Species-specific differences in the fine structure of learning walk elements in Cataglyphis ants

Hymenopteran central place foragers – such as bees, wasps and ants – perform so-called learning flights or walks whenever they have to memorize the landmark panorama surrounding a particular place, e.g. the location of their nest. Since the first descriptions over a century ago (Peckham and Peckham, 1898; Wagner, 1907), learning flights have been investigated in great detail in wasps (Tinbergen, 1932; Zeil, 1993a,b; Zeil et al., 1996), honeybees (Becker, 1958; Capaldi and Dyer, 1999; Lehrer, 1991, 1993; Opfinger, 1931; Vollbehr, 1975) and bumblebees (Collett et al., 2013; Hempel de Ibarra et al., 2009; Philippides et al., 2013; Robert et al., 2017) using increasingly sophisticated techniques like harmonic radar (Capaldi et al., 2000; Degen et al., 2015, 2016; Osborne et al., 2013) or high-speed cameras (Riabinina et al., 2014; Stürzl et al., 2016). Much less is known about learning walks of ants (Fleischmann et al., 2016; Jayatilaka, 2014; Müller and Wehner, 2010; Muser et al., 2005; Nicholson et al., 1999; Stieb et al., 2012; Wehner et al., 2004). However, these few studies document that – like flying insects – some ant species exhibit a striking behavioral sequence at the beginning of their foraging life. Desert ants in particular, which are well known for their navigational performance (Ronacher, 2008; Wehner, 2008; Wehner and Rössler, 2013; Wehner et al., 1996), use a considerable amount (up to 3 days; Fleischmann et al., 2016; Stieb et al., 2012) of their short lives (often less than a week; Schmid-Hempel and Schmid-Hempel, 1984) outside the nest to perform learning walks. Cataglyphis ants are thermophilic scavengers searching for dead arthropods during their extensive foraging excursions. Cataglyphis fortis inhabiting North African chotts and sebkhas cover the largest distances compared with other desert ant species (Wehner, 1983, 1987a). Their paths can be longer than 1 km and their maximal foraging distances more than 300 m from the nest (Buehlmann et al., 2014). Cataglyphis noda, native to Greece, search for food closer to the nest, but still cover large distances (Wehner et al., 1983), as do Cataglyphis bicolor [Wehner, 1987a; note that in that publication (and others: Harkness, 1977; Wehner et al., 1983), C. noda belonging to the bicolor species group, i.e. to the same phylogenetical group as C. bicolor, is named C. bicolor]. Before the successful completion of these extensive foraging trips, Cataglyphis ants perform learning walks characterized by several typical features: naive ants (called ‘novices’ below) move slowly around the nest entrance and explore all directions around it in subsequent learning walks (Fleischmann et al., 2016; Wehner et al., 2004). These ants include repeated body turns in their small excursions and they do not bring back any food items (Fleischmann et al., 2016; Stieb et al., 2012; Wehner et al., 2004). However, in novices, the characteristics of the learning walks, including circular movements as well as saccadic rotations with distinct stopping phases, have not yet been analyzed in detail. Only a few studies have recorded these striking elements during learning walks in different desert ant species and report on their rotational motifs (C. fortis: Stieb et al., 2012; Melophorus bagoti: Wystrach et al., 2014; Myrmecia croslandi: Jayatilaka, 2014). Namibian desert ants, Ocymyrmex robustior, perform ‘re-learning walks’ when presented with a new landmark situation around the nest (Müller and Wehner, 2010). These walks contain marked pirouettes with well-defined nest-fixating stopping phases.

Here, we used high-speed video recordings to reveal temporal and spatial characteristics of the learning walk of novices in different Cataglyphis species inhabiting different types of environments (cluttered and open). Our analyses show that the ant species performed different types of turn. Only ants inhabiting a cluttered environment (namely C. noda and Cataglyphis aenescens) performed both voltes (small walked circles) and pirouettes (tight rotations about the vertical body axis). They paused during the latter to gaze in the nest direction. In contrast, ants inhabiting a featureless habitat (C. fortis) only performed voltes and did not gaze back to the nest entrance even if artificial landmarks offered a prominent panorama.

To compare learning walks of different Cataglyphis species inhabiting different environments, we performed experiments in Tunisia and Greece. The experiments with C. noda (Brullé 1832) were performed in July and August 2015 in the Schinias National Park near Marathon, Greece (38°08′N, 24°02′E). We used two nests located on different clearings in the surrounding pine forest. The trees around the clearings offer prominent skylines with natural landmarks (Fig. 1A). In addition, during July 2016, we recorded learning walks of a smaller species, C. aenescens (Nylander 1849), inhabiting the same clearings as C. noda. The experiments with C. fortis (Forel 1902) were carried out in June and July 2014 at a saltpan near Menzel Chaker, Tunisia (34°57′N, 10°24′E). The saltpan offers an almost landmark-free habitat. We used two different colonies for the video recordings, both situated towards the middle of the saltpan where almost no natural landmarks occurred (Fig. 1B).

All foragers were marked with the same color (Motip Lackstift Acryl, Motip Dupli GmbH, Haßmersheim, Germany) over a period of 3–4 days before video recordings started. We considered unmarked ants appearing on subsequent experimental days at the nest entrance as novices and recorded their learning walks for 2–3 days. As these novices are easily disturbed and disappear immediately into the nest (Wehner et al., 2004), we usually did not mark them. Therefore, we usually could not decide whether a learning walk was the first or a subsequent one. However, as the walks were only included in the data if the ants stayed within the recording area, and the ants move further away with more experience (Fleischmann et al., 2016), these walks were most probably early learning walks. In one colony of C. noda, we marked some of the naive ants with an individual color code in order to record subsequent learning walks of individual ants. The experimental setup consisted of cameras mounted on one or two tripods (Manfrotto 190, Lino Manfrotto+co. Spa, Cassola, Italy) placed north of the nest entrance. A Full HD video camera (HDC-SD300, Panasonic Corp., Osaka, Japan) recorded the nest entrance and its surroundings (i.e. an area of about 100 cm×60 cm) at 25 frames s⁻¹ and minimal zoom throughout the entire day. We used these videos to trace the learning walks in the case of one individually marked ant (see Fig. 2D–F), because its paths left the high-speed recording area. The high-speed camera (Lumix DMC-FZ200, Panasonic Corp.) recorded a smaller area (about 25 cm×43 cm) at 100 frames s⁻¹ temporal and 1080×720 pixel spatial resolution. The observer sitting 5 m east of the nest watched the ants through binoculars (Condor 10×50 1821050, Bresser GmbH, Rhede, Germany), and started and stopped the high-speed recordings via a remote control (Wireless Remote Release DCCS System Base, HAMA GmbH & Co KG, Monheim, Germany) whenever learning walks occurred.

In Tunisia, we performed three experiments with C. fortis: no artificial landmarks, one artificial landmark or three artificial landmarks were placed near the nest entrance. In the first case, the ants did not have any additional landmarks (Fig. 1B). In the second case, one black cylinder (height: 33 cm, diameter: 20 cm) was positioned 0.4 m north of the nest entrance (vertical angular size: 48 deg, horizontal angular size: 28 deg; Fig. 1C). In the third case, three black cylinders (height: 38 cm, diameter: 22 cm) were placed 2 m from the nest at 0 deg (north), 120 deg (southeast) and 240 deg (southwest) (for each cylinder: vertical angular size: 11 deg, horizontal angular size: 6 deg; Fig. 1D).

We cut the videos to the length of one learning walk or one turn and converted them into image sequences using QuickTime Player Pro 7.7.5 (Apple Inc., Cupertino, CA, USA). Both learning walks and individual turns were easily identifiable and were extracted by eye. A learning walk was defined as an explorative trip, i.e. no digging or foraging, of an unmarked ant from exiting the nest entrance until its return into the nest. A turn was defined as a circular movement during which an ant changed its gaze direction until eventually re-establishing its previous walking direction. We analyzed the image sequence frame by frame with the Matlab (MathWorks, Inc., Natick, MA, USA) application Digilite (Jan Hemmi and Robert Parker, The Australian National University). We tracked the ant's position by marking the mandibles as well as the thorax. Furthermore, the position of the nest and the north direction were marked. We used these four types of coordinates to determine the ant's position and its viewing direction relative to the nest (the nest direction was defined as 180 deg). We defined a stopping phase as a period of at least 100 ms during which the ant did not move forward and gazed in one direction. Of course, the duration of 100 ms is an arbitrary criterion. However, as O. robustior ants pause for about 150 ms (Müller and Wehner, 2010), and both Cataglyphis and Ocymyrmex desert ants are very fast (Wehner and Wehner, 2011), we are confident that we captured the vast majority of relevant stopping phases with this criterion. It is also noteworthy that the duration of the stopping phases in Ocymyrmex is temperature dependent (Müller and Wehner, 2010). However, the ground temperatures in Tunisia and Greece were similar [Tunisia: 41.7±9.3°C, 36.5–45.8°C; Greece: 47.2±21.6°C, 38.8–60.4°C; median±interquartile range (IQR: Q1–Q3), ranging from 31.3 to 51.4°C in Tunisia and from 26.6 to 66.3°C in Greece]. Concerning the gaze direction, we tolerated a deviation of 10 deg during a stopping phase. Multiple analyses of the same sequence showed that the accuracy for determining gaze direction is within this 10 deg limit. Based on these data, we also calculated the angular velocities of the turns.

Because our data were not normally distributed, we report medians± IQR, calculated by subtracting the lower quartile (Q1) from the upper quartile (Q3). We compared the duration of learning walks and the number of turns per learning walk of our experimental groups using the Kruskal–Wallis test. The significance level was α=0.05. If necessary, i.e. if there was a significant difference indicated by the Kruskal–Wallis test, we performed post hoc pairwise comparisons with the Mann–Whitney U-test with Bonferroni correction. The significance level was α=0.05 (i.e. α=0.0084 with Bonferroni correction, because we compared the four experimental groups: ‘C. fortis with no artificial landmark’, ‘C. fortis with one artificial landmark’, ‘C. fortis with three artificial landmarks’ and ‘C. noda’ pairwise with each other, resulting in six performed tests). We compared the number of stopping phases and the angular velocity of different types of turn pairwise within species using the Mann–Whitney U-test (α=0.05). All these statistical tests were performed with Matlab R2014b (MathWorks, Inc.). To analyze gaze direction, we used the Rayleigh test to test for uniformity, i.e. whether the data were randomly distributed over the 360 deg. If the data were significantly directed (α=0.05), we calculated the 95% confidence interval and checked whether the expected value, i.e. the direction of the nest (180 deg) was between the limits. All circular statistics were performed with Oriana (version 4.02, Kovach Computing Services, Pentraeth, UK).

Cataglyphis desert ants leave their nest for quite different reasons. When observing the happenings at the nest entrances of three Cataglyphis species (C. noda, C. fortis and C. aenescens) in two completely different environments (Fig. 1), three main reasons were identifiable, i.e. performing learning walks, digging/waste disposal and foraging (Fig. 2A–C). All these outdoor behaviors were clearly distinguishable. In this study, the focus was on the learning walks performed by the ants before starting their foraging careers (Fig. 2A). However, ants also left the nest to carry out nest-related tasks like waste disposal or digging behavior (Fig. 2C; see Harkness, 1977; Stieb et al., 2012), and of course to search for food (Fig. 2B). In addition, some ants carried out single nest mates or stood next to the nest entrance and guarded it from time to time as has been described in detail elsewhere (Harkness, 1977). Indoor workers that deposited soil or waste outside the nest moved very fast. They ran in a straight line outside the nest, dropped their item a few centimeters away [9.9±3.7 cm, 8.5–12.2 cm, median±IQR (Q1–Q3) dropping distance from the nest entrance, n=20], turned around by 180 deg immediately and ran back into the nest (Fig. 2C). Foragers also left the nest quickly and in a straight line. In contrast to diggers and learners, foragers moved far away from the nest entrance and left the recording area rapidly (Fig. 2B). Conversely, novices performed their learning walks at the beginning of their foraging career in close vicinity to the nest entrance, and therefore they remained inside our selected recording area, enabling us to analyze entire learning walks using high-speed video recordings. These explorative trips were very short [C. noda n=25: 10.4±9.3 s, 6.8–16.1 s; C. fortis without artificial landmarks n=12: 12.0±5.4 s, 11.2–16.6 s, median±IQR (Q1–Q3); Fig. 3A, Kruskal–Wallis test: duration of learning walks =3.98, n=66, P=0.27]. No significant differences between the durations of the learning walks of the different species or setups were apparent (Fig. 3A). We marked some individual C. noda ants with a unique color code at their first appearance so that we could record subsequent learning walks of the same ants (Fig. 2D–F). Typically, C. noda moved further away from the nest entrance with increasing experience until they eventually became foragers. The most obvious characteristic of C. noda learning walks were the frequent turns, during which the ants usually looked back to the nest. The ants performed up to 17 turns per learning walk in our recording area. The average number of turns was 3±3 [1–4, median±IQR (Q1–Q3), n=25; Fig. 3B]. In comparison to the Greek C. noda ants living in a pine forest, the Tunisian C. fortis ants inhabiting a saltpan performed fewer turns during their learning walks [0.5±1, 0–1, median±IQR (Q1–Q3), n=12; Fig. 3B; Kruskal–Wallis test: number of turns per walk =30.75, n=66, P<0.05, and post hoc pairwise comparison with Mann–Whitney U-test with Bonferroni correction: C. noda versus C. fortis without an artificial landmark (LM) z=3.41, n_Cn=25, n_Cf0LM=12, P<0.0084; C. noda versus C. fortis with one artificial landmark z=4.03, n_Cn=25, n_Cf1LM=15, P<0.0084; C. noda versus C. fortis with three artificial landmarks z=4.27, n_Cn=25, n_Cf3LM=14, P<0.0084].

To check whether C. fortis did not perform frequent turns because of the absence of visual landmarks, we manipulated its natural habitat with two different settings of artificial landmarks. In one experiment, we installed a black cylinder 0.4 m north of the nest entrance (Fig. 1C; Movie 1); in another, three cylinders were placed at 2 m distance around the nest entrance (Fig. 1D). However, there was not any noticeable difference between the learning walks of C. fortis under the different conditions (Fig. 2G–I). The duration of the learning walks remained the same (Kruskal–Wallis test: duration of learning walks =3.98, n=66, P=0.27; Fig. 3A), and the number of turns per learning walk of C. fortis did not increase [one artificial landmark: 0±0.75, 0–0.75, median±IQR (Q1–Q3), n=15; three artificial landmarks: 0±0, 0–0, median±IQR (Q1–Q3), n=14; Fig. 3B]. There was no significant difference between the number of turns per walk in the three conditions under which C. fortis ants performed their learning walks (Fig. 3B; Kruskal–Wallis test: number of turns per walk =30.75, n=66, P<0.05, and post hoc pairwise comparison with Mann–Whitney U-test with Bonferroni correction: C. fortis without an artificial landmark versus C. fortis with one artificial landmark z=1.09, n_Cf0LM=12, n_Cf1LM=15, P=0.28; C. fortis without an artificial landmark versus C. fortis with three artificial landmarks z=1.55, n_Cf0LM=12, n_Cf3LM=14, P=0.13; C. fortis with one artificial landmark versus C. fortis with three artificial landmarks z=0.38, n_Cf1LM=15, n_Cf3LM=14, P=0.71). As stated above, C. noda made significantly more turns per learning walk than C. fortis even when artificial landmarks were available for the latter species (Fig. 3B).

Beside the difference in the number of turns per learning walk between C. fortis and C. noda, our high-speed recordings also revealed striking qualitative differences of turns across and within species. Cataglyphis fortis and C. noda included different types of turns in their learning walks (Fig. 4). Cataglyphis fortis ants performed only one type of turn during their learning walks: small walked circles, we termed voltes. During this type of turn, the ants moved forward on a circumference. Voltes were only rarely interrupted by stopping phases longer than 100 ms. The median number of stopping phases per volte of C. fortis was 1±2 [0–2, median±IQR (Q1–Q3), n=20; Fig. 5A]. Cataglyphis noda also performed this type of turn (Fig. 4A,D; Movie 1). During these voltes, C. noda stopped 2±1.5 times [0.75–2.25, median±IQR (Q1–Q3), n=17; Fig. 5A, Mann–Whitney U-test: C. noda voltes versus pirouettes z=−3.99, n_CnV=17, n_CnP=85, P<0.05]. However, C. noda much more frequently performed another, more saccadic, type of turn we termed pirouettes (as in Müller and Wehner, 2010). Over 80% of C. noda turns were pirouettes (85 out of 102 analyzed turns were pirouettes and only 17 were voltes). In contrast to a volte, a pirouette does not include any forward motion. Instead, to perform a pirouette, an ant stops walking and turns on its vertical axis without any translation (Fig. 4B,C; Movie 1). To re-establish its former walking direction, an ant has to turn back either continuing or reversing its direction of rotation. Therefore, pirouettes may be full 360 deg turns or partial turns (Fig. 4E,F; Movie 1). Usually, the ants took the shorter way back to re-establish their initial angular position – only in 12 partial pirouettes did ants cover an angle larger than 180 deg when turning back (n=74). Furthermore, the angular velocities of the back-turns were usually higher than those of the in-turns (Fig. 5C). Pirouettes were frequently interrupted by stopping phases defined as phases of at least 100 ms during which the ants did not move forward and gazed in one direction (±10 deg to compensate for tracking inaccuracies). Cataglyphis noda pirouettes included 4±3 [2–5, median±IQR (Q1–Q3), n=85] stopping phases, significantly more than their voltes (Fig. 5A). In addition, C. noda pirouettes had a lower angular velocity than their voltes (Fig. 5B, Mann–Whitney U-test: C. noda voltes versus pirouettes z=2.84, n_CnV=17, n_CnP=85, P<0.05). Interestingly, a smaller Cataglyphis species (C. aenescens) inhabiting the same clearings as C. noda in the Greek pine forest also performed both types of turns. Similar to C. noda, pirouettes in C. aenescens contained more stopping phases than voltes (Fig. 5A, Mann–Whitney U-test: C. aenescens voltes versus pirouettes z=−2.50, n_CaV=5, n_CaP=15, P<0.05); however, there was no significant difference in their angular velocity (Fig. 5B, z=0.00, n_CaV=5, n_CaP=15, P=1). In both C. noda and C. aenescens, the turning-in movements of the partial pirouettes were slower than the turning-out movement (Fig. 5C, Mann–Whitney U-test: C. noda turning-in versus turning-out z=−6.79, n_Cni=74, n_Cno=74, P<0.05; C. aenescens turning-in versus turning-out z=−2.04, n_Cai=14, n_Cao=14, P<0.05).

The striking feature of the saccadic pirouettes is that the ants turn back and look towards the nest entrance. As has been described for other desert ants (see C. bicolor: Wehner et al., 2004; O. robustior: Müller and Wehner, 2010), C. noda ants turned back to the nest (defined as the direction 180 deg), faced the nest entrance (which was invisible to them) and paused for a few hundred milliseconds [longest stopping phase per turn: 162±68 ms, 134–202 ms, median±IQR (Q1–Q3); minimum by definition 100 ms, maximum 469 ms, n=83]. However, C. noda ants only paused and fixated the position of the nest entrance during one type of turn, i.e. during pirouettes (Rayleigh test: z=30.48, n=83, P<0.05; 187.5 deg, mean vector, 174.7–200.2 deg, 95% confidence interval; Fig. 6A). In contrast, the gaze direction of C. noda ants during the voltes was randomly distributed (Rayleigh test: z=1.29, n=13, P=0.28; Fig. 6B). The same was true for C. fortis ants – they also did not fixate the nest direction during voltes (Rayleigh test: z=2.46, n=11, P=0.08; Fig. 6C). In contrast to C. noda, C. fortis never performed any pirouettes during which they stopped to look back to the nest entrance (Fig. 5). Cataglyphis aenescens ants inhabiting the same clearings in the Greek pine forest as C. noda also performed pirouettes and looked back to the nest entrance during the longest stopping phases (Rayleigh test: z=10.17, n=11, P<0.05; 183.5 deg, mean vector, 172.5–194.4 deg, 95% confidence interval; Fig. 6D).

Conspicuous turns have been observed in several desert ant species (C. fortis: Fleischmann et al., 2016; Stieb et al., 2012; C. bicolor: Wehner et al., 2004; C. aenescens: Petrov, 1993; Cataglyphis bombycina: Wehner, 1994; Wehner and Wehner, 1990; M. bagoti: Wystrach et al., 2014; O. robustior: Müller and Wehner, 2010). However, these turns have never been directly compared with each other. In this study, we showed that both C. noda and C. aenescens inhabiting a pine forest in Greece (Fig. 1A) performed voltes as well as pirouettes during their learning walks. These different types of turns can be easily distinguished qualitatively – the former is a walked circle whereas the latter consists of a full or partial rotation about the body axis. However, there were also quantitative differences – the number of stops per turn was higher in pirouettes than in voltes in both species. Most importantly, C. noda as well as C. aenescens only stopped to gaze back at the nest entrance during the pirouettes. Furthermore, in partial pirouettes, the turning-in rotations were slower than the turning-out rotations as is the case in O. robustior (Müller and Wehner, 2010). In contrast, C. fortis inhabiting a featureless saltpan in Tunisia (Fig. 1B) did not perform any pirouettes and did not look back at the nest. We manipulated the natural panorama around the nest entrance using artificial landmarks (Fig. 1C,D). However, even this enrichment of the visual environment did not induce the performance of pirouettes in C. fortis with pauses directed to the nest entrance, even though experiments showed that the ants learn these landmarks reliably during their learning walks (Fleischmann et al., 2016). It is noteworthy, however, that the artificial landmarks did not enrich the visual panorama comparable to the natural environment of C. noda (Fig. 1A versus B). Therefore, one might speculate that the panorama was not visually enhanced enough to induce the performance of pirouettes. We think this is unlikely, because other desert ants that live in cluttered but less structured environments than the pine forest of the Greek Cataglyphis ants in this study also perform turns and look back to the nest entrance repeatedly during their learning walks (Müller and Wehner, 2010; Wehner et al., 2004). Why is it that C. fortis does not invest any time in stopping and looking back at the nest entrance? As its natural habitat usually does not offer a prominent panorama and all directions look alike, it may not be worth making distinct snapshots of the panorama, as has been suggested for the Namibian desert ant O. robustior (Graham et al., 2010). These ants only stop once or twice during their back turns (Müller and Wehner, 2010). This suggests a possible correlation between the richness of the visual scene in the natural environment and the number of stopping phases during the frequent turns of the desert ants performing learning walks [i.e. C. fortis: no landmarks (saltpans and chotts) – exclusively voltes without stopping phases, O. robustior: few landmarks (a skyline of distant dunes and acacia trees) – only nest-centered stopping phases during pirouettes; Müller and Wehner, 2010; C. noda and C. aenescens (forest): many landmarks, i.e. a lot of visual information available – two types of turns with several stopping phases during pirouettes]. Further analyses should investigate whether fixation directions are associated with panorama features. Wood ants (Formica rufa), for example, look back to fixate a landmark when becoming familiar with a new feeder (Nicholson et al., 1999). Therefore, it is possible that desert ants might fixate prominent landmarks during the other stopping phases. However, systematic fixations of specific landmarks could not be observed by qualitative judgments, whereas the look back to the nest entrance behavior was as obvious in the Greek Cataglyphis species as in the Namibian O. robustior.

Besides the question of the purpose of the other stopping phases during the pirouettes, there are two other pressing questions: (1) what enables the ants to look back precisely towards the nest entrance?; and (2) why do Cataglyphis desert ants perform different types of turns? Concerning the first question, it has been proposed that the main navigational tool, i.e. path integration, offers the reference system needed to precisely gaze at the nest (Graham et al., 2010; Müller and Wehner, 2010). Further experiments need to be conducted to verify that path integration enables the Greek Cataglyphis ants in their landmark-rich habitat to turn back to the nest, as opposed to another mechanism, such as landmark guidance. Concerning the second question, it may be that there are anatomical reasons for performing different types of turn (Wehner, 1994). However, the behavioral trait to perform pirouettes is independent from morphological traits enabling different desert ant species to raise their gasters to improve their mobility (McMeeking et al., 2012). This indicates that the habitat, rather than anatomical constraints or phylogenetic relationships, exerts a selection pressure on the performance of the different types of turns.

Therefore, the second question has to be reformulated: what might be the function of the voltes performed exclusively (C. fortis) or in addition to other turns (C. noda, C. aenescens)? As mentioned above, the main navigational toolkit of Cataglyphis is path integration (Ronacher, 2008; Wehner, 2008). To determine the direction of their home vectors, the ants mainly rely on celestial cues, especially the polarization pattern and the azimuthal position of the sun (Wehner and Müller, 2006). In principle, compass information can be deduced from the polarization pattern by either a sequential or an instantaneous method (Wehner and Labhart, 2006). Voltes could provide the rotatory component necessary for the sequential method (Wehner, 1987b, 1994). Moreover, voltes may be used by desert ants to learn the configuration of landmarks close to the nest entrance even though they do not include stopping phases like pirouettes. Regardless, C. fortis ants learn the artificial landmarks surrounding the nest entrance even though they do not perform any pirouettes or other turns with distinct stopping phases (Fleischmann et al., 2016). It is not known how ants – or other animal species – store snapshots or other visual information in their brains. However, it was shown that short-term light exposure at the beginning of the ants' foraging careers has long-term effects on the synaptic architecture in visual subregions of the mushroom bodies, high order sensory integration and learning and memory centers (Stieb et al., 2010, 2012). Similarly, microglomerular synapses in the lateral complex, the last synaptic relay station in the neural pathway of the sky compass are influenced by light particularly in the UV region of the spectrum during first exposure (Schmitt et al., 2016). We hypothesize that visual information gained by different types of turns during learning walks may represent crucial elements triggering neuroplastic calibrations in visual pathways.

Interestingly, circular movements and rotatory motifs are important not only in the learning walks of desert ants but also in the orientation behavior of other arthropods. Learning flights of flying hymenoptera include repeated arcs, loops and turn-backs (honeybees: Becker, 1958; Capaldi and Dyer, 1999; Capaldi et al., 2000; Degen et al., 2015, 2016; Lehrer, 1991, 1993; Opfinger, 1931; Vollbehr, 1975; wasps: Peckham and Peckham, 1898; Stürzl et al., 2016; Tinbergen, 1932; Zeil, 1993a,b; Zeil et al., 1996; bumblebees: Collett et al., 2013; Hempel de Ibarra et al., 2009; Philippides et al., 2013; Riabinina et al., 2014; Robert et al., 2017; Wagner, 1907). Dung beetles perform rotations about their vertical axis before rolling a ball away from the dung pile (Baird et al., 2012), during which they take a snapshot of the celestial scenery (el Jundi et al., 2016). Desert spiders also perform learning walk-like behavior: they leave their burrows in sinusoidal paths when departing to unfamiliar terrains (Nørgaard et al., 2012).

In conclusion, different types of turns performed by desert ants (and other arthropods) during learning walks (or flights) are likely to serve different functions. Pirouettes during which the desert ants look back to the nest entrance are only performed by ants inhabiting cluttered environments (C. noda and C. aenescens). The stopping phases during pirouettes are most suited to taking snapshots (Graham et al., 2010; Müller and Wehner, 2010), suggesting that the ants take snapshots of the distant panorama around the nest entrance. Additionally, all Cataglyphis species investigated so far performed voltes, which may provide the rotational movement needed to successfully calibrate the celestial cues as compass tools or serve other navigational purposes like memorizing the configuration of nearby landmarks.

We thank the Tunisian and the Greek governments for the permits to conduct our experiments in these two wonderful home countries of Cataglyphis. We especially want to thank the management board of the Schinias National Park for permission to perform our research in the Schinias National Park in Marathon. Furthermore, we are deeply grateful for the indispensable support on site by Maria Trivourea and Olga Papigioti. We also want to thank Christos Georgiadis for introducing us to the Greek Cataglyphis ants and for his help with administrative issues. In addition, Christos made contact with two students from the University of Athens, Thanos Tsiatouras and Tom Zisis, who kindly enriched our field team in 2016 and helped to record the learning walks of C. aenescens. We also want to thank all members of the field teams of 2014, 2015 and 2016, without whom the experiments could not have been performed in this way. We thank Bernhard Ronacher, who supported our research excursions with his expertise concerning Cataglyphis fieldwork and by lending us his VW bus, Jochen Zeil for help with data analysis and the introduction to the Digilite application, for many discussions during the project and for his remarks concerning the manuscript, and Martin Strube-Bloss for helping to solve Matlab problems. We thank two anonymous reviewers for their very helpful comments.