Desert ants benefit from combining visual and olfactory landmarks

Landmark ambiguity during navigation can be reduced either by increasing sensory sensitivity or by integrating information deriving from several modalities (Wessnitzer and Webb, 2006). The desert ant Cataglyphis fortis Fabricius 1793 has, so far, been a model organism for studying mainly visual orientation (Wehner, 2003). The individually foraging ants leave their nests for long-range foraging trips. Once they encounter a food item they return to the inconspicuous nest entrance on a straight path following their home vector. The ant's home vector is the continuously updated and reversed sum of all directional and translational movements (Wehner, 2003; Wittlinger et al., 2006). Owing to the egocentric nature of the ants' path integration system, errors accumulate and the home vector leads the ants only to the approximate vicinity of the nest. However, for the survival of the foragers it is essential to return to the nest as fast as possible and to minimize the time spent outside the nest in the hostile habitat. Therefore, to reliably locate the nest entrance the ants make use of visual landmarks (Wehner, 2003). The ants learn the visual surroundings of the nest and use this knowledge when returning from their foraging trips. Unlike in, for example trail-laying ant species, olfactory orientation was thought to play no major role in C. fortis, with the exception of when locating food sources (Wolf and Wehner, 2000). Only recently, it was shown that, in addition to the visual panorama, C. fortis also memorizes environmentally derived olfactory cues and uses this information for close-range navigation (Steck et al., 2009; Steck et al., 2010).

This study investigated whether C. fortis is able to use the combination of visual and olfactory cues and whether combined cues improve the homing performance of C. fortis.

Field experiments were performed in the ants' natural habitat, the dried-out salt lakes in Northern Africa. The experimental site was located near the village of Menzel Chaker, Tunisia (Sebkhet Bou Jemel, 34°96′N, 10°41′E).

Foraging ants were trained to associate the nest entrance with an olfactory cue, a visual cue or a combination of both cues. Each training and test was conducted in linear channels with a U-shaped cross section. The channels were composed of aluminium modules (length, 1 m; width, 7 cm; height, 7 cm). The channel walls were covered with a homogeneous light-coloured adhesive tape and the channel floor was coated with quartz sand to eliminate reflections. The training and the test channel were aligned in parallel, with the wind blowing from the nest entrance to the feeder (Fig. 1A).

We enclosed an ants' nest with a low bucket and connected it via a tube to the training channel (Fig. 1A). The ants were trained to visit a feeder positioned 1 m downwind of the inconspicuous nest entrance in the channel floor. The entrance hole was marked by one of the following nest-defining cues.

As a visual cue we used two pieces of black cardboard (each 10 cm wide, 7 cm high) that were placed adjacent to the nest entrance on the channel walls (Fig. 1B). Because C. fortis has a visual resolution of approximately 3 deg (Labhart, 1986; Zollikofer et al., 1995) the black cardboard became visible to an approaching ant from a distance of approximately 35 cm. The pieces of cardboard covered the largest part of the ant's visual field (i.e. were most salient) at the position of the nest entrance. We tested whether the visual cue provided additional olfactory information by analyzing the air above the cardboard using gas chromatographic analysis (Agilent Technologies, Santa Clara CA, USA; model 7890A GC-MS). We did not detect any cardboard-derived olfactory cues (data not shown).

As an olfactory cue we applied 2 μl of diluted indole (1:50 in hexane) directly at the nest entrance. In order to ensure the presence of the cue at all times during the training, the olfactory cue was reapplied every 20 min; for a detailed description see Steck et al. (Steck et al., 2009). The range of the olfactory cue was determined using a photoionization detector (Aurora Scientific Inc., Ontario, Canada; model 200A). Using this device we detected the volatiles until 50 cm downwind of the odour source, with the olfactory cue being strongest directly at the odour source, i.e. at the nest entrance. For a detailed description of the plume structure in the channel see Steck et al. (Steck et al., 2010).

The combined cue consisted of the visual cue (two pieces of black cardboard) and the olfactory cue (2 μl of diluted indole).

After the nest had been connected to the training channel the ants located the feeder usually within a few minutes. Food crumbs were provided in a feeder trap, i.e. a cup placed in a hole in the ground in the centre of the channel. The ants could enter the feeder trap but were not able to leave it without assistance. Having entered the feeder trap for the first time, the ants were marked individually with a two-colour code (dots of enamel paint applied on the gaster) and were released from the feeder trap. For each marked ant we counted how many training runs it had conducted. For each training situation we used a different nest. That means that a nest that had been used for training with, e.g. a visual cue, was never used for training with another nest-defining cue again.

For the test, marked ants were caught at the feeder and were released for their homebound run in a test channel 2 m downwind of one of the nest-defining cues (Fig. 1). Only ants that had gathered a food item were caught and tested. The test channel was not connected to a nest and therefore, did not contain an exit hole. The ants started their homebound run and then began a systematic nest search. The ants' home vector was 1 m in length, because they had covered the distance from the nest to the feeder (i.e. 1 m) before being displaced from the feeder to the test channel (Fig. 1A). This experimental setup ensured that there was a conflict between the path integrator information and the landmark navigation information. Therefore, we were able to tell which information the ants were relying on: would the ants run off their home vector and search at the nest position as defined by the path integrator, or did they learn to associate the nest entrance with one of the landmarks that was positioned 1 m behind the path integrator position?

An ant having run off its home vector without encountering the nest switches to a systematic and well-investigated nest search (Mueller and Wehner, 1994; Wehner and Srinivasan, 1981). During this nest search C. fortis carries out loops around the position where it assumes the nest entrance to be, returning at regular intervals to the estimated position of the nest (Wehner and Wehner, 1986).

As a result of the linear channels the ants' systematic nest search is reduced to one dimension and therefore, is characterized by subsequent turning points (Fig. 1A and Fig. 2) (Cheng and Wehner, 2002). We recorded the first six turning points with an accuracy of 5 cm by placing a measuring tape alongside the test channel. To display the search pattern we calculated search density plots for each experimental situation (Fig. 3). Therefore, we divided the total distance of the test channel into virtual 5 cm bins. The visits per bin were cumulated for each ant during its nest search and summed up for each experimental group. Consequently, the peak of the search density plot depicts the bin most often visited, i.e. the estimated position of the nest entrance. The search density plots were normalized to the total number of bin visits.

Apart from the position of the assumed nest entrance, the shape of the search density plots reveals additional information. The ‘sharpness’ of the peak, i.e. the narrowness of the nest search reflects the ants' confidence in the nest position, as the loops decrease in size with increasing confidence in the estimated position of the nest (Merkle et al., 2006). We used the median distance between the first six turning points and the position of the nest-defining cue as a measure of confidence in the estimated position of the nest, i.e. the search accuracy (Fig. 1). A nest search centred closely around the position of the nest-defining cue results in a short median distance between turning points and cue, whereas a broad search pattern is reflected in a long median distance.

In order to investigate how experience affects the homing performance, we tested naïve ants (i.e. ants that arrived at the feeder before the nest-defining cues had been installed) 1-run ants (i.e. ants that arrived for the first time at the feeder after the nest-defining cues had been installed), 5-run ants and 15-run ants.

Statistical differences between the experimental groups were based on the median distances between the first six turning points and the position of the nest-defining cue, i.e. the search accuracy, and were calculated with GraphPad Instat (GraphPad Software Inc., La Jolla, CA, USA).

We first tested how naïve animals respond to the visual, the olfactory or the combined cues. The turning points of naïve ants were ranged widely around the path integrator position (naïve ants in Fig. 2) and search patterns were broad (naïve ants in Fig. 3). Moreover, animals that were confronted with the olfactory cue for the first time seemed to be slightly repelled as the search density plot exhibits a local minimum at the position of the nest-defining cue (Fig. 3).

Having experienced one of the cues, be it the visual, the olfactory or the combined cue, at least once in the training, the search density plots exhibit the maximum at the position of the cue and not at the path integrator position (Fig. 3).

Ants that were tested with an olfactory or a visual cue after they had experienced it only once displayed broad search patterns (1-run ants in Fig. 3) with low search accuracies, i.e. long median distances between turning points and the nest-defining cues (1-run ants in Fig. 4A,B). With increasing experience of the cue the ants exhibited an increased search density at the position of the cue (Fig. 3), and displayed the most focused search after 15 training runs (Fig. 3). Accordingly, the median distances between the turning points and the cues became shorter with increasing experience (Fig. 4A,B), i.e. the ants became more confident about the position of the nest. At no experiential stage did the search accuracies of visually trained and tested ants differed from those of olfactory trained and tested ants (Fig. 4A,B). In contrast, when trained with the combined visual and olfactory cue, the animals displayed a focused search at the position of the combined cue after the very first training run (Fig. 3). Despite the ants' single experience with the combined cue their searches were as focused as after 15 training runs with the individual cues (Fig. 4, compare 15-run ants in A and B with 1-run ants in C). Although the ants needed 15 training runs with the single cues to reach the same accuracy as ants trained once with a combined cue, the final accuracy was the same irrespective of whether the ants were trained and tested with the individual or the combined cues (15-run ants in Fig. 3 and Fig. 4A–C).

We finally investigated whether the ants would recognize a single olfactory or visual cue when they had been trained with a combination of both cues. Therefore, we again trained ants to associate their nest entrance with a combined cue, but now tested them either with the single olfactory cue or the single visual cue. Even after one single training experience with the combined cue the ants concentrated their search on the single test cues (Fig. 3 and 1-run ants in Fig. 4C–E). However, having experienced the combined cues several times, the ants lost their confidence in the singly presented cues, i.e. displayed a broad search pattern with long distances between the turning points and the cues (Fig. 3 and 15-run ants in Fig. 4D,E).

Desert ants have been shown to use visual and olfactory cues for navigation. Although the visual system has been well studied (Bisch-Knaden and Wehner, 2003a; Wehner, 2003), the use of olfactory landmarks has been described only recently (Steck et al., 2009; Steck et al., 2010). In the present study we examined whether and how the visual and the olfactory navigation systems interact. We investigated whether the ants benefit from nest-defining cues that combine visual and olfactory information or only relied on single visual or olfactory cues.

All nest-defining cues used in this study did not attract naïve ants (naïve ants in Fig. 3). Moreover, the ants that had never experienced the olfactory cue before appeared to pass the olfactory cue less frequently (Fig. 3). Because naïve ants had no external cue, i.e. visual and/or olfactory landmarks, they could rely solely on the information derived from the path integrator. Therefore, the maxima of the search density plots for these experimental groups were at the position as calculated by the path integrator (control ants and naïve ants in Fig. 3).

The ants' responses to the cues in the test changed considerably when there had been a cue installed during training and the ants were asked to associate their nest entrance with the cues. We observed a shift of the maxima of the search density plots towards the position of the nest-defining cue even after the very first training run (1-run ants in Fig. 3). However, the search accuracies of the visual- or olfactory-trained and tested 1-run ants did not differ significantly from the search accuracies of naïve ants (Fig. 4A,B). It needed five to 15 experiences with the individual cues before the ants started to focus their search on the cues, with the highest accuracy being reached only after 15 experiences (Fig. 4A,B). This acquisition rate is comparable to findings in desert ants that had to learn visual landmarks in the open field (Bisch-Knaden and Wehner, 2003b; Narendra et al., 2007). It is difficult to precisely quantify the salience and the operating distance of the visual and the olfactory cues that we used in the present study. However, both cues could be detected by the ants at a distance of between 35 and 50 cm from the nest entrance and, most important, both cues reached their highest intensity directly at the nest. Our findings that the ants learned the association between their nest entrance and the visual or the olfactory cue equally fast (Fig. 4A,B) points at a comparable salience of these two cues. Furthermore, they show that C. fortis, which so far was considered to be a mainly visually guided navigator, is able to make use of olfactory cues equally well.

The ants that were trained with the combined visual and olfactory cues exhibited a search accuracy that did not differ from ants that had repeatedly experienced individual cues. However, being trained with the combined cues the ants focused their search on the combined test cue after just a single training experience (1-run ants in Fig. 4C). The rapid location of the nest entrance after a long-lasting foraging trip in the inhospitable desert is crucial for an ant. Therefore, faster acquisition of the combined cue might be of significant benefit for C. fortis, especially as ants of the genus Cataglyphis usually conduct only 20–50 foraging runs during their brief lifetime (Harkness, 1977; Schmid-Hempel and Schmid-Hempel, 1984). The benefit of multisensory training in subsequent unimodal tests has been shown for humans (Shams and Seitz, 2008) and Drosophila (Guo and Guo, 2005). The ants' accelerated acquisition of the combined cue thus supports the enhanced perception (Chow and Frye, 2008; Goyret et al., 2007; van Swinderen and Greenspan, 2003) and learning (Reinhard et al., 2006; Rowe, 2002; Waeckers and Lewis, 1994) of bimodal signals.

We also tested how ants that had been trained with the combined cue responded to the single olfactory or visual cues. We found that after only one single outbound run with the combined cue the ants concentrated their search on the single olfactory or visual cues as accurately as on the combined cue (1-run ants in Fig. 4C–E). Hence, although being trained to a combination of both cues, the ants accepted the single cues as nest-defining landmarks. The finding that the ants learned the single cues faster in a bimodal background (1-run ants in Fig. 4A,B,D,E) again supports the enhanced perception and learning of bimodal signals (Chow and Frye, 2008; Goyret et al., 2007; Pearce and Bouton, 2001; Reinhard et al., 2006; Rowe, 2002; van Swinderen and Greenspan, 2003; Waeckers and Lewis, 1994).

However, after extended training with the combined cue the ants displayed a broad search pattern when tested with the individual cues (Fig. 3, and 15-run ants in Fig. 4D,E). Apparently, with increasing experience the ants had realized that the single cues were only valid when detected in combination. Navigating subjects are often confronted with similar, i.e. ambiguous visual landmarks (Wessnitzer and Webb, 2006), which could become unequivocal if a second modality is added. In the habitat of C. fortis visual landmark information can be ambiguous, i.e. several landmarks (e.g. different shrubs of halophytic plants) can be similar in shape and size. Therefore, the use of bimodal cues could prevent confusion arising from environmental ambiguity. C. fortis has previously been a model organism in which to study orientation because of its recognized visually based navigation system. However, we found that olfactory cues are learned as fast as visual cues, whereas combined visual and olfactory landmarks are learned much faster. Apparently, the hostile habitat of C. fortis pushes the evolution of a sophisticated navigational machinery, exploiting and combining input deriving from different sensory modalities.

We thank A. Fiedler for help in the field.