Introduction

Rivers are among the most threatened ecosystems worldwide (Grzybowski & Glińska-Lewczuk, 2019). In Europe, most rivers have been deeply modified by anthropogenic structures like dams and weirs, impacting negatively the ecological processes and aquatic species (Gough et al., 2018). Moreover, it is estimated that more than 1.2 million obstacles fragment European rivers, most of them low head barriers (< 2 m in height; Belletti et al., 2020). These structures hinder or block the movement of aquatic fauna, especially fish, which is a vital element of their life history (Strayer & Dudgeon, 2010; Lothian et al., 2020).

Often river obstacles create velocity barriers, zones of high velocity flows that may exceed the ability of some or all individuals to pass, e.g. ramped dams, gauging stations, culverts, weirs, and fishways (Sanz-Ronda et al., 2015; Bravo-Córdoba et al., 2021). Therefore, velocity barriers bring about spatio-temporal environmental variation that affects the gene flow and selection (Cooke & Hinch, 2013). This results in a variation in fish migratory behaviour and associated traits within and among individuals, reducing the phenotypic and genetic variance in populations (Tamario et al., 2019). In particular, for migratory fish, velocity barriers may cause sublethal or nonlethal nonnatural selective pressure, leading to exclusion of fish from spawning areas, which in turn impacts population level (Morita & Yamamoto, 2002; Maynard et al., 2017). Successful passage of fish through such barriers is governed by a combination of swimming capacity, behaviour, and motivation (Castro-Santos et al., 2013). Artificial selection caused by barriers could become underlying force generating this variation, by modifying their dispersal behaviour (Apgar et al., 2017; Branco et al., 2017). This variation may influence the phenotypic and genetic structure of fish populations, as much for upstream movements as for downstream ones (Silva et al., 2018).

The selection pressure on fish populations and the characterization of the evolutionary impacts of velocity barriers have barely been tackled from a phenotypic perspective and never by quantifying morphological characters. In fact, most of them have been about salmonids and take into account features like size using traditional morphometrics (Marcus, 1990), but not morphological traits by removing size effect (Maynard et al., 2017; Lothian et al., 2020). Knowing the relationship between shape (sensu Benson, 1975) and swimming capacity of fish is essential to understand the selective pressure that velocity barriers in rivers and streams can exert on these organisms. Langerhans (2008) recognized gaps to understand the effect of water flow in phenotypic diversity of fish and found that more empirical tests are required to study hypothesis about linking morphology, locomotor performance, and fitness in fish. Therefore, it is fair to draw attention to the fact that the loss of phenotypic diversity can be observed in a few decades (Haugen et al., 2008). Approaches based on traits such as morphological or physiological characters, aspects of behaviour, or genome-level features, which combine functional and evolutionary information, are used in ecology to describe populations and communities and their responses to natural or anthropogenic changes. Body shape is the main interlocutor of an organism to interact with the environment (Benson, 1975). Shape can affect the interactions with conspecifics, the responses to environmental forces, and how the organism integrates with the surrounding community (Wootton, 1990; Rincón et al., 2007). Swimming performance, as a specific body shape ability, is the main character determining fitness and survival in many fish species (Schaefer et al., 2011), and is obviously, determined by phylogenetic, interactions with habitat and other species, individual constraints, and it may strongly influence the ability of a fish to obtain food, find a mate, select suitable habitats, migrate, reduce competition, and avoid predation or unfavourable habitat conditions (Plaut, 2001; Ohlberger et al., 2006; Langerhans & Reznick, 2010; Alexandre et al., 2014). It is considered to represent an adaptation to the prevailing environmental conditions (Pakkasmaa & Piironen, 2001; Leavy & Bonner, 2009; Franssen, 2011; Franssen et al., 2013). In this kind of studies, traditional morphometric techniques have been widely used. However, these techniques are based on measures intensively dependent of size, in consequence, many patterns may remain hidden. For this purpose, quantitative traits, as provided by geometric morphometric, are more practical, objective, informative, and size independent than qualitative ones, and they usually show considerable variation within and among populations (Caillon et al., 2018; Walsh & Lynch, 2018).

Therefore, the aim of this paper is to relate morphological traits to swimming performance by using for the first time geometric morphometrics in a swimming flume to determine whether velocity barriers could exert a selection pressure on mature specimens of fish populations. The cyprinid Northern straight-mouth nase, Pseudochondrostoma duriense (Coelho, 1985), hereafter referred to as “nase”, is a potamodromous medium-bodied water column fish, which inhabits running waters (Doadrio, 2002). It is endemic to the Iberian Peninsula and categorized by International Union for Conservation of Nature (Crivelli, 2006) as vulnerable. The obtained results could be essential to detect the existence of fish morphology selection and movement limitations because of velocity barriers, to enhance conservation efforts aimed at improving fishways and culvert design and establish management strategies to recover the natural diversity of fish.

Methods

Facilities

The study was carried out in a swimming flume located in the field at Vadocondes hydroelectric power plant (Burgos, Spain), on the Duero River (ETRS89 41,63,664,663 N 3,57,280,384 W). The flume was made of polished concrete to reduce turbulence, flow friction, and boundary layer effect. It had a zero slope and consisted of three modules (Fig. 1): head tank, swimming flume, and staging area. A slide gate system regulated the discharge and set the flow velocity and depth in the flume. A broader description of the flume can be seen in Sanz-Ronda et al. (2015) and Ruiz-Legazpi et al. (2018).

Fig. 1
figure 1

Modified from Ruiz-Legazpi et al. (2018)

Open channel flume facility in Vadocondes showing head tank, swim speed flume, and staging area as well as flow control mechanism (spillway, gates, and depth loggers) and monitoring system (PIT antennas and cameras).

Full size image

During experimentation, flow monitoring was performed continuously: depth was monitored every minute (Orfeo-Mini probe, OTT Hydrome GmbH, Kempten, Germany; Accuracy: 0.002 m), flow rate every 30 min (Model 2100 speedometer, Swoffer Instruments Inc., Washington, USA; Accuracy: 0.01 m·s−1), and physicochemical parameters every hour (Multiparameter Water Quality Meter Hanna Instruments HI9829).

Fish sources

In this study, nases were engaged to carry out swimming performance experiments in an open-channel flume. We captured the specimens by electrofishing (Erreka Model; DC 300 V) and trapping in a fishway (closing water flow and collecting fish with hand nets) from the Arlanza River, a tributary of the Duero River. The captures were carried out in late autumn (28th and 29th of November 2013) when cyprinids move along the river searching winter habitats (Lucas et al., 2001).

Within 2 h of capture, fish were transported to the flume installations in 100 L aerated tanks and held in an acclimation pond at ambient water temperatures (5 °C). This pond consisted of two consecutive fish ladder pools, adjacent to the experimental flume. The pools measured 1.6 m in width, 1 m in depth, and 2.2 m in length each, and were supplied with water (50 L·s−1) directly from the Duero River.

All fish were anaesthetized with tricaine methanesulfonate 60 mg·L−1 (MS-222, Argent Chemicals, Redmond, WA), weighed (Body Mass BM ± 1 g), photographed, and surgically implanted (IP) with half-duplex PIT tags (TIRIS model RI-TRP-WRHP; Texas Instruments). These tags were 23 mm long and 3.85 mm in diameter and weighed 0.6 g, or < 1% of the body mass of the smallest tagged fish. This method has been shown to have negligible effects on growth, survival, and behaviour of many species (Brown et al., 1999; Ostrand et al., 2011; Castro-Santos & Vono, 2013). Acclimation period before trials was < 48 h, and neither fish died, nor erratic behaviour was detected. Fish were not fed during experiments, although the river water that supplied the experiment contained numerous food items that the fish could eat. Condition factor (CF = 100·BM·FL−3 in g·cm−3, where BM represents body mass and FL represents fork length) was calculated as a potential predictor of nase’s swimming performance in a swimming flume. In total, 32 nases, with FL ranging from 20.2 to 30 cm (Mean ± Standard Deviation: FL = 25.45 ± 2.75 cm; BM = 196.03 ± 67.59 g; CF = 1.29 ± 0.07 g·cm−3) participated in the experiments.

Image acquisition and shape analysis

A standardized protocol was used to acquire digital images for morphometric analyses. To avoid any damage while taking pictures, prior to photographing, all fish were anaesthetized as mentioned above. To avoid potential arching effects (Valentin et al., 2008), photographs were taken using a standardized method in which all fish were placed on their right side in a relaxed position on a flat surface and photographed from directly above using midline as a reference to linearity. A scale ruler was included in each photograph so that fork length (FL ± 0.01 cm) could be measured by using tpsDig v.2.31 (Rohlf, 2013a), and centroid size (CS), the square root of the summed square distances of each landmark from the centroid of the landmark configuration (Mean ± Standard Deviation: CS = 26.28 ± 3.02), was computed using tpsRelw version 1.70 (Rohlf, 2013b).

Morphometric analysis of body shape was carried out by means of geometric morphometrics (Bookstein, 1991; Zelditch et al., 2004). A total of 10 two-dimensional landmarks (Fig. 2) were obtained on digitized pictures by using as tpsDig v.2.31 (Rohlf, 2013a); of these, 9 landmarks had an unequivocal anatomic significance, and 1 homologous landmark (number 5) was geometrically determined but presented a clear anatomical undertone (Sánchez-González & Nicieza, 2017). Landmark configurations were superimposed, aligned, scaled, and rotated to a consensus shape. Thin-Plate Spline (TPS) analyses (Zelditch et al., 2004) were conducted by using tpsRelw version 1.70 (Rohlf, 2013b) to obtain 16 (2 k-4, where k is the number of landmarks) partial warp scores (14 uniform and 2 non-uniform components of shape), which are our geometric shape variables and used as a proxy of body size.

Fig. 2
figure 2

Collection of landmarks used for the morphometric analyses of Northern straight-mouth nase’s shape: (1) tip of upper jaw, (2) posterior supraoccipital notch, (3) anterior insertion of dorsal fin, (4) and (6) anterior junction of dorsal and ventral membrane from caudal fin, (5) intersection of lateral line and membrane of caudal fin, (7) and (8) posterior and anterior insertion of anal, (9) origin of pectoral fin, and (10) ventral insertion between operculum and the body outline

Full size image

Experiments

A volitional fish swimming study was undertaken in a swimming flume. The study was configured to allow fish to swim against three different nominal flow velocities: 1.5, 2.5, and 3.0 m·s−1 (Trial1.5 [low velocity], Trial2.5 [medium velocity], and Trial3.0 [high velocity], respectively). Data dependency and the learning and fatigue dependence issues have been dealt by randomly splitting fish into the two experimental groups, and potential differences have been evaluated. The sequence of trials was selected randomly to test if learning or fatigue affect fish behaviour. The experiment was conducted between 30th November and 2nd December 2013. Fish were randomly divided into two experimental groups of 16 fish each, (n1 = 16 and n2 = 16). Each group of 16 individuals was exposed at once to the full range of flow velocity treatments, so that fish of each experimental group participated in every three trials. Each trial lasted 4 h, starting at 11:00 h (local time). Trial sequence was established to assure resting periods of > 18 h between consecutive events and fish were hold in in two resting areas placed in the fishway pools used for acclimation and in the flume-staging area (Fig. 1).

Swimming capacity was estimated with video recording and telemetry systems using the method described in Sanz-Ronda et al. (2015) and Ruiz-Legazpi et al. (2018), and the following variables were calculated: maximum distance (Dmax, maximum distance in m travelled through the flume by fish and trial), fatigue time (Ft, time in s employed in reaching Dmax from the start of the flume), and swimming speed (Us = Uf + Ug in m·s−1; where Uf is the mean flow velocity through which fish actually swam, and Ug is the ratio between Dmax and time). We considered a single attempt for each fish per trial, registering the one in which Dmax was reached, since it maximizes swimming capacity. Furthermore, we recorded this information as the binary variable response (with response [for swimming fish, i.e. when fish attempted to ascend], registering a Dmax, Ft, and Us, versus without response [for non-swimming fish, when fish did not attempt]).

The presence of shifts between prolonged and sprint swimming modes (Beamish, 1978) was also tested by applying the moving-point regression approach (Castro-Santos, 2005). The prolonged swimming mode corresponds to the speed that fish can maintain for around 20 s and ends in fatigue (sensu Castro-Santos et al., 2013; Sanz-Ronda et al., 2015), and sprint swimming mode corresponds to the highest speed attainable by fish and can be maintained for only short periods.

Experimentation was planned and developed under the international principles established in the Directive 2010/63 UE relative to the protection of animals used for scientific purposes.

Statistical analyses

We conducted a Principal Component Analysis (PCA) on the 16 partial warp scores (using prcomp function in stats package; R Development Core Team, 2020). This PCA was performed to obtain a multidimensional and empirical morphospace with orthogonal and, in consequence, uncorrelated independent variables (Pielou, 1977; Everitt, 2007), since original geometric shape variables are neither biologically nor statistically independent (Rohlf & Slice, 1990; Zelditch et al., 2004). PCA was performed on the covariance matrix (partial warp scores; alpha = 0) (Dryden & Mardia, 1998; McGarigal et al., 2000) to confer more weight to variables that describe global aspects of shape (Zelditch et al., 2004).

The analyses of which results are presented have been performed within trials, not between trials, so there is no data dependence between the three trials. To check for morphological and size differences (BM, FL, CS, and CF) between fish with and without response at the three different velocity trials, and to ensure there are not differences between the two 16-inviduals groups, we conducted the t student test or the nonparametric alternatives, such as Welch test and Wilcoxon rank sum test for non-homogeneous and non-normal data, to compare results between the same velocity trials. Previously and in all cases, normality was tested with the Shapiro–Wilk test, and homogeneity of variances was evaluated using the Levene’s test. These statistical analyses were conducted using R software (R Development Core Team, 2020).

In the second step, we compared shapes of the swimming reaction groups for the three different velocity trials to test the hypothesis of morphological differences between fish with and without response. With this aim, we performed three Discriminant Analyses (DA) on partial warps to assess differences between groups (McLellan & Endler, 1998), to describe the linear combination of variables that maximally discriminate between groups. To discard potential problems associated with small sample sizes, cross-validation of the discriminant functions was performed by random resampling of the experimental individuals; a subsample containing 75% of the cases was used for analysis; and the remainder 25% constituted the holdout sample in a loop of 1,000 random samples. We conducted three multivariate analysis of variance (MANOVA) tests on first three components of PCA to analyse significant morphological differences between fish groups according to their swimming response at the three trials (McGarigal et al., 2000). DA and MANOVA assumptions of homogeneity of variances and multivariate normality (McGarigal et al., 2000) were evaluated by using the Levene test and Shapiro–Wilk’s multivariate test, respectively, from the R package mvnormtest (Jarek, 2012).

After an exploratory correlation matrix analysis using chart.Correlation function from library PerformanceAnalytics (Peterson & Carl, 2020) for the three trials and cor.test and lm functions, we implemented general linear model (GLM) by using the glm and stepAIC functions in the MASS package (Venables & Ripley, 2002), with Dmax, Us, Ft, and response at Trial1.5, Trial2.5, and Trial3.0 as dependent variables, and CS, CF, BM, FL, and shape components (Principal Components from PCA; i.e. PC1, PC2, PC3, etc.) as explanatory variables; moreover, residuals were also analysed. We compared the performance of these models using the second-order Akaike Information Criterion (AICc). To assess the relative strengths of each candidate model, we used ΔAICc and calculated AICc weights and evidence ratios (ER) (Akaike, 1974; Burnham & Anderson, 2002; Richards et al., 2011). In case, Δi values were greater than 10, models were considered uninformative, and when Δi were less than 2, models were assumed to be equivalent to the best model (Burnham et al., 2011; Richards et al., 2011).

Results

Swim speed trials and parameters are included in Table 1. Moreover, distance ascent and fatigue time did not differ between the same velocity trials (Dmax: Wilcoxon rank sum test, p > 0.27; Ft: Wilcoxon rank sum test, p > 0.41). In general, fish (93%) swam in a prolonged swimming mode for Trial1.5, and in a sprint mode for Trial2.5 and Trial3 (64% and 100%, respectively). Shift in swimming mode was detected at Us = 2.6 m·s−1.

Table 1 Swim speed trials and parameters: hydraulic, biometric, and environmental (Mean ± Standard Deviation)
Full size table

Size analysis

At Trial1.5, no significant differences were found between fish with and without swimming response (CS: t test, t30 = 1.5282, p = 0.137; BM: Wilcoxon test, W = 179, p = 0.05409; FL: t test, t30 = 1.639, p = 0.1117; CF: t test, t30 = 1.47, p = 0.152). However, it is important to point out that the BM difference was marginally significant: Fish with a lower body mass showed a higher response. At Trial2.5, swimming fish were significantly larger, longer, heavier, and had the highest CF (CS: t test, t30 = -2.5133, p = 0.01757; BM: Wilcoxon test, W = 62.5, p = 0.02649; FL: Wilcoxon test, W = 57, p = 0.01334; CF: Wilcoxon test, W = 62, p = 0.02357). Finally, at Trial3.0, non-significant differences in size were detected for BM and CF (BM: Welch test, t30 =  − 1.5668, p = 0.1311; CF: t test, t30 = -0.88133, p = 0.3851) but marginally different for CS and FL (CS: Welsch test, t30 =  − 1.874, p = 0.07269; FL: Welsch test, t30 =  − 1.7874, p = 0.08681).

Shape analysis

The first three principal components absorbed 60.89% (PC1: 34.23%; PC2: 15.03%; PC3: 11.63%) of the whole shape variability. Specifically, PC1 arranged body shape variation on the relative head position (landmarks 1, 2, and 10; Fig. 2). Fish with a higher relative head position scored positive values of PC1, whereas fish with lower positioned head scored negative.

PC2 explained body depth or elongation; thus, shapes or morphological variation are ordered according to dorsal muscular mass development versus elongated, streamlined, or fusiform shapes. Deeper specimens, with an overdeveloped dorsal area and with a higher muscular mass in the dorsofrontal area (landmark 3), scored positive values for PC2. On the opposite side of PC2, specimens with elongated, streamlined, or fusiform shapes were placed.

Finally, PC3 summarized variation in the relative position of tail and the shape of the caudal peduncle (landmarks 4, 5, and 6; Fig. 3). Shorter peduncle and a lower position of the caudal peduncle scored negative values for PC3, whereas fish with higher-positioned, longer, and thinner peduncles had positive scores.

Fig. 3
figure 3

Scores of the principal component analysis (PCA) performed on the variances–covariances matrix for the morphological variables. First principal component (PC1; 34.2%): Positive scores are related to a higher-positioned head; second principal component (PC2; 15.1%): Positive scores correspond to deep-bodied and more robust shapes; third principal component (PC3; 13.7%): negative scores mean lower positioned tails and shorter peduncles. Red dot represents fish without any response and blue dot represents ascending fish in Trial3.0. Blue arrow indicates the observer’s perspective. This graph has been plotted using function plot_ly from plotly library (Sievert, 2020), and it has been published at https://rpubs.com/Xurde/623783

Full size image

According to MANOVA results, we did not found differences in shape, either in Trial1.5 (Wilk’s Lambda = 0.94299, F3,28 = 0.56422, p = 0.6431) or in Trial2.5 (Wilk’s Lambda = 0.90668, F3,28 = 0.96067, p = 0.425), but we detected morphological differences in Trial3.0 (Wilk’s Lambda = 0.78099, F3,28 = 2.6172, p = 0.07062). These differences are mainly due to PC3 (F1,30 = 6.0906, p = 0.01951); fish with a higher relative position of their tails and a longer peduncle swam upstream during the experiments.

We conducted three DAs by calculating prior probabilities for response (with response and without response) in Trial1.5 (0.46875 and 0.53125, respectively), Trial2.5 (0.375 and 0.625, respectively), and Trial3.0 (0.21875 and 0.78125, respectively). DA functions (in a loop of 1,000 random samples) successfully classified 66.67% accuracy of fish for Trial1.5, 52.59% accuracy for Trial2.5, and 83.33% accuracy for Trial3.0.

Shape and size as swimming performance predictors

A potential positive relationship between PC3 and Us, Ft, and Dmax, was observed in Trial3.0 (Ft: S = 3418.7, ρ (Rho) = 0.3734, p = 0.03528; Dmax: S = 3176.6, ρ (Rho) = 0.4178, p = 0.01735; Us: S = 3051, ρ (Rho) = 0.4408, p = 0.01157) (Fig. 4) and marginally significant relationship between Dmax and shape in Trial1.5 (PC2: S = 302.31, ρ (Rho) = 0.4602, p = 0.0844) and Trial2.5 (PC1: S = 261.83, ρ (Rho) = 0.0845). However, even though preliminary analyses provided significant results, a detailed evaluation showed that neither PC1 nor PC2 were correlated to Dmax, Us, Ft in Trial1.5 and Trial2.5 (p > 0.05).

Fig. 4
figure 4

Linear relationship between the third principal component (PC3) and 95% confidence interval (in bluish grey), where specimens are arranged according to the relative position of their tails, shape of the caudal peduncle, and Swimming Speed (Us) in Trial3.0 (S = 3051, ρ (Rho) = 0.4408, p = 0.01157). Deformation grids (tpsSpline version 1.20 by Rohlf, 2004) show shape variation between the consensus and the individuals with the extreme PC3 values

Full size image

GLM’s R-squared ranged from 0.29 to 0.39 (Online Resource). The GLM models that best explained fish response, Us, Dmax, and Ft can be partially explained by size (CS, FL, and BM) and, to a lesser extent, shape (PC2) for Trial1.5. For Trial2.5; those traits (CS, FL and, to a lesser extent, PC2) become determinant for explaining response, Us, Dmax, and Ft. For Trial3.0, shape (PC2 and PC3) became an essential descriptor of swimming performance, and size became a partial descriptor (CS and FL; Table 2 and Online Resource) (See Fig. 5, as a summary picture).

Table 2 Comparison of Akaike Information Criterion (AICc) deltas and weights for binary general linear models testing the effects of centroid size (CS), condition factor (CF), fork length (FL), body mass (BM), and shape components (PC1–PC3) on response (swimming up or displaying no rection in Trial1.5, Trial2.5, and Trial3.0)
Full size table
Fig. 5
figure 5

Number (n = deep + elongated) and shape of fish with a response at the three water velocity trials (1.5, 2.0, and 3.0 m·s−1). This is a conceptual scheme where numbers and distances are kept, but shapes should not be considered as binary but as gradient. Dark blue represents elongated fish with thinner and longer peduncle, and light blue represents deep-bodied fish with wider and shorter peduncles

Full size image

Discussion

Predicting the viability of organisms and populations to rapid environmental changes, modifications, habitat destruction, and fragmentation is a central issue for the conservation of populations (Fagan, 2002; Hanski, 2011). Velocity barriers constitute important alterations of lotic ecosystems and cause obstructions to fish movements, such as dispersal processes and fish migration (Castro-Santos et al., 2013). Global shape aspects, such as elongation and/or streamlining, and other morphological traits, like relative tail and fin positions, mediate swimming performance and, in consequence, determine fish capability to swim up through those barriers (Beamish, 1978). In this study, we analysed those relationships and their possible consequences on nase population due to selective pressure in velocity barriers.

Despite the variance absorbed by the first 3 components (60.89%), our results suggest the presence of morphological differences, vague in Trial1.5, and conspicuous in Trial3.0. Thus, these results reveal that at the lowest tested velocities (1.5 m·s−1 and prolonged swimming mode), being large, deep-bodied, and, apparently, strong, is optimal for the individuals of the studied nase population (Fig. 5, Table 2, and Online Resource). Under higher water velocities (2.5 and 3 m·s−1 and sprint swimming mode), a change is observed and, being larger (from a lateral point of view) becomes optimal. At higher flow velocities, fusiform, elongated, and streamlined body shapes with a higher relative position of the tail and a narrower caudal peduncle are more efficient (Figs. 4, 5, Table 2, and Online Resource). This is because fish face high velocities by reducing drag forces and increasing thrust propulsion, reducing the energy expenditure required to swim up (Müller et al., 2000; Langerhans et al., 2003; Alexandre et al., 2014; Rubio-Gracia et al., 2020). Traditionally, fusiform, streamlined, and elongated shapes are linked to steady swimming in respirometer tests (Yan et al., 2013), where they develop swimming velocities close to the lowest values of prolonged speeds (Sanz-Ronda et al., 2015; Ruiz-Legazpi et al., 2018). In our case, fish did not perform prolonged swimming in Trial2.5 and Trial3.0, but sprint; under those conditions (Uf > 2.5 m·s−1), fusiform individuals could display a more efficient swimming performance than deep-bodied fish. Multimodel inference also suggested, to some extent that BM had relatively high importance at lower tested velocities, but shape gained relative weight at higher velocities. Furthermore, when flow velocity is lower, neither size nor shape is as strong a determinant as elongated shape is under high water velocity, and size cannot be considered as a plausible predictor of swimming at high flow velocities. Our results, despite multimodel constraints, show that larger deep-bodied individuals display better swimming performance at the lowest tested velocity, whereas the higher velocities could select individuals with a higher relative position of the tail, longer peduncle, and elongated shapes. This last statement is congruent with the results found by Alexandre et al. (2014) and Haas et al. (2015) in two different cyprinids species.

In general, morphometric variability reflects the adaptation to local conditions (Sánchez-González & Nicieza, 2017) and specifically to water velocity in salmonids streams (Beacham & Withler, 1985). Related to cyprinids, populations of blacktail shiner (Cyprinella venusta), living in habitats with higher mean annual run-off, exhibited, among others, characteristics, a more slender body, and caudal peduncle (Haas et al., 2015). Alexandre et al. (2014) experimented in a swimming tunnel with Iberian barbel (Luciobarbus bocagei), also a rheophilic cyprinid like nase, using individuals inhabiting both temporary river and flowing river. They found a lower critical swimming speed in temporary river fish and, as an interesting finding related to morphological traits that a combination of more fusiform body shape and a narrower caudal peduncle is better suitable for slow movement and manoeuvring than for swimming in flowing areas.

In contrast, at 2.5 and 3.0 m·s−1 flow velocities, when swimming mode changes from prolonged to sprint, elongated or fusiform shapes with thinner peduncles and a higher relative position of tails become more efficient to swim up (Figs. 4, 5, Table 2, and Online Resource). This pattern could be explained by the fact that deep-bodied shapes are related to burst swimming, territorial and hierarchical behaviour, and dominance (Monet et al., 2006; Rouleau et al., 2009; Sánchez‐González & Nicieza, 2021). In consequence, overdeveloped dorsal muscular mass, robust bodies, and caudal peduncles attain faster burst velocities (Cadrin, 2000; Pakkasmaa & Piironen, 2001), which, apparently, profit from a lower flow velocity, where burst swimming mode, short but explosive, could be more advantageous.

This study discards CF as a useful predictor of nase’s swimming performance in concordance with previous works (Reidy et al., 2000; Romão et al., 2012; Sanz-Ronda et al., 2019). However, Alexandre et al. (2016) obtained the opposite result, i.e. a significant positive linear relationship between critical swimming and CF. Since most studies used traditional analysis of morphometry, mainly with salmonids, most of the explained patterns remained unobserved so using geometric morphometric becomes essential to detect these patterns (Pakkasmaa et al., 1998; Cramon-Taubadel et al., 2005; Fraser et al., 2007). Moreover, geometric morphometric analyses are empirical, which means they are more robust than CF for hydrodynamics purposes, as their results depend on the analysed sample. Therefore, comparing results is difficult and can never be performed directly. Since the last decade, the use of geometric morphometric techniques has increased significantly, providing a plethora of contrastable outcomes.

Regarding the response variables Dmax, Us, and Ft, our study showed that they were mediated by shape and size with different patterns at different flow velocities. However, these results might be interpreted with caution due to the R-squared values, and we cannot discard the fact that other relationships were confirmed, or not, in a higher sampling size. The obtained outcomes are equally interesting since they can have important consequences for traditional management practices to deal with habitat fragmentation, dams, weirs, and other hydromorphological alterations that determine dispersal processes and connectivity. In fact, these results give reasons to doubt about the effectiveness and the effects of velocity barriers. Some of these velocity barriers permit upstream migration and dispersal processes, but then the following questions arise: which fish swim upstream? which sizes? which shapes? In this context, an important question to be explored in the future is whether these obstacles exert a selection pressure on the fish populations. Our results seem to point out that these velocity barriers, even when they are apparently working properly, can exert a selection pressure on mature specimens of this cyprinid species. In fact, we showed that ascending behaviour can be predicted by individual size and shape, which identifies correlated morphological and behavioural traits that support the pace-of-life syndrome (POLS) hypothesis. Then, these results and the found selection processes could imply substantial effects for the persistence of migratory and rheophilic fish species constrained to disperse within dendritic riverine networks with mainly unidirectional gene flow (Fagan, 2002). In consequence, velocity barriers assessment and fishways design might consider these results, by identifying the most suitable flow velocities during the nase’s migration, and, if it is required, by removing those velocity barriers of which selection pressure acts against migratory processes (Goerig et al., 2020).

To the best of our knowledge, this is one of the first times, geometric morphometrics has been used to evaluate the relationship between fish shape and swimming performance in velocity barriers, and it has been evidenced as a reliable tool for this kind of analysis, showing that at lower velocities, large individuals swim up more efficiently and that when water velocity increases, due to drag forces, streamlined body shapes with a higher relative position of the tail and a narrower caudal peduncle are more efficient. These results agree with our prediction that velocity barriers could represent a selection for certain body shapes of the individuals in a fish population, as previous studies have already confirmed (Wilkes et al., 2019; Goerig et al., 2020). This study contributes by explaining how this selection process take place at morphological level, thanks to geometrics morphometric. It could be interesting to figure out if velocity barriers by selecting shapes are causing behavioural alterations and, morphological selection, and selective fragmentation of which consequences are having an effect this selection is being produced at the genetic level, to verify the existence of a selection pressure on a migratory species within the context of a dendritic ecosystem causing an effective fragmentation of fish populations, as Goerig et al. (2020) already pointed out. In conclusion, obtained results confirm individual size and body shape as predictors of swimming performance for nase adults, with more potential consequences on fish population due to selection pressures associated with velocity barriers than might be considered by stakeholders and decision makers dealing with restoration of river connectivity and improvement of design specifications for culverts and fishways.