Alternatives for the biomonitoring of fish and phytoplankton in tropical streams

Biomonitoring programs need to balance accurate responses in assessments of changes in biological communities with sampling that is fast and low cost. In this study, we evaluated the concordance among fish and phytoplankton communities of streams. We tested the cross-taxa surrogacy, taxonomic, numerical resolution and ecological substitute group (habitat use and trophic guilds) resolution with Procrustes analyses aim of simplifying the biomonitoring process. We collect a total fish abundance of 8,461 individuals, represented by the ecological classes of habitat, including benthic, nektonic, nektobenthic, marginal and trophic guilds by detritivore, terrestrial invertivore, aquatic invertivore, piscivore, algivore and herbivore. We sampled a phytoplankton total density of 1,466.68 individuals/ ml, represented by four Morphology-Based Functional Groups and nine Reynolds Functional Groups. Our results don’t support the use of substitute groups among fish and phytoplankton. For fish, habitat Neotropical Biology and Conservation 14(2): 361–380(2019) doi: 10.3897/neotropical.14.e38088 Copyright Hugo de Oliveira Barbosa et al. This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. RESEARCH ARTICLE Hugo de Oliveira Barbosa et al. 362 use and trophic guild are good surrogates for species-level data. Additionally, our results don’t support the use of functional groups as surrogates for phytoplankton. We suggest the use of higher taxonomic levels (genus and family) and record only the occurrence of species and/or genus for fish and phytoplankton. Our findings contribute to decreasing the costs and time of biomonitoring programs assessments and/or conservation plans on fish and phytoplankton communities of headwater streams.


Introduction
Environmental changes are primarily caused by anthropogenic drivers. As a consequence, we experience an accelerated global loss of species (Ceballos et al. 2015(Ceballos et al. , 2017Crist et al. 2017). In this sense, it is urgent to develop strategies to understand and monitor the consequences of environmental changes on biodiversity (Maćkiewicz et al. 2018). Sampling species along an area of interest and over time is often used as a tool for environmental assessment and monitoring. However, financial and human resources used for these activities are often limited. These resources require optimizing sampling protocols that are less expensive yet do not compromise the quality of information (Kallimanis et al. 2012;Maćkiewicz et al. 2018).
Biodiversity estimates can be time-consuming and expensive (Bates et al. 2007), mainly in regions with large territorial extensions, such as Brazil (Bessa et al. 2011). These factors impose several limitations when associated with the low availability of financial resources for these purposes (Kallimanis et al. 2012). Therefore, many studies investigate alternative approaches to simplify the procedure of obtaining information in biomonitoring, i.e., to reduce the time of identification and associated costs (Landeiro et al. 2012). Proposals for the simplification of biomonitoring programs may include: i.) taxonomic resolution, where higher taxonomic levels (genus or family) are used as surrogates for species (Heino and Soininen 2007;Carneiro et al. 2010;Machado et al. 2015); ii.) numerical resolution, where species presence/ absence data are used as surrogates for abundance data (Landeiro et al. 2012;Rosa et al. 2014;Gomes et al. 2015); iii.) substitute groups, where a group can replace another when they are present in concordance with species distribution variation (Bini et al. 2007;Gioria et al. 2011;Ruhí and Batzer 2014); iv.) ecological substitute group, where the community composition is based on easily recognizable ecological characteristics (e.g. ecomorphology) and a strong power to replace the taxonomic composition, reducing the need for identification at the species level (Carneiro et al. 2010;Trigal et al. 2014;Machado et al. 2015).
Despite the importance and utility of aquatic groups in biomonitoring programs, a great part of simplification protocols have investigated only isolated taxonomic groups. This approach ignores the interaction between assemblages and the potential concordance (Padial et al. 2014). The spatial concordance among aquatics groups can occur because groups respond to similar drivers; for example, water transparency and nutrient limitation (phosphorus and nitrogen) for phytoplankton and periphyton (e.g., Rodrigues and Bicudo 2004). Furthermore, groups with different drivers (e.g. fish and phytoplankton) (Mazaris et al. 2010;Sharma et al. 2016;Erős et al. 2016) may also have concordance due to their links in the trophic web (Thomson et al. 2014). Concordant spatial distribution among different taxocenoses could provide the opportunity for sampling simplification by sampling a single taxonomic group.
The proposed simplification protocols for sampling, identification and characterization of species have involved different water bodies (e.g., streams, rivers, lakes, ponds) and several aquatic groups, such as phytoplankton (Gallego et al. 2012;Carneiro et al. 2010;Machado et al. 2015), zooplankton (Gomes et al. 2015;Vieira et al. 2017), macroinvertebrates (Sanchez-Moyano et al. 2006;Tataranni et al. 2009) and mollusks (Zuschin et al. 2015). However, it is noteworthy that most studies mainly focus on evaluating the use of substitute groups. In a few cases, studies have used other approaches, such as numerical and taxonomic resolution (Grenouillet et al. 2008, see more in Suppl. material 1: Table S1). In this context, we evaluated the concordance of the spatial ordering among fish and phytoplankton communities of streams through four strategies (substitute group, taxonomic, numerical resolution and ecological substitute group) with the aim of simplifying the biomonitoring process. Therefore, our objectives were guided by the following questions: i.) Is there a spatial concordance between species and coarser taxonomic resolution (i.e., genus, family and order)? ii.) Is it possible to replace the abundance/density of species/genus/family data by merely the presence/absence of data? iii.) Is there a concordance between the taxonomic level (species) and the ecological classification of species? And iv.) is there concordance between the spatial distribution of fish and phytoplankton species?

Study area
Sampling was performed during the dry period of 2013 in 29 Cerrado streams, in the Tocantins River basin (Fig. 1). The Tocantins River extends over 1,960 km, with spring in the Goiás plateau, at about 1,000 m altitude. It is formed by the union of Almas and Maranhão rivers, with its mouth in Marajó Bay (Meirelles et al. 2007). The predominant climate of this region is tropical and humid, with two well defined seasons. The rainy period dominates between October and April and the dry period between May and September (MMA 2006). The average annual temperature of the Tocantins-Araguaia basin is 26 °C (MMA 2006). Sampled streams in the subbasin of the Santa Teresa River, North region of Goiás state are mostly of headwater ranging from first to third order (Strahler 1957). In general, the sub-basin of the Santa Teresa River presents a near pristine condition with the maintenance of 75% of the native vegetation remaining in its watershed (Borges et al. 2016). Therefore, although the sampled streams reach a short environmental gradient, 40% of the streams present more than 20% of the land use with anthropogenic activities for cattle raising in the watershed (see Suppl. material 1: Fig. S1). The stretches of streams studied are mainly mesohabitats of backwaters and riffles, a neutral pH, low concentrations of nutrients and chlorophyll-a (Suppl. material 1: Table S2).

Biological data
We used a protocol widely employed for sampling of phytoplankton and fish communities (Uieda and Castro 1999;Bicudo and Menezes 2006;Bellinger and Sigee 2010). Since the planktonic communities are widely dispersed, and greater diversity and abundance occur mainly in lentic environments, we prioritize the mesohabitats of backwaters for phytoplankton collection. In each stream, a phytoplankton sample was performed by collecting approximately 100 ml of water in the subsurface (0.5 m depth) and storing the sample in dark jars. After collection, each sample was fixed with modified acetic acid (Vollenweider 1974). Individual counting was performed in an inverted microscope (Zeiss Axiovert 25), at 400× magnification, following the Utermöl sedimentation method (Utermöhl 1958). Identification was conducted to the lowest possible taxonomic level and organisms were classified according to genus, family and order according to Round (1965), Round (1971) and Round et al. (1990). Afterwards, fish were collected with an electrofishing apparatus along an 80 meter stretch in each stream. Electrofishing was conducted from downstream to upstream, exploring all types of microhabitats along the 80 m reaches (Barbosa et al. 2019). The individual fish captured were fixed in a 10% formalin solution and transferred to 70% ethanol after 72 h. The species were identified to the lowest possible taxonomic level by consulting specialized literature (Claro-García and Shibatta 2013; Lima and Caires 2011; Miranda and Mazzoni 2003) and specialist (Carvalho FR).

Ecological classification data
Phytoplankton species were grouped into two functional groups (Morphology-Based Functional Groups -MBFGs and Reynolds Functional Groups -RFGs) according to the classification proposed by Kruk et al. (2010) updated by Reynolds et al. (2014) and Reynolds et al. (2002) revised by Padisák et al. (2009). The MBFG approach classifies species into eight functional groups according to their morphological characteristics, such as cell size (e.g., biovolume and surface area), silica structure, biological form (unicellular, colonies, filaments) and the presence of flagella, aerotops, heterocysts and mucilage. The RFG group classifies the species by their morphological characteristics, environmental and physiological tolerances, habitat preferences and life history.
The fish species were classified ecologically into three groups: i.) habitat use guilds, according to ecomorphological characteristics, ii.) trophic data, according to diet information, and iii.) habitat use in conjunction with trophic data (see Suppl. material 1: Table S3). In relation to habitat use, fish were classified as benthic, nektonic, nektobenthic and marginal, based on their ecomorphological characteristics, as well as from the literature (Teresa and Casatti 2012). The benthic fish usually have a dorso-ventrally flattened body, long caudal peduncle and ventral mouth (Breda et al. 2005;Oliveira et al. 2010;Negret 2016). The nektonic fish usually present a compressed body, a high, compressed caudal peduncle and a short, terminal mouth (Oliveira et al. 2010). The species of nektobenthic fish are more diverse and generally have a relatively low cylindrical body and a subterminal mouth (Breda et al. 2005), but are also composed of fish with a very high and compressed caudal peduncle, mainly those species of the family Cichlidae. The marginal fish guild is composed of fish with morphological characteristics of other guilds (benthic, nektonic and nektobenthic), but they explore predominantly stream margins and not the main channel. The trophic information was obtained by analyzing the stomach content of fish of the size considered to be at adult stage. For this, three steps were followed: i.) the stomachs were extracted by dissection; ii.) empty stomachs were excluded from the analysis; and iii.) the contents of the remaining stomachs were exposed to Petri dishes and examined under a microscope. The food items were separated into six trophic categories: detritivore, terrestrial invertivore, aquatic invertivore, piscivore, algivore and herbivore. The volume of food items larger than 1mm in height was estimated from the known water volume displacement in the water column. For items smaller than 1mm, the volume was estimated from the area occupied in Petri dish with the aid of the graph paper (Hyslop 1980). To describe the species diet, the Alimentary Index (AI i ) was calculated for each food item (Kawakami and Vazzoler 1980), determined by the following equation: where AIi = alimentary index, n = food item, Fi = frequency of occurrence (%) of each item, Vi = volume of each item in percentage.
The matrices of ecological classification were obtained by multiplying the abundance data (species relative abundance by site matrix) by habitat use (species by habitat use matrix), trophic data (species by diet items matrix) or habitat use together with trophic data (see Suppl. material 1: Table S3) resulting in matrices of ecological groups (habitat use, diet or habitat use together with diet) by site (Lavorel et al. 2008).

Data analysis
The abundance of fish, habitat use and trophic data, and density data of phytoplankton, MBFGs and RFGs were log-transformed (x+1) to minimize the effect of extreme values (Legendre and Legendre 2012). Then, these data were used to build distance matrices using Bray-Curtis (abundance/density data) and Jaccard (presence/absence) coefficients. Subsequently, we applied the Principal Coordinate Analysis (PCoA) to build the fish and phytoplankton matrix to be compared in Procrustes analysis (see more details in Suppl. material 1: Fig. S2). We used a Procrustes analysis to evaluate the concordance between the compared matrices (Legendre and Legendre 2012).
In order to evaluate the taxonomic resolution (proposal of using higher levels), we performed pairwise comparisons of species, genera, families and orders matrices. In this case, we compared the different taxonomic resolutions with fish abundance and phytoplankton density. For numerical resolution (proposal of using occurrence data), the matrices compared were: abundance matrix versus presence/absence matrix for fish and density matrix versus presence/absence matrix for phytoplankton. Numerical resolution was also performed among all taxonomic levels combinations (i.e., species, genera, families and orders). In order to analyze the ecological substitute group (proposal of using ecological classifications), the matrices compared for fish were: species abundance matrix versus habitat use guild, trophic guild, and combined trophic data with habitat use. The analysis of the ecological substitute group for phytoplankton compared the following matrices: density matrix versus MBFGs and RFGs ecological classification. In order to evaluate the concordance between fish and the spatial distribution of phytoplankton (surrogate group proposal), we used both fish abundance matrices versus phytoplankton density and the presence/absence matrices for fish versus presence/absence matrix for phytoplankton.
The Procrustes analysis correlation values (r) range from 0 to 1 (Legendre and Legendre 2012). In general, we considered values of r ≥ 0.7 to indicate highly concordant values (Heino 2010). The significance of the r values was evaluated through 9,999 permutations. All analyses were performed in the R (R Core Team 2018) statistical software. For PCoA ordination analysis, we used the cmdscale function in the stats package. For Procrustes analysis, we used the protest function in the vegan package (Oksanen et al. 2016).

Results
We identified 47 fish species, comprising 39 genera, 16 families and five orders, with a total abundance of 8,461 individuals (see more details about distribution by point in Suppl. material 1: Fig. S3). The most abundant families were Characidae with 5,012 individuals, Crenuchidae with 1,353 and Loricariidae with 1,314 individuals. Regarding the fish ecological classification, based on the position in water column (habitat use guild), we observed 17 benthic, 10 nektonic, 13 nektobenthic and seven marginal species and for trophic guilds we provide the proportions for items found in fish stomachs (Suppl. material 1: Table S3). Regarding phytoplankton, we identified 65 species, distributed in 27 genera, 20 families and 10 orders. The taxonomic class with highest density was Bacillariophyceae, with 1,371.65 individuals/ml, followed by Chlorophyceae, with 91.23 individuals/ml (see more details about distribution by point in Suppl. material 1: Fig. S4). We observed representatives of four MBFGs (Groups III, IV, VI and VII) and nine RFGs (Groups J, F, MP, P, A, D, B, H1, S1). We observed representatives of five MBFGs (Groups III, IV, VI, VII and VIII, Suppl. material 1: Fig S5) and the epilithon species of group VI showed the highest abundance. We also found nine RFGs (Groups J, F, MP, P, A, D, B, H1, S1, Suppl. material 1: Fig. S6), being the MP group, represented by planktonic and epilithon species, the most abundant in most areas of streams.
In general, the taxonomic resolution presented significant results up to the order level (Table 1), although the concordance values decreased with the increase in taxonomic resolution level. Species versus genus and species versus family resolutions were highly correlated for fish (presented greater r values than 0.8). For phytoplankton, only taxonomic resolutions of species versus genus showed significant correlations and r > 0.7 (Table 1). For numerical resolution, although the fish and phytoplankton groups also presented significant results up to the level of order, both indicated the possibility of using presence/absence data as a surrogate for species abundance/density data, genus and family (r > 0.7; P < 0.05) ( Table 1).
Among the ecological classification groups for fish, the concordance tests between species abundance and ecological classification by habitat use together with trophic guilds presented a high concordance (r > 0.7; P < 0.05). The test of concord-ance between species abundance and habitat use group presented significant results with correlation coefficient marginally lower than 0.7 (Table 1). The concordance between species abundance and trophic guild was significant (r > 0.7; Table 1). For phytoplankton, despite significant results for MBFGs and RFGs, correlations were low (r < 0.6) ( Table 1). Fish and phytoplankton data generated non-concordant ordination patterns (P > 0.05) ( Table 1).

Discussion
The distribution patterns of fish and phytoplankton species are maintained at the taxonomic level of genus comparable to those revealed at the species level. For fish, we find concordance with similar predictability power at the family level. Nu-  (Heino 2010). In this sense, we consider these results promising and suggest further tests to evaluate the reliability in using this information as an efficient alternative approach. We support the substitution of taxonomic and numerical resolutions for fish and phytoplankton, as well as ecological substitute group for fish. We also highlight that our results do not support the use of fish as a substitute group for phytoplankton. Nevertheless, we acknowledge the need of caution when applying these coarser measures and suggest them as a potential tool for use at species and genus level. These alternatives are indicated mainly for adverse situations when there is a shortage of resources necessary to implement environmental assessment and monitoring strategies and/or difficulty in accessing expert taxonomists for the identification of fish and phytoplankton.

Taxonomical resolution
Our results indicated that the use of coarser taxonomic resolutions of genus (phytoplankton and fish) and family (fish) may be possible when rapid environmental assessments are required for streams. This is because reaching the species level during identification may be a problem for inexperienced researchers (Williams et al. 2006) and it takes less time to identify taxonomic levels such as genus and family (Kallimanis et al. 2012). In addition, some groups are difficult to identify, generating a dependence on taxonomists. The identification of species by a taxonomist is important; however, access to specialists is not always easy for ecologists and conservation biologists Halme et al. 2015). For example, for fish identification in the Loricariidae family, it is necessary to observe bony structures that require diaphanization with more costly and time-consuming procedures (see Loricariidae fish identification keys in Covain and Fisch-Muller 2007;Vera-Alcaraz et al. 2012). Identification is also difficult for phytoplankton, as it may involve recognizing structures that are not always present in samples (e.g., depending on the algae sample fixation and preservation, some structures can be lost; for example, formalin causes flagella to fall off, hindering the identification of flagellated organisms) (Bicudo and Menezes 2006). Moreover, the identification process also involves groups with different morphologies and requires knowledge of the life cycle (e.g., type of sexual or asexual reproduction) (Carneiro et al. 2013). Therefore, with accelerated biodiversity loss and from a biological and statistical point of view, coarser identification may be more suitable than incorrect species identification (James et al. 1995;Rimet and Bouchez 2012).
Other studies have also found similar results regarding the use of higher taxonomic levels for aquatic organisms, such as benthic macroinvertebrates and diatoms (Heino and Soininen 2007), phytoplankton (Carneiro et al. 2010(Carneiro et al. , 2013, aquatic Nepomorpha (Giehl et al. 2014), zooplankton (Gomes et al. 2015;Mis-sias et al. 2017), phytoplankton, periphyton, zooplankton, aquatic macrophytes and fish (Ribas and Padial 2016). A study with marine mollusks has shown that beta diversity is maintained for genus and family ), demonstrating that heterogeneous patterns are maintained at coarser levels. The use of family as a surrogate for species has already been approached in another study as an important way to decrease the effect of dominant species in a sample (Khan 2006). In case of replacement by a higher level, many studies show that the closer this ratio to 1, the better the higher taxa approach (Bevilacqua et al. 2012;Rosser 2017). Thus, our results support the use of genus for phytoplankton, and genus and family for fish.

Numerical resolution
Our results show that the use of abundance and presence/absence matrices generates concordant patterns. Other studies with different groups such as phytoplankton (Carneiro et al. 2010), plants (Landeiro et al. 2012) and zooplankton (Gomes et al. 2015) agree with the use of presence/absence data instead of abundance/density data. Therefore, the use of presence/absence data for fish and phytoplankton is recommended for streams at species and/or genus levels. As r value for family was very close to 0.7, we understand that in this case the use of numerical resolution for family is a bold and optimistic proposal (Bevilacqua et al. 2012;Rosser 2017). In any case, it is interesting to record solely the occurrence of the species and/or genus, reducing analysis time by dispensing the complete sample count. It is also worth highlighting that the numerical resolution of the presence/absence data approach requires fewer animals to be captured, avoiding unnecessary sacrifice.

Ecological substitute group
Our results found a significant concordance between species abundance/density with the ecological classification for fish and phytoplankton, respectively. However, the low correlation coefficient value of MBFG and RFG ecological classification indicates a non-correspondence of the ecological ordination for phytoplankton species. Therefore, we do not suggest its use as a surrogate for species taxonomic information (Gallego et al. 2012;Machado et al. 2015). For fish, the ecological classification of the habitat, together with trophic guilds, showed higher concordance with species abundance data. Whenever possible, it's interesting to prioritize the application of this classification to monitor streams. A negative point that should be highlighted is that the data obtained may be more complex, especially trophic data. Classification with the trophic guild also presented a high concordance, but to analyze the diet of fish requires a more experienced professional with specific knowledge. We understand that other classifications considering functional fish traits related to breeding and life cycle would contribute to a more robust un-derstanding of the biological response to environmental changes. However, these approaches demand knowledge about the species and this is unavailable in some cases. Therefore, it would not facilitate the application in biomonitoring programs by a less qualified person. Considering the cost-effectiveness of monitoring biological communities in streams, using this ecological classification may be more interesting to complement the assessment of anthropic impacts than to replace species-level taxonomic resolution. The ecological substitute group based on habitat use presented r values marginally lower than 0.7. Thus, we highlight the advantages of this proposal for future studies to facilitate the ease of classification of fish, for the independence in identifying fish at the species level, and the possibility of it being conducted by a trained technician.
Classifying fish in terms of their ecomorphological pattern considers their morphotypes (Casatti and Castros 2006;Oliveira et al. 2010). That is, body characteristics can be easily visually observed (e.g., body compression, caudal peduncle compression and mouth orientation). Nektonic fish, for example, have a terminal mouth, compressed caudal body and a peduncle (i.e., laterally flattened) (Gatz 1979;Watson and Balon 1984;Oliveira et al. 2010). Therefore, a trained technician would be able to work in different regions, without having prior knowledge of a species that occurs in different river basins. Moreover, biological monitoring with an ecomorphological approach for fish encompasses a variety of ecological niches and can provide an impact measurement on aquatic ecosystems, as it may reflect the effect of anthropic pressures (Karr et al. 1986;Barbour et al. 1999;Oliveira et al. 2010). Therefore, the ecological classification of fish is a potential tool that requires further testing to ultimately support this approach in biomonitoring assessments and conservation plans.

Surrogate group
Our results showed no concordance between phytoplankton and fish. Thus, the distribution patterns of both groups can respond differently to preferences and adaptations to available environmental factors. In addition, biotic interactions are possibly weak and may be related to different life history traits (Heino 2002;Bowman et al. 2008;Guareschi et al. 2015). Biological groups that are not phylogenetically related, such as phytoplankton and fish, tend to be irreplaceable (Morais et al. 2018), as corroborated by our results. This finding highlights the unique importance of each taxonomic group in environmental monitoring and biodiversity assessments (Heino 2015). In addition, the lack of concordance among taxa is due to their response to the local environment at different scales (Backus-Freer and Pyron 2015), i.e., the scale may be influencing the lack of concordance. The concordance between groups tends to be larger at large scales, such as several river basins (Grenouillet et al. 2008;Gioria et al. 2011;Backus-Freer and Pyron 2015). In our study, samplings were carried out at a scale relevant to management (landscape scale), and thus require caution to generalize the results to other scales (Landeiro et al. 2012).
Other studies in aquatic ecosystems also support that environmental monitoring based on a single taxonomic group cannot be easily applied to other biotic groups (Heino et al. 2005;Dolph et al. 2011;Larsen et al. 2012;Padial et al. 2012;. Caution when using this type of approach stems from the high variability in levels of concordance between substitutes' groups tested in other studies (Morais et al. 2018). Our results suggest that care should be taken when approaches based on one group are extrapolated to other groups in the assessment of the environmental conditions of headwater streams since the power to generate useful predictions regarding other taxonomic groups may be limited. Therefore, from a practical point of view, we suggest that biologists, environmental consultants, environmental managers and conservation planners rely on approaches of monitoring integrating producer groups (e.g. phytoplankton and periphyton) and consumers (e.g. fish). With these approaches, it will certainly be possible to develop more comprehensive and sustainable conservation strategies (e.g., meeting social, environmental and economic demands).

Conclusions
The best alternative approaches for the biomonitoring of fish and phytoplankton in headwater streams are using higher taxonomic levels (genus and family) and recording only species and/or genus occurrence. For fish, the ecological classification provides useful information, but with a lower level of concordance. In a cost-effective perspective, habitat use could be a good option due to its simplicity in classifying fish independently of taxonomic identification, which could make the biological assessment easy for a less qualified professional. The results found for taxonomic and numerical resolution have been consistent in the literature and are therefore strongly recommended.