1. Physics of Living Systems

A microfluidic device for inferring metabolic landscapes in yeast monolayer colonies

  1. Zoran S Marinkovic
  2. Clément Vulin
  3. Mislav Acman
  4. Xiaohu Song
  5. Jean-Marc Di Meglio
  6. Ariel B Lindner  Is a corresponding author
  7. Pascal Hersen  Is a corresponding author
  1. UMR 7057 CNRS and Université de Paris, France
  2. U1001 INSERM, France
  3. Université de Paris, France
  4. ETH Zürich, Switzerland
  5. Eawag, Switzerland
https://doi.org/10.7554/eLife.47951
Share this article
Cite this article
  1. Zoran S Marinkovic
  2. Clément Vulin
  3. Mislav Acman
  4. Xiaohu Song
  5. Jean-Marc Di Meglio
  6. Ariel B Lindner
  7. Pascal Hersen
(2019)
A microfluidic device for inferring metabolic landscapes in yeast monolayer colonies
eLife 8:e47951.
https://doi.org/10.7554/eLife.47951
  2. Download BibTeX
  3. Download .RIS

Abstract

Microbial colonies are fascinating structures in which growth and internal organization reflect complex morphogenetic processes. Here, we generated a microfluidics device with arrays of long monolayer yeast colonies to further global understanding of how intercellular metabolic interactions affect the internal structure of colonies within defined boundary conditions. We observed the emergence of stable glucose gradients using fluorescently labeled hexose transporters and quantified the spatial correlations with intra-colony growth rates and expression of other genes regulated by glucose availability. These landscapes depended on the external glucose concentration as well as secondary gradients, for example amino acid availability. This work demonstrates the regulatory genetic networks governing cellular physiological adaptation are the key to internal structuration of cellular assemblies. This approach could be used in the future to decipher the interplay between long-range metabolic interactions, cellular development and morphogenesis in more complex systems.

https://doi.org/10.7554/eLife.47951.001

Introduction

Structured cellular populations are complex, dynamic systems and their composition, expansion and internal structure are the result of interactions between the cells and their microenvironment. Cells absorb and metabolize nutrients and also produce and secrete metabolites, creating spatial gradients of nutrients and metabolites. Thus, cells at the outskirts of a multicellular assembly do not experience the same microenvironment as the cells deeply buried within. Reciprocally, cellular physiology is dependent on the cell’s position within a colony. Such variations in cellular physiology are consistently observed in a variety of multicellular systems – from bacterial and yeast colonies (Vulin et al., 2014; Cáp et al., 2012) to biofilms (Nadell et al., 2016) and tumors (Carmona-Fontaine et al., 2013; Delarue et al., 2014) – and are reflected by altered gene expression levels and cellular phenotypes as growth rates, nutrient uptake rates and metabolic activity. Such variations presumably emerge because of long-range metabolic interactions between cells, in that the cellular microenvironment at one position depends on the nutrient uptake rate at another position.

Notably, multicellular communities (Shapiro, 1998; Shou et al., 2007; Xavier and Foster, 2007) exhibit various adaptive benefits, including higher cell proliferation, improved access to resources and niches (Koschwanez et al., 2011), collective defence (e.g., against antagonists, drugs, antibiotics) (Nadell et al., 2016) resulting in optimization of population survival when confronted with averse physical, chemical, nutritional or biological challenges (Palková and Váchová, 2006). These examples indicate the importance of understanding the emergence and maintenance of complex spatial multicellular structures from ecological (Antwis et al., 2017; Gonzalez et al., 2012; Widder et al., 2016), medical (Bryers, 2008; Gilbert et al., 2018; Estrela and Brown, 2018) and evolutionary (Ratcliff et al., 2012; Nadell et al., 2010; Kim et al., 2008) perspectives. Yet, despite the obvious contrast between homogeneous environments and the pronounced environmental heterogeneity of microbial cellular assemblies, the majority of scientific research to date has either focused on single cells in homogeneous environments or populations of cells grown in batch or continuous liquid cultures. This is mostly due to the complexity of designing an experiment that would allow monitoring, over long time, the development of a spatially defined extended multicellular assembly. This is in particular the case for the widely used eukaryotic model organism yeast Saccharomyces cerevisiae, despite the numerous calls in recent reviews to study its nutrient sensing, signaling, and related growth and development control within the natural colony context. (Conrad et al., 2014; Broach, 2012; Horák, 2013).

As microorganisms in nature tend to live in multicellular communities, devising an experimental approach that captures this complexity while being easy to use and amenable to different experimental needs and conditions should further our understanding of complex gene regulatory networks in the context of microbial evolution and ecology.

Current direct observations of three-dimensional colonies and biofilms are cumbersome and often constrained by existing technologies (Nadell et al., 2016). For example, two-photon microscopy of sliced agarose-encapsulated yeast colonies was required to show that yeast cells may adopt different physiologies – and possibly different cell types – depending on their position within a colony (Cáp et al., 2012). In another example, nanospray desorption electrospray ionization mass spectrometry (nanoDESI MS) was used to study growing bacterial colonies on agar plates and showed a wide diversity and complexity of compounds that characterize microbial chemical ecology (Watrous et al., 2012; Traxler and Kolter, 2012). Such complex methodologies are not amenable to time-lapse imaging, nor to observation of the temporal variations in gene expression and growth rates of single cells over relevant time and length scales. An alternative is to grow microbial cells in microfluidic devices to spatially constrain the growth of the cells and to control the delivery of nutrients (Bennett and Hasty, 2009; Cookson et al., 2005; Robert et al., 2010; Ni et al., 2012; Llamosi et al., 2016). Microfluidic experimental research is typically designed to ensure that the cells being studied experience a homogeneous environment. This can be done at the single cell level, as it has been demonstrated in studies of aging of yeast (Jo et al., 2015) and bacteria (Yang et al., 2019) where single cells had to be trapped and kept under constant nutrient flow for long term observations to capture their death. Alternatively, a small cell assembly can be trapped in dead-end chambers under assumption that a quick diffusion of nutrients will keep the environment in chambers homogeneous. With that approach cell lineages were tracked for bacteria (Wang et al., 2010) (the widely used ‘mother machine’) and yeast (Xu et al., 2015), cells were subjected to fluctuating environments of different carbon sources to study non-genetic memory in bacteria (Lambert et al., 2014), and bacterial colonies were synchronized through quorum sensing and gas-phase redox signalling over centimeter-length scales to produce oscillating colony ‘biopixels’ (Prindle et al., 2011). Although effective for their specific applications, unfortunately, such devices do not fully capture emerging properties at a colony level, that is spatial variations in growth rates, microenvironments and phenotypes. Recently, there have been a few attempts to use microfluidics to study collective properties of bacterial colonies grown as a microcolony. Hornung et al. (2018) grew two-dimensional bacterial microcolonies in a 75 µm long device perfused with a very low concentration (up to 585 µM) of protocatechuic acid as the only carbon source from both sides of the cell chamber. They observed heterogeneous growth in agreement with a combination of a reaction-diffusion model and particle-based simulations. In a similar setup, a 60 µm long device perfused with a very low concentration of glucose (up to 800 µM) from one side was used to study the emergence of microscale gradients that resulted in metabolic cross-feeding between glucose-fermenting and acetate-respiring subpopulations of bacteria and antibiotic tolerance by slow growing subpopulation (Co et al., 2019). Wilmoth et al. used microwells up to 100 µm in diameter to look at spatial patterns of H1-Type VI secretion system (T6SS) mutants of Pseudomonas aeruginosa accompanied with an agent-based model depicting the two observed subpopulations (Wilmoth et al., 2018). They found that spatial constraints and local concentrations of growth substrates affect the spatial organization of cells. Finally, Liu et al. grew Bacillus subtilis biofilms perfused with glycerol and glutamate media. They discovered collective oscillations which emerge as a consequence of long-range metabolic co-dependence between cells in the interior and cells at the periphery of a biofilm, presumably to maximize the availability of nutrients and survival of interior cells (Liu et al., 2015). While the use of microfluidics gave rise to the discovery of collective properties of microbial assemblies and facilitated quantitative observations, such attempts are currently limited by inherent difficulties tied to fabrication of efficient microfluidic devices and choice of model organism. These limitations include small device dimensions (<100 µm) (Hornung et al., 2018; Co et al., 2019; Wilmoth et al., 2018), use of low nutrient concentrations (<1 mM) (Hornung et al., 2018; Co et al., 2019), limited scope of nutrient types (Hornung et al., 2018; Co et al., 2019; Liu et al., 2015) and in some cases inability to access single cell level measurements (Wilmoth et al., 2018). Therefore, it is challenging to apply such devices for the general case of the study of a large monolayer of cells in standard range and scope of nutrients, often used in biological research in liquid cultures. Additionally, it is tempting to reconstruct the emergence of gene expression landscapes on a global scale (e.g., within structured populations) from local (e.g., single cell) properties, given the extensive knowledge accumulated on single-cell gene regulatory networks. However, the variations in the microenvironment within a multicellular assembly and their interconnections with gene expression and cell metabolism are rather poorly known.

In attempt to overcome the above limitations in current methodologies and observe emerging properties at a colony level in larger dimensions and standard nutrient conditions, we developed a microfluidic device to grow thin, extended arrays of yeast cell monolayers that are perfused with nutrients from a single direction. We demonstrate the emergence of heterogeneous microenvironments and quantify spatial variation in cellular growth rate and the formation of gene expression landscapes for key metabolic genes involved in glucose transport and utilization, across the nascent 2D microcolony in 800 µm long cell chambers and up to 444 mM glucose concentration. Interestingly, the gene expression landscapes exhibited a high degree of spatial correlation over a wide range of glucose concentrations. Notably, we show that a growing extended assembly of cells presents a robust, steady state spatial structure, transitioning between fermentative (high glucose environment, fast growth, rapid glucose utilization) and respiratory (low glucose environment, slow growth, slow but efficient glucose utilization [Pfeiffer and Morley, 2014; Hagman and Piškur, 2015]) regimes, located close to and far from the nutrient source, respectively. This spatial structure emerges from the interplay between how cells individually adapt to the microenvironment and, at the same time, alter their surroundings as a result of their metabolic activity. Said differently, structured cells create and experience a spatially structured micro-environment through the interplay of nutrient diffusion and uptake without any obvious inherent biological program that would imply cell-cell communication and coordinated community action.

Results

Growing extended yeast monolayers

Microfluidic systems are usually designed to ensure a homogeneous microenvironment for all cells (Bennett and Hasty, 2009). In contrast, in this study, we designed a microfluidic device – dubbed the ‘yeast machine’ – to grow long, narrow yeast monolayers with the aim of observing the emergence of nutrient gradients and spatial variations in cellular growth and gene expression landscapes. We used soft lithography techniques to fabricate a multi-layered microfluidic device composed of a large channel (to flow nutrients) and an array of perpendicular, extended (800 µm-long), narrow (50 µm-wide), flat (4.5 µm-high) dead-end chambers in which yeast cells can grow as monolayers while the media is supplied by a pressure pump-based system with flow control (Figure 1, Figure 1—figure supplement 1). The length of the dead-end chambers was optimized to induce significant variations in the nutrient concentrations within the chambers due to cellular nutrient uptake. The chamber width was large enough to avoid jamming during cell growth due to geometric constraints and small enough to avoid generation of complex, cell-recirculating flows induced by cell growth (Boyer et al., 2011). The chamber height was comparable to – but slightly larger than – the average size of a yeast cell, so the cells were vertically constrained to facilitate single-cell imaging and time-lapse fluorescence microscopy (Figure 1—figure supplement 2).

Figure 1 with 2 supplements see all
Download asset Open asset
Microfluidic device setup and design.

(a) Media reservoirs are pressurized with the help of Fluigent MFCS pressure pump resulting in flow through the flow sensors, into the chip and then to waste. Flow sensors and pressure pump are connected to the flow-rate control module, which maintains a constant flow through the system. Nutrient supply and media conditions can be changed in real time. (b) Each single ‘yeast machine’ has two sets of cell chambers of various widths (5 µm, 10 µm, 25 µm and 50 µm). The Cell chambers are connected perpendicularly to a large flow channel (1 mm wide, 25 µm high). This design facilitates adaptation for different model systems (e.g. bacteria, yeast, mammalian cells) and high-throughput depending on the of predefined flexible length, width and height flexible adapted dimensions. (c) A close-up sketch of a set of cell chambers used in our experiments. They are 800 µm long, 50 µm wide, and 4.5 µm high. A single cell chamber fits a monolayer of up to 2500 yeast cells. The whole setup is mounted on a microscope for time-lapse fluorescent imaging.

https://doi.org/10.7554/eLife.47951.002

The cells were injected into the main channel of the ‘yeast machine’ and then forced into the dead-end chambers by centrifugation using a homemade 3D-printed holding device attached to a spin coater (see Figure 1—figure supplement 1; Materials and methods). The main channel was washed with yeast synthetic complete growth medium to remove excess cells; cells that were trapped in the dead-end chambers by centrifugation were not removed by the washing step. Nutrients were flowed through the main channel and could passively diffuse into the array of dead-end chambers. The cells formed growing monolayers that extended from the closed end of the chamber and collectively progressed towards the nutrient source (i.e. the open end of the chamber) as the cells pushed each other while growing (Figure 2a,b; Video 1). Cells eventually filled each chamber, forming an extended two-dimensional colony composed of about 2500 cells (Figure 2b, Figure 1—figure supplement 2), typically ~10 cells wide and ~200 cells long. Cells could be observed locally at high magnification (100 × objective), while the whole assembly could be seen at low magnification (10 × objective). We recorded the cellular expansion and subsequent internal dynamics of these long monolayers, as well as the landscape of expression of key fluorescently tagged endogenous genes, over time and over an almost 1000-fold range of glucose concentrations (from 0.01% to 8% w/vol).

Video 1
Download asset
Emergence of metabolic landscapes in yeast monolayer colonies.
https://doi.org/10.7554/eLife.47951.005
Figure 2 with 2 supplements see all
Download asset Open asset
Expansion and dynamics of extended cellular monolayers.

(a) The microfluidic device is perfused with nutrients using a pressure-driven system (see Figure 1). Yeast monolayers extend within the long chambers: front velocity (VF) and local velocity (Vz) are determined by cellular growth and division. (b) Example of a time-lapse collage of yeast monolayer expansion along an 800 µm-long chamber (2% w/vol. glucose, 5 × amino acid concentration). Front velocity increases and reaches a plateau (indicated by flattening of the slope of the green curve). When the front approaches close to the open end of the chamber (i.e., 0 µm), the over-spilling cells are constantly washed away by the nutrient flow within the main channel. ( c) Front velocity reaches a maximum when the position of the front becomes close to the open end of the chamber indicating that after expanding by a typical distance (~400 µm here for 2% w/vol. glucose), the maximal number of cells that receive glucose and can participate in expansion has been reached. 340 velocity data points binned into 10 equally spaced position points were extracted from n = 12 colony front trajectories (2% w/vol. glucose). The error bars denote standard deviations of each bin (~15–30 velocity data points). (d) Front velocity as function of external glucose concentration. Data comes from the bin closest to the open end of the chamber as measured in Figure 2c for each glucose concentration (n > 5). Error bars denote standard deviations. (e) Local cellular motion can be assessed by computing the standard deviation of pixel intensities across a stack of time-lapse images. Here, white areas indicate variations in movement across the time-lapse for cells below 400 µm, while the cells above do not move. Averaging over several channels (n = 9), we obtained an indicator of cell motion and thus an estimate of the glucose penetration distance, H (~400 µm for 2% glucose). (f) Local velocity decreases for cells deeper within the chamber. Local velocity also increases with external glucose concentration. Velocity Data, that were binned into 16 equally spaced position, comes from the analysis of >100 cell trajectories. Error bars denote standard deviations.

https://doi.org/10.7554/eLife.47951.006

Monolayer in expansion displays regions of fast and slow growth

Expansion of the monolayers of cells was observed by microscopy at low magnification (10× objective). Under standard glucose-rich conditions (2% w/vol; 111 mM) and excess amino acids (5× CSM, see Materials and methods), the front velocity, VF, increased during the first 2-4 h and eventually reached a steady-state close to 100 µm.h−1 (Figure 2c,d, Video 1). Front velocity is the sum of the contribution of every cell to colony expansion. Therefore, VF depends on the quantities of glucose and other nutrients that penetrate inside the yeast monolayer, which impact both the number of cells that grow and their growth rates. Initially, the monolayer is sparsely populated and sufficient glucose is expected to reach all cells. After growth and division, a larger number of cells can participate in global expansion of the population. Thus, the front velocity is expected to quickly increase over time. However, at some point, as the size of the monolayer increases, the cells close to the dead end of the chamber will stop growing (due to absorption and metabolism of available nutrients by cells closer to the nutrient source/chamber opening) and the front velocity will plateau. Hence, after the cell chamber populates with cells completely, a steady-state is reached where a constant number of cells with access to glucose continue to divide and move passively towards the nutrient source, while the number of cells at the dead end of the chamber deprived of glucose (and other nutrients) remains unchanged. If we consider the ideal case in which yeast cells are 4 µm-wide and divide every 90 min in the presence of glucose, each cell layer leads to an expansion of 4 µm every 90 min, or 2.6 µm.hr−1. The observed terminal front velocity of 94 ± 8 µm.hr−1 (Figure 2) can be attributed to the first 36 ± 3 layers of cells, that is the first 140 µm of the colony. The glucose penetration distance can be approximated by assuming (Vulin et al., 2014) that glucose – of which the concentration is maintained at C0 at the front of the monolayer – freely diffuses within the assembly with a diffusion coefficient D ~ 100 µm (Cáp et al., 2012).s−1 and is absorbed by cells at a constant rate, q0, of ~ 1 mM.s−1. Diffusion law dictates that the glucose concentration is expected to decrease significantly after a typical distance, H, that scales with DC0q0 ~ 100 µm. Our direct observation (Figure 2e) showed that for a layer of growing cells, H is around 400 µm at 2% w/vol glucose. Notably, both estimations are in agreement, albeit they underestimate the observed size of the growing layer. These discrepancies result from discarding the decay in the cellular growth rate at decreasing glucose concentrations and the variation in the specific cellular uptake rate, q, with glucose concentration. Indeed, the interplay between glucose diffusion and uptake is central to structuration of the colony as it affects both the number of cells that have access to glucose and the glucose concentration in the microenvironment of each region, and thus determines which cells actually participate in colony expansion and by how much (Vulin et al., 2014). The true glucose penetration distance is therefore likely to be larger than the above ‘guesstimate’. Yet, inferring the true penetration distance would require a detailed model of the dependency of both cellular glucose absorption and the growth rate on the glucose concentration, as well as experimental measurements of the glucose concentrations within the monolayer. This outlines the difficulty of predicting the internal structure of a simple yeast monolayer due to our limited understanding of how yeast cells interact with nutrients and the difficulty of obtaining quantitative details of the microenvironment landscapes within a yeast monolayer. In the following text, we quantify the expression of different glucose concentration-dependent transporters as a possible proxy for intra-colony glucose concentration. We even ventured further, to study how landscapes of cellular growth and expression of key genes involved in glucose transport self-emerge from long-range metabolic interactions within the yeast colony.

Front velocity increases with glucose concentration

Increasing the glucose concentration (from 0.01% to 8% w/vol) led to higher terminal front velocities (Figure 2d), in agreement with the fact that at higher concentrations, glucose will penetrate further by diffusion in the colony (Figure 2a). Thus, increasing the concentration allows a larger number of cells to access glucose and participate in the growth of the colony. Yet, the front velocity does not increase linearly with glucose concentration, and plateaus at very high glucose concentrations (>4% w/vol). One interpretation is that at this concentration range, sufficient glucose reaches the dead end of the chamber, allowing all cells to participate in the growth of the assembly. However, based on VF ~ µL, where L is the length of the dead-end chamber and µ is the average cell growth rate, one would expect a saturating front velocity of 368 µm.h−1, much larger than the measured value of 100 µm.h−1.

Glucose is not the only nutrient required for cellular growth; amino acids can be a limiting factor for auxotrophic strains such as the one employed in this study (S288C background). This is why we used an excess of amino acids (5 × CSM) compared to classic SC medium for yeast cell cultures. Indeed, using standard amino acid concentrations in the media resulted in significantly lower terminal front velocities, even at high glucose concentrations (Figure 2—figure supplement 1). This suggests that amino acid availability can limit cellular growth, which is especially visible in the presence of high glucose concentrations, where glucose is no longer limiting but amino acids are. As all experiments were performed under 5-fold higher amino acid concentrations than normal SC medium, other metabolites that are consumed are likely to form gradients within colony and might become rate-limiting for growth. Taken together, we conclude that the spatial variations in all metabolic components of the microenvironment need to be taken into account in order to fully understand microbial colony growth. With that in mind, building a mathematical model to account for the observed expansion of a spatially structured colony is barely achievable, and we will not address this question here. Rather, we opted to further characterize the development of glucose gradients as a specific and critical component of the emergence of the metabolic landscape of the colony.

Local expansion rate decreases with distance from the nutrient source

Once the dead-end chambers were filled with cells, we found that similar growth pattern emerged across parallel chambers, specific to each glucose concentration. The cells closer to the open end of the chambers continued to divide, pushing cells out that were washed away by the flow in the nutrient channel. Cells closer to the dead end (y ~ 800 µm) did not move, grow nor divide. At standard glucose conditions (2% w/vol) and a high amino acid concentration (5 × CSM), significant cell motion was not observed after y ~ 400 µm, indicating that very limited glucose is available to the cells that are beyond this region. By tracking single cell trajectories, we measured the velocity field within the yeast monolayers over a range of glucose concentrations. We extracted >100 single cell trajectories per concentration, resulting in thousands of velocity data points (see Materials and methods). As expected, increasing the glucose concentration in the nutrient channel (from 0.01% to 8% w/vol) led to higher local velocities deeper in the colony (Figure 2f, Figure 2—figure supplement 2). Concomitantly, velocity also increased closer to the chamber opening when cells experienced a higher glucose concentration.

In summary, our setup captures the essence of structured colonies, with the emergence of a landscape of growth divided into a non-growing area and actively growing area. This spatial separation is the result of the formation of glucose (and other nutrient) gradients. These gradients emerge as a result of cellular metabolic activity, which in turn affects the cellular growth rate and physiology at the local scale.

Cellular metabolic activity creates gene expression landscapes

The emerging glucose (and other nutrient) gradients are expected to both trigger and be governed by differential gene expression landscapes. To this end, we studied the expression of seven key glucose transporters (HXT1-7) whose expressions are regulated by the extracellular glucose concentration. We employed yeast strains in which these endogenous glucose transporters were tagged with GFP (Materials and methods), and recorded the fluorescence signals at the global scale using a low-magnification objective (10×) and local cellular scale using a high-magnification objective (100×). Cells were loaded into the chambers as described above and observed after the establishment of a quasi-steady state (starting 10 hr after the chamber was filled with cells, Figure 3—figure supplement 1). We observed the formation of different landscapes of gene expression for each of the seven transporters, each with marked territories of low and high expression (Figures 3 and 4; Materials and methods). In particular, HXT1 and HXT7 displayed inversely correlated landscapes of gene expression (e.g., Figure 3a and g for 2% w/vol glucose). Both patterns demonstrate the formation and maintenance of a glucose gradient that emerges from cellular metabolic activity. HXT1 is a low-affinity glucose transporter mainly expressed under high-glucose conditions, while HXT7 is a high-affinity glucose transporter expressed under low-glucose conditions only (Figure 3b and f) (Reifenberger et al., 1997; Diderich et al., 1999; Maier et al., 2002). Concomitantly, HXT1 was expressed at the highest levels in the cells close to the chamber opening (i.e., in the highest glucose concentration), while HXT7 expression peaked further away in the chamber, indicating a transition to a low-glucose region. We examined the cells at higher magnification (60×) to assess the localisation of HXT7 gene expression. As expected, in the cells expressing the highest levels of this gene, the fluorescence was localized to the cell membrane, indicating HTX7 played an active role in glucose transport in these cells. In contrast, deeper in the colony, we observed lower levels of HTX7 fluorescence due to the long lifetime of GFP-fused proteins and absence of dilution through cell division, though this fluorescence was localized in vacuoles, indicating the transporter had been targeted for degradation by the cells (Hovsepian et al., 2017) (Figure 3a). Assuming the observed peak of HXT7 fluorescence matches the peak fluorescence observed in batch culture at a glucose concentration of 0.016% w/vol. (Figure 3b,c, Figure 3—figure supplement 2), we could locate the position in the yeast monolayer at which the glucose concentration reached 0.016% w/vol. This position was around Hf ~ 500 µm from the front, in good agreement with the transition in cell motion (Figure 2, Hm ~400 µm).

Figure 3 with 3 supplements see all
Download asset Open asset
Landscapes of gene expression self-emerge in extended yeast monolayers.

(a) Expression profile of HXT7-GFP along the chamber (average fluorescence levels, n = 9; standard deviation shown as the envelope) for an external concentration of 2% w/vol glucose. Membrane localization of HXT7 was only observed in the cells surrounding the area of peak HXT7 expression, localized at ~500 µm at 2% w/vol. glucose. (b) FACS measurements of HXT7-GFP expression in batch culture (average of three replicates) showing a single intensity peak at C0 = 0.016%. This peak value can be mapped back to the spatial landscape of 3a to infer the glucose concentration in the region of peak HXT7-GFP fluorescence. n = 3–6 per glucose concentration (c) On varying the glucose concentration in the nutrient channel, we observed a transition in peak HXT7-GFP fluorescence within the 2D colony. At a concentration of 4% w/vol and above, the peak was located close to the dead end of the chamber or not visible, indicating sufficient glucose was available throughout the chamber (color code normalized to maximal expression level). Data obtained from n = 8–17 replicates per glucose concentrations (see also Figure 3—figure supplement 2). (d) Compared with 3b, it is possible to roughly define areas of glucose presence in the monolayer for a range of glucose concentrations (n = 8–17, per glucose concentrations, error bars denote ± one standard deviation). (e) Landscape of HXT1-GFP gene expression over a range of glucose concentrations (color code normalized to maximal expression); n = 8–9 per glucose concentrations (Figure 3—figure supplement 3). (f) FACS measurements of HXT1-GFP over a range of glucose concentrations; n = 3 replicates. 3 g. Overlay of HXT1 (red) and HXT7 (green) gene expression landscapes at three external glucose concentrations, showing that the expression landscapes of these transporters were inversely correlated, in agreement with their different glucose-dependent expression patterns (compare 3b and 3 f).

https://doi.org/10.7554/eLife.47951.009
Figure 4
Download asset Open asset
Using the fluorescence landscapes of glucose transporter gene expression to infer glucose concentration gradients.

(a) FACS measurements for HXT1-GFP to HXT7-GFP in batch culture over a range of glucose concentrations. The expression levels of each HXT show a specific dependence on glucose concentration (n = 3–6 replicates per glucose concentration). (b) Landscapes of gene expression for all HXTs-GFP at an external glucose concentration of 2% w/vol. HXTs are ordered by their relative glucose specificity: HXT1 is expressed under high-glucose conditions, while HXT5 is only expressed at very low-glucose conditions. Assuming a progressive spatial decay in the glucose concentration away from the chamber opening, all maps of gene expression are in perfect agreement with the intensity profiles observed in batch culture (n = 8–10 replicates per glucose concentration). (c-d) Method of glucose gradient reconstruction. The fluorescence landscape of HXT7 (resp. HXT1) shows a peak Fmax (resp. a minimum, Fmin) at a given location. The fluorescence intensity at the opening of the chamber, F0, corresponds to the external glucose concentration, C0. Using the FACS measurements of HXT7 (resp. HXT1) as a function of glucose concentration, one can define the concentration of glucose that matches the peak Fmax (respective to the minimum Fmin), and the fluorescence intensity that corresponds to C0. This allows us to linearly map all other fluorescence intensities for a given glucose concentration from the batch culture to the fluorescence intensities inside the colony, allowing the glucose concentration across the entire cellular monolayer to be reconstructed. Data comes from previously mentioned HXT1 and HXT7 microfluidics and flow cytometry measurements. (e-f) Reconstruction of glucose concentration obtained from HXT7 (e, f) and HXT1 (f) fluorescence data and various external glucose concentrations.

https://doi.org/10.7554/eLife.47951.013

Gene expression landscapes depend on the glucose source concentration

Increasing the glucose concentration in the nutrient channel changed the gene expression landscape of all seven glucose transporters (Figures 3 and 4). In particular, at 1% w/vol glucose, HXT1 was only expressed at low levels at the growing front of the colony (y < 60 µm). In contrast, at the highest glucose concentration (8% w/vol; Figure 3e, Figure 3—figure supplement 3), HXT1 was expressed at high levels throughout the whole colony, demonstrating glucose was available throughout the chamber. As HXT1 is mainly expressed under high-glucose conditions (>1% w/vol glucose) in batch culture (Diderich et al., 1999), this observation indicated the glucose penetration distance (within the chamber) increased with the external glucose concentration. This is in agreement with the increase in local velocity with the external glucose concentration in Figure 2, with the size of the growing area also increasing with the external glucose concentration.

In contrast, HXT7 exhibited a peak-like expression pattern, and was repressed under both high-glucose conditions and when no glucose was present. At low-glucose concentrations (0.1% w/vol), a peak in HXT7 expression was observed at the very beginning of the colony (y ~ 20 µm), indicating glucose was quickly absorbed by the cells closest to the chamber opening, thus these were the only cells with access to sufficient carbon resources to grow and divide. The peak of HXT7 expression moved deeper into the colony as the glucose concentration increased and disappeared completely at 8% w/vol glucose, again indicating sufficient glucose could diffuse to the end of the chamber under high-glucose conditions (Figures 3 and 4).

Reconstructing glucose concentration landscapes using glucose transporter gene expression levels

We assessed the expression profiles of HXT1-7 in batch culture as a function of glucose concentration (see Materials and methods) to obtain a qualitative idea of the glucose concentrations within the microfluidic device. The data for HXT7 was particularly revealing: its rather sharp, well-defined expression peak at 0.016% w/vol allowed to define the distance in the microfluidic device at which the glucose concentration is close to that value (Figure 3a,c). This concentration boundary separates the yeast monolayer into two regions with different properties, that is actively dividing and growth arrest. The position of this boundary moved deeper into the colony as the external glucose concentration increased (Figure 3d).

We extended this idea further and used the complete HXT7 expression profile to infer the glucose concentrations at all positions within the chambers. Assuming that the local level of HXT7 expression is only set by the local glucose concentration, we can use batch culture measurements of HXT7 expression (based on flow cytometry) to determine the glucose concentration at a given chamber position (Figure 4c and d). However, this only allows us to reconstruct the glucose concentration gradient up to 0.016% w/vol., that is in the domain where cells are actively dividing. The idea is simply to linearly map the two sets of measurements (in batch culture and in the microfluidic device) based on the fluorescence levels that correspond to the maxima Fmax and F’max and HXT7-GFP fluorescence levels at the chamber entry F0 and F’0. Using the data for HXT7 in Figure 3, we were able to reconstruct the glucose gradient for different initial glucose concentrations (Figure 4e). When applied to HXT1, the same inference led to very similar results (Figure 4f). In both cases, glucose concentrations decay very quickly moving away from the chamber opening and then exhibit a relatively long tail moving deeper into the colony.

Gene expression landscapes of other genes and transcription factor activity confirm the inferred glucose gradients

The fact the seven glucose transporters exhibited varied, robust spatial expression patterns under identical conditions (e.g., Figure 4a), together with the observed growth rate landscapes (Figure 2), suggests cellular metabolic state varies significantly across the longitudinal axis of the yeast monolayers. This variation was further assessed by mapping the expression and localisation of additional key genes involved in glucose metabolism.

MIG1 is a key transcription factor involved in glucose repression that localizes to the nucleus in the presence of glucose, to repress genes that participate in parallel carbon metabolic pathways (e.g., galactose) (Conrad et al., 2014; Broach, 2012). Observing the cells at high magnification, we quantified the distance after which MIG1 fluorescence was not present in the nucleus of the cells (Figure 5b). This distance, around 400 µm at C0 = 2% w/vol glucose, was in excellent agreement with the data obtained by HXT7 profiling. Interestingly, the spatial transition from nuclear MIG1 to cytoplasmic MIG1 localisation was very sharp and occurred over just a few cells.

Figure 5 with 3 supplements see all
Download asset Open asset
Other landscapes of genes involved in glucose metabolism.

(a) Landscape of HXT5 expression. HXT5 is expressed under very low and no glucose conditions and appears to be a good marker of growth arrest. At C0 = 2% w/vol, HXT5 expression is in good agreement with the observed absence of cellular division (see Figure 2, Figure 5—figure supplement 1). (b) Landscape of MIG1 activity. MIG1 fluorescence was located in the nucleus in the presence of glucose, with a sharp transition in nuclear localization observed (middle picture, at 2% w/vol glucose in the nutrient channel), confirming the existence of a glucose gradient (n = 3 replicates). Total number of cells and cells with nuclear localization of fluorescence were annotated manually and binned into 25 µm bins (see also Figure 5—figure supplement 2). (c) HXK1 and HXK2 are hexokinases involved in glucose metabolism. Their landscape of expression exhibited peaks that indicate a transition from high to very low glucose levels (n = 8–9 replicates per glucose concentration). (d) FACS measurements of HXK1 and HXK2 expression over a range of glucose concentrations (n = 3–6 replicates per glucose concentration).

https://doi.org/10.7554/eLife.47951.014

In agreement with the batch culture observations, we found HXT5 was only expressed in regions with very low or no glucose concentrations where the cells did not seem to divide over several hours (Figure 5a). Therefore, HXT5 appears to be an excellent marker of growth arrest in this context (Verwaal et al., 2002).

The expression landscapes of two hexokinases involved in glucose metabolism, HXK1 and HXK2 (Figure 5c) that are expressed when cells are grown on non-glucose carbon sources, were also consistent with the batch measurements (Figure 5d, Figure 5—figure supplement 3) and further validated the existence of a glucose gradient. For each profile, we extracted the position of maximal expression and inferred the glucose concentration at this position from the FACS measurements of batch cultures. The batch measurements indicated maximal HXK1 and HXK2 expression were observed at a glucose concentration of about 0.016% w/vol. As expected, neither enzyme was expressed at very high glucose concentrations. The HXK1 and HXK2 expression maxima were similar at the two other glucose concentrations studied, around 300 µm at C0 = 1% and 500 µm at 2% w/vol. Again, these data are in very good agreement with the positions of HXT7 peak expression at the same glucose concentrations.

Finally, we examined the expression of PDC1 and SDH2, which are overexpressed in fermenting and respiring cells, respectively (Otterstedt et al., 2004; Bonander et al., 2008; Ohlmeier et al., 2004). Their expression landscapes were inversely correlated (Figure 6a, Figure 6—figure supplement 1), indicating a transition from fermentative metabolic activity at the nutrient front of the colony to respiratory metabolic activity towards the dead end of the chamber where glucose is scarce. These expression maps are in good accordance with our previous results (Figures 2, 3 and 5) and the levels of PDC1 and SDH2 expression in batch culture (Figure 6b and c).

Figure 6 with 1 supplement see all
Download asset Open asset
Impact of the glucose gradient on yeast physiology and the emergence of a landscape of phenotypes.

(a) Overlay of the landscapes of gene expression of PDC1 (blue) and SHD2 (pink). PDC1 is known to be expressed when yeast cells ferment, SDH2 is mainly expressed in respiring cells (see also Figure 6—figure supplement 1). (b) FACS measurements of PDC1 expression over a range of glucose concentrations in batch culture (n = 3). (c) FACS measurements of SDH2 expression over a range of glucose concentrations in batch culture. The inverse correlation between PDC1 and SDH2 expression observed in batch culture is in good agreement with the inversely correlated spatial expression patterns within yeast cell monolayers (n = 3).

https://doi.org/10.7554/eLife.47951.018

Multiple gene expression landscapes are spatially correlated

We decided to compare the landscapes of gene expression for the entire set of reporter genes by aligning the different landscapes across varied nutrient conditions (Figure 7a). Strikingly, all landscapes showed a high level of spatial correlation. Two major landscapes emerged: peaking (e.g., HXT7) and switching (e.g., HXT1 or MIG1). We defined and extracted the typical lengths of the peaking and switching landscapes (Figure 7b) and plotted them as function of the external glucose concentration (Figure 7c). The typical lengths of all of these landscapes for different reporter genes were remarkably close, despite the fact that we looked at different cellular components: a transcription factor (MIG1), glucose transporters (HXTs), metabolic enzymes (HXKs) and metabolic state reporters (SDH2, PDC1). Notably, we gained a global view of gene expression landscapes and their interrelationships along a monolayer colony. All data showed the colonies were structured into two regions with very different properties (Figure 7d): an actively growing region, where cells divide abundantly and ferment glucose, and a quiescent area, where cells do not divide much and have switched to respiratory metabolism to compensate for the very low glucose availability. While it is not surprising to see the expression levels of metabolic genes vary with the glucose concentration, our approach demonstrates genetic programs not only allow individual cells to adapt to changes in the nutrient environment, but also enable multicellular assemblies to spatially self-organize through long-range metabolic interactions governed by physical rules of diffusion and uptake. This sheds new light on the coordinated actions of these genes in individual cells in a biologically relevant multicellular context that has impact on ecology, evolution, development and emergence of multicellularity.

Figure 7
Download asset Open asset
Global view of the emergence of landscapes of gene expression.

(a) The different landscapes of gene expression presented in this study are aligned, regrouped and displayed over a range of glucose concentrations. This simple view sheds light on the macroscopic spatial correlations between these different landscapes, which are both setting and traces of the establishment of glucose gradients. (b) For each gene expression landscape, we identified the fluorescence peak (HXT7, HXK1, HXK2) or the position of the transition between low and high expression (HXT1, HXT5, SDH2, PDC1) or activity of the transcription factor (MIG1). (c) Landscapes of gene expression delimit two regions in which cells are physiologically different. Phase I indicates active growth by fermentation in the presence of glucose; Phase II indicates growth arrest or very limited growth via respiratory metabolism at zero or close to zero glucose concentrations. The transition between the two phases typically takes place relatively sharply, over a hundred micrometers or ~20 cells.

https://doi.org/10.7554/eLife.47951.020

Overall, we studied how cells within a monolayer colony collectively shape their microenvironment through long-range metabolic interactions. This is a complex process, in which individual cells adapt locally, and shape a spatial landscape of gene expression as a global phenotype. As a whole, the structure of an assembly of cells and the microenvironment landscapes emerge as the result of local cellular metabolic activity of individual cells.

Discussion

Here, we took an alternative point of view compared to traditional systems and single-cell biology. Rather than studying single-cell metabolic properties in a well-mixed, homogeneous environment, we designed a microfluidic chip to force yeast cells to grow and shape their microenvironment, solely by fixing the properties of the microenvironment at the boundary of the monolayer. This approach allowed us to measure simultaneously properties at both the single-cell scale and structured population scale and holds potential for establishing a quantitative link between these scales.

In standard batch culture, as cells exhaust the media, their adaptation time, limited by sensing, transcription and translation, may lag behind the decrease of glucose in the media. In contrast, in our 2D colony device, once steady-state gradient is settled, cells’ residence time within a given range of glucose that corresponds to stable HXT expression is significantly longer, assuring that even though cells are continuously pushed away, they spend sufficient time within a given concentration to equilibrate their response to the gradient and therefore faithfully report the glucose concentration in their environment. As example, at 1% glucose the mean velocity at Hxt7 expression peak, spanning over >200 µm (Figure 3a, varies between 5–15 µm/hour (Figure 2f), suggesting cells’ residence time of >10 hr, pointing to the advantage of working in this setup.

Specifically, we showed that cells self-generate nutrient landscapes that in turn influence cellular metabolism and gene expression profiles. This behavior, based on nutrient uptake adaptation, is generic and feeds back on the behavior of other cells through what we call non-specific long-range metabolic interactions. Indeed, the microenvironment sensed by cells a few hundred micrometres inside a colony is very different from the microenvironment experienced by the cells at periphery. The resulting patterns may emerge from the individual adaptive properties of the cells without the need to evoke specific higher-level community properties as cell-cell communication. Notably, gradients emerge over relatively short distances, and this process may possibly affect studies of cellular populations within microfluidics settings. More importantly, quantitative description of gene expression landscapes is critical if one wants to understand the establishment and behavior of cellular populations, whether these are as simple as yeast colonies or more complex, such as biofilms and complex microbial ecosystems in which several types of cells cohabit and may specifically communicate and interact. Indeed, in addition to the described long-range metabolic interactions, many other environmental and genetic determinants such as intercellular communication, cell surface properties, cell-cell adhesion strength and secretion of extracellular matrix components have been shown to participate in the emergence of the complex morphology (Nadell et al., 2016; Granek and Magwene, 2010; Flemming et al., 2016) and internal structure of microbial colonies in such complex situations. The nature of many of these interactions could also be studied using similar microfluidic devices to identify the relative contribution and relationship of environmental and genetic determinants to the metabolically generated microenvironment.

We have made notable advances in the study of emerging properties of yeast colony growth, microenvironment formation and gene expression compared to previously published studies (Cáp et al., 2012; Maršíková et al., 2017; Palková et al., 2014). These studies have shown fascinating differentiation and diversity within yeast colonies grown on agar but their relevance to study the dynamical emergence of complexity in microbial colonies is limited by their methodology (e.g., growth on a single specific medium with no dynamic control of environmental changes, two-photon microscopy, unsuitable for live time-lapse microscopy, obligation to section colonies etc.) which does not allow detailed spatiotemporal analysis of cellular growth, microenvironment and gene expression landscapes at a relevant single-cell scale. Our approach is designed to access the dynamics of large microbial colonies, and while we did not report it here, it is straightforward with microfluidics to dynamically change in frequency and composition the external environment, and as such to analyse how colonies adapt their internal organization to such stresses.

Our results are in most part in line with the knowledge of glucose metabolism obtained in batch culture. Yet, our methodology sheds quantitative description of the spatial expression of genes involved in the glucose metabolism and its correlation with the cell local growth rates. Our results show that even in the simple context studied here, reconstructing the microenvironment spatial structure from single-cell measurement is not trivial. A proper model should take into account how the growth rate and specific absorption rate vary with the glucose concentration and the microenvironment. Modeling the entire complexity of the microenvironment is hardly possible, even with today’s knowledge. Thus, we decided to take a different approach and use key genes involved in glucose metabolism to infer the glucose concentration gradient. We showed that different reporter genes consistently reported the same glucose gradient. We envision that the data extracted from relevant fluorescent reporters could be fed into an agent-based or mean-field models that take cell-cell interactions, mechanics and spatial diffusion of metabolites into account to fill the gap between data generated from single cells to data that is relevant to evolution and ecology, that is at the colony scale. We anticipate that linking local properties to macroscopic, global behavior will help to understand the architecture of microbial communities and how evolution shapes the development of these architectures through long-range metabolic interactions and possible inherent biological programs that coordinate it. Of note, in another rare attempt to study emergence of population level phenomena in yeast S. cerevisiae Campbell et al. (2015) looked at the synthetic ‘self-establishing communities’ that were able to cooperatively exchange metabolites. They inoculated on agar plate auxotrophic S. cerevisiae strain that had different auxotrophic markers on plasmids. As cells were dividing, some of the plasmids that complemented yeast auxotrophy and therefore rescued their growth were lost, resulting in a colony which is composed of yeast that are auxotrophic for a certain amino acid. However, they were able to grow because they used amino acids that were released in the environment by other yeast that were producing it, effectively generating a very heterogeneous colony that sustained growth through metabolite exchange. Interestingly, previous efforts to co-culture complementary auxotrophs had limited effectiveness in supporting co-growth in liquid cultures, indicating the importance of spatial structure in facilitating cooperation and makes our system very attractive for study of such phenomena (Wintermute and Silver, 2010).

Furthermore, while the spatial microenvironment is not fully characterized, we have shown that the emergence of gradients, and simultaneously gene expression landscapes, are robust and reproducible features of the colony. Moreover, the landscapes can be compared to extract correlation patterns and infer how gene regulatory networks act in synchronicity to establish the microenvironment within the colony. This approach may provide a relatively simple, yet effective method of screening for ‘organismic’ properties that have been shaped by evolution and are only relevant in a multicellular context.

Our future efforts to extend the application of this setup will be dedicated to the study of how the microenvironment dynamically changes when external conditions are altered, an uncharted territory at the scale of a multicellular assembly that is central to the understanding of microbial ecosystem resistance to stress, environmental fluctuations and adaptation. We anticipate that similar approaches could be used to study aging, cooperation and competition, cell memory or evolutionary dynamics, as well as quantitative characterization of (synthetic) ecological systems and mixtures of cells relevant to ecology and chemical biology.

Materials and methods

Yeast strains

Request a detailed protocol

All experiments were performed using haploid S. cerevisiae strains derived from the S288C background - BY4741: MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0. See Supplementary file 1 for a detailed list of the yeast strains used in this study.

Microscopy

View detailed protocol

We used an inverted fluorescence microscope (IX81, Olympus) equipped with an EMCCD camera (Evolve 512, Photometrics) and X-Cite exacte fluorescence light source (Lumen Dynamics). Optical filters from Chroma Technology Corporation ET-EGFP (U-N49002; Ex 470/40 nm Di495 Em 525/50 nm) and ET-DsRed (U-N49005; Ex 545/30 nm Di570 Em620/60 nm) were used to observe GFP and RFP fluorescence. Cells were observed using Olympus 10× (Plan 10x/0.25 NA), 60× (PlanApo N 60x/1.42 NA Oil) and 100× (UPlanFL N 100x/1.3 NA Oil) objectives. Open-source µManager (Edelstein et al., 2014) microscopy software was used to control all of these components and setup multi-dimensional acquisition. The temperature inside the microscope incubation chamber that contained the media and cells was maintained at 30°C (Life Imaging Services). Fluorescence intensity was set to 10% of maximum output, fluorescence exposure was set to 1000 ms and camera gain was set at maximum. The time interval between each acquisition cycle was 6 min.

Microfluidics and cell loading

View detailed protocol

Microfluidic devices were constructed using soft lithography techniques. Photomasks were drawn using L-Edit software (Tanner) and printed on a high-resolution glass substrate (Delta Mask). A master wafer was created using SU-8 2000 (MicroChem) epoxy-based photoresist that was spin-coated to the appropriate thickness and exposed to UV light using an appropriate photomask to create the desired pattern. Multi-layered patterns were aligned and exposed to UV light using a MJB4 manual mask aligner (SUSS MicroTec) and the dimensions of the master wafer were checked using a Dektak 150 surface profiler (Veeco). The master wafer was treated with 95% (3-mercaptopropyl)-trimethoxysilane (Sigma) for 1 hr in the vapor phase. Microfluidic chips were created by casting a degassed 10:1 mix of polydimethylsiloxane (PDMS) and curing agent (Sylgard 184 kit; Dow Corning) on the master wafer, followed by at least 2 hr curing at 65°C. Each chip was gently cut and peeled off the master wafer; the entry/exit ports were punched out. The chip and a glass coverslip (24 × 50 mm #1; Menzel-Gläser) were treated with O2 plasma for 1 min in a plasma cleaner (Harrick Plasma), bonded together and incubated at 65°C for 10 min. Before loading cells, the chips were coated with 1% Pluronic F-127 (Sigma) for 30 min. Cells were precultured overnight in 5 mL of synthetic complete (SC) medium containing 2% w/vol glucose in a shaking incubator at 30°C, diluted 50-fold into 50 mL of SC +2% w/vol glucose, cultured for 5–6 hr in a shaking incubator at 30°C to an OD600 of 0.2–0.4, collected by centrifugation, and loaded into the microfluidic system with a pipette. The microfluidic system was centrifuged for 2 min at 1000 rpm using 3D-printed adaptors (Laurell WS-650 spin coater) to force the cells into the dead-end cell chambers. Liquid media was flowed rapidly through the flow channel to remove excess cells and the flow rate was set to 5 µL/min. A pressure-based microfluidic flow control system (MFCS; Fluigent) coupled with a flow rate platform (Fluigent) and a flow rate control module (Fluigent) that measured the flow rate and kept it constant by adjusting the pressure through a feedback loop was used to push liquid media through the flow channel. The output was kept at a constant pressure of 100 mbar above atmospheric pressure to minimize formation of air bubbles inside the flow channel.

Flow cytometry

Request a detailed protocol

Flow cytometry experiments were performed on a Gallios Flow Cytometer (Beckman Coulter) using a 488 nm excitation laser and 530/30 nm FL1 emission filter to detect GFP fluorescence. Data analysis was performed using Kaluza Flow Cytometry Analysis Software (Beckman Coulter). Approximately 104 cells were inoculated in 10 mL of SC medium containing various glucose concentrations (log2 dilutions from 8% to 0.0078125%, and 0% w/vol glucose) and cultured in a shaking incubator at 30°C to an OD600 of ~0.02–0.2 depending on the starting glucose concentration. Cells were then diluted 10-fold into 10 mL of fresh SC media containing the same starting glucose concentration and grown for 4–5 hr in a shaking incubator at 30°C, centrifuged at 4000 rpm for 10 min, re-suspended in 300 µL of PBS pH 7.4 buffer (Gibco) and fluorescence was measured using the flow cytometer. The supernatant of each sample was collected, and the glucose concentration was measured using the Glucose (HK) Assay Kit (Sigma) to confirm that the glucose concentration remained at similar levels during the growth phase (Figure 5—figure supplement 3). This is a modification of a previously published protocol (Youk and van Oudenaarden, 2009) for measuring expression of glucose transporters in different glucose concentrations modified to minimize glucose depletion in batch culture prior to expression measurements by working with sufficiently diluted batch populations. We thus expect that these measures represent expression levels that correspond to the starting glucose concentration, and would therefore be close to the measure within the 2D colonies (see discussion).

Image analysis

View detailed protocol

Image analysis was performed using open-source ImageJ 1.51 p software (Schneider et al., 2012). To obtain front velocity, we applied a threshold (Otsu) to detect the bottom frontier over time after flattening the background using a FFT band-pass filter. The image signal is decomposed by FFT into a spectrum of its constituent frequencies. Because some operations can be more easily performed on the spectrum than on the original image, the FFT bandpass algorithm filters out large structures (shading correction) and small structures (smoothing) of the specified size by gaussian filtering in Fourier space. The default parameters are set at 40 pixels for large structure and five pixels for small ones. To compute the local speed of the cells inside the cell assembly, we used the plugin TrackMate (Tinevez et al., 2017) v3.5.1 to track cell trajectories. TrackMate was set to DoG detector with estimated blob diameter of 4 µm and threshold of 4, while tacking was set to linear motion LAP.

Data availability

All data presented in the figures, article, and supplementary information are available on Zenodo (http://dx.doi.org/10.5281/zenodo.2557396).

The following data sets were generated

References

    1. Bryers JD
    (2008) Medical biofilms
    Biotechnology and Bioengineering 100:1–18.
    https://doi.org/10.1002/bit.21838

Decision letter

  1. Kevin J Verstrepen
    Reviewing Editor; VIB-KU Leuven Center for Microbiology, Belgium
  2. Naama Barkai
    Senior Editor; Weizmann Institute of Science, Israel
  3. Hans Steenackers
    Reviewer

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

[Editors’ note: a previous version of this study was rejected after peer review, but the authors submitted for reconsideration. The first decision letter after peer review is shown below.]

Thank you for submitting your work entitled "Emergence of metabolic landscapes in yeast monolayer colonies" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and a Senior Editor. The reviewers have opted to remain anonymous.

Our decision has been reached after consultation between the reviewers. Based on these discussions and the individual reviews below, we regret to inform you that your work will not be considered further for publication in eLife.

We appreciate how your study uses a microfluidics setup to study the behavior of S. cerevisiae in a glucose gradient. The combination of this setup with GFP tagged version of hexose transporters, a glucose sensitive transcription factor (MIG1), and other reporters of metabolic state (PDC1, SHD2) allows studying spatial patterning as a function of glucose availability. The results show that patterns of expression of hexose transporters with different affinities are spatially arranged in a manner consist with a gradient of glucose availability with highest amounts of glucose available near the "source" (i.e. periphery of the artificial colony), and lowest levels near the "sink" (i.e. center of the artificial colony). Perhaps more interestingly, interpretation of the data suggests a relatively sharp transition in the colony between zones of growth and slow/no-growth, which they ascribe to zones of fermentative and aerobic metabolism.

Overall, the reviewers agree that this is a particularly elegant study. As you can see in the individual reports (below), reviewer 1 is significantly more positive than reviewers 2 and 3. However, after discussing the individual reviews, all reviewers question the amount of novel biological insight that is presented in your paper.

As such, I am afraid that we are not able to offer publication in eLife (unless of course you believe that we missed the major message and its biological relevance). On the other hand, everyone also agrees that the experiments and results are beautiful, and that the setup may be usable for other studies. As such, we wonder whether it would be conceivable to re-write your paper as a "Tools and resources" paper? If so, you would need to emphasize the design and operation of the system, detail its use for a broad range of studies and compare it to other, existing systems such as the ones describes by Hornung et al., 2018, and Wilmoth et al., 2018. The Hornung et al. paper in particular seems to use a rather similar setup…

Reviewer #1:

The authors applied a microfluidic device to study how long-range metabolic interactions influence the internal structure of simple yeast cell monolayers asymmetrically exposed to an external nutrient source. They showed spatial variation in nutrient levels, cellular growth, and expression of several metabolic genes to self-emerge. Notably, the gene expression landscapes exhibited a high degree of spatial correlation over different levels of external nutrient level and were indicative of a clear spatial transition between fermentative and respirative growth.

I really enjoyed reading this work. Although the experimental set-up is simple and the conclusions intuitive and straightforward, the depth of the analyses and the clarity of the interpretations are impressive. The authors combined growth measurements, single cell tracking, fluorescence based expression analysis with FACS and microscopy and advanced data representation to study spatial relations between nutrients, growth and metabolism with an unprecedented level of detail. This work will therefore undoubtedly provide inspiration to many other researchers interested in spatial heterogeneity of microbial communities. I therefore recommend publication in eLife.

Reviewer #2:

Marinkovic et al. address single-cell response to glucose gradients within structured communities. They present an intriguing method to look at metabolic landscapes in a microbial community. However, I do believe the novelty of the results need to be clarified significantly better, and the conclusions drawn must be discussed more critically. Alternatively, the work may be more suitable for a methods focused publication, as using the gene expression of the transporters to gauge the extracellular concentration of glucose within a colony is an interesting tool.

Major concerns:

A) The glucose gradients emerge in the end artificially, due to the design of the microfluidic chamber and due to the fact the yeast's consume glucose, and not due to community action. The method of the author's pictures of how the cells respond to different glucose availability as a function of glucose concentration in different areas of the closed channels. But the effects are the same if one cultivates the cells in batch cultures, and tunes glucose availability in this way. One does hence not learn anything about yeast communities, only, that yeast in batch culture and in a microfluidic chamber respond similarly to glucose depletion. Hence, is there anything new in the study, other than nicely illustrating the well-established fact that yeast activates different glucose transporter at different glucose concentration?

B) Batch cultures are used to tune gene expression sensors. But, in batch culture, glucose is constantly and rapidly depleted, while gene expression, needs some time to adjust. At any given time, in batch culture, the glucose concentration in the media, and the gene expression programme, are hence asynchronous (i.e. gene expression lags behind the glucose concentration). Chemostat experiments would be needed, to have a glucose concentration that corresponds to the gene expression programme at a particular point in time.

Method:

Figure 1: I have not found the mention of the exact number of dead-end chambers within one microfluidic chip (I'm guessing 16 from the figures?). However, looking at the n in different experiments (e.g. Figure 1, n=12, n>5 and n=9) I am wondering in how many separate experiments these were done, and how much of the collected data was discarded and for which reasons?

Was it excluded that nutrient concentrations from the previous chamber affected the later ones? Especially in the stages when the chambers were already filling up/cells were being carried away?

Results and Conclusion:

Subsection “Cellular metabolic activity creates gene expression landscapes”: "Both patterns demonstrate the formation and maintenance of a glucose gradient that emerges from cellular metabolic activity." The pattern of the gene expression demonstrates that the cells adapt to the glucose gradient that emerges through supply of fresh media from only one side. "Formation and maintenance by cellular metabolic activity" in the sense that, yes, the cells close to the opening do import the glucose they require seems a weak point to make. In other words, there is evidence for the contrary – there seems no different action of cells in the community as in the batch culture, when it comes to the expression of the transporters.

Discussion paragraph four: What is the dependence/independence of the different gene expression looked at here? Can we talk about synchronicity between networks if the studied genes are known to dependent on the same external factor?

Figure 6: Landscapes of gene expression is not shown at all glucose concentrations for all genes, why?

Reviewer #3:

This manuscript details the results of a study using a microfluidic system to study growth and gene regulation in monolayers of yeast cells. This system is presented as a tractable model for studying pattern formation in microbial colonies, via spatial and temporal measurement of cell division rates and measurement of gene expression via fluorescently tagged reporters.

The authors use this experimental setup, in combination with GFP tagged version of hexose transporters, a glucose sensitive transcription factor (MIG1), and other reporters of metabolic state (PDC1, SHD2) to study spatial patterning as a function of glucose availability.

Unsurprisingly, they find that patterns of expression of hexose transporters with different affinities are spatially arranged in a manner consist with a gradient of glucose availability with highest amounts of glucose available near the "source" (i.e. periphery of the artificial colony), and lowest levels near the "sink" (i.e. center of the artificial colony). Perhaps more interestingly, their interpretation of their data suggest a relatively sharp transition in the colony between zones of growth and slow/no-growth, which they ascribe to zones of fermentive and aerobic metabolism.

While I appreciate the elegance of the microfluidic colony model that is presented, the key findings are rather modest -- i.e. there are gradients of nutrient availability in microbial colonies. This doesn't really have the impact I expect from eLife papers.

Below I detail a number of concerns/comments about the study, the analyses, and whether the authors have adequate related their work to the larger literature in this area.

Detailed comments:

1) Several other recent papers (Hornung et al., 2018,; Wilmoth et al., 2018) have used microfluidic setups to study microbial colonies. The Hornung et al. paper is the most similar to the current experimental design.

2) If goal was to measure the rate of glucose uptake and availability it seems like there may be other more direct approaches than inferring this from hexose transporter expression, such as the use of 2-NBDG, a non-metabolizable, fluorescent glucose analog (Roy et al. 2015, doi: 10.1371/journal.pone.0121985).

3) The following paper, while not at the fine spatial scale provided by the microfluidics, presents a more complete and compelling view of metabolic differentiation in yeast colonies: Maršíková et al., 2017.

4) The spatial analyses focus primarily on peaks of maximum expression, but the data is potentially much richer and more interesting if the authors consider not only the global maxima but more complex spatial patterns. For example, in Figure 4B MIG1 nuclear localization appears to be multimodal. What is going on here?

5) There is very little effort to contextualize this work in the larger body of studies of yeast colony structure, physiology, or regulation of glucose responsive TFs. Illustrative of this, there are only three citations total in the entire discussion. There is only a single reference to the regulation of hexose transporters in the entire manuscript. Specific examples:

- Subsection “Cellular metabolic activity creates gene expression landscapes”: " HXT1 is a low-affinity glucose transporter mainly expressed under high-glucose conditions, while HXT7 is a high-affinity glucose transporter expressed under low-glucose conditions only" Citations?

- Subsection “Gene expression landscapes depend on the glucose source concentration”: "As HXT1 is mainly expressed under high-glucose conditions (> 1% w/vol glucose) in batch culture…" Citation?

- Subsection “Gene expression landscapes of other genes and transcription factor activity confirm the inferred glucose gradients”: "MIG1 is a key transcription factor involved in glucose repression that localizes to the nucleus in the presence of glucose, to repress genes that participate in parallel carbon metabolic pathways (e.g., galactose)." Citation?

- and in the same subsection: "…we examined the expression of PDC1 and SDH2, which are overexpressed in fermenting and respiring cells, respectively" Citation?

[Editors’ note: what now follows is the decision letter after the authors submitted for further consideration.]

Thank you for resubmitting your work entitled "A microfluidic device for inferring metabolic landscapes in yeast monolayer colonies" for further consideration at eLife. Your revised article has been favorably evaluated by Naama Barkai (Senior Editor), a Reviewing Editor, and three reviewers.

We really appreciate the efforts of re-writing this manuscript as a "resource" paper. We feel that the paper has been much improved, but there are some remaining issues that need to be addressed through textual changes before acceptance, as outlined below:

1) We ask that the authors make it clear to the reader that what they call a "community action" does not imply that cells are responding differently in the community than they would as single cells in planktonic growth. The observed effects (gradients and response to these gradients) are merely the consequence of a high concentration of more or less static cells. So, perhaps calling it "population effect" would be more appropriate?

2) We would appreciate if you could insert a succinct discussion of the difference between a batch culture and this steady-state (where in principle, the cells could be better adapted as they are not always "running behind" in their response to a variable environment, as they would be in batch). (see reviewer 2's comments below for more details)

3) We again ask to include more references and to not overstate the advantage of the presented device (see reviewer 3's comment below).

4) Please provide more details about the reproducibility/variability and statistics of the results and analyses.

The above points summarize what we believe to be essential changes, and we only ask for textual changes instead of further experiments. We are including the full reviews for your reference below.

Reviewer #1:

In this revised manuscript, the authors – as requested- presented their work as a tool that can be generally applied to quantitatively and dynamically study growth landscapes, metabolic landscapes and gene expression landscapes within extended monolayers of cells. They extensively elaborated on the similarities and differences of their approach compared to other studies that applied agar colony models or microfluidics devices to study spatiotemporal patterns in colonies, and clearly pinpointed advantages of their approach. A figure was added to clarify the experimental set-up. Also my other (minor) comments were addressed to a sufficient extent. I therefore recommend publication.

Reviewer #2:

Although the authors have put efforts in their revision, they have not addressed my two main points. I'm hence wondering if I have not expressed myself clearly enough, so it's perhaps my own fault – or it also could be a different use of language between disciplines, so that the authors did simply not understand my questions. Long story short, I don't want to sound negative as it may be my own use of language, but the revision has not addressed the two simple concerns I had. Perhaps the other reviewers could comment if they agree or disagree with me, I'm happy if my comments are ignored, in case my concerns are not shared.

I think both points still are relevant, though. The first one could be fixed simply in writing. 'Community action' implies to me, and I guess to many others in the field, that there are interactions between the cells that emerge in the community – implying that cells would behave differently in the community as if they are not in the community. When it comes to glucose consumption, this seems not to be the case – there seems nothing which indicates that the individual cell's glucose import would be depending on interaction terms between the cells. The manuscript is written as if glucose gradients would emerge because of 'community action', but in fact, they emerge from cells that consume the glucose, in a way that does not require a different action as if cells would act independently to one another. In other words, the cells deplete the glucose in the MF channels, and within the channel, this creates a gradient, to which the cells respond. This alone does not indicate that cells act differently in the community. as they would act in Isolation.

The second point may remain relevant, however. The sensors are tuned in batch culture. But in batch culture, one expects a time differential, between the glucose concentration present at a point in time, and the average gene expression program activated at that same moment (as gene expression needs time to adjust, but glucose is constantly depleted. I.e. if gene expression needs 30mins to adjust to a changed glucose concentration, the gene expression program reflects the glucose concentration that was in the batch 30mins earlier before the timepoint was taken). I had suggested the authors to control this in chemostats, where gene expression and concentration are a time-wise aligned. I agree this would have been time-consuming to do and could hence be difficult for the authors to do. But it would have been great if they could have come up at least with some idea, and quantify how big this effect is; One would assume the differential between glucose concentration and gene expression are strongest in the early and late exponential phase. If chemostats are too difficult, one could suggest also much simpler surrogate experiment, for instance. One could, for instance, re-supplement exhausted media with glucose, perform a time-course, and estimate in this way the time needed until the gene expression has re-adjusted to the supplied glucose level. The sensor data would need to be adjusted by the time differential.

Reviewer #3:

This revised manuscript has been appropriately rewritten as a "Tools and Resource" article.

1) The authors overstate the uniqueness of their approach. For example, in the Introduction in reference to previous studies using similar microfluidics designs:

" While the use of microfluidics gave rise to the discovery of interesting collective properties of microbial assemblies, such attempts were too specific and had to deal with some of the limitations like small device dimensions (<100 100 μm), use of low nutrient concentrations (<1 mM), limited scope of nutrient types, inability to access single cell level – and therefore cannot be transposed to the general case of the study of a large monolayer of cells in standard range and scope of nutrients employed in biological research."

Few of these critiques hold up. For example, low nutrient concentrations and limited scope of nutrient types are simply specifics factors applied to the different organisms study. This critique applies equally to the system presented here. Furthermore, the Hornung et al. article certainly allows access to the single cell level. The text should be revised to reflect a more accurate summary of this earlier related work.

2) The rebuttal states "We agree that more contextualization was necessary so we expanded our Introduction and Discussion part, as well as provided additional citations, including all the ones proposed by the reviewer" I can find no references our discussion of prior literature at the line numbers provided.

3) One of the points the authors make several times regards the reproducibility of their assays. For example: "We demonstrate a novel capacity to reproduce and

quantify…" and "…we found the growth pattern was highly reproducible across parallel chambers.." However in the eLife "transparent reporting form" the authors state "We did not use statistical inference or comparison between data sets." How then did the authors estimate reproducibility/repeatability?

https://doi.org/10.7554/eLife.47951.026

Author response

[Editors’ note: the author responses to the first round of peer review follow.]

We thank the editors and reviewers for their careful review of our work. We concur that our work does not provide novel genetic network mechanistic insights but reveals and quantifies time- and length-scales at play in the emergence and the maintenance of gene expression and cellular growth patterns in a growing multicellular assembly. It also provides an elegant and straightforward method for quantitative spatiotemporal exploration of microbial colonies, as you outlined in your editorial decision. As such, and following your recommendations, we rewrote our article as a “Tools and Resources” submission, outlining the methodology and discussing how this could be used to gain novel insights in a broad range of multicellular systems. We changed the title, the abstract, rewrote part of the introduction and discussion sections and added one figure with methodological details of the microfluidic system and a few supplementary figures.

Reviewer #1:

The authors applied a microfluidic device to study how long-range metabolic interactions influence the internal structure of simple yeast cell monolayers asymmetrically exposed to an external nutrient source. They showed spatial variation in nutrient levels, cellular growth, and expression of several metabolic genes to self-emerge. Notably, the gene expression landscapes exhibited a high degree of spatial correlation over different levels of external nutrient level and were indicative of a clear spatial transition between fermentative and respirative growth.

I really enjoyed reading this work. Although the experimental set-up is simple and the conclusions intuitive and straightforward, the depth of the analyses and the clarity of the interpretations are impressive. The authors combined growth measurements, single cell tracking, fluorescence based expression analysis with FACS and microscopy and advanced data representation to study spatial relations between nutrients, growth and metabolism with an unprecedented level of detail. This work will therefore undoubtedly provide inspiration to many other researchers interested in spatial heterogeneity of microbial communities. I therefore recommend publication in eLife.

We do thank the reviewer for his positive assessment of our work. In light of the other reviewers comments, we recall here that our goal was indeed to get a quantitative analysis of cellular growth and gene expression landscapes in an extended monolayer of cells. We did not expect to gain novel mechanistic insights on, say, the glucose genetic regulatory network, which has been studied in details in the past. On the contrary, we used this knowledge (e.g., on HXTs) to extract the lengthscales on which gradients emerged as a function of the external glucose concentration. Providing with a quantitative, detailed, description of these landscapes is the main scientific contribution of our work. Methodologically, our contribution is probably broader, since we describe a tool for researchers interested in the emergence of metabolic gradients in multicellular systems (microbial colonies, biofilms, multicellular tissue, tumor, etc…). As suggested by the editors, we now addressed this in the discussion part of our paper.

Reviewer #2:

Marinkovic et al. address single-cell response to glucose gradients within structured communities. They present an intriguing method to look at metabolic landscapes in a microbial community. However, I do believe the novelty of the results need to be clarified significantly better, and the conclusions drawn must be discussed more critically. Alternatively, the work may be more suitable for a methods focused publication, as using the gene expression of the transporters to gauge the extracellular concentration of glucose within a colony is an interesting tool.

We thank the reviewer for his careful review. Our goal was both to provide with a simple method to study the emergence of gradients within multicellular systems (in contract with classic studies that tend to study isolated single cells) and to obtain a quantitative description of an extended monolayer of yeast cells. As such, our analysis does not bring novel mechanistic understanding of the biology of yeast, but rather focus on bringing quantitative data that could be used to model yeast colonies.

Major concerns:

A) The glucose gradients emerge in the end artificially, due to the design of the microfluidic chamber and due to the fact the yeast's consume glucose, and not due to community action. The method of the author's pictures of how the cells respond to different glucose availability as a function of glucose concentration in different areas of the closed channels. But the effects are the same if one cultivates the cells in batch cultures, and tunes glucose availability in this way. One does hence not learn anything about yeast communities, only, that yeast in batch culture and in a microfluidic chamber respond similarly to glucose depletion. Hence, is there anything new in the study, other than nicely illustrating the well-established fact that yeast activates different glucose transporter at different glucose concentration?

We disagree. The gradients do not appear artificially but because of the yeast cells. The design of the microfluidic chamber does not influence the gradients at all.If there were no cells, no gradients would be observed. We would argue that the gradients emerge precisely because there is community action. Cells at different positions consume nutrients at different rates and this is the main driving force of nutrient availability within a colony. Community structure (density, spatial structure, metabolic and genetic heterogeneity, etc.) will determine how, where and when different emergent properties arise. To put it differently, the spatial description of a yeast community cannot be learnt from single cell /batch analysis. Therefore, we do learn (quantitatively) something; the spatial pattern that emerges from single cell programs when cells are together, shaping a metabolic environment at a much larger scale than their own size.

Note that the gradient of glucose is not known inside the cell monolayer (and this is not mentioning the other nutrients) and cannot be inferred from existing models. Methods set aside, the scientific novelty of our study is that we now know quantitatively where and when different glucose transporters are expressed in space and time in a community context. It is a good fundament for further studies and models which could have hardly been extrapolated from batch culture glucose transporter expression data.

B) Batch cultures are used to tune gene expression sensors. But, in batch culture, glucose is constantly and rapidly depleted, while gene expression, needs some time to adjust. At any given time, in batch culture, the glucose concentration in the media, and the gene expression programme, are hence asynchronous (i.e. gene expression lags behind the glucose concentration). Chemostat experiments would be needed, to have a glucose concentration that corresponds to the gene expression programme at a particular point in time.

We are not sure to understand this comment. Our goal is NOT to study a particular gene in a chemostat condition. On the contrary, our method embraces the complexity of a dynamically growing cell monolayer and propose to observe and analyze its emergence and maintenance in terms of landscapes of growth and gene expression.

If this comment is directed to the way we quantified HXTs level of expression in batch culture, we do agree that the most elegant way to do this would be by using a chemostat, or even better, a microfluidic device that can trap single cells subjected to perfusion of media with a constant nutrient concentration. However, we made sure to measure glucose depletion in batch culture and adjusted our experiments accordingly (number of inoculated cells and transfer to a fresh medium) so that cells do not experience a significant drop in glucose concentration. We show in Figure 6—figure supplement 1 the concentration of glucose in supernatant right before flow cytometry measurements, and it was more than 90% of initial glucose concentration.

Method:

Figure 1: I have not found the mention of the exact number of dead-end chambers within one microfluidic chip (I'm guessing 16 from the figures?). However, looking at the n in different experiments (e.g. Figure 1, n=12, n>5 and n=9) I am wondering in how many separate experiments these were done, and how much of the collected data was discarded and for which reasons?

We used different microfluidic chips but most of the experiments presented here contain 5 – 10 chamber in parallel. The number of experiments is the sum of all “growing monolayers” for a given condition, observed in several, separated experiments (2-3). Sometimes it can happen that either no or not enough cells are loaded in one or two chambers and therefore they remain empty or take a long time to fill up compared to other well-loaded chambers. Or, it can happen that during the growth some chambers get clogged by cells and then we do not take these chambers into account for analysis. In the end, we took all growing monolayers into account and did not discard any data in the analysis we provide.

Was it excluded that nutrient concentrations from the previous chamber affected the later ones? Especially in the stages when the chambers were already filling up/cells were being carried away?

Yes. The dimensions of the feeding channels and the flow rate (5μL/min) ensures that the concentration seen by the cells in different chambers are independent. In line with this consideration, we did not observe any decays of the growth rate of cells or of the speed of the monolayer front, nor any drift in the gene expression landscapes from one chambers to the next.

Results and Conclusion:

Subsection “Cellular metabolic activity creates gene expression landscapes”: "Both patterns demonstrate the formation and maintenance of a glucose gradient that emerges from cellular metabolic activity." The pattern of the gene expression demonstrates that the cells adapt to the glucose gradient that emerges through supply of fresh media from only one side. "Formation and maintenance by cellular metabolic activity" in the sense that, yes, the cells close to the opening do import the glucose they require seems a weak point to make. In other words, there is evidence for the contrary – there seems no different action of cells in the community as in the batch culture, when it comes to the expression of the transporters.

We are not sure we understand the point made by the reviewer. We do not intend to show that there are “surprising” differences between what is known of yeast biology and what occurs in the monolayers. Our point, is to say that our (quantitative) knowledge of single cell (obtained e.g. in batch culture) is not enough to infer quantitatively the landscape of gene expression observed. On the contrary, we USE the landscape of gene expression to infer the glucose gradients.

Discussion paragraph four: What is the dependence/independence of the different gene expression looked at here? Can we talk about synchronicity between networks if the studied genes are known to dependent on the same external factor?

Again, this is a difference of perspective. Given that the global microenvironment is not known, it is not obvious that the different genes that we studied display a high level of coordination in their spatial patterns. For example, you could imagine that the ethanol, pH, metabolite by products, or more simply the varying dilution rate of cells due to their variable growth rate, may affect the gene expression levels. We just show that this is the case in an experimental context that is relevant for multicellular systems.

Figure 6: Landscapes of gene expression is not shown at all glucose concentrations for all genes, why?

We did not show the landscapes that were flat or non-informative. The point was to show that the landscapes were spatially correlated and adopt a structure with two main regions displaying different metabolic activity.

Reviewer #3:

This manuscript details the results of a study using a microfluidic system to study growth and gene regulation in monolayers of yeast cells. This system is presented as a tractable model for studying pattern formation in microbial colonies, via spatial and temporal measurement of cell division rates and measurement of gene expression via fluorescently tagged reporters.

The authors use this experimental setup, in combination with GFP tagged version of hexose transporters, a glucose sensitive transcription factor (MIG1), and other reporters of metabolic state (PDC1, SHD2) to study spatial patterning as a function of glucose availability.

Unsurprisingly, they find that patterns of expression of hexose transporters with different affinities are spatially arranged in a manner consist with a gradient of glucose availability with highest amounts of glucose available near the "source" (i.e. periphery of the artificial colony), and lowest levels near the "sink" (i.e. center of the artificial colony). Perhaps more interestingly, their interpretation of their data suggest a relatively sharp transition in the colony between zones of growth and slow/no-growth, which they ascribe to zones of fermentive and aerobic metabolism.

We understand that the reviewer found our analysis “unsurprising”, but it was not our intention to show a surprising effect (which could have been seen in a different context, for example by changing the frequency and concentration of nutrients supply or type, by perturbing gene networks or by controlling one or more additional nutrients, to name a few. Some of which are currently a subject of investigation in the lab using this microfluidic system). Here our goal was to propose a quantitative description of these landscapes. This is not only about observing two populations. It is about observing the continuum of populations and the emergence of a spatially stable structure which results from growth landscape, metabolic landscape and gene expression landscape. Methodologically, we propose a novel method that can be used to study these landscapes dynamically, which no other systems proposed so far.

While I appreciate the elegance of the microfluidic colony model that is presented, the key findings are rather modest -- i.e. there are gradients of nutrient availability in microbial colonies. This doesn't really have the impact I expect from eLife papers.

This is not the key finding of the article. It was expected that we observed gradients, but we did not know their spatial extent and their stability and maintenance. Consider, that it is not possible (to date) to propose a model that would estimate this gradient for any metabolite. This is due to our limited quantitative data that describe multicellular systems, and our limited knowledge about multicellular effects (potential synergistic or antagonistic effects, interactions with other nutrients, effects of secreted metabolites etc…). Here, we do propose quantitative measurements of these landscapes, providing in return real data to fine tune computational models. We also are able to use this to infer the shape of the glucose gradients and estimate the number of cells that can grow in such conditions. Additionally, we show that the gradients are very different depending on the amino acid concentration (Figure 2—figure supplement 1), indicating that the global microenvironment, which is also composed of complex gradients needs to be taken into account to properly model the dynamics of cellular assemblies. It is one thing to assume that there are gradients. It is not the same to propose a robust experiment that allows to quantitatively study them. This is even more important if we want to get closer to the context that occurs in natural, non-laboratory, environments where cells often find themselves in everchanging environments composed of many interacting cells of same or different type.

Below I detail a number of concerns/comments about the study, the analyses, and whether the authors have adequate related their work to the larger literature in this area.

Detailed comments:

1) Several other recent papers (Hornung et al., 2018,; Wilmoth et al., 2018) have used microfluidic setups to study microbial colonies. The Hornung et al. paper is the most similar to the current experimental design.

This article specifically deals with “small” chambers. There are other articles that try to model colonies, the originality of our system is that we are able to let the cells create the gradients themselves through metabolic activity over almost 1000 fold concentration range of nutrients spanning from very high concentrations that are commonly used in laboratory and that appear in nature to complete lack of nutrients. And additionally, we can track on global and single cell level the activity of cells in space and time. It is thus different and applicable to a wider range of contexts and questions. In the revised manuscript, we changed the Introduction / Discussion to mention other articles that started to address the problem of spatial organization of multicellular assemblies in microfluidics even there are only a few to our knowledge.

2) If goal was to measure the rate of glucose uptake and availability it seems like there may be other more direct approaches than inferring this from hexose transporter expression, such as the use of 2-NBDG, a non-metabolizable, fluorescent glucose analog (Roy et al. 2015, doi: 10.1371/journal.pone.0121985).

We did consider the use of 2-NBDG, however it is not as straightforward as proposed by the reviewer. As a non-metabolizable glucose analog, it is likely to be progressively absorbed by cells until they cannot absorb it anymore, leaving it to freely diffuse further in the colony. Here, glucose (and other metabolizable nutrients) show a decrease of concentration precisely because they are used and converted in biomass. Using an analog to infer the glucose concentration would likely not work.

3) The following paper, while not at the fine spatial scale provided by the microfluidics, presents a more complete and compelling view of metabolic differentiation in yeast colonies: Maršíková et al., 2017.

The interesting differentiation patterns shown in this paper and in other papers published by the same group have one shortcoming, which is that they are observed in colonies grown on glycerol or glycerol/ethanol supplemented agar plates as the major carbon source. As they are not fermentable, it is likely that the insight in this major component of yeast metabolism is limited. This method also has limited spatiotemporal scale as noted by the reviewer and inability to dynamically change environments or provide constant environments. The advantage is that it is possible to perform analysis like RNAseq, which is nicely demonstrated in this paper. In our experiments we use media that has precisely quantified components (yeast nitrogen base, complete supplement mixture of amino acids and additional carbon source at precise concentration) as opposed to media based on yeast extract (basically yeast growing on yeast, which nota bene, would be an interesting possibility to study the effects of cannibalization of dead yeast in our microfluidic colony system). Also, they do not systematically quantify the landscape of gene expression, but they showed evidence that there are different metabolic areas in a colony. We did mention this study in our revised article.

4) The spatial analyses focus primarily on peaks of maximum expression, but the data is potentially much richer and more interesting if the authors consider not only the global maxima but more complex spatial patterns. For example, in Figure 4B MIG1 nuclear localization appears to be multimodal. What is going on here?

We do agree and this is the main point that we make by quantifying the full landscape of expression of many genes in similar conditions that emerges from the metabolic activity of cells. With that regards, the conclusions may not be surprising but they take place in a physiological setting. Regarding MIG1, we do not observe a multimodal localization, but a sharp transition from a localized to a cytoplasmic state. This is used as an indicator of a shift in glucose concentration.

5) There is very little effort to contextualize this work in the larger body of studies of yeast colony structure, physiology, or regulation of glucose responsive TFs. Illustrative of this, there are only three citations total in the entire discussion. There is only a single reference to the regulation of hexose transporters in the entire manuscript. Specific examples:

We agree that more contextualization was necessary so we expanded our Introduction and Discussion part, as well as provided additional citations, including all the ones proposed by the reviewer.

- Subsection “Cellular metabolic activity creates gene expression landscapes”: " HXT1 is a low-affinity glucose transporter mainly expressed under high-glucose conditions, while HXT7 is a high-affinity glucose transporter expressed under low-glucose conditions only" Citations?

See above

- Subsection “Gene expression landscapes depend on the glucose source concentration”: "As HXT1 is mainly expressed under high-glucose conditions (> 1% w/vol glucose) in batch culture…" Citation?

See above

- Subsection “Gene expression landscapes of other genes and transcription factor activity confirm the inferred glucose gradients”: "MIG1 is a key transcription factor involved in glucose repression that localizes to the nucleus in the presence of glucose, to repress genes that participate in parallel carbon metabolic pathways (e.g., galactose)." Citation?

See above

- and in the same subsection: "…we examined the expression of PDC1 and SDH2, which are overexpressed in fermenting and respiring cells, respectively" Citation?

See above

[Editors' note: the author responses to the re-review follow.]

We really appreciate the efforts of re-writing this manuscript as a "resource" paper. We feel that the paper has been much improved, but there are some remaining issues that need to be addressed through textual changes before acceptance, as outlined below:

1) We ask that the authors make it clear to the reader that what they call a "community action" does not imply that cells are responding differently in the community than they would as single cells in planktonic growth. The observed effects (gradients and response to these gradients) are merely the consequence of a high concentration of more or less static cells. So, perhaps calling it "population effect" would be more appropriate?

We did not intend to imply a “community” effect (with the meaning proposed by reviewer 2) and have exchanged all relevant instances of “community” with the word “population” as you suggested. Notably, the cells themselves are not static, i.e. they continuously grow and therefore passively pushed to be washed away from the channel. However, a steady state gradient is achieved so that cells, transiently occupying a geographic location along the gradient axes, exhibit a steady state response, mirroring the established stable gradient.

2) We would appreciate if you could insert a succinct discussion of the difference between a batch culture and this steady-state (where in principle, the cells could be better adapted as they are not always "running behind" in their response to a variable environment, as they would be in batch). (see reviewer 2's comments below for more details)

i) In standard batch culture, as cells exhaust the media, their adaptation time, limited by sensing, transcription and translation (on the order of 30 minutes; response time referred to by reviewer #2), may lag behind the decrease of glucose in the media. In contrast, in our 2D colony device, once steady-state gradient is settled, cells’ residence time within a given range of glucose that corresponds to stable Hxt expression is significantly longer, assuring that even though cells are continuously pushed away, they spend sufficient time within a given concentration to equilibrate their response to the gradient and therefore faithfully report the glucose concentration in their environment. As example, at 1% glucose the mean velocity at Hxt7 expression peak, spanning over >200μm (Figure 3A, varies between 5-15μm/hour (Figure 2F), suggesting cells’ residence time of >10 hours, pointing to the advantage of working in this setup.

ii) In our batch culture experiments (modified from a previously published protocol reported by Youk and van Oudenaarden, 2009), we intently used very diluted cultures to assure that within the experiment span cells both adapt to the nutrient condition and do not consume sufficiently to change significantly the glucose concentration. Thus, we deem the chemostat experiment is not useful.

We addressed point (i) by inserting the paragraph above to the Discussion section, and point (ii) in a succinct manner in the Materials and methods section of our revised manuscript.

3) We again ask to include more references and to not overstate the advantage of the presented device (see reviewer 3's comment below).

We apologize to the reviewer for the confusion about the references that we actually added in the revised manuscript but not pointing to the correct line numbers. The lines were referring to the lines in the original manuscript that the reviewer pointed out. In the current revised manuscript they can be found in subsections “Cellular metabolic activity creates gene expression landscapes” and “Gene expression landscapes of other genes and transcription factor activity confirm the inferred glucose gradients”.

Indeed, our method compliments other approaches cited in the manuscript and expands researchers’ arsenal for quantitative observation of 2D colony growth. We addressed this in the revised manuscript, toned down the advantages of our device and discussed its limitations in view of other published devices more clearly (see Introduction section).

4) Please provide more details about the reproducibility/variability and statistics of the results and analyses.

Each result is represented with standard deviation as error bars as descriptors of how disperse measurements are. When we mentioned the reproducibility of our experiments, it is based on the fact that error bars are small compared to the range of variation of the mean fluorescence peak location when the control parameter (here the external glucose concentration) is varied. Given the significant in-between concentration differences as compared to the error bars we see no added value in computing further statistics such as p-values. All figure captions contain the number of samples and the description of the error bars displayed. In addition, the complete dataset used in this study is available on an online Zenodo archive.

https://doi.org/10.7554/eLife.47951.027

Article and author information

Author details

  1. Zoran S Marinkovic

    1. Laboratoire Matière et Systèmes Complexes, UMR 7057 CNRS and Université de Paris, Paris, France
    2. U1001 INSERM, Paris, France
    3. CRI, Université de Paris, Paris, France
    Contribution
    Data curation, Investigation, Writing—original draft, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-7720-2802

  2. Clément Vulin

    1. Laboratoire Matière et Systèmes Complexes, UMR 7057 CNRS and Université de Paris, Paris, France
    2. Institute of Biogeochemistry and Pollutant Dynamics, ETH Zürich, Zürich, Switzerland
    3. Department of Environmental Microbiology, Eawag, Dübendorf, Switzerland
    Contribution
    Investigation, Writing—original draft
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-3914-0708

  3. Mislav Acman

    1. Laboratoire Matière et Systèmes Complexes, UMR 7057 CNRS and Université de Paris, Paris, France
    2. CRI, Université de Paris, Paris, France
    Contribution
    Visualization, Methodology
    Competing interests
    No competing interests declared

  4. Xiaohu Song

    U1001 INSERM, Paris, France
    Contribution
    Software, Image analysis
    Competing interests
    No competing interests declared

  5. Jean-Marc Di Meglio

    Laboratoire Matière et Systèmes Complexes, UMR 7057 CNRS and Université de Paris, Paris, France
    Contribution
    Supervision, Writing—original draft, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-0446-4550

  6. Ariel B Lindner

    1. U1001 INSERM, Paris, France
    2. CRI, Université de Paris, Paris, France
    Contribution
    Supervision, Writing—original draft, Writing—review and editing
    For correspondence
    ariel.lindner@inserm.fr
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-5015-6511

  7. Pascal Hersen

    Laboratoire Matière et Systèmes Complexes, UMR 7057 CNRS and Université de Paris, Paris, France
    Contribution
    Conceptualization, Supervision, Funding acquisition, Writing—original draft, Writing—review and editing
    For correspondence
    pascal.hersen@univ-paris-diderot.fr
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-2379-4280

Funding

European Commission (724813)

  • Pascal Hersen

Agence Nationale de la Recherche (ICEBERG-ANR-10-BINF-06-01)

  • Pascal Hersen

Agence Nationale de la Recherche (ANR-16-CE12-0025-01)

  • Pascal Hersen

Agence Nationale de la Recherche (ANR-11-LABX-0071)

  • Pascal Hersen

Agence Nationale de la Recherche (ANR-11-IDEX-0005–01)

  • Pascal Hersen

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

The authors would like to thank Sébastien Léon (IJM, CNRS) and their respective team members for their critical reading of this manuscript. This work was supported by the Agence Nationale de la Recherche (ICEBERG-ANR-10-BINF-06–01; ANR-16-CE12-0025-01), the interdisciplinary program of the University Sorbonne Paris Cité, the Who am I? Laboratory of Excellence (ANR-11-LABX-0071 and ANR-11-IDEX-0005–01) and the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 724813).

Senior Editor

  1. Naama Barkai, Weizmann Institute of Science, Israel

Reviewing Editor

  1. Kevin J Verstrepen, VIB-KU Leuven Center for Microbiology, Belgium

Reviewer

  1. Hans Steenackers

Publication history

  1. Received: May 5, 2019
  2. Accepted: June 30, 2019
  3. Accepted Manuscript published: July 1, 2019 (version 1)
  4. Version of Record published: July 11, 2019 (version 2)

Copyright

© 2019, Marinkovic et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 2,790
    Page views
  • 394
    Downloads
  • 17
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Zoran S Marinkovic
  2. Clément Vulin
  3. Mislav Acman
  4. Xiaohu Song
  5. Jean-Marc Di Meglio
  6. Ariel B Lindner
  7. Pascal Hersen
(2019)
A microfluidic device for inferring metabolic landscapes in yeast monolayer colonies
eLife 8:e47951.
https://doi.org/10.7554/eLife.47951

Categories and tags

Research organism

Further reading

    1. Physics of Living Systems

    Vein fate determined by flow-based but time-delayed integration of network architecture

    Sophie Marbach, Noah Ziethen ... Karen Alim
    Research Article Updated

    Veins in vascular networks, such as in blood vasculature or leaf networks, continuously reorganize, grow or shrink, to minimize energy dissipation. Flow shear stress on vein walls has been set forth as the local driver for a vein’s continuous adaptation. Yet, shear feedback alone cannot account for the observed diversity of vein dynamics – a puzzle made harder by scarce spatiotemporal data. Here, we resolve network-wide vein dynamics and shear rate during spontaneous reorganization in the prototypical vascular networks of Physarum polycephalum. Our experiments reveal a plethora of vein dynamics (stable, growing, shrinking) where the role of shear is ambiguous. Quantitative analysis of our data reveals that (a) shear rate indeed feeds back on vein radius, yet, with a time delay of 1–3 min. Further, we reconcile the experimentally observed disparate vein fates by developing a model for vein adaptation within a network and accounting for the observed time delay. The model reveals that (b) vein fate is determined by parameters – local pressure or relative vein resistance – which integrate the entire network’s architecture, as they result from global conservation of fluid volume. Finally, we observe avalanches of network reorganization events that cause entire clusters of veins to vanish. Such avalanches are consistent with network architecture integrating parameters governing vein fate as vein connections continuously change. As the network architecture integrating parameters intrinsically arise from laminar fluid flow in veins, we expect our findings to play a role across flow-based vascular networks.

    1. Physics of Living Systems

    Vasculature: Exploring the dynamics of vascular adaptation

    Thomas S Shimizu, E Toby Kiers, Howard A Stone
    Insight

    A combination of in toto imaging and theory suggests a new mechanism for the remodeling of veins in vascular networks.

    1. Medicine
    2. Physics of Living Systems

    Comparable in vivo joint kinematics between self-reported stable and unstable knees after TKA can be explained by muscular adaptation strategies: A retrospective observational study

    Longfeng Rao, Nils Horn ... Pascal Schütz
    Research Article Updated

    Background:

    Postoperative knee instability is one of the major reasons accounting for unsatisfactory outcomes, as well as a major failure mechanism leading to total knee arthroplasty (TKA) revision. Nevertheless, subjective knee instability is not well defined clinically, plausibly because the relationships between instability and implant kinematics during functional activities of daily living remain unclear. Although muscles play a critical role in supporting the dynamic stability of the knee joint, the influence of joint instability on muscle synergy patterns is poorly understood. Therefore, this study aimed to understand the impact of self-reported joint instability on tibiofemoral kinematics and muscle synergy patterns after TKA during functional gait activities of daily living.

    Methods:

    Tibiofemoral kinematics and muscle synergy patterns were examined during level walking, downhill walking, and stair descent in eight self-reported unstable knees after TKA (3M:5F, 68.9 ± 8.3 years, body mass index [BMI] 26.1 ± 3.2 kg/m2, 31.9 ± 20.4 months postoperatively), and compared against 10 stable TKA knees (7M:3F, 62.6 ± 6.8 years, 33.9 ± 8.5 months postoperatively, BMI 29.4 ± 4.8 kg/m2). For each knee joint, clinical assessments of postoperative outcome were performed, while joint kinematics were evaluated using moving video-fluoroscopy, and muscle synergy patterns were recorded using electromyography.

    Results:

    Our results reveal that average condylar A-P translations, rotations, as well as their ranges of motion were comparable between stable and unstable groups. However, the unstable group exhibited more heterogeneous muscle synergy patterns and prolonged activation of knee flexors compared to the stable group. In addition, subjects who reported instability events during measurement showed distinct, subject-specific tibiofemoral kinematic patterns in the early/mid-swing phase of gait.

    Conclusions:

    Our findings suggest that accurate movement analysis is sensitive for detecting acute instability events, but might be less robust in identifying general joint instability. Conversely, muscle synergy patterns seem to be able to identify muscular adaptation associated with underlying chronic knee instability.

    Funding:

    This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.