HAL Id: hal-02917086
https://hal.archives-ouvertes.fr/hal-02917086
Submitted on 18 Aug 2020
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Comparing taxon-and trait-environment relationships in stream communities
Victor Satoru Saito, Tadeu Siqueira, Luis Mauricio Bini, Raul Costa-Pereira, Edineusa Pereira Santos, Sandrine Pavoine
To cite this version:
Victor Satoru Saito, Tadeu Siqueira, Luis Mauricio Bini, Raul Costa-Pereira, Edineusa Pereira San-
tos, et al.. Comparing taxon-and trait-environment relationships in stream communities. Ecological
Indicators, Elsevier, 2020, 117, pp.106625. �10.1016/j.ecolind.2020.106625�. �hal-02917086�
Author contributions: VSS and TS formulated the initial idea with inputs from LMB and RCP.
Comparing taxon- and trait-environment relationships in stream
1
communities
2
Victor Satoru Saito¹, Tadeu Siqueira², Luis Mauricio Bini³, Raul Costa-Pereira
4, 3
Edineusa Pereira Santos
5, Sandrine Pavoine
64
5
1-Corresponding author. Departamento de Ciências Ambientais – Universidade 6
Federal de São Carlos, – Rodovia Washington Luís, São Carlos, 13565-905, Brazil - 7
victor.saito@gmail.com 8
2-Instituto de Biociências, Universidade Estadual Paulista (UNESP), Rio Claro, Brazil 9
3-Departamento de Ecologia – Universidade Federal de Goiás, Goiânia, Brazil 10
4-Departamento de Biologia Animal, Instituto de Biologia, Universidade Estadual de 11
Campinas, Campinas, Brazil 12
5-Programa de Pós-Graduação em Ecologia e Biodiversidade, Instituto de Biociências, 13
Universidade Estadual Paulista (UNESP), Rio Claro, Brazil 14
6-Centre d'Ecologie et des Sciences de la Conservation (CESCO), Muséum national 15
d’Histoire naturelle (MNHN), Centre National de la Recherche Scientifique (CNRS), 16
Sorbonne Université, CP135, 57 rue Cuvier, 75005 Paris, France 17
18
Highlights:
19
1- Trait-environment relationships are weaker than taxon-environment 20
relationships 21
2- Traits are interrelated and do not respond monotonically to environmental 22
changes 23
3- Species with different suite of traits can respond similarly to the environment 24
4- Taxonomic composition is more related to spatial variables than trait 25
composition 26
5- Statistical developments are still needed to improve trait-environment analyses.
27
Abstract
28
Traits define how organisms interact with their surrounding environment and with other 29
organisms. Thus, trait composition of biological communities is expected to change 30
predictably along environmental gradients. Because organisms’ traits, but not 31
taxonomic identity, determine their fitness, trait-environment relationships should 32
provide a better way to elucidate how biodiversity respond to environmental change.
33
Here, we used data on tropical streams embedded in a landscape of intensive agriculture 34
to investigate trait-environment and taxon-environment relationships in a set of 91 35
mayfly communities from southeastern Brazil. We expected that trait-environment 36
relationships would be stronger than taxon-environment relationships and that the 37
linkage between traits and environmental variables would provide mechanistic insights 38
on environmental filtering. We found that variation in both species composition and 39
traits were correlated to salinity, highlighting the influence of water salinization on 40
mayfly communities due to agricultural practices. Surprisingly, using analogous 41
statistical methods, in general, we found that the strengths of trait-environment 42
relationships were lower than that of taxon-environment relationships. Further, (1) 43
species responses to gradients were not correlated to similarity in their traits and (2) 44
some species with different trait composition responded similarly to environmental 45
variation, indicating that different suite of traits can cope with similar environmental 46
contexts. Besides some cautionary results about trait-based approaches, results from 47
taxon-based approaches indicated that variation in composition was more related to 48
spatial variables, suggesting that dispersal limitation undermine its use for large scale 49
assessments. Our results suggest that both taxon- and trait-based approaches have 50
weakness and strengths and deciding between them for biomonitoring purposes will 51
depend on spatial scales, trait interrelationships, and analytical methods.
52 53
Keywords: distance-based redundancy analysis, community weighted mean, species 54
trait, mayflies, tropical streams, trait-environment relationship 55
56
57
58
1. Introduction
59
The ‘habitat templet’ concept is a major principle of modern trait-based 60
approaches (Townsend and Hildrew 1994; McGill et al. 2006; Dray and Legendre 61
2008), which states that species evolve ecological strategies that “maximize the numbers 62
of their descendants in their habitat” (Southwood 1977). Accordingly, ecologists have 63
sought for relationships between species’ strategies and environmental variables under 64
the assumption that this would reveal how the environment arranges natural 65
communities (McGill et al. 2006; Dray and Legendre 2008). A prevalent approach to 66
define species’ strategies is to measure their functional traits, i.e., phenotypic 67
characteristics of organisms that are expected to influence their fitness (McGill et al.
68
2006). Species with functional traits that maximize their population growth should be 69
abundant, reflecting successful recruitment under a given environment (Tilman 2004;
70
Shipley 2010; Shipley et al. 2016). Therefore, the relative representation of traits in 71
local communities should change along environmental gradients, which have been 72
called the trait-environment relationship (Townsend and Hildrew 1994; Dray and 73
Legendre 2008).
74
Trait-based approaches are thought to have two main advantages when 75
compared to taxonomic approaches. First, variation in trait composition along 76
environmental gradients have the potential to provide mechanistic understanding of 77
community assembly, as traits should respond in predictable ways given their role in 78
shaping individual fitness (Shipley 2010). Second, trait-based approaches enable 79
comparing biodiversity patterns along gradients (e.g. human pressures) at large spatial 80
scales, while taxonomic analyses are contingent to the biogeographic context of the 81
species pool because dispersal limitation also influences species distribution (McGill et 82
al. 2006).
83
The potential advantages of trait-based approaches boosted their use in 84
freshwater ecology in the last three decades. Morphological, ecological, and life-cycle 85
traits have been traditionally used in biomonitoring protocols for diverse freshwater taxa 86
(Resh et al. 1994; Townsend and Hildrew 1994; Poff 1997 Dolédec, Statzner and 87
Bournard 1999; Verberk, Van Noordwijk and Hildrew 2013). It has been suggested that 88
using trait-based approaches would improve our ability to elucidate the sources of 89
environmental change (e.g. organic pollution, flow alteration, salinization) (Dolédec, 90
Statzner and Bournard 1999, Aspin et al. 2019). However, although some studies have 91
indeed found strong trait-environment relationships (e.g., macroinvertebrates: Dolédec, 92
Statzner and Bournard 1999; fishes: Leitão et al. 2016), many others have found weak 93
or no relationship (e.g. Resh et al. 1994; Davies et al. 2000; Saito et al. 2016; see a 94
review in Hamilton et al. 2019). Importantly, explanations for weak or absence of trait- 95
environment relationships in freshwater communities are still unclear (Statzner, Dolédec 96
and Hugueny 2004). Thus, although trait-based approaches have been present in the 97
literatures for decades, comparing their predictive performance against taxonomic 98
approaches remains a challenge, especially in tropical regions where freshwater 99
biomonitoring is still under development (Saito et al. 2015a).
100
In this study, we investigated trait- and taxon-environment relationships in 101
tropical mayfly (Ephemeroptera) communities along a gradient of land use 102
intensification. We focused on mayflies but not multiple insect orders because each 103
taxonomic group is likely to have developed evolutionary solutions to environmental 104
changes so that merging multiple groups could mask the responses of individual orders 105
(Resh et al. 1994). We studied streams embedded in a landscape with different levels of 106
forest cover and anthropogenic influence (mainly sugar cane plantations and pasture) in 107
Brazil. We aimed to investigate two assumptions of trait-based approaches: (i) traits
108
provide mechanistic explanations of environmental filtering that taxon-based 109
approaches do not; (ii) taxon-environment relationships are dependent on the species 110
pool given the influence of species dispersal for taxa distribution while trait approaches 111
are not. Thus, we expected that (1) trait-environment relationships would be stronger 112
than taxon-environment relationships. Because traits should allow a mechanistic 113
explanation of the role of environmental filters, we expected that (2) gill traits (shape 114
and number), which are strongly associated with osmoregulation and respiration in 115
mayflies (Wingfield et al. 1939), would respond to salinity and dissolved oxygen levels;
116
and (3) feeding strategies would change according to stream width and riparian forest 117
cover because these variables determine the organic matter inputs in streams (Vannote 118
et al. 1980). Finally, we expected that (4) variation in trait composition would be less 119
constrained by the spatial structure of the landscape than the variation in taxonomic 120
composition, since trait distribution should not be strongly associated to dispersal.
121
122
2. Methods
123
2.1 Sampling sites 124
We used data from 91 streams in the Corumbataí River Basin, state of São 125
Paulo, Brazil (Fig. 1). The basin has an area of ~170,000 ha and deforestation was the 126
primary impact in the beginning of the 1900s. After that, the region has been intensively 127
used for sugarcane production and cattle ranching (Valente 2001).
128
Sampled streams were distributed along a gradient of forest cover, which is a 129
major predictor of variation in stream insect communities in Brazil (Siqueira et al.
130
2012a,b, 2015; Saito et al. 2015a). We calculated the proportion of different land uses in 131
the catchment area upstream the sampling sites. We calculated the area covered by 132
pasture, sugar cane, and native forest using the software QGIS (Quantum GIS
133
Development Team, 2019). Catchment forest cover ranged from 0 to 83% within the 134
river catchment area.
135
136
Figure 1. Location of studied streams within the Corumbataí River Basin. Streams are 137
represented by dots.
138 139
We measured the following in situ stream variables to characterize local 140
environmental variation: % of shading at the stream stretch (visually estimated), average 141
stream width and depth (cm), flow velocity (m/s), water temperature (
oC), pH, 142
conductivity (µS/cm), salinity (mS cm
-¹), turbidity (NTU), dissolved oxygen (mg/L) and 143
substrate composition (sand, gravels, pebbles, cobbles, and boulders; %). Water 144
characteristics were measured using a multiparameter sensor (Horiba device U‐50).
145
Additionally, water samples were taken from each stream for the laboratory analysis of 146
total phosphorus (mg/L) and total nitrogen (mg/L) using methods described in 147
Golterman, Clymo and Ohnstad (1978) and Mackereth, Heron and Talling (1978),
148
respectively. Based on a correlation matrix between all variables, we decided to exclude 149
the variables conductivity and % of pebbles from subsequent analysis due to collinearity 150
(Pearson’s correlations = 0.83 and -0.60 with salinity and % of boulders, respectively).
151
We decided to keep salinity instead of conductivity given the alarming impact that 152
freshwater salinization may have over mayflies and given our specific predictions about 153
salinity and gill traits (Kaushal et al. 2005; Cañedo-Arguelles et al. 2016). To 154
summarize the environmental data (standardized), we applied a Principal Component 155
Analysis (PCA) to reduce the dimensionality of the environmental variables data (prior 156
to standardization, continuous variables, except pH, were log-transformed and 157
percentage variables – e.g., substrate composition – were transformed using the square 158
root of arcsine).
159
160
2.2. Mayfly data 161
Between May and August (dry season) of 2015 we sampled mayflies using a two 162
minutes kick-net procedure in each sampling site. We counted and identified all nymphs 163
to the lowest possible taxonomic level (morphotype and species level). Based on 164
previous studies and the natural history of Neotropical mayflies, we selected six key 165
traits: feeding strategy, gills characteristics, body shape, locomotion strategy (all fuzzy 166
coded trait), number of gills (quantitative discrete trait), and wing size (quantitative 167
continuous traits). We compiled trait values from previous studies and databases 168
(Tomanova, Goitia and Helešic 2006; Colzani et al. 2013; Saito et al. 2015a).
169
Additionally, for each species, we measured the mean body size (quantitative 170
continuous trait) of nymphs (last instar, minimum of five individuals) under a 171
stereomicroscope (Table S1).
172
173
2.3 . Taxon- and trait-environment relationships 174
To describe trait syndromes in mayfly traits (i.e., trait combinations that are 175
consistent among different species), we applied a PCA to the trait data after 176
standardizing fuzzy and quantitative traits by zero mean and unit variance. We then 177
compared taxon- and trait-environment relationships using the following multivariate 178
methods (Kleyer et al. 2012):
179
1) Redundancy analysis (RDA) estimates the relationship between a multivariate 180
explanatory matrix and a multivariate response matrix (Legendre and Legendre 2008).
181
We used the environmental variables (described above) as the explanatory matrix and 182
the community composition dataset ( Hellinger transformed ) as the response matrix in 183
the taxon-based approach (RDA-Composition). For the trait-based approach we used 184
the community-weighted mean (CWM) values of traits as the response matrix (RDA- 185
CWM). CWM is the mean value of a trait, weighted by species abundances, considering 186
all species present at a site (Lavorel et al. 2008), being one of the most used metrics to 187
test trait-environment relationships (Lavorel et al. 2008; Bruelheid et al. 2018, but see 188
Peres-Neto, Dray and ter Braak 2017). We applied a forward selection procedure with 189
double stopping criterion (P-values and adjusted R²; see Blanchet, Legendre and 190
Borcard 2008, Bauman et al. 2018) to select a subset of environmental variables to be 191
used in the RDA. We used an ANOVA-like procedure with 999 permutations 192
(Legendre, Oksanen and ter Braak 2011) to test for the significance of RDA axes.
193
2) Distance based-RDA (db-RDA) is a variation of RDA to cope with 194
dissimilarity matrices as response matrices (Legendre and Anderson 1999). db-RDA 195
finds linear relationships between all the Principal Coordinate Analysis (PCoA) axes 196
(summarizing a dissimilarity matrix between sites) and environmental variables. We 197
used the Bray-Curtis dissimilarity as response matrix in the taxon-based approach (db-
198
RDA-Composition). For the trait-based approach, we used the trait dataset to calculate a 199
dissimilarity matrix among sites that provides maximum dissimilarity between 200
completely dissimilar communities (D
βcoefficient - equation 12 in Pavoine and Ricotta, 201
2014) (hereafter the db-RDA-Traits approach). This coefficient relies on differences in 202
traits between species, which were calculated using the modified Gower coefficient 203
(Pavoine et al. 2009). We also used estimated P-values through an ANOVA-like 204
procedure with 999 permutations (Legendre, Oksanen and ter Braak et al. 2011) to test 205
for the significance of the axes. In all analyses described above, we used the adjusted 206
coefficient of determination (Adj. R²) to measure the strength of relationship between 207
the response and explanatory matrices (Peres-Neto et al. 2006). We also added a 208
constant to dissimilarities to avoid negative eigenvalues during ordination of matrices 209
(Cailliez transformation, Legendre and Legendre 2008).
210
3) In order to find a direct link between taxa, traits and the environment, we 211
applied a RLQ analysis that simultaneously evaluate the relationships among sites, 212
species, environmental variables and traits using simultaneous ordination (Dolédec et al.
213
1996). RLQ relies on covariance measures between environmental variables and traits 214
linked by the abundance of species in sites. For this approach, we considered the 215
continuous traits number of gills, wing size and body size as fuzzy coded, separated in 216
three distinct blocks with affinities sum to unity for a given species. We did that 217
because the implemented RLQ statistical test cannot cope with the mixed type of data 218
we used (feeding fuzzy-coded traits and body size; see also Thioulouse et al. 2018).
219
Significance testing of the general relationships between species and the environment 220
and between traits and the environment was based on a sequential test through a Monte- 221
Carlo procedure with 9999 permutations that randomizes sites and species (model 6 in 222
Dray et al. 2014, which combines in sequence the randomizations of model 2 and 4).
223
Model 2 tests the relationship between species and the environment, while model 4 tests 224
the relationship between traits and the environment . For this method, we used only the 225
environmental variables selected by forward selection in the previous RDAs.
226
4) In order to find specific relationships between individual traits and 227
environmental variables, we used the fourth-corner method (Dray and Legendre 2008) 228
to estimate the Pearson’s correlation coefficient between each trait and each 229
environmental variable. Again, significance levels were assessed by the model 6 230
described by Dray et al. 2014, which combines in sequence the randomizations of 231
model 2 and 4.
232
5) We tested the association between the trait distance matrix (modified Gower 233
distance) and the environmental response distance matrix (Euclidean distance matrix 234
from the species scores along the first three axes of the RDA-Composition) using a 235
Mantel test (9999 permutations). The first matrix describes the similarities between 236
species in terms of traits composition, while the second matrix describes the similarities 237
between species in terms of their responses to environmental gradients. We also applied 238
a hierarchical cluster analyses using the Unweighted arithmetic mean (UPGMA) 239
method to the species trait distance and then compared it to a hierarchical cluster 240
(UPGMA) based on the species environmental responses matrices.
241
Finally, to compare how taxon- and trait-environment relationships were 242
constrained by the spatial structure of the sampling sites, we used a variation 243
partitioning approach (Peres-Neto et al. 2006). First, we applied distance-based Moran’s 244
Eigenvector Maps (db-MEM, Dray, Legendre and Peres-Neto 2006) to the geographic 245
coordinates of sampling sites, which transforms spatial distances into a rectangular 246
matrix, where orthogonal vectors describe the various forms of spatial variation among 247
sites. The first vectors represent the largest scale of spatial variation, whereas the last
248
represent fine scale spatial variation. Then, we partitioned the variation in community 249
response matrices among two explanatory matrices – environmental (all environmental 250
variables) and spatial (db-MEM vectors) (Peres-Neto et al. 2006). We used the strength 251
of partial relationships (Adj. R²; Peres-Neto et al. 2006) to measure the relative 252
importance of the environmental and spatial variables in explaining the variation in the 253
response matrices. Significance levels were assessed with an ANOVA like permutation 254
(999 permutations, as explained above).We used this technique to partition variation of 255
four response matrices in the redundancy approaches described above (i.e., RDA- 256
Composition, RDA-CWM, db-RDA-Composition and db-RDA-Traits).
257
258
3. Results
259
We identified 18,803 nymphs distributed into 23 morphotypes/species. The PCA 260
depicted complex environmental gradients in the studied region, with the first two axes 261
explaining 33% of the environmental variation across sites (Fig. 2). Along the first PCA 262
axis (PCA1), streams in catchments dominated by sugarcane cultivation were, in 263
general, deeper, wider, with a predominance of cobbles, and higher streamflow. In turn, 264
streams in catchments dominated by pasturelands presented substrates dominated by 265
sand and gravel, with lower depth and flow velocity. Along the second axis (PC2), the 266
streams were mainly differentiated according to shading; streams positively related to 267
shading and high dissolved oxygen concentration in forested catchments were 268
negatively related to pH and salinity. The trait-PCA depicted a phylogenetic constraint 269
pattern in species trait as species of the same family were tended to be grouped together.
270
Along the first axis, we found a separation between Baetidae (with negative scores) and 271
Leptophlebiidae (with positive scores), while Caenidae and Leptohyphidae were 272
separated from the other families along the second axis (Fig. 2).
273
274
275
Figure 2. Principal component analysis biplots of the environmental (left panel) 276
gradient and trait (right panel) datasets. Left: Ordination of environmental variables 277
using Principal Components Analysis (PCA) in the Corumbataí River Basin. Right:
278
PCA depicting the trait syndromes in mayfly species from four different families. Black 279
and colored dots represent streams and taxonomic units (species or morphospecies), 280
respectively.
281 282 283
RDA-Composition and RDA-CWM detected weak relationships between 284
species abundance and environmental variables and between abundance-weighted trait 285
mean (CWM) and environmental variables, respectively (Table 1). However, the 286
strength of the relationship detected by the RDA-Composition (Adj. R² = 0.14, 287
P=0.001) was higher than that detected by the RDA-CWM (Adj. R²=0.08, P=0.015).
288
Forward selection indicated that the best set of variables to explain variation in species 289
composition included salinity (F=8.93, P=0.002), stream width (F=3.02, P=0.008), and 290
percentage of sand (F=2.22, P=0.046). Forward selection applied to RDA-CWM 291
indicated that the best set of variables to explain variation in CWM included stream 292
width (F=3.97, P=0.012), salinity (F=3.49, P=0.012) and water temperature (F=2.61, 293
P=0.032). The db-RDA-Composition also showed slightly higher explanatory power
294
(all axes used, Adj. R²=0.17, P=0.001) than simple RDA-Composition, but with the 295
same variables retained by forward selection. Interestingly, db-RDA-Traits contrasted 296
the general trend and had the highest adjusted coefficient of determination (all axes 297
used, adj. R²=0.21, P=0.003) of all methods, and with the same variables retained in db- 298
RDA-Composition (salinity, stream width and percentage of sand) (Table 1).
299
The RLQ analysis, which was computed considering only salinity, stream width, 300
water temperature and percentage of sand as environmental variables (i.e., retained by 301
forward selection in RDAs), did not detect a general relationship between the three data 302
(i.e., species abundances, species traits and environmental variables; model 6 combining 303
results of model 2 and 4; see Dray et al. 2014). However, model 2 indicated a 304
relationship between species abundances and environmental variables, while model 4, 305
which tests the null hypothesis of a random distribution of traits along environmental 306
axes, did not. Fourth-Corner correlations also did not detect relationships between 307
specific traits and specific environmental variables (Table 1); thus, the specific 308
predictions regarding gill traits and feeding habits were not supported.
309
310
311
Table 1. Results of trait-environment and taxon-environment relationships in mayfly 312
communities from Corumbataí River basin. NS = non-significant. * Adjusted P-value.
313
**S.E.S. = Standardized effect size.
314
Method Association statistics Value P-value
RDA-Composition Adj. R² 0.141 0.001
RDA-CWM Adj. R² 0.088 0.015
db-RDA-Composition Adj. R² 0.173 0.001
db-RDA-Trait Adj. R² 0.216 0.003
RLQ Model 2 S.E.S.** 6.219 0.001
Model 4 S.E.S. 1.476 0.082
Model 6 S.E.S. - 0.084
Fourth-Corner method Pearson correlation all <0.10 all NS*
315
We found a lack of relationship between dissimilarity in species traits and 316
dissimilarity in species’ environmental responses (Mantel’s r = -0.02, P = 0.522). The 317
comparison of the two hierarchical clusters also demonstrated a lack of association 318
between similarity in species traits and similarity in environmental responses (Fig. 3).
319
The high number of crossing lines when comparing the two clusters depicts how species 320
with very distinct trait composition can respond in similar ways to the environment. For 321
example, – e.g. the trait composition of Baetodes sp.1 differed from that of Thraulodes 322
sp. and yet both taxa responded similarly to the environmental gradients (Fig. 3). The 323
opposite pattern can also be seen, as Baetodes sp.1 and Zelusia sp. were similar in trait 324
composition and yet responded differently to the environmental gradients (Fig. 3).
325
326
327
Fig. 3. UPGMA clustering based on species traits (left panel) and on their responses to 328
environmental gradients (left panel). The cophenetic correlation coefficients were 0.94 329
and 0.95, respectively. The species A. alphus and A. longetron are from the genus 330
Americabaetis (Baetidae).
331
332
Variation partitioning indicated that taxonomic composition was more related 333
to spatial variables than traits, supporting our expectations. Environmental and spatial 334
variables explained 16% (F=1.58, P=0.004) and 11% (F=1.26, P=0.059) of the 335
variation in species composition and 9% (F=1.28, P=0.120) and 1% (F=1.02, P=0.452) 336
of the variation in CWM, respectively. Using db-RDA the results were similar. For db- 337
RDA-Composition, 15% (F=1.57, P=0.010) of variation was explained by the 338
environment, 10% explained by db-MEM axes (F=1.24, P=0.097) and 4% of shared 339
explanation. As for the trait-based approach (db-RDA-Traits), the environmental and 340
the spatial variables explained 22% (F=10.65, P=0.028) and 5% of the variation 341
(F=9.87, P=0.005), respectively (Figure S1).
342
343
4. Discussion
344
By applying diverse multivariate methods, we found that mayfly trait-based 345
approaches do not provide a better and clearer understanding of community assembly 346
along environmental gradients in comparison to a taxonomic-based approach based on 347
species identities and abundances. While trait-based approaches have provided valuable 348
insights about biodiversity variation that would not be possible using species identities 349
alone (McGill et al. 2006; Pavoine and Bonsall 2011), they often have limited 350
explanatory power on how communities respond to environmental variation (Davies et 351
al. 2000; Horrigan and Baird 2008).
352
Trait-based approaches rely on multiple methodological choices that are still 353
under debate in comparison to compositional approaches. For instance, the decision 354
about which traits to measure and how is discussed in the literature (Ackerly and 355
Cornwell 2007; Saito et al. 2016), as well as the organization level needed to measure 356
them (e.g., species vs. individual level, Cianciaruso et al. 2009). In this regard, weak 357
trait-environment relationships in empirical studies may emerge due to methodological 358
aspects. First, a weak relationship between traits and environmental conditions may 359
arise simply due to the use of imprecise data gathered from the literature – as opposed to 360
the use of precise measurements of traits from locally sampled organisms (Cano- 361
Barbacil, Radinger and García-Berthou 2020). Second, we usually cannot be sure 362
whether all traits are relevant to determine community assembly and, so, using 363
irrelevant traits could obscure the detection of environmental filtering (Colwell and 364
Winkler 1984; Saito et al. 2016). Third, in general, we do not know the best resolution 365
and scale to measure traits in order to detect trait-environment relationships (Violle et 366
al. 2012). Fourth, trait measurements have trade-offs of information quality and 367
time/resources costs (Saito et al. 2016). For example, while we have considered wing 368
length, what explains species flight ability is the whole design of the wing, a very
369
challenging and time-consuming trait to measure. Thus, because trait-based approaches 370
are dependent on a higher number of methodological steps and decisions, when 371
compared to taxonomic-based approaches, they are also more prone to methodological 372
issues that will ultimately affect our ability to detect trait-environment relationships.
373
We expected that some traits would respond specifically to some environmental 374
variables (e.g., gill traits to dissolved oxygen). However, we did not find clear single- 375
trait vs. single-environmental characteristic relationship to support the mechanistic 376
assumption of trait-based approaches. In this sense, it is likely that trait composition, 377
rather than individual traits, affects individual fitness (Resh et al. 1994; Verberk, Van 378
Noordwijk and Hildrew 2013; Laughlin 2014; Pilière et al. 2016). In general, from an 379
evolutionary perspective, traits do not evolve independently in response to 380
environmental pressures but as specific combinations of traits that enable spin-offs 381
(combination of traits maximizing fitness) and trade-offs (investments in one trait 382
reducing investments in another traits) (Verberk, Van Noordwijk and Hildrew 2013).
383
The strength of such evolutionary constraint was exemplified by the phylogenetic signal 384
pattern in the PCA of the trait data, where genera from the same family shared similar 385
suite of traits. While these ideas address the importance of trait combinations, one 386
should also argue that there are different suites of traits able to maximize species 387
abundances in a given environment (Resh et al. 1994). For example, water-penny 388
beetles with flattened bodies, caddisflies with strong anal claws, and blackfly larvae 389
using glues to attach to the substrate all use different strategies to deal with the same 390
environmental filter, high streamflow. Our results support these ideas, since we found 391
that genera with distinct trait composition (suite of traits) can respond similarly along 392
the same environmental gradient (e.g., Farrodes and Leptohyphes).
393
Both trait-based and taxon-based approaches revealed a clear relationship 394
between mayfly communities and salinity. Our results suggest that even a small 395
variation in salinity (from 0.01 to 0.05 mS cm-¹) may indicate impacts able to modify 396
mayfly communities, which in our case is related to land use practices in the study area.
397
Salinity was present in streams surrounded by sugarcane plantations but not in forested 398
streams, suggesting that agricultural practices are driving freshwater salinization. This 399
salinization process is currently a major concern in aquatic ecology (Kaushal et al.
400
2005; Cañedo-Argüelles et al. 2016; Kefford 2018) and our findings suggest that small 401
increases in salinity can modify the taxonomic and trait composition of mayfly 402
communities. This environmental filter has been hypothesized to be important because 403
of the probable interference in pH regulation and osmoregulation costs that are mediated 404
by gill traits (Kefford 2018). However, the specific effects over gill traits at the 405
community level were not evidenced given the lack of a relationship between salinity 406
and gill traits in fourth-corner correlations.
407
Compared to species, trait variation should be less related to spatial variables 408
and this is a key assumption to the generalized use of trait-based approaches. We found 409
some support to this assumption as variation in taxonomic composition was more 410
related to db-MEM vectors. Pure spatial signals in partial constrained ordination have 411
been commonly associated with both the effect of unmeasured spatially structured 412
environmental variables and dispersal on species composition (Diniz-Filho et al. 2012, 413
Siqueira et al. 2012b, Saito et al. 2015b). Because we measured a high number of 414
environmental variables that has been previously shown to drive the distribution of 415
mayflies, we suggest the first explanation is less likely (effect of unmeasured spatially 416
structured environmental variables). In contrast, mayflies tend to disperse over short 417
distances because adults usually live only hours or few days and are considered weak
418
fliers in comparison to other taxa (e.g., Odonata) with low ability to direct their 419
trajectories, making them prone to have limited geographic distribution (Malmqvist 420
2000). For example, Caudill (2003) showed that most females emerging from one lake 421
lay their eggs in the same lake, despite the occurrence of several other lakes in the 422
vicinity (~250 m). The stronger spatial structure detected in composition data, as 423
compared to trait data, adds support to the use of trait-based approaches for large scale 424
assessments.
425
In general, biomonitoring tools in the Neotropics are still under development and 426
our results raise a question on whether we should focus on trait-based approaches.
427
Previous studies found relationships between functional diversity, trait composition, and 428
environmental change (Saito et al. 2015a; Castro, Dolédec and Callisto 2017). However, 429
our findings indicate that both trait and taxonomic composition responded to the same 430
environmental variables, indicating the importance of the same environmental gradients 431
(e.g., salinity) in structuring the local communities. We highlight that traits are 432
inherently interrelated and therefore may not respond monotonically and invariably to 433
environmental variation, but instead present complex responses depending on the set of 434
traits a species has (Resh et al. 1994). Also, trait-based approaches indicated a weaker 435
spatial signal on trait composition variation, potentially being more appropriate for large 436
scale assessments, a desirable attribute for biomonitoring tools. For biomonitoring 437
purposes, choosing between approaches remains a challenge as trait- and taxonomic- 438
based approaches have advantages and disadvantages and could, in practice, give 439
similar information regarding community response to environmental change.
440
441
5. Acknowledgments
442
We thank many colleagues who stimulated interesting discussions during the 443
elaboration of this study at the Brazilian Symposium of Limnology in 2017, as well as 444
colleagues from the Conference of the Iberian Association of Limnology in 2018. We 445
also thank two anonymous reviewers that helped us during the preparation of this 446
manuscript. The study was partly funded by grants #13/50424-1, #18/02074-5 and 447
#19/04033-7, São Paulo Research Foundation (FAPESP). Work by LMB has been 448
supported by CNPq (304314/2014-5) and the National Institute for Science and 449
Technology (INCT) in Ecology, Evolution and Biodiversity Conservation (proc.
450
465610/2014-5 MCTIC/CNPq and FAPEG).
451
452
6. References 453
Ackerly, D. D., & Cornwell, W. K. (2007). A trait‐based approach to community 454
assembly: partitioning of species trait values into within‐and among‐community 455
components. EcologyLetters, 10(2), 135-145.
456
Aspin, T. W., Khamis, K., Matthews, T. J., Milner, A. M., O'Callaghan, M. J., Trimmer, 457
M., ... & Ledger, M. E. (2019). Extreme drought pushes stream invertebrate 458
communities over functional thresholds. Global Change Biology, 25(1), 230-244.
459
Bauman, D., Drouet, T., Dray, S., & Vleminckx, J. (2018). Disentangling good from 460
bad practices in the selection of spatial or phylogenetic 461
eigenvectors. Ecography, 41(10), 1638-1649.
462
Blanchet, F. G., Legendre, P., & Borcard, D. (2008). Forward selection of explanatory 463
variables. Ecology, 89(9), 2623-2632.
464
Cañedo-Argüelles, M., Hawkins, C. P., Kefford, B. J., Schäfer, R. B., Dyack, B. J., 465
Brucet, S., ... & Coring, E. (2016). Saving freshwater from salts. Science, 351(6276), 466
914-916.
467
Cano‐Barbacil, C., Radinger, J., & García‐Berthou, E. Reliability analysis of fish traits 468
reveals discrepancies among databases. Freshwater Biology. In press.
469
Castro, D. M., Dolédec, S., & Callisto, M. (2017). Landscape variables influence 470
taxonomic and trait composition of insect assemblages in Neotropical savanna 471
streams. Freshwater Biology, 62(8), 1472-1486.
472
Caudill, C. C. (2003). Measuring dispersal in a metapopulation using stable isotope 473
enrichment: high rates of sex‐biased dispersal between patches in a mayfly 474
metapopulation. Oikos, 101(3), 624-630.
475
Cianciaruso, M. V., Batalha, M. A., Gaston, K. J., & Petchey, O. L. (2009). Including 476
intraspecific variability in functional diversity. Ecology, 90(1), 81-89.
477
Colwell, R. K., & Winkler, D. W. (1984). A null model for null models in 478
biogeography. Ecological communities: conceptual issues and the evidence, 344-359.
479
Colzani, E., Siqueira, T., Suriano, M. T., & Roque, F. O. (2013). Responses of aquatic 480
insect functional diversity to landscape changes in Atlantic Forest. Biotropica, 45(3), 481
343-350.
482
Davies, K. F., Margules, C. R., & Lawrence, J. F. (2000). Which traits of species predict 483
population declines in experimental forest fragments?. Ecology, 81(5), 1450-1461.
484
Diniz‐Filho, J. A. F., Siqueira, T., Padial, A. A., Rangel, T. F., Landeiro, V. L., & Bini, 485
L. M. (2012). Spatial autocorrelation analysis allows disentangling the balance between 486
neutral and niche processes in metacommunities. Oikos, 121(2), 201-210.
487
Dolédec, S., Chessel, D., Ter Braak, C. J. F., & Champely, S. (1996). Matching species 488
traits to environmental variables: a new three-table ordination method. Environmental 489
and Ecological Statistics, 3(2), 143-166.
490
Dolédec, S., Statzner, B., & Bournard, M. (1999). Species traits for future 491
biomonitoring across ecoregions: patterns along a human‐impacted river. Freshwater 492
Biology, 42(4), 737-758.
493
Dray, S., & Legendre, P. (2008). Testing the species traits–environment relationships:
494
the fourth‐corner problem revisited. Ecology, 89(12), 3400-3412.
495
Dray, S., Legendre, P., & Peres-Neto, P. R. (2006). Spatial modelling: a comprehensive 496
framework for principal coordinate analysis of neighbour matrices (PCNM). Ecological 497
Modelling, 196(3-4), 483-493.
498
Dray, S., Choler, P., Doledec, S., Peres-Neto, P. R., Thuiller, W., Pavoine, S., & ter 499
Braak, C. J. (2014). Combining the fourth‐corner and the RLQ methods for assessing 500
trait responses to environmental variation. Ecology, 95(1), 14-21.
501
Golterman, H. L., Clymo, R. S., & Ohnstad, M. A. M.(1978). Methods for Physical and 502
Chemical analysis of freshwater. IBP. Hand book NO. 8. Black well Scientific 503
Publications. Osney Nead.
504
Hamilton, A. T., Schäfer, R. B., Pyne, M. I., Chessman, B., Kakouie, K., Boersma, K.
505
S., ... & Bierwagen, B. (2019). Limitations of Trait‐Based Approaches for Stressor 506
Assessment: the Case of Freshwater Invertebrates and Climate Drivers. Global Change 507
Biology.
508
Horrigan, N., & Baird, D. J. (2008). Trait patterns of aquatic insects across gradients of 509
flow-related factors: a multivariate analysis of Canadian national data. Canadian 510
Journal of Fisheries and Aquatic Sciences, 65(4), 670-680.
511
Kaushal, S. S., Groffman, P. M., Likens, G. E., Belt, K. T., Stack, W. P., Kelly, V. R., ...
512
& Fisher, G. T. (2005). Increased salinization of fresh water in the northeastern United 513
States. Proceedings of the National Academy of Sciences, 102(38), 13517-13520.
514
Kefford, B. J. (2018). Why are mayflies (Ephemeroptera) lost following small increases 515
in salinity? Three conceptual osmophysiological hypotheses. Philosophical 516
Transactions of the Royal Society B, 374(1764), 20180021.
517
Kleyer, M., Dray, S., Bello, F., Lepš, J., Pakeman, R. J., Strauss, B., ... & Lavorel, S.
518
(2012). Assessing species and community functional responses to environmental 519
gradients: which multivariate methods?. Journal of Vegetation Science, 23(5), 805-821.
520
Laughlin, D. C. (2014). The intrinsic dimensionality of plant traits and its relevance to 521
community assembly. Journal of Ecology, 102(1), 186-193.
522
Lavorel, S., Grigulis, K., McIntyre, S., Williams, N. S., Garden, D., Dorrough, J., ... &
523
Bonis, A. (2008). Assessing functional diversity in the field–methodology 524
matters!. Functional Ecology, 22(1), 134-147.
525
Legendre, P., & Anderson, M. J. (1999). Distance‐based redundancy analysis: testing 526
multispecies responses in multifactorial ecological experiments. Ecological 527
Monographs, 69(1), 1-24.
528
Legendre, P., & Legendre, L. F. (2008). Numerical ecology (Vol. 24). Elsevier.
529
Legendre, P., Oksanen, J., & ter Braak, C. J. (2011). Testing the significance of 530
canonical axes in redundancy analysis. Methods in Ecology and Evolution, 2(3), 269- 531
277.
532
Leitão, R. P., Zuanon, J., Villéger, S., Williams, S. E., Baraloto, C., Fortunel, C., ... &
533
Mouillot, D. (2016). Rare species contribute disproportionately to the functional 534
structure of species assemblages. Proceedings of the Royal Society B: Biological 535
Sciences, 283(1828), 20160084.
536
Mackereth, F. J. H., Heron, J., & Talling, J. F. (1978). Water Analysis of 537
Freshwater. Biological Assessment Publications, 36, 120.
538
Malmqvist, B. (2000). How does wing length relate to distribution patterns of stoneflies 539
(Plecoptera) and mayflies (Ephemeroptera)?. Biological conservation, 93(2), 271-276.
540
McGill, B. J., Enquist, B. J., Weiher, E., & Westoby, M. (2006). Rebuilding community 541
ecology from functional traits. Trends in Ecology & Evolution, 21(4), 178-185.
542
Pavoine, S., & Bonsall, M. B. (2011). Measuring biodiversity to explain community 543
assembly: a unified approach. Biological Reviews, 86(4), 792-812.
544
Pavoine, S., & Ricotta, C. (2014). Functional and phylogenetic similarity among 545
communities. Methods in Ecology and Evolution, 5(7), 666-675.
546
Pavoine, S., Vallet, J., Dufour, A. B., Gachet, S., & Daniel, H. (2009). On the challenge 547
of treating various types of variables: application for improving the measurement of 548
functional diversity. Oikos, 118(3), 391-402.
549
Peres - Neto, P. R., Dray, S., & ter Braak, C. J. (2017). Linking trait variation to the 550
environment: critical issues with community‐weighted mean correlation resolved by the 551
fourth‐corner approach. Ecography, 40(7), 806-816.
552
Peres-Neto, P. R., Legendre, P., Dray, S., & Borcard, D. (2006). Variation partitioning 553
of species data matrices: estimation and comparison of fractions. Ecology, 87(10), 554
2614-2625.
555
Pilière, A. F. H., Verberk, W. C. E. P., Gräwe, M., Breure, A. M., Dyer, S. D., 556
Posthuma, L., ... & Schipper, A. M. (2016). On the importance of trait interrelationships 557
for understanding environmental responses of stream macroinvertebrates. Freshwater 558
Biology, 61(2), 181-194.
559
Poff, N. L. (1997). Landscape filters and species traits: towards mechanistic 560
understanding and prediction in stream ecology. Journal of the North American 561
Benthological Society, 16(2), 391-409.
562
Quantum GIS Development Team (2019) Quantum GIS Geographic Information 563
System. Open Source Geospatial Foundation Project. http://qgis.osgeo.org 564
Resh, V. H., Hildrew, A. G., Statzner, B., & Townsend, C. R. (1994). Theoretical 565
habitat templets, species traits, and species richness: a synthesis of long‐term ecological 566
research on the Upper Rhône River in the context of concurrently developed ecological 567
theory. Freshwater Biology, 31(3), 539-554.
568
Saito, V. S., Siqueira, T., & Fonseca-Gessner, A. A. (2015a). Should phylogenetic and 569
functional diversity metrics compose macroinvertebrate multimetric indices for stream 570
biomonitoring?. Hydrobiologia, 745(1), 167-179.
571
Saito, V. S., Soininen, J., Fonseca‐Gessner, A. A., & Siqueira, T. (2015b). Dispersal 572
traits drive the phylogenetic distance decay of similarity in Neotropical stream 573
metacommunities. Journal of Biogeography, 42(11), 2101-2111.
574
Saito, V. S., Cianciaruso, M. V., Siqueira, T., Fonseca‐Gessner, A. A., & Pavoine, S.
575
(2016). Phylogenies and traits provide distinct insights about the historical and
576
contemporary assembly of aquatic insect communities. Ecology and Evolution, 6(9), 577
2925-2937.
578
Shipley, B. (2010). Community assembly, natural selection and maximum entropy 579
models. Oikos, 119(4), 604-609.
580
Shipley, B., De Bello, F., Cornelissen, J. H. C., Laliberté, E., Laughlin, D. C., & Reich, 581
P. B. (2016). Reinforcing loose foundation stones in trait-based plant 582
ecology. Oecologia, 180(4), 923-931.
583
Siqueira, T., Bini, L. M., Roque, F. O., & Cottenie, K. (2012a). A metacommunity 584
framework for enhancing the effectiveness of biological monitoring strategies. PLoS 585
One, 7(8), e43626.
586
Siqueira, T., Bini, L. M., Roque, F. O., Marques Couceiro, S. R., Trivinho‐Strixino, S., 587
& Cottenie, K. (2012b). Common and rare species respond to similar niche processes in 588
macroinvertebrate metacommunities. Ecography, 35(2), 183-192.
589
Siqueira, T., Lacerda, C. G. L. T., & Saito, V. S. (2015). How does landscape 590
modification induce biological homogenization in tropical stream 591
metacommunities?. Biotropica, 47(4), 509-516.
592
Southwood, T. R. (1977). Habitat, the templet for ecological strategies? Journal of 593
Animal Ecology, 46(2), 337-365.
594
Statzner, B., Dolédec, S., & Hugueny, B. (2004). Biological trait composition of 595
European stream invertebrate communities: assessing the effects of various trait filter 596
types. Ecography, 27(4), 470-488.
597
Thioulouse, J., Dray, S., Dufour, A. B., Siberchicot, A., Jombart, T., & Pavoine, S.
598
(2018). Relating Species Traits to Environment. In Multivariate Analysis of Ecological 599
Data with ade4 (pp. 223-237). Springer, New York, NY.
600
Tilman, D. (2004). Niche tradeoffs, neutrality, and community structure: a stochastic 601
theory of resource competition, invasion, and community assembly. Proceedings of the 602
National Academy of Sciences, 101(30), 10854-10861.
603
Tomanova, S., Goitia, E., & Helešic, J. (2006). Trophic levels and functional feeding 604
groups of macroinvertebrates in neotropical streams. Hydrobiologia, 556(1), 251-264.
605
Townsend, C. R., & Hildrew, A. G. (1994). Species traits in relation to a habitat templet 606
for river systems. Freshwater Biology, 31(3), 265-275.
607
Valente, R.D.O.A. (2001). Análise da estrutura da paisagem na bacia do rio 608
Corumbataí, São Paulo, Brazil, Master thesis – São Paulo University.
609
Vannote, R. L., Minshall, G. W., Cummins, K. W., Sedell, J. R., & Cushing, C. E.
610
(1980). The river continuum concept. Canadian Journal of Fisheries and Aquatic 611
Sciences, 37(1), 130-137.
612
Verberk, W. C., Van Noordwijk, C. G. E., & Hildrew, A. G. (2013). Delivering on a 613
promise: integrating species traits to transform descriptive community ecology into a 614
predictive science. Freshwater Science, 32(2), 531-547.
615
Violle, C., Enquist, B. J., McGill, B. J., Jiang, L. I. N., Albert, C. H., Hulshof, C., ... &
616
Messier, J. (2012). The return of the variance: intraspecific variability in community 617
ecology. Trends in Ecology & Evolution, 27(4), 244-252.
618
Wingfield, C. A. (1939). The function of the gills of mayfly nymphs from different 619
habitats. Journal of Experimental Biology, 16(3), 363-373.
620
Table S1. Mayfly traits from Neotropical streams. Collector, shredder, scrapper and 621
filterer encompass the feeding habit of mayflies using a fuzzy code where we provide an 622
affinity (between 0 and 3) of each species for each trait level. The number of gills is a 623
discrete trait. The levels of the second gill trait (flat gill, sharp gill, gill fringes and 624
opercular gill) are coded as binary variables. Body format is fuzzy coded within the 625
levels streamlined, flat body and cylindrical body. The locomotion trait is separated in 626
swimmer and scrawler and also fuzzy coded. Wing size, posterior wing size and the size 627
of the immature are continuous traits measured in millimeters.
628
Taxon
Collector Shredder Scrapper Filterer number of gills flat gill sharp gill gill fringes opercular gill Streamlined flat body cillindrical body swimmer scrawler wing size (mm) posterior wing size (mm) immature size (mm)
Americabaetis
longetron 3 1 2 0 12 1 0 0 0 1 0 3 3 2 5.8 0 7.05
Americabaetis
alphus 3 1 2 0 12 1 0 0 0 1 0 3 3 2 4.9 0 5.60
Baetodes sp.1 3 0 3 0 10 1 0 0 0 1 0 3 3 2 4.0 0 4.13
Baetodes sp.2 3 0 3 0 10 1 0 0 0 1 0 3 3 2 3.0 0 3.34
Caenis chamie 3 1 1 0 8 1 0 1 1 1 1 2 1 3 3.8 0 5.50
Caenis
pflugfelderi 3 1 1 0 8 1 0 1 1 1 1 2 1 3 2.5 0 5.30
Callibaetis
sellacki 3 1 2 1 28 1 0 0 0 1 2 3 2 3 7.0 1.2 7.80
Cloeodes
maracatu 2 0 1 0 14 1 0 0 0 1 0 3 2 3 4.0 0 5.10
Farrodes sp.1 2 1 1 1 28 0 1 0 0 1 2 0 1 2 5.7 0.8 6.25
Paracloeodes
sp.1 2 0 3 1 14 1 0 0 0 1 0 3 3 2 4.2 0.9 4.08
Traverhyphes
edmundisi 3 1 2 0 28 1 0 0 1 0 1 2 0 3 3.9 0.78 4.99
Tricorythopsis
sp.1 3 1 3 0 32 1 0 0 1 1 3 1 0 3 2.0 0 2.16
Zelusia sp. 3 0 2 1 12 1 0 0 0 1 0 3 3 2 4.0 0 3.95
Thraulodes
sp. 0 3 0 1 28 1 0 0 0 0 2 1 1 3 9.0 0 8.50
Hylister sp. 1 0 2 3 24 1 0 1 0 1 2 0 1 3 10.4 2.2 10.96
Leptohyphes
sp. 3 1 3 0 42 1 0 0 1 0 1 2 0 3 4.5 0 6.10
Hagenulopsis
sp. 2 1 1 1 28 0 1 0 0 1 2 2 1 3 5.6 0 6.30
Waltzoyphius
sp. 2 0 3 1 14 1 0 0 0 1 0 3 3 2 4.0 0 3.40
Miroculis sp. 2 1 1 1 28 0 1 0 0 1 2 0 1 2 5.6 0 5.00
Ulmeritoides
sp. 0 3 1 1 28 1 0 1 0 1 2 0 0 3 8.0 0 10.00
Apobaetis sp. 3 1 2 1 14 1 0 0 0 1 1 3 3 1 6.0 0 6.87
Tricorythodes
sp. 3 0 1 1 28 1 0 0 1 1 3 1 0 3 4.4 0 5.17