Marginal link between sociality and gut microbiome in disparate mammalian hosts

Abstract

Studies in model organisms and wild populations have uncovered manifold links between the gut microbiome and sociality, which, considering the adaptiveness of social behaviour, suggest a potentially generalized coevolution between microbiomes and social behaviour. Here, we leverage phylogenetically and ecologically diverse data from the Earth Microbiome Project to test the generality of the links between sociality and the gut microbiome in wild animals. We find evidence of a small but significant link between sociality and microbiome beta diversity, but not alpha diversity, in mammalian taxa, potentially due to socially mediated microbial transmission. Our work highlights the value of leveraging large-scale multi-study datasets to test fundamental questions about the role of sociality in host–microbiome coevolution.

Despite the strong laboratory evidence for a link between gut microbiome and sociality, a cross-species analysis detects only marginal evidence that this pattern is generalized across natural populations. Created in BioRender. Lebeuf-Taylor, E. (2026) https://BioRender.com/wexvs50.

Introduction

The adaptive value of social behaviour and the bidirectional link between sociality and the gut microbiome have generated considerable interest in understanding what role, if any, the gut microbiome may have played in the evolution of sociality. The gut microbiome—the community of microbes that inhabits animals’ gastrointestinal tract—can affect host fitness (Gould et al. 2018, O’Brien et al. 2024) and is linked to host metabolism (Martin et al. 2019), immunity (Belkaid and Hand 2014), and the gut–brain axis (Carabotti et al. 2015, Bonaz et al. 2018). Less is known about the association between sociality and gut microbiomes. Studies in model organisms and wild populations have uncovered various links between the microbiome and social behaviour (Sarkar et al. 2020, Baniel and Charpentier 2024, Debray et al. 2024); however, whether the association between sociality and the microbiome is generalized across diverse host taxa remains an important outstanding question.

Laboratory studies certainly suggest a strong bidirectional link between sociality and microbiome in a handful of host species. In mice, the gut microbiome is essential for the development of normal social behaviour (Desbonnet et al. 2014), and social conditions can affect microbiome composition (Bailey et al. 2011, Bharwani et al. 2016). Social behaviour is linked to the microbiome in fruit flies (Venu et al. 2014, Chen et al. 2019) and honeybees (Liberti et al. 2022). Impaired sociality can be rescued with the administration of specific ‘prosocial’ bacteria in both mice and zebrafish (Desbonnet et al. 2014, Buffington et al. 2016, Davis et al. 2016, Wu et al. 2021). While the bidirectionality of these findings suggests a strong association, studies in non-model organisms are critical to understanding their ecological reality, which is the subject of ongoing debate (Greyson-Gaito et al. 2020).

Field studies investigating the association between sociality and microbiome are far less conclusive about the importance of the host–microbiome link as pertains to sociality. Unlike laboratory work, field studies mostly focus on the link between social traits and microbiome at the level of individuals. Generally, social structure (i.e. the social organization of a population) can affect microbiome composition (reviewed in Davidson et al. 2020), as the horizontal transmission of bacteria among conspecifics drives individual microbiome community assembly (Gilbert 2015, Raulo et al. 2021, Yarlagadda et al. 2021, Murillo et al. 2022). The social structuring of the microbiome in wild populations is usually related to social behaviours that favour physical proximity, such as allogrooming in baboons (Papio cynocephalus) (Gilbert 2015), time spent together in chimpanzee dyads (Pan troglodytes) (Moeller et al. 2016), and contiguous roosting and meal sharing in bats (Desmodus rotundus and Eptesicus fuscus) (Yarlagadda et al. 2021, Lebeuf-Taylor et al. 2025). So far, however, it remains unclear whether there are generalized patterns linking sociality and the microbiome.

Given the evidence that sociality influences the microbiome, the social exchange of microbes, and the link between sociality and fitness in some populations (Alberts 2019), it has been suggested that microbiomes and hosts have coevolved differently in social, compared with solitary, host taxa (Debray et al. 2024). Within this proposed symbiotic framework, social interactions enable the transmission of beneficial microbes (Baniel and Charpentier 2024) and drive the emergence of symbionts that promote sociality (Münger et al. 2018, Sherwin et al. 2019, Debray et al. 2024). Coevolution based on sociality may favour both host and microbe fitness by group-living benefits for the former (Lombardo 2008) and providing a larger pool of potential hosts for the latter (Rubenstein 1978). Put another way, have social species coevolved with microbes to the point that they share a microbiome profile, distinct from that of solitary species?

If the microbiome of social host taxa does indeed differ from that of solitary ones, it may be enriched in ‘prosocial’ microbes. Candidate prosocial microbes can be found among those known to improve social behaviour in experimental studies. Species of the fermenting bacteria Limosilactobacillus (formerly Lactobacillus) and Bifidobacterium have consistently been found to improve social deficits in laboratory studies on mice (Sgritta et al. 2019a, Dooling et al. 2022, Wang et al. 2024) and zebrafish (Bruckner et al. 2022). Both genera are also abundantly found in eusocial bees while being completely absent from a solitary outgroup (Kwong et al. 2017) ). It is thought that Limosilactobacillus and Bifidobacterium may double as modulators of social behaviour in addition to aiding host digestion (Sherwin et al. 2019). The detection of sociality-linked microbes in such disparate host species as mice, zebrafish, and bees warrants a wider exploration of the generalizability of these findings.

The detection of generalized links between sociality and microbiome across phylogenetically diverse hosts, including the presence of prosocial bacteria, presents several challenges. Foremost, sociality is a vague term that encompasses structures and behaviours that are both complex and diverse (Prox and Farine 2020). Social groups can range from pairs of individuals to collectives numbering in the thousands, even encompassing many different species as in mixed-species bird flocks. Furthermore, some species exhibit seasonal or sex-dependent variation in social structure. Finally, it can be difficult to make meaningful comparisons between studies due to the wide diversity of approaches used for sampling and analysis, as well as the lack of a standard format for published data.

Nevertheless, studies across diverse host taxa—rather than within populations—are crucial to understanding whether there are broad-scale patterns linking the microbiome and sociality. The earth microbiome project (EMP) provides an ideal framework to address the challenge of combining methodologically heterogeneous microbiome studies (Thompson et al. 2017). The EMP’s goal of mapping microbial life on Earth is achieved by crowdsourcing microbiome data from a wide range of environments—including various animal hosts—and standardizing the data, allowing for large-scale analyses that combine multiple different studies. Thus, the framework offered by the EMP enables researchers to investigate microbiome patterns on a global scale. Here, we leverage the EMP’s extensive gut microbiome dataset to test fundamental questions about the link between microbes and sociality in a large and taxonomically diverse group.

Our questions consider both directions in the host–microbiome association. First, we investigate the effect sociality may have on the microbiome: (1) Do the microbiomes of social species exhibit higher alpha diversity, due to potentially higher horizontal transmission among conspecifics? (2) Are the within-species microbiomes of social hosts more homogeneous, given potential intraspecific homogenization through social transmission? (3) Does the microbiome composition of social species differ from that of solitary species? Next, we consider the potential effect of the microbiome on social behaviour: (4) Are candidate prosocial bacteria such as Limosilactobacillus and Bifidobacterium more abundant in social species?

This work has several important caveats. We are inherently limited to identifying differences in community composition based on operational taxonomic units (OTUs), rather than a finer scale classification, as this is the nature of the EMP dataset. Additionally, as with any taxonomical inventory, we lack metagenomic information on the observed bacterial community and are thus limited to testing differences in microbial communities rather than functional profiles. Lastly, while much social microbiome research in wild animal populations includes interindividual interaction data, this is not included in the EMP dataset.

Methods

Data acquisition

We obtained microbial community data generated from the V4 region of the 16S rRNA gene and curated by the EMP. We used the closed reference SILVA 16S observation (OTU) table (emp_cr_silva_16S_123.release1.biom), quality control-filtered metadata (emp_qiime_mapping_qc_filtered_20170912.tsv), and the SILVA reference tree with 97% similarity (silva_123.97_otus.tre) provided in the EMP release. We filtered the EMP dataset to non-human vertebrate gut metagenome samples and omitted samples for which it was impossible to identify the host due to conflicting metadata information. Since only one species belonged to class Reptilia (Iguana iguana), we limited our analyses to classes Mammalia and Aves. To control somewhat for vastly different sample sizes between species, we conducted all analyses on a subsample of up to 41 samples per species, corresponding to the mean number of samples per species. Our analysis is also informed by two major known drivers of interspecific variation in gut microbiomes: host phylogeny (Brooks et al. 2016) and diet (Delsuc et al. 2014, Ingala et al. 2021), which result in distinct clusters among herbivores, carnivores, and omnivores (Muegge et al. 2011). Previous work on the EMP dataset has shown that phylogeny and diet are significant factors driving internal microbiome diversity in this dataset (Woodhams et al. 2020). We therefore included both host phylogeny and diet (herbivorous, carnivorous, omnivorous) in our models.

Sociality metric

Given the diversity of sociality types among the EMP host species and the exploratory nature of this work, we broadly define social species as species that tend to live in groups for a considerable length of time (Ward and Webster 2016). This general definition allows us to use a common metric for species with different types of sociality, e.g. foxes who hunt cooperatively and bats who share a daytime roost, since in both cases these species form aggregates through social attraction. While this definition may be validly criticized for being too vague, its generality is precisely what makes it functional in an analysis that comprises a wide range of species with diverse forms of sociality. Indeed, broad catch-all definitions of sociality are commonly used in the field of social microbiome research (Sherwin et al. 2019); although they often conflate diverse social behaviours with vastly different underlying mechanisms, they nevertheless allow for large-scale studies that encompass a wide range of species, such as this one.

We employed a simple sociality classification for the host species in the EMP gut microbiome dataset, categorizing species as social or solitary, while those that could not be confidently defined as either were classified as intermediate (Table 1). The identities of the EMP host species warranted this simplicity. For one, many display seasonal and sex-dependent variation in gregariousness and social behaviours (including those that may affect whether and how gut microbes are exchanged). However, because the EMP metadata does not include date of sample collection or sex-disaggregated data, it is impossible to reliably assign a finer-scaled assessment of social behaviour. The limited literature on social behaviour for several EMP species further precludes the use of a richer sociality metric.

Table 1.

Sociality of host species in the EMP gut microbiome dataset.

Host common name	Host scientific name	Sociality level	Sociality description
sambar deer	Rusa unicolor	intermediate	sex-dependent; live alone or in small family groups (Eisenberg and Lockhart 1972, Leslie Jr 2011)
kangaroo	Macropus sp.	social	highly social; live in large groups known as mobs, from small groups to more than a hundred individuals (Kaufmann 1975)
European rabbit	Oryctolagus cuniculus	social	stable social groups; large groups with multiple breeding pairs (Cowan and Bell 1986, DiVincenti and Rehrig 2016, Sawyers et al. 2022)
short-eared possum	Trichosurus caninus	solitary	thought to be solitary but bonded pairs can sometimes share the same den; overlap in denning ranges (Lindenmayer 1997)
cat	Felis catus	solitary	generally solitary (Barratt 1997)
domestic dog	Canis lupus familiaris	intermediate	both solitary and social (Wynne 2021)
goose	Anser sp.	social	form flocks (Guggenberger et al. 2022)
red fox	Vulpes vulpes	social	group-living but forage alone; at high densities, form groups of up to 10 individuals; grooming within family groups promotes social bonds (Dorning and Harris 2019)
emu	Dromaius novaehollan-diae	solitary	solitary; can form loose flocks associated with food sources (Patodkar et al. 2009)
dingo	Canis lupus dingo	social	hunt in packs but little physical contact (Letnic et al. 2012)
screaming hairy armadillo	Chaetophractus velle-rosus	solitary	solitary (Amaya et al. 2019)
pink fairy armadillo	Chlamyph-rus truncatus	solitary	solitary (Borghi et al. 2011)
Linnaeus’ two-toed sloth	Choloepus didactylus	solitary	solitary (Adam 1999)
Hoffmann’s two-toed sloth	Choloepus hoffmanni	solitary	solitary (Voss and Fleck 2017)
nine-banded armadillo	Dasypus novemcinctus	solitary	solitary (McDonough 2000)
Chinese pangolin	Manis pentadactyla	solitary	solitary (Sun et al. 2018)
sloth bear	Melursus ursinus	solitary	solitary, but can be seen in groups when resources are plentiful (Amici et al. 2017)
giant anteater	Myrmecophaga tridactyla	solitary	solitary (Shaw et al. 1987)
aardvark	Orycteropus afer	solitary	solitary (Knöthig 2005)
aardwolf	Proteles cristatu	intermediate	live in breeding pairs but forage alone (Koehler and Richardson 1990)
southern tamandua	Tamandua tetradactyla	solitary	solitary (Navarrete and Ortega 2011)
little yellow-shouldered bat	Sturnira lilium	intermediate	solitary or small groups (Fenton et al. 2000)
Sowell’s short-haired bat	Carollia sowelli	social	roost with Glossophaga commissarisi (Kelm et al. 2021)
greater sac-winged bat	Saccopteryx bilineata	social	usually roost in groups, although males are often solitary (Bradbury and Vehrencamp 1977)
Parnell’s mustached bat	Pteronotus parnellii	social	highly social, roost in tight clusters (Clement and Kanwal 2012)
Pallas’s long-tongued bat	Glossophaga soricina	social	social (Hernández-Aguilar and Santos-Moreno 2020)
Davy’s naked-backed bat	Pteronotus davyi	social	roost with many other bat species; rarely roosts alone (Adams 1989)
greater dog-like bat	Peropteryx kappleri	social	roost in colonies (Giral et al. 1991)
greater bulldog bat	Noctilio leporinus	social	roost in groups and forage in groups (Brooke 1997)
great fruit-eating bat	Artibeus lituratus	social	roost together, live in harem groups (Muñoz-Romo 2006)
pygmy round-eared bat	Lophostoma brasiliense	social	3–7 individuals found in 3 roosts (Esquivel et al. 2020)
velvety myotis	Myotis simus	social	shares roosts with Myotis midastactus (Moratelli and Wilson 2014)
common vampire bat	Desmodus rotundus	social	engage in social grooming, food sharing via regurgitation (Wilkinson 1986)
Jamaican fruit bat	Artibeus jamaicensis	intermediate	some form harem groups while others are solitary (Ortega and Arita 1999)
ghost-faced bat	Mormoops megalophylla	social	large roosts (Santos-Moreno and Hernández-Aguilar 2022)
Sinaloan mastiff bat	Molossus sinaloae	intermediate	mixed sociality; solitary males, multiple-male groups, multiple-female groups (Jennings et al. 2002)
proboscis bat	Rhynchonycteris naso	social	roost in groups of 5–11 (Plumpton and Jones 1992)
lesser long-tongued bat	Choeroniscus minor	intermediate	solitary or small groups (2–4) (Solmsen and Schliemann 2008)
rufous-collared sparrow	Zonotrichia capensis	intermediate	immatures and non-breeding individuals gather in small flocks (Rising and Jaramillo 2020)
buff-throated saltator	Saltator maximus	solitary	singles or pairs; may associate with mixed-species flocks (Schulenberg 2010)
common ground dove	Columbina passerina	solitary	usually found singly or in pairs (Bowman 2002)
black-hooded thrush	Turdus olivater	solitary	may congregate at fruiting trees (Collar and Kirwan 2020)
palm tanager	Thraupis palmarum	social	usually travel in pairs or small groups (Hilty 2020a)
lance-tailed manakin	Chiroxiphia lanceolata	solitary	usually found singly but several individuals can gather at fruiting trees or leks (Snow 2020)
silver-beaked tanager	Ramphocelus carbo	social	travel in groups (Hilty 2020b)
ruddy ground dove	Columbina talpacoti	social	non-breeders roost in social groups (Stiles et al. 1989)

Statistical analyses

All statistical analyses were conducted in R Statistical Software version 4.4.2 (R Core Team 2024). We used the packages biomformat (McMurdie and Paulson 2024) and phyloseq (McMurdie and Holmes 2013) to manipulate EMP data. We generated a host phylogeny using the rotl package (Michonneau et al. 2016) and Open Tree of Life data (OpenTreeOfLife et al. 2019).

(1) Do the microbiomes of social species exhibit higher alpha diversity, due to potentially higher horizontal transmission among conspecifics?

We tested the effects of sociality and diet on alpha diversity metrics (Observed OTUs, Chao1, Shannon, Faith PD) while accounting for phylogenetic relatedness among hosts using phylogenetic generalized linear mixed models (PGLMM), implemented in the function pglmm() from the package phyr (Ives et al. 2020). Sociality and diet were included as fixed effects, while host species was included as a random effect with phylogenetic covariance structure. Models assumed Gaussian error distributions and were fitted using restricted maximum likelihood (REML). We used untransformed alpha diversity values since log transformation did not improve model fit nor assumptions; diagnostics revealed that residuals were homoscedastic and deviated only slightly from a normal distribution (SI Fig. 1). To confirm that alpha diversity was not influenced by sequencing depth, we tested for correlation between sequencing depth and diversity metrics using Spearman’s rank correlation. We found no association across all metrics (SI Fig. 2).

(2) Are the microbiomes of social species more homogeneous between individuals, given potentially greater intraspecific microbiome homogenization through social transmission?

To test within-species dispersion, we restricted our analysis to species with a minimum of five samples, which yielded 308 observations across 15 host species. First, we quantified within-species dispersion by calculating the distance of each sample to its species-specific centroid in multivariate space using the functions vegdist() and betadisper() from the package vegan (Oksanen et al. 2012). We calculated distances using the robust Aitchison metric, which accounts for the sparse compositional nature of microbiome data and is robust to uneven sequencing depth due to its centered log ratio transformation (Martino et al. 2019). Then, we used a linear model to test mean dispersion per sociality level, with ‘social’ as the reference level, which we weighted by sample size to account for the fact that species with more samples have a more reliable dispersion estimate.

(3) Does the microbiome composition of social species differ from that of solitary species?

To isolate the possible effect of sociality on microbiome composition, we used a distance-based redundancy analysis (db-RDA) to test the effect of sociality after accounting for known drivers of variation (host phylogeny, diet, and batch effects). First, we calculated beta diversity using the robust Aitchison metric. Then, we performed a Principal Coordinates Analysis (PCoA) on the cophenetic distance matrix derived from the host phylogenetic tree. The first five principal coordinate axes, which explained ~90% of phylogenetic variance, were included as conditioned terms in our db-RDA. Our final model, implemented using the function capscale() from the package vegan (Oksanen et al. 2012), included sociality as a predictor variable conditioned upon host phylogenetic relatedness, diet, and study ID. We assessed statistical significance using a permutational ANOVA with 999 permutations.

As a complement to our db-RDA, we conducted a permutational multivariate analysis of variances (PERMANOVA) using the function adonis2() from the vegan package (Oksanen et al. 2012). The model included host sociality, diet, and five principal components summarizing phylogenetic relatedness on robust Aitchison distances. We assessed statistical significance using permutation tests (999 permutations), testing the marginal effect of social behaviour after accounting for all covariates. Lastly, we tested dispersion differences between social categories using permutational analysis of multivariate dispersion (PERMDISP), implemented in the function betadisper() on robust Aitchison distances calculated with the function vegdist(), both from the package vegan (Oksanen et al. 2012).

(4) Are social behaviour-promoting bacteria such as Limosilactobacillus and Bifidobacterium more abundant in social species?

To test for the presence of candidate sociality-promoting microbes, we identified the per-sample prevalence and relative abundance of Bifidobacterium and Limosilactobacillus OTUs in samples belonging to social, intermediate, and solitary host species. We performed PGLMMs on prevalence and log-transformed relative abundances, with ‘social’ as the reference level. We included host species and host phylogeny as random effects. Presence/absence models assumed a binomial error distribution and relative abundance models a Gaussian error distribution; models were fitted using REML.

Results

The subsampled EMP vertebrate gut metagenome data used in this study comprised 375 samples, encompassing 13 085 267 reads, which were mapped to 34 451 OTUs. Forty-five host species were represented across 375 samples, comprising 35 mammal and 10 bird species (n_samples = 312 and 63, respectively). These species belonged to 13 orders. Sample size varied widely across host species, with a mean of 41 samples per species (median 3; min. 1; max. 905). Mammal samples had a mean of 1019 OTUs per sample, significantly more OTUs than birds, which averaged 807 OTUs per sample (Wilcoxon signed-rank test, W = 6841, p = 0.0001). Mammal samples comprised 85% of all reads, within which 92% of all OTUs were present, while bird samples contributed 15% of reads and 40% of all OTUs.

Mammal and bird gut microbiomes were dominated by the same phyla in different proportions: Firmicutes (39.5% and 39.6%, respectively), Proteobacteria (31.1% and 33.8%, respectively), and Bacteroidetes (18.4% and 14.4%, respectively) (Fig. 1). Differences between mammals and birds were more pronounced at the class level, with mammalian samples dominated by classes Clostridia (24.8%), Gammaproteobacteria (16.8%), Bacilli (13.7%), Bacteroidia (9.3%), and Alphaproteobacteria (7.1%), while avian samples were dominated by Bacilli (39.6%), Clostridia (17.5%), Gammaproteobacteria (17.3%), Alphaproteobacteria (7.3%), and Betaproteobacteria (6.5%).

Figure 1.

Phylum-level abundances of normalized OTU counts in mammals (a) and birds (b) from the EMP gut metagenome dataset.

Alpha diversity and social homogenization

Overall, we did not find that social species exhibited higher alpha diversity. PGLMMs on alpha diversity metrics, with sociality and diet as fixed effects and host species and phylogeny as random effects, revealed no significant effect of host sociality across alpha diversity metrics (Fig. 2; SI Table 1) with one exception: solitary hosts exhibited higher Shannon diversity than social ones (PGLMM, β = 1.43, SE = 0.58, p = 0.014).

Figure 2.

Alpha diversity metrics according to sociality level. No significant difference in alpha diversity was found between sociality levels (PGLMMs with sociality and diet as fixed effects and host phylogeny as a random effect), with the exception of social vs solitary hosts when measured in Shannon diversity (PGLMM, β = 1.43057, SE = 0.58, P = 0.014).

Given the potential for increased horizontal transmission within social species, we hypothesized that dispersion would be lower among social species compared to intermediate or solitary species. However, we found no link between sociality level and mean distance from species-specific centroid (linear model, reference level = social, F = 1.10, P = 0.37) (Fig. 3).

Figure 3.

Within-species dispersion for host species with at least five observations. Distances were calculated using robust Aitchison metric. Mean dispersion did not differ between sociality levels (linear model, F = 1.10, P = 0.37). Sample size (right) ranged from 5 to 41 samples per species.

Beta diversity and differential abundance

After accounting for host phylogeny, diet, and batch effects, sociality was a significant, though small, predictor of microbiome composition (db-RDA, F = 1.77, P = 0.001, 999 permutations), explaining 0.39% of variation (Fig. 4a). This finding was driven by samples belonging to the mammalian subset (F = 1.85, P = 0.001, 999 permutations); the same model on the avian subset was not significant (db-RDA, F = 0.71, P = 0.91, 999 permutations) (SI Fig. 3). Additional testing with PERMANOVA confirmed the significant effect of host sociality on gut microbiome composition (F = 1.55, R² = 0.008, P = 0.004), after accounting for study, diet, and phylogenetic covariates. In this model, diet also explained a small but significant portion of variation (F = 1.35, R² = 0.007, P = 0.015). As a whole, the model accounted for ~5% of total compositional variance. Because diet and sociality were correlated across host species (χ² = 101.8, df = 4, P < 0.001), these factors may capture partially overlapping ecological effects. Lastly, a PERMDISP analysis revealed that sociality levels were associated with differences in variability in microbiome profiles rather than composition (PERMDISP, F = 20.71, P < 0.001) (Fig. 4b). Solitary taxa displayed lower within-group dispersal than both social (Tukey HSD, mean difference = 7.04 units, 95% CI [4.16, 9.92], P < 0.001) and intermediate taxa (Tukey HSD, mean difference = 7.83 units, 95% CI [4.43, 11.23], P < 0.001), whereas there was no difference in dispersion between social and intermediate taxa (Tukey HSD, p = 0.823) (Fig. 4c).

Figure 4.

Drivers of gut microbiome assembly in vertebrate hosts. (a) Host sociality significantly predicted microbiome composition (db-RDA, F = 1.77, R² = 0.039, P = 0.001, 999 permutations). (b) Distance to group centroid (diamonds) in PCoA space by sociality group. Sociality levels were associated with differences in dispersion (PERMDISP, F = 20.71, P < 0.001). (c) Post-hoc analyses revealed that social and intermediate hosts had significantly higher dispersion than solitary hosts (Tukey HSD, mean difference = 7.04 units, 95% CI [4.16, 9.92], P < 0.001 and mean difference = 7.83 units, 95% CI [4.43, 11.23], P < 0.001, respectively). All distances calculated using the robust Aitchison metric.

Candidate prosocial microbes (Limosilactobacillus and Bifidobacterium) were present in the dataset, representing 0.45% of all sequences (mapped to 226 OTUs) in mammals and 0.021% of all sequences (14 OTUs) in birds, respectively. They were, however, neither significantly more prevalent nor more abundant in samples belonging to social species after accounting for host species identity and phylogeny (PGLMM on presence/absence and relative abundance, all P > 0.1) (Fig. 5, SI Table 2).

Figure 5.

Occurrence of candidate prosocial microbes (Bifidobacterium and Limosilactobacillus OTUs) by sociality level. Neither prevalence (presence/absence) (a and c) nor relative abundance (b and d) within individual samples was significantly associated with sociality level after accounting for other drivers of variation (PGLMM with sociality as a fixed effect, host species and host phylogeny as random effects, all P > 0.1).

Discussion

We found that across phylogenetically and ecologically disparate mammalian taxa, sociality is a small but significant predictor of microbiome composition after accounting for known important drivers of gut microbiome variation, namely phylogenetic relatedness and diet. Differences in dispersion among samples, rather than a distinct microbial profile among social taxa, underlie these findings. We found no evidence that prosocial microbes, identified in multiple laboratory studies, are significantly more prevalent or abundant in social species. The significance of our results, despite the very small effect sizes, nevertheless warrants a closer evaluation of how sociality—its modes and mechanisms—may affect host-associated microbiomes.

Generally, we did not find greater microbial community richness in social species (Fig. 2), contrary to our expectation that individuals from social species would harbour a richer microbiome given the increased potential for socially mediated horizontal transmission (Baniel and Charpentier 2024). While horizontal transmission of symbionts may still partly shape microbiome assembly in social species, other physiological and/or environmental factors may be more important in determining the diversity of host-associated microbiomes. Immune system complexity is one such physiological factor that has been shown to be positively correlated with gut microbial diversity (Woodhams et al. 2020). Although we did not include immune system variables in our analysis due to the wide diversity of host taxa, future work focused on monophyletic taxa (to reduce other sources of variation) may be valuable to explore the role of sociality in shaping microbial diversity independent of immunity. Microbiome richness is also associated with diet type, wherein herbivores have more diverse microbiomes than carnivores (de Jonge et al. 2022). Our results confirm the significant effect of diet as a predictor of microbiome composition and suggest that work undertaken within dietary guilds may be better suited to explore the more subtle drivers of microbiome variation, including sociality.

The microbiomes of solitary species were not more dissimilar in intraspecific dispersion than those of social or intermediate species (Fig. 3). We posited that sociality may lead to homogenization of the microbiome between individuals through social transmission and, conversely, that intraspecific variation would be greater in solitary species due to stronger environmental effects on the microbiome. While our results show no greater intraspecific microbiome similarity at higher sociality levels, there is an important caveat: we lack behavioural data. It is impossible to know whether samples from the same species come from interacting individuals who have the potential to exchange microbes. Although challenging to integrate in large-scale and multi-study datasets like the EMP, future work that incorporates social behaviour data will be crucial to understanding the effect of social networks on microbiomes on broader scales.

We found a small but significant effect of sociality on microbiome composition, which was detected in two different statistical tests (db-RDA and PERMANOVA; Fig. 4a). More specifically, we found that dispersion was lower in solitary taxa compared to both social and intermediate taxa (PERMDISP, Fig. 4b, c). Higher interindividual variation in social and intermediate taxa may be due to several factors, including socially mediated microbial transmission. If hosts’ microbial communities are partly assembled through environmental uptake, social interactions may provide additional opportunities for microbial transmission (reviewed in (Sarkar et al. 2020)); given the large size of microbial communities—which number in the thousands—social behaviour may in fact be a diversifying force. In other words, rather than homogenizing microbiomes, social contact may instead increase variability between individuals who have acquired different microbes from the local microbial pool. This may occur without increasing alpha diversity if there is some upper limit to intraindividual microbial diversity. Conversely, social and intermediate taxa may interact with their environment differently from solitary taxa in a manner unrelated to microbial exchange among conspecifics; studies within ecologically similar but behaviourally divergent hosts will be important to explore this further.

Our intentionally simple classification of sociality increases noise within the social category by conflating species that have different potential mechanisms of social microbial transmission. A finer-scaled analysis that considers the type and frequency of social contact—and thus the degree of potential interindividual transmission—is necessary to detect whether, and how, social microbial exchange occurs. Physical proximity and fluid exchange play a major role in horizontal transmission among conspecifics (Gilbert 2015, Yarlagadda et al. 2021, Rose et al. 2023), but the degree to which these occur varies widely among social species. While the diversity of species in the EMP dataset makes the use of a more detailed sociality metric challenging, this should be a focus of future studies. This could be undertaken, for instance, via monophyletic analyses that include a spectrum of potential transmission mechanisms modelled after a hierarchical pathogen transmission framework, from infrequent physical contact to fluid exchange (Collier et al. 2022). In this way, vampire bats who regurgitate in the mouths of conspecifics can be differentiated from herds of deer or flocks of birds, both of which form social aggregates but with very different levels of social contact and potential microbe exchange.

The very small effect size of sociality on microbiome beta diversity may be due to several factors. For instance, there may be seasonal and sex-dependent effects of sociality on the microbiome that are not captured by the EMP metadata. Temporal variation in microbiomes is widespread among wild animals, where it has been attributed to seasonal changes in food availability or metabolic requirements (Lewis et al. 2016, Ren et al. 2017); it is therefore no stretch to speculate that seasonality in social behaviour may also play a role in shaping microbiome composition. Stronger links between microbiome and sociality may exist at particular times of the year when individuals form social groups, e.g. in the case of breeding season ungulate harems or bat maternity colonies. Disentangling temporal and social effects will require longitudinal data collection that explicitly considers seasonal variation in social behaviour.

A cornerstone of laboratory-based microbiome research lies in identifying the effects of specific microbes on sociality, for which we find no evidence here. We did not find that candidate prosocial OTUs belonging to the genera Bifidobacterium and Limosilactobacillus were either more widespread or more abundant in samples belonging to social species (Fig. 5). Our findings challenge the ecological reality of laboratory findings showing the positive effects of Limosilactobacillus and Bifidobacterium on social behaviour (Sgritta et al. 2019b, Buffington et al. 2016, Yang et al. 2017, Dooling et al. 2022), which may be restricted to highly controlled and highly contrived laboratory settings. That said, causal effects of individual microbial taxa may still be present but not detectable among the complex microbiome community dynamics of natural host populations. It remains therefore worthwhile to continue searching for these candidate taxa in field studies by focusing on populations where sociality-based host-microbiome coevolution may have occurred—that is, in host species where prosocial behaviour both confers fitness benefits and affects microbiome composition (Debray et al. 2024).

Despite our detection of a marginal but significant signal of sociality on gut microbiome composition, this work nevertheless relies on broad generalizations that obscure the mechanisms underlying social behaviour. Often, the diversity of social behaviours found across the animal kingdom invites generalizations (Sherwin et al. 2019) that ignore the highly divergent basis for different social behaviours (Rubenstein and Hofmann 2015). Although we followed common practice by employing such a generalization here, a framework that considers the mechanisms underlying social behaviours (e.g. antipredator, foraging efficiency, mate attraction) will be crucial to disentangling the nature and directionality of the link between sociality and microbiome. Recent recommendations aim to distinguish between the incidental effects of sociality on the microbiome (i.e. due to shared environments or diets) from social transmission processes that could underpin the microbiome as an adaptive benefit of sociality (Debray et al. 2024). Although the nature of the EMP dataset does not allow such a distinction, our detection of small but significant results suggests that a finer-scale analysis where a richer assessment of sociality is possible may be promising.

Yet other generalizations were necessary to allow an analysis that combined multiple independent studies. For instance, our analysis pooled together all samples categorized as ‘gut metagenome’, although this combines samples from different parts of the gastrointestinal tract, along which the microbial community is known to vary (Lkhagva et al. 2021). There is undoubtedly variation in collection methods among individual EMP studies (e.g. gut dissection after euthanasia, collection of faeces). Standardized prospective sampling protocols, rather than post-hoc merging of studies, will be invaluable for future comparisons of microbiomes across a range of host taxa.

Conclusion

We detected a marginally significant link between sociality and gut microbiome composition across highly diverse mammalian host species. Sociality level explained a small degree of variation in microbiome composition, with solitary taxa displaying less dispersion than social or intermediate taxa. This may be due to socially mediated structuring of the microbiome; however, finer-scale studies that account for differences in the type and frequency of social contact are needed to understand the mechanisms and consequences of these findings. Contrary to our expectations, the microbiomes of social taxa did not have higher alpha diversity, nor were they enriched in candidate prosocial microbes. Overall, we show that while the EMP dataset provides a valuable framework in which to test the link between microbiome composition and sociality at a high taxonomic level, its broad scope also requires significant generalizations that can obscure analyses. Given the strong laboratory evidence for links between sociality and microbiome composition, large-scale analyses involving diverse natural populations are essential to test the ecological reality of laboratory-based inferences; it is telling that our multi-taxon analysis did not detect any such strong association. As the number of field microbiome studies increases exponentially, standardization of methods, results, and rich metadata collection will enable increasingly sophisticated analyses at taxonomic levels above that of the population. Questions that arise from laboratory findings can thus be investigated through large datasets like the EMP, deepening our understanding of the complex factors governing differences in host–associated microbiomes.

Conflicts of interest

None declared.

Funding

This work was supported by a Vanier Canada Graduate Scholarship to ELT and a Natural Sciences and Engineering Research Council of Canada Discovery Grant to K.C.