Migratory orientation in a narrow avian hybrid zone

Study site and orientation trials

Between August 15th and September 12th of 2011 we captured migratory yellow-rumped warblers (n = 181) near Kananaskis, Alberta, a site at the center of the Audubon’s/myrtle warbler hybrid zone (Brelsford & Irwin, 2009). Our sampling was concentrated in two areas: along the southern edge of Barrier Lake (51.0235°N, 115.0608°W) and within the hamlet of Lac Des Arc (51.0536°N, 115.1589°W; Fig. 1C). We set up mist-nets before dawn and used passive netting along with song playback (using a tape of a variety of yellow-rumped warblers as well as begging calls from nestlings; previous experience showed that similar recordings could be used to attract all phenotypes in the hybrid zone) to increase our likelihood of catching individuals of our target species. These sites, timing, and capture methodology were chosen to capture birds that were beginning their migratory movements, but it is possible that some of the individuals sampled were close to their breeding territories or natal areas. Immediately after capturing each individual we took morphometric measurements (bill, tarsus, wing and tail length; according to Pyle, 1997), photographs, a blood sample (10–40 µL), and applied a unique aluminum USFWS leg band. Birds were then transported approximately three kilometers from the capture site to the location where the orientation trials were performed (51.0286°N, 115.0242°W). This site is a large, recently clear-cut field (∼400 m²; <10 years old) near the Kananaskis Biogeoscience Institute and has clear views in all directions.

Each individual was placed individually into one of 12 outdoor holding/orientation cages (Fig. 2) and given water and small mealworms throughout the day. The cages were a modified design based on Fitzgerald & Taylor (2008), who used similar cages to study orientation in yellow-rumped warblers (Fig. 2). The cages were leveled, oriented with a compass, and spaced approximately 3–5 m apart. The cage frames were made out of pine boards, with the top and bottom of the cages made from composite plywood. The perch was made with a 9″ plastic embroidery hoop, held up by four 7/16″ dowels, and placed horizontally such that birds could perch on the hoop (Fig. 1C). Since research suggests that birds may use magnetic cues for navigation, the wood pieces were joined with non-magnetic brass screws and stainless steel staples were used to affix screen mesh to the top and sides of the cage (Alerstam, 1993; Muheim, Moore & Phillips, 2006). During the day the sides of the cages were covered with a blue, opaque tarp for sun shelter, with part of the tarp rolled up to provide sufficient airflow (Fig. 2A). Using the same cages to both hold the individuals during the day and run the orientation trials in the evening has a number advantages over traditional use of Emlen funnels (Emlen, 1970). Most importantly, this likely reduces the stress inherent in holding individuals indoors in artificial environments and moving individuals between holding and assay cages shortly before the trials, as is necessary with funnels (Emlen, 1970). We were careful in assessing the welfare of the birds: if individuals did not immediately fly to the perch, showed any signs of stress (e.g., panting), or were not eating properly they were released during the day (less than 15% of birds captured).

Before we started each orientation trial, we removed the food and water dishes and closed each tarp around the sides of each cage, allowing each individual to see only out of the top of the cage, which had a full view of the sky. Through a hole in the bottom of each cage we attached a D-Link Wireless Network Camera (DCS-932L) pointing directly up, with the top of the camera oriented northwards. Unlike cameras that point down (e.g., Muheim et al., 2014), these upward-facing cameras do not obstruct a bird’s view of the sky. The cameras we used have an infrared LED light for illumination during low light conditions. This light was applied consistently across all of the trials, and we are aware of no evidence that birds can see infrared light. The cameras were set such that the video image was recorded with the right side of the video representing the west side of the cage and the left side of the video representing the east side of the cage. We recorded 320p x 240p 30 frame-per-second video from each cage simultaneously using the D-ViewCam software on a PC laptop via a D-Link router. We began recording approximately one hour before sunset and ran the trials until approximately 30 min past sunset. Each evening we ran the trials until the last individual stopped moving, after which we released all of the individuals. We tested each bird once and recorded the behaviour of one to 10 individuals each evening, averaging six birds per evening, with a total of 124 individuals for which we collected complete orientation data. All animal care and experimentation was conducted according to the University of British Columbia protocol Nos. A11-0054 (Project title: Orientation in migratory songbirds) and A09-0131 (Project title: Geographic variation in birds of western Canada). Field permits were provided by the Canadian Wildlife Service, Prairie and Northern Region office (AB Scientific Permit 11-AB-SC023) and Alberta Parks (11–107).

Video analysis

From the D-ViewCam software we exported each video in “.asf” format, noting the start and end time of each trial. We analyzed the video data with radR, an open source platform developed for acquiring and analyzing radar data (Taylor et al., 2010) that was more recently adapted to analyze video files. In brief, radR uses contrast to score individual pixels, and then uses movement, area and intensity to define objects. We sampled the videos at three frames per second. For each detected object we extracted the X and Y coordinates of its centroid, as weighted by the area of the object. Given that the light conditions change over the evening, we used three groups of parameters to accurately and consistently identify the bird as the primary object relative to the background (see Supplemental Information 1 for parameters used). We defined the center of the circular perch as the center of our analytical coordinate system and created an exclusion zone within the circular perch, such that the program would identify the bird only in the region from the perch outwards (Fig. 3). This makes our results more comparable to previous funnel studies, which only recorded jumps where the individual contacts the side of the funnel (Fitzgerald & Taylor, 2008). From this data radR generates a list of time-stamped X and Y coordinates for points where it identified an object (see Table S1 for an example of data extracted for a single individual).

Using R 3.0.3 (R Development Core Team, 2013) we applied a number of additional filters, primarily to remove noise (i.e., objects that were not the bird) and remove times when the bird was not moving (i.e., sitting on the perch). To remove noise, we first removed all data where there were three or more objects identified at a single time-stamp, as three or more objects were invariably an artifact of background noise. Sometimes radR identified the bird as two separate objects, primarily when the individual was above one of the four perch dowels, such that parts of the bird stuck out on either side of the dowel. For time-stamps with two objects, we averaged the XY values that were less than 500 pixels apart to include only those times that an individual was above the dowels (i.e., the dowels are slightly less than 500 pixels wide in the video image) and visually inspected the data to confirm this was accurate. To ignore points where the bird was sitting motionless on the perch we removed consecutive time points that had a lower XY distance than a predetermined threshold, which we calculated by studying videos of individuals sitting quietly on the perch (50 pixels for the current analysis). Each of the points that passed these filters were then transformed into an angle relative to North (N = 0°, E = 90°, S = 180° and W = 270°) and given a timestamp (in seconds) relative to the time of sunset for that evening.

From these data we estimated three behavioural traits for each individual: mean orientation, rho and activity, defined as follows. We calculated mean orientation using the “circular statistics” package in R (Agostinelli & Lund, 2013). We used the mean of angles observed over the entire observational period (restricting the time range to only those times when the birds were more active did not meaningfully change the results). We used the same R package to estimate rho, a measure of angular variance that varies between 0 and 1 (i.e., a measure of the concentration of points, with a value of 1 being perfectly concentrated). Finally, we used the total number of time points at which an object (i.e., a moving bird) was detected by radR over the trial as the total activity for each individual.

To assess robustness of estimating the mean orientation for each individual using the data generated from radR, we chose five orientation videos at random and scored them by eye, blind to the output from radR. For this we visually estimated the angle of the bird (if present in the frame) every 30 s, over the entire video, and also recorded whether the bird was on the perch or in mid-flight. We then calculated the difference in angle between this estimate and that obtained from radR for the full dataset and also from only those points where the bird was observed to be in flight. Our data filtering appeared to be effective as our data from a random selection of videos analyzed by eye (for five birds) was consistent with the output from radR. When we compared all of the points observed by eye, including those where the individual was sitting on the perch, the resultant mean angles were within ±22° relative to the output from radR after filtering. If we included only those points where the individual was in flight (by excluding those times we observed by eye for which the individual was on the perch), which our filters within radR were designed to remove, the resultant mean angles were within ±11° of the radR output.

Molecular analysis

Blood samples were taken using a small needle and capillary tube from the brachial vein, stored in Queen’s lysis buffer (Seutin, White & Boag, 1991), and left at ambient temperature until returned to the laboratory. DNA was extracted using a standard phenol-chloroform procedure and resuspended with 50–200 µL of buffer (depending on the size of the pellet) containing 10 mM Tris–HCl and 1 mM EDTA, at pH 8.0, and stored at 4 °C. We sexed individuals molecularly using the procedure described in Fridolfsson & Ellegren (1999). We genotyped individuals at three molecular markers. The full PCR and genotyping protocol for two of the three nuclear markers (CHD1Z and numt-Dco1) was presented in a previous publication (Brelsford & Irwin, 2009). For the third nuclear marker, an 850 base pair fragment of RIOK2, we used the forward primer 5′-ATGGGTGTTGGCAAAGAATC-3′, the reverse primer 5′-GCTCCTCTTCRTTWGCAACA-3′, and a PCR annealing temperature of 60 °C. The enzyme XmnI cuts an allele common in Audubon’s warblers, but leaves intact an allele common in myrtle warblers. To generate a genetic hybrid index we scored a zero for each Audubon’s allele and a one for each myrtle allele for the three markers and divided this by the total number of alleles (six for males, four for females) resulting in an index that ranges between 0 (all Audubon’s alleles) to 1 (all myrtle alleles).

Combining genotype and migratory behaviour

To test whether orientation is associated with genetic background, we used two statistical approaches. First, we used a circular linear model to test whether orientation varied linearly with hybrid index (Agostinelli & Lund, 2013). Second, we used a circular ANOVA to test whether mean orientation angle differed in any way (not necessarily linearly) among these five genetic groups: those with all Audubon’s alleles (h-index = 0; Group A), those with mostly Audubon’s alleles (0 < h-index < 0.5; Group AH), those with mixed genotypes (h-index = 0.5; Group H), those with mostly myrtle alleles (0.5 < h-index < 1; Group MH), and those with all myrtle alleles (h-index = 1; Group M). While the linear model would assume that each hybrid class as an intermediate migratory orientation compared to the two classes on either side of it, the ANOVA does not make that assumption, allowing situations in which hybrid classes might have more extreme orientations than either parental group.

To test whether there was a significant mean orientation of all of individuals considered together, regardless of their genetic background, we used a Rayleigh test. This is a procedure to test the null hypothesis that the orientation angles are distributed randomly, with the alternative being that the distribution is clumped in certain direction(s). The test statistic is r, an estimate of rho, and is the magnitude of the mean vector (Fitzgerald & Taylor, 2008).

Molecular data

We obtained genotypes from 166 of the 181 yellow-rumped warblers captured at the beginning of fall migration through the Kananaskis area, including 123 individuals with orientation data. Based on the three genetic markers, our data set consisted of individuals with a broad range of genetic backgrounds, spanning from Audubon’s warblers (h-index = 0) through a broad array of hybrid genotypes to myrtle warblers (h-index = 1), consistent with the expectation based on prior research of a broad mixture of hybrid classes (F1’s, F2’s, backcrosses, etc.) and pure-type individuals (Brelsford & Irwin, 2009). The daily composite allele frequencies varied relatively little over the study period: the daily proportion of myrtle alleles was usually between 0.3 to 0.6, with an average over the study period of 0.4 (Fig. 4A). Our molecular sexing of individuals indicated that we captured an excess of male birds with 73% of all of the individuals identified as males (Fig. 4B). This was likely due to our use of song playback during mist netting.

Orientation trials

In the orientation trials, initial observations of yellow-rumped warblers in the orientation chambers indicated that their activity began to increase approximately one hour before sunset. During this time of increased activity their behavior also changed qualitatively, from primarily sitting on the perch and/or flying occasionally to and from the bottom of the cage, to performing more short flights from the perch to the top of the cage, consistent with zugunruhe (i.e., migratory restlessness; Emlen, 1970). The activity of the birds increased to a peak at around 20–30 min before sunset (Fig. 5), and then gradually declined. Following sunset their activity sharply declined, such that we recorded virtually no movements after 40 min post-sunset. At the end of most evenings individuals usually stopped moving within 5–10 min of each other with remarkable consistency (the steep decline to the right of Fig. 5).

Of the 123 individuals where we had genotype and orientation data, we found that 96 birds showed strong evidence of orientation behavior (i.e., within-individual r > 0.1; total activity >500). We used all 123 individuals in analyses; analyses using only those individuals with strong orientation did not qualitatively change results. Of the five genetic groups, our sample contained n = 6 for Group A (i.e., genetically Audubon’s), n = 45 for Group AH, n = 41 for Group H, n = 22 for Group MH and n = 9 for Group M (i.e., genetically myrtle). We did not detect a significant difference in mean orientation between the genetic groups (F = 1.23, P = 0.09) and there was no significant linear relationship between mean orientation and genetic hybrid index (t = 0.779, P = 0.22). Hence we could not reject the null hypothesis of no association between genetic background and orientation.

While there was considerable variability in orientation among individuals, there was a significant mean orientation towards 25° or NNE (Fig. 6A; n = 123, among-individual r = 0.320, P < 0.01), with a 95% confidence interval between 5° and 46°. Separating the birds by their genetic hybrid index, each group had mean orientations similar to N or NE, although only Groups M (h-index = 1) and H (h-index = 0.5) showed evidence of significant mean orientations compared to the null of an even distribution of orientation angles: Group A, 38° (n = 6, r = 0.634, P = 0.09); Group AH, 3° (n = 45, r = 0.230, P = 0.09); Group H, 54° (n = 41, r = 0.373, P < 0.01); Group MH, 359° (n = 22, r = 0.321, P = 0.10); and Group M, 15° (n = 9, r = 0.733, P < 0.01).