The soil in our microbial DNA informs about environmental interfaces across host and subsistence modalities

Abstract

In this study, I use microbiome datasets from global soil samples and diverse hosts to learn whether soil microbial taxa are found in host microbiomes, and whether these observations fit the narrative that environmental interaction influences human microbiomes. A major motivation for conducting host-associated microbiome research is to contribute towards understanding how the environment may influence host physiology. The microbial molecular network is considered a key vector by which environmental traits may be transmitted to the host. Research on human evolution seeks evidence that can inform about the living experiences of human ancestors. This objective is substantially enhanced by recent work on ancient biomolecules from preserved microbial tissues, such as dental calculus, faecal sediments and whole coprolites. A challenge yet is to distinguish authentic biomolecules from environmental contaminants deposited contemporaneously, primarily from soil. However, we do not have sound expectations about the soil microbial elements arriving to host-associated microbiomes in a modern context. One assumption in human microbiome research is that proximity to the natural environment should affect biodiversity or impart genetic elements. I present evidence supporting the assumption that environmental soil taxa are found among host-associated gut taxa, which can recapitulate the surrounding host habitat ecotype. Soil taxa found in gut microbiomes relate to a set of universal ‘core’ taxa for all soil ecotypes, demonstrating that widespread host organisms may experience a consistent pattern of external environmental cues, perhaps critical for development. Observed differentiation of soil feature diversity, abundance and composition among human communities, great apes and invertebrate hosts also indicates that lifestyle patterns are inferable from an environmental signal that is retrievable from gut microbiome amplicon data.

This article is part of the theme issue ‘Insights into health and disease from ancient biomolecules’.

1. Introduction

Soil encapsulates some of the highest biodiversity per unit volume of material as compared with other environmental mediums [1,2]. It houses microorganisms such as bacteria, microeukaryotes, fungi and invertebrate microfauna, while also being an anchor point for plants and a multifaceted resource for larger animal organisms, who derive food or shelter from its crevices. One can think of soil much like water, as a medium that organisms live not only on but throughout, and like occupants of aquatic environments, those that inhabit soil ecosystems take on components of their environmental [3,4]. The microbiome, that is, the microscopic organisms including viruses, bacteria, archaea and single-celled fungi or eukaryotes, is a living component of habitable ecosystems that permeates the structural web of organic matter. Although organisms are defined by the epithelial boundaries that protect internal tissues, defending self from other, infiltration from external elements still occurs. Soil microbiomes are distinctive communities among diverse soil ecologies, and their compositions are the fingerprints of broader ecosystems, such as forests, grasslands, shrublands or deserts [5]. Naturally, the soil microbiome can be found in organisms that consume soil such as termites and earthworms, but soil microbes also infiltrate other arenas that are simply proximate to soil spaces, such as air currents, plant leaf surfaces, animal skin and human domiciles [6–8]. Where these microbes rest on, they also reside in, and host-associated microbiomes sustain a certain small percentage of persistent inoculation from the environment. Whether or not contaminants remain transient or become endemic is still under investigation [9], but they do participate in gene transfers [10–13], if not directly in wholescale community cell replacement. Through faecal microbiome data from human populations spanning a range of living-condition modalities, it might be possible to understand the variation in soil transmission—in this case of soil taxa—to the gut ecosystem.

(a). Anthropological perspective on human microbiome data can inform widely

The question of environmental influence on human host microbiomes can be studied from the robust data generated from dozens of human societies around the world, many of whom are small-scale Indigenous populations occupying the global ‘south’ with varying limited affects from built environments, urbanization, environmental toxins or integration into market economies [14]. Of particular interest to researchers in anthropology is how over time human ancestors have shaped their environment to suit their needs. Studies on human biology in evolution look at the interface between humans and the environment, but the focus is usually on the impact of human activity on the external world [15]. However, humans adapt to environmental challenges such as climate, resource availability, pathogens or xenobiotics, often by modifying their behaviours, which can impact auxiliary genomes found internally in human microbiomes. Owing to a strong predilection for constructed spaces, or built environments, the human-environmental connection depends on a multitude of factors tethered to geography, subsistence and culture. The intensity of association (physical contact, duration, permeation) may determine the prevalence of an environmental signal (environmental microbial molecules) within a host-associated ecosystem. Factors relevant to exposure circumstances include whether environmental interaction comes from a wild, cultivated or urban setting, whether exposure vectors can enter the airways, oral cavity and gastrointestinal tract, or whether exposure is only superficial, such as on washed surfaces. Variations in these modalities would likely impart more distinguishing features defining microbial community structure or diversity across host communities. Humans have dramatically altered how they occupy their environments primarily through subsistence and economic transitions, which has both positive and negative consequences on our own health [16], but overwhelmingly negatively impacts the environment through the loss of biodiversity [17]. When we learned how to kill infectious microbial agents, we were still blind to the widespread benefits their counterparts might perform, and how integrated our own physiologies are with trillions of microscopic organisms that colonize every surface of our bodies and our environments. Renewed efforts to catalogue the (micro)biological world around us has shown us that human activity is costing us the variability inherent to nature [18], and that these costs undercut the stabilizing dynamics of natural ecosystems.

In the wake of massive globalization of hyper-industrialized societies, it is of paramount importance to take action against the collapse of macro- and micro-ecosystems due to the loss of biodiversity. This means that we should look to ancient biomolecules as an archive, offering a glimpse of the ecological status prior to widespread anthropogenic disturbance to our internal and external ecosystems. However, ancient specimens are rare, and those that exist may be too degraded for sufficient recovery of ancient DNA to reconstruct informative ecologies. One major concern when interrogating ancient microbiomes is the issue of contamination, since both authentic-endogenous and contaminant-exogenous molecules are predominately of bacterial origin [19]. Yet we should expect some transient or even opportunistically colonizing bacteria naturally to be found in host-associated microbiome ‘tissues’, such as in dental plaque and calculus, and in faeces, through daily exposure to environmental (air, water, soil) microbiota [2]. In order to know whether ancient microbiomes retain authentic signals of the host environment, we need to know to what extent the environment intrudes upon present-day host-associated microbiomes. Investigating environmental inputs on host-associated microbiomes is also important for understanding the intrinsic connection between organisms and their environment, and can be used as an indicator of disturbance hotspots that are likely to impact host health.

(b). The nature of entanglement between soils in human environments and microbiomes

Soil is ubiquitous in natural landscapes, and while it may be often hidden, it is still the foundation of the built environment in urban settings, and is among the implied actors of ‘environmental interactions'. How humans occupy their landscapes surely impacts their physiology, since the organism–environment interface determines the type of molecular information exchanges that can occur, where those molecules come from, and how they are conveyed. The microbiome is a potent reservoir of rapidly adapting microorganisms that accumulate traits through discrete chemical interactions, most notably horizontal gene transfer [20]. For host-associated microbiomes, these can be regarded as an impressionable symbiotic organ, being at once an integral part of host physiology as well as a signal of the environment, and therefore be an indicator of the environmental interactions that humans experience. More than a decade of research into human host-associated microbiomes has shown that as environments and behaviours with health-mediating effects change, so too are associated microbiome communities, to varying degrees, perturbed [21].

Host-associated microbiomes congregate at epithelial surfaces or in fluids, most notably in breast milk and along the mouth, airway, skin, urogenital tract and intestines [22]. Visualization of the community is accomplished most directly with metagenomics, a kind of horizontal interrogation of multiple species, and represents a transpose of the vertically oriented information field one obtains from single organism genomics. The ability to recover ancient molecules such as proteins [23,24], DNA [25] and lipids [26,27] opens the door to deep time and to date has informed most notably how organisms have evolved, what they have produced, and to whom they related. With recovery of ancient host-associated microbiome molecules [23,28,29], the reconstruction expands to include also features of the environmental, insofar as they pertain to host activities. The combination of genomic and microbiomic information is highly complementary towards understanding who human ancestors were (genomics: population history), the immediate circumstances of their physiological condition (proteomics: immune factors), and their environmental experiences (microbiomics: genetic and taxonomic arrangements).

This analysis probes the hominid (humans and great apes) gut microbiome for similarities to a set of soil microbiome core taxa from diverse ecosystems. The presence, diversity and composition of shared taxa may be evidence of soil microbial infiltration into the host-associated microbial ecosystems. The aim is to apply these findings to expectations about whether ancient DNA of human microbiomes could tell us about the human–environmental interface in evolution. Great ape data from bonobos, chimpanzees and gorillas, as well as termite data from a multitude of taxa (including cockroaches, crickets and one scarab beetle), were used to inform about the presence of soil taxa due to phylogeny, feeding habits and host habitat (see table 1 for sample data summary and electronic supplementary material, table S1, for full sample metadata). In particular, through this work, I aim to expand the ecological and evolutionary plausibility for the existence of soil taxa in the gut microbiome of humans and their ancestors by including data from African great apes, and to benchmark expectations about soil taxa incorporation into host-associated microbiomes in an extreme example by using datasets from many diverse termite species and other invertebrates.

Table 1.

Summary of host-associated datasets included in this study.

host	N (samples with soil core taxa)	study	platform	16S region	community (as specified in original study)
beetle	1	Dietrich [30]	454	v3v4	n.a.
bonobo	70	Moeller [31]	Illumina	v4	LA, ML
chimp	44	Moeller [31]	Illumina	v4	Gombe
cockroach	7	Dietrich [30]	454	v3v4	n.a.
cricket	2	Dietrich [30]	454	v3v4	n.a.
fungal comb	2	Li [32]	454	v3v4	n.a.
gorilla	181	Moeller [31]	Illumina	v4	Campo, Belgique, Djoum, Bipindi
human	27	Ayeni [33]	Illumina	v3v4	Bassa, urban Nigerian
	23	Duboise [34]	Illumina	v4	Montreal, Nunavut
	6	Li [35]	Illumina	v4v5	Tibetan, Han
	78	Obregon-Tito [36]	Illumina	v4	Matses, Norman, Tunapuco
	43	Pires [37]	Illumina	v4	Puruzinho, Buiucu, Rio
	1	Schnorr [38]	454	v4	Hadza
	107	Smits [39]	Illumina	v4	Hadza
	1	Stagaman [21]	Illumina	v4	Shuar
	529	Yatsunenko [34]	Illumina	v4	Malawi, USA, Venezuela
termite	77	Bourguignon [40]	Illumina	v3v4	n.a.
	3	Chew [41]	Illumina	v3v4	n.a.
	15	Dietrich [30]	454	v3v4	n.a.
	14	Kohler [42]	454	v3v4	n.a.
	153	Li [32]	454	v3v4	n.a.
	43	Schnorr [43]	Illumina	v4	n.a.

2. Methods

A methodological work-flow diagram of the following procedures is provided in figure 1.

Figure 1.

Methods workflow of bioinformatics process using 16S rRNA amplicons containing the V4 hypervariable region. The schematic shows the process undertaken to derive the soil-core OTUs and the subsequent abundance tables of host-associated sample data after searching for the soil-core OTUs among their sequenced communities. The first step was to obtain the full set of OTUs at 97% identity using the USEARCH UPARSE algorithm (totalling 16 936 OTUs), and then these OTUs were filtered for only those that achieved a presence in at least 80% of the soil samples from the Global Soil Dataset (a total of 112 OTUs). Then, both the Global Soil Dataset and the Host-Associated Data were searched for the soil-core OTUs, deriving the abundance tables that were used for analysis in R. The taxonomy summaries derived from VSEARCH SINTAX call against the SILVA v132 16S database were used for sourcetracker2 analysis, and the taxonomy derived from the BLAST search against the NCBI 16S rRNA database was used otherwise for annotating the soil-core OTUs. Colour boxes indicate a microbiome dataset (green), a process or software (gold), or a reference database (blue).

(a). Deriving soil-core taxa

The 16S V3-V4 global soil dataset [44] was downloaded from the National Center for Biotechnology Information Short Read Archive (NCBI SRA) and curated following the original study methods, which involved clipping 20 bases from both ends of each read. Unique sequences with a minimum of 5 reads (less than 5 were considered spurious) were de novo clustered into operational taxonomic units (OTUs) at 97% similarity using USEARCH UPARSE [45] resulting in 16 936 OTUs, and the OTU table was built for all soil samples using the VSEARCH ‘usearch_global’ call [46] at 97% identity to the OTU representative sequences. A set of 112 core OTUs were found among the soil samples, defined as OTUs that were present in at least 80% of samples (183 out of 229 samples). Taxonomy of the soil-core OTUs was obtained by a BLAST search against the NCBI prokaryotic 16S rRNA database, taking top hits only, with a per cent ID score greater than 84% (electronic supplementary material, table S2).

(b). Finding soil-core taxa among host-associated datasets

Host-associated 16S rRNA datasets (electronic supplementary material, table S1) that sequenced at least a portion of the V4 region were selected in order to search for presence and abundance of the derived ‘soil-core’ OTU dataset [21,30,32–34,36–39,41–43,47–49]. The soil-core OTUs were used as the reference database with VSEARCH ‘usearch_global’ call at 97% identity (terminal gaps are excluded from the identity score), resulting in dataset-specific OTU tables of the counts of soil-core OTU sequences per sample. The resulting soil-core OTU count tables were merged and used for downstream analysis in R.

(c). Verifying soil-core taxa

To validate that the soil-core OTUs are not environmental contaminates, sourcetracker2 [50] was used to identify the soil-core components of the original soil samples from other host-associated and environmental microbiomes. Data used for the ‘sinks’ included the original global soil dataset samples [44]; datasets used as the ‘sources’ included human hunter–gatherer (HG) [36,38,51], rural, urban and pastoralist gut microbiome data [37,51–53]; human skin microbiome data [54]; great ape microbiome data [48]; and two soil microbiome datasets (including the original global soil dataset) [5,44,55]. All datasets were run through de novo OTU picking at 97% using VSEARCH implementation of USEARCH UPARSE, and taxonomy mapped against the SILVA v132 16S database [56] using VSEARCH SINTAX with confidence cut-off specified at 0.8. All resulting relative abundance taxa tables were summarized at the genus level (L5) and then merged using QIIME merge OTU tables [57]. The unassigned bin was removed prior to running sourcetracker2. The sourcetracker2 gibbs command was run with default parameters except that source rarefaction was set to 0 because of prior normalization.

(d). Tracing compositional similarity of soil-core from host sample sets

To understand whether soil-core taxa found among the host-associated microbiomes could be identified in composition to soil profiles, sourcetracker2 was again used, this time on the soil-core taxa tables from the V4 region 16S rRNA datasets where some samples could be rarefied to 100 sequences [30,32,33,39,40–42,47,48]. The taxonomy for the soil-core OTUs was found using the VSEARCH SINTAX call against the SILVA v132 16S database (done previously for verifying the soil-core taxa), and then the taxonomy annotations were added to the soil-core OTU tables, rarefied to 100, summarized at the genus (L5) level, and then merged. Sources included three soil microbiome datasets [5,44,55], a human skin dataset [54] and human gut microbiome for three subsistence specifications (foraging, rural farming, industrial agriculture) [36]. Source-tracked proportion tables were sorted by the proportion of ‘unknown’ composition and samples evaluated if at least 25% of the community profile could be traced to a source dataset.

(e). Analyses

Analyses conducted on the merged soil-core count table and the alpha-diversity calculations were handled in R v. 3.4.1 using the {vegan}, {made4}, {propr} and {reshape2} packages [58–63]. Visualization of results were achieved using the base R graphics alongside the {vioplot}, {scales}, {colorspace} and {ggplot2} packages [64–66]. The input core count table was normalized relative to the number of total reads for each sample. In the human samples, where age was provided with the associated metadata, samples were binned into three lifestages: ‘child’ (aged 0–10 years), ‘adolescent’ (aged 11–15) and ‘adult’ (aged greater than 15). Statistical significance for separation between groups was calculated using permutational multivariate analysis of variance (PERMANOVA) with adonis in the {vegan} package.

(f). Alpha-diversity analysis

Alpha diversity for observed OTUs and Shannon diversity were calculated using QIIME on the merged soil-core count table from multiple rarefactions to a depth of 1000 observations, and resulting matrices were imported to R for aggregation and visualization. Rarefaction curves were created using the rarecurve function in {vegan} on the normalized non-rarefied count table (electronic supplementary material, figure S1A). The observed OTUs and Shannon diversity matrices were transformed using {reshape2} to long form. First, by-sequence alpha-diversity values were obtained by aggregating the mean values across all samples at each sequence depth and plotted to visualize the effect of sampling, indicating that diversity value is highly impacted up to a depth of approximately 300 sequences (electronic supplementary material, figure S1B), given that few samples among all queried datasets had greater than 1% of soil-core sequences (electronic supplementary material, figure S1C). Alpha-diversity values at a depth of 300 observations were visualized in boxplots, but this strategy omitted 712 out of 1015 samples. Instead, the maximum diversity values were found for each individual sample and used for boxplot visualization, but as a result, these calculations should be taken with caution (electronic supplementary material, figure S1D). Given these limitations, a projection of the diversity values was accomplished using the {vegan} richness estimator function, specaccum, which shows species accumulation curves based on the ‘exact’ method of expected/extrapolated accumulation from a moment-based estimation of the mean. Richness estimations projects the number of expected species by site (as opposed to by sequence in rarefaction curves), which was accomplished on subsets of the original count table (by-host, by-subsistence and by-age).

(g). Beta-diversity analysis: ordination, heatplots, composition and prevalence

Principal components analysis (PCA) was conducted on the centre log-ratio (clr)-transformed data, first using the {propr} package and propr command with [ivar = ’clr’] to obtain the log-ratio data frame, and then using the capscale function from the {vegan} package, specifying the LHS as the log-ratio data frame and RHS as ‘approximately 1’ with [distance = ’euclidean’] to obtain PCA with no constraints. Since there was a high degree of sparsity and 0 counts, the table was also filtered for samples that have at least 0.1% of total reads assigned to a soil taxon prior to ordination and clustering of the whole dataset, which reduced the number of samples from 1278 to 324, but gave a more accurate representation of sample clustering by composition. Feature gradients represent the number of unique soil-core OTUs (features), which were fitted onto PCA plots using ordisurf function of {vegan}. Heatmaps were generated from the clr-transformed data using the {made4} package with the ‘ward.D2’ clustering method. Prevalence heatmaps were generated by taking the top-half most-abundant OTUs (56 total out of 112) for the host-associated samples. Host groups with less than 10 individuals were removed (beetle, cockroach, cricket and fungal comb), leaving large host groups for prevalence mapping. The heatmaps show the per cent of prevalence (presence/absence) of soil-core OTUs for each host group and each human community, respectively, with the soil-core prevalence in the original soil samples as comparative.

3. Results

In the earlier work by Delgado-Baquerizo et al. [44], which looked at the dominant bacterial communities from a global atlas of soil ecosystems, roughly 2% (approx. 500) of all bacterial OTUs found among soil microbiomes represented nearly half of the actual bacterial abundance in these communities. The ‘core’ OTUs were thus defined as the top 10% most abundant OTUs that were found in greater than 55% of soil samples. Here, in the present analysis, a highly abundant set of ‘soil-core’ OTUs were found using a more stringent selective criteria, aimed at capturing truly ubiquitous soil taxa that may be traceable to host-associated microbiomes. The soil-core OTUs were determined here to be OTUs found in at least 80% (183 out of 229) of the global soil microbiomes from the same soil samples previously reported in Delgado-Baquerizo et al. [44], encompassing a wide array of ecosystems from dry grasslands to tropical forests. These soil-core OTUs represented 0.66% of the total number of OTUs (112 out of 16 936 OTUs), but on average, 27.5% (7.4%) of total OTU abundance of total identified OTUs among all soil samples (minimum 6.5%, maximum 42.5%) (electronic supplementary material, figure S2A). Sourcetracker2 analysis was used to trace soil-core profiles of each global soil sample in order to understand whether soil-core taxa were actually reflective of soil ecosystems, or whether they were contaminants, perhaps from reagents or airborne particles. The soil-core OTU abundances for global soil sample types do in fact map on to soil microbial ecosystems rather than host-associated microbiomes (approx. 99% soil identity), and recapitulate the soil ecosystem from which the samples came (electronic supplementary material, figure S2B). The soil-core OTUs served as the representative OTU database against which diverse human- and animal-host gut microbiome communities were mapped at 97% identity (see figure 1 for workflow depiction). The resulting abundance tables demonstrate that there is some component of soil-core taxa that are prevalent in host-associated gut microbiome communities (figure 2), potentially deriving from direct contact with soils (a proxy for the intensity of interaction with the natural environment).

Figure 2.

Heatmaps of prevalence of 56 top-most abundant OTUs. The prevalence (the per cent of samples with the OTU present) of the top-half most abundant soil-core OTUs as measured among host samples (a) and then within humans (b) shows the differential distribution of soil-core bacteria based on the host category and community, respectively. Gorillas and then chimpanzees appear to have the most numerous consistently present taxa, while the bonobo, human and termite host samples have lower group-based prevalence, indicating a more dispersed, soil-core representation. The soil-core OTU prevalence in the soil samples is expectedly near to 100% across all OTUs, reinforcing the notion that these soil-core OTUs are indeed ubiquitous in global soil ecosystems. Taxonomic annotation of OTUs is based on top BLAST hit from the full NCBI 16S rRNA database.

Overall, diverse hosts, which included humans, African great apes, termites, cockroaches, crickets, and one species of beetle, significantly differed in the amount, diversity and distribution of soil-core OTUs (PERMANOVA, R² = 0.223, p < 0.001). An overall picture of the effect of sequencing depth on the diversity of soil-core taxa indicates that the evenness of soil-core taxonomic abundance decreases as more new taxa are found, which can be expected from sparse or rare data. Since Shannon diversity decreases concordant with an increase in OTU count for soil-core taxa (electronic supplementary material, figure S1B), then indeed the event of soil-core taxonomic presence in host-associated gut microbiomes is rare, and the distribution is sparse. To drill down into more specific host effects, it was necessary to use diversity projection curves. Richness estimations, or species accumulation curves, project the number of expected species with the addition of each new site (individual host). The non-human great apes exhibit the highest estimation of species richness (but with varying accumulation rates) while humans (aggregated) and the termite species (distinguished by whether or not they farm fungus) have more gradual accumulation (figure 3a). Humans and non-fungus-farming termites appear to have a slower rate of accumulation but less severe plateau, while the non-human great apes and the fungus-farming termites indicate plateau phases following varying rates of early accumulation. These differences likely relate to the fact that the human and non-fungus-farming termites have more heterogeneous intraspecific subsistence practices within their grouping [67,68]. The mean relative abundance of the soil-core taxa summarized by host also show distinctive variations in the distributed proportions of present soil-core taxa (figure 3b; see electronic supplementary material, table S2 for by-host soil-core OTU proportions) with some hosts dominated by just one or two taxa (beetle, cricket, fungal comb, gorilla and human) and others having higher representation by many taxa (bonobo, chimp, cockroach and termite). Additionally, the proportion of sample sequences that mapped to soil-core OTUs ranged between less than 1% to 20%, and were highest surprisingly among the chimpanzee, gorilla and human hosts, rather than among termites and other invertebrates (figure 3c), which was unexpected and may indicate any number of phenomena such as feeding habit, host foraging selectivity, environmental disturbance from host activity, or potentially, contamination during sample collections. The soil-core taxonomic compositions are clearly distinguished between environmental (soil) and host microbiomes, and also separated by host species (figure 3d; PERMANOVA, R² = 0.192, p < 0.001), but these profiles are actually best explained by the more granular categorization of ‘community’ (i.e. country, region or ethnic group) (PERMANOVA, R² = 0.32, p < 0.001). When richness estimates are viewed for human data alone, stratification is apparent based upon subsistence type, but also for lifestage, defined as ‘child’, ‘adolescent’ and ‘adult’ (see Methods for binning criteria; figure 4a,b). Rural-farming and hunter–gatherer (HG) groups have greater accumulation of taxa per site than do urban-living humans, which conforms to the hypothesis that direct interaction with earth soil is a necessary condition for soil bacteria to infiltrate the gastrointestinal tract. Curiously, samples assigned to children also accumulate soil-core taxa richness faster than those assigned to adolescents and adults. Children tend to interact with their environment more, often orally, at very young ages, but also through play and exploration phenotypes that characterize this lifestage, and which may continue into adolescence. It is expected that children and adolescents would have greater soil-core taxonomic richness than adults; however, adolescents are observed to have a slightly lower richness accumulation rate than adults, but still fall entirely within the 95% confidence area. This effect may simply be a matter of a small age range for adolescents, therefore encompassing a smaller span of (and fewer) individuals than the other binned lifestage categories. Viewed by age, the number of soil-core taxa observed appears to have a general negative association with increasing age in humans (Pearson correlation, ρ = − 0.09, p = 0.018). When this relationship is looked at specifically within individuals from three different countries (USA, Malawi, Venezuela; from Yatsunenko et al. [34]), the pattern is upheld mainly due to observations on individuals from the USA (Wilcoxon, p = 0.019; figure 4c). This result is good evidence for the notion that while an urban lifestyle may reduce soil-bacteria exposure overall, children may still be exposed to soil from natural environmental exploration, and the contrast between children and adults in urban settings is most apparent. There are also more sampled individuals from the USA, which may bias this analysis. Human community microbiomes from populations living as hunter–gatherers, subsistence farmers and foragers, pastoralists, and in urban-industrialized settings also showed marked variation in the abundance and distribution of soil-core taxa, which is also best explained by community-level assignments (PERMANOVA, R² = 0.37, p < 0.001; figure 4d), yet taxonomic distribution did not have any apparent relation to age or lifestage (PERMANOVA, R² = 0.039, p = 0.129).

Figure 3.

Soil-core OTU distribution among host samples. The projected richness from the number of soil-core taxa found by site in host-microbiome communities (a) shows African great apes to harbour the highest richness per site, but human and termite sample projections indicate greater variation in by-site richness. The soil-core relative abundance averaged by host (b) shows distribution of abundance across the 112 soil-core OTUs. The proportion of soil-core OTUs found in all sequences is shown averaged by host (c). Clustering based on soil-core abundance tables shows that samples cluster by host, but stochastically, and are highly interspersed, depending on whether taxonomic abundance is concentrated in a few or many OTUs (d).

Figure 4.

Human soil-core taxonomic profiles by subsistence, community and age. Rural farming communities and hunter–gatherer communities have much higher richness projections than urban communities (a), while rural arctic and pastoral communities are under sampled but appear to have a lower curve trajectory. Children have higher richness than adolescent or adult individuals (b), which corresponds to a negative linear trend in the number of soil-core OTUs by age and is particularly well-represented in lifestage groupings among samples from the USA (c). Distribution of samples by clr-transformed abundance table of soil-core OTUs shows clusters by subsistence (top colour bar, left-hand colour key) and by community (bottom colour bar, right-hand colour key), primarily differentiating hunter–gatherer and some rural farming samples from urban and other rural groups (d).

Ordinations were used to visually explore the link between different hosts and their soil-core microbial distribution to understand whether patterns might suggest an effect of lifestyle and environment, or whether the soil-core presence is an extended phylogenetic trait, as has been suggested of gut microbiomes in past works [30,47,69,70]. Continuing this exploration with the addition of the great ape and termite datasets helps to discern the ecological (termite) and evolutionary (great apes) plausibility for soil taxa to arrive in the human gut microbiome as a result of proximity to the environment. Principal component analysis of Euclidean distance of the clr-transformed soil-core abundance table, filtered for individuals achieving at least 0.1% total soil-core sequence abundance, indicates host-dependent clustering to some degree. The full dataset shows clustering contingent on ecology (soil versus host-associated microbiomes) and host species (figure 5a). Of interest, there is interspersion between some human samples among great-ape and termite samples, but extreme separation between some human and gorilla samples obliterates finer differentiation among other hosts (figure 5a). When the dataset is reduced to only those studies containing individuals identified as having a hunter–gatherer subsistence [36,38,39], it is possible to see some meaningful compositional-based stratification of samples, which in part relates to the community and subsistence practice (figure 4d). These human samples derive entirely from datasets contributing hunter–gatherer, mixed subsistence, or rural farming group sampling, alongside samples from westernized urban communities for comparative work. Isolating the human data for ordination again depicts clustering by community and country of origin for the individuals (figure 5b), which of course relates to subsistence, lifestyle and geography.

Figure 5.

Ordination of clr-transformed soil-core OTU abundance table. Soil-core clustering showing soil and host-associated soil-core OTU distribution indicates major separation between the environment and the host-associated ecologies, as well as between host organisms (a). Human community datasets were evaluated independently and also depict some community-based clustering (b and c). Comparative analysis looking at human sample stratification within the same study (that depicted both small-scale and westernized communities) clearly differentiates the hunter–gatherer groups (Hadza and Matses) from rural farmers and westernized urban-living individuals (Tunapuco and Norman; b). Comparing all human samples from across all studies (filtered for samples achieving at least 0.1% of soil-core OTUs in the total microbiome community) shows HG and some rural farming groups separate from a tight cluster of mostly individuals from the USA and some from small-scale Venezuelan communities (c). The deviations from clear subsistence patterned clustering here may be due to the inclusion of young children. The gradient (green topographical lines) indicates the OTU features (richness) distribution across the plot space, and is not instructive for explaining the distributions.

Finally, the soil-core taxonomic composition of interrogated sample datasets was analyzed using sourcetracker2 software in order to understand whether these profiles bore any relation to non-host-associated soil microbiome communities versus host-associated human gut communities, and human skin communities, or if instead they appear as only random accumulations of environmental contaminants. Datasets submitted for source tracking included only those whose sample soil-core OTU tables could be rarefied to a minimum of 100 sequences, since extremely sparse tables would not be traceable to a source community profile. Even with such a low threshold criteria, only two human datasets consisting of individuals from rural and urban Nigeria (Ayeni et al. [33]) and individuals from Malawi, Venezuela and the USA (Yatsunenko et al. [34]) could be included on account of the overall high-frequency of a low per cent of soil-core OTU prevalence in accessed datasets (figure 3c; electronic supplementary material, figure S1C). In addition, the African great ape and termite datasets [30,32,41,46] were included for comparison and to better understand the environmental predictive power of the recovered soil-core community in each host. Of the 388 ‘sink’ samples, a total of 34 contained a greater than 25% identifiable taxonomic profile, all of which map to soil environments rather than human-associated microbiome communities (figure 6). The highest proportion environmental match also closely matched the actual environment from which the sample originates: human communities from rural Nigeria, Venezuela and Malawi were traced to dry forest soil communities; bonobo and chimp samples also matched to dry forests, while gorillas largely matched temperate forest communities; termite samples matched dry forest, dry grassland and temperate forest communities. These results are encouraging evidence towards the hypothesis that the environment can be a source of endogenous molecular diversity found among gut microbiome genetic data, in modern and in ancient samples. Furthermore, universally shared soil microbial traits are apparently spread throughout environments and animal host-associated microbial ecosystems, implying that we should expect the environment to be a reservoir of conserved microbial information (e.g. taxa, genes, proteins, metabolites). According to these results, supported through source-tracking analysis, microbial molecular information does arrive in the host-associated gut microbiome, and in patterns suggestive of a fairly uninhibited pathway through the animal digestive tract, dependent upon patterns of exposure that relate to community-level lifestyle properties. Therefore, I answer the question posed by Blum and colleagues, ‘Does soil contribute to the human gut microbiome?’ in the affirmative [2].

Figure 6.

Sourcetracker2 identification of soil-core profiles from host-associated microbiomes. For samples that achieved greater than 75% identification from sourcetracker2 analysis, the soil-core profiles are maximally representative of soil communities (shown in the cool to warm colour gradient) and do not map on to any host-associated microbiome sources provided in the analysis (discrete categorical colours: sky blue, human skin; tan, Tunapuco rural farmers; green, Matses HG; red, westernized urban-living individuals in Norman, Oklahoma, USA).

4. Discussion and conclusion

The concept of a core microbiome community is widely applied to host microbiomes, but it also remains valid for environmental microbiome communities where certain structural or metabolic parameters are shared. This assumption has been previously confirmed through identification of a core soil microbiome community that is shared among globally distributed soil microbiomes from diverse ecosystems [44].

The results of this initial study on the presence and pattern of a soil-core microbial component in host-associated microbiomes suggest that a fingerprint of an individual's environment in the form of soil-associated bacterial taxa may be found among gut microbiome sequence data. This has important implications for studying materials for traces of ancient molecules, particularly from sources that may contain host-associated microbial elements such as latrine sediment, coprolites, soil sediments and dental calculus. Identification of soil bacteria in endogenous sample data may be troubling from the standpoint that soil microbial elements integrated within a microbiome are not readily distinguishable from environmental contamination (in fact, the definition of contamination is hazy in this regard), but this work serves as a proof of concept that more sensitive analyses to detect soil-origin microbiota would be a rewarding venture for a variety of reasons, including ancient environmental reconstruction, understanding factors in microbiome assimilation, and clarifying the medium of environmental molecular transmission to host-associated ecosystems.

Soil and other ecological features such as water and air are important reservoirs for microbes that impact host physiology [6,7,10,18] or seeding obligate symbioses [71,72]. A study of edible fungus-farming termites found one novel termite Treponema strain that clustered separately with Treponema strains found in human hunter–gatherer and rural farming groups, lending support to the idea of a shared universal reservoir for rare or novel bacterial strains in host-associated microbiomes [43]. Such a finding hints at the possibility that environmental taxa play important roles in structuring host-associated microbiomes, and that ‘infiltrators’ are regularly transiting the gastro-intestinal tract and contributing to the molecular cross-talk, or even directly participating in metabolic activities within the lumen. One potential issue of note, however, is the uncertain possibility of post-depositional environmental contamination, since sample collection methods across studies are not consistent and are rarely specified in the original publications. Human and African great ape data came from faecal samples, and invertebrate data came from homogenized whole-gut isolates. The African great ape samples were collected from deposits on the forest floor, and so it is plausible that these could be contaminated with soil bacteria after evacuation. However, if that were the case, a more consistent base-line level of soil-core bacterial presence in the African great ape specimens would be expected, which would be set apart from the rest of the samples. Instead, not only do bonobo samples fail to show a comparable proportion of soil-core sequences (less than 2%), there is a species-specific distribution of soil-core OTUs among the great apes with some overlap (figure 3b–d), similar to what is seen in the original paper results on the full taxonomic distribution of the microbiome [31]. The omnivorous and fungus-farming invertebrates also intersperse with a cluster of the great ape data (figure 5a), suggesting a separate mode of integration of soil-core taxa, since invertebrate samples were removed whole from intact specimens. Still, if soil taxa inoculation to host-associated microbiomes is to be studied further, more careful documentation on the manner of sample collection, including the intermittent context of the sample, between the deposit and the stabilized collection vessel, is critical.

Of the main findings in the present analysis, a number of observations stand out that warrant further exploration. First, it is notable that using a fairly constrained sample set (only datasets for which the V4 region of the 16S rRNA gene is available), it was still possible to witness host-species, subsistence and community-based clustering of the soil-core abundance tables. This means that a number of factors influence the infiltration of soil taxa to the host gut microbiome that can be informative for attempting to reconstruct the ecology of an unknown microbiome sample from ancient remains. Individuals from rural human societies that farm or forage for subsistence contain a higher abundance of total reads mapping to soil-core taxa, and also greater richness of taxonomic units, than do urban-living individuals. However, samples deriving from individuals living in a rural arctic community or who live in a pastoralist society had very few reads assigned to soil-core OTUs and could not be used in subsequent analyses due to the sparsity of data. Technical factors could certainly influence this finding, but so far this observation is consistent with the overall picture from previous works of gut microbiome biodiversity found among individuals from these communities. The gut microbiome biodiversity from some high-altitude pastoralist groups as well as traditional populations in arctic environments is comparable with individuals from market-integrated communities [34,73]. In addition, environmental factors (such as sourced drinking water) can swamp variation due to diet or subsistence practice [52]. Rarefaction curves shown by host are congruent with a scenario in which close-association or feeding on soil (such as in the case for termites) permits high amounts of soil taxa to mingle with host-associated microbiomes.

A second important finding is that the ability to harbour high abundance and diversity of soil taxa in a hominid gut is phylogenetically feasible, given that chimpanzees had the greatest number of sequences assigned to soil-core taxa and had the greatest richness by sequence and by site. This finding is interesting but puzzling, since both bonobos and chimpanzees typically feed from ripe canopy fruits and share activity times occupying terrestrial and arboreal spaces, while gorillas are ground floor occupants and feeders [74]. Further exploration of great ape data for similar soil taxa should be highly illuminating towards understanding what aspects of hominid host ecologies stimulate environmental inoculations. Also expanding this exploration to include cercopithecine monkeys would be important for understanding the role of diet versus activity habits on patterns of soil taxa presence [75,76].

Third, the soil-core taxa profiles in human samples varied depending on human community (a combination of subsistence, lifestyle and geography) and depending on lifestage (child, adolescent or adult) in patterns that are predicted from a hypothesis of environmental exposure-mediated bacterial incorporation. The samples from rural farmers contain the greatest abundance and richness of soil-core taxa, even more than the two HG communities included here. Again, though, there are technical limitations—fewer HG samples and fewer total reads from those samples. Still, more soil-core taxa found among rural farmer samples may make sense if these people must work with soil to cultivate plants for their livelihood. Predictably, samples from urban-living individuals are well below richness estimates of the other subsistence categories. Ordination clustering indicates that the composition of the soil-similar OTUs found in humans is driven by their social community category, which can be a rough proxy for subsistence. Regarding differences seen by lifestage, children (aged 0–10) had the highest richness by site as is expected because they tend to explore their environment more, and this proclivity transcends whether they were from rural or urban or hunter–gatherer groups. This universal pattern is likely a critical stimulus towards the development of microbiome diversity and host immune system training. Curiously though, the composition of soil-core taxa did not relate to lifestage, only the quantity and richness of taxa were significantly influenced by age.

Finally, sourcetracker analysis on the soil-core abundance tables derived from host gut microbiome data demonstrates that the soil-core component of host gut microbiomes are not aberrant environmental contamination. The identifiable components of soil-core profiles maximally represent soil ecosystems and negligibly resemble other host gut or skin microbiomes. When the soil-core profiles of the original global soil samples themselves were analysed with sourcetracker, the soil-core profiles were overwhelmingly identified as soil ecosystems and recapitulated their origin environments. If the sourcetracker analysis on the soil-core profiles from the original soil datasets had resulted in unmapped or unknown compositional identities, then the soil-core taxa may have been volatile environmental contaminants rather than actual conserved ubiquitous soil taxa.

On the basis of these findings, two exciting prospects come to light: (i) soil taxonomic information recovered from unknown host microbiomes may help identify an environmental setting in which the host may have lived, aiding the endeavours to learn about ancient host environment, physiology and microbial ecologies; and (ii) universal soil bacteria (or other global soil molecular traits) may have had a consistent role in guiding the development of the human microbiome and immune system throughout evolution despite human global dispersals.

Data accessibility

Data accessibility is provided from accessions in the original manuscripts.

Competing interests

The author declares no competing interests.

Funding

This work was supported by the National Science Foundation SBE Postdoctoral Research Fellowship (award no. 1810060).