SUPPLEMENTARY INFORMATION

1. Control Tests for Effect of RFID Tags

Methods

Using three additional colonies, we tested the effect of RFID tagging on ant behaviour. We tagged 70% of the workers and observed behaviour within the nest. To test effects on mobility, for each colony we removed external workers (three tagged and three non-tagged) and transferred the workers one by one to a glass arena with a floor marked out in 5mm x 5mm squares. After 1 minute for acclimatisation, ant movement was video recorded (Panasonic NV-DS55B) from directly above the arena for 1 minute. The video was analysed using Adobe Premier Pro (©1991-2000 Adobe Systems Incorporated) and Ant Track software which allows the location of an ant to be pinpointed every 0.8 seconds. From this the mean speed of each ant over 1 minute was calculated. We also calculated the sinuosity of the path using the ratio between the distance from the starting point to the end point and the total path length travelled. This index ranges from 0 to 1, with 1 representing a completely straight path. 95% confidence intervals are given for p-values of non-significant tests (Colegrave and Ruxton 2003; Colegrave and Ruxton 2005).

Results

Tagged ants were observed carrying out usual tasks inside the nest (including self- and allogrooming, brood tending, brood carrying, queen tending, trophallaxis, debris removal) and outside the nest (including exploring arena, collecting protein, sugar solution, water). During emigration into new nests, tagged ants were again involved in all the usual tasks, including exploring the new nest, leading and following tandem runs, carrying other workers and brood, and being carried. No differences between the behaviour of tagged and un-tagged ants were apparent.

Running speed was not affected by the presence of an RFID tag (Supplementary Fig. 1). Tagged mean speed 13.0mms-1 (range 6.3-19.4mms-1); non-tagged mean speed 13.3mms-1 (range 7.2-16.0mms-1) t-test: t=0.17, df=15, p=0.87 (95% Confidence Intervals for p: 0.75-1 for the observed difference between means of 0.29mms-1).

Path sinuosity was also unaffected by the presence of an RFID tag (Supplementary Fig. 1). Tagged ants path sinuosity mean± SD 0.24 ± 0.08; non-tagged ants path sinuosity mean± SD 0.23 ± 0.09, t-test: t=0.24, df=15, p=0.82 (95% Confidence Intervals for p: 0.64-1 for the observed difference between means of 0.092).

Supplementary Figure 1. Boxplots of speed and sinuosity of path of tagged and non-tagged ants. Boxes contain interquartile range; horizontal lines indicate median value, w indicate mean value, * indicates outliers.

We also used the data collected in the main experiment to compare the final corpulence measurements of ants which had retained their tags versus those which had lost them during the experiment. There was no significant difference between the gaster dry weights of tagged ants (mean ± SD: 0.076 ± 0.03mg) compared to non-tagged ants (0.072 ± 0.03mg), t-test t=1.33, df=657, p=0.184. Note that the tags were affixed to the thorax, so the tags and glue do not confound corpulence measurements on the gaster.

Our data do not indicate any effect of RFID tagging on speed of movement, directness of path or corpulence. They also do not indicate any effect of corpulence on likelihood of retaining RFID tags. Finally, it should be noted that our analyses in the main experiment are comparisons among the tagged ants, so any minor effects of tagging (if any occur) would be expected to be expressed across the board and therefore would not be expected to influence the results of this study.

2. Control Experiment: Weight

To control for the weight loss of individuals held in individual vials during the removal period, a control experiment was carried out using a further six colonies. All ants which were outside the nest were collected over a 10 minute period. Each ant was placed into an individual vial, and allocated at random to one of two treatments. Treatment 1 ants were immediately frozen; Treatment 2 ants were kept in vials at room temperature for 60 hours then frozen. For both treatments after 2 days in the freezer the gasters were weighed according to the same protocol as in the main experiment.

There was a significant difference between the two treatments (ANOVA: F=6.75, df=1,156, p<0.05) with the ants isolated for 60 hours (treatment 2) having a mean gaster weight of 12% ±2.8% below that of ants frozen immediately (treatment 1). We used this control data to apply a correction to our weight data, increasing the gaster mass of ants isolated for the full 60 hours by 12% and applying a conservative linear function for ants isolated for shorter periods. Applying a different correction (exponential decay functions or a step function) made no qualitative difference to the outcome of the GLMMs. All subsequent weight data are corrected using the linear function.

3. Calculating Trip Intervals

As we could not distinguish between exits and entrances, treating all exits and entrances equally could cause apparent clustering of activity, if periods outside the nest were usually much longer or much shorter than periods in the nest. To avoid this problem, a set of three consecutive RFID readings for the same individual was taken to make up a complete trip. Intervals therefore include the duration of an external trip, and the period in the nest before the next trip, in either order (see Supplementary Figure 2).

Supplementary Figure 2. For each ant the data could be grouped in two ways, set 1 starting with the first RFID reading (AB-AB-AB-AB); set 2 starting with the second reading (BA-BA-BA-BA).

Individual ID / Colony / t / df / p / Significance at 5% / Significance at 0.2%
2e / E / 0.18 / 42 / 0.86 / NS / NS
46 / B / 1.51 / 6 / 0.18 / NS / NS
58 / A / 0.13 / 40 / 0.90 / NS / NS
61 / A / 0.04 / 21 / 0.97 / NS / NS
64 / A / 0.16 / 46 / 0.88 / NS / NS
7c / A / 0.00 / 29 / 1.00 / NS / NS
a2 / C / 2.23 / 4 / 0.09 / NS / NS
104 / E / 0.55 / 22 / 0.59 / NS / NS
107 / A / 1.49 / 38 / 0.14 / NS / NS
15c / B / 0.55 / 5 / 0.61 / NS / NS
1a4 / C / 0.38 / 15 / 0.71 / NS / NS
1ea / A / 0.06 / 34 / 0.95 / NS / NS
1f2 / B / 3.00 / 7 / 0.02 / ** / NS
275 / A / 0.00 / 49 / 1.00 / NS / NS
281 / E / 3.17 / 3 / 0.05 / NS / NS
2a4 / C / 0.60 / 6 / 0.57 / NS / NS
2b0 / B / 0.33 / 18 / 0.74 / NS / NS
2e3 / D / 0.00 / 12 / 1.00 / NS / NS
310 / B / 0.22 / 3 / 0.84 / NS / NS
353 / D / 0.23 / 2 / 0.84 / NS / NS
38c / B / 0.21 / 26 / 0.84 / NS / NS
38e / A / 0.00 / 33 / 1.00 / NS / NS
3c2 / A / 0.14 / 49 / 0.89 / NS / NS
3d4 / A / 0.02 / 41 / 0.99 / NS / NS
3eb / B / 0.17 / 28 / 0.86 / NS / NS

Supplementary Table 1: Results of comparison between gradient coefficients of set 1 and set 2 data for each ant are given, with significance before and after the Bonferroni correction (α=0.2%). All individuals for which 6 data points or more are available are included.

There were no differences between these two ways of pairing the data for each ant, so for further analyses, set 1 data were used for all ants.

4. Distribution of Ant Weights

Pooled weight data from all ants in all 5 colonies fit a Weibull distribution (Anderson Darling 2.42, R=0.99) with parameters:

β (shape) =1.908

α (scale) = 0.06812

c (threshold) = 0.01448

Figure 4 in main document illustrates the weight distribution.

References:

Colegrave N, Ruxton GD (2003) Confidence intervals are a more useful complement to nonsignificant tests than are power calculations. Behav Ecol 14:446-450

Colegrave N, Ruxton GD (2005) What hypothesis tests are not: a reply to Johnson. Behav Ecol 16:325

1