Environmental modulators of algae-bacteria interactions at scale

Abstract

Interactions between photosynthetic and heterotrophic microbes play a key role in global primary production. Understanding phototroph-heterotroph interactions remains challenging because these microbes reside in chemically complex environments. Here, we leverage a massively parallel droplet microfluidic platform that enables us to interrogate interactions between photosynthetic algae and heterotrophic bacteria in >100,000 communities across ~525 environmental conditions with varying pH, carbon availability, and phosphorous availability. By developing a statistical framework to dissect interactions in this complex dataset, we reveal the dependence of algae-bacteria interactions on nutrient availability is strongly modulated by pH and buffering capacity. Furthermore, we show that the chemical identity of the available organic carbon source controls how pH, buffering capacity, and nutrient availability modulate algae-bacteria interactions. Our study reveals the previously underappreciated role of pH in modulating phototroph-heterotroph interactions, and provides a framework for thinking about interactions between phototrophs and heterotrophs in more natural contexts.

Graphical Abstract

eTOC blurb

Interactions between photosynthetic and heterotrophic microbes occur in chemically diverse environments and are vital to the global carbon cycles. Our high-throughput screening of >100,000 algae-bacteria communities in ~525 environments reveals pH, buffering capacity, and carbon source identity modulate algae-bacteria interactions by impacting the dependence of growth on nutrient availability.

INTRODUCTION

Microbial communities occupy nearly every niche on Earth, from animal hosts, to soils, and oceans. These complex consortia often contain many interactions between members whereby one species impacts the abundances of another. Interactions in these communities can determine the outcome of invasions¹, metabolic processes such as carbon and nitrogen remineralization², or the phenotype of the host³. Crucially, however, interactions between members of a microbial consortium depend on the environmental context. For example, changes in pH, nutrient availability, temperature, or toxic metabolic byproducts, can strongly modulate interactions between members of a collective^4–8. As a result, an important question in ecology is understanding how environmental parameters impact interactions.

Understanding how environmental parameters influence ecological interactions between pairs of taxa in communities is challenging. The physicochemical environment in natural microbial communities is high-dimensional in the sense that there are many possible parameters that change in time and space and can impact the outcome of an interaction⁹. This high-dimensionality means that experimentally interrogating how interactions depend on the environment is a daunting task. For example, to measure growth of a single strain in across all possible combinations of four different environmental variables at 10 levels for each variable (say, pH, carbon, nitrogen, phosphorous availability) would require 10⁴ experiments. To determine interactions between just two taxa would require measuring their growth alone and in pair-culture in each one of these conditions – meaning 30,000 measurements would be required, a huge undertaking.

Here we address this problem using a massively parallelized droplet microfluidic platform¹⁰ to interrogate interactions between a photosynthetic alga (phototroph) and a heterotrophic bacterium. Phototrophs form the basis of primary production in many environments, and heterotrophic bacteria play an important role in the growth of phototrophic populations both in natural ecosystems and engineered bioreactors.^11,12. One of the key features of phototroph-heterotroph interactions is that they occur between distinct metabolic strategies. Phototrophs are capable of fixing inorganic carbon using light, while heterotrophic organisms require chemical energy often in the form of reduced carbon to generate energy and biomass. Phototrophs excrete some fraction of the carbon they fix and this provides the chemical substrates upon which heterotrophic microbes depend. As a result, prior work on phototroph-heterotroph interactions has often focused on the exchange of organic carbon between these two metabolic strategies^11,13.

While the exchange of organic carbon between phototrophs and heterotrophs is important, other environmental factors also play a role and are less well studied. For example, in some nutrient-rich environments such as estuaries or coastal ecosystems, organic carbon is available to heterotrophs through the decay of organic matter rather than the direct excretion of carbon from phototrophs¹⁴. In addition, interactions between phototrophs and heterotrophs depend on a host of other environmental factors. For example, the dynamics of phototrophs in association with heterotrophs can depend on the availability of exogenously supplied carbon, nitrogen, and phosphorous or temperature, light, pH, and small molecule exchanges^15–23. Therefore, it appears critical to understand how environmental factors affect interactions in these communities, even when carbon exchange is not a central factor. Most, but not all¹⁶, previous studies in laboratory model systems have focused on carbon exchange^11,24–27, leaving a gap in knowledge. Addressing this gap comes with the challenges posed above of measuring interactions at a large enough scale to assess the role of multiple environmental factors. Thus, while carbon exchange between phototrophs and heterotrophs is important, there are also contexts where carbon for bacterial growth is supplied exogenously. In addition, nutrients and environmental variables beyond carbon can play a defining role in the outcome of phototroph-heterotroph interactions but the role of these variables is less well-studied.

To address this problem, we interrogate phototroph-heterotroph interactions in a context where carbon exchange does not play a dominant role in the growth and proliferation of the community, but the role of other environmental factors can be readily assayed at a massive scale. To accomplish this, we use a microfluidic platform that leverages nanoliter droplets, with contents barcoded using fluorescent dyes, to measure abundance dynamics in >20,000 cultures in a single experiment. Using this approach, we measure the interaction between the model alga Chlamydomonas reinhardtii and the bacterium Escherichia coli in ~525 environmental conditions in >10 replicates each for both monoculture and pair-culture. Using this platform, we quantify the dynamics of algal and bacterial growth over a period of 4 days. On this timescale, the excretion of organic carbon by the alga is small²⁸, and so we provide exogenous organic carbon to permit bacterial growth to occur.

Within the nanoliter droplets, we measure algae-bacteria abundance dynamics via microscopy across a range of organic carbon sources and concentrations, phosphorus concentrations, pH, and buffering capacities. The resulting dataset proves amenable to statistical analysis where a regression reveals the key environmental drivers of algae-bacteria interactions. While previous studies suggest that nutrient availability is the key driver of interactions between phototrophs and heterotrophs, we find that pH and buffering capacity qualitatively alter how the availability of nutrients impacts the interaction between algae and bacteria. Thus, we show that across a huge range of environmental conditions, pH and the ability of the environment to resist changes in pH (buffering capacity), act as important regulators of the interaction between phototrophs and heterotrophs. Finally, the role of the environmental factors - pH, buffering capacity, and nutrient availability in regulating interactions is modified by the chemical identity of exogenously available organic carbon. These results suggest that the chemical composition of organic carbon and pH interact to qualitatively determine the outcome of algae-bacteria interactions.

RESULTS

The model system and environmental conditions

The microbial community under study comprises the alga, C. reinhardtii, commonly found in soils and freshwater²⁹, as the phototroph, and the host-associated and soil-dwelling bacterium^30,31, E. coli, as the heterotroph. We note that these microbes are not known to coexist in the wild and so we expect no strong co-evolutionary history between these organisms. Despite this, these two species represent the essential metabolic strategies of phototrophs and heterotrophs. The alga fixes CO₂, and the bacterium utilizes complex carbon sources for energy and biomass. In addition, given that the droplet microfluidic platform is not readily amenable to longer-term growth assays (>5 days) the relatively rapid growth of the alga (doubling time 8-12 hours) enables us to use the platform to interrogate the interaction between these two taxa. So while these two species do not represent an interacting pair of wild microbes, they are representative of the orthogonal metabolic strategies of phototrophs and heterotrophs while being amenable to measurements at scale. Thus, our intention here is to utilize these taxa as representatives of these metabolic strategies while being cognizant that the insights we gain here will need further validation in other ecological contexts. Despite this limitation, these two microbes have been widely used in studies as model phototrophs and heterotrophs due to their thorough biological characterization, ease of cultivation, and accessibility to molecular techniques and quantitative measurements. Previous studies of closed microbial communities including these two microbes, in addition to a ciliate, have revealed strongly deterministic dynamics on timescales of months and rich spatiotemporal and phenotypic processes^32,33. Another study demonstrated the presence of higher-order interactions between this alga and bacteria mediated by a ciliate.¹. Thus, the interactions between these two model organisms constitute a tractable test bed for understanding phototroph-heterotroph interactions.

In this study, interactions between the algae, C. reinhardtii, and the bacteria, E. coli were assayed in the modified 1/2x Taub media (a freshwater mimic media) that varied in five environmental factors - initial pH, buffering capacity, phosphorus concentration, carbon concentration, and carbon source identity. The chosen environmental factors are among those that significantly contribute to chemical variation across natural environments^34,35. Although resource competition and exchange are identified as key players in driving phototroph-heterotroph interactions^{11,12,16–19,22,36}, several studies have reported a strong correlation between microbial communities compositions and environmental factors such as pH and concentration of nutrients - carbon, nitrogen and phosphorus^37–39. Additionally, it is well known that the identity of the carbon source affects E. coli metabolism via impacting the growth rate and the nature of the metabolic products, which could potentially lead to different interactions with C. reinhardtii ^40–42. Therefore, we reasoned that a multitude of abiotic factors such as pH, buffering capacity, light level, including nutrient concentration and type may contribute to phototroph-heterotroph interactions (Fig. 1). Hence, we chose the above 5 factors. The values of each of the environmental factors were chosen to be in biologically plausible ranges: 6.1-7.5 for initial pH, ~0-3.5 mM for buffering capacity, 0.01 mM - 4 mM for phosphorus concentration, 2 mM - 10 mM (carbon atoms) for the carbon concentration (STAR Methods). We chose five different carbon sources (glycerol, glucose, galactose, pyruvate, acetate) to assay both gluconeogenic and glycolytic carbon sources as well as a carbon source that is known to support mixotrophic growth of the alga (acetate). For each of these carbon sources, the algae-bacteria interactions were assayed in a total of ~105 environmental conditions for monoculture of each taxon and co-culture of both.

Figure 1. Dependence of algae-bacteria interactions on environmental factors.

Illustration of our hypothesis that diverse interactions between algae and bacteria are altered by a multitude of chemical factors in the environments such as concentration of nutrients, pH, buffering capacity, light level, and temperature.

High-dimensional characterization of phototroph-heterotroph interactions

In this study, we used droplet-based microfluidic chip (”kChip” with k=2) to rapidly assay the phototroph-heterotroph interactions in hundreds of environmental conditions in parallel. The kChip platform has previously been utilized for drug discovery, pathogen detection, and the study of bacterial interactions^10,43–45. Briefly, the experiment proceeds by first generating a library of environmental conditions that vary in the initial pH, buffering capacity, concentration of phosphorus, and concentration of carbon of a chemically defined minimal medium (Fig. 2A; STAR Methods). Initial pH refers to the starting pH of the environment, which we varied by using buffers and titration. To vary the buffering capacity of the environment, we added different concentrations of organic buffers (Tris or MOPS) that cannot be used as nutrient sources by algae or bacteria (Fig. S21). We fluorescently barcoded each environmental condition using three fluorescent dyes in low concentrations and added algae and bacteria independently. Using these precultures, a commercial droplet generator was used to create thousands of nanoliter water-in-oil droplets containing algae or bacteria in each of the predefined nutrient conditions. These droplets were then pooled and loaded into a kChip microfluidic chip platform which contains ~25,000 microwells, each of which randomly groups two droplets containing microbes in predefined media conditions, resulting in the formation of all possible combinations of communities (monocultures and cocultures) and environmental conditions (Fig. 2A; STAR Methods). The chip is then imaged to identify the fluorescent dye barcodes and thereby infer the environmental conditions present in each microwell (STAR Methods). Subsequently, the droplets in each microwell were merged via exposure to an alternating electric field, leading to the formation of the phototroph-heterotroph communities in hundreds of environmental conditions. Thereafter, the kChip was incubated at 30 °C under light (68.5 μmol m²s⁻¹) to allow for growth. The chip was then imaged at regular intervals (approximately 0 h, 12 h, 21 h, 45 h, 68 h) to track the growth of the microbes using chlorophyll fluorescence for C. reinhardtii and genetically encoded GFP fluorescence for E. coli (Fig. 2B; STAR Methods and Dataset 4). Algal and bacterial abundances over time were determined by analyzing the microscopy images, generating microbial growth curves, and estimating growth as the difference between the initial abundances and the final abundances at the end of the experiment. We performed this analysis for both the phototroph and heterotroph, in >100,000 microbial communities constructed in the kChip experiments for all the carbon sources (Fig 2C; STAR Methods and Dataset 1). We quantified total growth, over the four days of the experiment rather than growth rates because the low temporal resolution of our measurement seriously limited our ability to quantify growth rates, especially for bacteria which often saturated before the first time point. We note that the abundance of the microbes may not be saturated at the end of the experiment in some of the environmental conditions. This is due to the duration of the experiment being limited by small droplet volumes and evaporation losses. The experiments on the kChip platform were found to be largely reproducible (Fig. S20).

Figure 2. A high-throughput droplet platform for measuring algae-bacteria growth in hundreds of environments.

(A) Setting up the microfluidic chip. Environments (media conditions) varying in the factors - initial pH, buffering capacity, phosphorus concentration, and carbon concentration, are prepared and barcoded using three fluorescent dyes (STAR Methods). After adding the bacteria (brown) and algae (green) independently to each barcoded media, nanoliter droplets of each of the microbes in the barcoded environments are generated. The generated droplets are pooled together and loaded on the microfluidic chip which randomly groups two droplets in each of its microwells. The chip is then imaged for fluorescent barcodes using a widefield fluorescence microscope, to infer the values of the environmental factors in the microwells via image processing (STAR Methods). Following exposure of the chip to an alternating electric field, droplets in the microwells merge to form replicates of bacterial monocultures, algal monocultures, and algae-bacteria cocultures in all combinations of the environments that were present in the initial droplets. The chip is then incubated at ${30}^{°}\mathrm{C}$ under light $\left(68.5\mu \mathrm{m}\mathrm{o}\mathrm{l}{\mathrm{m}}^2{\mathrm{s}}^{-1}\right)$. (B) Microscopy images of a single microwell showing the growth of algae and bacteria over time. The GFP fluorescence image representing the bacteria (in brown) and the chlorophyll fluorescence image representing the algae (in green) are overlayed in these images. The first image shows the bacteria and the algae in the separate compartments of the well, prior to the merging of the droplets. The later images show the increase in the abundance of the algae and bacteria at 12 h, 21 h, and 45 h. (C) Example growth curves of algae and bacteria in monoculture and coculture in an environmental condition. The images of the chip are analyzed to infer the abundances of the microbes in the microwells over time (STAR Methods). The growth $Y$ of algae and bacteria are then obtained by estimating the increase in their respective abundances at 68 h from their abundances at 0 h (black arrow labeled ”GROWTH $\left(Y\right)$” right panel).

Previous studies utilizing this platform studied bacteria. So, we modified existing protocols to make the measurement compatible with algae. Specifically, we added the functionality for imaging chlorophyll fluorescence to track the growth of C. reinhardtii and devised a computational pipeline to remove the bleed-through between chlorophyll fluorescence and one of the barcoding dyes (STAR Methods). This expanded the number of fluorophores that can be probed on the kChip from four to five.

Patterns in interactions between algae and bacteria

To begin, we compared the growth of both algae and bacteria in cocultures to their growth in monocultures. To visualize this, we plotted the growth in cocultures against the growth in monocultures. The dashed lines indicate equal growth in coculture and monoculture. Points below the dashed line indicate competitive or inhibitory interactions and points above the dashed line indicate facilitation. The bacterial growth in cocultures was lower than their respective growth in monocultures in all the environmental conditions, suggesting inhibition of E. coli by C. reinhardtii (Fig. 3A; Fig. S8). Additionally, the E. coli cells show greater aggregation in monocultures than in cocultures (Fig. S6). These results are consistent with a previous study that showed that introducing bacteria into algal cultures results in the inhibition of bacterial growth and the dispersal of bacterial aggregates¹.

Figure 3. Complex dependence of algae-bacteria interactions on environmental factors.

(A) Panels show bacterial growth in monoculture (x-axis) and co-culture (y-axis). Each point indicates median growth (Fig. 2C) of E. coli in co-culture and monoculture computed across replicates of each environmental condition. Error bars indicate the standard error of the mean growth. The median number of replicates per environmental condition ranges from 35-70 for the different culture conditions. The dashed line indicates equal growth in monoculture and coculture. Note the fact that all points lie below this line indicating the pervasive inhibition of bacteria by algae. The data in each panel are the same, but the colorbar represents each of the four environmental factors - Initial pH (top left), buffering capacity (top right), phosphorus concentration (bottom left), and carbon concentration (bottom right). The colorbar for phosphorus is logarithmic. The carbon source is glycerol. See Fig. S8 and Fig. S9 for the data in other carbon sources. (B) Identical plots as in (A) but for algal growth in monoculture and co-culture. The fact that most data lie near the dashed line indicates overall weaker impacts on algal growth by bacteria. Negative values of growth correspond to death where the number of cells detected declines from the beginning to the end of the experiment.

C. reinhardtii, on the other hand, has similar growth in cocultures and monocultures in most cases, indicating a weak effect of E. coli on the growth of C. reinhardtii (Fig. 3B; Fig. S8). There do exist a few environments where C. reinhardtii is suppressed or enhanced in co-culture relative to monoculture (points lying considerably below or above the dashed line in Fig. 3B), indicating an impact of the presence of the bacteria.

Further, we observe that interactions tend to be inhibitory or competitive in conditions where monoculture growth is substantial and facilitative when monoculture yields are low. To see this examine Fig. 3B (and Fig. S9) where algal growth tends to lie below the dashed line at high values along the x-axis and the opposite at low values. This trend is conserved across all environmental conditions. We observe a similar trend for the bacteria. While E. coli is inhibited by algae in all conditions assayed (all points are below the dashed line, Fig. 3A), the inhibition is stronger at high values of monoculture growth and the inhibition is weaker at low values of monoculture growth. If higher values of monoculture growth are interpreted as indicative of more permissive environments, this pattern supports the stress-gradient hypothesis, which posits that interactions should tend to be competitive in permissive environments and facilitative in stressful environments (see Discussion).

Finally, despite the overall reproducibility of our kChip experiments (Fig. S20), we observe higher variability in E. coli growth than in C. reinhardtii growth. Detailed analyses at the single-droplet level reveal this variability to be associated with the stochasticity in the initial cell densities in the kChip wells (STAR Methods and Fig. S19).

Algae-bacteria interactions show complex dependence on the environmental factors

Next, we sought to understand the dependence of algae-bacteria interactions on environmental factors. To visualize this, we plotted the growth in cocultures against the growth in monocultures, color-coding the data for each of the four environmental variables considered - Initial pH, buffering capacity, concentration of carbon and phosphorus (Fig. 3; Fig. S8, Fig. S9). These plots show no distinct grouping of the data based on any of the four environmental factors and indicate a complex dependence of algae-bacteria interactions on the environmental factors. For example, in the case of E. coli, while the low carbon concentration (the light green points in Fig. 3A bottom right panel) sets the growth in monocultures to low values, the variation in other environmental factors (pH, buffering capacity) causes the coculture growth to span from low to high values. There also exist cases where a single environmental factor largely determines monoculture and coculture growth. For example, low buffering capacity, not initial pH or nutrients’ concentration, appears to give rise to the death of C. reinhardtii (light green points have growth less than zero) (Fig. 3B top right panel).

When we compute correlations between the environmental factors and growth, we see significant statistical relationships between multiple factors and the bacterial or algal growth (Fig. S7) across carbon sources. These correlations reinforce the idea that there is a complex interplay between nutrient concentration, pH, buffering capacity, and the identity of the carbon source in determining algae-bacteria interactions.

One important observation from Fig. 3 is that initial pH and buffering capacity are shown to affect algae-bacteria interactions. This result agrees with surveys of communities in the wild which show that pH is an important environmental factor in determining community structure^37–39. In contrast, most previous experimental interrogations of interactions between phototrophs and heterotrophs focus on the role of nutrient concentration and competition^{15,17–19,25}. We expect that pH and buffering capacity are likely affecting interactions by influencing physiology including nutrient uptake rates.

Next, we sought a framework to quantify the interaction between algae and bacteria in our experiment. We considered consumer-resource models to quantify competition for carbon, nitrogen, and phosphorus. However, the interactions in our community cannot be described by a model that considers only these nutrients. For example, the overall inhibition of E. coli does not depend in a simple way on the concentration of nutrients. Similarly, variations in pH are not naturally modeled in a consumer-resource framework. Hence, a simple consumer-resource model approach is not suitable for dissecting the interactions in our data. We, therefore, took a statistical approach using simple linear regressions to model interactions as a function of the environmental factors.

Quantifying algae-bacteria interactions statistically

Our goal is to quantify how the presence of algae or bacteria impacts the growth of the other species across all the environmental conditions tested. To do this, we developed a simple framework for estimating interactions in the algae-bacteria communities via regression analyses. Specifically, we used a linear regression formalism to predict algal or bacterial growth (Fig. 2C) using environmental factors (pH, buffering capacity, phosphorus concentration, and carbon concentration) as independent variables. We designed a regression approach based on Monod’s growth law (STAR Methods), that allowed us to estimate the role of environmental factors and the presence and absence of the other species on growth while retaining a high level of interpretability. We performed independent regressions to predict algal and bacterial growth across all conditions.

Our regression approach can be explained mathematically using a simple example. To do this, first consider communities of algae and bacteria where the total growth is affected by a single environmental factor $\left(X\right)$ and by the presence of the other species via an interaction. Our regression was designed to measure the change in the growth of the target species (algae or bacteria) in response to changes in $X$. In this scenario, the model for predicting the growth of E. coli in monoculture and coculture takes the following form:

$$Y^{Ec}=\left({\beta }_{1,M}^{Ec}+I{\beta }_{1,I}^{Ec}\right)+X\left({\beta }_{X,M}^{Ec}+I{\beta }_{X,I}^{Ec}\right)$$ (1)

where $Y^{Ec}$ is the growth of E. coli and the ${\beta }_{*,*}^{Ec}$ are regression coefficients. $I$ is a variable that indicates the presence $(I=1)$ or absence $(I=0)$ of C. reinhardtii. The coefficient ${\beta }_{X,M}^{Ec}$ represents the change in growth in monoculture per unit change in $X$ and $\left({\beta }_{X,M}^{Ec}+{\beta }_{X,I}^{Ec}\right)$ represents the change in growth in coculture per unit change in $X$ (Fig. 4A). Similarly, ${\beta }_{1,M}^{Ec}$ is the growth at $X=0$ in monoculture and ${\beta }_{1,M}^{Ec}+I{\beta }_{1,I}^{Ec}$ is the growth at $X=0$ in co-culture. Hence, ${\beta }_{X,I}^{Ec}$, estimates the average change in growth per unit $X$ in coculture relative to monoculture (Fig. 4A, right panel). In other words, ${\beta }_{X,I}^{Ec}$ represents the effect of C. reinhardtii on E. coli as $X$ increases, in coculture. A positive coefficient would represent an enhancement of E. coli growth by C. reinhardtii as $X$ increases (Fig. 4B left panels). Similarly, a negative coefficient would represent suppression of E. coli growth by C. reinhardtii as $X$ increases (Fig. 4B right panels). An identical regression is used to estimate the impact of E. coli on C. reinhardtii growth.

Figure 4. Quantifying algae-bacteria interactions via regression.

(A) Formulation of the regression model for predicting growth from environmental conditions, here using E. coli as an example. $Y^{Ec}$ is the growth of E. coli in monocultures and cocultures and $X$ is an environmental factor that determines the growth. The indicator variable $I$ is set to 0 for growth in monoculture and 1 for growth in co-culture. The coefficient ${\beta }_{X,M}^{Ec}$ represents the change in growth in monoculture with $X$ and is referred to as a monoculture coefficient. The coefficient ${\beta }_{X,M}^{Ec}+{\beta }_{X,I}^{Ec}$ represents the change in growth in coculture with $X$ (shown schematically in the plot on the right). Hence, the coefficient ${\beta }_{X,I}^{Ec}$ represents the change in the effect of $X$ on growth in coculture relative to monoculture. The coefficient ${\beta }_{X,I}^{Ec}$ is dubbed an interaction coefficient. (B) Illustration of enhancement and suppression of E. coli growth by C. reinhardtii as $X$ increases. The growth of E. coli in monoculture (in brown) and coculture (in red) vs the environmental factor $X$ plotted in the case of enhancement (top left panel) and suppression (top right panel) of E. coli growth by C. reinhardtii as $X$ increases. The panels on the bottom row show the corresponding regression coefficients. The monoculture coefficient ${\beta }_{X,M}^{Ec}$ (in brown) and interaction coefficient ${\beta }_{X,I}^{Ec}$ (in magenta) in the case of enhancement (bottom left panel) and suppression (bottom right panel) of E. coli growth by C. reinhardtii as $X$ increases.

We extended the above model to include the effect of multiple environmental factors in determining the growth of both species (STAR Methods). For our dataset comprising of four environmental factors - initial $\mathrm{p}\mathrm{H}\left(pH\right)$, buffering capacity $\left(BC\right)$, phosphorus concentration $\left(\right[P\left]\right)$, and carbon concentration $\left(\right[C\left]\right)$, the model includes the following terms: $\left[P\right],\left[C\right],pH\left[P\right],pH\left[C\right],BC\left[P\right],BC\left[C\right],\left[P\right]\left[C\right]$. For each term, we estimated a coefficient for monoculture and interaction as described above. For simplicity, we refer to coefficients of features without the indicator variable $I$ as monoculture coefficients and coefficients of features with the indicator variable as interaction coefficients.

We did not include linear terms in $pH$ or $BC$ in our model because biologically $pH$ alone does not generate biomass, but instead modulates the ability of cells to grow on the available nutrients. Thus, we included only interaction effects between nutrients and $pH$ or $BC$. Therefore, the coefficient ${\beta }_{pH\left[P\right]}^{Ec}$ represents the susceptibility of growth to phosphorus concentration modulated by pH. The feature $\left[P\right]\left[C\right]$ was included to capture interactions between nutrients. Additionally, our model being simple, cannot capture nonlinearities in the growth as a function of a nutrient concentration. Despite these limitations, this statistical approach allows us to achieve a unified and interpretable picture of interactions between these microbes across a wide range of environmental conditions.

Finally, to account for the fact that algae globally inhibit bacterial growth in our experiment, we standardize the growth of both E. coli and C. reinhardtii prior to performing the regression above (STAR Methods). Thus, our regressions describe variation in bacterial growth after removing the effect of this global inhibition. It is important to recognize that in no condition do the bacteria actually grow better in monoculture than in co-culture (Fig. 3A). To facilitate interpretation, we also standardized all the independent variables in the regression. As a result, the regression coefficients describe the relative change in growth per unit change in each environmental factor. This standardization also allows us to compare coefficient values for regressions performed on different carbon sources despite variation in the growth on those nutrients. To perform the regression, we fit the growth measured in each well using a weighted least-square approach (STAR Methods).

In general, we find that this model provides good predictions of growth across environmental conditions in our experiment, the fits being better for some carbon sources (glucose, glycerol, acetate) than others (galactose) (Fig. S10). We successfully validated a few coefficients obtained from the regression model in microtiter plates (STAR Methods and Dataset 3; Fig. S22). Further, we note that the buffering ability of the phosphorus source in our experiments did not significantly affect our regression results (STAR Methods). Lastly, we found that a more complex model, such as a decision tree regression, gives superb fits to the data at the expense of interpretability (Fig. S15, Fig. S16).

pH and buffering capacity modulate nutrient dependence of algae-bacteria interactions

Using the linear regression approach outlined above, we modeled the dependence of algal and bacterial growth on the environmental factors for each of the five carbon sources in monoculture and coculture. We first looked at the regression coefficients describing the growth of E. coli in one particular carbon source (glycerol, Fig. 5A). Of all the monoculture coefficients (brown bars in top panel Fig. 5A) obtained from fitting E. coli growth in glycerol, the coefficient of $BC\left[C\right]$ is the largest, suggesting a strong interaction of buffering capacity with carbon concentration in determining the monoculture growth. Thus when $BC$ is high, there is a substantially higher growth per unit $\left[C\right]$ than when $BC$ is low. These results are consistent with the greater acidification of the environment at lower buffering capacity observed in the microtiter plate experiments (STAR Methods); this greater acidification likely negatively impacts E. coli. Therefore, the E. coli growth is expected to be higher at a higher buffering capacity for the same carbon concentration, which is reflected in the high value of the $BC\left[C\right]$ coefficient. In addition to $BC\left[C\right]$, there also exist statistically significant interactions between pH and carbon concentration, and buffering capacity and phosphorus concentration, with the magnitude of the coefficients of $pH\left[C\right]$ and $BC\left[P\right]$ being comparable or greater than the coefficients of $\left[P\right]$ and $\left[C\right]$ alone. Mechanistically interpreting each of these coefficients is beyond the scope of the present work, but could be pursued via additional experiments in the droplet platform or lower throughput batch cultures.

Figure 5. pH and buffering capacity modulate nutrient dependence of algae-bacteria interactions.

(A) The coefficients for regressions predicting algal and bacterial growth in coculture and monoculture in glycerol. The results for the other carbon sources are shown in Fig. S11 and Fig. S12. The top panel reports the monoculture coefficients ${\beta }_{X,M}^{Ec}$ (brown bars) and the interaction coefficients ${\beta }_{X,I}^{Ec}$ (magenta bars) of the corresponding features on the x-axis obtained for the regression model predicting the growth of E. coli in monocultures and cocultures. The interaction coefficients (magenta bars) indicate the effects of C. reinhardtii on E. coli growth with an increase in the corresponding features in coculture. The bottom panel reports the monoculture coefficients ${\beta }_{X,M}^{Cr}$ (green bars) and the interaction coefficients ${\beta }_{X,I}^{Cr}$ (cyan bars) of the corresponding features on the x-axis obtained from the regression model predicting the growth of C. reinhardtii in monoculture and coculture. The interaction coefficients (cyan bars) indicate the effects of E. coli on C. reinhardtii growth with an increase in the corresponding features in coculture. The error bars represent the $95\mathrm{\%}$ confidence intervals. ** indicates a p-value $<0.001$ and * a p-value $<0.05$. (B) Example data illustrating modulation of the effect of carbon concentration on the growth of E. coli by pH and buffering capacity. The median bacterial growth in monoculture and coculture are plotted as a function of carbon concentration at $\left[\mathrm{P}\right]\sim 1.51$ mM in the left and right panels respectively. The experimental data are represented by circles and connected with dashed lines. The error bars represent the standard error about the mean bacterial growth, with the number of replicates ranging from $\sim 14-114$ for the different conditions. The solid lines represent the model prediction. Darker or thicker lines represent the results at low $\mathrm{p}\mathrm{H}\left(6.98\right)$ and high buffering capacity $\left(2.56\mathrm{m}\mathrm{M}\right)$ and lighter or thinner lines represent the results at high pH (7.34) and low buffering capacity (0.76 mM).

Next, among the interaction coefficients containing the factors $pH$ and $BC$ (magenta bars in Fig. 5A top panel), the coefficients of $pH\left[C\right],BC\left[P\right]$ and $BC\left[C\right]$ are non-zero and compare in magnitude with their respective monoculture coefficients. This reveals that the effects of $pH\left[C\right],BC\left[P\right]$, and $BC\left[C\right]$, on bacterial growth in coculture are significantly different compared to their effects in monoculture. We conclude from this that the interaction between C. reinhardtii on E. coli is strongly impacted by pH and buffering capacity. This is a central finding of our study.

The fact that pH and buffering capacity of the environment can strongly influence interactions is illustrated by looking at a specific example from the data (Fig. 5B). Choosing a subset of data corresponding to a specific phosphorus concentration ($\left[P\right]\sim 1.51\mathrm{m}\mathrm{M}$), we compared the change in growth with carbon concentration in monocultures and cocultures at the different pH and buffering capacities values. The change in E. coli growth in monoculture with carbon concentration at the different buffering capacities shows different behavior (Fig. 5B, left panel). Particularly, the increase in the growth with carbon concentration is observed to be higher in the condition with high buffering capacity (and low pH) compared to the increase in the condition with low buffering capacity (and high pH) as expected, with the trends in the model and the data being in good agreement. Next, we compare these results to E. coli growth in coculture. The trends in E. coli growth with carbon concentration in coculture is distinct from monoculture and depends on the pH and buffering capacity values (Fig. 5B right panel). The growth appreciably declines with carbon concentration in the condition with low pH (and high buffering capacity) whereas there is an increase in growth with carbon concentration at high pH (and low buffering capacity), with the model reasonably capturing the trend in the data. These results agree with the positive coefficient of $pH\left[C\right]$ and ∼0 coefficient of $BC\left[C\right]$ obtained when the model is evaluated for E. coli growth in coculture (sum of brown and magenta $pH\left[C\right]$ and $BC\left[C\right]$ bars in Fig. 5A top panel; Fig. S13A).

In terms of interactions between E. coli and C. reinhardtii, the result can be summarized as follows: while an enhancement of E.coli growth is observed as carbon concentration increases in monoculture, the effect on E. coli by C. reinhardtii in coculture as carbon concentration increases is inhibitory at low pH and high buffering capacity, but facilitatory at high pH and low buffering capacity (evidenced by the positive interaction coefficient of $pH\left[C\right]$ and the negative interaction coefficient of $BC\left[C\right]$ obtained from regressing E. coli growth, purple bars in Fig. 5A top panel). This example illustrates that C. reinhardtii modulates the capacity of E. coli growth on carbon in a manner that depends on pH and buffering capacity of the environment.

Algal abundance dynamics also depend strongly on pH and buffering capacity. The regression coefficients for predicting algal growth on glycerol in monoculture and co-culture are shown in Fig. 5A bottom panel. In this regression, we observe a similar interplay between pH and buffering capacity and nutrient concentration i.e the monoculture coefficients of $pH\left[P\right]$, $pH\left[C\right],BC\left[P\right]$, and $BC\left[C\right]$ (green bars in the bottom panel (Fig. 5B)), are all non-zero and statistically significant, showing the presence of a modulation effect of pH and buffering capacity on nutrient concentration in determining C. reinhardtii growth in monoculture. Here again, the largest monoculture coefficient is for the $BC\left[P\right]$ term indicating an increase in the growth of C. reinhardtii with phosphorus concentration and buffering capacity. While the growth of C. reinhardtii is known to increase with phosphorus concentration⁴⁶, we speculate that the increased phosphorus uptake leads to increased N utilization (the N source here is ammonium). Ammonium utilization by algae causes acidification of the environment⁴⁷, which is known to negatively affect the growth of C. reinhardtii⁴⁸. Therefore, we reason that the environments with high buffering capacity potentially prevent this acidification and hence favor increased growth of C. reinhardtii, as reflected in the high coefficient of $BC\left[P\right]$.

The modulation of algal growth by bacteria also depends on pH and buffering capacity in a fashion similar to what we observe with bacteria. For example, the interaction coefficients of $BC\left[P\right]$ and $BC\left[C\right]$ (cyan bars in the bottom panel of Fig. 5B), being significant means that the impacts of E. coli on C. reinhardtii growth is altered by an interplay between buffering capacity and nutrient concentration. In other carbon sources, the impacts of E. coli on C. reinhardtii growth modified by an interplay between both pH and buffering capacity and nutrient concentration are observed (Fig. S11, Fig. S12).

Overall, the result that the interactions between algae and bacteria are impacted by pH and buffering capacity, through their differential impacts on nutrient dependence on monoculture and coculture growth holds across carbon sources (Fig. S11, Fig. S12).

Effect of environmental factors on algae-bacteria interactions depends on the identity of carbon source

Finally, we investigated if the dependence of algae-bacteria interactions on the environmental factors - pH , buffering capacity, phosphorus concentration, and carbon concentration, is further modulated by the identity of the carbon source available in the communities. Between several carbon source pairs, we found some apparent differences in the effect of environmental factors on algae-bacteria growth. For example, differences in several of the monoculture and interaction coefficients (which quantify the effect of environmental factors on growth and interactions) between glucose and galactose are clearly observed (Fig. 6A). While the feature $BC\left[C\right]$ has the highest effect in predicting E. coli growth in the case of glucose, $BC\left[P\right]$ is the feature with the highest importance in the case of galactose. And the effect of $BC\left[C\right]$ in predicting the E. coli growth is the opposite between glucose and galactose. Additionally, for E. coli, the coefficients of $\left[P\right]$ and $\left[C\right]$ show different patterns in glucose and galactose, with generally negative coefficients in glucose and coefficients of opposing sign for monoculture and interaction coefficients in galactose. Qualitatively similar patterns are observed in coefficients describing algal growth (Fig. S11, Fig. S12, Fig. S17). These observations suggest that the identity of the carbon source modulates how environmental factors impact algae-bacteria interactions.

Figure 6. Effect of environmental factors on algae-bacteria interactions depends on the identity of carbon source.

(A) Comparison of the regression coefficients between glucose and galactose. The monoculture coefficients ${\beta }_{X,M}^{Ec}$ (brown bars) and the interaction coefficients ${\beta }_{X,I}^{Ec}$ (magenta bars) of the corresponding features on the x-axis obtained from the regression model predicting the growth of E. coli in monocultures and cocultures for glucose (on the left) and galactose (on the right). ** indicates a p-value $<0.001$ and * a p-value $<0.05$. (B) Hierarchical clustering of carbon sources by the monoculture and interaction coefficients obtained from the regression models predicting the growth of E. coli and C. reinhardtii. The matrix showing correlations between the regression coefficients of the different carbon sources on the left and the resulting dendrogram from hierarchical clustering based on the correlation matrix on the right (See STAR Methods). (C) Hierarchical clustering of carbon sources by the median growth of algae and bacteria in monocultures and cocultures in all the environmental conditions. The correlation matrix computed for the hierarchical clustering on the left and the resulting dendrogram on the right (See STAR Methods). The colors in the heatmap correspond to the correlation values indicated by the color bar in linear scale, on the right.

To interrogate these patterns further, we classified carbon sources based on their modulation of the effect of environmental factors on algae-bacteria growth. To do this, we computed correlations between the regression coefficients (which quantify the effect of environmental factors on growth and interactions) obtained for predicting algae-bacteria growth, between all pairs of carbon sources. We performed hierarchical clustering of the carbon sources based on the monoculture and interaction coefficients of $\left[P\right],\left[C\right],pH\left[P\right],pH\left[C\right],BC\left[P\right],BC\left[C\right]$ and $\left[P\right]\left[C\right]$, obtained from the regressions for the carbon sources (STAR Methods). The correlation matrix computed for the hierarchical clustering showed that glycerol is most similar to glucose, galactose is most similar to pyruvate, and acetate has no strong correlation with any of the carbon sources (Fig. 6B left panel). Therefore, hierarchical clustering identified three clusters of carbon sources in our dataset, with glucose and glycerol forming one cluster, galactose and pyruvate forming another cluster, and acetate forming a cluster of its own (Fig. 6B right panel).

We wondered why these different carbon sources would have such divergent impacts on interactions. We first examined the metabolic pathways associated with these carbon sources but found no correlation between the nature of the carbon sources (glycolytic/gluconeogenic) and the observed clustering pattern in carbon sources. We then suspected that bacterial utilization of distinct carbon sources could have differing impacts on pH. To test this idea, we grew E. coli in plates in each of the 5 carbon sources and measured the final pH. We found that glucose and glycerol both showed large drops in pH while the other three carbon sources did not (STAR Methods). Thus, we speculate that heterotrophic utilization of organic carbon might play a key role in modulating pH and thus the interactions between algae and bacteria.

Finally, we wanted to check whether this result was dependent on the details of the regression formalism we defined for quantifying growth across environments. To do this, we quantified similarities in growth across environments in a model-independent fashion. We classified carbon sources based on the similarity in algae-bacteria growth. The classification of the carbon sources was done by computing the correlation between carbon sources in algae-bacteria growth across all the environmental conditions and culture conditions (Fig. 6C; STAR Methods). Here again, we found the carbon sources within the same clusters - glycerol and glucose, galactose and pyruvate, to have the greatest correlation in the algae-bacteria growth with each other than with any other carbon sources. We concluded that this apparent clustering of carbon sources does not depend on the details of our model specification.

DISCUSSION

By using a high-throughput droplet microfluidic platform, we were able to perform a massively parallel screening of algae-bacteria interactions in several hundreds of environmental conditions varying in pH, buffering capacity, phosphorus availability, carbon availability, and carbon source identity. To our knowledge, this is the largest screen exploring the combinatorial effect of environmental factors on phototroph-heterotroph interactions in a systematic way via a bottom-up approach. Studies in the past have tested for the effect of nutrient availability on phototroph-heterotroph relationships, but have been mostly limited to only a handful of nutrient types/availabilities or have involved uncontrolled experimental conditions such as uncharacterized phototrophic and heterotrophic species, often in the presence of organisms from other trophic levels^15,19,49,50. Our observation of the complex dependence of algae-bacteria interactions on environmental factors underscores the importance of undertaking such high-dimensional studies. This is especially important in light of the chemical complexity of environments wild microbial communities are exposed to⁹.

Our study is also novel with respect to exploring the effect of the chemical properties of the environment - pH and buffering capacity, on algae-bacteria interactions. The central finding of the study is that pH and buffering capacity substantially alter algae-bacteria interactions by manipulating the impact of nutrient availabilities on growth. For most carbon sources, the role of pH and buffering capacity in determining algae-bacteria interactions were comparable to, or significantly higher than, the effect of nutrient availabilities alone, underscoring the importance of the effects of pH and buffering capacity on algae-bacteria interactions. This result suggests that chemical factors in the environments play an important role in impacting phototroph-heterotroph interactions which are largely considered as being driven by resource exchange and competition^{15–17,19,22,25,36}.

In the context of photosynthetic metabolism, pH, buffering capacity, and alkalinity are known to be important factors impacting the availability of inorganic carbon in the environment and the physiology of autotrophs. Changes in pH alter the equilibrium between CO₂ and bicarbonate both of which can be taken up by the alga^51,52. In addition, photosynthesis alters the pH of the environment via the utilization of inorganic carbon⁵³, but this effect on pH can be altered by the presence of organic carbon at high concentrations (> 100mM) in the environment⁵⁴. Similar to bacteria, algal growth is also inhibited by large changes in pH⁵⁵. These observations are consistent with our regressions in that we find the buffering capacity to support significant and positive regression coefficients in monoculture for algae across carbon sources assayed (Fig. S14). Thus buffered media enable more robust algal growth presumably by enabling resistance to changes in pH driven by photosynthetic activity in the absence of bacteria.

Recently, microbial ecologists have encouraged the use of statistical modeling approaches to derive general governing principles in ecology^56,57. In this regard, we highlight the apparent agreement between the results of our statistical modeling and the known mechanistic processes in literature. Our statistical approach for predicting algae-bacteria growth in different environments permitted us to dissect the contribution of the different environmental factors on the inter-species interactions. Even though our modeling approach is largely agnostic to the detailed mechanisms of the effect of environmental factors on algae-bacteria interactions, we find that the regression results do align qualitatively with some known processes. For example, E. coli can acidify its environment when growing on glycolytic substrates at sufficiently high growth rates through the process of overflow metabolism⁵⁸. In this case, the bacterium could be acidifying the medium in conditions where buffering capacity is weak and carbon levels are relatively high. However, overflow occurs at relatively high growth rates of approximately 0.71/h-0.81/h, and microtiter measurements indicate that our strain in these conditions grows slower than this (STAR Methods, Table S5). Further, our measurements cannot accurately capture bacterial growth rates in droplets due to limited temporal sampling, but we cannot rule out the possibility that overflow causes growth to modify pH in the droplets. Similarly, it is known that C. reinhardtii will acidify the environment due to ammonia uptake and this may also play a role in the importance of pH and buffering capacity in determining growth in these experiments. It remains an important avenue for future work to uncover the mechanisms underlying the interactions discovered here. Our hope is that large-scale screens like those enabled by this platform can contribute new insights into the mechanisms by which environmental factors contribute to algae-bacteria interactions.

Our observation that interactions tend to favor competition (facilitation) in permissive (stressful) conditions is one example of how large-scale screens can help to identify general patterns in interactions. This observation (Fig. 3) generally supports the stress-gradient hypothesis⁵⁹. Earlier efforts to validate the Stress Gradient Hypothesis (SGH) within the realm of plants and microbes underscore the difficulties linked to the inherent ambiguity of the hypothesis⁶⁰. Specifically, it is evident that not all stressors seem to favor facilitative interactions⁶¹. Moreover, in instances where the SGH has been identified⁶², uncertainties persist regarding the existence of any shared underlying mechanisms. Our study points to the possibility that common environmental factors such as pH or buffering capacity might give rise to persistent patterns in interactions across environments. Validating this proposal would require additional measurements and physiological insights into the origins of the observed pattern.

Our exploration of the impact of carbon source identity on algae-bacteria interactions showed that the effect of the environmental factors - pH, buffering capacity, and nutrient availability, on the interspecies interactions depends on the carbon source identity. This result suggests that the chemical identity of the available reduced organic carbon plays a key role in determining how algae-bacteria interactions play out. Therefore, considering the role of individual nutrients such as phosphorous⁶³ in these interactions might be too simple a picture. Additionally, our analyses revealed three groups of carbon sources, showing that the impact of the environmental factors - pH, buffering capacity, and nutrient availability, on algae-bacteria interactions was approximately conserved between the carbon sources within the same group. Such an apparent similarity between the different carbon sources within the groups hints that there may be some relatively simple structure in how the carbon source identity and the other environmental factors conspire to determine the outcome of an interaction. Whether this is the case or not awaits a broader survey of additional carbon sources, mixtures of carbon sources, and a deeper mechanistic understanding of the physiology underlying these processes.

While kChip offers a massive throughput advantage to perform a screen of this magnitude, the interactions inferred in the confined environments of droplets on the kChip could potentially differ from the interactions in the well-mixed, open, environments in the lab or the wild. For example, the rate of gas exchange, particularly O₂, and CO₂ will determine respiration, photosynthesis, and pH and thereby modulate interactions in the droplets. In fact, a recent microfluidic-based study has shown that droplet size substantially modifies the degree of syntrophic interaction between bacterial species⁶⁴. Consistent with these findings, we observe differences in bacterial growth between microtiter plates and droplets (Fig. S18). Hence, it remains an important avenue for future work to understand how confinement impacts the algae-bacteria interactions observed here, as this process could well be important in the wild.

As our study of phototroph-heterotroph interactions was undertaken in a community of algae and bacteria that are not known to associate in the wild, it remains to be seen how our results relate to communities of phototroph and heterotroph with wild associations and shared evolutionary history. For example, the mechanism by which C. reinhardtii inhibits E. coli growth is not precisely known, and it is unclear whether other bacterial taxa would also be subjected to similar strong inhibitory effects. Studies between several strains of the phototroph, Prochlorococcus, and of oligotrophic and copiotrophic bacteria, have revealed strain-dependant interactions^65,66. Thus, it would be interesting to repeat these experiments with a broader sampling of bacterial taxa including those that are known to associate with the alga in the wild⁶⁷. By expanding this study to wild associations, we would hope to more broadly capture the relevance of these findings for consortia in complex environments.

STAR METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Seppe Kuehn (seppe.kuehn@gmail.com).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• Microscopy data and other datasets reported here are publicly available. DOIs are listed in the key resources table.

• All code required to reproduce the analyses has been deposited at Zenodo and is publicly available. DOIs are listed in the key resources table.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and Virus Strains
Escherichia coli MG1655-motile	Coli Genetic Stock Center (CGSC)	8237
Fluorescent dyes
Alexa Fluor 555	ThermoFisher Scientific	A33080
Alexa Fluor 594	ThermoFisher Scientific	A33082
Alexa Fluor 647	ThermoFisher Scientific	A33084
Deposited Data
Microscopy and other datasets reporting algal-bacterial abundances in different media conditions and the code to reproduce the analyses	This work	https://doi.org/10.5281/zenodo.12151777
Experimental Models: Organisms/Strains
Chlamydomonas reinhardtii UTEX 2244	University of Texas culture collection of algae	2244
Other
kChip droplet microfluidic platform	Kulesa et al., 2018	https://www.pnas.org/doi/abs/10.1073/pnas.1802233115

Highlights.

• Phototrophic and heterotrophic microbes reside in chemically complex environments

• A microfluidic chip enables measurements on >100,000 phototroph-heterotroph cultures

• Screen algae-bacteria interactions in ~525 environments in monoculture and co-culture

• pH, buffering capacity, and carbon source identity modulate algae-bacteria interactions