Retention and expulsion of oviduct-inserted transmitters
We tested the assumptions that oviduct-inserted transmitters are retained until spawning and then expelled with the eggs (Assumptions 1 and 2) in a laboratory tagging study. Forty-eight hatchery-raised, female lake trout (Seneca strain) from Sullivan Creek National Fish Hatchery (Brimley, MI, USA) were transferred to Hammond Bay Biological Station (Millersburg, MI, USA) on 6 August 2013. The trout were housed in a 6,800 L rectangular (6.7 × 1.7 × 0.9 m) holding tank, supplied with chilled (water temperature between 7°C and 13°C) water from Lake Huron. Trout were offered a diet of 13 mm brood stock fish pellets until mid-September, when the trout ceased feeding.
Trout were tagged with oviduct transmitters on 17 September 2013. To test the hypothesis that insertion of transmitters through the oviduct does not affect retention and expulsion of transmitters, relative to those that have been surgically implanted, we inserted transmitters via the oviduct in half the fish (n = 24; hereafter referred to as ‘inserted’) and surgically implanted transmitters just anterior to the genital pore in the rest (n = 24; hereafter referred to as ‘implanted’). Half of each tagging group received V6 dummy transmitters (VEMCO, Nova Scotia, Canada; 16.5 × 6 mm, 1 g in air) and the remaining trout received V7 dummy transmitters (VEMCO; 20 × 7 mm, 1.6 g in air). Dummy transmitters contained a passive integrated transponder (PIT) tag so that individual fish could be identified. All trout were anesthetized in 40 L of clove oil solution (0.8 mL/L of 1:9 clove oil:ethanol solution) and then placed dorsal side down on a v-board prior to undergoing the tagging procedure. Gills were perfused with aerated lake water throughout the procedures. In trout receiving the transmitter via oviduct-insertion, transmitters were gently inserted through the oviduct and pushed approximately 5 cm deep using a 10 cm piece of sterilized 6.5 mm diameter Tygon tubing. The procedure was performed by a single researcher and took, on average, 20 to 30 s. In trout receiving surgically implanted transmitters, a small incision (1.5 to 2.0 cm) was made slightly off the ventral midline, approximately 5 cm anterior to the genital pore. After the transmitter was inserted into the body cavity, the incision was closed using two simple, interrupted sutures (Ethicon, Inc.; 3-0 polydioxanone monofilament). Surgeries were performed by a single surgeon and took, on average, 2 to 3 min. Following tagging, trout were returned to the holding tank to recover from anesthesia.
The holding tank was searched daily, September to November, for expelled oviduct transmitters, as well as expelled loose eggs that would indicate trout were ovulating. During daily searches, the fish were assessed for signs of odd behavior (for example, failure to respond to the researcher’s movements or inability to maintain equilibrium) and signs of poor health (for example, dermal discoloration or presence of fungus). If a transmitter was found, the associated PIT tag number, date, and time were recorded. Lake trout did not spawn naturally under laboratory conditions, so manual stripping was used as a surrogate to determine if transmitters would be expelled with eggs. Our assumption was that, while not identical to natural spawning, the manual stripping procedure was representative of the processes that occur naturally during egg expulsion. Manual stripping was conducted on 25 November 2013, to coincide with the end of the Lake Huron lake trout spawning period. The ovulating trout were removed from the holding tank one at a time and anesthetized as before to prevent excessive stress. Anesthetized trout were held dorsal side up with heads and tails pointed slightly upward, as occurs during natural egg deposition [33, 34]. Their bellies were then massaged from anterior to posterior until the oviduct tag was expelled, or the trout stopped expelling eggs.
Differences in retention of oviduct transmitters were compared using time-to-event analysis (R package ‘survival’; α = 0.05), which allowed us to not only compare the proportion of individuals that retained the transmitter, but also the temporal pattern of premature expulsion. The proportion of lake trout that expelled the oviduct transmitter during manual stripping was compared among groups using Fisher’s Exact Test (R package ‘stats’; α = 0.05), and the number of stripping motions required to expel the transmitters was compared using a t-test (R package ‘stats’; α = 0.05).
Estimating time and location of spawning using oviduct transmitters
Inferences from a single oviduct transmitter in each fish
The ability of the VEMCO Positioning System (VPS) to accurately position expelled oviduct acoustic transmitters in substrate (Assumption 3) was investigated in a field study at known lake trout spawning reefs in the Drummond Island Lake Trout Refuge, northern Lake Huron, North America (Figure 1). Three small VPS arrays (approximately 100 × 100 m) were constructed on three separate near-shore (0.6 to 1.3 km from shoreline) spawning sites. The three sites ranged in depth from approximately 1.5 to 7 m and consisted of rocky substrate. The approximate 100 m spacing of receivers was based on recommendations from the manufacturer (VEMCO) on the maximum detection range of the less powerful 180 kHz transmitters. Each array consisted of four receiver locations and a centrally-located synchronization transmitter site, which held transmitters used to synchronize the clocks on receivers. Because the V6 and V7 acoustic transmitters transmit at different frequencies, we installed two acoustic receivers (VR2W-69 kHz and VR2W-180 kHz) at each receiver location on a single mooring. The upper receiver was positioned upside down so that the top-mounted hydrophones on the two receivers were located as closely as possible to one another.
Five test transmitter locations were chosen for each VPS array. The first was at the center of the array. The remaining four were scattered throughout the array so that they were located at random distances from the center of the array. Three transmitters of each type (V6 and V7) were deployed at each site. Transmitters were programmed to transmit with a fixed delay of 1 min and initialization of transmitters was staggered by 20 s to ensure that no signal code collisions would occur between transmitters. One of each transmitter type was suspended 1 m above the substrate to act as a control, representing a transmitter that was still within the trout. The two remaining transmitters of each type were placed in the cobble substrate (5 to 20 cm in diameter) by scuba divers to represent transmitters that had been expelled during spawning. Divers ensured that the transmitters were placed in the space between rocks, as this is most likely how the negatively-buoyant transmitters would settle when expelled naturally during a spawning event. Test transmitters were left in place for 20 min (allowing for a total of 20 transmissions from each transmitter) before being retrieved and moved to the next test transmitter site. At the end of the study, receivers were retrieved and downloaded, and log files from the receivers were sent to VEMCO for postprocessing using their proprietary hyperbolic positioning algorithms [42].
Probabilities of detecting and positioning control and oviduct transmitters were compared across transmitter type using linear mixed models (R package ‘lme4’; α = 0.05). Treatment (control transmitter or oviduct transmitter) and transmitter type (V6 or V7) were fixed effects and transmitter identification code and array number were random effects. Variables related to poor detectability of transmitters in the substrate were also investigated using a linear mixed model (R package ‘lme4’; α = 0.05). Distance between transmitter and receiver, total area of obstruction between transmitter and receiver, tag type (V6 or V7), and treatment (control transmitter or oviduct transmitter) were fixed effects; transmitter identification code and receiver serial number were random effects. Area of obstruction, the total cross-sectional area of substrate located at higher altitude than the line of sight between transmitter and receiver, was calculated using high-resolution (1 m horizontal, 10 cm vertical) multibeam sonar bathymetry (Seabat 7101 system, Teledyne RESON Inc.).
Pairing oviduct transmitters with larger transmitters to track fish after spawning
The feasibility of using transmitter separation in a paired-transmitter design to accurately estimate time and location of oviduct transmitter expulsion (Assumption 3) was tested using random walk models to simulate a variety of possible lake trout spawning behaviors [43]. The paired-transmitter design called for each simulated trout to be implanted with two transmitters; a small oviduct transmitter that would be expelled with the eggs during spawning, and a larger fish transmitter that would remain within the fish after spawning. Time and location at which the two transmitters began to behave differently (point of transmitter separation) was used to estimate time and location of spawning.
Each random walk model (referred to hereafter as ‘trout track’) consisted of 43,200 steps, each 1 m in length (total track = 43.2 km). Steps of the trout track were straight lines, with headings sampled from a random distribution with mean turn angle equal to 0 and standard deviation σ. For simplicity, the models assumed that trout remained within the VPS array for the entire time series. However, in practice, the only requirements are that the oviduct transmitter be expelled in the array and then the trout remain within the array long enough to determine that the two transmitters had separated. Numerous possible spawning behaviors were simulated to determine how different behaviors, spanning a range of species, could affect ability to accurately estimate time and location of spawning using a paired-transmitter design. To determine how degree of spawning site residency affected our ability to accurately estimate the time and location of transmitter separation (that is, spawning location) four different values were used for σ (5, 20, 45, and 90; hereafter referred to as low, medium, medium-high, and high residency groups, respectively). Distance travelled from spawning site was inversely related to σ (Figure 3A and C). For example, simulated trout with σ = 5 (low residency group) travelled, on average, 17 times further from the spawning site than simulated lake trout with σ = 90 (high residency group). Mean (±SE) distance between all positions for an individual and its actual spawning site was 1,976 ± 536 m, 493 ± 141 m, 223 ± 63 m, and 116 ± 33 m for simulated tracks with σ = 5, 20, 45, and 90, respectively. We also varied mean swim speed of individual lake trout between 0.25 m · s-1 and 1.00 m · s-1 (0.25, 0.50, 0.75, or 1.00 m · s-1) to cover a range of realistic swimming speeds.
Transmitter positions, such as those that would be triangulated using the VPS system, were overlaid on each trout track. The result was two transmitter tracks, one for the fish transmitter and another for oviduct transmitter (Figure 3B and D). Transmitters were programmed to transmit with mean transmission delay of 60, 180, 300, or 420 s. The exact delay of each transmission was sampled from a uniform distribution with bounds delay - delay/2 and delay + delay/2. These delays were assumed to represent effective transmitter delays, which incorporated not only the nominal delay of the transmitters as programmed by the manufacturer, but also any variation due to environmental variables like wave noise and signal code collisions with other nearby transmitters. For the sake of realism, we incorporated positioning error into each of the transmitter positions by sampling from a bivariate normal distribution with mean equal to the true location (x and y coordinates) of the trout at time t (based on trout track) and a standard deviation of 15 m. At the halfway point of each trout track, the oviduct transmitter was expelled into the substrate as a simulated spawning event. Based on our field test of the ability of the VPS system to position transmitters expelled into the substrate, we assumed the probability of positioning the oviduct transmitter in the substrate from this point forward was reduced to 10% (that is, only every 10th transmission could be positioned).
In total, 100 trout tracks were simulated for each combination of σ, swim speed and transmission delay (6,400 trout tracks in total). We used the R statistical package ‘changepoint’ to estimate time and location of transmitter separation for each simulated trout track [29]. Changepoint analysis, an emerging tool in movement ecology [44, 45], uses maximum likelihood methods to determine points in a time series where the mean, variance, or mean and variance of a response variable change. In our analyses, we tested for a single changepoint, and visual inspection of results indicated that the ‘changepoint’ function ‘cpt.meanvar’ (that is, changepoint of mean and variance; number of changepoints (Q) = 1, test.stat = ‘Normal’) was best able to identify when the two transmitters separated.
We tested two separate methods for estimating time and location of transmitter separation. The first method was to test for a change in the distance between the two transmitters. For each time in the fish transmitter track, we used linear interpolation to estimate the location of the oviduct transmitter at that time. The linear distance between the fish transmitter and oviduct transmitter was calculated and changepoint analysis was used on those values to determine when in the time series the mean and variance of the distance between the two transmitters changed. The second method took advantage of the reduction in positioning probability of expelled oviduct transmitters that we observed in our field trial. For each time in the fish transmitter track, we calculated the ratio of the time since the previous position for the oviduct transmitter and the time since the previous position for the fish transmitter (that is, time since last oviduct transmitter position / time since last fish transmitter position). Changepoint analysis was then used to determine when in the time series the mean and variance of that ratio changed. Change in ratio of positioning probability was used rather than simply testing for a change in the positioning probability of the oviduct transmitter itself because positioning probability can vary naturally due to environmental variability. The ratio method can accommodate this variation because any environmental variation should affect both transmitters in a similar manner. Both methods outlined above can return estimates of time and location of transmitter separation if the oviduct transmitter continues to be positioned (even at a lower rate). However, only the latter method (change in relative positioning probability) can return an estimate of time and location of transmitter separation if the oviduct transmitter ceases to be positioned after expulsion.
The output of the changepoint analysis was an estimate of the location in the fish transmitter track time series where the two transmitters separated (that is, began to behave differently). Two metrics were calculated to evaluate the ability of the paired-transmitter design to accurately estimate the time and location of oviduct expulsion: (1) difference in time (hereafter ‘time error’) between estimated transmitter separation and true time of spawning (that is, the midway point of the trout track), and (2) distance between estimated location of transmitter separation and true location of spawning (hereafter ‘location error’). A series of general linear models (R package ‘stats’; α = 0.05) were used to determine the effects of σ, transmission delay and swim speed on time and location error of estimated transmitter separation. Effects were modeled separately using both the change in distance between transmitter method and change in relative positioning probability method. Because the effect of spawning site residency on accuracy of transmitter separation estimates was of interest, the continuous variable mean distance travelled from spawning site (that is, mean distance between all positions in a given trout track and its true spawning location) replaced σ in the models. Interactions between distance travelled from spawning site and transmission delay, distance travelled from spawning site and swim speed, and transmission delay and swim speed were also included in the models.