Freshwater macroinvertebrate traits assessment as complementary to taxonomic information for mining impact detection in the northern Peruvian Andes

We studied the distribution of freshwater macroinvertebrate taxa and traits to distinguish ecological gradients among the mining‐controlled and natural headwaters, and rural and urban economic activity influences.


| INTRODUC TI ON
The exploitation of freshwater ecosystems has been largely detrimental (MEA, 2005), as evinced by the increased aquatic biodiversity loss (Harrison et al., 2018). Among the many economic activities threatening freshwater ecosystems, mountaintop mining poses complex environmental pressures that result in a cascade of socialecological impacts (Palmer et al., 2010;Vela-Almeida et al., 2015).
Large-scale mining operations affect key mountainous ecosystem functions such as aquifer recharge and habitat quality maintenance, which in turn affect freshwater flow regimes, biodiversity and ecosystem services (Jeronimo et al., 2015;Mercado-Garcia et al., 2018).
The consideration of both direct and indirect influences on ecological interactions is thus crucial to design more sustainable mining approaches (Xu et al., 2018) and mining legacies (Zhao et al., 2018;Zipper et al., 2011). In the exploitation phase, mining watersheds undergo freshwater quality assessments (FWQA) regularly to guarantee the suitability of food production, recreation and biodiversity protection downstream of mining sites. Such a holistic purpose of FWQA requires a mechanistic understanding of the aquatic communities' responses to environmental stressors rather than addressing freshwater quality components separately. Hence the need for ecologically-oriented assessments in the headwaters as well as at downstream reaches of mining watersheds (Mercado-Garcia et al., 2019). Moreover, simple but scientifically sound techniques for FWQA are preferred to induce environmental policy changes (Metze et al., 2017).
A major contribution from the scientific community to FWQA is the simplification of ecological information into one-dimensional values, namely a biological quality or multimetric index, making environmental assessments more accessible to non-expert stakeholders. Taxonomy-based approaches have contributed greatly to such developments (Li et al., 2010;Villamarín et al., 2013). Unfortunately, FWQA in mining contexts lacked progress in bioindicators, multidisciplinary integration and policy relevance aspects (Mercado-Garcia et al., 2019). For instance, the need for calibrating the occurrence of aquatic macroinvertebrates taxa to the site-specific environmental conditions hinders FWQA throughout vast and scarcely studied regions like the Andes (Villamarín et al., 2013). Moreover, although the mining impacts on freshwater macroinvertebrates have been widely reported (Battaglia et al., 2005;Byrne et al., 2012;Dedieu et al., 2015;Ruse & Herrmann, 2000;Zhao et al., 2018), these were mostly demonstrated in the presence of either extreme acidity, peaking metal concentrations or contaminated sediment loads (Gerhardt et al., 2004;Tarras-Wahlberg et al., 2001;Wright & Ryan, 2016).
Furthermore, several 'good-quality' macroinvertebrate taxa (e.g. Plecoptera and Coleoptera orders) have been found in streams with low pH and metal pollution (Gerhardt et al., 2004;Mercado-Garcia et al., 2019;Ruse & Herrmann, 2000;Wright & Ryan, 2016). Studies of mining impacts on macroinvertebrates also argue on whether to aim for higher (Goldschmidt, 2016) or lower taxonomic resolutions (Wright & Ryan, 2016), leaving room for improvement of freshwater bioassessment tools.
The assessment of macroinvertebrates via trait-based approaches (TBA) is promising for advancing FWQA (Menezes et al., 2010). By looking at specific traits in river monitoring sites, the influence of environmental gradients and anthropogenic pressures on biological communities translates into changes in the relative abundance of macroinvertebrate traits. Moreover, TBA offers higher sensitivity because, in the presence of stressors, the distribution of the traits tends to change earlier than the taxonomic composition (Culp et al., 2011). Similar to taxa, traits are influenced by environmental stressors: the traits that impart resistance to stressors prevail, whilst the vulnerable traits fade . Since the potential generalization of TBA relies on the use of consensual ecological knowledge, a major limitation remains in the lack of studies in neotropical regions (Van den Brink et al., 2011) and in mining contexts.

| Aim of this study
The responses of macroinvertebrate traits to mining pressures in the Andes have been still scarcely studied. Generic mining impacts on macroinvertebrates traits include the absence of leaf-like or filamentous gills due to suspended particles (Dedieu et al., 2015), or higher proportions of endobenthic burrowers, collector filterers and operculated gills, which protect against mining debris (Mocq & Hare, 2018), or acid mine drainage causing less functional diversity (He et al., 2015) and smothering by metal flocs. The latter in turn hampers macroinvertebrates that are temporarily attached to substrates, filter-feeders, water column swimmers, free-egg layers or with exposed gills (Reich et al., 2019).
The management of riverine habitats requires substantial ecological characterization and information systematization across multiple geographical regions and stream types (Chapman et al., 2016;Pottgießer & Sommerhäuser, 2008). Since TBA provides mechanistic insights on macroinvertebrate community changes, its application is suited for neotropical ecosystems where freshwater ecological monitoring and knowledge are rather scarce. We conducted a TBA to differentiate gradients in macroinvertebrate communities under different ecosystem types and anthropogenic pressures. The present study complements a previous study that demonstrated a lower biological quality in the mining-impacted headwaters, with insights beyond an excellent physicochemical quality in the headwaters impacted by mining (Mercado-Garcia et al., 2019).

K E Y W O R D S
freshwater quality, hydrological impacts, macroinvertebrate traits, mining, Peruvian Andes 2 | ME THODS

| Study area
The Mashcon catchment is in the northern Peruvian Andes (Figure 1), hosting large-scale gold mining at mountaintop besides rural and urban economic activities downstream (Mercado-Garcia et al., 2019). Two high-altitude ecosystems can be differentiated in the catchment, namely jalca and western-slope relict mountainous forest (WSRMF) (MINAM, 2019). The jalca (3500-4000 m.a.s.l.) is an Andean transition between the northern paramo and the humid puna further south, thus having intermediate climatic characteristics (i.e. more humid than the puna but not as heavily rainy nor as cloudy as the paramo) and hosting mostly herbaceous vegetation and evergreen shrubs (Britto, 2017).
Besides having floral assemblages like the puna and paramo, the jalca is rich in endemic genera such as Agrostis, Poa, Festuca and Arcytophyllum, among others. Contrary to the paramo, the jalca's mountainous landscapes are continuous (MINAM, 2019).
The WSRMF, previously classified within Quechua (Vidal, 2014) or meso-Andean (Britto, 2017) ecosystems, refers to altitudes between 1400 and 3000 m.a.s.l. and it is characterized by epiphytes and 15 metres high canopies (MINAM, 2019). The WSRMF hosts a combination of evergreen shrubs, arboreal and perennial herbaceous vegetation (Vidal, 2014). Due to the temperate climate, human activities proliferate in the WSRMF more than in the jalca.
Part of the WSRMF has been converted into agricultural plots and residential areas. The headwaters are formed in the hydrological centre Cajamarca-Hualgayoc, where major gold and copper deposits exist (Bissig et al., 2015). Open-pit gold mining has impacted the freshwater physicochemical quality between 1991 and 1993 (Bury, 2004), besides decreasing the groundwater table in the jalca. Consequently, the Yanacocha mining company installed advanced water treatment facilities (Miller et al., 2008) and an artificial system to recharge the main river headwaters, as required by downstream rural communities and 70% of Cajamarca's municipal water (Vela-Almeida et al., 2016).

| Biophysical data collection
The freshwater quality was assessed in the dry season of 2016 at 40 locations selected according to potential landuse impacts (e.g. urban, agriculture or mining) and reference conditions (i.e. for dissolved metal analyses. The latter step was repeated without filtering for total metal analyses. A Thermo Fisher Scientific ® iCAP F I G U R E 1 The Mashcon watershed in northern Peru, its digital elevation model and the 40 sampling sites divided into 4 watershed sections. The headwaters correspond to the high mining (Grande River) and high tributaries groups. The latter consists of the Grande River's tributaries out of direct mining influence. The peri-urban area includes the Porcon River, which mixes with the Grande River to form the Mashcon River (urban area) 6000 ICP-OES ® was used to measure concentrations of cadmium, barium, copper, chromium, iron, nickel, manganese, lead, zinc, cobalt and arsenic. Geocoordinates and altitude were recorded using a Garmin eTrex ® HC series GPS.
Freshwater macroinvertebrates were collected with a standardized kick-sampling method, by simultaneously kicking and sweeping the river substrate using a hand-net (frame size 20 × 30 cm; mesh size 500 µm) to sample as many microhabitats as possible, and for 5 minutes at each location (Acosta et al., 2008). Macroinvertebrates were collected from the substrate samples and stored in 75% ethanol. The taxa were counted and identified at the family level by an expert taxonomist, except for the Acari subclass. The level of taxonomic resolution was selected as being sufficient to detect environmental and biodiversity gradients of freshwater macroinvertebrates communities (Mueller et al., 2013).

| Estimation of macroinvertebrate trait abundances
The macroinvertebrate taxa counts were complemented with traits information. Eight traits were selected (Table 1) according to the availability of databases of sufficient taxonomic resolution and suited to Andean ecosystems. Traits of macroinvertebrates were obtained from Tomanova et al. (2008), Tomanova and Usseglio-Polatera (2007) and Tachet et al. (2000). These consisted of five biological (maximal body size, food source, body form and respiration) and three ecological (mobility and attachment to substrates, feeding habits, saprobity and pH preferendum) traits (Table 1). A traits matrix was created containing the modalities' affinity scores for 43 taxa (S3). Six families (Blephariceridae, Chordodidae, Curculionidae, Gripopterygidae, Hyalellidae and Hydrobiosidae) were excluded due to lacking traits information.
A fuzzy coding procedure was applied to estimate the weighted traits abundance (Forio et al., 2018), allowing multiple affinities for certain taxa. For example, Baetidae feeds by scraping fine particulate organic matter, microphytes and microinvertebrates from surfaces, but it is also able to feed on deposited debris sometimes.
Thus, considering the modalities 'x','y' and 'z' for a given trait, and supposing there are 100 individuals of a taxon, and knowing that this taxon has an affinity of 3 for modality 'x', 1 for modality 'y' and no affinity for modality 'z', then the modality 'x' was given an abundance of 75 counts, modality 'y' 25 counts and modality 'z' 0 counts. In addition, a dominant-trait coding procedure was applied to the initial traits matrix, disregarding the non-dominanttrait modalities. In the above-mentioned Baetidae example, modality 'x' dominates the others, thus the dominant-trait approach assigns 100 counts for modality 'x'. If more than one modality were equally dominant in a taxon, the affinity scores had to be assigned by equal fractions totalling 1, allocating equal counts to the equally dominant modalities. In both quantification methods, the abundance per sampling site was estimated by adding up the modality counts from each taxon.

| Data processing
The biophysical data of 40 sampling sites were compared among watershed sections using the procedure depicted in Figure S4. The abiotic measurements (S1) were divided in four groups: high mining (n = 6), high tributaries (n = 8), midstream (n = 14) and urban (n = 12). The medians and value ranges for each group were identified. The Wilcoxon Rank Sum test and (Welch's) Two-sample t-test were applied to compare the groups by pairs. The Shapiro-Wilk, ANOVA and Kruskal-Wallis tests were used to verify the tests' assumptions. Statistically significant differences were identified at either p < .05 or p < .001.
The taxa's average abundance per watershed section was assessed as: where a i is the sampling site's abundance of taxon 'i'; x 1 is the first sampling site of the watershed section; X n is the last sampling site of the watershed section; n is the number of sites in the watershed section; and AA i is the average abundance of taxon 'i' per sample at the watershed section. Taxa resulting in averaged abundances ≤0.2 were excluded from the traits quantification.
Regarding the traits, the relative abundance per watershed section was assessed as: where a j is the sampling site's abundance of a given trait modality 'tm'; x 1 is the first sampling site in the watershed section; X n is the last sampling site in the watershed section; ∑m 1 is the sum of the watershed section sites' abundances for modality 1 of the trait; ∑m n is the sum of the watershed section sites' abundances for modality n of the trait; and RA tm (%) is the relative abundance of trait modality 'tm' in the watershed section. The latter result was used to generate bar plots, enabling a comparison of traits composition among watershed sections as well as between the weighted and dominant-traits quantification.
The ecological interpretability of the ungrouped data was verified using the R software (R Core Team, 2021) and the 'vegan' package (Oksanen et al., 2020) to conduct a constrained ordination.
A detrended correspondence analysis was performed on the ecological data (i.e. environmental variables and biological counts per sampling location, either taxa or traits) to find the length of gradient ('axis length' in R) using the 'decorana' function. Axis lengths below three indicate a redundancy analysis (i.e. linear method) is needed and above three a Canonical Correspondence Analysis (i.e. unimodal method). The distance measures for the ecological data were ranked using 'rankindex' to select the highest index needed for a distancebased Redundancy Analysis (dbRDA). The highest ranked distance indexes (i.e. either Chao or Cao for this study) were specified in the function 'capscale' to produce a dbRDA model. Collinearity of environmental variables was identified by calculating their variance inflation factors. The latter uses 'vif.cca' on the dbRDA model and removes abiotic variables until all the variance inflation factors are lower than three (Forio et al., 2017). The final three dbRDA models, for the taxa, the weighted and the dominant-traits abundances, were plotted using the 'plot' function. 8. pH preferendum  ≤4 pH1

| Freshwater abiotic conditions
>4-4.5 pH2  (Table 2). Moreover, the concentrations of barium and manganese were always higher in dissolved forms than in particulate forms. Table 3 shows the averaged-per-sample abundance in each watershed section for each taxon, including identifiers serving as labels for Figure 3b. An upstream to a downstream increment of the total averaged taxa abundances was found, with a contrasting reduction in the high mining section (i.e. one-fourth of the averaged abundance found in the higher-altitude tributaries

| Differentiating traits distributions
The traits bar plots (Figure 2) show the relative abundance of the modalities (percentages) in each watershed section for the weighted TA B L E 2 Abiotic conditions measured at the sampling locations: medians (variation ranges) and significant differences among watershed sections

| Distance-based redundancy analysis (dbRDA)
The dbRDA plots (Figure 3) yielded fair representations of the sampling sites, although to a lesser extent when using taxonomic ( Figure 3b) and weighted traits information (Figure 3c). The ranking families tightly clustered at the centre of the plot (Figure 3b). An overlap between the high tributaries and midstream groups is visible in the traits-based plots (Figure 3a and c), and to a lesser extent when using taxonomic information (Figure 3b). According to the environmental vectors, the high tributaries and high mining groups followed the direction of increasing altitude and velocity and decreasing pH; the urban section was in the direction of increasing temperature, dissolved metals except zinc, chlorophyll-a and pH; and the midstream section was in the direction of increasing oxygen saturation and dissolved zinc.

| Bioassessment of the gold-mined catchment
From both the naturally and artificially recharged headwaters in the jalca, to the midstream rural settlements and downstream Cajamarca city, the watershed sections were characterized by the differences in abiotic variability (Table 2), macroinvertebrate assemblages ( Figure 3b) and averaged abundances (Table 3). By including traits information, we identified an upstream to downstream evolution of phenotypic traits and patterns among watershed sections for the ecological and physiological traits (   Table 1 (mbs = maximal body size, mas = mobility and attachment to substrate, food = food sources, fhs = feeding habits, bf = body form, res = mode of respiration, sap = saprobity or sensitivity to organic pollution, pH = pH preferendum). The taxa-type (grey labels, b) abbreviations are those from Table  3 (e.g. r = Helicopsychidae, z = Hydranidae). A few not-shown modalities were overlapping at the plot's centre (i.e. mbs ≤0.25 cm and >2-4 cm; mas full-water swimmer, endobenthic burrower and temporary attached; food coarse detritus, dead animals and microinvertebrates; res plastron and stigmata; bf spherical; fhs shredder and piercer; sap oligosaprobic and pH > 5.5) or remained unreadable near to the plot's centre (i.e. mbs > 1-2 cm and >4-8 cm; food macrophytes and macroinvertebrates; fhs scraper and predator). The scattered trait modalities include mbs ≤ 0.25 and >2-4 cm; mas full-water swimmer, endobenthic burrower and temporary attached; food coarse detritus, dead animals and microinvertebrates; res plastron and stigmata; bf spherical; fhs shredder and piercer; sap oligosaprobic and pH > 5.5-6, >6 watershed thanks to the slope-induced mechanical oxygenation, as reported in high-altitude streams of rough topography (Zhou et al., 2017).
Although the maintenance of physiological and ecological traits makes them less suitable for human impact detection (Menezes et al., 2010), a few slight changes occurred in the anthropized sections. For instance, the removal of riverbank vegetation in urban areas is reflected in fewer shredders as in Villamarín et al. (2013).
A decrease in crawlers and increasing polysaprobic shares reflect the deterioration of river bottoms and water quality in the urban section. Conversely, the decrease of tegument respiration in the oxygen-rich high mining section (Figure 2e) is unprecise towards the type of anthropogenic impact since the excluded, yet increased, Gripopterygidae family (Table 2) uses cutaneous respiration in its initial nymphal stages (Gutiérrez-Fonseca, 2010). Andean autecological studies would be needed at different developmental stages of Gripopterygidae. Moreover, the smallest macroinvertebrates were exclusive to both high-altitude sections (>3000 m.a.s.l.), and these have been associated with reference conditions in French Guianese watersheds impacted by artisanal gold mining (Dedieu et al., 2015).
The traits shares found in the high tributaries, while different in the high mining section, recovered in the midstream section ( Figure 2) coinciding with higher and more fluctuating values of sulphates and electric conductivity nearby rural settlements (Table 2).
Mining pollutants are rather complex in types, sources and ecological effects. The correlation between the traits-based ordinations and the altitude, DO%, stream velocity and dissolved metal measurements points to hydromorphological and geochemical factors steering the community traits. A study of high-altitude water springs in the Cajamarca province found groundwater flows and hydrogeochemical processes, such as pyrite weathering and calcium carbonate dissolution, causing the natural variability of pH and ionic concentrations (Cooper et al., 2010), relatable to the physicochemical quality of the high tributaries and the presence of acidophilic modalities (Figure 2h).
Unfortunately, the typification of Andean freshwater ecosystems lacks the characterization of gradients in the physical environment and floral assemblages (Mercado-Garcia et al., 2019;Sylvester et al., 2017;Villamarín et al., 2013). Naturally occurring ranges of salinity and pH in waterbodies enable the in-situ adaptation of microbial assemblages (Waiser & Robarts, 1995), and functional compensations in macroinvertebrate communities despite their changing taxonomic composition (Péru & Dolédec, 2010;Petrin et al., 2008). For instance, the clustering of sampling sites after translating the taxonomic information into weighted traits ( Figure 3)  information, making the study of traits trade-offs and compensations (Boersma et al., 2014;Menezes et al., 2010;Schmera et al., 2017) a necessary addition to traits assessments.
The high mining section's physical environment differed from the other rural sections in the sense of having larger boulders in the substrates, less shading, concrete riverbanks and channels, the most fluctuating stream velocities and the narrowest physicochemical quality ranges. Such distinctive abiotic characteristics coincided with a reduced macroinvertebrate community dominated by Elmidae and Gripopterygidae (Table 3) (Table 3) belongs to the Amphipoda order and looks similar to Gammarus, which is known to have delicate current velocity requirements (Vadher et al., 2018). The hydrological regimes, and the differences in stream size, order and slopes play an important role in the definition of reference conditions for traits compositions (Horrigan & Baird, 2008

| Physicochemical quality control impacts
The absence of microorganisms as a food source in the high mining group (Figure 2d) might be reflecting a deleterious effect of the controlled physicochemical quality, since chlorination is used at the mining camp to eliminate traces of cyanide before recharging the headwaters (Miller et al., 2008). Microorganisms play an important role in symbiotic (e.g. insect's gut microbiota) and mutualistic (e.g. litter decomposition) interactions with macroinvertebrates (Emilson et al., 2016;Wielkopolan & Obrępalska-Stęplowska, 2016). Likewise, the electric conductivity thresholds for the Hyalellidae's crust formation (Colla & César, 2019) make it a sensitive organism to slight changes in physicochemical water quality. Moreover, the lower share of scrapers (Figure 2) suggests that the mining section lacks periphyton, which is a biologically complex biofilm supporting higher trophic levels. The growth of periphyton in freshwaters of intermediate-tolow acidity and low metals concentrations is governed by thresholds in the water alkalinity and nutrients. For example, calcareous geology reduces algal diversity in periphyton (Smucker et al., 2014).
For the Mashcon watershed, although the reactivity, bioavailability and ecotoxicity of metals are expected to increase in dissolved forms (Kravchenko et al., 2014), the physicochemical quality variations constituted healthier freshwater habitats in the high tributaries when compared to the mining section. Presumably, the stringent physicochemical quality in the mining section misrepresents the jalca's natural gradients in physicochemical water quality (Cooper et al., 2010).

| Future research
The lack of macroinvertebrate traits databases for (neo)tropical regions and the growing evidence that bioindicative traits vary among the different genera and species of ubiquitous macroinvertebrates families (Colla & César, 2019;Goldschmidt, 2016;Serra et al., 2017; Wielkopolan & Obrępalska-Stęplowska, 2016) point to a need for autecological studies in Andean mining regions of Peru. The latter is particularly for the scarcely characterized Grypoterigidae and Hyallelidae families, and for ubiquitous taxa like Acari and Chironomidae. Horrigan and Baird (2008) recommended the characterization of freshwater nutrients in flow-sensitive traits assessments, acknowledging the complexity of freshwater ecological interactions. In fact, the sources and requirements of the freshwater nutrients, metals bioavailability and microbial enzymatic activities would be more insightful measures of ecosystem processes (Smucker et al., 2014).

| CON CLUS ION
We found ecological impairment (less macroinvertebrate diversity and changing phenotypic traits compositions) in the mine-impacted headwaters despite the excellent physicochemical quality of the water discharged by the mine. Presumably, the environmental norms for large-scale mining operations are unaligned with the ecological requirements of Andean freshwater ecosystems. In addition to the generic mining impacts on freshwater quality, traits-based assessments help understanding the ecological impacts of artificial flow regimes and hydromorphological changes induced by mining operations.

ACK N OWLED G EM ENTS
This work was supported by FONDECYT-CONCYTEC (grant contract number 002-2016-FONDECYT) and a VLIR-TEAM programme (ZEIN2013PR395: 'Impact on surface water resources and aquatic biodiversity by opencast mining activities in Cajamarca, Peru'). The authors would like to thank Dr. Koen Lock for the identification of macroinvertebrates, and the colleagues from the National University of Cajamarca for facilitating the field campaign in Peru.

CO N FLI C T O F I NTE R E S T
The authors declare they have not known competing financial interests nor personal relationships that could have appeared to influence the work reported in this paper.

DATA AVA I L A B I L I T Y S TAT E M E N T
The raw macroinvertebrate taxa counts per sampling location, the affinity scores for both the weighted and dominant-trait approaches, and the traits counts are in the S3, and the abiotic measurements in the S1 file. The hydromorphological assessment and geocoordinates of the sampling locations are in the S2. Datasets available in https:// doi.org/10.5061/dryad.ngf1v hhwj.