First, the precision of social network data depends on reliable, error-free video tracking. The number of errors accumulated by video tracking is dependent on the level of contrast between the flies and the arena background, the length of the videos, and the number of flies (Robie et al., 2017). The most common tracking platform across the Drosophila social network literature is Ctrax, an open-source machine vision tracker (Branson et al., 2009). An inconvenient limitation of Ctrax is the requirement to tediously review the tracking data for errors that involve, for example, inconsistent identification of the same individual fly or changes in the size and orientation of the tracks. Each of these errors requires manual review and correction. In a recent experiment that repeatedly filmed 10 flies in an arena for 15 min, an automated error fixing script was applied to edit the tracking errors from Ctrax (Bentzur et al., 2020). An alternative called Flytracker, has been developed that claims to produce error-free tracking (Liu et al., 2018). Recently, Wice and Saltz (2021) cross-evaluated the performance of Flytracker with manual annotation of fly identities from 700 random frames and reported a strong correlation between automated tracking and manual tracking. While these alternatives may increase the speed of data collection, there is always the danger of harboring tracking errors that could lead to a loss of precision and integrity of the SNA. When considering a video tracking pipeline, the speed versus the precision should be weighed appropriately depending on the research objectives. For example, if social networks are generated from interactions defined by distance and angle parameters between flies, then it may be worth thoroughly reviewing and fixing tracking errors that swap identities, and that alter the size and orientation of the fly tracks, as done by multiple studies (Schneider et al., 2012; Jezovit et al., 2020; Rooke et al., 2020; Alwash et al., 2021). If the objective is to generate social networks from interactions defined exclusively by the distance between flies, errors in the orientation of the tracks, for example, can be tolerated and ignored.

Three recent studies analyzed Drosophila social networks using the more conventional static network approach (Liu et al., 2018; Bentzur et al., 2020; Wice and Saltz, 2021). This method generates a single network that represents the entire history of social interactions within a single video (visualized in Figure 1). The number of connections within these networks varies depending on the number of interactions observed. All three of these studies weighted the networks, offering additional information on the strength of connections. Four other Drosophila social network studies published to date utilized a dynamic iterative approach (Schneider et al., 2012; Jezovit et al., 2020; Rooke et al., 2020; Alwash et al., 2021). This method, published by Schneider et al. (2012), generates directed and unweighted iterations of networks in groups of flies. Unlike static networks, the iterative approach generates multiple networks from a single video at a controlled network density from a sliding boxcar filter (Kossinets and Watts, 2006; visualized in Figure 1). To summarize, one network iteration is built exclusively from a threshold number of unique interactions. When an additional unique interaction is observed, the oldest unique interaction is removed from the network and the newest interaction is added and this forms the second network iteration. This pattern continues and can produce hundreds or thousands of social network iterations in a single video, all offering snapshots of changing network structure over time. To score and compare the network measures of different types of fly groups, each iteration is standardized to thousands of random network permutations with equal in-degree and out-degree distributions. This normalization by degree distribution is then followed by averaging all iterations to summarize the network measure to a single data point. The result is an averaged z-score of all network iterations per video. This use of the z-score normalization attempts to evaluate properties of the group-wide behavioral interaction patterns independent of the observed individual interaction patterns (degree distribution). Overall this iterative method removes the confounds of network density and degree distribution when comparing networks across different treatment groups (Schneider et al., 2012).

The static and iterative methods each have their advantages and disadvantages (see Table 4). An advantage of the static network approach is that the results are intuitive, whereas the iterative approach is far more abstract to interpret. For example, in a static network, the betweenness centrality score (defined in Table 1) can be compared to each network node and the node with the highest score can be interpreted as being critical to the cohesion of the network. In the iterative approach, every node’s score is averaged in each network iteration and all iterations are averaged. As a result, the iterative analysis sacrifices information about the individual fly in exchange for measuring the overall group. Another distinction between the two methods lies in the network density. The iterative method controls the density of networks by capping the number of social interactions per network and analyzing iterations of density-controlled networks. It was found that iterative networks capped at a 25% network density for groups of 12 flies, which are networks consisting of 33 unique interactions, were more robust than other network densities (Schneider et al., 2012). The static approach simply allows all social interactions in a single video trial to fill out into a single network and therefore network density may vary between video trials. However, network density can be an informative behavioral measure of the animal group since denser networks indicate more social activity and this is not directly measured through the iterative approach. Instead, to gauge differences in network density in the iterative approach the researcher can compute the number of iterations, which is associated with higher social activity. Finally, the static and iterative approaches may be combined as seen in Pasquaretta et al. (2016a) where static networks were generated every 15 min from multiple hours of footage. This combines the simplicity of the static network methods with the advantage of measuring dynamic group activity over multiple time points.

While the methods of generating and analyzing Drosophila social networks differ, one question many of the studies address is how social experience affects the group dynamics of flies (Schneider et al., 2012; Liu et al., 2018; Bentzur et al., 2020). Schneider et al. (2012) examined the effects of 3-day social isolation by measuring the network properties of groups of flies that were all separately reared in isolation, compared to groups of flies reared in a socialized environment. There were no significant differences found in the average network measures between these treatments (Schneider et al., 2012). A limitation to this study was the use of the same fixed criteria (2 body lengths, 90 degrees, and 1.5 s) for the isolated flies and socialized flies. This did not consider potential differences in the social interaction patterns of isolated versus socialized flies. Therefore, to better understand how social experience influences the behavior of flies, we re-analyzed the Schneider et al. (2012) data using automated criteria (Figure 2). Indeed, we find that socially isolated flies tend to engage in longer social interactions than socialized flies (housed-together treatment). On the other hand, a treatment of socialized flies that were combined from separate housing groups (mixed-together treatment) tend to have a shorter interaction time compared to the housed-together treatment. We then generated social networks using the iterative approach and found that social isolation significantly affects the network structure. For example, global efficiency (defined in Table 1) is significantly higher in isolated flies, indicating that isolated flies have more redundant connections in their networks. Isolated flies also display a significantly lower betweenness centrality, indicating that there are fewer central individuals serving as a hub in the network. Across all measures, we observe greater variability in networks of isolated flies compared to the controls, particularly in assortativity and clustering coefficient (defined in Table 1). Lack of social experience in these groups of isolated flies may be contributing to these less predictable network measures. Other behavioral measures, such as the average interaction rate and percentage of interactions reciprocated, were also significantly lower in groups of isolated flies. The new analysis shown here underscores the importance of the automated criteria and makes the findings of Schneider et al. (2012) consistent with recent studies that have addressed these questions in other ways (Liu et al., 2018; Bentzur et al., 2020).

Liu et al. (2018) took a different approach to the same question. Rather than isolate virgin flies for 3 days, 9-day old flies were isolated for 6 days. Replicates of static, directed, and weighted social networks were generated from multiple video sources and then averaged. Liu et al. (2018) revealed groups of 16 flies that had been isolated tend to be more active, interact more often and produce networks with a higher clustering coefficient than groups of socialized flies. Additionally, a time course of 1-day to 6-day long isolation treatments showed that the average clustering coefficient is significantly greater than that of socialized flies at all time points. This suggests that a single day of isolation is sufficient to alter the clustering coefficient of flies, and this may be robust since we also report a higher clustering coefficient in isolated flies (Figure 2). In fact, Liu et al. (2018) also report a higher global efficiency in isolated flies, which also agrees with our re-analyzed data (Figure 2). This similarity illustrates that network measures are robust in flies of different ages since groups of 3-day old flies raised in isolation produce similar networks to 9-day old flies.

A more recent experiment by Bentzur et al. (2020) also found social isolation affects fly networks. Like Schneider et al. (2012), these authors collected flies as virgins and isolated or socialized them for 3 days before recording their behavior. In this study, the authors generated static networks, and they were analyzed two ways: (1) weighting nodes by the number of interactions to emphasize short and acute social patterns; (2) weighting nodes by the length of interactions to emphasize long-lasting social interaction patterns. The authors found that isolated flies displayed a lower average betweenness centrality than socialized flies in networks weighted by the number of interactions and the length of interactions, and our re-analyzed data further validates this finding (Figure 2). Bentzur et al. (2020) also found that, on average, isolated flies have a lower modularity score (defined in Table 1) in networks weighted both ways, indicating that isolated flies produce networks that are less compartmentalized. This can be attributed to their finding that social isolation leads to flies being more active as measured by increased velocity, and decreased instances of flies physically clustering and interacting over long periods. However, higher locomotor activity may lead to more frequent social interactions, resulting in a higher network density, and a higher average degree/strength. Indeed, this was reported in isolated flies (Liu et al., 2018; Bentzur et al., 2020) and may serve as a confound when assessing network structure. Alternatively, the aggregate clustering observed in socially experienced flies could reduce the number of social interactions, skewing the data to a lower network density. Interestingly, in networks weighted by the length of interactions, Bentzur et al. (2020) found no significant difference in the network density between isolated and experienced flies. Perhaps measuring networks weighted by length of interactions reduces the confounds that arise from differences in locomotor activity and frequency of social encounters because these networks favor connections between flies that spend longer periods of time socializing.

Despite the differences in methodology, three studies overlap in showing how social experience affects the group dynamics of flies. Two recent publications were the first to report these effects (Liu et al., 2018; Bentzur et al., 2020), and re-analyzing data from Schneider et al. (2012) further validates these two independent studies. Flies isolated for 3 days form social networks with a lower betweenness centrality (Bentzur et al., 2020) (Figure 2H). This results in less cohesive social networks with fewer central individuals holding the group together. Additionally, isolated flies form networks with a lower modularity (Bentzur et al., 2020), which indicates social isolation leads to less complex network structures. Taken together, these studies show that isolating flies hinders their ability to socialize within groups. This appears to contradict the finding that flies isolated for 1–6 days form networks with a higher clustering coefficient (Liu et al., 2018) (Figure 2F), indicating isolated flies on average may form cliquish groups. However, an automated classifier trained to detect instances of multiple flies physically aggregating found that isolated flies aggregate less than socialized flies (Bentzur et al., 2020). This illustrates the point that social networks capture patterns not necessarily intuitive to the human eye and future experiments would benefit by applying machine learning classifiers to measure additional qualities of social interactions. Measuring a wealth of behavioral classifiers, as done by Bentzur et al. (2020), would help validate and interpret the more abstract social network measures. Another recent experiment by Sun et al. (2020) studied the social attraction of free-walking flies by measuring their proximity to immobilized flies in arenas. With this assay, the authors found evidence that isolated flies exhibited a decrease in social attraction when compared to socialized flies. Finally, we find evidence that isolated flies are just as active as socialized flies and engage in fewer social interactions on average (Figure 2), which contradicts other studies (Liu et al., 2018; Bentzur et al., 2020). This highlights the benefit of automated criteria for generating social networks (Table 4). The behavior of the flies filmed may fluctuate based on experiments being completed at different times. The automated algorithm (Schneider and Levine, 2014) can take these behavioral fluctuations into account and estimate social space criteria reflective of the flies’ behavior in the current experiment. Additionally, automated criteria can take behavioral differences between experimental treatments into account. For example, if socially isolated flies tend to interact differently than socially experienced flies, the automated criteria can correct for this and generate social networks that best represent the social environment being measured.

Each study that compared social networks of isolated and socialized flies examined groups of different sizes. A recent experiment by Rooke et al. (2020) demonstrated that group size affects features of social networks by comparing groups of 6, 12, and 24 flies across three different arena sizes. First, the authors found that the average locomotor activity of flies was similar across different group sizes and arena sizes, suggesting flies regulate their movement to compensate for decreased space. In terms of the social networks, the authors generated iterative, unweighted, and directed networks at controlled network density as published by Schneider et al. (2012). Rooke et al. (2020) found that groups of 6 and 12 flies form networks with a significantly lower average clustering coefficient than groups of 24 flies, and this was consistent across three different arena sizes. Additionally, groups of 12 and 24 flies form networks with a significantly higher average betweenness centrality than groups of 6 flies. This suggests that larger groups, on average, have more flies that are central and maintain greater cohesion across the group. Although the number of social interactions increases as the arena size and group size increase, properties of the social networks remain consistent across the same group size. Since the social networks were all generated at a controlled network density, and all flies were reared with equal social experience, differences in the network measures can be attributed to differences in group size. No matter how confined or dispersed a group of flies may be, the properties of the group shift only when the size of the group changes. Perhaps individual flies may be sensitive to changes in group size based on visual feedback and through the perception of pheromone concentration and organize themselves in the group according to these signals.

With D. melanogaster being one of the most popular organisms for behavioral genetic experiments, the wide availability of mutant strains and genetic tools to manipulate gene expression have been applied to social network experiments. To date, social networks have been generated for flies with disrupted visual, olfactory, gustatory, and acoustic modalities (see Table 5). Schneider et al. (2012) reported that the gustatory mutant poxn^ΔXBs6 displayed an extreme reduction in the ability to form social networks (Schneider et al., 2012). More specifically, 40% of the videos filmed of these mutants did not harbor enough social interactions to form a single iterative network (Schneider et al., 2012). Recently, Jiang et al. (2020) also reported that poxn mutants, in addition to a range of other gustatory mutants, displayed an impaired ability to form physical social clusters. Together, this suggests chemosensory receptors are crucial for maintaining the sociality of flies.

Schneider et al. (2012) also demonstrated that hearing-impaired inactive mutants (iav¹) produced social networks that were not significantly different from wild-type flies. Surgical removal of ariste to ablate auditory perception in flies also had no effect on social clustering behaviors (Jiang et al., 2020). However, Jiang et al. (2020) reported that iav¹ mutants form more dispersed social clusters, unlike wild-type flies that are more tight-knit. This is also reflected in social space criteria for iav¹ mutants where the distance parameter was estimated to be larger than wild-type flies (Schneider and Levine, 2014). Although auditory mutants may socially interact and cluster less than wild-type flies, there is currently no evidence that manipulating auditory cues within a group of single-sex flies affects measures of social network structure (Schneider et al., 2012).

To disrupt vision, experiments have been conducted on flies in the dark. Schneider et al. (2012) reported that groups of flies filmed in the dark display a lower clustering coefficient and higher betweenness centrality, but these effects were not considered significant when accounting for multiple test correction (Schneider et al., 2012). Bentzur et al. (2020) found that groups of socially isolated flies behave more similarly in the light and dark compared to socially experienced flies. The authors reported that in networks of socially experienced flies, visual disruption leads to a significantly lower average betweenness centrality, opposite of what was reported by previous studies (Schneider et al., 2012). Despite disagreement in the social network data when subjecting flies to darkness, multiple studies report similarities in how flies aggregate and physically cluster. Using automated behavioral classification, Bentzur et al. (2020) reported that groups of flies in the dark aggregate less often and for shorter periods of time on average. Data by Jiang et al. (2020) also found that wild-type flies in the dark, along with norpA33 visual mutants, cluster together less than wild-type flies. These two recent studies reinforce observations by Schneider et al. (2012) that darkness decreases the average interaction duration among groups of flies.

Arguably olfaction is the dominant sensory mechanism Drosophila depends on to locate foraging sites and conspecifics. Ablating olfaction is complex because Drosophila insects possess multiple olfactory receptors that are encoded by multiple genes. The olfactory mutant, orco, is known to have a severe loss of smell because it is deficient for a co-receptor that complexes with a variety of odorant receptors (Vosshall and Hansson, 2011). Social networks of orco mutants have been shown to have a significantly lower global efficiency than wild-type flies, with orco heterozygotes displaying an intermediate score (Schneider et al., 2012). This may indicate that the copy number of the orco gene leads to social interactions that, on average, result in a greater social distance between individuals in the network. In the same study, the orco mutants displayed a higher clustering coefficient and a higher assortativity compared to controls, although the differences were not statistically significant after multiple test correction (Schneider et al., 2012). Also, the orco mutant aggregates less with conspecifics compared to wild-type flies (Jiang et al., 2020). Overexpressing an orco transgene in the olfactory system of these mutants led to the flies aggregating like wild-type flies (Jiang et al., 2020). This is similar to an observation of ant orco mutants that displayed a reduction in their ability to follow pheromone trails and cluster with other ants (Trible et al., 2017). This cross-species reduction in aggregation suggests that olfaction is crucial for the sociality of numerous insects, and it is no surprise that olfactory mutants produce social networks different from wild-type flies.

In addition to studying the social behavior of fly mutants, the Drosophila model system offers genetic tools to manipulate the expression of genes in a tissue-specific manner through the GAL4-UAS system (Elliott and Brand, 2008). This system was applied to recent social network studies to examine the downstream behavioral effects of ablating specific olfactory sensing cells (Bentzur et al., 2020; Rooke et al., 2020). One experiment examined the social networks of flies where the olfactory receptor neurons Or65a and Or67d were inhibited by driving the expression of kir2.1 in those cells. These olfactory receptors are known to be sensitive to cVA, a male-specific pheromone that mediates aggressive and copulatory behaviors in male flies (Bontonou and Wicker-Thomas, 2014). Interestingly, flies with inhibited Or67d neurons did not produce social networks drastically different from wild-type flies despite there being evidence that Or67d plays a role in social attraction (Bentzur et al., 2020; Sun et al., 2020). However, the inhibition of Or65a neurons leads to a significantly decreased average betweenness centrality (Bentzur et al., 2020). Another experiment focused on inhibiting the olfactory support cells that express the gene lush, which is expressed in trichoid sensillae of flies and aids in the binding of ligands to olfactory receptors (Rooke et al., 2020). By driving the expression of kir2.1 in all lush-expressing cells, Rooke et al. (2020) found that lush-inhibited flies produce social networks different from controls in larger group sizes. More specifically, groups of 6 lush-inhibited flies formed social networks with an average clustering coefficient and betweenness centrality that resembles the wild-type controls. However, in groups of 12 and 24, the lush-inhibited flies formed social networks with a significantly higher betweenness centrality and clustering coefficient than wild-type controls (Rooke et al., 2020). Together, results of these studies indicate that different olfactory genes, expressed in different tissues, may play different roles in regulating group-wide social connections in flies.

Transgenic tools have also been used to manipulate the foraging (for) gene in a recent SNA study. This gene expresses natural polymorphisms in flies that influence behavioral phenotypes in the larval stage called rovers and sitters (Sokolowski, 1980). Alwash et al. (2021) demonstrated that networks of adult rovers and sitters form different social networks, suggesting this gene influences the behavior of adult flies. Sitter flies were shown to display a higher interaction duration and were more likely to reciprocate interactions, whereas rover flies were more active and displayed higher interaction rates. Compared to rovers, sitters formed networks with a higher assortativity and clustering coefficient, as well as a lower global efficiency suggesting there is less efficient information flow within these groups of flies. Alwash et al. (2021) also used separate transgenic lines, generated by Allen et al. (2017; see for details), that carry 1 copy, 2 copies and 4 copies of the for allele, respectively. By comparing social networks across these lines, it was found that for gene dosage affects the average assortativity, clustering coefficient and global efficiency measures. Additionally, the average interaction duration, the average rate of interactions, the proportion of interactions reciprocated and the activity of flies all changed across different dosages of for. The authors confirm that many of the social network differences observed between rovers and sitters are influenced by the for locus. These findings characterize the influence of a specific gene on social network dynamics in Drosophila, shedding light on the genetic underpinnings of sociality.

Multiple independent experiments that measured the social behavior of Drosophila mutants and transgenic flies with inhibited neurons revealed that sociality of flies is multisensory. In unisex groups of D. melanogaster, auditory sensory systems do not appear to play a role in social organization (Table 5). Visual, gustatory, and olfactory manipulations cause flies to behave differently than wild-type flies in several ways (Schneider et al., 2012; Bentzur et al., 2020; Jiang et al., 2020; Rooke et al., 2020). However, these studies investigating the sensory mechanisms behind collective behavior are limited to unisex groups. It is possible that mixed groups of male and female flies may generate social structures that depend on a wider range of sensory systems since, for example, auditory cues are critical for courtship in flies (von Schilcher, 1976). Future studies should consider manipulating the composition of the social groups when inhibiting genes of interest to widen our knowledge of Drosophila social structures, like how Rooke et al. (2020) studied flies with lush inhibition at a variety of group sizes. It remains difficult to define how differences in precise network measures of mutants translate to differences in social organization, especially since various social network experiments utilize different methods of generating and analyzing networks. However, experiments that focused on social attraction and aggregation of flies used similar mutants and transgenic tools and found overlapping results to social network studies (Bentzur et al., 2020; Jiang et al., 2020; Sun et al., 2020). For example, Sun et al. (2020) reported that a combination of both vision and olfaction are crucial for the social attraction behavior of flies. This suggests that social networks capture some aspect of group-level social organization that is genetically and neurologically controlled. Recent work has demonstrated that the for gene plays a role in the social organization of adult flies since different polymorphisms are associated with differences in social network structure and manipulating the for gene influences this structure (Alwash et al., 2021). Further experimentation in social attraction and aggregation of flies at the neuronal and genetic level can assist in unraveling how abstract social network measures translate to a real-world group structure.

To date, one group analyzed Drosophila social networks to directly study information flow, like many social network studies on ant colonies (Blonder and Dornhaus, 2011; Mersch et al., 2013; Stroeymeyt et al., 2018). Social communication within Drosophila groups can inform naïve flies about the presence of oviposition sites (Battesti et al., 2014) and the presence of predatory insects (Kacsoh et al., 2018). Pasquaretta et al. (2016a) applied SNA to examine how information spreads within a group of informed and naïve flies. This was done by video recording a 4-h training phase designed to inform focal flies of an oviposition site. Then social networks were generated within groups consisting of 8 informed flies and 4 uninformed flies (Pasquaretta et al., 2016a). Static, directed, and weighted social networks were generated every 15 min from 4 h of video footage. Afterward, every trained and untrained female fly was subjected to an oviposition site choice assay to determine if the mean and variance of social network measures predict whether uninformed flies follow or avoid the choices made by informed flies. Uninformed flies followed the correct choice when informed flies had less variable network distances from other individuals, as measured by weighted closeness centrality (defined in Table 1). Uninformed flies also followed when informed flies had a similar number of social contacts, as measured by eigenvector centrality (defined in Table 1) and when informed flies exchanged information to a similar extent, as measured by information centrality index (defined in Table 1). On the contrary, uninformed flies were less likely to follow the correct choice when they had a high betweenness centrality in the social network. Taken together, this suggests that when informed flies participate in most social interactions within the group, the uninformed flies are more likely to follow, and information is passed from the informed to the uninformed flies. In groups where uninformed flies were central to group cohesion (high betweenness centrality), the informed flies had less influence in transmitting the site preference. The authors also reported a remarkable finding where informed flies were more likely to avoid the media they were trained to prefer if they formed clusters, measured by a higher mean clustering coefficient. Properties of a social group are complex, and this highlights how individual foraging preferences can shift based on social associations within a group.

Diseases can also be transmitted via social interactions within a group. Utilizing the SNA approach in bumblebee colonies, for example, shed light on the relationship between interaction rate and parasitic transmission (Otterstatter and Thomson, 2007; Naug, 2008). So far, no studies to date have used SNA to explore how social interaction and network properties affect disease transmission in Drosophila. However, one study by Dawson et al. (2018) investigated how the social environment affects cancer progression in flies. In a homogenous group, cancerous flies were found to have higher interaction rate and duration than in heterogeneous groups consisting of cancerous and healthy flies. Additionally, Dawson et al. (2018) showed that tumor progression is slower when cancerous flies are kept in a homogenous group, and tumor progression is faster when cancerous flies are in isolation or within a group of healthy individuals. The use of the SNA approach can allow us to investigate the relationship between disease progression and social interactions even further by analyzing global network measures.

Recently, a social network comparative study was conducted on 20 drosophilid species. Generating iterative, directed, and unweighted networks from groups of 12 male flies and groups of 12 female flies across all species, Jezovit et al. (2020) found no phylogenetic patterns for the species differences observed in assortativity, clustering coefficient, betweenness centrality, and global efficiency. This mirrors the results of a social network comparative analysis conducted on primates that also reported no evidence of phylogenetic signal in species-specific social networks (Pasquaretta et al., 2014). However, significant phylogenetic signal was found for the variation observed in social distance Jezovit et al. (2020). Social distance also correlated with the relative leg length of each species, suggesting morphological traits can influence behavioral evolution in flies. Next, the authors extracted averaged climate data from the geographic range of each drosophilid species and tested for correlations with each species’ averaged social network score. The authors found that variation in the climate data predicted species differences in the social network measures better than the differences found in the flies’ general behavioral characteristics such as average locomotor activity, average interaction duration, and average tendency to reciprocate interactions. Considering that each fly species descended from an inbred stock domesticated to the laboratory environment, it is surprising that factors of each species’ environment predicted differences in their social network measures. From these findings, we hypothesize that group-level organization is a behavioral trait that adapted to the abiotic selective pressures of each species’ habitat. For example, Drosophila species from tropical environments tend to have shorter cuticular hydrocarbons and rely more on visual sensory modalities than arid-adapted species (reviewed by Jezovit et al., 2017) and these ecological categories may also be relevant to species’ social structures measured by SNA. Finally, Jezovit et al. (2020) collected two independent datasets of social networks for 5 species, separated by 2 years at the time of collection. Consistent trends in the relative species’ differences were found for average assortativity, clustering coefficient, betweenness centrality, and global efficiency. This replication shows that species-specific social networks are robust and may represent phenotypes that emerge from physiological and behavioral mechanisms in individual flies.

Another recent comparative study by Wice and Saltz (2021) investigated the evolutionary relevance of social network measures across 20 different D. melanogaster strains. The authors generated static and directed social networks from mixed groups of flies using fixed criteria (see Table 2). These groups consisted of 10 males and 10 females, and each individual was genotypically distinct. The authors measured in-strength, out-strength, betweenness centrality, clustering coefficient, and eigenvector centrality for each fly within the group, and then compared the distribution of these measures for each genotype. This study stands out from other social network studies in that the authors were focused on measuring the characteristics of individual nodes and not the overall network structure. By comparing average network measures across numerous strains, the authors reported the broad sense heritability measure of clustering coefficient, betweenness centrality, and eigenvector centrality. Interestingly, betweenness centrality displayed the highest broad sense heritability score where genotypic differences account for 16.6% of the variation in this network measure (Wice and Saltz, 2021). This corroborates a prediction made by Schneider et al. (2012) that betweenness centrality may be a heritable trait based on robust differences observed between two D. melanogaster strains. To study environmental effects on social network measures, the authors reared the flies in various environments differing in calorie concentration and in the ratio of protein to carbohydrate content. The authors found no effect of environmental variation on betweenness centrality, similar to what Jezovit et al. (2020) found when comparing social networks across multiple Drosophila species. There is also evidence that various drosophilid species maintain consistent group structures across separate experiments, reinforcing the idea that social networks are emergent properties built from some genetic foundation shared by the individuals in the group (Jezovit et al., 2020). This view is strengthened by emerging evidence of specific genes accounting for differences in social networks within Drosophila (Bentzur et al., 2020; Rooke et al., 2020; Alwash et al., 2021). If social networks measure some heritable aspect of social behavior, then we can begin to consider that these properties are phenotypes that diversified through evolutionary selection mechanisms.