DipG, P5FD, ANEM, %Match, %NoIns, %GOLDG) presented the lowest potential to distinguish sediment quality classes. Although functional traits metrics performed slightly better than usual taxonomical metrics [49
], they did not emerge as relevant in this sorting exercise. In addition, the database of traits available for macroinvertebrates of the St. Lawrence River is still incomplete. Consequently, trait approach is not currently used for bioassessment monitoring due to difficulty of their application in ecotoxicological risk assessment.
3.3. Relationships between Macroinvertebrates Metrics and Indices and Sediment Characteristics and Contamination
To assess how the selected indices and metrics were related to sediment characteristics and contamination, we performed RDA analyses (Figure 4
) and supplemented them with correlation analyses (Table S5
). Only the results of RDA analysis without outlier sites are presented to provide a more comprehensive illustration of the relationships with sediment variables and the spatial distribution of sampling sites (see Figure S3
, for the results with the outlier sites). In general, most of the indices and metrics showed higher significant correlations with sediments characteristics than contamination (Table S5
Overall, the RDA model relating indices and metrics to sediment characteristics explained 50% of the total variation in the macroinvertebrate community. We could dissociate two groups of metrics and indices depending on sediment composition, depth, nutrients and inorganic and organic carbon (Figure 4
A). The significant explanatory variables were sand and NH3
on axis 1, DOC and sulfur on axis 2 that explained 39% of the variance. Globally, along the first axis, metrics based on the abundance of tolerant taxa (AGOLD, AOL, %GOLD, ANEM) were associated with nutrient-poor sandy sediments (Sand, low NH3
) in shallow sites; other similar metrics (%NoIns, %OliG) were also associated with nutrient-poor sandy sediments (TN, Ninorg) and organic carbon (DOC, TOC, %TOC, TC). Most of these metrics had higher scores in the Montreal Harbour, Lake Saint-Louis, and Lake Saint-Pierre sites, which were considered to be the most impaired. On the second axis, diversity metrics (DM, SHANG), and metrics based on the relative abundance of ubiquist and sensitive taxa (% Match, %DipG) had higher scores in sediments composed of silt and gravel, with higher pH, sulfur, organic carbon (TOC, %TOC) and nitrogen (TN, Ninorg) levels. Taxa richness metrics (NBTETG, NBTCOBG) had higher scores in shallow sediments poor in nutrients (TP, Pass) and dissolved organic carbon (DOC). Most of these metrics had higher scores in the sites of Lake Saint-Louis and Lake Saint-François, which were considered as the less impaired.
Overall, the RDA model relating indices and metrics to sediment inorganic contamination explained 55.7% of the total variation in macroinvertebrate community. Given the trends observed with inorganic contaminants (Figure 4
B), the sediments most polluted with metals and metalloids were found in the Montreal Harbour and the Lake Saint-Louis (see also Figure S3B
). The significant explanatory variables were Mn and Cd on axis 1 and Cu on axis 2, which explained 50% of the variance. Diversity indices (DM, SHANG) and metrics based on richness or abundance of intolerant taxa (NBTETG, NBTCOBG, %Match, %DipG) were associated with sediments rich in calcium (Ca), mercury (Hg) and arsenic (As) but less contaminated in metals (Cd, Cu, Pb, Zn), most of which were found in the Lake Saint-François sites. The metrics based on tolerant taxa (%OliG) were associated with sediments rich in Cu at the Lake Saint-Pierre sites and other metals mainly at Montreal Harbour sites (relationships are better seen in Figure S3B
). Spearman correlation analysis indicated significant positive or negative relationships between indices and metrics and inorganic contaminants (Al, As, Cd, Cr, Cu, Fe, Hg, Ni, Pb, Zn; Table S5
Overall, the RDA model relating indices and metrics to sediment organic contaminants explained 28.1% of the total variation in the macroinvertebrate community. Given the trends observed for organic contaminants (Figure 4
C), the sediments of the Montreal Harbour were the most contaminated by hydrocarbons and butyltins (PAHs, PAH High, PAH low, C10
Petroleum Hydrocarbons, BT, TBT). The significant explanatory variables were C10
petroleum hydrocarbon, PAH with high molecular weight (HAP high) and total PAH (PAHs). Here again, two types of metrics were opposed (better seen in Figure S3C
). Metrics based on diversity indices and sensitive taxa (DM, SHANG, NBTETG, NBTCOBG, %Match, %DipG) were associated with the less contaminated sites in Lakes Saint-François, Saint-Louis, and Saint-Pierre. On the other hand, metrics based on tolerant taxa (AOL, %OliG, %GOLD, %NoIns) were associated with the most oil- or butyltin-polluted sediments in the Montreal Harbour.
3.4. Responses of Metrics and Indices to Contaminant Thresholds
Cascading homogenous grouping thresholds were determined for the 14 selected indices and metrics using inorganic and organic contaminants in regression tree models. For each index and metric, the estimated thresholds were compared to (i) the criteria established to assess the sediment quality in a remediation context [55
], and (ii) natural concentrations in sediments during preindustrial period < 1950 and postglacial clays [55
] (See Table S6A,B
). A total of 10 over 14 indices and metrics showed robust tree regression models (r2
> 60%). There were divided into two groups: (1) diversity (DM, SHANG), richness (NBTETG, NBTCOBG) and dominance (P5FD) metrics based on ubiquitous and sensitive taxa changed mainly with inorganic contaminants (Pb, Zn, Hg, Ca) rather than with organic contaminants (PCBS, PAHs), (2) metrics based on the abundance of tolerant taxa (%GOLD, %OliG, %Match, %Noinsc) were more related to organic contaminants (PCBS, PAHs) than inorganic contaminants (Cu, As, Cd). An example of a regression tree model for each type of indices and metrics is presented in Figure 5
. The other models are presented in the Figure S4, Supplementary Materials
According to the criteria established for the sediment quality [55
], most of the thresholds determined by the regression trees were below Probable Effect Level (PEL) or Frequent Effect Level (FEL), but still above concentrations in preindustrial sediments except for As (6.6 mg/kg), and in postglacial clays except in some case for Ni, Cr and Cu (75, 150 and 54 mg/kg respectively) or ambient levels except for As (2–7 mg/kg) and Cr (52–93 mg/kg) [55
For the DM model (r2 = 0.65), Pb concentration below the PEL was the first node discriminating 51 sites with a concentration below 61 mg/kg with DM values of 1.35 ± 0.40, followed by calcium dividing these sites into two blocks with 23 sites below and 28 sites above 19,000 mg/kg of calcium. In sites with lower Ca concentration, PCBs were taken into account for the classification of the sites, most sites with PCBs content below 0.0423 mg/kg were discriminated a second time by calcium (11,000 mg/kg) and PAHs (1.66 mg/kg). In sites with Ca concentration above 19,000 mg/kg, contamination by Cd, Hg and Cu below the PEL threshold completed the site classification.
For the NBTETG model, a Pb concentration below the PEL was again the first node discriminating 47 sites below and 12 sites above 45 mg/kg. The 12 sites with higher Pb concentration were divided by a Zn concentration between PEL and FEL (398 mg/kg). The group of 7 sites with higher Pb and Zn contamination had the lowest NBTETG value (0.14 ± 0.38). The 47 sites below the threshold of 45 mg/kg was followed by a separation of sites in two blocks based on PCBs contamination threshold (0.026 mg/kg) below the PEL. At the sites with PCBs concentrations below this threshold (22 sites), Hg, PAHs and Al explained the classification of sites. At sites with higher PCBs concentrations (25 sites), Hg, Cd and Cu explained the classification of the sites.
For the %OliG model, a Cu concentration below the PEL was the first node discriminating 49 sites with higher Cu concentrations (100 mg/kg). For sites with higher Cu concentrations, the Cd concentration was the second node (2.1 mg/kg) separating the sites in two equal numbers (5 sites each). The group of 5 sites with higher Cu and Cd contamination had the highest %OliG value (90.20 ± 13.10%). At the sites with low Cu content, As (3.5 mg/kg) at the lower than pre-industrial concentration was the second node dividing the sites into two equal blocks. At sites with higher As content, PCBs was the third node, and metal contamination by Cd or Pb was the last node. At sites with low As, organic contaminants (PAHs, C10-C50 Petroleum hydrocarbons) were the last nodes.
For the %GOLD model, organic contamination by PAHs was the first node separating 48 sites with higher PAHs concentrations. Pb concentration (63 mg/kg) below the PEL segregated another 41 stations with lower Pb concentration and 7 stations with higher concentration. The 7 sites with higher PAHs and Pb concentration had the highest %GOLD value (90.61 ± 9.71%). The 41 sites with low Pb concentration the second node was divided by Cr (28 mg/kg). The sites with concentration below this threshold (12 stations) was divided by PAHs and the sites with concentration above this threshold (29 stations) were divided by Fe (22 stations), PAHs contamination (17 stations), followed by metal contaminants (Ni, As).
Overall, models for the diversity indices and metrics based on richness (DM, SHANG) and the number and abundance of ubiquitous or intolerant taxa (NBTETG, NBTCOBG, P5FD, %Match) were more related to changes in inorganic Pb and Zn contamination for the first nodes, and organic contamination by PCBs for the second nodes. Metrics based on relative abundance of tolerant taxa (%OliG, %NoIns, %DipG) were also related to changes in inorganic Cu contamination. Metrics based on the abundance of tolerant taxa (%GOLD, AOL, AHIR, ANEM, AGOLD) were more related to organic contamination by PAHs contamination for the first nodes and inorganic contamination by Pb, Cu, and Mn for the second nodes.