Ants regulate colony spatial organization using multiple chemical road-signs

Communication provides the basis for social life. In ant colonies, the prevalence of local, often chemically mediated, interactions introduces strong links between communication networks and the spatial distribution of ants. It is, however, unknown how ants identify and maintain nest chambers with distinct functions. Here, we combine individual tracking, chemical analysis and machine learning to decipher the chemical signatures present on multiple nest surfaces. We present evidence for several distinct chemical ‘road-signs' that guide the ants' movements within the dark nest. These chemical signatures can be used to classify nest chambers with different functional roles. Using behavioural manipulations, we demonstrate that at least three of these chemical signatures are functionally meaningful and allow ants from different task groups to identify their specific nest destinations, thus facilitating colony coordination and stabilization. The use of multiple chemicals that assist spatiotemporal guidance, segregation and pattern formation is abundant in multi-cellular organisms. Here, we provide a rare example for the use of these principles in the ant colony.

which fewer stimuli are present are not statistically likely. We first test the null hypothesis that none of the chamber surfaces obtained any special characteristics during the priming period.
Under this assumption, floor segments cannot be distinguished by the ants (see "Permutation test" in the Methods). Therefore, using the ants' locations to generate a guess regarding which of the four new nest locations {N,S,E,W} contains the Q floor segment (i.e. the floor segment on which the queen resided before the manipulation) would only succeed with a chance of 25%. The chance that the correct assignment is achieved over six independent trials would be (weight of the tail of the corresponding Binomial distribution). Our experimental measurements enables locating the chamber which contains the Q floor segment in six out of six trials. This procedure that yields this is simple: of the 24 possible permutations, the six top scoring permutations were those in which chamber Q was strongly associated with a single location in the nest (this includes all six permutations in five experiments and four out of six permutations in one experiment). In all six experiments, the Q floor segment was indeed shuffled into this location. Again, the chances for this to happen randomly are very small.
In fact, one can reject the null hypothesis with even higher statistical significance. In all six experiments, among all possible permutations, σ, the one that corresponded to the actual experimental permutation, ∑, ranked within the top 12.5% (SI figure 2, "Permutation test" in the Methods). If the probability of returning to a chamber was completely independent of floor characteristic then the actual experimental permutation would be no different than any other randomly generated permutation. In this case, the chances that the correct association appears in the top 12.5% (or 1/8) of all permutations would simply be 1/8. The probability that this would happen by chance six time is, therefore, . We can, therefore, reject the null hypothesis that ants do not rely on floor characteristics when they return to the nest.
The previous results provide strong support for the fact that at least one type of stimulus, obtained during the priming period, affects the ants' spatial distribution in the nest after the manipulation.
Next we show that, given our experimental results, a single stimulus is also unlikely. To do this, we assume four hypothetical situations and then refute them one by one.
 The first case is the one in which the same single stimulus is present on all four floor segments. In this case, the segments are indistinguishable from each other. This is equivalent to the case of no stimuli at all which we have already negated in the previous paragraphs with .
 The second case we consider is the one in which a single stimulus is present on two of the floor segments. In this case, there are two pairs of floor segments: one pair contains the stimulus and the other does not. By the null hypothesis, the two floor segments belonging to the same pair are indistinguishable by the ants (otherwise, there would have been two distinct stimuli in contrast to our assumption). We focus on the pair which contains the floor segment corresponding to chamber Q. Assume, without loss of generality, that these are floor segments {Q,1} and that they were inserted into locations {N,W}. We expect that after the manipulation these two segments would be equivalent from the point of view of the colony. In other words, the spatial distribution of ants after the manipulation could not hold any information regarding which of two chambers {N,W} contains floor segment Q. Any assignment thus has 50% chance of being correct.
However, as shown above, in six out of six times it is actually possible to correctly accurately identify the nest chamber that contains the queen's floor segment Q. The probability that this happen, if the null hypothesis holds, is (weight of the tail of the corresponding Binomial distribution).
 The third case we consider is the one in which only one floor segment (Q) contains a stimulus while all other chambers contain no stimulus. To verify that the queen's chamber is not the only source of order in the system we removed the corresponding floor segment (and its location in the new nestwithout loss of generality we name it E) from the analysis and repeated the permutation procedure described above. The goal is now to associate floor segments {1,2,3} with their locations in the new nest {N,S,W}.
Again, we find that the permutation, σ, which is consistent with the experimental manipulation, ∑, ranks high (see "Permutation test" in the Methods) .In fact, it is ranked highest in four cases, and second and third in one case each. Similar to the calculation in the previous paragraph the probability that such high ranking would be achieved at random given the null hypothesis of just one floor segment containing a stimulus can be computed by adding the probabilities to reach these ranks or a better combination of ranks. Specifically, we calculate the probabilities to reach the combinations of ranks shown in SI table 1.
 The final possibility is the one in which one stimulus exists and is present in three out of the four floor segments. In this case the queen's segment must be the only one that does not contain the stimulus while the other three chambers are indistinguishable. This is precisely equivalent to the case of one stimulus in a single chamber, a case which was refuted above.
To summarize our results show that the null hypothesis stating that no information is transferred in the floor segments can be rejected with probability Moreover, the second null hypothesis that a single type of stimulus is present on some or all of the floor segments can also be rejected. A p-value for each possible case was presented above and the highest of these is p < 2e-2. Therefore, our results support the fact that, following the priming period, at least two stimuli are present in the floor segments that are distinguishable by the ants and affect their final distribution in the new nest.

Extraction and analysis method performance
We used a linear hydrocarbon mix to assess the SNR, reproducibility, and linearity of this method (SI figure 5). Samples were analyzed on a 7890 Agilent gas chromatograph equipped with a fused silica column (DB5-MS 30 m × 0.25 mm × 0.25 µm, Agilent) and coupled to an FID as described in section "GC-FID analysis" in the Methods. The SNR, per chromatogram peak, was defined as the ratio between the measurements' mean value to the blank signal mean value (taken over all relevant peaks). We find a mean SNR of 1.4e5 for 6 ng of standard weight on a surface of 20 cm 2 (SI figure 5a). Reproducibility was quantified by calculating the coefficient of variation (standard deviation divided by the mean) over three replicate measurements (SI figure 5b). As an example, for 6 ng of standard weight on a surface of 20 cm 2 for alkanes with 15 carbons or more yields (N=25) a median coefficient of variation of 0.065. Finally, our measurements exhibit a linear relation between the amount of analytes and the output GC signal (SI figure 5c) this relation persists over a wide range of total injected mass (low discrimination, SI figure 5c inset).

Task group distribution in asymmetrical nests
The artificial nest structure used for the chemical characterization of nest floors was chosen since it induces a clear association between different chambers and the task groups that occupy them.
Indeed, the more internal chambers are occupied by the queen, the brood, and workers with distended abdomen (the intersegmental membranes become visible giving the abdomen a striped appearance) (72% ± 9%, N=60 time points in 3 different experiments) that are characteristic of corpulent nurses, whereas the entrance chamber is occupied by non-striped ants (86% ± 6.5%,

N=60 time points in 3 different experiments) associated with lean foragers.
The correlation between body weight and task group affiliation has been previously demonstrated (48). To verify the association between corpulence and stripe pattern the body weight of striped workers 19.3±3.6mg (N = 10) was compared to that of plain workers 9±1.3mg (N = 10).

Silica plates cleaning procedures
Silica on glass thin layer chromatography (TLC) plates (Analtech) were cut to 6x5 or 3x5 cm2 and placed in one layer in a glass baking pan (Pyrex). The plates were subsequently cleaned using: ethyl acetate, hexane and acetone in the following manner: the plates were covered in solvent and heated to the boiling point of the solvent for 5 minutes. The plates were then transferred to a clean baking pan and the same procedure was repeated with the next solvent.

Ants which are associated with the queen's chamber show higher fidelity
Out of the ants that spent over 70% of their time inside the nest 90 where strongly associated with the queen's chamber and 43 with a different chamber. The fidelity of each ant to the floor segment she occupies was evaluated using the manipulation described in Results section "Nest surfaces affect ant spatial organization". Each ant was given a score that is a measure of her fidelity to her floor segment (For details see Methods section "Permutation test"). Ants associated with the queen's room had a mean score of 0.8±0.28 which is significantly different from the scores of the other group 0.4±0.3 (Kolmogorov-Smirnov test, p<4e-10). This increased fidelity to the queen's chamber may be attributed to a stronger attraction of the relevant ants to the scents associates with the queen's chamber or perhaps to a tendency of ants to remain for longer periods in denser areas. It cannot be attributed to the attraction of workers to the queen itself since, the queens were not present at the stage in which the ants were reintroduced into the nest.

Supplementary Note 2 Subterranean road signs can be removed by hexane
The raw data for of the experiments described in section "Hexane soluble compounds as subterranean road signs" of the main text is shown the tables below.  For details see section "Hexane soluble compounds as subterranean road signs".  For details see section "Hexane soluble compounds as subterranean road signs".

Supplementary Note 3 Silica plates' floorings induce the same behavior as paper floorings
Silica plates were chosen over paper for the chemical experiments because of their low background noise. In this section we verify that silica plates' floorings accumulate chemical orientation signatures that are recognized by the ants. This is done in order to show continuity between the behavioral experiments presented in section "Nest surfaces affect ant spatial organization", which use paper floorings and the chemical experiments presented in "Classifying chamber function by its chemical signature", which use silica plates. To do this we housed colonies (N = 4) in symmetric four chambered nests containing silica floorings in which only one chamber was accessible to the ants. After five days the ants were removed from the nests and the silica floorings were transferred to new nest structures in which all chambers were accessible.
The ants were then introduced to the new nests and the number of ants in each chamber was counted after 24 hours. In all experiments (N = 6) the chamber that contained the largest fraction of ants was the chamber that was originally accessible. This implies that the silica plates floorings contain an orientation stimulus (p< 2e-04, by the weight of the tail of the corresponding Binomial distribution).

Supplementary Note 4 Refuse area floor extraction
Colonies were placed in artificial nests with silica floorings where they created refuse piles which were randomly placed along the walls of the inner chambers (see SI figure 4a). In fact over the 16 experiments we conducted the refuse pile was in the same location only twice. After 5 days, we transferred the colony to a new nest to which a hexane extract taken from the original refuse area was added to a 2-3 cm strip randomly chosen near one of the inner chambers walls (total of 64 cm of nest wall). The ants recreated the new refuse pile on top of the hexane extract in 4 out of 16 repeats (see SI figure 4). Although this is not a high percentage, if one considers all possible locations of the refuse pile, the signal is significantly above random (p< 0.005).
Supplementary figure 4: ants react to refuse pile extraction. a) Refuse pile natural distribution overlay where the refuse pile location of each experiment is uniquely colored b) Refuse pile recreated on top of its floor extraction.