For the purpose of the study, and in order to avoid possible differences in feeding behaviour among populations due to differences in physical habitat characteristics, samples were collected in wadeable riffle sections with similar environmental characteristics. In total, eight neighbouring rivers of Galicia (NW Spain) were sampled (Figure 1) during June 2003 (Rois, Santa Lucía, Sar and Traba rivers) and September 2007 (Anllóns, Furelos, Lengüelle and Tambre rivers). Prior to electrofishing, samples of potential prey items (benthic and drifting invertebrates) were collected to study prey availability in the environment. Benthic invertebrates were collected from riffles using a 0.1-m2 Surber sampler (n = 3), and a Brundin net (250-μm mesh size, 1 m long, 30-cm mouth diameter) was used to collect drifting invertebrates. Drift nets were set at sunrise (8:00 a.m.) and retrieved after at least 2.5 h (ranging between 179 and 200 min). After collection, we fixed samples using 4% formalin and stored them for later processing. Information on prey availability is only provided for rivers surveyed in 2007.
Trout were collected using pulsed D.C. backpack electrofishing equipment (ELT60II, Hans Grassl GmbH, Schönau am Königssee, Germany). Fishes were killed immediately with an overdose of anaesthetic (benzocaine) and transported in cool boxes (approximately 4°C) to the laboratory, where they were frozen at −30°C until processing. In the laboratory, fishes were measured for fork length (FL; nearest 1 mm) and weighed (nearest 0.01 g), and the stomachs were removed. Estimates of fish age were made by scale examination and by using Petersen's length-frequency method (Bagenal and Tesch 1978). Age-4+ individuals were not included in the diet analysis because only one specimen was captured in the River Furelos. No empty stomachs were found, and the stomach fullness index (f) was calculated as f = (Ws/W) × 100, where Ws is the total stomach content mass (g) and W is the fish mass (g).
Potential (benthic and drifting invertebrates) and actual prey items were counted and measured (total length) with a digital micrometer (0.01-mm resolution, Mitutoyo Absolute, Mitutoyo Corporation, Takatsu-ku, Japan). The number of fragmented or partially digested invertebrates was estimated by counting body parts resistant to digestion. In those cases, prey length was estimated from the width of the cephalic capsule (see Rincón and Lobón-Cerviá 1999), which was normally the best-preserved part.
The similarity between the size distributions of potential prey in the environment (benthos and drift) and those consumed by trout was assessed using the Bray-Curtis similarity index. The data were first transformed by Y = log(x + 1), and the similarity index was calculated using the PRIMER statistical package version 5.0 (Clarke and Gorley 2001). In the present study, in order to assess whether prey size selection is dependent upon the size frequency distribution of available prey, we clustered all trout regardless of age at each sampling site. Additionally, to explore ontogenetic shifts, the similarity matrix was calculated separately for each age class.
Finally, with the aim of exploring the possibility of a nonlinear relationship between prey size and the other analysed variables (prey numbers, stomach fullness and fish size), the curve estimation procedure was used using pooled data, which compared eight different models (linear, logarithmic, inverse, quadratic, exponential, power, compound and S-curve). The model with the highest adjusted Pearson's rank correlations coefficient (R) was chosen. The data were not normally distributed, so in order to analyse differences among age classes in the studied variables (prey size, prey numbers and stomach fullness), nonparametric analyses (Mann-Whitney and Kruskal-Wallis tests) were used. The Mann-Whitney U-test was used to compare differences between two independent groups because in some rivers (Rois, Santa Lucía, Sar, Traba and Tambre rivers) only two age classes were analysed (1+ and 2+). Kruskal-Wallis test was used to detect differences among four groups (0+, 1+, 2+ and 3+) in the other rivers (Anllóns, Furelos and Lengüelle). Statistical analyses were conducted using the programme IBM SPSS Statistics 20 software (IBM Corporation, Armonk, NY, USA). All of these tests were considered statistically significant at P level < 0.05.