Cell-to-cell bacterial interactions promoted by drier conditions on soil surfaces

Significance

Despite a high number of microbial cells and species present in small volumes of soil, detailed observations suggest that most bacteria interact with only a few other individuals. These interactions between cells are crucial to many soil processes including imparting genetic traits (e.g., antibiotic resistance) and microbial evolution, yet our understanding of the controlling environmental factors remains sketchy. Based on evidence from experiments in soil microcosms and results of a mathematical model, we demonstrate that a ubiquitous physical factor such as fragmentation of the aqueous phase commonly found in unsaturated soils affects the ranges and frequency of cell-to-cell bacterial interactions. Our findings thus help reveal some of the basic principles that control microbial life and diversity in soil environments.

Abstract

Bacterial cell-to-cell interactions are in the core of evolutionary and ecological processes in soil and other environments. Under most conditions, natural soils are unsaturated where the fragmented aqueous habitats and thin liquid films confine bacterial cells within small volumes and close proximity for prolonged periods. We report effects of a range of hydration conditions on bacterial cell-level interactions that are marked by plasmid transfer between donor and recipient cells within populations of the soil bacterium Pseudomonas putida. Using hydration-controlled sand microcosms, we demonstrate that the frequency of cell-to-cell contacts under prescribed hydration increases with lowering water potential values (i.e., under drier conditions where the aqueous phase shrinks and fragments). These observations were supported using a mechanistic individual-based model for linking macroscopic soil water potential to microscopic distribution of liquid phase and explicit bacterial cell interactions in a simplified porous medium. Model results are in good agreement with observations and inspire confidence in the underlying mechanisms. The study highlights important physical factors that control short-range bacterial cell interactions in soil and on surfaces, specifically, the central role of the aqueous phase in mediating bacterial interactions and conditions that promote genetic information transfer in support of soil microbial diversity.

Bacterial cell-to-cell interactions sustain key evolutionary and ecological processes in all environments, including horizontal gene transfer mediated by conjugative pili (1, 2), nutrient cross-feeding (e.g., in syntrophy; ref. 3), and chemical signaling that shapes social behaviors (4). To interact, microorganisms have to remain in direct physical contact or within a range permitting effective molecular exchanges by diffusion, i.e., most interactions occur at the scale of individual microbes (5). Ultimately, cumulative cell-to-cell exchanges determine the overall bacterial activity in a given habitat, affecting large-scale fluxes and, hence, impacting global ecosystem processes (6).

Bacteria that inhabit natural porous media such as soil experience life in a complex 3D pore network and on rough surface architecture, with heterogeneous nutrient resources and fragmented aqueous niches that limit their distribution, dispersion, and contact with neighbors, but can locally increase cell density and proximity (7–11). Evidence suggests that (i) the distribution of microorganisms in soil is highly patchy, with nonrandom microhabitats colonized by single cells or microcolonies (12–15), and that (ii) although soil microbial density is high in comparison with other habitats, cells occupy no more than a few percent of the pore spaces of an average soil and only a small fraction (<1%) of the available soil surfaces (9, 16, 17). This spatial context—together with microbial growth and motility (18)—controls the probability for a bacterial cell to encounter another cell, and therefore to interact, in soil. Based on a detailed analysis of bacterial distributions in hundreds of soil thin sections and on statistical modeling, Raynaud and Nunan (19) have concluded that the potential for cell-to-cell interactions in soil is relatively limited, both in terms of the number of interacting cells and of the number of different species to interact with. The vicinity of a soil bacterium typically harbors few neighbors (10 to 100 cells within a radius of 20 μm). For certain high cell density “hotspots,” e.g., in the rhizosphere (19), the number of neighbors can increase significantly. Although the distance to the nearest neighboring cell can be relatively small on average (∼10 μm in densely populated topsoils), it is highly spatially variable, and, since soil bacterial distributions show a high degree of clustering, colonized microsites are often spatially isolated (19). To sum up, the level of bacterial cell-to-cell interactions in soil may be far more limited than would appear from the high cell density values commonly reported (10⁹–10¹⁰ cells·g⁻¹) (20) due to the high surface area per volume of soil and the clustered spatial distribution of soil bacteria. In this background of limited opportunities for cell-to-cell interactions, additional environmental factors regulate the frequency and duration of cell-to-cell encounters in soil. Developing a better mechanistic understanding of how key factors affect encounters among bacterial cells is important for revealing basic principles that govern soil microbial community diversity, dynamics, and functioning.

In this study, we hypothesize that the probability of bacterial cell-to-cell encounters and interactions on hydrated rough surfaces such as found in soil is modified by soil properties and hydration conditions. We thus seek to link macroscopic variables such as the soil water potential, soil pore spaces, and internal surface area to biological interactions that take place in aqueous microhabitats. Specifically, we propose that cell-to-cell bacterial interactions could be promoted under hydration conditions that locally increase the likelihood and duration of cell encounters while limiting cell dispersal. To systematically investigate this question, we have used the exchange of a conjugative plasmid as a proxy for close encounters between bacterial cells, with the soil bacterium Pseudomonas putida (21) selected as a donor and recipient of a broad-host range plasmid isolated from soil environments (Fig. 1). Sand microcosms were used as simple and well-defined porous environments for assessing conjugation events as function of hydration conditions, as determined by a prescribed matric potential (Fig. 2 and SI Appendix, Supplementary Methods for details). In unsaturated soil, the matric potential results from capillary and adsorptive interactions that retain water in pores and within roughness elements (22, 23). The matric potential is often expressed as a negative pressure (relative to atmospheric pressure): where zero value marks complete water saturation, while progressively negative values correspond to drier conditions (22, 23). We use the simple experimental system (Figs. 1 and 2) to directly study the relation between a biological cell-to-cell interaction (bacterial conjugation) and a physical parameter (matric potential) relevant to soil. In addition to experiments, mathematical modeling provides a means for generalization and enables evaluation of microscopic mechanisms that are inherently difficult to observe. Models of plasmid transfer on surfaces have been reported (24, 25) but remain limited due to oversimplification of the aqueous habitats in unsaturated soil (26). More recently, individual-based models (IBMs) (26–28) have evolved and permit mechanistic study of cell-to-cell interactions at relevant microscales, considering physical and geometrical complexity found in natural habitats. Here, we have used an IBM that explicitly integrates salient physical properties of soil and terrestrial habitats (29–31), such as the distribution of the aqueous phase held under capillary forces on model rough surface. Both experiments and simulations show that the fragmentation of the aqueous phase induced by drier conditions led locally to higher rates of cell encounters and increased the number of direct cell interactions (plasmid transfer). Overall, these results permit us to delineate a causal link between macroscopic variables (matric potential) and the probability of bacterial interactions that take place at the microscale.

Fig. 1.

Bacterial conjugation to study cell-to-cell interactions. (A) Illustration of the conjugation process (star) on a surface between donor cells (pink) and recipient cells (orange), which requires direct physical contact and results in plasmid transfer to recipients, which become transconjugants (green) over time. In this simplified sketch, all contacts between donor and recipient cells lead to conjugation; however, in reality, various physical and biological factors may lower plasmid transfer efficiency. (B) Fluorescence micrograph shows an experiment with P. putida donor and transconjugant bacterial cells immobilized on an agar surface, respectively, shown in pseudocolors magenta and cyan. Recipient cells are not fluorescent and, hence, not visible. (C) Control for conjugation efficiency on solid homogeneous surfaces (0.1× TSB agar plates) and in liquid environment without shaking (5 mL of 0.1× TSB). Initial R:D ratio was 10:1. After 20 h of incubation at 25 °C, the cells were diluted and plated on selective agar media followed by cell counts (measured as colony-forming units or cfu; see SI Appendix, Supplementary Methods for details). Individual results from triplicate cultures are shown. Transconjugants were only detected in one of the liquid replicates.

Fig. 2.

Sand microcosms with controlled hydration conditions. (A) Photograph shows an assembled system, with microcosms connected to a liquid medium reservoir (0.1× TSB). The height of the liquid column (h) prescribes a fixed suction to the sand phase via the ceramic plate, which mimics the effects of matric potential in porous media (e.g., h = 10-cm produces a water matric potential of ∼−1 kPa). Inset shows stereomicroscope image of the quartz sand layer. (B) Illustration of the microcosm unit, with the quartz sand grains (0.5 g of dry sand, ∼1 mm thick) in contact with the saturated ceramic plate. Up to six units were connected to a single reservoir bottle.

Results

Effects of Hydration Conditions on Bacterial Interactions in Controlled Environments.

To study the frequency of cell-to-cell contacts, we used the soil bacterium P. putida as the donor and recipient of a conjugative plasmid. We specifically used a plasmid that transfers best on surfaces (Methods), showing high rates of transfer on agar plates but very poor rates in liquid cultures (Fig. 1). A suspension of recipient (R) and donor cells (D), mixed in a ratio of 10:1 (R:D), served to inoculate quartz sand microcosms (Fig. 2) that were kept at constant temperature and matric potential values for the duration of the experiment (20 h). Hydration conditions in the microcosm ranged from relatively wet (matric potential value of −1.2 kPa) to relatively dry (−6.5 kPa), representing a wide range of saturation levels within such coarse porous medium (SI Appendix, Figs. S1 and S2). Fig. 3A shows the resulting number of recipient, donor, and transconjugant cells (i.e., cells that have acquired the plasmid after cell-to-cell interaction with a donor) that were isolated from the microcosms and enumerated at the end of the incubation period. Bacteria survived and grew in all microcosms, with an average population increase of ∼13-fold (corresponding to an average of three to four cell doublings during 20 h). Data showed no significant difference in total population sizes observed across the different matric potentials, although we noted a slight tendency toward less cell doublings with lower matric potential values (SI Appendix, Fig. S3). Therefore, the relatively limited range of matric potential values used (approximately −1 to −7 kPa) corresponds to relatively wet conditions in most soils (23) and conducive to bacterial growth, which confirmed that changes in transfer frequency were mainly due to variations in constraints imposed by aqueous distribution in the microcosms. Transconjugants were detected in all microcosms and their absolute numbers increased with decreasing matric potential values (Fig. 3A), although the difference was significant only when comparing results from the lowest and highest matric potential values (P = 0.03 with a one-tailed t test). Common metrics of plasmid transfer efficiency confirmed that conjugation rates increased with lower matric potential values in the microcosms by about one order of magnitude (SI Appendix, Table S1). We were able to visualize some of the donors and transconjugants in the sand environment at the microscale, and the observations confirmed the occurrence of direct contacts between cells that newly acquired the conjugative plasmid (transconjugants) and donor cells growing as microcolonies (Fig. 3B). Fig. 4 illustrates a conceptual view of the cell probability of interaction in various environments, including the role played by matric potential in porous medium. We developed a probabilistic model for cell-to-cell interactions on hydrated surfaces that relies on simple spatial statistics (32) and that integrates changes in hydration conditions (SI Appendix, Fig. S4); however, this simple model was not sufficient to faithfully capture the effects of hydration on conjugation rates.

Fig. 3.

Bacterial cell-to-cell interactions in sand microcosms. (A) Sand microcosms with controlled conditions (Fig. 2) were inoculated with ∼1 × 10⁷ cfu of a mixed suspension of P. putida recipient and donor cells (R:D ratio was 10:1). After 20 h of incubation at 25 °C, the entire sand fraction was harvested to count the final number of recipients, donors, and transconjugants. Individual and mean results from triplicate microcosms are shown. (B) Overlay fluorescence micrographs show donor cells (magenta) and transconjugants (cyan) observed after 45 h of incubation at 25 °C in a sand layer. Sand particles were kept at a matric potential of −3.6 kPa. Donors are observed as single cells or as microcolonies, and transconjugants are only seen in direct contact to donors (Inset). Recipient cells and sand particles are present but not visible in the fluorescence channels.

Fig. 4.

Conceptual illustration of physical factors influencing cell interactions. (A) In liquid culture, local cell density fluctuates due to cell motility, and donors fail to encounter recipients for a sufficient amount of time. On agar surfaces, cells are immobilized but can spread by cell division. At high cell density, chances become high that a donor comes to contact with a recipient and conjugates. (B) On hydrated rough surfaces (e.g., in sand microcosms), the conditions are somewhat intermediary. As the rough surface dries (lower matric potential values leading to unsaturated conditions), the aqueous domain becomes fragmented, the cells are partially immobilized, and the probability of cell-to-cell interaction increases.

Modeling of Aqueous Fragmentation, Cell Encounters, and Conjugation.

We developed a mechanistic modeling framework for systematic evaluation of bacterial conjugation in soil-like surfaces under different hydration conditions (see Methods for details). The main assumption in the context of the study is that the spatial physical environment (pore geometry and aqueous connectivity) ultimately controls the rate of cell-to-cell encounters necessary for bacterial conjugation. In the model, bacteria were represented as individual agents that disperse by flagellated motion within water films, grow and divide, or die, depending on the local conditions that they experience. These agents populated an idealized rough surface made of connected bonds that retain liquid by capillary forces and adsorption depending on their geometry and on the matric potential prescribed to the system (29–31). In contrast to nearly saturated conditions, lower matric potential values (i.e., drier conditions) resulted in numerous fragmented aquatic domains forming spatially isolated bacterial subpopulations (Fig. 5A). As a result, conjugation events (and the resulting transconjugant cells) tended to localize in spatial hotspots where (i) water was retained and (ii) both donor and recipient cells were present (Fig. 5A). Model estimations of conjugation rates were consistent with the experimental observations and showed that the number of transconjugants increased under lower hydration conditions (Fig. 5B). This pattern was consistent across a range of cell velocity; however, an imposed slower speed produced results that better matched the empirical data (Fig. 5B). The simulations were conducted with similar initial cell densities and ratios (10:1, R:D) as in the experiments. Simulation results showed the effect of increased cell motility on enhancing conjugation process by promoting cell-to-cell encounters and, therefore, increasing the number of transconjugant cells. However, when hydration conditions approached saturation, the benefit of enhanced motility was counterbalanced by larger spatial domains visited by cells with shorter close encounter durations (resembling conditions in bulk liquid), resulting in a decrease in conjugation rates.

Fig. 5.

Individual-based modeling of conjugation on hydrated surfaces. (A) Exemplary simulated spatial patterns of bacterial distribution in hydrated pore networks under almost water saturated (−0.2 kPa) or relatively dry (−8.0 kPa) hydration conditions. Images are close-up views of the model representation. Water retained in the bonds of the pore network is represented with shades of blue coloration. Each bacterium is an individual agent (colored circle) whose physiological behavior is influenced by its local physical environment. Cells can only move in and occupy bonds that contain water. For this reason, local spatial hotspots of conjugation are observed under drier conditions. (B) Simulated number of recipients, donors, and transconjugants per unit volume and comparison with experimental data (Fig. 3). Simulation values are averages calculated from 15 realizations (five replicates with three independent networks). Error bars show one SD. Simulations were conducted for two cell motilities (0.1 and 1 µm/s). For the sake of comparison, the number of cells was represented per unit volume of the experimental and modeling systems.

Individual-based modeling permits direct links between specific physical processes and the probability of cell-to-cell interactions. In particular, the model revealed how, all else being equal, matric potential controls both the size and abundance of the microbial aqueous habitats (Fig. 6A), i.e., the aqueous fragmentation of the system. The level of aqueous phase fragmentation at a given matric potential may be used to estimate the probability of donors and recipients cooccurrence in a given aqueous cluster. This probability increases with cell density and for wetter conditions (Fig. 6B). Prerequisite for interactions such as conjugation events is a donor-to-recipient encounter. We thus used the model to quantify in detail the effects of aqueous phase fragmentation (driven by hydration conditions and surface geometrical features) on the number of cell encounters. Fig. 6C details the encounter times as a function of the largest aqueous cluster size (representing the fragmentation state of the network). In these simulations, individual donor and recipient cells are tracked and their number of encounters is accumulated over time. Results indicate a substantial increase in the number of encounters as the aqueous phase became more fragmented (Fig. 6C). The simulation results only considered donor and recipient cells that experienced at least one encounter after the total simulation time (12 h). As shown in Fig. 6C and SI Appendix, Fig. S5, the results indicate that for connected networks (wet) where cells swim freely, most encounters between donors and recipients last only a small fraction of the total simulation time, while in a fragmented environment (drier conditions), more and longer-lasting encounters are in general observed, which explains why more conjugation events are measured under drier conditions. This habitat fragmentation and spatial confinement effect progresses until the system becomes so dry that the number and size of aqueous clusters rapidly diminishes, leading to the decline in the total number of cell-to-cell interactions (at matric potential values lower than −20 kPa, see SI Appendix, Fig. S5). Consistent with cell-to-cell encounter time, the number of conjugation events was affected by aqueous phase fragmentation, cell density, and decrease in the ratio of donor to recipient cells (SI Appendix, Fig. S6). Finally, an exponential relationship between the cell density and the number of conjugation events was observed (SI Appendix, Fig. S6).

Fig. 6.

Modeling the effects of matric potential on aqueous fragmentation and cell encounters. (A) Model hydrated porous networks were used to determine the largest aqueous cluster size in the domain (normalized by network size, approximately 3 × 3 mm) as well as the total number of aqueous clusters as function of the matric potential prescribed to the network. Here, a (connected) aqueous cluster is defined by the capacity of a bacterial cell to travel across the cluster through liquid films supporting flagellar motility. Mean values from 10 model realizations are shown. Error bars are one SD. (B) Individual-based model simulations show the probability that at least one donor cell and one recipient cell cooccur in a given aqueous cluster, as a function of matric potential and number of donors (R:D ratio is constant as 10:1). Mean values from 20 model realizations are shown. Error bars are one SD. (C) Simulation results show the relationship between cells encounter time and aqueous cluster size. The number of simulation time steps with encounter of a donor and a recipient cell are counted for a total simulation time of 12 h. The encounter time t is calculated for more than 1,000 donor/recipient cells and is normalized by total simulation time and averaged over the time distributions. Insets show exemplary results for the number and duration of donor-to-recipient encounters under relatively wet (−0.5 kPa, connected) and relatively dry (−8.0 kPa, fragmented) conditions, as well as the corresponding distribution of individual connected aqueous domains (marked by different colors). (Matric potential of −0.5 kPa resulted in a single blue cluster, hence we represent a slightly lower potential value of −2.0 kPa for illustrative purposes.)

Discussion

We have used bacterial conjugation in P. putida (Fig. 1) as a proxy for cell-to-cell interaction in hydration-controlled sand microcosms (Fig. 2), and the results supported our hypothesis that fragmentation of the aquatic microbial habitats induced by lower water potential could locally promote encounters and interactions of bacterial cells (Fig. 3 and SI Appendix, Fig. S7). These findings go against intuition that aqueous phase fragmentation would limit opportunities for gene transfer in soil. A numerical individual-based model of bacterial conjugation with realistic representation of the physical domain and aqueous phase distribution corroborated these experimental results (Fig. 5) and offered mechanistic explanation based on the distribution of water content in the system (Fig. 6). The model confirms the intuitive expectation that beyond a certain dehydration threshold, most cell interactions cease due to vanishing aqueous habitats (SI Appendix, Fig. S5). Experiments and simulations suggest two main effects of matric potential reduction on increased conjugation rate. First, motile bacteria (such as P. putida cells) are slowed down or even immobilized within thin liquid films at low matric potential. This increases the average duration of cell-to-cell contacts, and it has been shown experimentally that bacterial swimming and mobility reduce the rate of plasmid transfer (33). Second, lowering matric potential in the system increases the travel time required for cell dispersal to a given distance (30) and constrains overall cells’ maximal dispersal range (34). (Incidentally, these constraints also apply to nonmotile bacteria, although their effects are less marked than with flagellated cells.) The biophysical processes described above are directly linked to the fragmentation of the aqueous domain, as shown by the individual-based model (Fig. 6). The connectivity of the retained aqueous phase (which underlies aqueous fragmentation) is defined here operationally by the minimum liquid film thickness that allows the passage of a bacterial cell of a given diameter. Lowering the matric potential in the porous system thus induces the formation of spatially isolated (“disconnected”) aquatic microhabitats (Figs. 5 and 6) that strongly limit bacterial dispersal and intermixing (30, 34). Such disconnected habitats represent microsites with a local higher probability of cell-to-cell contacts, or “pockets of interactions” to use the words of Raynaud and Nunan (19). Because most soils remain under unsaturated conditions most of the time (11), we argue that the fragmentation of the soil aqueous phase and its associated constraints for microorganisms may be a key factor in explaining soil bacterial diversity. Moreover, the matric potential values used in our study were closer to water saturation than values commonly measured in soil, with the so-called field capacity often ranging between −10 and −33 kPa [field capacity corresponds to the water content retained in soil once internal drainage has ceased (23), for example a few days after rainfall]. The presence of multiple concurrent soil microhabitats in which species (temporarily) coexist and genetically interact has profound consequences for microbial evolution in the long term, as fragmented aqueous habitats may represent a substantial barrier to gene flow between bacterial populations (35, 36). However, horizontally transferred genes (e.g., via a plasmid) are likely more stable in small and isolated bacterial populations due to reduced competition. Finally, spatially and genetically isolated bacterial communities found in any volume of soil, which theoretically harbor low intracommunity but high intercommunity diversity, could contribute to the unequalled diversity observed at all scales, a hypothesis recently proposed by Rillig et al. (35) (albeit focusing on soil aggregates). Overall, our modeling and experimental findings suggest a simple physical mechanism—aqueous phase fragmentation and associated reduction of connectivity—that directly influences the occurrence of bacterial encounters, interactions, and genetic exchanges in unsaturated habitats such as found in most soils.

The results permit an important distinction between the likelihood of cell coexistence and that of cell interaction within a connected aqueous cluster. Simulations have shown that matric potential (a macroscopic soil hydration state) and cell density jointly control the probability of cooccurrence of two individuals of different types (e.g., strains, species) within an aqueous cluster (Fig. 6B), a prediction supported by previous modeling studies (37, 38). However, if cooccurrence of cell types is a prerequisite for cell–cell interaction, it does not represent evidence for short-range contact, because two cells can coexist in an aqueous cluster while still being spatially separated by hundreds of microns. For example, in our experiments, conjugation was clearly associated with cell-to-cell physical contact (Fig. 3B), which was supported by previous studies with P. putida showing that the vast majority of plasmid transfer with rigid pili occur via direct contact (<1 μm donor-recipient distance) (39). In that context, we should not assume direct interaction between species based solely on their cooccurrence in soil volumes typically sampled at macroscopic scales (>cubic centimeters) (7, 40). Such spatial considerations are essential for better representation of the behavior of complex microbial communities (41, 42), and enabling discrimination between “potential” species interactions (based on cooccurrence in a sample) and “realized” interactions. We note that, although we measured genetic interactions (plasmid transfer) in this study, the conclusions may be applicable to metabolic or signaling interactions, as they most often rely on solute diffusion over microscopic distances (5, 43, 44).

Admittedly, we did not explore here the wide variety of known conjugative plasmids (45) and bacterial hosts (46), nor other factors that could influence the rate of bacterial conjugation, such as properties of soil surfaces, pH and oxygen levels (47), the proximity of plant roots (48), the presence of microbiota (49), or the presence of fungal hyphae (33). Any effect of the water potential on conjugation rates in soil will thus likely remain part of a mosaic of effects that stem from the abovementioned factors and which lies beyond the scope of this study. Instead, we highlighted the role of the physical habitat—especially the dynamic aqueous phase—as a “gatekeeper” of bacterial interactions, enabling or preventing genetic and metabolic exchanges among and between communities in soil. Such simple physical processes should not be overlooked in the discussion of community interactions in the microbial world as they can play an overarching role in the realization of biological functions.

Methods

Bacterial Strains, Plasmid, and Culture Conditions.

We used P. putida KT2440 as recipient strain and P. putida KT2440::lacI^q-pLpp-mCherry-Km^R (46) as donor strain of the cryptic broad-host range plasmid pIPO2tet::Plac::gfp (50), which encodes resistance to tetracycline. This plasmid was isolated from the rhizosphere (51), and it shows high rate of transfer on surfaces and rates of transfer lower by several orders of magnitude in liquid cultures (Fig. 1), which is due to the type of pili (short and rigid) encoded by pIPO2 and used by P. putida to exchange the plasmid (52). The donor constitutively expresses the mCherry fluorescent protein as well as the LacI^q repressor of the Plac promoter, which prevents expression of GFP from the pIPO2tet::Plac::gfp plasmid in the donor cell. Since the recipient strain lacks the lacI^q gene, GFP can be expressed in transconjugants (53) (Fig. 1B). Both strains were routinely grown on tryptic soy broth (TSB) (VWR International) at 30 °C with shaking at 280 rpm or on TSB agar plates at 30 °C. Tetracycline (15 μg/mL) was added in cultures of the donor strain to ensure plasmid maintenance.

Conjugation Experiments in Sand Microcosms.

A suspension of recipients and donors with a final ratio of 10:1 (R:D) based on optical density measurements (OD₆₀₀) was added (100 μL with ∼10⁷ cfu) to the saturated microcosms. A fixed matric potential was prescribed (Fig. 2 and SI Appendix, Supplementary Methods for details). After equilibration, microcosms were sealed with parafilm (to prevent evaporation) and incubated at constant temperature (25 °C) for 20 h. Then, the entire sand phase from each microcosm was collected by pipetting and transferred to 15-mL centrifugation tubes containing 10 mL of PBS. To detach cells from the sand particles, tubes were vortexed at maximum speed for 10 s, followed by 2 min in an ultrasonic water bath (Branson 5800; Branson Ultrasonics Corporation). Various dilutions were plated on TSB agar plates (with or without tetracycline at 15 μg/mL) to enumerate viable bacteria as cfu, and GFP and mCherry fluorescence was used to discriminate between transconjugants and donors (SI Appendix, Supplementary Methods). Visualization of bacterial conjugation at the microscale was performed using a slightly modified microcosm setup that permitted closer access to the sand surface with the microscope objective (SI Appendix, Supplementary Methods).

Individual-Based Modeling of Bacterial Conjugation on Idealized Hydrated Surfaces.

The mathematical model abstracts the natural soil structure into 2D regular pore networks consisting of v-shaped channels (30, 54). The network is composed of 40 × 40 sites that together simulate a physical domain of ∼3 mm × 3 mm. The structural heterogeneity of soil is implemented in the network model by varying the angularity of v-shaped bonds through a statistical probability distribution (29–31). The aqueous volume retained in a network bond is function of matric potential and bond angular geometry (29). Neighboring network bonds are defined as “connected” (and hence grouped in a “connected aqueous cluster”) when the thickness of the liquid films retained in the bonds are above a threshold value sufficient to sustain bacterial flagellar motility (30). Therefore, in our model “connectivity” is defined operationally from the subjective viewpoint of a bacterial rod cell of a given width. (This means that thin residual liquid films existing between “disconnected” aqueous clusters may still allow for nutrients diffusion.) The pore network environment is combined with an agent-based model that represents the behavior of individual bacteria in soil. Bacterial growth and substrate uptake rates at the single-cell level are based on Monod-type kinetics (see SI Appendix, Table S2 for the growth parameters used in the model). Nutrients transport and uptake are modeled based on Fick’s law of diffusion and mass conservations for the whole network. A 1D reaction-diffusion model is solved for each network bond (31). Bacterial cell motility is often restricted in water-limited soil environments where viscous drag in thin water films and capillary pinning at the air-water interfaces are enhanced. To account for this phenomenon, we have used a previously developed model that balances self-propulsion forces of flagellated cells (F_M) in bulk water with physical restrictions (cell-wall hydrodynamic interactions F_λ and capillary pinning force F_Ca) as a function of water film thickness δ. A proportional relationship is considered between the cell velocity within a single bond V in the network and its velocity in bulk water, V₀, as follows:

$$V(\psi )=V_0\hspace{0.17em}\frac{F_{\mathrm{M}}-F_{\mathrm{\lambda }}\left(\delta \left(\psi ,\alpha \right)\right)-F_{\mathrm{Ca}}\left(\delta \left(\psi ,\alpha \right)\right)}{F_{\mathrm{M}}}.$$

The water film thickness δ is quantified as a function of matric potential ψ and bond angularity α (29, 30). Note that the simulations are performed assuming entirely random cell motility without chemotaxis. To mimic the nutrient flux in the experimental setup (porous surfaces), nutrient flux in the model is assumed to be supplied from the bottom of the network directly to individual bonds, and nutrients are distributed uniformly in the network. In the current modeling framework, plasmid conjugation from a donor to a recipient bacterium takes place if the encounter lasts more than a minimum required time. We assumed that the conjugation occurs if the short distance between a donor and recipient bacterium (less than 2 µm in our simulations) persists long enough (called “conjugation duration”). The probability of conjugation P increases as a function of encounter duration t, and it is assumed to follow Poisson probability distribution

$$\{\begin{array}{l}P\left(t\right)=0\begin{array}{ccc}& & t<t_{\min }\end{array}\\ P\left(t\right)=\frac{{\mathrm{\lambda }}^{(t-t_{\min })}\exp (-\mathrm{\lambda })}{\left(t-t_{\min }\right)!}\begin{array}{cc}& t\ge t_{\min }\end{array}\end{array},$$

where λ is the sample mean of encounter duration t, and $t_{\min }$ is the minimum encounter time required to observe first conjugation. The minimum encounter time is considered to be 5 min based on previous observations (55).