Experiment 1 and 2
Different groups of ants were trained in one of three different training paradigms (Figure 1A–C). These consisted of walking within a channel system either to a feeder at ground level (flat training; Figure 1A), or to an elevated food source ("ramp training", Figure 1B), or to pass a hill while being trained to a feeder at ground level (Λ training, Figure 1C). This last training was supposed to result in the same global vector between feeder and nest as in the flat training, but to include a vertical component in the path, as in the ramp training.
The numbers of passes and descents that ants made during their homebound runs are classified according to the ramp positions in Figure 3. Only the ramp training resulted in a non uniform distribution of descents between the ramps on offer, caused by an extraordinarily high number of descents at ramp No. 6, the first decision point for a returning ant (Figure 3B). In the case of flat and Λ training (Figure 3A,C), the overall numbers of full descents were too small to identify heterogeneous distributions of the ants' choice between ramps. As can be seen by the length of the bars in Figure 3, the number of total decisions made at the ramps decreased markedly from ramp 6, which was the first one encountered on the homebound trip, towards ramp 1. This was particularly true for the ramp and Λ training situations, where a high proportion of ants descended the full length of the first ramp that they encountered, thus ending their respective test runs.
If we applied a softer criterion by comprising all descents that went further than 20 cm in the "descent" category, this did not change the outcome considerably (Additional file 1). Please note that the total numbers of decisions for each ramp do not correspond exactly between the two criteria. The underlying reason is that under the two criteria the analysis of an ant's run ended at different points. Consider for example an ant that chooses on its first descent to climb down a ramp for 50 cm. According to the "hard" criterion (depicted in Figure 3), this incomplete descent was considered as a decision against this respective ramp, counted consequently as a pass, and the ant's path would be analyzed further. Under the "soft" criterion however (documented in Additional file 1), a descent of 50 cm (i.e., of more than 20 cm) was seen as a decision in favor of this ramp, counted as a descent, and the run was considered to be terminated.
After flat training, ramp choice was distributed homogenously (p > 0.1, χ2 homogeneity test and squared standardized residuals). Ramp training resulted in a heterogeneous distribution (p Λ training was still not sufficient to detect heterogeneity.
Tests in a flat channel
If the ants operate exclusively in 2-D (hypothesis A), they should always search for their nest at the correct ground distance when released for their homebound run in a flat test channel – irrespective of the training paradigm. If, on the other hand, Cataglyphis couples the memory of an ascent or descent with certain values of her home vector, one could expect an increased search density at a distance corresponding to the position of the ramp during training. This would be akin to the "procedural knowledge" that could be demonstrated in 2-D experiments [19]. In order to check these predictions, we trained ants to a feeder, following the same three paradigms as described above (flat, ramp, Λ). Ants were then taken from the feeder and placed into a flat channel, and the first U-turns of their homebound runs were recorded.
The distances at which ants begun their search differed between the training paradigms "flat", "ramp", and "Λ" (Figure 4). Flat training (median of first U-turns: 11.7 m) resulted in search distances that came closest to the relative position of the nest, which corresponded to the 12 m mark in the test channel. All other training situations resulted in animals displaying shortened homing distances. This "undershooting" was most pronounced after Λ training (median of first U-turns: 8.3 m). After ramp training, the animals also carried out a slightly truncated search for the nest (median of first U-turn: 10.45 m), although a larger scatter leads to non-significant differences to the results from flat training. In order to unequivocally decide whether the ants searched for the nest or for the beginning of the descent, we included an additional training, in which the ramp was located at 9 m distance from the nest. Under this condition, an ant had experienced on its homebound run the beginning of the ramp already 2.5 m after the feeder. As Figure 4 shows, the centers of search (median for "ramp at 9 m": 10.6 m) did not differ between the two forms of ramp training, i.e. they were not influenced by the position of the ramp.
Experiment 2: Outbound tests
After training in almost identical fashion to the Homebound test (see the Materials and Methods section), we tested ants that were on their outward trip from the nest to a feeder that they had frequently visited before. Individual ants were given access to a test channel by means of a switch near the nest entrance. The animals had six ramps at which they could choose either to ascend or to pass through a central gateway (Figure 1E, inset). As in the Homebound test, the type of training that the ants underwent prior to testing had a marked influence on the number of individuals that stepped onto a ramp and continued to climb on it for more than 20 cm (Figure 5A). Fewer individuals chose to climb up after flat training than after ramp or Λ training.
The distribution of distances that ascending ants walked on the ramps before turning around for the first time mirrors the results of the Homebound tests (compare Figure 2B with Figure 5B). After flat training, most of the ants' U-turns were located in the lower half of the ramps (Figure 5B; median = 30 cm). In contrast, almost all of the ascending ants from ramp training and the majority of animals that underwent Λ training, climbed the full length of the ramps (Figure 5B; both medians = 150 cm). A statistical comparison of the ascent distances confirmed strong differences between all three training paradigms.
The distribution of ascents over the six ramp positions is shown in Figure 6. In the case of flat training (Figure 6A), only 3 out of 370 total decisions of ants at the different test ramps resulted in a complete ascent, prohibiting any conclusions about their distribution.
In spite of the high relative value of ascents at ramp No. 6 (the ramp furthest away from the nest), where 6 out of 26 decisions resulted in climbing up, ramp training did not cause a heterogeneous distribution in the number of ascents compared to passes (Figure 6B). In the case of Λ trained ants, we could also observe a trend to prefer ramp No. 6 (Figure 6C). However, the overall number of ascents was also too small to confirm a heterogeneous distribution. Here too, we applied a softer criterion for defining ascents, in order to rule out that our strict definition of ascents masked any effects of the different training situations, and considered all ascents that continued for more than 20 cm (Additional file 2). Now, the data retrieved from all training paradigms was sufficient to carry out a statistical analysis of heterogeneity. In the case of flat and ramp training, no preference or rejection of any ramp could be detected (p > 0.05, χ2 homogeneity test). In Λ training however, the trend observed under the strict criterion could be confirmed, with an over proportional number of ants climbing up on ramp No. 6 (p χ2 homogeneity test and squared standardized residuals).
Experiment 3: Induction of a negative vertical vector component
If the ants do acquire a true 3-D vector (hypothesis C), it should be possible to specifically and separately influence the vertical component of the 3-D vector. The rationale of this experiment was to train ants essentially in a flat channel to a feeder on level ground (Figure 7A, "Training"). After several visits at the feeder, ants were transferred for the critical experiment to the elevated end of the test channel (Figure 7A, "Critical test"; "R" marks the release point), so that on their homebound run they first experienced a descent. At the nest position, the test channel ended in an ascending ramp (2 m long), and the length of ascents on it was recorded. If during the "enforced" descent within the test channel the ants had built up a negative vertical vector, they should be eager to ascend again on the test ramp. As a control, a different group of ants was released on level ground within the test channel at the same distance from the nest as in the critical test (Figure 7A, "Control test"; "R" marks the release point). Note that the training was always completely flat. Hence, according to the results gained from flat-trained animals in experiments 1 and 2, the expectation was that ants in the control should not ascend very far on the ramp located at the nest position. In order to facilitate the descents in the critical test, in this experiment both the training and the test channels were lined with landmarks that were inconspicuous on the outbound runs, but conspicuous on the homebound runs (see Methods section).
The results are very clear (Figure
7B): As expected, ants that were trained and tested in the flat channels did not ascend very high on the test ramp (median: 65 cm). Note that in this test channel the ants had no opportunity to continue their path in the horizontal plane (as in experiment 1 or 2), but were forced to continue on the ascending ramp. This probably explains the larger ascent height, as compared to Figure
5B. Most importantly, however, the median ascent height did not differ between the critical test (with forced descent, N = 37) and the control (without descent, N = 39). This shows clearly that the ants did not accumulate a negative vertical vector component during this "enforced" descent, which would have prompted them to compensate for this vertical vector proportion on the ascending ramp. This experiment renders a vertical vector component, and thus a true 3-D vector, very unlikely.
Answers to all your Biology Questions
