Supplementary Methods
Fly Stocks:
Canton-S and Oregon-R wild-type flies were obtained from J. C. Hall (Emeritus at
Brandeis University, Waltham, MA). Auditory-impaired iav1 flies were from the
Bloomington stock centre (#6029). Orco2 was obtained from L. Vosshall (The
Rockefeller University, New York, NY) and established in a Canton-S background.. The
taste-deficient mutant (w ΔXBs6; PoxnΔM22-B5) and its control (w; PoxnΔM22-B5 SuperA-207-2)
were obtained from M. Noll (University of Zurich, Zurich, Switzerland).
Detailed Methods:
Generation of ‘null’ datasets:
For each treatment group, ‘null’ trials were constructed by selecting 12 trials at random.
From each of those 12 trials, one fly was chosen at random, and their entire trajectories
were normalized in space (to compensate for any image distortion during acquisition
relating to cameras/lenses) and combined.
Establishing social space:
The angle and distance between each fly and all other flies' centre of mass is established
for all trials within a treatment and 500 'null' trials of the same treatment. The maximum
distance was set to 20 body lengths. It is important to set the distance to a large enough
value that the distance and angle frequencies reach their peak (see Figure S1). These
frequencies are normalized, and then the angles are summed per distance bin. It is
important to note that we normalize the two dimensional angle-distance prior to summing
and subtracting as we see a non-uniform distribution in the ‘null’ dataset that would not
be accounted for by simply calculating distances. Subtracting the ‘real’ distances from
the ‘null’ distances creates the two dimensional bar graph seen in Figure 1A.
Determining the interaction values:
Working within the social space:
1) Generate an estimate of the spatial heat map by subtracting the ‘null’ distribution
from the ‘real’ heat map.
2) Set all negative values of the spatial heat map from 1) to zero (negative values
represent configurations that appear more frequently in ‘null’ trials).
3) Identify ‘hot’ locations where the distribution is greater than the 3rd quartile of all
positive values (to ensure their inclusion in the interaction criteria).
4) Eliminate hot spots identified in 3) that are not connected to the main grouping of
‘hot’ locations by 2 bins (to eliminate outlier locations).
5) Create an initial estimate of criteria by calculating the minimal angle and distance
bins that are inclusive to the connected ‘hot’ spots.
6) Refine the initial estimate by increasing values (distance, angle, or both) if the
increase of that value increases the mean of the included bins.
7) Repeat 6) until no further increase in a value increases the mean of the included
bins.
Determining the cut-off time of an interaction:
All potential interactions satisfying the distance and angle criteria in both the ‘real’ and
‘null’ datasets were flagged. We note that for both ‘real’ and ‘null’ datasets, what we
term ‘encounters’ (potential interactions lasting between 0-0.1 seconds) were necessarily
set to 1 following normalization (as they were cumulative distributions), but the shape of
the distribution differed, allowing us to define potential interaction times that showed
signs of being actively regulated by the flies.
Repeat these steps (establish null dataset, establish social space, establish interaction
values, establish the cut-off time) 500 times, choosing a different subset of 15 trials each
time.
Reliability of the median measurements:
Estimates of interaction values were first standardized (converted to z-scores). For each
‘iteration’ estimate, the appropriate number of estimates was randomly chosen and their
median was taken. This was re-done 100 times per treatment, and then all treatments
were combined. When treatments failed to generate sufficient estimates (i.e. poxnΔXBs6
which generated less than 100 valid estimates), they were not counted for higher
‘iteration’ estimates.
Figure S1. Generating non-social or ‘null’ datasets. For each ‘null’ dataset, 12 trials
were selected at random. From each of those 12 trials, one fly was chosen at random, and
their entire trajectories were combined. These null datasets therefore control for flies
moving within our arena in the presence of 11 other flies, but remove any social cues.
Once we have used these datasets to generate interaction values, we can ‘flag’ when
putative interactions occur in both ‘real’ and ‘null’ datasets and compare their duration
(putative interactions are identified with boxes). Reproduced from Schneider et al. [1].
Figure S2. Establishing the social distance and visualizing the sample dependency in
Canton-S males. (A-B) The angle and distance between each fly and all other flies'
centre of mass is established for (A) all trials and (B) 1000 'null' trials out to 20 body
lengths. These frequencies are normalized, and then the angles are summed for each body
length to create the two dimensional bar graph seen in Figure 1A. (C-J) Visualizing the
effect of sample size on the estimation of the interaction space. We first establish the
angle-distance frequencies and plot them, and then subtract the angle-distances
frequencies for the ‘null’ trials. Then the angles and distances that are less than the 3rd
quartile are eliminated and the result plotted. (C-D) Example of a single trial. (E-F) 5
trials chosen at random. (G-H) Ten trials chosen at random. (I-J) All 43 trials together.
10 trials appear to be the minimum re-sampling number that generates fair agreement
with the maximum sample size of 43 male Canton-S.
Figure S3. Reliability of the estimate of the median versus number of iterations
performed. There is no large benefit to performing more than 100 iterations for the
estimate of characterizing social interactions. (A) Estimates of social distance converge
slightly after 300 iterations. (B) While there is a slight improvement in reliability between
325-450 iterations, it is not a robust convergence as variability increases again after 475
iterations. (C) Estimates of angle are slightly more variable after 375 iterations. (D)
Estimates of time durations for interactions remain constant from 75 iterations onwards.
Qualitatively this suggests that there is no benefit to more than ~100 iterations.
Figure S4. Establishing female Canton-S's interaction space based on repeated
spatial-temporal behaviour. This figure presents an illustration of the method using the
entire dataset of Canton-S females (n=26). (A) The social distance is identified per group
based on the large over-representation of close fly-fly distances in ‘real’ compared to
‘null’ data (see Figure S1). The red dotted line indicates the social distance cut-off (see
methods). (B-C) The angle and distance between each fly and all other flies' centre of
mass is established for (B) all trials and (C) 500 'null' trials (see methods). Numbers
around heatmap indicate angle. (D) The normalized frequency from the ‘null’ dataset is
subtracted from the normalized frequency of the ‘real’ dataset, which reveals spatial
positions seen more often in our assay than one would expect from non-social organisms.
The angle and distance that captures the majority of this over-representation is
established and plotted in red (see supplementary methods). (E) Using the angle and
distance criteria from (D), we count the number of interactions that last at least a
specified time duration. The normalized histogram of ‘null’ data subtracted from the
histogram of ‘real’ data is plotted, and the first positive time bin (red dotted line)
indicates the time duration for which the putative interactions occur more often in the
‘real’ data.
Figure S5. Establishing male Oregon-R’s interaction space based on repeated
spatial-temporal behaviour. This figure presents an illustration of the method using the
entire dataset of Oregon-R males (n=29). (A) The social distance is identified per group
based on the large over-representation of close fly-fly distances in ‘real’ compared to
‘null’ data (see Figure S1). The red dotted line indicates the social distance cut-off (see
methods). (B-C) The angle and distance between each fly and all other flies' centre of
mass is established for (B) all trials and (C) 500 'null' trials (see methods). Numbers
around heatmap indicate angle. (D) The normalized frequency from the ‘null’ dataset is
subtracted from the normalized frequency of the ‘real’ dataset, which reveals spatial
positions seen more often in our assay than one would expect from non-social organisms.
The angle and distance that captures the majority of this over-representation is
established and plotted in red (see supplementary methods). (E) Using the angle and
distance criteria from (D), we count the number of interactions that last at least a
specified time duration. The normalized histogram of ‘null’ data subtracted from the
histogram of ‘real’ data is plotted, and the first positive time bin (red dotted line)
indicates the time duration for which the putative interactions occur more often in the
‘real’ data.
Figure S6. Establishing female Oregon-R’s interaction space based on repeated
spatial-temporal behaviour. This figure presents an illustration of the method using the
entire dataset of Oregon-R female (n=23). (A) The social distance is identified per group
based on the large over-representation of close fly-fly distances in ‘real’ compared to
‘null’ data (see Figure S1). The red dotted line indicates the social distance cut-off (see
methods). (B-C) The angle and distance between each fly and all other flies' centre of
mass is established for (B) all trials and (C) 500 'null' trials (see methods). Numbers
around heatmap indicate angle. (D) The normalized frequency from the ‘null’ dataset is
subtracted from the normalized frequency of the ‘real’ dataset, which reveals spatial
positions seen more often in our assay than one would expect from non-social organisms.
The angle and distance that captures the majority of this over-representation is
established and plotted in red (see supplementary methods). (E) Using the angle and
distance criteria from (D), we count the number of interactions that last at least a
specified time duration. The normalized histogram of ‘null’ data subtracted from the
histogram of ‘real’ data is plotted, and the first positive time bin (red dotted line)
indicates the time duration for which the putative interactions occur more often in the
‘real’ data.
Figure S7. Establishing male Canton-S in the dark’s interaction space based on
repeated spatial-temporal behaviour. This figure presents an illustration of the method
using the entire dataset of Canton-S males in the dark (n=28). (A) The social distance is
identified per group based on the large over-representation of close fly-fly distances in
‘real’ compared to ‘null’ data (see Figure S1). The red dotted line indicates the social
distance cut-off (see methods). (B-C) The angle and distance between each fly and all
other flies' centre of mass is established for (B) all trials and (C) 500 'null' trials (see
methods). Numbers around heatmap indicate angle. (D) The normalized frequency from
the ‘null’ dataset is subtracted from the normalized frequency of the ‘real’ dataset, which
reveals spatial positions seen more often in our assay than one would expect from nonsocial organisms. The angle and distance that captures the majority of this overrepresentation is established and plotted in red (see supplementary methods). (E) Using
the angle and distance criteria from (D), we count the number of interactions that last at
least a specified time duration. The normalized histogram of ‘null’ data subtracted from
the histogram of ‘real’ data is plotted, and the first positive time bin (red dotted line)
indicates the time duration for which the putative interactions occur more often in the
‘real’ data.
Figure S8. Establishing iav1/Canton-S s interaction space based on repeated spatialtemporal behaviour. This figure presents an illustration of the method using the entire
dataset of iav1/Canton-S males (n=26). (A) The social distance is identified per group
based on the large over-representation of close fly-fly distances in ‘real’ compared to
‘null’ data (see Figure S1). The red dotted line indicates the social distance cut-off (see
methods). (B-C) The angle and distance between each fly and all other flies' centre of
mass is established for (B) all trials and (C) 500 'null' trials (see methods). Numbers
around heatmap indicate angle. (D) The normalized frequency from the ‘null’ dataset is
subtracted from the normalized frequency of the ‘real’ dataset, which reveals spatial
positions seen more often in our assay than one would expect from non-social organisms.
The angle and distance that captures the majority of this over-representation is
established and plotted in red (see supplementary methods). (E) Using the angle and
distance criteria from (D), we count the number of interactions that last at least a
specified time duration. The normalized histogram of ‘null’ data subtracted from the
histogram of ‘real’ data is plotted, and the first positive time bin (red dotted line)
indicates the time duration for which the putative interactions occur more often in the
‘real’ data.
Figure S9. Establishing iav1's interaction space based on repeated spatial-temporal
behaviour. This figure presents an illustration of the method using the entire dataset of
iav1 males (n=26). (A) The social distance is identified per group based on the large overrepresentation of close fly-fly distances in ‘real’ compared to ‘null’ data (see Figure S1).
The red dotted line indicates the social distance cut-off (see methods). (B-C) The angle
and distance between each fly and all other flies' centre of mass is established for (B) all
trials and (C) 500 'null' trials (see methods). Numbers around heatmap indicate angle. (D)
The normalized frequency from the ‘null’ dataset is subtracted from the normalized
frequency of the ‘real’ dataset, which reveals spatial positions seen more often in our
assay than one would expect from non-social organisms. The angle and distance that
captures the majority of this over-representation is established and plotted in red (see
supplementary methods). (E) Using the angle and distance criteria from (D), we count the
number of interactions that last at least a specified time duration. The normalized
histogram of ‘null’ data subtracted from the histogram of ‘real’ data is plotted, and the
first positive time bin (red dotted line) indicates the time duration for which the putative
interactions occur more often in the ‘real’ data.
Figure S10. Establishing Orco2/Canton-S’s interaction space based on repeated
spatial-temporal behaviour. This figure presents an illustration of the method using the
entire dataset of Orco2/Canton-S males (n=26). (A) The social distance is identified per
group based on the large over-representation of close fly-fly distances in ‘real’ compared
to ‘null’ data (see Figure S1). The red dotted line indicates the social distance cut-off (see
methods). (B-C) The angle and distance between each fly and all other flies' centre of
mass is established for (B) all trials and (C) 500 'null' trials (see methods). Numbers
around heatmap indicate angle. (D) The normalized frequency from the ‘null’ dataset is
subtracted from the normalized frequency of the ‘real’ dataset, which reveals spatial
positions seen more often in our assay than one would expect from non-social organisms.
The angle and distance that captures the majority of this over-representation is
established and plotted in red (see supplementary methods). (E) Using the angle and
distance criteria from (D), we count the number of interactions that last at least a
specified time duration. The normalized histogram of ‘null’ data subtracted from the
histogram of ‘real’ data is plotted, and the first positive time bin (red dotted line)
indicates the time duration for which the putative interactions occur more often in the
‘real’ data.
Figure S11. Establishing Orco2’s interaction space based on repeated spatialtemporal behaviour. This figure presents an illustration of the method using the entire
dataset of Orco2 males (n=23). (A) The social distance is identified per group based on
the large over-representation of close fly-fly distances in ‘real’ compared to ‘null’ data
(see Figure S1). The red dotted line indicates the social distance cut-off (see methods).
(B-C) The angle and distance between each fly and all other flies' centre of mass is
established for (B) all trials and (C) 500 'null' trials (see methods). Numbers around
heatmap indicate angle. (D) The normalized frequency from the ‘null’ dataset is
subtracted from the normalized frequency of the ‘real’ dataset, which reveals spatial
positions seen more often in our assay than one would expect from non-social organisms.
The angle and distance that captures the majority of this over-representation is
established and plotted in red (see supplementary methods). (E) Using the angle and
distance criteria from (D), we count the number of interactions that last at least a
specified time duration. The normalized histogram of ‘null’ data subtracted from the
histogram of ‘real’ data is plotted, and the first positive time bin (red dotted line)
indicates the time duration for which the putative interactions occur more often in the
‘real’ data.
Figure S12. Establishing poxnΔM22-B5SuperA-158’s interaction space based on
repeated spatial-temporal behaviour. This figure presents an illustration of the method
using the entire dataset of poxnΔM22-B5SuperA-158 males (n=21). (A) The social distance
is identified per group based on the large over-representation of close fly-fly distances in
‘real’ compared to ‘null’ data (see Figure S1). The red dotted line indicates the social
distance cut-off (see methods). (B-C) The angle and distance between each fly and all
other flies' centre of mass is established for (B) all trials and (C) 500 'null' trials (see
methods). Numbers around heatmap indicate angle. (D) The normalized frequency from
the ‘null’ dataset is subtracted from the normalized frequency of the ‘real’ dataset, which
reveals spatial positions seen more often in our assay than one would expect from nonsocial organisms. The angle and distance that captures the majority of this overrepresentation is established and plotted in red (see supplementary methods). (E) Using
the angle and distance criteria from (D), we count the number of interactions that last at
least a specified time duration. The normalized histogram of ‘null’ data subtracted from
the histogram of ‘real’ data is plotted, and the first positive time bin (red dotted line)
indicates the time duration for which the putative interactions occur more often in the
‘real’ data.
Figure S13. Establishing poxnΔXBs6's interaction space based on repeated spatialtemporal behaviour. This figure presents an illustration of the method using the entire
dataset of poxnΔXBs6 males (n=20). (A) The social distance is identified per group based
on the large over-representation of close fly-fly distances in ‘real’ compared to ‘null’ data
(see Figure S1). The red dotted line indicates the social distance cut-off (see methods).
(B-C) The angle and distance between each fly and all other flies' centre of mass is
established for (B) all trials and (C) 500 'null' trials (see methods). Numbers around
heatmap indicate angle. (D) The normalized frequency from the ‘null’ dataset is
subtracted from the normalized frequency of the ‘real’ dataset, which reveals spatial
positions seen more often in our assay than one would expect from non-social organisms.
The angle and distance that captures the majority of this over-representation is
established and plotted in red (see supplementary methods). (E) Using the angle and
distance criteria from (D), we count the number of interactions that last at least a
specified time duration. The normalized histogram of ‘null’ data subtracted from the
histogram of ‘real’ data is plotted, and the first positive time bin (red dotted line)
indicates the time duration for which the putative interactions occur more often in the
‘real’ data.
Figure S14. Movement rate of Drosophila melanogaster. The average movement rate
for all flies over 30 minutes within a trial was calculated. Although the olfactory mutant
Orco2 (n=23) had lower movement and a higher rate of failure (30%) to generate valid
criteria, we see that poxnΔM22-B5SuperA-158 (n=21) has a similarly low rate of movement
with lower rates of failure (5%), indicating that these movements speeds are sufficient, on
average, to generate reliable estimates of interactions, and that the lack of proper
interactions cannot be entirely caused by low rates of movement.
1.
Schneider J., Dickinson M.H., Levine J.D. 2012 Social structures depend on
innate determinants and chemosensory processing in Drosophila. Proc Natl Acad Sci U S
A 109 Suppl 2, 17174-17179. (doi:10.1073/pnas.1121252109).