Social network position is a major predictor of ant behavior, microbiota composition, and brain gene expression

Abstract

The physiology and behavior of social organisms correlate with their social environments. However, because social environments are typically confounded by age and physical environments (i.e., spatial location and associated abiotic factors), these correlations are usually difficult to interpret. For example, associations between an individual's social environment and its gene expression patterns may result from both factors being driven by age or behavior. Simultaneous measurement of pertinent variables and quantification of the correlations between these variables can indicate whether relationships are direct (and possibly causal) or indirect. Here, we combine demographic and automated behavioral tracking with a multiomic approach to dissect the correlation structure among the social and physical environment, age, behavior, brain gene expression, and microbiota composition in the carpenter ant Camponotus fellah. Variations in physiology and behavior were most strongly correlated with the social environment. Moreover, seemingly strong correlations between brain gene expression and microbiota composition, physical environment, age, and behavior became weak when controlling for the social environment. Consistent with this, a machine learning analysis revealed that from brain gene expression data, an individual's social environment can be more accurately predicted than any other behavioral metric. These results indicate that social environment is a key regulator of behavior and physiology.

Fig 1. System overview.

(A) Ants were tagged with unique 1.4 mm² matrix barcodes and paint-marked to indicate their age. (B) Head and body regions were defined around each tag. (C) Worker age distribution across the 4 colonies. See Fig A in S1 Text for equivalent distributions per colony. The code and data used in this figure are available on Zenodo (doi.org/10.5281/zenodo.8043085 - data: “Fig 1C&S1.csv”; code: “02-Main.R”).

Fig 2. Social network position, time spent foraging, and age.

(A) The social networks for each of the 4 colonies (rows) with workers colored according to time spent foraging (column 1), social maturity (column 2), and age (column 3). Lowest values are yellow; highest values are dark blue. Queens are colored magenta. Edge color intensity and width correspond to edge strength. Layouts are calculated with the Fruchterman–Reingold algorithm [32] using R package “iGraph” [33]. (B) Scatter plots relating proportion of time spent foraging, social maturity, and age. The code and data used in this figure are available on Zenodo (doi.org/10.5281/zenodo.8043085 - data: all 4 “Fig 2A…” csv files and “Fig 2B.csv”; code: “02-Main.R”).

Fig 3. Similarity in social network position, physical environment, microbiota, brain gene expression, behavior, and age.

(A) A 5-layer multiplex network constructed from behavior, brain gene expression, microbiota, the physical environment, and social interactions. In each layer, each node represents a worker. Nodes are colored according to behavior (cyan = nursing; yellow = cleaning; magenta = foraging; black = guarding). Intralayer edges are unweighted and connect pairs whose interaction strength exceeds the upper quartile of the edge–weight distribution. Interlayer edges connect each worker with itself in the adjacent layers. (B) Graphical representation of the correlation (R² values) between the 5 layers and age (in blue). Edge width is proportional to edge strength. Layout is calculated with the Fruchterman–Reingold algorithm [32], and vertices are colored according to the layer labels in panel (A) and sized according to their strength (i.e., the sum of their weighted connections). The code and data used in this figure are available on Zenodo (doi.org/10.5281/zenodo.8043085 - data: all 11 “Fig 3A…” txt files and “Fig 3B.csv”; code: “Multiplex.py” and “04-InterlayerCorr.R”).

Fig 4. Validation of predictive accuracy.

(A) Box plots of the R² values between observed and predicted values for the proportion of time spent performing each behavior individually, for position along PC1 of behavioral space, for age, and for social maturity. Black lines indicate median values; boxes and whiskers indicate upper and lower quartiles and 1.5× IQ range, respectively. (B) Scatter plot of the predicted versus observed social maturity scores for 10 randomly selected iterations. Color indicates iteration. The code and data used in this figure are available on Zenodo (doi.org/10.5281/zenodo.8043085 - data: “Fig 4A.csv”; code: “05-ML.R”).