Homing abilities of the Australian intertidal ant Polyrhachis sokolova

At the end of a foraging trip, ants return home typically using one of three navigation strategies: (1) following a pheromone trail, (2) taking the shortest path home, a strategy known as path integration (for reviews, see Collett and Collett, 2000b; Wehner and Srinivasan, 2003), or (3) returning using familiar landmark information (Collett et al., 2007; Mangan and Webb, 2012). In trail following, individuals who travel along a particular route produce a pheromone trail secreted by one of their abdominal glands. A follower who also travels on the same route detects these pheromones from the chemoreceptors present on their antennae to follow the trail until they reach the destination (e.g. Hölldobler and Möglich, 1980). In path integration, an individual leaving the nest keeps track of the distance travelled and the angles steered on the outbound journey and upon finding food integrates this information to compute the shortest home vector (HV) (Collett and Collett, 2000a; Collett and Collett, 2000b; Müller and Wehner, 1988; Wehner and Srinivasan, 2003). Distance travelled is estimated by a stride integrator (Wittlinger et al., 2006; Wittlinger et al., 2007) and compass information is derived from the pattern of polarised skylight (Wehner, 2001; Wehner and Müller, 2006). For landmark guidance, visual landmark information first needs to be learnt. For this, ants carry out a systematic learning routine when leaving the nest and also when leaving newly discovered food sources (Müller and Wehner, 2010; Nicholson et al., 1999). These learning walks are most likely crucial in pinpointing goals using visual landmarks (Narendra et al., 2007; Wehner and Räber, 1979).

From the ants studied so far, it is clear that the above navigation strategies are not species specific. For instance, inexperienced foragers of the trail-following ant Paraponera clavata use information from the pheromone trails, whereas experienced individuals switch to relying on visual landmark information (Harrison et al., 1989). Among solitary foraging ants, for instance in the Australian desert ant Melophorus bagoti, individual ants typically establish and adhere to routes that meander around specific bushes and tussocks, but when familiar landmarks are unavailable they switch to relying on a path integrator (Kohler and Wehner, 2005; Narendra, 2007a; Narendra, 2007b). The converse holds true for the African desert ant Cataglyphis fortis, which typically relies on path integration, but switches to relying on landmark information when available (Collett and Collett, 2000a; Graham et al., 2003; Knaden and Wehner, 2005; Müller and Wehner, 1988; Müller and Wehner, 2010; Wehner et al., 1996). The ability to orient using the pattern of polarised skylight and visual landmarks is not restricted to ants active in bright light conditions only. The large-eyed bull ant Myrmecia pyriformis is a case in point. These ants navigate to specific trees during the evening twilight and return to their nest in the morning twilight (Narendra et al., 2010), the periods of the day when the pattern of polarised skylight is most simple because the sun is at the horizon (Cronin et al., 2006). These ants derive compass information from both the pattern of polarised skylight and visual landmarks even in the dim light conditions under which they operate (Narendra et al., 2013b; Reid et al., 2011). It is becoming increasingly evident that the information content available in the environment strongly influences the navigation strategies used by ants and the navigational decisions they make (Zeil, 2012). One of the most compelling pieces of evidence for this comes from M. bagoti, in which ants that inhabit landmark-rich habitats relied on their path integrator to travel approximately 43% of their HV, and this increased to 70% in ants that inhabit landmark-poor habitats (Cheng et al., 2012; Cheung et al., 2012; Narendra, 2007a).

Since animals can use more than one navigational strategy, it is of great interest to identify how animals resolve situations when a conflict occurs between multiple navigation strategies. Studies on cue conflict in ants can be categorised as follows: (1) conflict between the HV and landmarks; (2) conflict between celestial and terrestrial cues, (3) conflict between celestial and idiothetic cues, and (4) conflict between the food vector and a vector derived from local landmarks. A conflict between the HV and visual landmarks appears to be resolved in two distinct manners. In one, animals ignore information from the HV and directly home in to the nest [Formica japonica (Fukushi and Wehner, 2004); Myrmecia croslandi (Narendra et al., 2013a)]. In the other, animals resolve the conflict by following a compromise direction between the HV and the landmarks [M. bagoti (Narendra, 2007b; also see Wystrach et al., 2012)]. Cue conflict between the celestial and terrestrial cues on an outbound journey is resolved by choosing a compromise direction indicated by the two cues [Myrmecia pyriformis (Reid et al., 2011)]. But when a conflict is created between the compass information provided by celestial cues and idiothetic cues, ants solely rely on the celestial cues [C. fortis (Lebhardt et al., 2012)]. When conflicts occur between the food vector and a vector attached to a local landmark (e.g. end of a channel), ants followed a direction that is a compromise between the directional estimates of the two cues [C. fortis (Collett, 2012)]. We here aim to identify the navigational strategies used by the intertidal ant Polyrhachis sokolova Forel and determine how they resolve conflicts between the HV and landmark information while returning to the nest. To the best of our knowledge, this is the first study that addresses navigational strategies in the Old World ant genus Polyrhachis, thus contributing towards a comparative analysis in the family Formicidae.

The intertidal ant P. sokolova is unique among ants in nesting at the land and ocean interface of the mangrove zone (Robson, 2009; Robson and Kohout, 2007), where their nests are regularly inundated by tides. The mangrove trees provide a distinct panoramic skyline in the landscape with no other conspicuous landmark features on the ground. The ants construct subterranean nests at the base of mangroves and when the tide water reaches the nest, loose soil particles surrounding the nest entrance collapse to form a plug to prevent water from entering the nest, thus keeping the interior chambers dry during high tides (Nielsen, 1997). These ants typically feed on bird droppings and crustaceans and bivalves (e.g. crabs) that get washed in by the tides (A.N., personal observation). They mostly travel on dry land but when they encounter a body of water, remarkably, they swim. While swimming, they use their first two pairs of legs as paddles and hold their hind legs on the water surface in line with the body (Fig. 1) (Robson, 2009). Given the nature of the substrate on which they forage (mud and/or water), it is unlikely that these ants use pheromone trails for navigation. The ants are found along the Australian east coast from Torres Strait to Gladstone in Queensland, and also in nearby tropical countries (Andersen et al., 2013; Kohout, 1988). Our study was carried out in the mangrove habitats of Pallarenda, Townsville, Queensland, Australia (Fig. 1). The study was carried out during the day at low tides in April and May 2012.

We trained ants to a feeder placed 7 m north of the nest. The typical food source used for ants, such as 10% sugar or honey solution and cookie crumbs, failed. After we noticed some ants feeding on dead crabs, we provided clams (i.e. pipis, Plebidonax deltoides; purchased from the local seafood stores) as a food source for these ants, which they drank from or tore and carried small pieces to the nest. Ants leaving from the feeder (full-vector ants) were captured individually in foam-stoppered Perspex tubes and transferred in the dark to one of seven release stations. Ants were released: (1) at the feeder, (2) 1 m lateral to the feeder (1 m lateral), (3) 1 m lateral to and 5 m behind the feeder (1 m lateral and 5 m behind), (4) 5 m lateral to the feeder (5 m lateral), (5) 5 m lateral to and 5 m behind the feeder (5 m lateral and 5 m behind), (6) 50 m lateral to the feeder (50 m lateral) or (7) >100 m away from the feeder (>100 m distant). To test the navigation abilities in the absence of vector information, we captured ants returning from the feeder close to the nest (zero-vector ants) and released them at the same seven release stations. Zero-vector and full-vector ants were released at one of the randomly chosen release stations. Tested ants were marked with a single colour to ensure they were recorded only once.

Ant paths were tracked by placing miniature flags every 10 cm behind a walking ant, carefully avoiding disturbing the ants' progress. Ants were tracked between 08:00 and 11:00 h and between 15:00 and 18:00 h. We avoided tracking when the sun was in the zenith because the polarisation pattern which is horizontal everywhere does not provide a reliable compass cue. The flag-marked path was later recorded using a Differential Global Positioning System (DGPS; NovAtel, Calgary, AB, Canada). The setup consists of a base station antenna (GPS-702-GG L1/L2, GPS plus GLONASS), a base station receiver (FLEXPAK-V2-L1L2-G GPS plus GLONASS RT-2), a rover antenna (ANT-A72GLA-TW-N 532-C) and a rover receiver (OEMV-2-RT2-G GPS plus GLONASS). In a DGPS, a stationary reference or base station calculates corrections for a mobile rover antenna, the position of which is determined with centimetre accuracy at least on a local scale, in our case an area of approximately 120 m radius. The stationary base station electronics and antenna were mounted on a tripod and set to integrate antenna position readings over 30 min. The rover receiver electronics were carried on a backpack and connected to the rover antenna that was mounted at the end of a long, hand-held stick, so that it could be moved close to the ground along a flag-marked path (Narendra et al., 2013a; Narendra et al., 2013b). The base station and rover communicate through a radio link, allowing the exchange of corrections that provide position accuracy of the rover antennae of 1–2 cm. Northing, easting and height coordinates in metres, together with the standard deviations of position error estimates, were recorded and monitored at 1 s intervals with a laptop computer and extracted with a custom-written MATLAB program (The MathWorks, Natick, MA, USA).

Ants displaced close to the familiar route were tracked until they reached the nest or for up to 20 min, which was well within the time required for ants to travel directly home from the largest displacement. When ants were displaced to unfamiliar locations, we tracked them until they began a search. To identify the start of search we measured the cumulative distance travelled from the release point. The location where this distance decreased for more than four consecutive points (equivalent to 40 cm) was identified as the start of search. Circular analyses of heading directions of ants from different release stations were carried out based on Batschelet (Batschelet, 1981) using Oriana (Kovach Computing Service, Anglesey, UK). We compared the distance travelled by full- and zero-vector ants that were released in unfamiliar locations, relative to their path length. For this, we used animals whose path length was ≤10 m.

We captured the panoramic scenes at each of the release locations on a single day between 10:00 h and 10:15 h using a Sony Bloggie camera (MHS-PM5). The camera was levelled using a spirit level. During the gathering of these images, reference directions (e.g. nest and/or northing) were identified. The circular panoramic images were unwarped to rectangular panoramas measuring 1759×198 pixels, which corresponded to 360×40.52 deg, using a custom-written MATLAB program (courtesy of Wolfgang Stürzl). Sun glare and reflection artefacts were removed using the colour replacement tool in Adobe Photoshop CS6 (Adobe Systems, San Jose, CA, USA) to copy adjacent pixel values into the corrupted areas. The images were converted to greyscale, shifted (MATLAB circshift function) to align the nest direction in the centre of each image, and low-pass-filtered with a 18×18 pixel Gaussian filter with a resolution of 6 pixels deg⁻¹ to match the interommatidial angle of the eye of P. sokolova (Narendra et al., in press). We compared view familiarity between different release sites by determining the rotational image difference function (rotIDF) (for details, see Stürzl and Zeil, 2007; Zeil et al., 2003). We did this by calculating the pixel differences for a shift in each pixel between the image at the feeder and the image at each release station using the MATLAB circshift function. The values were then squared and averaged. For each image shift, the root mean squared pixel differences were determined. The minimum derived by such a comparison (i.e. least difference between images) indicates the home direction for each release location.

The initial heading direction of ants released at the feeder location (θ=93.50 deg; Fig. 2A) was close to that indicated by the HV (θ=100.0 deg; indicated by a black arrow in circular plots in Fig. 2). All ants from this location returned to the nest directly without exhibiting any search (Fig. 2A). The initial heading direction of ants released 1 m lateral (θ=74.14 deg; Fig. 2B) was directed slightly away from both the true nest direction (θ=92.0 deg) and the HV (θ=100.0 deg). All ants from this release location returned to the nest directly without exhibiting any search (Fig. 2B). Ants released 1 m lateral and 5 m behind also exhibited an initial orientation (θ=81.45 deg; Fig. 2C) away from both the true nest direction (θ=91.0 deg) and the HV (θ=100.0 deg). Nearly 90% of the ants released at this location returned to the nest (Fig. 2C). Few ants (four out of 10) engaged in a brief search (U-turns) before heading directly to the nest. The initial heading direction of ants released 5 m lateral (θ=65.69 deg; Fig. 2D) was directed between the true nest direction (θ=58.0 deg) and the HV (θ=100.0 deg). Only 50% of the individuals released at this location returned to the nest (Fig. 2D). The initial heading direction of ants released 5 m lateral and 5 m behind (θ=94.58 deg; Fig. 2E) was also directed between the true nest direction (θ=73.0 deg) and the HV (θ=100.0 deg). Nearly 72.72% of ants released at this location returned to the nest (Fig. 2E). Most ants from this group travelled a distance greater than that indicated by their path integrator and subsequently corrected their heading to travel towards their familiar foraging corridor, before reaching the nest.

At all release stations, the initial heading directions of the full-vector ants were not significantly different from the directions predicted by the path integrator (V-test, P_s<0.001) and by the true nest (V-test, P_s<0.001; Fig. 2A,E).

At all release stations, the initial heading directions of the zero-vector ants were not significantly different from the potential directions predicted by the path integrator (V-test, P_s<0.001) and by the true nest (V-test, P_s<0.001; Fig. 2F,J).

All ants released at the feeder (Fig. 2F) and at the 1 m lateral location (one exception; Fig. 2G) successfully returned to the nest. Among the ants released 1 m lateral and 5 m behind (Fig. 2H), only one ant returned to the nest and the majority of ants searched close to the release location. Among the ants released 5 m lateral (Fig. 2I), only 20% of the individuals successfully returned to the nest. Of the two successful ants, one was initially directed towards the fictive nest and corrected its heading to travel to the familiar corridor and subsequently to the nest. The second ant was initially directed towards the habitual route, but then corrected its heading to travel towards the fictive nest and then corrected its heading once more to travel directly to the familiar corridor and then the nest. The unsuccessful ants in this group drifted in the direction towards the fictive nest but turned back and began searching after travelling a maximum of 3 m (Fig. 2I). Among the ants released 5 m lateral and 5 m behind (Fig. 2J), only two ants reached the nest. Both these ants walked towards the fictive nest and travelled a distance greater than the feeder–nest distance. Both ants corrected their heading direction at nearly the same spatial location, following which they headed to the habitual route and then reoriented to head to the nest. Most of the unsuccessful ants in this group (Fig. 2J) drifted towards the feeder location, but turned back after travelling a maximum distance of 5 m.

The initial heading direction of full-vector ants was directed towards the fictive nest at both 50 m lateral (nest=90 deg, θ=88.42 deg; Fig. 3A) and >100 m distant locations (nest=90 deg, θ=90.88 deg; Fig. 3C). The initial heading direction of zero-vector ants was not directed towards the fictive nest at the 50 m lateral (nest=90 deg, θ=62.97 deg; Fig. 3B) or >100 m distant locations (θ=57.49 deg; Fig. 3D), but occurred within ±90 deg of the fictive nest position relative to the release location. All full-vector ants, except two, travelled distances shorter than that indicated by their path integrator before beginning a search (middle row in Fig. 3A,C). Zero-vector ants released at the 50 m lateral and >100 m distant locations began searching immediately upon release. This is evident when distance travelled from the release location was compared between zero- and full-vector ants with path lengths of ≤10 m. At the 50 m lateral location, the full-vector ants travelled farther from the release point (3.98±0.62 m; mean ± s.e.m.) compared with the zero-vector ants (1.54±0.24 m; t-test, t=3.066, d.f.=23, P<0.01; Fig. 3, bottom panels). At the >100 m distant location, the full-vector ants also travelled farther from the release point (3.90±0.38 m) compared with the zero-vector ants (1.44±0.25 m; t-test, t=4.621, d.f.=24, P<0.001; Fig. 3, bottom panels). The zero-vector ants also returned repeatedly either to or close to the release point (Fig. 3, bottom panels, arrows).

Panoramic views available at all release stations are shown in Fig. 4A,B. Images are aligned such that the nest or the fictive nest position is in the centre of the image. We compared view familiarity of the feeder location with all release locations using the rotIDF. A comparison of the view from the feeder with itself produced a minimum, i.e. least image difference (black curve in Fig. 4C), which coincided with the south direction, which indicated the nest direction. A detectable minima was present at all the local release locations (locations 2–6 in Fig. 4B). However, the depth of the minimum decreased as the distance from the nest increased (blue, green, brown, red in Fig. 4C). Strangely, a detectable minimum was available at even the two distant release locations (light blue and yellow in Fig. 4C). To investigate this, we compared views from the feeder with the 50 m lateral location (Fig. 5A) and views from the feeder with the distant location (Fig. 5B) at different elevations. A comparison of views with elevation between 10 and 40 deg (upper part of the panorama), which was predominantly information from the sky (red box and red curves), produced a pronounced minimum. In contrast, comparison of views with elevation between 0 and 10 deg (lower part of the panorama), which has landmark information (brown box and brown curves), did not provide a distinguishable minimum, at least for the distant site.

The intertidal ant P. sokolova is a solitary foraging ant that nests at the base of mangroves and forages along the mudflats during low tide. Ants when displaced to a distant location travelled following their HV, providing evidence for path integration. Zero-vector ants when displaced either at or near the feeder location headed directly to the nest, providing evidence that P. sokolova ants can home in using landmark information. When a conflict between the direction indicated by the HV and landmark information was created, as was the case at the 5 m lateral and the 5 m lateral and 5 m behind release locations, ants initially head in a direction that is a compromise between that indicated by the HV and landmark information. When zero-vector ants were displaced to local release stations, only a small proportion of animals returned to the nest. This suggests that at these local release locations, vector information coupled with landmark information is required for most ants to home in to their nest.

At the 1 m lateral location (Fig. 2B), there was little discrepancy between the directions indicated by the HV and the landmarks indicating the true nest position. The initial orientation of these ants appeared to be directed towards the familiar route between the feeder and the nest and was not directed to the nest or following the HV. Once animals were in the familiar corridor they headed directly to the nest. It was, however, difficult to pinpoint whether ants were guided by the HV or by landmark information. Hence releasing zero-vector ants at the same location (Fig. 2G) demonstrated with certainty that P. sokolova can return home using landmark information alone and without vector information. Their ability to path integrate was evident when full-vector ants were displaced to unfamiliar locations, where ants walked in the direction towards the fictive nest (Fig. 3A,C). The rotIDFs indicate that the view similarity between the feeder location and distant location was primarily driven by information from the sky and not from the landmarks (Fig. 5). It is perhaps because of this dominant sky information that animals released at the distant location rely on their path integrator (Fig. 3C). For path integration, these ants most likely derive compass information from the pattern of polarised skylight (Labhart and Meyer, 1999; Wehner and Labhart, 2006), which draws support from the specialised photoreceptors in the dorsal rim area of their eyes (Narendra et. al., in press).

Another release location where there was little discrepancy between the directions indicated by the HV and landmarks was the 1 m lateral and 5 m behind location (Fig. 2C,H). At this release location, most of the full-vector ants (90%) returned to the nest, exhibiting individualistic paths. Full-vector ants, however, had HV information only for 7 m and hence should have started searching at 7 m or earlier (Cheung et al., 2012), which they did not. In most cases, animals continued to travel in a well-directed path to the nest. Some ants carried out a short U-turn, following which they headed directly to the nest. Thus the second half of their journey was most likely visually guided. As indicated from the rotIDFs (Fig. 4C), the similarity in the views increases as one gets closer to the nest, thus providing animals a reliable visual cue to locate home (Stürzl and Zeil, 2007; Zeil et al., 2003). But few zero-vector animals returned to the nest (Fig. 2H), indicating that this release location was too far away for ants to return home using landmark information alone. Most ants continued searching for the 20 min recorded duration. This is despite there being a detectable minimum when comparing the views from the feeder and the 1 m lateral and 5 m behind location (brown curve in Fig. 4C). Similarity in views appears insufficient to trigger homing in majority of the zero-vector ants, suggesting that perhaps vector information is required to some extent even to rely on visual landmark information.

The maximum discrepancy between the compass directions of the HV and landmark information is at the 5 m lateral (Fig. 2D,I) and at the 5 m lateral and 5 m behind locations (Fig. 2E,J). Full-vector ants released at both these locations could have again travelled their HV partially or completely before beginning a search (Cheung et al., 2012), but they did not. We noticed three distinct phases in the homing paths of these ants: Phase I, paths immediately after release; Phase II, search and reorientation; and Phase III, final approach. In Phase I, animals travelled in a direction intermediate between that indicated by the HV and landmark information. Here, ants travelled in a straight line until they were approximately 2.6 m from the nest position, indicating that ants released at the 5 m lateral and 5 m behind location (Fig. 2E) travelled a distance greater than their HV. Phase II began at nearly the same spatial location for ants from both the groups. Here, ants began to search and in most cases reoriented to head towards the familiar foraging corridor. Some ants (especially those for which data are shown in Fig. 2D) searched extensively and did not find the familiar corridor within the recording duration and were considered ‘lost’. In Phase III, ants reoriented as soon as they reached the familiar corridor (~1.3 m from the nest) and headed directly to the nest. In the case of zero-vector ants released at these two locations (Fig. 2I,J), the majority of the ants were considered to be lost after 20 min of recording. However, the initial heading direction of these ants also was directed towards the fictive nest. Few zero-vector ants from these two release stations were successful in finding the nest (red lines in Fig. 2I,J). Interestingly, these successful ants again displayed a three-phase homing behaviour similar to the full-vector ants, with the reorientation at the second phase starting at the same spatial location. The view similarity of the 5 m lateral and the 5 m lateral and 5 m behind release locations with the views at the feeder appear to be sufficient for a few zero-vector ants to return to the nest (green and red curves in Fig. 4C).

The initial heading direction of zero-vector ants towards the fictive nest at the local release stations (5 m lateral, and 5 m lateral and 5 m behind; Fig. 2I,J) is quite puzzling. It is unlikely that a residual vector caused this behaviour. This is because zero-vector ants were captured 5–10 cm from the nest entrance while returning from the feeder and this residual vector is less than the distance where the initial heading direction (50 cm from release) of ants was noted. If ants could recall their recent home vector when faced with unfamiliarity, this could explain their behaviour. The initial heading direction of zero-vector ants in the unfamiliar locations (50 m lateral and distant; Fig. 3B,D) occurred only in the southern hemicycle, i.e. ±90 deg of the fictive nest position. Perhaps even the little landmark information that was available in the southern half of the panorama (see bottom panel in Fig. 4B) resulted in animals heading only towards the southern half.

Foragers of P. sokolova rely on both visual landmarks and on vector information for homing to their nest. When a conflict occurs between the HV and landmark information, ants initially head in a direction that is a compromise between the two and then correct to head towards the nest. In some cases, landmark information alone appears to be insufficient for homing, and vector information coupled with landmark information is necessary. We are currently identifying the navigational tactics used by these ants while swimming.

We are grateful to Jochen Zeil for his support and advice in all stages of this project. We thank the Defence Science and Technology Organization, Australia, for the initial loan of the DGPS system, Javaan Chahl, Matt Garratt and Mark Snowball for designing, constructing and maintaining DGPS electronics, and Janelle Evans and Mahenddra Raj for field support. We are grateful to Jan Hemmi and Wolfgang Stürzl for their advice and help with MATLAB programming and camera calibration.