Wait and snap: eastern snapping turtles (Chelydra serpentina) prey on migratory fish at road-stream crossing culverts

Abstract

There is growing evidence that culverts at road-stream crossings can increase fish density by reducing stream width and fish movement rates, making these passageways ideal predator ambush locations. In this study, we used a combination of videography and δ¹³C stable isotope analyses to investigate predator–prey interactions at a road-stream crossing culvert. Eastern snapping turtles (Chelydra serpentina) were found to regularly reside within the culvert to ambush migratory river herring (Alosa spp.). Resident fish species displayed avoidance of the snapping turtles, resulting in zero attempted attacks on these fish. In contrast, river herring did not display avoidance and were attacked by a snapping turtle on 79% of approaches with a 15% capture rate. Stable isotope analyses identified an apparent shift in turtle diet to consumption of river herring in turtles from culvert sites that was not observed in individuals from non-culvert sites. These findings suggest that anthropogenic barriers like culverts that are designed to allow passage may create predation opportunities by serving as a bottleneck to resident and migrant fish movement.

1. Introduction

A variety of predator taxa have been observed capitalizing on the aggregation of fishes at migratory barriers. Most studies take place at dams and include fish predators [1–3], but other taxa (e.g. pinnipeds, birds) have also been documented depredating migratory fish [4–6]. Foraging theory predicts that the increased prey density and reduced prey movement capacity at barriers will create preferred foraging habitat [7], thus contributing to increased predation risk of anadromous fishes. Such constraints occur at both fishways and culverts as fish search for passage routes at migratory barriers. Furthermore, while data are lacking on predator recognition and avoidance by anadromous fishes, naive fishes may lack predator recognition or avoidance abilities present among fish that have been previously exposed to a predator [8–13]. Because anadromous fishes are naive visitors to freshwater, they may share these vulnerabilities.

The eastern snapping turtle (Chelydra serpentina) is an ambush predator of many taxa, including fish, and one of the most widely distributed turtles in North America [14]. They are apex predators with cascading effects in freshwater ecosystems [15–17]. Most individuals stay within narrow home ranges for extended periods of time [18]. Snapping turtles and other predators often use culverts to move between habitats, but data are lacking on specific movement and predation behaviours within these structures [19–22].

In this study, we used a combination of stable isotope analyses (SIA), videography and passive integrated transponder (PIT) telemetry to investigate predator–prey dynamics at a road-stream crossing culvert. The goals of this study were to determine if predators use a culvert as a location to ambush migrating anadromous fishes and how anadromous and resident prey fishes respond to the occupancy of predators at the culvert.

2. Material and methods

(a). Study sites

We sampled five sites in and near the Herring River (HR) system in Wellfleet, MA. The culvert site was the upstream-most culvert in the HR which was a 4.3 m long × 0.75 m diameter culvert with 20–30 cm of water depth. This culvert was located on a small stream connecting two kettle ponds (electronic supplementary material, figure SF1). Non-culvert sites were sub-divided into two groups: landlocked ponds which do not support anadromous fish (‘landlocked’); and ponds connected to the HR but not near a culvert (i.e. ‘non-culvert’).

(b). Videography

(i). Motion-triggered cameras

We used motion-triggered cameras to detect birds and mammals preying on river herring or resident fishes (e.g. Lepomis spp.; Perca flavescens; Micropterus salmoides). These cameras were deployed throughout the 2016–2017 herring migratory periods at the culvert site and two non-culvert sites in the HR system (i.e. the outlet of the downstream-most pond and in a river reach with high mortality based on PIT data from a companion study). Motion-triggered cameras were set to record 7 s videos with a 10 s delay between triggers.

(ii). Underwater videography

Action cameras were deployed daily opportunistically during daylight hours approximately 10 cm inside the entrance of the culvert via a magnetic tripod (electronic supplementary material, table ST1). The cameras were positioned to record continuous timestamped video through the interior of the culvert. All videos were reviewed and the following events were recorded when observed: turtle entry and departure from the culvert; fish species entering culvert, direction of movement (i.e. upstream or downstream), and whether they passed or failed to pass; turtle strike attempts; fish species attacked; capture success/failure; number of individual herring in a school; and how many individuals passed through the culvert or returned.

(iii). Fish positioning

To test whether fish recognize the occupancy of a turtle and actively avoid the attack cone [23], single frame images were extracted from underwater action camera footage to determine the cross-sectional position of the fish within the culvert. ImageJ 1.46r [24] was used to determine the pixel coordinates of five landmarks in the cross-sectional plane which were normalized to a standard length from the bottom-left to top-right corners, correcting for tilt and maintaining relative distances between landmarks (see Methods in the electronic supplementary material).

We tested for multivariate normality in the distribution of fish position within the culvert using the Henze–Zirkler test and univariate normality using the Anderson–Darling test [25] in R 3.6.1 [26] for each condition. This allowed us to test for differences in fish distribution by species and turtle occupancy.

(c). Stable isotope analyses

To test the hypothesis that culverts create a predation opportunity on herring, we used SIA to detect a diet shift towards greater proportions of marine derived nutrients (MDN) in culvert-dwelling turtles than non-culvert-dwelling and landlocked turtles. To do so, turtle blood fractions, river herring muscle tissue and aquatic vegetation δ¹³C values were compared. A pond connected to the HR system less than 0.2 km upstream from the culvert study site with no other culverts in the vicinity was used as a control pond for SIA. Two landlocked ponds less than 1.5 km from the culvert site were also sampled as negative controls for SIA. PIT telemetry at culvert sites was used to identify culvert usage by individual turtles for classification in SIA (see Methods in the electronic supplementary material).

(i). Biological sample collection

Snapping turtles (n = 11) were captured May–June 2016–2017 using baited hoop traps (electronic supplementary material, table ST2). Turtles were tagged subcutaneously with 12 mm half-duplex PIT tags on the ventral side of the tail anterior to the cloaca. A 4 ml blood sample was taken from the dorsal coccygeal vein using a 3.8 cm 21 G needle and 6 ml heparinized syringe [27]. Turtles were then released at the place of capture. Blood samples were placed on ice for transportation; plasma was separated from red blood cell (RBC) fraction by centrifugation at 1000g for 10 min. Blood samples from two individuals that did not contain sufficient volume to fractionate were left as whole blood samples (n = 1 landlocked; n = 1 non-culvert). Blood samples were frozen at −18°C until SIA were performed.

Migratory river herring were collected using seine and dip nets. Fish were collected heading in both upstream (April 2017) and downstream (May–June 2017) directions (i.e. likely pre- and post-spawn, respectively; see electronic supplementary material). Captured river herring were then sacrificed in 0.2 g l⁻¹ MS-222 buffered with NaHCO₃. A muscle sample was taken from approximately 1 cm ventral to the first dorsal spine using a scalpel blade and placed in a sealable plastic bag. Muscle samples were placed on ice prior to freezing at −18°C until SIA were performed.

It was assumed that the δ¹³C signature of the aquatic vegetation within each pond was representative of the δ¹³C of other potential freshwater-derived turtle diet items in comparison to the marine-derived δ¹³C signature of river herring muscle tissue [28]. Aquatic plants and algae were collected in June 2017 from all sites where snapping turtles were captured (electronic supplementary material, table ST3). Plant and algae specimens were dried in a drying oven at 60°C for 24 h then ground to a powder using mortar and pestle. Dried samples were stored at room temperature until SIA were performed.

All biological samples were processed and analysed for δ¹³C using a Thermo Delta V isotope ratio mass spectrometer interfaced to a NC2500 elemental analyser by Cornell Isotope Laboratory (COIL; Ithaca, NY). A pairwise Mann–Whitney–Wilcoxon test with Bonferroni correction was used to test the hypothesis that the MDN δ¹³C distribution from river herring muscle tissue was distinct from the FDN δ¹³C distribution of freshwater vegetation from landlocked and HR ponds. The Scheirer–Ray–Hare extension of the Kruskal–Wallis test [29] was used to test the hypothesis that the δ¹³C signature of culvert-dwelling turtles would more closely resemble the MDN δ¹³C signature of river herring, while non-culvert anadromous and landlocked snapping turtles would more closely resemble FDN δ¹³C. A two-way 2 × 3 factorial design followed by pairwise Mann–Whitney–Wilcoxon post hoc tests with Bonferroni correction were used to test for differences between blood fractions and turtle group (i.e. culvert-dwelling, non-culvert-dwelling, or landlocked). Non-parametric tests were used because the data violated assumptions of parametric tests.

3. Results

(a). Predation at culvert

Underwater videography documented 35.4 h of turtle culvert occupancies in 115 h of footage. Snapping turtles were observed attempting 118 attacks on river herring, 0 attacks on any other fish species and one attack on a relatively large animal at the surface (possibly a muskrat, Ondatra zibethicus) that could not be positively identified due to low light availability. A predation attempt was made on 79% of approaching schools of herring when turtles were available to strike (i.e. not currently consuming other prey or not with their head above water breathing) with a capture rate of 15%. The occupancy of a turtle in the culvert and an attempted attack on a school of herring resulted in fewer individuals within a school passing through the culvert (5–7%) than when no turtle was present (77–80%; table 1). Furthermore, trends in turtle occupancy within the culvert closely followed the number of herring observed per day throughout the season (table 2).

Table 1.

Number of herring passing the culvert or turning back with snapping turtles absent, present but not attacking, or present and attacking herring. Per cent passage range represents 95% confidence interval derived from the binomial distribution.

	pass	turn back	total	pass %
no turtle present	1716	473	2189	77–80
turtle present
not attacked	304	234	538	52–61
attacked	85	1418	1503	5–7
total	2105	2125	4230	48–51

Table 2.

The cumulative number of hours that snapping turtles were observed within and the number of herring observed approaching the culvert by date.

date	turtle hours	herring count
17 May 2016	2.0	562
18 May 2016	0.0	42
19 May 2016	0.0	572
21 May 2016	0.0	0
22 May 2016	5.0	1347
23 May 2016	1.2	0
24 May 2016	2.1	81
25 May 2016	4.2	754
26 May 2016	5.4	251
27 May 2016	4.3	318
28 May 2016	0.0	117
30 May 2016	0.4	0
31 May 2016	0.6	0
1 Jun 2016	0.0	0
2 Jun 2016	2.5	9
3 Jun 2016	0.5	122
4 Jun 2016	0.0	0
6 Jun 2016	0.6	0
7 Jun 2016	0.0	3
8 Jun 2016	0.0	0
9 Jun 2016	0.0	0
10 Jun 2016	2.7	98

Multiple species of potential mammalian predators (e.g. raccoon Procyon lotor, coyote Canis latrans) of river herring were detected at the culvert sites by motion-sensor cameras. Depredation of river herring by these mammalian species were observed at non-culvert sites but no evidence of attempted predation by mammalian or avian predators was observed at the culvert during the periods that motion-sensor cameras were deployed.

(b). Fish positioning

In the absence of snapping turtles, river herring displayed an approximately normal distribution around the centre of the culvert about the x-axis (Anderson–Darling test, p = 0.082) but a skewed distribution about the y-axis favouring the lower portion of the water column (Anderson–Darling test, p < 0.001; figure 1, top-left). In the presence of snapping turtles, the river herring cross-sectional distribution remained normally distributed about the x-axis (Anderson–Darling test, p = 0.311) but displayed an increase in variance relative to when no turtles were present. Additionally, the distribution about the y-axis continued to be skewed towards the bottom of the water column with turtles present as with no turtle present (Anderson–Darling test, p < 0.001; figure 1, top). In the absence of snapping turtles within the culvert, resident fishes displayed a multivariate normal distribution around the centre of the culvert cross-section (Henze–Zirkler test, p = 0.247). In contrast to river herring, when a turtle was present the distribution of resident fish positions shifted to a bimodal distribution favouring the two side walls of the culvert and heavily skewed towards the surface on the y-axis (figure 1, bottom).

Figure 1.

Cross sectional position of resident fishes (bottom) and migratory herring (top) within a culvert when a snapping turtle was occupying the culvert (right) or not (left) with marginal histograms for the x- and y-axes. Colours indicate whether a turtle, if present, attempted to attack the fish as it passed and the result of that attack. Scales of x- and y-axes are relative to total width and depth.

(c). Stable isotope analyses

Vegetation from landlocked and HR ponds did not differ in δ¹³C values (p = 0.61). Herring muscle tissue δ¹³C values were higher than freshwater vegetation δ¹³C values from both landlocked and HR ponds with nearly non-overlapping ranges (p < 0.0001). Furthermore, turtles that were detected residing in a culvert had elevated δ¹³C signatures when compared to both landlocked and non-culvert turtles from the same system (p = 0.086 and 0.171, respectively). Though these differences were not significant at α = 0.05, statistical power was low due to small sample sizes. Additionally, the ranges of observed δ¹³C values from culvert turtles did not overlap with the ranges of either of the other groups. Snapping turtles from the HR system that did not enter a culvert during the time of the study did not display elevated δ¹³C values in comparison to landlocked turtles (p = 1.00). Finally, there was some evidence that the δ¹³C of plasma was elevated in comparison to RBCs of culvert turtles (p = 0.20; figure 2). Again, these differences were not significant at α = 0.05, but the interquartile ranges did not overlap between the blood fractions.

Figure 2.

δ¹³C values for aquatic vegetation representing FDN, river herring muscle tissue representing MDN, and blood fractions of landlocked snapping turtles, turtles from HR system but did not enter culverts, and turtles that entered culverts. Two whole blood samples that did not have enough volume to be fractionated were grouped with plasma fractions.

4. Discussion

Dams and culverts are both ubiquitous features of rivers and streams around the world. While fish passage at these structures has been investigated for decades, the effects of predators using these structures are rarely considered and less frequently quantified. We found that even an easily passable culvert can create a novel predation opportunity for snapping turtles and alter the behaviour of resident fishes.

Predators often target migrants, perhaps in part because their life-history drives them to accept risk they might otherwise avoid [30]. Foraging theory predicts that both concentration and restricted movement of prey will create preferred foraging habitat for snapping turtles, prompting them to increase the contribution of herring to their diet [7]. The elevated δ¹³C signatures of culvert-dwelling individuals suggest that herring represent an important component of their diet; whereas, non-culvert-dwelling and landlock individuals did not exhibit elevated δ¹³C signatures. Divergence between plasma and RBC fractions in culvert-dwelling individuals might indicate a recent diet shift toward higher δ¹³C prey items as plasma has a shorter half-life than RBCs [31]. Owing to the challenges of collecting these specimens, our sample size was necessarily small, resulting in low statistical power. The apparent patterns from the stable isotope analyses, however, are suggestive of an important diet shift, which is supported by the PIT telemetry and videography data. These findings suggest that culverts may create a predation opportunity that likely would not exist in their absence. This has ecological consequences, potentially increasing the deposition of MDN into the system by increasing instream mortality and delaying emigration [32–34].

This study documented multiple individual turtles regularly using a culvert to ambush river herring. Migratory herring did not substantially alter approach behaviour in response (electronic supplementary material, video SV1). The increased variance in cross-sectional position within the culvert could have been expected of river herring passing an inanimate object similar in size to the turtles and did not resemble the response of resident fish avoiding a threat. By contrast, resident fish showed attack cone avoidance of the turtles, diverting their path to swim in the top corner of the culvert as they passed the turtle's head (electronic supplementary material, video SV2). This strategy was effective at preventing turtle strike attempts with no observed attacks of resident fish during this study. Even if river herring are unable to detect or recognize chemical alarm cues used by some resident species [35–37], herring did not appear to learn from repeated visual cues of turtles attacking and at times killing conspecifics within a school. Despite failing to enact pre-strike avoidance behaviours seen in resident fishes, herring reactions to a strike were largely effective at evading capture. However, these post-strike avoidance manoeuvres typically resulted in retreating rather than passing through the culvert, thus impeding movement between habitats.

Future fish passage studies should consider the possibility of predators creating additional delays and mortality at migratory barriers. Additionally, future studies could investigate predator recognition and avoidance abilities of seasonal migrants compared to resident fishes.

Ethics

All animal handling methods were reviewed and approved as ethical by Institutional Animal Care and Use Committees for the University of Massachusetts Amherst (no. 2015-002; 2016-0019; 2017-0003), National Park Service (NER_CACO_Alcott_Fish.SnappingTurtle_2017.A2) in all years that animals were handled. All animals were collected with approval from the Massachusetts Division of Fisheries and Wildlife (no. 154.16SCRA; 070.17SCRA) and the Division of Marine Fisheries.

Data accessibility

Data and code for this study are available at: https://doi.org/10.6084/m9.figshare.12074901 [38].

Authors' contributions

D.A. designed the experiment, carried out the fieldwork, performed data analyses and drafted the manuscript. M.L. assisted in fieldwork, data collection, and helped draft the manuscript. T.C.-S. helped design the study, coordinated the study, assisted in data analysis and critically revised the manuscript. All authors gave final approval for publication and agree to be held accountable for the work performed therein.

Competing interests

We declare we have no competing interests.

Funding

This work was funded by the U.S. Geological Survey, the graduate programme in Organismic and Evolutionary Biology at the University of Massachusetts Amherst, University of Massachusetts Amherst Dissertation Research Grant, and the Nickerson Fellowship of the Cape Cod National Seashore.