trophic level variance in invaded riverine fish assemblages ... highest trophic level (TL) of food chains (i.e. biomagnification; LeBlanc, 1995). ...... Second Edition.
Environmental correlates of food-chain length, mean trophic level and trophic level variance in invaded riverine fish assemblages
Alberto Maceda-Veiga1,2, Ralph Mac Nally3, Adolfo de Sostoa1
1
Department of Evolutionary Biology, Ecology and Environmental Sciences & Institute of Research in Biodiversity, Universitat de Barcelona (IRBio-UB), 08028 Barcelona, Spain
2
Department of Integrative Ecology, Estación Biológica de Doñana (EBD-CSIC), 41092 Sevilla, Spain
3
Institute for Applied Ecology, University of Canberra, Bruce 2617, ACT, Australia
1
ABSTRACT
2
Examining how the trophic structure of biotic assemblages is affected by human
3
impacts, such as habitat degradation and the introduction of alien species, is important
4
for understanding the consequences of such impacts on ecosystem functioning. We used
5
general linear mixed models and hierarchical partitioning analyses of variance to
6
examine for the first time the applicability of three hypotheses (ecosystem-size,
7
productivity and disturbance) for explaining food-chain length (FCL) in invaded fish
8
assemblages. We used Fishbase trophic level (TL) estimates for 16 native and 18 alien
9
fish species in an extensive riverine system in north-eastern Spain (99,700 km2, 15
10
catchments, 530 sites). The FCL of assemblages ranged from 2.7 to 4.42. Ecosystem
11
size-related variables (Strahler stream order, physical habitat diversity) and human-
12
disturbance (conductivity) made the largest contribution to the explained variance in the
13
FCL model after accounting for spatial confounding factors and collinearity among
14
predictors. Within-assemblage TL also was positively associated with Strahler stream
15
order, suggesting that large rivers have the highest trophic diversity. High conductivity
16
was negatively associated with FCL, as did with the mean TL of fish assemblages.
17
However, an inverse association was found between mean TL and Strahler stream order,
18
possibly because the presence of fish species of high TL may be offset by larger
19
numbers of alien species of lower TL in large rivers. Given that there may be trophic
20
replacements among native and alien species, this inference needs to be addressed with
21
detailed trophic studies. However, reducing water conductivity by improved wastewater
22
treatment and better agricultural practices probably would help to conserve the fish
23
species on the apices of aquatic food-webs.
24
Keywords: alien species; disturbance; ecosystem size; fish assemblages; food-chain
25
length; trophic levels
26 27
1
28 29
1. Introduction The structure of biotic assemblages is determined by many factors (Townsend et
30
al., 2003) of which trophic interactions are among the most important (Pimm, 1982;
31
Pauly et al., 2002; Layman et al., 2015). Trophic interactions dictate the number of
32
trophic transfers of energy from basal resources to consumers (i.e. food-chain length,
33
FCL; Thompson et al., 2012). FCL is a central characteristic of food-webs, which
34
affects important ecosystem properties including nutrient cycling, species diversity and
35
atmospheric carbon exchange (Pace et al., 1999; Persson et al., 1999). Moreover, FCL is
36
associated with the risk of pollutant concentrations increasing from the lowest to the
37
highest trophic level (TL) of food chains (i.e. biomagnification; LeBlanc, 1995).
38
Identifying the factors that determine FCL long has attracted the attention of
39
ecologists (Lindeman, 1942; Takimoto & Post, 2013; Ward et al., 2017), who have
40
erected three main hypotheses to explain FCL. The ecosystem-size hypothesis predicts
41
that FCL increases with spatial scale because larger ecosystems have greater habitat
42
availability and diversity, which supports larger populations of more diverse arrays of
43
prey and predators (Post et al., 2000). The productivity hypothesis states that nutrient-
44
poor ecosystems should have shorter FCL than nutrient-enriched ecosystems because of
45
the energy loss in each trophic transfer, which limits the population sizes of species at
46
higher trophic levels (TL) (Schoener, 1989). The disturbance hypothesis predicts that
47
FCL will be reduced by frequent, intense changes in environmental conditions because
48
the longer the food chain, the more vulnerable the ecosystem is to perturbations (Pimm
49
& Lawton, 1977). Moreover, the most energy-demanding consumers (i.e. those at the
50
highest TLs) are not abundant in ecosystems and, so, their probability of being lost
51
during disturbance events is greatest (Pimm & Kitching, 1987). Although none of these
52
hypotheses has received universal support, ecosystem size has been identified as a
53
crucial determinant of FCL, even though the three hypotheses rarely have been explored
54
concurrently (reviewed by Takimoto & Post, 2013). Moreover, changes in food-web
55
structure are complicated by processes such as species additions or replacements from
56
the introduction of alien species (Eby et al., 2006; Byrnes et al., 2007).
57
Rivers are important testing grounds for exploring FCL hypotheses because their
58
food-webs are as complex and species-rich in any other ecosystem type (Winemiller,
59
1990). Rivers have long gradients of variation in size, productivity and disturbance
60
(McHugh et al., 2010; Sabo et al., 2010; Mor et al., 2018). Moreover, alien species often
61
are a major proportion of regional species pools in rivers (Chandra & Gerhardt, 2008; 2
62
Tockner et al., 2009), which make rivers pertinent study systems to explore the
63
contemporary environmental determinants of FCL. Studies examining the three
64
hypotheses in uninvaded rivers found little support for any hypotheses (Warfe et al.,
65
2013) or a neutral effect of productivity, a positive effect of ecosystem size and a
66
negative effect of disturbance on FCL (Sabo et al., 2010; McHugh et al., 2010).
67
However, these few studies can provide little indication of the generality of the effects
68
of the rival hypothesis.
69
We used data on riverine fish in an extensive area of north-eastern Spain (99,700
70
km2, 15 catchments) to explore variation in FCL along environmental gradients and to
71
identify the changes in fish-food chains responsible for such variation. Many studies
72
exploring FCL variation have used a single measure (e.g. maximum or mean TL;
73
Takimoto & Post, 2013), which provides limited ability to distinguish among the rival
74
ideas. A single measure can be the result of very different food-web architectures, but
75
much more discernment might be achieved if one also were to explore within-
76
assemblage TL variation (Fig. 1). Fish assemblages in north-eastern Spain have diverse
77
trophic strategies, ranging from low-order (e.g. Cyprinus carpio, Alburnus alburnus) to
78
high-order consumers (e.g. Micropterus salmoides, Anguilla anguilla), which exemplify
79
the trophic guilds of native and alien species in many if not most rivers around the
80
world (e.g. Kottelat & Freyhof, 2007). Variation in fish-species composition in north-
81
eastern Spain mostly has been attributed to river size and to other geographical factors
82
(e.g. elevation and basin name) alongside water quality and habitat degradation
83
(Maceda-Veiga et al., 2017). A study of functional diversity in the same study system
84
mirrored these findings but ranked habitat degradation before water quality variables,
85
which included nutrient concentrations (Colin et al., 2018). Therefore, studies of the
86
fish assemblages we consider have the potential to explore how the trophic structure of
87
food webs depends on environmental factors, including disturbance, which has been the
88
FCL hypothesis with least support to date (Takimoto & Post, 2013).
89
By using multimodel inference, we identified the best environmental descriptors
90
of TL variation in fish assemblages along gradients of geographical, hydraulic, physical
91
habitat and water quality variables in north-eastern Spain. We examined the extent to
92
which alien and native species richness affects the TL of fish assemblages given that the
93
percentage of alien fish species in Mediterranean-type rivers, including north-eastern
94
Spain, is high (Leprieur et al., 2008). If FCL and trophic diversity increase with
95
ecosystem size and productivity (Post et al., 2000; Schoener, 1989), then we expected 3
96
river size and river productivity to be positively associated with the maximum TL and
97
the range of TL of fish assemblages, but negatively related to within-assemblage TL
98
variation. If human disturbances induce opposite effects, then TL measures should be
99
inversely related to measures of poor environmental conditions. Our results have
100
management implications beyond our study system considering the many rivers with
101
Mediterranean-climate around the world (Bonada & Resh, 2013), and the rarity of fish
102
predators in these ecosystems prior to alien fish introductions (Doadrio et al., 2011).
103 104
2. Methods
105
2.1. Study area
106
Fish and environmental data were from several projects conducted in north-eastern
107
Spain from 2002 to 2009 (e.g. Figuerola et al., 2012; Maceda–Veiga et al., 2017) to
108
assess the ecological status of rivers as required by the Water Framework Directive of
109
Europe (EU Directive 2000/60/EC). The study area was part of the Garonne basin and
110
all catchments drain to the Mediterranean Sea from the Sènia to Muga basins (Fig. 2).
111
The native and alien fish richness in each basin ranges from 1 to 10 and from 1 to 11,
112
respectively (Maceda-Veiga et al., 2010). Sampled sites (N = 530) were selected for
113
accessibility and also to include broad gradients of elevation, river size, in-stream
114
hydraulic, physical habitat and water quality (Table 1). The rivers drain relatively well-
115
preserved riparian forests to grasslands, weedy areas and extensive agriculture areas,
116
including regions vulnerable to nutrient pollution (the EU Nitrates Directive). There are
117
many industries in the surroundings of large cities close to the coast that release metals
118
and other biologically recalcitrant compounds into rivers at harmful concentrations for
119
fish and other aquatic taxa (e.g. Maceda-Veiga et al., 2013b; Colin et al., 2016). The
120
water quality of rivers is degraded by many urban sewage treatment plants that lack
121
tertiary treatment, which results in eutrophication and in the release of the so-called
122
emerging pollutants (see Muñoz et al., 2009; Munné et al., 2012). The hydrological
123
regime of most of these streams is typically ‘Mediterranean’, with droughts common in
124
summer and some torrential flooding in autumn. Large rivers peak in flow in spring
125
from snowmelt. We surveyed in low-flow conditions, because this is when fish can be
126
sampled most effectively using electrofishing (see below).
4
127
2.2. Environmental descriptors of rivers
128
Each sampling site was characterized using environmental variables related to the three
129
main hypotheses for explaining food-chain length (stream size, productivity and
130
disturbance) (Tables 1 and 2). Disturbance was defined broadly as any reduction in
131
water and physical habitat quality because they may alter the taxonomic composition
132
and functional diversity of fish assemblages (e.g. Maceda-Veiga et al., 2017; Colin et
133
al., 2018). Changes in water and physical habitat quality are the result of sudden (e.g.
134
diffuse pollution, riparian removal) or long-term impacts on rivers (e.g. chronic
135
pollution) (Freedman, 1995). The environmental variables were grouped as
136
geographical, hydraulic, physical habitat and water quality factors because these
137
variables in each group tend to be managed concurrently by water agencies (Table 1).
138
The basin name and elevation (m.a.s.l.) were recorded as geographical factors
139
using Google Earth®. Elevation was used to represent natural spatial factors affecting
140
aquatic organisms (e.g. Maceda–Veiga et al., 2013b). River size was determined by
141
calculating the Strahler stream order (Strahler, 1964) on a map (1:50000).
142
Eight water-quality variables were measured in situ prior to the conduct of each
143
fish survey. Temperature (ºC), conductivity (µS/cm) and pH were measured using a
144
digital multiparametric YSI® probe, as was general water hardness. The concentrations
145
of four nutrient types (i.e. ammonium, nitrite, nitrate, and phosphate–P; mg/l) were
146
measured using the colorimetric test kit VISOCOLOR®. The levels of nutrients, pH, and
147
conductivity in rivers naturally vary from upstream to downstream and are altered by
148
sewage effluents and run-off from urban areas, industries and agricultural areas
149
(Camargo & Alonso, 2006; Kefford et al., 2012; Colin et al., 2016). Clean waters in the
150
studied region have conductivity values ≤ 600 µS/cm (Sánchez-Montoya et al., 2012),
151
being particularly low in upstream compared to downstream clean river reaches. High
152
concentrations of nitrate and phosphate–P, but particularly of nitrite and ammonium at
153
high pH values, are lethal to fish (e.g. Camargo & Alonso, 2006; Smallbone et al., 2016;
154
Noga, 2011). Nutrient pollution results in eutrophication, while detritus accumulates
155
from high levels of eutrophication, agricultural runoff and the release of detritus from
156
sewage treatment-plant outflows. We estimated the accumulation of detritus as the % of
157
the streambed covered by brown particles < 1 mm (Townsend et al., 2008).
158
Physical habitat was described using six measures. We used an index of riparian
159
quality (QBR) as an integrated measure of river hydro-morphology and conservation
160
status of riparian areas, which is used widely by Spanish authorities in environmental 5
161
risk assessment (Munné et al., 2003). The QBR index is a composite of: (1) presence of
162
riparian areas; (2) connectivity among such areas; (3) links to adjacent woodlands; (4)
163
percentage of riparian trees, shrubs and emerged aquatic plants; (5) presence of alien
164
riparian species; and (6) presence of weirs, channels, and stream-side walls. QBR
165
ranges between 0 and 100 (see the protocol in Munné et al., 2003). The percentage of
166
river margins with ferns and herbs was estimated because the herbaceous layer is not a
167
component in the QBR but is likely to be important for river functioning (e.g.
168
preventing run-off, provision of shelter for organisms). We estimated the percentage of
169
dead wood, macrophytes (e.g. filamentous algae such as Cladophora spp., submerged
170
aquatic plants such as Potamogeton spp.) and rock hides. We used an integrated
171
measure of microhabitat diversity, which ranked the presence and diversity of structural
172
refuges in the riverbed (e.g. roots, caves) on a scale from 0 to 20.
173
Last, the mean flow velocity (m/s) and water level (m) were measured as
174
indicators of in-stream hydraulics. Three transects were set perpendicular to the water
175
flow at 20-m intervals along the river reach surveyed in each sampling site.
176 177
2.3. Fish surveys
178
We used an international standardized fish sampling method (CEN standards EN 14962
179
and EN 14011). We used a single–pass electrofishing with a portable power unit that
180
generated up to 200V and 3 A pulsed direct current to survey the whole wetted width of
181
at least 100-m long reaches (mean±standard error=140±30 m) at each site in an
182
upstream direction. Each sampling site only was sampled once, often had an area larger
183
than 100 m2 (560±50 m2) and included one riffle-run-pool sequence (e.g. Kennard et al.,
184
2007; Maceda–Veiga et al., 2017). All sites were wadeable stream reaches because the
185
efficiency of electrofishing is limited in deeper waters. The studied rivers exemplified
186
the small and medium-size streams typical of Mediterranean-climate regions (Sabater et
187
al., 2009). The survey of large rivers (e.g. main Ebro river channel) would have required
188
the use of fishing boats and, so, fish data would not have been comparable among
189
sampling sites. Our sampling procedure detects 80–100% of the fish species and
190
captures 60–90% of the individuals compared with estimates from four-pass
191
electrofishing (A.S., unpublished data). Fish captures were expressed as captures per
192
unit of effort (CPUE, individuals min-1 m-2 = number of individuals caught / (fishing
193
time in minutes x the area surveyed in m2).
6
194
The conduct of fish surveys was authorized by the Spanish Government and the
195
Autonomous Government of Catalonia (AP/003). All fish were placed in buckets and
196
identified to species. A random sample of each fish species (up to 40 individuals if
197
possible) was collected for recording weight (0.01 g), which was used to express CPUE
198
in biomass. Fish were anaesthetized with buffered MS-222 (0.02%, Sigma-Aldrich, St.
199
Louis, MO, USA). All fish were allowed to recover in buckets provided with air pumps
200
and returned alive at the site of capture. The status of fish species was defined as: (1)
201
alien, if the species did not occur historically in a basin and in the Iberian Peninsula
202
(IP); or (2) native translocated, if their presence is the result of an introduction from
203
another basin within the studied basins in which the species is native (Table 3). Alien
204
and native translocated species were pooled together and regarded as ‘introduced’.
205
2.3. Trophic level measures
206
Five trophic level (TL) measures were calculated for the fish assemblage in each
207
sampling site: (1) maximum TL; (2) mean TL; (3) TL range; (4) TL variance; and (5)
208
weighted-mean TL. Mean TL is the arithmetic mean calculated using the TL of all fish
209
species. TL range is the difference between the fish species with the minimum TL and
210
the one with the maximum TL. TL variance is the variance calculated using the TL of
211
all fish species. Weighted mean TL is calculated: WmTL = (Y1X1 + Y2X2 + … +
212
YnXn)/(Y1+ Y2 + … + Yn), where Yi is the relative biomass of each fish species (CPUE
213
expressed in biomass) in each sampling site, and Xi is the corresponding TL.
214
The TL value of each fish species was gathered from the online database
215
FishBase (Froese & Pauly, 2018; see also Branch et al., 2010; Romanuk et al., 2011).
216
The FishBase TL is estimated using the TROPH routine (Pauly et al., 2000): TLi = 1 +
217
∑𝑆𝑗=1 𝐷𝐶 ijTLj, where TLi is the fractional TL of prey j, DCij is the fraction of prey j in the
218
diet of the fish species i and S is the total number of prey species consumed by the fish
219
species i. The resultant TL estimates are correlated highly with those from stable
220
isotope ratios (Kline & Pauly, 1998) and are consistent with the classification of our fish
221
species pool into trophic guilds using regional fish atlases (e.g. omnivores, piscivores;
222
de Sostoa et al., 1990; Doadrio et al., 2011). Therefore, notwithstanding the limitations
223
of Fishbase data, Fishbase TLs are likely to be good proxies for the trophic position of
224
fishes (Mancinelli et al., 2013; Amezcua et al., 2015). However, we acknowledge that
225
fish species diet may vary even within the same river (Colin et al., in preparation; Mas-
7
226
Martí et al., 2010). Nonetheless, such detailed stomach content data were not available
227
for many of the 34 fish species captured in our extensive surveys.
228 229
2.5. Data analyses
230
Analyses were done in R v 3.3.3. (R Core Team, 2016) using the packages,
231
functions and steps outlined below. Spatial gradients of variation in environmental
232
variables were used in models instead of classifying sampling sites as ‘pristine’ and
233
‘disturbed’ because community approaches based on gradient analyses are more
234
realistic and objectively defined representations of natural systems (Friberg et al., 2011).
235
Rivers are a continuum of community structures varying along environmental gradients
236
(Vannote et al., 1980). Moreover, the actual toxicities of conductivity, nutrients and pH
237
are largely unknown for fish species in north-eastern Spain. Therefore, a pristine vs
238
disturbed dichotomy would be arbitrary, especially if upstream and downstream fish
239
species are expected to have different tolerances to the same water quality variables (see
240
Maceda-Veiga et al., 2018). Nonetheless, nutrient concentrations were aggregated based
241
on their toxic severity (Camargo & Alonso, 2006; Noga, 2011), such as ammonium and
242
nitrite (TN) and nitrate and phosphate–P (NP), with the latter being the least toxic at
243
low concentrations.
244 245 246
2.5.1. Pair-wise correlations and data transformation Visual inspection of the distribution of all variables showed that continuous
247
variables were right-skewed and, so, log-transformations were undertaken. Spearman’s
248
correlation coefficients were used to select relatively uncorrelated variables and to
249
reduce collinearity among all ordinal and continuous environmental variables. There
250
were only two highly correlated pairs, namely, temperature and elevation, and pH and
251
general water hardness (both Spearman’s
252
elevation represents better within-day and inter-seasonal variation in water temperature
253
than a snapshot measure at the time of sampling fish. Conductivity is less affected than
254
pH by natural mineral content (Spearman’s
255
as indicator of water pollution.
256
> 0.80). We excluded temperature because
= 0.42), so that conductivity was retained
The retained environmental variables (Appendix S2) were standardized (mean =
257
0, standard deviation [SD] = 1) to be modelled on comparable scales (units of SD).
258
Collinearity among these environmental variables was of little statistical concern (All
8
259
Variation Inflation Factors < 2; Appendix S1; Zuur et al., 2010). The environmental
260
variables were grouped into four sets (i.e. geography, in-stream hydraulic, physical
261
habitat and water quality) to reflect that resource managers’ actions are likely to treat all
262
variables in a set together (e.g. an advanced tertiary treatment will act to reduce all
263
nutrients and conductivity rather than, say, just nitrate). However, environmental
264
variables within each set were included individually in models.
265
Spearman’s
correlation coefficients were used to check redundancy among the
266
five TL measures, and to determine relationships between the least correlated TL
267
measures and the richness of native, native translocated and alien species. The five
268
trophic measures used to characterize fish assemblages (maximum TL, mean TL, TL
269
range, TL variance and weighted-mean TL) were all significantly correlated (Table 4).
270
However, TL variance and TL range and mean TL and weighted-mean TL were
271
strongly correlated (Spearman’s
272
variance for next steps.
> 0.80) and, so, we retained only mean TL and TL
273 274
2.5.2. Overview of modelling techniques
275
We used general linear mixed models (GLMMs) and hierarchical partitioning
276
analyses of variance (HP) as two complementary approaches to evaluate associations
277
between TL measures and environmental variables. Environmental variables are always
278
correlated, even when the most highly correlated are excluded (Appendix S2), which
279
can bias regression estimates of GLMMs (Mac Nally, 2000; Freckleton, 2011). GLMMs
280
can include random effects, so we included basin and year as fixed factors. While HP
281
cannot accommodate random effects nor include categorical variables (e.g. basin), the
282
method is one of the best ways to allocate independent explained variance among
283
correlated predictors (Mac Nally, 2000), and so complements GLMMs.
284
The support for the three FCL hypotheses and the underlying trophic structure
285
was determined by assessing the effect size of each environmental variable identified as
286
important by both GLMM and HP models to explain variation in trophic measures.
287 288
2.5.2.1. General linear mixed models
289
Series of Gaussian linear mixed models were built using the function glmer in the lme4
290
package (Bates et al., 2015) to identify the riverine environmental variables that best
291
represented the variation in the TL measures of fish assemblages. All models included
9
292
random intercepts (collectively referred to as ) for year and river basin, which
293
accounted for potential systematic differences among components of these factors. The
294
most plausible model for each TL measure was the one with the lowest AIC value
295
(models with ΔAIC < 2.0; Appendix S2). The candidate models included all possible
296
additive combinations of the four sets of environmental variables, with individuals
297
variables in each set entered separately (Appendix S3). Geographical variables (river
298
size and elevation) were included with interactions with the other environmental
299
variables to account for potentially different effects of productivity or human
300
disturbance on the fish depending on size or elevation (Appendix S4). Rivers can differ
301
in productivity due to size (Schoener, 1989), and the effects of human disturbances on
302
ectotherms may intensify in lowlands because warmer waters increase their metabolism.
303
However, the inclusion of such interactions is unlikely to be warranted statistically. For
304
example, other analyses at the assemblage scale using the same data-set showed a low
305
fraction of variance shared between the sets of environmental variables (e.g. < 0.06;
306
Maceda-Veiga et al., 2017). Similarly, additive models of environmental variables on
307
other response variables from the same data-set had lower AIC values than did ones
308
with interactions, so that the probability of interactions being important is low (Maceda-
309
Veiga et al., 2018).
310
The selection of environmental variables in the best models for each TL measure
311
was refined further by eliminating predictors that were not significant, or whose 95%
312
confidence intervals included 0, to avoid over-parameterizing models. Significance was
313
assessed using 𝜒 2 tests at P ≤ 0.05 (the function Anova in the car package, Fox &
314
Weisberg, 2011). These final models were checked for overall adequacy using the
315
function r.squaredGLMM (R2) in the MuMIn package (Barton, 2013), and the visual
316
inspection of model residuals following Zuur et al. (2010). Fits were decomposed into
317
the shared variance between random and ‘fixed’ factors (R2c) and the variance
318
explained by the ‘fixed’ factors alone (R2m).
319 320
2.5.2.2. Hierarchical partitioning analysis of variance
321
The relative importance of the environmental variables selected by the AIC approach in
322
explaining variation in trophic level measures was evaluated by using hierarchical
323
partitioning analysis of variance (HP; function hier.part, Walsh et al., 2003). HP
324
partitions the independent contributions of each environmental variable to variation in
325
the response variable, and distinguishes these from the joint contribution with other 10
326
environmental variables (i.e. collinearity). The function rand.hp was used to assess the
327
significance of HP models using a randomization test for hierarchical partitioning
328
analysis based on the upper 95% confidence limit.
329
3. Results
330
3.1. Characteristics of fish assemblages and trophic-level measures
331
We caught 16 native and 21 introduced fish species, including three native-translocated
332
species and two uncertain taxa (Phoxinus spp. and Gobio spp.) (Table 3). Based on
333
traditional-diet analyses, 66% of the introduced fish species were defined as omnivores
334
and 33% of them were mostly invertivores that also prey upon small vertebrates (i.e.
335
piscivores in Table 3). Conversely, omnivory is the most frequent guild (75%) among
336
native fish species, including native translocated (Table 3). Such differences in trophic
337
guilds were mirrored in their Fishbase trophic-level estimates (TL), with native fish
338
having lower mean and maximum TL values (mean±standard error = 3.27±0.08;
339
minimum-maximum range = 2.76-3.80) than alien fish (3.40±0.12; 2.70-4.42) (Table 3).
340
Native fish richness had stronger associations with the least correlated TL
341
measures (maximum and mean TL and TL variance) than did native translocated and
342
alien fish richness (Fig. 3). Maximum TL and TL variance increased with all three fish
343
richness measures, while mean TL reduced (Fig. 3).
344 345
3.2. Associations among environmental variables and trophic level measures
346
The most parsimonious general linear mixed models (GLMMs) for the TL measures
347
were ones including only additive combinations of environmental variables (Appendix
348
S3) rather than those also including interactions with elevation and Strahler stream
349
order (Appendix S4). That is, the inclusion of interactive effects of geographical
350
variables with the other environmental variables to the base additive model increased
351
model complexity without much improving model fit, so that the additive base model
352
was considered the more probable (Appendix S4). All final GLMMs included water and
353
physical habitat as important variable sets for TL measures after accounting for the
354
confounding effects of geographical factors (Appendix S3 and S5). Model fits (R2),
355
including the explained variance of random and fixed effects, ranged from 0.21 to 0.35.
356
The maximum TL of fish assemblages (i.e. food-chain length, FCL) was
357
positively associated with Strahler-stream order and physical-habitat diversity, while
358
conductivity was negatively associated with FCL (Table 5, Fig. 4). Similarly, there was
359
a significant inverse association between mean TL and conductivity, dead wood and 11
360
Strahler stream order (Fig. 4). Macrophyte cover was positively associated with mean
361
TL (Table 5; Fig. 4). Last, there was a positive association between within-assemblage
362
TL variance and pH, conductivity and Strahler stream order (Fig. 4).
363
Outputs of GLMMs are unlikely to be much affected by collinearity among
364
environmental variables. Spearman rank correlation coefficients and variation inflation
365
factors were low (Appendix S2), and the outputs of GLMMs, particularly for FCL, were
366
consistent with those of HP models, which specifically accounted for collinearity (Table
367
6). However, GLMMs for mean TL included macrophyte cover as significant factor but
368
HP models did not (Table 6). Similarly, GLMMs for TL variance identified
369
conductivity as important predictor but HP models did not (Table 6). Considering the
370
independent contribution of environmental variables related to FCL hypotheses (Tables
371
1 and 6), variation in the FCL of fish assemblages mostly was attributed to conductivity
372
and to Strahler stream order and, to a minor degree, physical habitat diversity (Table 4).
373
Variation in mean TL was much associated with conductivity and, also to a lesser
374
extent, with Strahler stream order and dead wood (Table 4). Within-assemblage TL
375
variance mostly was attributable to Strahler stream order and to pH, while conductivity
376
had little explanatory power (Table 4).
377 378
4. Discussion
379
Fish assemblages in north-eastern Spain have up to 18 alien fish species, which had an
380
overall higher trophic level (TL) than native fish species. Fish piscivores were rare in
381
these catchments but have increased with alien fish introductions due to the demands of
382
recreational fishing (e.g. Micropterus salmoides, Sander lucioperca; Doadrio et al.,
383
2011). Despite this, our results provided some support for two food-chain length (FCL)
384
hypotheses, namely the ecosystem size (Post et al., 2000) and disturbance (Pimm &
385
Lawton, 1977) hypotheses, which previously had been explored concurrently only in
386
uninvaded ecosystems (Takimoto & Post, 2013; Warfe et al., 2018). Much of the
387
variance in our models remained unexplained (69–79%), possibly because fish are
388
highly mobile (Radinger & Wolter, 2014) and regional-scale processes may obscure the
389
effects of local environmental variables on food-webs (Holt, 2002; Calcagno et al.,
390
2011; Warfe et al., 2018). However, such levels of explained variance are the norm for
391
complex models typical in modern ecological research (Low-Décarie et al., 2014; Heino
392
et al., 2015), including many food-web studies (e.g. Baiser et al., 2011; Takimoto &
393
Post, 2013; Warfe et al., 2018). Nevertheless, the support for the two FCL hypotheses 12
394
was apparent after accounting for spatial confounding effects and collinearity among
395
environmental variables (Mac Nally, 2000).
396
Strahler stream order and conductivity made the largest independent
397
contributions to variation in the maximum TL of fish assemblages, which provided
398
evidence for the ecosystem-size (Post et al., 2000) and disturbance FCL-hypotheses
399
(Pimm & Lawton, 1977) but not for the productivity hypothesis (Schoener, 1989). Our
400
results are consistent with the only two studies that have explored the three FCL-
401
hypotheses in temperate streams (McHugh et al., 2010; Sabo et al., 2010). Conductivity
402
was more influential than Strahler stream order, which may indicate that disturbance
403
may be even more important than has been suggested to explain variation in FCL (see
404
McHugh et al., 2010). Human-disturbance effects on fish are likely to be difficult to
405
distinguish from the effects of Strahler stream order because larger, lower catchment
406
streams often are the most human-altered in our region (Maceda-Veiga et al., 2017).
407
However, the use of modelling techniques that specifically deal with collinearity, such
408
as hierarchical partitioning analyses, helps to distinguish the individual effects of
409
correlated environmental variables (Mac Nally, 2000). We note that there were broad
410
ranges in water quality in streams from all Strahler orders (Appendix S6), so our
411
findings are unlikely to be much influenced by systematic differences in water quality
412
as a function of the catchment position of streams.
413
Sabo et al. (2009) suggested that a marked association between FCL and
414
disturbance might be detectable only if extreme ranges are considered. Our relatively
415
strong FCL-disturbance association may be due to the broad range of conductivity
416
values in our region (20–5220 µS/cm) (Table 1). Despite variation in conductivity being
417
affected by natural differences in mineral content, we are confident that our
418
conductivity values indicated intense, frequent human-disturbance due to the multiple
419
inputs from sewage-treatment plants, agricultural lands and potash mines in north-
420
eastern Spain (Prat & Munné, 2000; Fernandez-Turiel et al., 2003). These activities load
421
rivers with many ions (e.g. chlorides, copper, sodium, potassium) that alter
422
conductivity, which can kill invertebrates (Castillo et al., 2017). Reduced prey
423
abundance may explain why the FCL of fish assemblages declined with conductivity,
424
possibly because species at the highest trophic levels are the most energy-demanding in
425
food-webs (Pimm & Kitching, 1987). However, these inferences require further
426
assessment because fish invertivores, which often also prey on small vertebrates in
427
north-eastern Spain (e.g. Anguilla anguilla, M. salmoides, Lepomis gibbosus), often 13
428
occur in polluted rivers (Segurado et al., 2011) with conductivity values >1000 µS/cm
429
(Maceda-Veiga et al., 2018).
430
Our inverse association between conductivity and FCL did not obscure the
431
positive association between FCL and Strahler stream order (i.e. the ecosystem-size
432
hypothesis), which is the most widely supported FCL-hypothesis (Post, 2002; Takimoto
433
& Post, 2013; although see Warfe et al., 2018). Our positive association between FCL
434
and physical habitat diversity is consistent with large rivers being spatially
435
heterogeneous and hence supporting longer food-chains compared to small streams
436
(Schoener, 1989; Post, 2002). The positive association between within-assemblage TL
437
variance and Strahler-stream order also was consistent with the ecosystem-size
438
hypothesis. The difference between the maximum and minimum TL of fish assemblages
439
paralleled difference in within-assemblage TL variance (Table 4), which is in broad
440
agreement with the idea that trophic diversity increases with ecosystem size (Post et al.,
441
2000). The longer, diverse fish food-chains in larger rivers may be due to these rivers
442
generally having flowing water all year and fish being more able to move more freely in
443
larger rivers than in smaller streams. Seasonal flow interruption occurs in some smaller
444
streams in Mediterranean climates, and fish often become isolated in pools during the
445
drought season until fish re-colonization can begin when flow is restored (Pires et al.,
446
2010; Doadrio et al., 2011). In our study, the lack of time series data for river flow may
447
have obscured the associations between in-stream hydraulics and TL measures
448
considering the effect that variation in river flow may have on the trophic structure of
449
fish assemblages in Mediterranean streams (Mas-Martí et al., 2010; Mor et al., 2018).
450
Strahler stream order was positively associated with the maximum TL of fish
451
assemblages in north-eastern Spain but had an inverse association with mean TL (Fig.
452
4). Such a pattern might arise if just a few high-TL alien fish species (e.g. piscivores
453
such as Silurus glanis, M. salmoides) are offset by larger numbers of low-TL alien fish
454
invaders (e.g. Cyprinus carpio, Alburnus alburnus, Rutilus rutilus) in larger rivers
455
compared to upland streams. There was a positive association between within-
456
assemblage TL variance and Strahler stream order (Fig. 4), suggesting an increased
457
range of TLs in larger rivers compared with smaller, low-Strahler order streams. Species
458
richness of alien fish was negatively associated with mean TL (Fig. 3), and large rivers
459
had the highest alien fish richness (Maceda-Veiga et al., 2017). Therefore, this
460
explanation is consistent with large streams being colonized by more low-TL alien fish
461
species in comparison with upstream, low-Strahler order streams (Appendix S1), where 14
462
the most pristine fish assemblages are found (e.g. Barbus haasi, Salmo trutta; de Sostoa
463
et al., 1990; Maceda-Veiga et al., 2017). This may include trophic replacements of
464
native with alien fish species in downstream sites, such as the extirpation of the natives
465
A. anguilla and B. haasi and the introduction of C. carpio (Fig. 1).
466
There was no evidence that models with interactions between elevation or
467
Strahler stream order and productivity-related variables (e.g. nutrients, macrophyte
468
cover, Table 2) provided better inferences for explaining variation in FCL than those
469
with just additive effects for all environmental variables (Appendix S4). Moreover,
470
additive models did not identify any productivity-related variables as important
471
predictors of variation in FCL, suggesting weak support for the productivity hypothesis
472
(Schoener, 1989). Our results may support the productive-space hypothesis, which
473
posits that FCL lengthens because basal resources increase with ecosystem size
474
(Schoener, 1989; Post, 2002). It is possible we would have obtained more definitive
475
results if we had been able to use more direct measures of productivity, such as the
476
chlorophyll-a concentration, but we could not measure chlorophyll-a concentration over
477
such a large-scale. Moreover, chlorophyll-a concentration is more variable in space and
478
time than, say, macrophyte cover, so a snap-shot measurement, which was potentially
479
possible to make, probably is not a reliable value for summarizing productivity. Despite
480
these limitations, the positive association between pH and within-assemblage TL
481
variance suggests that productivity may promote trophic diversity in fish food-webs.
482
Other factors than those we have not been able to consider might influence TL
483
measures in this and other river systems much affected by human disturbances.
484
Fisheries can shape the structure of fish assemblages (Pauly et al., 2002), and all the
485
studied native and alien fish species are potentially valuable in angling (Doadrio et al.,
486
2011). Changes in the trophic structure of assemblages may arise from water
487
impoundments (e.g. Hoeinghaus et al., 2008; Mor et al., 2018), although we attempted
488
to account for such an effect with hydrological measures (e.g. water speed, river
489
morphology). Within-river physical barriers are threatening migratory fish around the
490
world, including large barbels (Luciobarbus graellsii), S. trutta and A. anguilla, being
491
the latter two the few native piscivores in rivers of north-eastern Spain (Doadrio et al.,
492
2011). Last, estimates from Fishbase were used as proxy of fish species TL and,
493
although these estimates reflect the TL of many fish records (Mancinelli et al., 2013;
494
Amezcua et al., 2015; Froese & Pauly, 2018), they do not necessarily account well for
495
intraspecific variability. There may be differences among age classes, the sexes, and 15
496
baseline isotopic signatures in different sub-catchments than may induce wide spreads
497
in the de facto TL of fish species (Layman et al., 2015). Knowledge about the
498
mechanisms behind our associations would be refined by more detailed trophic analyses
499
(gut-contents and isotopic analyses).
500 501
5. Management implications
502
The reasonable support for the ecosystem size and disturbance FCL-hypotheses
503
provides guidance at scales relevant to resource managers. High conductivity remains a
504
major environmental issue in rivers up to 10 years after the enforcement of the Water
505
Framework Directive in Europe (EC, 2000). Improving water quality long has been
506
advocated for its multiple benefits in biodiversity conservation, including the
507
advantages that rivers provide to human society (Sabater et al., 2009; Hering et al.,
508
2015). Our study shows that another benefit would be the conservation of fish from
509
apical TLs. This is especially important in small Mediterranean-type rivers, in which
510
one or two fish species often exert top-down control of the aquatic food-web (e.g.
511
Rodríguez-Lozano et al., 2015). Despite the largest rivers being the most human-altered,
512
we found evidence of them still having the most trophic diverse and largest food-webs.
513
Alien fish species may be occupying the trophic level left empty by native species in
514
marked decline (e.g. Anguilla anguilla; Maceda-Veiga et al., 2010; Colin et al., 2018).
515
However, these inferences and the associated implications for the functioning of
516
riverine ecosystems require in-deep study.
517 518
Acknowledgements
519
We are grateful to all people who assisted in the field and to the useful advice provided
520
by five anonymous referees. We are also in debt with Irene Jiménez and the Laboratory
521
of GIS and Remote Sensing at EBD for help with the figures. The surveys were
522
financed by regional water agencies (e.g. Agència Catalana de l’Aigua, ACA and
523
Confederación Hidrográfica del Ebro, CHE), natural parks (e.g. Sant Llorenç del Munt i
524
Serra de l’Obac, Collserola, Montseny) and municipalities (Ajuntament de Sabadell).
525
A.M.V. was funded during analyses and writing by a grant from ‘Fundació Barcelona
526
Zoo-Ajuntament de Barcelona’ and a Severo-Ochoa contract at the EBD-CSIC.
527
16
528
References
529 530 531
Amezcua, F., Muro-Torres, V., & Soto-Jiménez, M. F. (2015). Stable isotope analysis versus TROPH: a comparison of methods for estimating fish trophic positions in a subtropical estuarine system. Aquatic Ecology, 49, 235-250.
532 533 534
Barbour, M. T., Gerritsen, J., Snyder, B. D., & Stribling, J. B. (1999). Rapid bioassessment protocols for use in streams and wadeable rivers: periphyton, benthic macroinvertebrates and fish(Vol. 339). Washington, DC: US Environmental Protection Agency, Office of Water.
535 536
Baiser, B., Ardeshiri, R. S., & Ellison, A. M. (2011). Species richness and trophic diversity increase decomposition in a co-evolved food web. PLoS One, 6, e20672.
537 538
Bonada, N., & Resh, V. H. (2013). Mediterranean-climate streams and rivers: geographically separated but ecologically comparable freshwater systems. Hydrobiologia, 719, 1-29.
539 540
Branch, T. A., Watson, R., Fulton, E. A., Jennings, S., McGilliard, C. R., Pablico, G. T. et al. (2010). The trophic fingerprint of marine fisheries. Nature, 468, 431.
541 542
Byrnes, J. E., Reynolds, P. L., & Stachowicz, J. J. (2007). Invasions and extinctions reshape coastal marine food webs. PloS One, 2, e295.
543 544 545
Calcagno, V., Massol, F., Mouquet, N., Jarne, P., & David, P. (2011). Constraints on food chain length arising from regional metacommunity dynamics. Proceedings of the Royal Society of London B: Biological Sciences, 278, 3042-3049.
546 547
Camargo, J. A., & Alonso, Á. (2006). Ecological and toxicological effects of inorganic nitrogen pollution in aquatic ecosystems: a global assessment. Environment international, 32, 831-849.
548 549 550
Castillo, A. M., Sharpe, D. M., Ghalambor, C. K., & De León, L. F. (2017). Exploring the effects of salinization on trophic diversity in freshwater ecosystems: a quantitative review. Hydrobiologia, 1-17.
551 552
Chandra, S., & Gerhardt, A. (2008). Invasive species in aquatic ecosystems: issue of global concern. Aquatic Invasions, 3(1), 1-2.
553 554 555
Colin, N., Maceda-Veiga, A., Flor-Arnau, N., Mora, J., Fortuño, P., Vieira, C. et al. (2016). Ecological impact and recovery of a Mediterranean river after receiving the effluent from a textile dyeing industry. Ecotoxicology and Environmental Safety, 132, 295-303.
556 557 558
Colin, N., Villéger, S., Wilkes, M., de Sostoa, A., & Maceda-Veiga, A. (2018). Functional diversity measures revealed impacts of non-native species and habitat degradation on speciespoor freshwater fish assemblages. Science of the Total Environment, 625, 861-871.
559 560
Doadrio, I. (Ed). Ictiofauna española. Bases para su seguimiento. Ministerio de Agricultura y Pesca (In Spanish). 2011.
563 564
Eby, L. A., Roach, W. J., Crowder, L. B., & Stanford, J. A. (2006). Effects of stocking-up freshwater food webs. Trends in Ecology & Evolution, 21, 576-584.
565 566 567
Fernandez-Turiel, J. L., Gimeno, D., Rodriguez, J. J., Carnicero, M., & Valero, F. (2003). Spatial and seasonal variations of water quality in a Mediterranean catchment: the Llobregat River (NE Spain). Environmental Geochemistry and Health, 25, 453-474. 17
568 569 570
Figuerola, B., Maceda-Veiga, A., & De Sostoa, A. (2012). Assessing the effects of sewage effluents in a Mediterranean creek: fish population features and biotic indices. Hydrobiologia, 694, 75-86.
571
Fox, J., & Weisberg, S. (2011). An R companion to applied regression. Sage Publications.
572 573 574
Freckleton, R. P. (2011). Dealing with collinearity in behavioural and ecological data: model averaging and the problems of measurement error. Behavioral Ecology and Sociobiology, 65, 91-101.
575 576
Freedman, B. (1995). Environmental ecology: the ecological effects of pollution, disturbance, and other stresses. Elsevier.
577 578 579
Friberg, N., Bonada, N., Bradley, D. C., Dunbar, M. J., Edwards, F. K., Grey, J. et al. (2011). Biomonitoring of human impacts in freshwater ecosystems: the good, the bad and the ugly. In Advances in Ecological Research (Vol. 44, pp. 1-68). Academic Press.
580 581
Froese, R. and D. Pauly (2018) FishBase. World Wide Web electronic publication. www.fishbase.org, (04/04/2018).
582 583 584 585
Hering, D., Aroviita, J., Baattrup‐Pedersen, A., Brabec, K., Buijse, T., Ecke, F. et al. (2015). Contrasting the roles of section length and instream habitat enhancement for river restoration success: a field study of 20 European restoration projects. Journal of Applied Ecology, 52, 1518-1527.
586 587 588
Kefford, B. J., Hickey, G. L., Gasith, A., Ben-David, E., Dunlop, J. E., Palmer, C. G. et al. (2012). Global scale variation in the salinity sensitivity of riverine macroinvertebrates: Eastern Australia, France, Israel and South Africa. PloS one, 7, e35224.
589 590 591
Kennard, M. J., Olden, J. D., Arthington, A. H., Pusey, B. J., & Poff, N. L. (2007). Multiscale effects of flow regime and habitat and their interaction on fish assemblage structure in eastern Australia. Canadian Journal of Fisheries and Aquatic Sciences, 64, 1346-1359.
592 593
Kline, T. C., & Pauly, D. (1998). Cross-validation of trophic level estimates from a massbalance model of Prince William Sound using 15 N/14 N data. In Alaska Sea Grant College.
594 595
Kottelat, M., & Freyhof, J. (2007). Handbook of European freshwater fishes. Publications Kottelat.
596 597 598
Heino, J., Melo, A. S., Bini, L. M., Altermatt, F., Al‐Shami, S. A., Angeler, D. G. et al. (2015). A comparative analysis reveals weak relationships between ecological factors and beta diversity of stream insect metacommunities at two spatial levels. Ecology and Evolution, 5, 1235-1248.
599 600 601
Hoeinghaus, D. J., Winemiller, K. O., & Agostinho, A. A. (2008). Hydrogeomorphology and river impoundment affect food‐chain length of diverse Neotropical food webs. Oikos, 117, 984995.
602 603
Holt, R. D. (2002). Food webs in space: on the interplay of dynamic instability and spatial processes. Ecological Research, 17, 261-273.
18
604 605 606
Layman, C. A., Giery, S. T., Buhler, S., Rossi, R., Penland, T., Henson, M. N. et al. (2015). A primer on the history of food web ecology: Fundamental contributions of fourteen researchers. Food Webs, 4, 14-24.
607 608 609
LeBlanc, G. A. (1995). Trophic-level differences in the bioconcentration of chemicals: implications in assessing environmental biomagnification. Environmental Science & Technology, 29, 154-160.
610 611 612
Leprieur, F., Beauchard, O., Blanchet, S., Oberdorff, T., & Brosse, S. (2008). Fish invasions in the world's river systems: when natural processes are blurred by human activities. PLoS Biology, 6, e28.
613
Lindeman R.L. (1942) The trophic-dynamic aspect of ecology. Ecology, 23, 399–417
614 615
Low-Décarie E, Chivers C, Granados M (2014). Rising complexity and falling explanatory power in ecology. Frontiers in Ecology and the Environment, 12, 412-418.
616 617 618
Mac Nally R. (2000) Regression and model-building in conservation biology, biogeography and ecology: The distinction between - and reconciliation of - 'predictive' and 'explanatory' models. Biodiversity and Conservation, 9, 655-671.
619 620 621
Maceda-Veiga, A., Monleon-Getino, A., Caiola, N., Casals, F., & de Sostoa, A. (2010). Changes in fish assemblages in catchments in north-eastern Spain: biodiversity, conservation status and introduced species. Freshwater Biology, 55, 1734-1746.
622 623 624
Maceda-Veiga, A., Monroy, M., Navarro, E., Viscor, G., & de Sostoa, A. (2013a). Metal concentrations and pathological responses of wild native fish exposed to sewage discharge in a Mediterranean river. Science of the Total Environment, 449, 9-19.
625 626 627
Maceda-Veiga, A., de Sostoa, A., & Sánchez-Espada, S. (2013b). Factors affecting the establishment of the invasive crayfish Procambarus clarkii (Crustacea, Decapoda) in the Mediterranean rivers of the northeastern Iberian Peninsula. Hydrobiologia, 703, 33-45.
628 629 630
Maceda-Veiga, A., Baselga, A., Sousa, R., Vilà, M., Doadrio, I., & de Sostoa, A. (2017). Finescale determinants of conservation value of river reaches in a hotspot of native and non-native species diversity. Science of the Total Environment, 574, 455-466.
631 632 633
Mancinelli, G., Vizzini, S., Mazzola, A., Maci, S., & Basset, A. (2013). Cross-validation of δ15N and FishBase estimates of fish trophic position in a Mediterranean lagoon: the importance of the isotopic baseline. Estuarine, Coastal and Shelf Science, 135, 77-85.
634 635 636
Mas-Martí, E., García-Berthou, E., Sabater, S., Tomanova, S., & Munoz, I. (2010). Comparing fish assemblages and trophic ecology of permanent and intermittent reaches in a Mediterranean stream. Hydrobiologia, 657, 167-180.
637 638
McHugh P.A., McIntosh A.R. & Jellyman P.G. (2010). Dual influences of ecosystem size and disturbance on food chain length in streams. Ecology Letters, 13, 881–890.
639 640 641
Mor, J. R., Ruhí, A., Tornés, E., Valcárcel, H., Muñoz, I., & Sabater, S. (2018). Dam regulation and riverine food-web structure in a Mediterranean river. Science of the Total Environment, 625, 301-310.
19
642 643 644
Munné, A., Prat, N., Sola, C., Bonada, N., & Rieradevall, M. (2003). A simple field method for assessing the ecological quality of riparian habitat in rivers and streams: QBR index. Aquatic Conservation: Marine and Freshwater Ecosystems, 13, 147-163.
645 646 647 648
Munné, A., Tirapu, L., Sola, C., Olivella, L., Vilanova, M., Ginebreda, A., & Prat, N. (2012). Comparing chemical and ecological status in Catalan rivers: analysis of river quality status following the Water Framework Directive. In Emerging and Priority Pollutants in Rivers (pp. 243-265). Springer, Berlin, Heidelberg.
649 650 651
Muñoz, I., López‐Doval, J. C., Ricart, M., Villagrasa, M., Brix, R., Geiszinger, A. et al. (2009). Bridging levels of pharmaceuticals in river water with biological community structure in the Llobregat river basin (Spain). Environmental Toxicology and Chemistry, 28, 2706-2714.
652
Noga, E. J. (2011). Fish disease: diagnosis and treatment. John Wiley & Sons.
653 654
Pace, M. L., Cole, J. J., Carpenter, S. R., & Kitchell, J. F. (1999). Trophic cascades revealed in diverse ecosystems. Trends in Ecology & Evolution, 14, 483-488.
655 656
Pauly, D., Christensen, V. & Walters, C. (2000) Ecopath, Ecosim and Ecospace as tools for evaluating ecosystem impact of fisheries. ICES Journal of Marine Science, 57, 697–706.
657 658
Pauly, D., Christensen, V., Guénette, S., Pitcher, T. J., Sumaila, U. R., Walters, C. J. et al. (2002). Towards sustainability in world fisheries. Nature, 418, 689.
659 660
Persson, L. (1999). Trophic cascades: abiding heterogeneity and the trophic level concept at the end of the road. Oikos, 385-397.
661
Pimm, S. L. (1982). Food webs. In Food webs (pp. 1-11). Springer, Dordrecht.
662 663
Pimm, S. L., & Lawton, J. H. (1977). Number of trophic levels in ecological communities. Nature, 268, 329.
664
Pimm, S. L., & Kitching, R. L. (1987). The determinants of food chain lengths. Oikos, 302-307.
665 666 667
Pires, D. F., Pires, A. M., Collares‐Pereira, M. J., & Magalhães, M. F. (2010). Variation in fish assemblages across dry‐season pools in a Mediterranean stream: effects of pool morphology, physicochemical factors and spatial context. Ecology of Freshwater Fish, 19, 74-86.
668 669
Post, D. M., Pace, M. L., & Hairston Jr, N. G. (2000). Ecosystem size determines food-chain length in lakes. Nature, 405, 1047.
670 671
Post, D. M. (2002). The long and short of food-chain length. Trends in Ecology & Evolution, 17, 269-277.
672 673
Prat, N., & Munné, A. (2000). Water use and quality and stream flow in a Mediterranean stream. Water Research, 34, 3876-3881.
674 675
Radinger, J., & Wolter, C. (2014). Patterns and predictors of fish dispersal in rivers. Fish and Fisheries, 15, 456-473.
676 677 678
Rodríguez‐Lozano, P., Verkaik, I., Maceda‐Veiga, A., Monroy, M., Sostoa, A., Rieradevall, M., & Prat, N. (2016). A trait‐based approach reveals the feeding selectivity of a small endangered Mediterranean fish. Ecology and evolution, 6, 3299-3310.
20
679 680
Romanuk, T. N., Hayward, A., & Hutchings, J. A. (2011). Trophic level scales positively with body size in fishes. Global Ecology and Biogeography, 20, 231-240.
681 682
Sabater, S., Feio, M. J., Graça, M. A., Muñoz, I., & Romaní, A. M. (2009). The Iberian Rivers. Rivers of Europe, 113-149.
683 684
Sabo, J. L., Finlay, J. C., Kennedy, T., & Post, D. M. (2010). The role of discharge variation in scaling of drainage area and food chain length in rivers. Science, 330, 965-967.
685 686 687
Sánchez-Montoya, M., Arce, M. I., Vidal-Abarca, M. R., Suárez, M. L., Prat, N., & Gómez, R. (2012). Establishing physico-chemical reference conditions in Mediterranean streams according to the European Water Framework Directive. Water Research, 46, 2257-2269.
688 689 690
Segurado, P., Santos, J. M., Pont, D., Melcher, A. H., Jalon, D. G., Hughes, R. M., & Ferreira, M. T. (2011). Estimating species tolerance to human perturbation: expert judgment versus empirical approaches. Ecological Indicators, 11, 1623-1635.
691 692
Schoener, T. W. (1989). Food webs from the small to the large: the Robert H. MacArthur Award Lecture. Ecology, 70, 1559-1589.
693 694 695
Smallbone, W., Cable, J., & Maceda-Veiga, A. (2016). Chronic nitrate enrichment decreases severity and induces protection against an infectious disease. Environment International, 91, 265-270.
696 697
Sostoa, A. D., Casals, F., Fernández, J. V., Lobón-Cervià, J., de Sostoa, F. J., Vinyoles, D. (1990). Peixos. Història Natural dels Països Catalans, 11 (In Catalan).
698 699
Takimoto, G. & D.M. Post. (2013). Environmental determinants of food-chain length: a metaanalysis. Ecological Research, 28, 675–681.
700 701 702
Thompson, R. M., Brose, U., Dunne, J. A., Hall Jr, R. O., Hladyz, S., Kitching, R. L. et al. (2012). Food webs: reconciling the structure and function of biodiversity. Trends in Ecology & Evolution, 27(12), 689-697.
703
Tockner, K., Uehlinger, U., & Robinson, C. T. (2009). Rivers of Europe. Academic Press.
704 705 706
Townsend, C.R., Begon, M., Harper, J.L. (2003). Part III Individuals, Populations, Communities and Ecosystems. In Essentials of Ecology. Second Edition. Blackwell Publishing. pp. 149-366.
707 708
Vannote, R. L., Minshall, G. W., Cummins, K. W., Sedell, J. R., & Cushing, C. E. (1980). The river continuum concept. Canadian Journal of Fisheries and Aquatic Sciences, 37, 130-137.
709 710
Walsh, C., Mac Nally, R., & Walsh, M. C. (2003). The hier. part package. Hierarchical Partitioning. R project for statistical computing. URL: http://cran. r-project. org.
711 712
Ward, C. L., & McCann, K. S. (2017). A mechanistic theory for aquatic food chain length. Nature Communications, 8, 2028.
713 714 715
Warfe, D. M., Jardine, T. D., Pettit, N. E., Hamilton, S. K., Pusey, B. J., Bunn, S. E. et al. (2013). Productivity, disturbance and ecosystem size have no influence on food chain length in seasonally connected rivers. PLoS One, 8, e66240.
21
716 717
Winemiller, K. O. (1990). Spatial and temporal variation in tropical fish trophic networks. Ecological monographs, 60, 331-367.
718 719
Zuur, A. F., Ieno, E. N., & Elphick, C. S. (2010). A protocol for data exploration to avoid common statistical problems. Methods in Ecology and Evolution, 1, 3-14.
720 721 722 723
22
Table 1 Median (±Q25-Q75interquartile range, I.R.) and minimum-maximum range of the environmental variables used to explore their associations with the trophic level measures of fish assemblages (N = 530 sampling sites). Descriptors/predictors Geography (geo) Elevation (m.a.s.l.) Strahler stream order In-stream hydraulics (hyd) Water depth (m) Water speed (m/s) Water quality (wq) Conductivity (µS/cm) pH Ammonium+Nitrite (mg/l) Nitrate-Phosphates (mg/l) Silt (%) Physical habitat (hab) Herbaceous margin cover (%) Macrophytes (%) Structural refuge (%) Physical habitat diversity (score) QBR (score) Dead wood (%) QBR: index of riparian forest quality
Median±I.R. (min-max) 405±466 (3-1814) 2±2 (1-8) 0.10±0.21 (0.10-1.37) 0.4±0.50 (0-2.5) 560±743 (20-5220) 8±1.20 (6.5-9.6) 0.02±0.17 (0-6.1) 26±4 (0-51) 3±5 (1-54) 28±59 (0-80) 14±20 (0-90) 12±25 (0-32) 9±10 (0-18) 25±35 (0-100) 5±10 (0-50)
Table 2 Rationale for including each environmental variable in the models regarding the three main food-chain length hypotheses and the expected associations to be found if supported. Note that direct and indirect effects are expected and that variables can be related to multiple hypotheses. Elevation was used as covariate as it represents the upstream-downstream gradient, which is an important factor governing the structure of aquatic assemblages. ECOSYSTEM SIZE Strahler stream order Water depth Structural refuge Physical habitat div.
FCL increases with spatial scale because larger ecosystems have greater habitat availability and diversity Positive: the higher is the Strahler order number, the larger is the river Positive: fish have more space in rivers with a high water column Positive: diversity and heterogeneity, including physical refuges, increases with river size Positive: habitat diversity and heterogeneity increases with river size
PRODUCTIVITY
Nutrient-poor ecosystems should have shorter FCL than nutrient-enriched ecosystems because of the energy loss in each trophic transfer Positive: they cause nutrient enrichment Positive: they cause nutrient enrichment Positive: indirect measure of productivity because photosynthesis reduces water CO2 and, so, pH increases Positive: we mostly found Potamogeton pectinatus, Ceratophyllum demersum, and Myriophyllum sp,. all of which indicate eutrophy Positive: river nutrients promote plant growth on riverbanks and riparian vegetation provides fish with terrestrial-sourced insects and leaf litter Positive: river nutrients promote riparian forest development and riparian forest provides fish with terrestrial-sourced food insects and leaf litter Positive: dead wood provides rivers with nutrients and habitat for fish prey
[NH3+NO2] [NO3+PO4] pH Macrophytes Herb cover QBR Dead wood DISTURBANCE Flow speed Conductivity Silt [NH3+NO2] [NO3+PO4] QBR Herb cover Dead wood
FCL is reduced with frequent and intense changes in environmental conditions because the longer the food chain, the less resistant the ecosystem is to perturbations, especially those species at the highest trophic levels Negative or positive: fish can be wiped out because of high flow velocity, which in turn promote habitat diversity and species co-occurrence Negative: extreme conductivity values in our data-set mostly indicate acute and diffuse pollution events Negative: silt accumulates because of sewage treatment plants overflooding, agricultural run-off and stagnant waters Negative: ammonium and nitrite can reach lethal concentrations or cause harmful man-made eutrophication Negative: nitrate and phosphate can reach lethal concentrations or cause harmful man-made eutrophication Negative: low scoring in this riparian index indicates severely altered riparian vegetation or no riparian vegetation due to human activities Negative: poor or no riparian vegetation cover is due to human activities Negative: high wood accumulation occurs in stagnant waters (native fish mostly are rheophilous) or after riparian tree removal
Table 3 Occurrence (%) of freshwater fish species in north-eastern Spain (N = 530 sites) showing their distribution and their food guild. Native translocated and alien fish species are referred to as introduced in main text. Fish species were assigned to the omnivorous guild if they consume invertebrates, plants, algae or detritus and to the piscivorous guild if they also prey upon small vertebrates according to regional fish atlases (de Sostoa et al., 1990; Doadrio et al., 2011). The trophic level estimates (TL) are based on Fishbase data (Froese & Pauly, 2018). Scientific name Achondrostoma arcasii Barbus haasi Cobitis calderoni Cobitis palludica Luciobarbus graellsii
Occurrence Distribution Atlases Diet Fishbase TL 4.72 Native Omnivorous 3.50 26.06 Native Omnivorous 3.35 2.45 Native Omnivorous 3.00 0.10 Native Omnivorous 3.00 t 28.11 Native Omnivorous 2.76
Parachondrostoma miegii 21.89 Nativet Omnivorous Anguilla anguilla 11.51 Native Piscivorous Barbatula quignardi 10.50 Native Omnivorous Barbus meridionalis 15.66 Native Omnivorous Cottus hispaniolensis 0.57 Native Piscivorous Gasterosteus aculeatus 0.94 Native Omnivorous Gobio lozanoi 19.06 Native Omnivorous Phoxinus bigerri 24.34 Native Omnivorous Salaria fluviatilis 4.53 Native Omnivorous Squalius laietanus 16.23 Native Piscivorous Salmo trutta 34.34 Nativet Piscivorous Alburnus alburnus 9.06 Alien Omnivorous Ameirus melas 0.75 Alien Piscivorous Barbatula barbatula 1.50 Alien Omnivorous Carassius auratus 1.51 Alien Omnivorous Cyprinus carpio 14.91 Alien Omnivorous Esox lucius 0.38 Alien Piscivorous Gambusia holbrooki 2.45 Alien Omnivorous Gobio occitaniae 19.06 Alien Omnivorous Lepomis gibbosus 4.15 Alien Omnivorous Micropterus salmoides 1.32 Alien Piscivorous Oncorhynchus mykiss 1.89 Alien Piscivorous Pseudorasbora parva 0.19 Alien Omnivorous Rutilus rutilus 0.38 Alien Omnivorous Sander lucioperca 0.38 Alien Piscivorous Scardinius erythrophthalmus 3.40 Alien Omnivorous Silurus glanis 1.89 Alien Piscivorous Gobio spp. 5.40 Alien Omnivorous Phoxinus spp. 6.60 Alien Omnivorous t indicates translocated native species in some catchments of northeastern Spain
2.79 3.62 3.45 2.93 3.20 3.38 3.17 3.27 3.50 3.61 3.80 2.70 3.75 3.45 2.86 3.05 4.40 3.26 3.17 3.27 4.42 3.53 3.10 2.87 4.04 2.89 4.12 3.17 3.27
Table 4 Spearman rank correlation coefficients among the trophic level (TL) measures of fish assemblages (maximum, mean and weighted-mean TL in biomass, and within assemblage TL range and variance (see methods for a full description). All pair-wise correlations are significant at P < 0.05 and the strongest correlations (|Spearman’s | > 0.7) are highlighted in bold. Maximum TL Mean TL TL range TL variance Maximum TL Mean TL 0.49 TL range 0.49 -0.39 TL variance 0.47 -0.32 0.91 Weighted-mean TL 0.36 -0.39 -0.39 0.84
Table 5 Final best general linear mixed models for each trophic level (TL) measure of fish assemblages after eliminating environmental variables that were not significant (𝜒 2 -tests at P < 0.05) or whose 95 % confidence intervals of standardized coefficients included 0s . All models include basin and year as random intercepts. R2c describes the proportion of variance explained by both fixed and random factors, and R2m shows the contribution of the fixed factors alone to the variation in TL measures using the function r.squaredGLMM. 𝜒2
P-value
33.39 7.07 17.62
