Extreme environmental conditions limit the occurrence of macroflora in deserts, resulting in more prominent roles for microbial communities in driving key biogeochemical processes (Makhalanyane et al., 2016). Specialized microbes are common in desert niches, such as those colonizing the ventral surfaces of rocks (Pointing and Belnap, 2012; Makhalanyane et al., 2015). In the zone below translucent rocks, referred to as the hypolithon, photoautotrophic, bacteria-dominated communities have crucial roles in biogeochemical processes by driving primary productivity (Friedmann and Ocampo, 1976) and biomass production (Chan et al., 2012). Photoautotrophs, such as Cyanobacteria, also contribute to soil carbon sequestration in deserts (Cockell and Stokes, 2004) while heterocystous species directly fix atmospheric nitrogen (Hutchins and Miller, 2017). There is direct evidence that these producers also mediate the transfer of nutrients and energy to heterotrophic bacteria (Tracy et al., 2010; Cowan et al., 2011). However, little is known regarding microbial community interactions, although it is recognized that these have implications for understanding biogeochemical cycling in soils (Valverde et al., 2015).

Hypoliths have been the focus of numerous studies on community assembly and microbial functionality owing to their relative trophic simplicity and pertinence in ecosystem service maintenance (Warren-Rhodes et al., 2006; Tracy et al., 2010; Azua-Bustos et al., 2011; Cowan et al., 2011). These studies have shown that hypolithic communities differ in composition from exposed desert soils (Makhalanyane et al., 2013; Stomeo et al., 2013) and other lithic communities, such as endoliths (Pointing et al., 2009; Van Goethem et al., 2016). Taxonomic differences across niches are mirrored at the functional level with recent studies highlighting the diverse nutrient cycling attributes in hypolithic communities in contrast to bare soils. These include, for instance, a wide spectrum of taxa driving nitrification and anammox pathways (Chan et al., 2013). Metagenomic studies have added to our understanding of hypolithic colonists by revealing numerous stress response mechanisms geared for coping with environmental challenges (Le et al., 2016; Vikram et al., 2016). Despite these advances, few studies have explored microbe–microbe interactions within hypolithic communities (Valverde et al., 2015). Such studies have the potential to reveal community dynamics and mechanisms of co-ordinating community function (Barberán et al., 2012; Shi et al., 2016). The spectrum of transcriptionally active taxa in hypolithic communities is also unknown.

Network analysis (measuring non-random co-variation patterns) is a powerful tool for exploring putative ecological associations among species (Dunne et al., 2002; Bissett et al., 2013). Studies of metacommunity networks have revealed the organization of microbial communities under different dispersal rates (Thompson and Gonzalez, 2017), and the inter-dependence of sub-communities that share metabolites, such as sugars and amino acids (Zelezniak et al., 2015). Microbial interactions within soil niche communities have also been explored recently (Barberán et al., 2012; Shi et al., 2016) and have shown that niche communities, such as the rhizosphere, are more complex in their interactions than surrounding soils. Defining the interactions that occur among active microorganisms is also important in understanding the factors that shape functional community dynamics (Gunnigle et al., 2017), particularly for identification of direct species interactions (Faust and Raes, 2012), or shared niches (Deng et al., 2012). Understanding community interactions has relevance for climate change predictions, for example, by indicating how the loss of low abundance microorganisms might alter community function (Lynch and Neufeld, 2015) and how the metabolic activity of important taxa might change (Tracy et al., 2010).

To better understand community interactions in niche environments, we used Illumina amplicon sequencing to assess the total and potentially active hypolithons. Total hypolithic community composition is represented by DNA-derived 16S rRNA gene sequences, whereas RNA-derived 16S sequences indicate protein synthesis potential at the time of sample collection (Blazewicz et al., 2013). We hypothesized that microbial taxa central to carbon and nitrogen fixation in desert niches, such as photoautotrophic Cyanobacteria (Chan et al., 2012), would be highly abundant in the active hypolithic community fractions.

Using this approach, we addressed the following questions: (i) Does the composition of potentially active hypolithic community members differ significantly from the total fraction of these communities? (ii) Are ecologically relevant microbes, such as photoautotrophic cyanobacteria, potentially active in hypoliths? and (iii) Which taxa serve as key components of potentially active hypolithic ecological networks?

The results presented here should be considered with caution given that the methodology used has clear limitations. For instance, desiccation induces changes in the number of ribosomes, and the amount and quality of ribosomes varies with desiccation rates (Angeloni and Potts, 1986), which are expected to be different among the different components of bacterial communities, due, for example, to variation in the production of exopolysaccharides that contributes to water uptake and retention. The number of ribosomes can be high in dormant cells (Blazewicz et al., 2013 and references therein), which are expected to be numerous in deserts ecosystems (Lennon and Jones, 2011).

Notwithstanding the above, exploring the potentially active members of soil communities provides a more integrated understanding than DNA studies of the ecological roles played by microorganisms, by identifying which taxa may carry out biogeochemical cycling in the soil subsurface, and by documenting the putative interactions between potentially active members of the soil environment. Our results extend observations that potentially active and total soil microbial communities differ significantly in structure and composition (e.g., Urich et al., 2008). Comparable results were obtained using different soils from a variety of ecosystem types across the United States (Carini et al., 2016). We found that ca. 76% of hypolithic taxa potentially synthesize proteins at any given time point (Supplementary Figure S1), suggesting that a number of taxa in desert microbial communities are inactive, dead or dormant, and are not expected to contribute to community functionality. Dormancy is a life strategy common to all microbes, allowing non-replicative cells to remain viable during unfavorable conditions (Lennon and Jones, 2011; Lebre et al., 2017). Desert microorganisms are known to adopt dormancy strategies to avoid disturbance or resource limitation (Jones and Lennon, 2010; Dalling et al., 2011), which highlights the importance of identifying transcriptionally active species in desert niches.

Community membership differed significantly at many taxonomic levels between potentially active and total hypolith communities. We provide evidence that cyanobacteria of the orders Pseudanabaenales and Oscillatoriales are among the most active hypolithic community members. The over-representation of these cyanobacterial groups in the potentially active community fraction reflects their probable importance in the maintenance of desert soil carbon and nitrogen budgets (Warren-Rhodes et al., 2007). Heterotrophic bacteria in micro-niches are thought to benefit from the production and accumulation of organic substrates by cyanobacteria, and subsequently contribute to organic matter transformation and mineralisation (Chan et al., 2012).

However, not all cyanobacterial taxa present in the total community were prominent in the potentially active fraction, suggesting that only a subset of cyanobacteria may contribute to carbon and nitrogen fixation in these communities during a light cycle. The distribution, abundance and activity of cyanobacteria is strongly linked to environmental factors, such as temperature, water availability, resource availability, and light (Garcia-Pichel and Pringault, 2001; Garcia-Pichel et al., 2013; Rajeev et al., 2013), and therefore differences in how distinct species of cyanobacteria respond to these factors may explain these patterns. Indeed, we did not find members of the Chlorococcales to be among the most potentially active taxa, and in some Chlorococcales taxa (e.g., Synechococcus and Prochlorococcus sp.) the transcription of ribosomal genes appears to be higher during dark periods compared with light periods (Lepp and Schmidt, 1998; Zinser et al., 2009).

Network graphs, in which nodes represent taxa and edges reflect significant co-occurrences between taxa (Barberán et al., 2012), are a powerful tool to explore taxon coexistence within complex communities (Montoya et al., 2006; Eiler et al., 2012; Williams et al., 2014). The potentially active hypolithic network was highly modular in structure, in agreement with a previous study on ‘total’ Namib hypolith association networks (Valverde et al., 2015). Modular network structures have also been found in microbial rhizosphere assemblages (Shi et al., 2016) and in soil microbes subjected to long-term warming regimes (Deng et al., 2012). The modules found in the potentially active hypolithic network could potentially indicate direct or indirect interactions between microorganisms (Faust and Raes, 2012), but alternatively may be due to shared niches or a high level of phylogenetic relatedness (Lewinsohn et al., 2006). Modularity has been suggested to be an important attribute for ecosystem stability and community resilience, as disturbance events impact modular networks less severely than they do in non-modular structures (Olesen et al., 2007).

Based on ecological theory, each node in the network has a unique role in shaping the topology, which in turn influences the dynamics of the system (Guimera et al., 2007). Most of the taxa that were central to community structure belonged to the Alphaproteobacteria, mainly representatives of the Sphingomonadales and Rhizobiales, which have been suggested to be sources of phototrophy in deserts (Pointing et al., 2009) and nitrogen fixation (Black et al., 2012), respectively. This is particularly noteworthy as most of these nodes had low abundances in the active communities, which may indicate disproportionately large roles for these taxa in the hypolithic niche (Berry and Widder, 2014). These low abundance taxa had multiple, strong, positive interdependencies with heterotrophic and phototrophic community members within the hypolithic niche (Figure 4 and Supplementary Figures S4A–E). Here, positive microbe–microbe interactions may lead to the transfer of energy between trophic levels, or facilitate the cycling of soil nutrients by multiple taxonomic groups (Steele et al., 2011; Berry and Widder, 2014). A similar result has been found in peatland communities, whereby an extremely rare sulfate-reducing bacterium was the most important functional member (Pester et al., 2010).

Positive co-occurrences were the dominant network connections (Figure 4). Supporting evidence suggests that positive interactions are relatively more abundant in desert ecosystems than in temperate biomes as microbial competition (negative interactions) is reduced (Fierer et al., 2012). This could be interpreted as a result of functional interdependencies among taxa or shared productivity under similar environmental stress conditions (Eiler et al., 2012). Soil communities from the hyperarid Atacama Desert also showed an extreme bias toward positive interactions, which indicated that correlated microbes responded similarly to environmental conditions (Neilson et al., 2017).

The loss of highly connected taxa may lead to the collapse of networks into disconnected sub-networks (Dunne et al., 2002), which has implications for community functionality. For example, the disappearance of keystone species may disturb community structure, potentially by disrupting important community functions that are carried out jointly by multiple species (Delgado-Baquerizo et al., 2016). Berry and Widder (2014) noted that the loss of a keystone species from simulated microbial communities directly affected co-occurring species that had positive interactions with keystone taxa. The loss of single functional groups can have detrimental impacts on community functionality, as soil communities experiencing losses in denitrifier diversity were shown to have reduced denitrification activity (Philippot et al., 2013). It is likely that the loss of active keystone members from hypolithic communities would have negative impacts on both the nature of community interactions and the biogeochemical processes that are supported in these niches.

We found that Namib Desert hypoliths seem to be transcriptionally active, but that around 25% of the taxa are in states of apparent quiescence. This further supports recent findings (Carini et al., 2016) stating that the analysis of microbial community DNA alone is a highly unreliable guide as to the composition of active microbial populations. Most transcripts were attributed to photosynthetic Cyanobacteria of the orders Oscillatoriales, Pseudanabaenales, and Synechococcales. Members of the Alphaproteobacteria were also prominent in the potentially active fraction of these communities, which suggests that both Cyanobacteria and Alphaproteobacteria may be central to biogeochemical cycling in these niches. Using ecological network analysis, we identified several keystone taxa within potentially active hypoliths, which are most probably important in maintaining hypolithic community structure. Co-occurrence network analysis emphasizes the predominantly positive relationships among potentially active microorganisms, which we suggest is an important component of community structure and organization.

The sequence data used in this study are available via the NCBI Sequence Read Archive under the BioProject number 395861.

MWVG conducted the analyses and wrote the first draft of the manuscript. DAC contributed formulation of ideas and edited the manuscript. AV and TPM conceived the project, assisted with analyses and edited the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.