Multiplexed microfluidic screening of bacterial chemotaxis

  1. Michael R Stehnach
  2. Richard J Henshaw
  3. Sheri A Floge
  4. Jeffrey S Guasto  Is a corresponding author
  1. Department of Mechanical Engineering, Tufts University, United States
  2. Department of Biology, Wake Forest University, United States

Abstract

Microorganism sensing of and responding to ambient chemical gradients regulates a myriad of microbial processes that are fundamental to ecosystem function and human health and disease. The development of efficient, high-throughput screening tools for microbial chemotaxis is essential to disentangling the roles of diverse chemical compounds and concentrations that control cell nutrient uptake, chemorepulsion from toxins, and microbial pathogenesis. Here, we present a novel microfluidic multiplexed chemotaxis device (MCD) which uses serial dilution to simultaneously perform six parallel bacterial chemotaxis assays that span five orders of magnitude in chemostimulant concentration on a single chip. We first validated the dilution and gradient generation performance of the MCD, and then compared the measured chemotactic response of an established bacterial chemotaxis system (Vibrio alginolyticus) to a standard microfluidic assay. Next, the MCD’s versatility was assessed by quantifying the chemotactic responses of different bacteria (Psuedoalteromonas haloplanktis, Escherichia coli) to different chemoattractants and chemorepellents. The MCD vastly accelerates the chemotactic screening process, which is critical to deciphering the complex sea of chemical stimuli underlying microbial responses.

Editor's evaluation

This manuscript presents a valuable new microfluidic tool that will allow researchers from different fields to rapidly quantify the chemotactic response of microbes to chemical gradients that have different strengths. Using planktonic bacteria, this paper convincingly shows that a multiplexed microfluidic device produces similar results to previously described microfluidic devices that generate only one gradient at a time. By performing on-chip dilutions, this device allows data for six different gradient strengths to be generated simultaneously, potentially reducing both experimental effort and biological variability.

https://doi.org/10.7554/eLife.85348.sa0

eLife digest

Many microorganisms such as bacteria swim to explore their fluid habitats, which range from the human digestive system to the oceans. They can detect minute traces of food, toxins and other chemicals in their environment, and – through a process called chemotaxis – respond by swimming towards or away from them. Chemical concentrations naturally decrease with distance away from their source, forming gradients. By sensing these chemical gradients, and adjusting their swimming direction accordingly, cells can locate nutrients and other resources in harsh environments as well as avoid toxins and potential predators.

Over the past 20 years, laboratory devices that manipulate minute volumes of fluid – known as microfluidics devices – have been indispensable for studying chemotaxis. They enable researchers to generate gradients of chemicals in carefully designed networks of microscopic channels, controlling the conditions that swimming cells are exposed to and mimicking their natural habitats. However, large-scale studies of chemotaxis have been limited by the sheer range of chemicals that are present at different levels in natural environments. Conventional microfluidic devices often compromise between distinguishing how individual cells behave, precise control over the chemical gradient, or the ability to execute multiple assays at the same time.

Here, Stehnach et al. designed a microfluidic device called the Multiplexed Chemotaxis Device. The device generates five streams of precise dilutions of a chemical and then uses these streams – alongside a control stream lacking the chemical – to measure chemotaxis in six different conditions at the same time. The device was tested using a well-studied bacterium, Vibrio alginolyticus, which is commonly found in marine environments. The device reliably examined the chemotaxis response of the population to various chemicals, was able to carry out multiple assays more rapidly than conventional devices, and can be easily applied to study the response of individual bacteria under the same conditions.

The Multiplexed Chemotaxis Device is relatively easy to manufacture using standard methods and therefore has the potential to be used for large-scale chemotaxis studies. In the future, it may be useful for screening new drugs to treat bacterial infections and to help identify food sources for communities of microbes living in marine environments.

Introduction

Motile cells of all types navigate complex environments through the detection of and response to chemical signals via chemotaxis (Adler, 1966; Berg and Brown, 1972; Stocker, 2012; Wadhams and Armitage, 2004). This fundamental survival mechanism regulates countless biological processes, such as microbial foraging in marine environments (Seymour et al., 2010; Stocker et al., 2008) and reproduction (Kaupp et al., 2008). Consequently, considerable effort has been invested into the study of microbial chemotaxis (Keegstra et al., 2022; Raina et al., 2019) to better understand their chemotactic motility (Lazova et al., 2011), detection sensitivity (Mao et al., 2003), and transport (Ford and Harvey, 2007). Microfluidic devices have become an indispensable platform for disentangling the intricacies of microbial chemotaxis by virtue of their precise control over the chemical environment at scales relevant to swimming cells (Keegstra et al., 2022; Ahmed et al., 2010). Specifically, microfluidics have been employed to physically model a range of chemical landscapes, such as nutrient patches (Stocker et al., 2008), and provide highly tunable concentration profiles (Li Jeon et al., 2002; Sugiura et al., 2010). Microfluidics have been broadly applied across microbial systems for both drug-dose response quantification (Sugiura et al., 2010) and infectious disease diagnostics (Welch et al., 2022). While microfluidic chemotaxis assays have evolved since their inception (Ahmed et al., 2010), the vast landscape of potential chemical compounds, combinations of compounds, and concentration gradient conditions that regulate these important processes necessitates the development of new high-throughput devices.

Faced with a broad range of chemostimulant concentrations and gradients in their natural environment (Stocker, 2012), microorganisms, specifically prokaryotes, have evolved exquisite chemosensing abilities with variable degrees of specificity to nutrients, dissolved resources, toxins, and signaling molecules (Adler, 1966; Berg and Brown, 1972; Stocker, 2012). Some bacteria exhibit a dynamic sensing range spanning five orders of magnitude (Lazova et al., 2011; Kalinin et al., 2009; Cremer et al., 2019) and can detect nano-molar attractant concentrations (Mao et al., 2003), while marine invertebrate spermatozoa have a reported detection limit approaching the femto-molar scale (Guerrero et al., 2010). Quantifying the strength of chemotactic responses across varying concentration and concentration gradient conditions presents a key challenge to understanding microbial driven processes, extending far beyond their search for optimal metabolic activity conditions (Keegstra et al., 2022). For example, in marine microbial communities, the natural phycosphere surrounding individual cells (Raina et al., 2022) contains a diverse spectrum and concentration of metabolite and organic material (Moran et al., 2022), which are taken up by chemotaxing microbes (Zimmerman et al., 2020). Viral infection of microbes augments this process and is a principal mechanism (Moran et al., 2022; Evans and Brussaard, 2012) for transforming live biomass to readily available organic matter. Lysis (Weinbauer et al., 2011) and exudation (Howard-Varona et al., 2022) by virus infected cells releases a diverse range and concentration of metabolite and organic material (Moran et al., 2022). Furthermore, chemotaxis is essential in initiating bacterial infections and pathogenicity for both animals and plants (Matilla and Krell, 2018). For example, in gastric infections pathogenic organisms rapidly colonize surfaces via chemotaxis, where a range of attractants from urea to amino acids and metals are presumed to enable localization and colonization on the host epithelium (Keilberg and Ottemann, 2016). Identifying the key metabolites and signaling chemicals which drive microbial chemotaxis necessitates new microfluidic tools capable of probing the wide scope and scale of chemotactic behaviors across a myriad of complex systems.

Microfluidic devices are widely accepted as an indispensable platform for targeted chemotaxis assays by enabling the quantification of both single cell and population-scale responses to precisely-controlled chemical gradients (Ahmed et al., 2010; Li Jeon et al., 2002). One class of chemotaxis microfluidic devices, termed stop-flow diffusion, relies on flowing a chemostimulant solution and buffer stream side-by-side in a microchannel. Upon halting the flow a slowly-evolving concentration gradient forms via diffusion (Figure 1a and b; Seymour et al., 2010; Stocker et al., 2008; Mao et al., 2003; Ahmed et al., 2010). Other devices generate steady chemical gradients by utilizing porous materials (Ahmed et al., 2010) or mimic diffusing marine hotspots using micro-well assays that entice and trap chemotactic microorganisms (Lambert et al., 2017; Raina et al., 2022). While these well-established assays accurately measure chemotactic motility in physically relevant concentration gradients (Stocker et al., 2008), they largely overlook the potential for high-throughput screening afforded by microfluidic devices. Recently, such high-throughput capabilities have been broadly showcased in other fields through the use of parallelized microfluidics for clinical testing of viruses (Welch et al., 2022), drug responses (Sugiura et al., 2010), and cell profiling (Prakadan et al., 2017). The development of an integrated microfluidic design - comprising parallelized chemotaxis assays on a single chip - would enable high-throughput characterization of microbial chemotactic responses. Relative to time-prohibitive conventional assays, rapid chemotaxis phenotyping could facilitate comparative studies and discoveries across different swimming microorganisms, chemostimulants, and concentration gradient conditions.

Figure 1 with 4 supplements see all
Multiplexed microfluidic device for simultaneous chemotaxis assays.

(a,b) Continuous flow through a microfluidic junction (a) stratifies chemostimulus, cell, and buffer solutions, demonstrated here with fluorescein, DI water, and DI water, respectively. Upon halting the flow (b) diffusion establishes a chemical gradient across the channel, which is repeated at each observation channel in the MCD (d, red and orange boxes). Scale bars, 0.1 mm. (c) Assembly of the MCD showing the PDMS dilution layer (blue) and cell injection layer (red) microchannels mounted on a glass slide (grey; Materials and methods). (d) Scaled drawing of the dilution layer, which receives chemical (pressure, pin,1) and buffer (pin,2) solutions. Initial chemical concentration (C0) is sequentially diluted 10-fold to each of four additional concentrations (C1-4), plus a control solution (C5=0). These six chemostimulus solutions are merged separately with additional cell (Ai) and buffer (Bi) solutions from the cell injection layer (e) for chemotaxis assays in respective observation channels (dashed black box, corresponding to c and f). (e) Scaled drawing of the cell injection layer which injects a cell suspension (pin,3) and buffer solution (pin,4=pin,3) into the dilution layer (Ai,Bi; Materials and methods). Scale bars d,e, 2 mm. (f) Photograph of the completed MCD with dyed water to visualize the chemical (yellow), cell (red), and buffer (blue) fluid streams in the channel network. Scale bar, 5 mm. (g) Stratified chemical (C0), cell, and buffer solutions in the first observation region (d, red box). (h) Dilution of the chemical (C0) by the buffer prior to mixing in the first micromixer (Stroock et al., 2002) to produce concentration C1 (d, purple box). (i), Stratified chemical solution after initial dilution (C1, green) in the second observation region (d, orange box). Scale bars g-i, 0.2 mm. (j) Measured chemical concentrations (see Materials and methods) generated from the dilution microchannels (d) for various driving pressures pin,1=pin,2=[100,150,200]mbar (square, circle, and triangle, respectively). (k) Measured evolution of the chemical gradient (b) produced in the C0 observation region (Figure 1—figure supplement 3; Materials and methods) by the MCD shows the chemical diffusion across the channel with increasing time t.

Here, we present a microfluidic multiplexed chemotaxis device (MCD) that enables high-throughput chemotaxis screening of swimming microorganisms to chemical stimuli across concentration gradient conditions that potentially span the microorganism’s entire sensitivity range. The two-layer device architecture comprises a serial dilution layer that produces logarithmically diluted chemostimulant solutions (Sugiura et al., 2010) and a cell injection layer that introduces swimming cells, whilst minimizing both the footprint and operational complexity of the device (Figure 1c–e). On a single chip, the MCD simultaneously performs six stop-flow diffusion chemotaxis assays (including control), which span five orders of magnitude in chemostimulant concentration. The dilution, mixing, gradient generation, and flow performance are fully characterized (Materials and methods), and the MCD is validated against a conventional chemotaxis device for a known marine bacterial chemotaxis system (Vibrio alginolyticus). To demonstrate the device’s efficiency, capabilities, and operational flexibility, the MCD is then used to rapidly quantify the chemotactic responses of different microbes to a variety of chemostimulants. Compared to existing microfluidic devices, the MCD enables chemotaxis studies with significantly higher throughput rates, and most importantly facilitates the simultaneous measurement of chemotactic responses across a range of concentration gradient conditions.

Results

Multiplexed microfluidic device as a platform for high throughput chemotaxis screening

To enable rapid and efficient chemotaxis screening of swimming microbes, we designed the multiplexed chemotaxis device (MCD) to perform six chemotaxis assays in parallel on a single microfluidic chip (Figure 1). The individual assays in the observation region are based on laminar flow patterning and established stop-flow diffusion methods (Ahmed et al., 2010; Kirby, 2010; Stehnach et al., 2021; Stocker et al., 2008; Nguyen et al., 2019), where rapid, parallel flow of chemostimulus (concentration, Ci) and buffer (C=0) solutions maintain initially stratified fluid regions. Upon stopping the flow, a chemostimulus gradient forms via diffusion (Figure 1a and b). For each chemotaxis assay (Figures 2 and 3), a swimming cell solution is injected between the chemostimulus and buffer so their response may be observed and recorded. The MCD (Figure 1c–i) performs six simultaneous assays comprising five logarithmically decreasing chemical concentrations (Ci=10-iC0,i[0,4]; Figure 1j), plus one control (C5=0). The device is fabricated from polydimethylsiloxane (PDMS) in two layers (Materials and methods; Figure 1c–e). The primary function of the dilution layer (Figure 1d) is to receive two fluid inputs - base chemostimulus solution (C0) and buffer (C=0) - and passively generate six pre-defined concentration conditions (Ci) via a serial dilution process (Li Jeon et al., 2002; Sugiura et al., 2010; Walker et al., 2007), which are dispensed to each of the six observation channels for chemotaxis assays. At each stage of the serial dilution process, the chemostimulus stream is combined with buffer in a 1:9 volumetric flow rate ratio, where efficient mixing of the solutions is necessary for accurate dilution and chemotaxis assays downstream. To ensure sufficiently mixed solutions, herringbone micromixers (Stroock et al., 2002; Figure 1d and h and Figure 1—figure supplement 1) were incorporated into each dilution stage. These structured microchannel surfaces generate a three-dimensional flow to induce chaotic mixing, and thus significantly reduce the mixing length (Stroock et al., 2002) and the overall footprint of the device (Materials and methods; Figure 1—figure supplement 1). To achieve the targeted chemostimulus concentrations in the dilution layer, the microfluidic network was designed using hydraulic circuit theory (Materials and methods; Figure 1—figure supplement 2), and the accuracy of the serial dilution process was experimentally confirmed (Figure 1j).

Figure 2 with 1 supplement see all
Validation of MCD and measurement of V. alginolyticus chemotactic performance toward serine.

(a) TEM image of V. alginolyticus (Materials and methods). Scale bar, 1 μm. (b) A single chemotaxis assay (SA) with a single conventional microfluidic device flows chemostimulus (top, blue), cell suspension (middle), and buffer (bottom) streams into the observation region (Materials and methods). (c) SA with chemostimulus (serine, C=100μM) showing measured cell positions (V. alginolyticus, black dots) at various times t after initial flow stratification (t=0min) relative to the chemostimulus distribution (blue, from measurements in Figure 1b). Cells migrate up the gradient (t=1min) followed by uniform dispersal as the gradient dissipates (t=10min). Degree of cell accumulation is determined from the number of cells, Np,n, in a 200 μm wide region on the chemostimulus side (positive; green dashed line) and buffer side (negative; red dashed line), respectively (Seymour et al., 2010; Stocker et al., 2008). (d) The measured cell distribution across the microchannel evolves over time (from c) and is represented as a conditional probability density of cell position, P(y|t) (shown as a kymograph). (e) P(y|t) for V. alginolyticus chemotactic response to serine from a series of SA devices having the same geometry as the MCD observation regions (Figure 1). SA measurements illustrate the transition from positive chemotactic response at high attractant concentration (C0-2) to no response at low concentration (C3-4) compared to control (C5=0μM) (Altindal et al., 2011). (f) Accumulation index, β(t), for SA measurements from e. (g) P(y|t) measured by the MCD under the same conditions of the SA. (h) β(t) measured from g accurately captures the behavior of V. alginolyticus to serine compared to SA results (f). (i) Sørensen similarity metric (Cha, 2007) comparing e and g, which is calculated at each time point and averaged. (j) Comparing MCD and SA peak chemotactic response quantified by max (β(t)) from f,h. (k,l) β(t) in the absence of a chemical gradient (k; Figure 2—figure supplement 1) and for fixed gradients of Ci=200μM (l; Figure 2—figure supplement 1b) across each observation channel in the MCD indicates no significant bias. No gradient (k) conditions (Ci=0) were obtained by injecting buffer into the chemical inlet (setting C0=0). Fixed gradient (l) conditions (Ci=200μM of serine) were obtained by injecting C0=200μM of serine into both the chemical and buffer inlets of the dilution layer. Shading in f,h,k,l indicates one standard deviation (N=3). Error bars in i,j are one standard deviation across biological replicates.

Figure 3 with 1 supplement see all
MCD enables rapid quantification of chemotactic responses across different chemostimulants and microbial species.

(a,b) The negative chemotactic response of V. alginolyticus to the repellent phenol (Homma et al., 1996) is evident in kymographs of cell position, P(y|t), and the accumulation index, β(t), respectively. Central 250 μm wide band, which often contains a significant subpopulation of non-motile cells, is omitted from P(y|t) for visualization purposes and has no impact on β(t). (c,d) MCD measurements demonstrate the positive chemotactic response of V. alginolyticus to leucine observed in P(y|t) and β(t), respectively. (e–h) P(y|t) in response to various concentrations of serine for bacteria P. haloplanktis (e,f) and E. coli (g,h). TEM images of P. haloplanktis (f, inset) and E. coli (h, inset). Scale bars, 1 μm. β(t) for P. haloplanktis (f) illustrates a monotonically increasing response to increased concentrations (extracted from e). Accumulation index, β(t), for E. coli (h) reveals a peak response to an intermediate serine concentration (C2=3.6μM) and delayed accumulation at higher concentrations (Bhattacharjee et al., 2021). Shaded regions are standard deviation (N=2 and N=3 for P. haloplanktis and E. coli, respectively) across biological replicates. Color bar corresponds to kymographs in a,c,e,g.

The cell injection layer (Figure 1e) introduces a suspension of swimming microbes and a sheathing buffer solution from two corresponding inlets into the six observation regions of the dilution layer (Figure 1d, dashed box). In the observation regions, the chemostimulus solution, cell suspension, and buffer streams comprise six standard stop-flow chemotaxis assays (Figure 1a), where each one incorporates a unique chemostimulus concentration. The observation channel width (W=1mm) and height (H=90μm) are similar to other microfluidic devices (Stocker et al., 2008; Ahmed et al., 2010; Stehnach et al., 2021) and ensure organisms with different sizes can be studied using the MCD. The initially steady flow rates stratify the three fluids in the observation region and localize the cells in a thin band in the channel center with equal width chemostimulus and buffer streams on either side (Materials and methods; Figure 1g and i and Figure 1—figure supplement 3). Upon halting the flow, a unique and highly reproducible chemical gradient is formed in each observation channel via diffusion, where the consistency of the transient concentration profiles across all observation channels were confirmed using fluorescence microscopy (Materials and methods; Figure 1k, Figure 1—figure supplement 3). The mixing effectiveness, fluid stratification, and gradient formation were validated over a range of applied pressures (approximately 100–200 mbar) and were found to be consistent across all six observation regions (Figure 1—figure supplement 1 and Figure 1—figure supplement 3). This efficient two-layer architecture reduces the operational complexity of the device by decreasing the total number of fluid inlets (four) and reduces the footprint of the microfluidic chip.

Both layers of the MCD are fabricated from polydimethylsiloxane (PDMS) through standard soft lithography techniques (replica molding; Materials and methods). The final microfluidic chip is assembled by plasma bonding the dilution layer to a standard double-wide microscope slide (75 mm × 50 mm×1  mm) and subsequently aligning and bonding the cell injection layer on top (Materials and methods; Figure 1c and f and Figure 1—figure supplement 4). This reusable device (Materials and methods) is driven by a single pressure pump which maintains flow stratification (1–2 min) prior to each assay to ensure consistent initial gradient conditions for measuring cell responses. Pump and microscope automation enables the chemostimulus gradient to be reset for rapid replicate measurements. The design and operation of the MCD can accommodate most single-celled microorganisms, and efficient micromixer channels facilitate the use of a wide range of dissolved chemostimulants (Figure 1—figure supplement 1; Stroock et al., 2002). Due to variations in the replica mold fabrication process, a one-time tuning of the applied pressure ratio (pin,1-2/pin,3-4) for the dilution and cell injection layers is required to ensure symmetric flow in the observation channel (Materials and methods). Further optimization of the device layout could enable the number of observation channels to be expanded, including a broader range or more refined sampling of concentration gradient conditions. Additionally, the serial dilution layer design can be easily modified to produce different concentration scalings (e.g. logarithmic, linear; Sugiura et al., 2010; Walker et al., 2007). The high degree of parallelization for chemotaxis screening, combined with the demonstrated consistency and repeatability of the chemostimulus gradients, represents a significant advance relative to existing microfluidic chemotaxis devices (Stocker et al., 2008; Ahmed et al., 2010; Li Jeon et al., 2002).

Validation of MCD performance against conventional chemotaxis assays

Having established the gradient generation performance, the MCD is compared to a conventional chemotaxis assay to (i) validate the chemotaxis measurements against a known microorganism-chemostimulus system and (ii) demonstrate the high-throughput capability of the MCD in comparison with existing devices. The chemotactic response of the monotrichous marine bacterium Vibrio alginolyticus (Figure 2a) was measured using both the MCD and a single microfluidic gradient generation device (referred to as ‘single assay’, SA) (Stehnach et al., 2021). V. alginolyticus swims with a run-reverse-flick motility pattern, and it was chosen due to its well-documented chemotaxis (Altindal et al., 2011; Son et al., 2013; Xie et al., 2011) and prevalence within marine microbial communities (Baker-Austin et al., 2018). The SA device has an identical observation channel geometry with three inlets and operates using stop-flow diffusion in the same manner as the MCD, where the chemostimulus, cells, and buffer are initially flow-stratified (Figure 2b). Upon halting the flow for both assays, the concentration gradient develops via diffusion, and time-lapse microscopy is used to measure the evolution of the spatial cell distribution over time, t (Figure 2c; Materials and methods). For example, in a gradient formed by the model chemoattractant serine (Altindal et al., 2011), cells initially confined to a central band migrate toward the attractant, before uniformly sampling the channel (Figure 2c and d) as the gradient dissipates within t10min.

From one parallelized assay, the MCD precisely reproduces the chemotactic responses of V. alginolyticus toward various serine concentrations, compared to multiple, conventional single assays (Figure 2e–h). Chemotactic behavior of the bacteria was measured across a range of manually adjusted attractant concentrations for the single assay, decreasing from C0=200μM and matching the serial dilution concentrations generated by the MCD (Figure 1j). For each concentration gradient, the spatial distribution (conditional probability) of cells, P(y|t), across the observation channel, y, is shown over time as a kymograph (Figure 2e and g). For ease of comparison, the aggregate chemotactic response is distilled through the accumulation (or chemotactic) index, β(t), which quantifies the relative fraction of cells responding to the chemostimulus (Seymour et al., 2010; Stocker et al., 2008). The accumulation index is defined as (Seymour et al., 2010; Stocker et al., 2008): β(t)=(Np(t)-Nn(t))/(Np(T)+Nn(T)), where Np,n(t) are the instantaneous number of cells at time t within predefined positive and negative accumulation regions (Figure 2c, green and red dashed lines, respectively), and normalized by the total number of cells in these two regions at the final time (T10min). The measured cell distributions from the MCD (Figure 2g) were compared to the single assay results (Figure 2e) by calculating the Sørensen similarity metric (Cha, 2007) for each attractant concentration, showing excellent statistical agreement between the two assays (Figure 2i). A comparison of the strength of cell accumulation (Figure 2j) further emphasizes the high degree of concordance between the two devices. In particular, V. alginolyticus exhibits strong accumulation for high serine concentrations (C0-2), above the previously reported chemotactic sensitivity threshold of 0.2 μM (Altindal et al., 2011). Consequently, no discernable response is observed for lower concentrations (C3-4), which are comparable to β(t) for the control (C=0). At the highest concentrations (C0-1) in both devices, the chemotactic motility exhibits a slight reversal at later times. This is most likely due to the relatively high concentration of V. alginolyticus, which rapidly consumes the available serine at the top of the channel. Combined with chemoattractant diffusion across the channel width, the concentration gradient flips direction at later times causing accumulation on the bottom side of the channel. The persistent central band of cells at later times is due to a sub-population of bacteria which remain non-motile over the course of each assay and are thus localised near y=0 (Figure 2e and g; Figure 3—figure supplement 1). Because any non-motile or motile bacteria in this region do not impact the calculation of β(t), the central band is omitted from future kymographs for visualization purposes. Finally, additional assays with no chemostimulus and with a fixed chemostimulus concentration confirmed the consistency of the chemotaxis assays across all observation regions (Figure 2k and l and Figure 2—figure supplement 1), which is expected from the verified gradient generation performance in each channel (Figure 1—figure supplement 3).

These validation assays serve to highlight the dramatically improved efficiency in chemical screening when compared to a standard single assay. In the single assay case, each chemical concentration requires: (i) manually diluting stock solutions, (ii) exchanging peripheral reservoirs for chemicals, and (iii) a new cell suspension for each concentration assay, all of which become extremely costly and time prohibitive, when considering the scope and scale of multi-chemical, -concentration, and -organism panel experiments. The single assay results (Figure 2e and f) required six different dilutions, cell solutions, and devices, and with three replicates per condition, required 18 individual assays. In contrast, the MCD collected the same data (Figure 2g and h) in only three automated assays, and did not require culture changes eliminating inconsistencies due to variations in growth media, dilution errors, or growth conditions. If replacing the cell suspension is necessary, the MCD can be easily reset by simply exchanging the cell suspension with a fresh suspension and restarting the flow. Because it uses a robust serial dilution process and requires the preparation of a single chemostimulus (C0), the MCD ensures consistent chemical concentrations across different experiments, a crucial factor when working with microorganisms having femto- to nanomolar chemotactic sensitivities (Mao et al., 2003; Guerrero et al., 2010; Altindal et al., 2011). The MCD fully screens both microbe and stimulus pairings with three replicates in ≈ 1 hour with a single cell culture, including the bench time associated with preparing the solutions (Figure 2h). In contrast, the panel of single assays (Figure 2f) requires nearly a full working day (6-7hours). Thus, the MCD is a powerful and much needed tool for large scale chemotactic panel studies, where consistency in stimulus concentration, elimination of biological variability, and the need for efficiency, are essential.

Multiplexed microfluidic device supports high-throughput chemotaxis screening for novel stimuli and various microorganisms

Beyond validation with the single assay device, we demonstrate the efficacy of the MCD by examining the response of V. alginolyticus to both a known repellent and novel chemostimulus. Chemorepellents serve as an early warning sign for microorganisms to evade predators and toxins for survival (Yang et al., 2015). A single MCD assay reveals that V. alginolyticus exhibits negative chemotaxis (β<0) to the model repellent phenol (Homma et al., 1996) with an observed detection threshold on the order of C=1-10μM (Figure 3a and b). Separately, the amino acid leucine has been identified as an important metabolite in human health (Yang et al., 2020) and marine environments (Ferrer-González et al., 2021; Johnson et al., 2020), and it serves as a measure of prokaryote heterotrophic activity from viral lysis in deep-sea environments (Winter et al., 2018). Previously reported as an attractant for marine prokaryotes (Barbara and Mitchell, 2003), we verify the positive chemotactic response of V. alginolyticus to leucine through rapid chemotaxis screening using the MCD (Figure 3c and d).

To further illustrate the capabilities and flexibility of the MCD, the chemotactic behavior of Pseudoalteromonas haloplanktis (Figure 3f, inset) and Escherichia coli (Figure 3h, inset) to serine was measured (Figure 3e and g) with no design changes to the MCD (see Materials and methods). P. haloplanktis is a rapid-swimming, monotrichous marine bacterium that is a model organism for chemotaxis to amino acids (Barbara and Mitchell, 2003) and exhibits strong chemotaxis towards cellular exudates (Seymour et al., 2009). Similar to V. alginolyticus, it utilizes a run-reverse-flick foraging strategy (Son et al., 2016) for efficient chemotaxis in patchy chemical landscapes (Stocker et al., 2008). P. haloplanktis exhibited a monotonically increasing chemotactic response (Figure 3f) with increasing serine concentration, qualitatively comparable to the response of V. alginolyticus to leucine (Figure 3d). In contrast to the marine prokaryotes, E. coli is a pathogenic bacterium, which uses the bundling and unbundling of its multiple flagella to perform run-tumble motility for migrating up or down chemical gradients (Adler, 1966; Berg and Brown, 1972). E. coli has served as the canonical organism for bacterial motility and chemotaxis (Adler, 1966; Berg and Brown, 1972; Cremer et al., 2019; Mattingly et al., 2021) and has been instrumental in our understanding of logarthimic-sensing (Lazova et al., 2011; Kalinin et al., 2009) and chemical navigation in complex physical environments (Bhattacharjee et al., 2021). A single MCD assay reveals that E. coli has a strong chemotactic response to intermediate serine concentrations (C2; Figure 3g and h). The response significantly diminishes for high (C0-1; Figure 3g and h) and low chemostimulant concentrations (C3-4; Figure 3g and h). This observation reflects E. coli’s affinity for serine (Yang et al., 2015), but toxicity at high concentrations (Neumann et al., 2014). Furthermore, at higher concentrations (C1), the initial accumulation is delayed in time (Figure 3h), a feature that is also consistent with the chemotactic sensitivity of E. coli to serine (Son et al., 2016; Bhattacharjee et al., 2021). Taken together, these three model swimming chemotactic microbes cover diverse foraging and motility strategies, whose range of chemotactic responses were efficiently screened using the MCD. These results demonstrate that the MCD can rapidly ascertain chemotactic responses across different chemostimulants and concentration ranges, which will be particularly valuable for studying the nano-molar and even femto-molar concentrations that characterize the detection thresholds of many microorganisms (Mao et al., 2003; Guerrero et al., 2010; Altindal et al., 2011).

The ability to simultaneously quantify a microbe’s response for a spectrum of attractant concentrations using the MCD now enables rapid comparative studies across microbial or chemical species (Figure 4a). The magnitude of a microbe’s response to a given concentration gradient is compactly summarized by the maximum (or minimum for negative chemotaxis) of their accumulation index, βmax=±max(|β(t)|), where the sign of βmax is determined by the sign of the chemotaxis. In the case of serine, V. alginolyticus and E. coli (Figure 4a) both appear to have developed a chemotactic affinity to an optimal concentration (Lazova et al., 2011; Bhattacharjee et al., 2021), which is evident from the local maxima of βmax and is intrinsically linked to their motility and sensing abilities. This analysis is complemented by examining the maximal response rate max(dβ/dt) (measured prior to βmax; Figure 4b). The peak accumulation of V. alginolyticus to serine occurred at C1=49μM while the fastest rate of response occurred at a lower chemostimulus concentration (C2=3.6μM). Despite its similar morphology and motility to V. alginolyticus, P. haloplanktis (Figure 4a) exhibits a monotonically increasing response to serine across the concentrations tested, with the fastest response rate also occurring at the highest concentration tested (Figure 4b; C=200μM). Separately, the lack of an optimal concentration in the response of V. alginolyticus to leucine likely indicates a higher saturation threshold in the relevant chemoreceptors, relative to its serine response. This demonstrated screening efficiency highlights the benefits of this new microfluidic platform for tackling large-scale chemotaxis studies in a complementary manner to existing tools (Lambert et al., 2017; Raina et al., 2022; Haring et al., 2020; Satti et al., 2020), for a diverse array of micro-organisms and compounds.

Summary of chemotactic responses across various species, chemostimulants, and concentrations measured using the MCD.

(a) The primary metric for quantifying the chemotactic response of the bacteria was the peak of the accumulation index βmax=±max(|β(t)|), where the sign is determined by the positive or negative chemotactic behavior for each chemostimulus concentration (from Figures 2h, 3b, d, f and h). (b) The maximal response rate max(dβ/dt) (prior to βmax) is indicative of the speed of cell accumulation. These metrics distinguish chemotactic behaviors, for example: The accumulation of V. alginolyticus to serine is greatest at high concentrations (a; C=49μM), but the fastest response occurs at weaker concentrations (a; C=3.6μM). Error bars are one standard deviation across biological replicates.

Discussion

Here, we have introduced a novel microfluidic multiplexed chemotaxis device for rapid quantification of bacterial responses to a range of chemostimulant concentrations. Identifying the diverse chemical compounds and concentrations responsible for driving microbial interactions that underpin important environmental and human health processes – for example, ecosystem scale nutrient cycling and disease transmission – has proven to be a tedious and monumental undertaking. A primary challenge is the sheer diversity of potential compounds and the extensive concentration range of microbial responses from micro- to femtomolar (Mao et al., 2003; Guerrero et al., 2010; Altindal et al., 2011). Existing chemotaxis assays, whilst able to screen multiple compounds, are ill-equipped to probe the temporal response of motile cells and are limited to low time-resolution (e.g. hours; Lambert et al., 2017). The multiplexed chemotaxis device (MCD) design presented here mitigates these restrictions by rapidly screening the chemotactic behavior of microbes across a spectrum of chemostimulus conditions simultaneously (Figure 1f) on significantly faster timescales. This work opens new avenues to large-scale, laboratory-based panel experiments previously inaccessible with existing microfluidic devices.

The MCD’s two-layer device architecture uses a single pressure source to serially dilute a chemostimulus with a single buffer input, producing five logarithmically separated chemical solutions (Figure 1j). Separately, a cell suspension and additional buffer stream are introduced into each of the six observation regions, where the chemotactic response of the cells is recorded. The MCD performs reliably for a wide range of applied pressures and only requires calibration once, provided that the PDMS microchannels are cast from the same molds (Materials and methods). The simultaneous chemotaxis assays are not only fast and efficient compared to conventional (single assay) devices, but they also eliminate variability due to multiple culture preparations and potential changes in cell cultures that may occur over time (Figure 2g). Taken together, our results illustrate that the MCD enables robust and efficient quantification of the chemotactic responses of various bacterial species to different chemostimulants, simplifying the labor-intensive chemotaxis screening process (Figure 4).

The MCD design is amenable to a range of modifications to tailor its application, including but not limited to, alteration of chemical concentrations, gradients and flows, and increased multiplexing. For example, the design could be altered to enable the retrieval of chemotaxing cells from the individual observation regions for additional downstream analysis (Mao et al., 2003; Takagi et al., 2005; Gurung et al., 2020; Konishi et al., 2020). The sensitivity of prokaryotes is intrinsically linked to the strength of the concentration gradient (Lazova et al., 2011; Mao et al., 2003; Kalinin et al., 2009; Mattingly et al., 2021), an effect which can be examined with the MCD. With no alterations to the design or operation, simply changing the concentration of the input chemostimulus (C0, Figure 1d) will shift the examined concentration range. Likewise, the background concentration across all observation channels is adjustable by exchanging the buffer solution input in the cell injection layer for a chemostimulus solution. The serial dilution layer hydraulic circuit design can be modified to produce specific dilution ratios (i.e. linear or logarithmic; Sugiura et al., 2010; Walker et al., 2007), where optimization of the device layout can expand the number of different chemical conditions probed. The device operation is demonstrated here using fast swimming bacteria. However, the MCD can be operated using steady flow (Ahmed et al., 2010; Li Jeon et al., 2002) to study chemotaxis in slow-moving, surface attached microorganisms in systems such as neutrophil chemotaxis (Li Jeon et al., 2002) and biofilm formation (Boyeldieu et al., 2020). While the work presented here focuses solely on prokaryotes, the current device geometry will accommodate larger eukaryotic cells (10μm) and in principle could be scaled up for larger multicellular microorganisms (Varennes et al., 2017).

Whilst the MCD does offer substantial improvement and throughput compared to existing alternatives, it is not without limitations. Typical microfluidic devices consist of a single layer and can be fabricated in most cleanrooms, but the multi-layer dilution layer (Figure 1—figure supplement 4) does require the precision of a mask aligner (Materials and methods). If such facilities are not available, microchannel molds can be fabricated using other techniques such as 3D printing (Su et al., 2023) and micromilling (Guckenberger et al., 2015). A two-layer PDMS approach was chosen here to reduce the overall footprint of the device, but care must be taken to align the two layers, as misalignment could render the device inoperable, for example due to misconnecting ports between layers. In the current mode of operation, each observation channel is only imaged once every approximately 8  s, which is usually insufficient to identify swimming speed changes or changes in motility patterns associated with chemotaxis. This issue could be mitigated by capturing a short video at each visit to the observation channels, but it will also decrease the overall sampling frequency across concentration conditions.

In summary, the MCD provides a robust, high-throughput tool for expediting microbial chemotaxis screening. This engineered microfluidic device could simplify the study of microbial chemotaxis, which is paramount to understanding and modeling diverse problems including global scale carbon and nutrient cycling (Moran et al., 2022). Such technology could also be applied to accelerate microfluidic approaches to human health studies (Song et al., 2018; Atmaramani et al., 2019; Harimoto et al., 2022; Chen et al., 2020). Microfluidics have been identified as a potential means for meeting the high-throughput demands of chemical synthesis, screening, and testing with living cells, applications that remain key issues in drug discovery (Dittrich and Manz, 2006) and in meeting the challenge of antibiotic-resistant microbes (Qin et al., 2021).

Materials and methods

Microfluidic device design

Hydraulic circuit framework

Request a detailed protocol

In analogy with electrical circuits, well-established hydraulic circuit theory (Oh et al., 2012) was used to design the complex microfluidic network of the multiplexed chemotaxis device (MCD; see Figure 1 and Figure 1—figure supplement 2). Briefly, for incompressible, laminar flow through a constant cross-section microchannel, the pressure drop, Δp, is linearly proportional to the volumetric flow rate, Q, and is given by the Hagen-Poiseuille law (Oh et al., 2012; Kirby, 2010): Δp=QR. The hydraulic resistance, R, is a function of the fluid viscosity (properties of water assumed for all fluids) and the channel geometry. Fabrication of microfluidic devices via the soft lithography method (McDonald et al., 2000) used here (see below) results in rectangular cross section microchannels (height, H; width, W; length, L). Exact expressions for R are tabulated for rectangular and other cross-section channels and provided in various resources (Pozrikidis, 2001). Combined with conservation of mass, ΣQi=0 at the junctions (nodes) between several channels, i, the Hagen-Poiseuille law enables us to design complex microfluidic networks (Figure 1—figure supplement 2) via the solution of a set of linear equations.

MCD design considerations

Request a detailed protocol

The primary goal of the MCD was to efficiently perform several stop-flow bacterial chemotaxis assays (Ahmed et al., 2010; Ford et al., 1991) simultaneously for a range of chemostimulus concentrations. The design requirements were to: (i) dilute and distribute five logarithmically spaced concentrations of chemostimulus plus one control buffer solution to each of six chemotaxis assays. (ii) Perform those six chemotaxis assays in parallel on the same microfluidic chip. And, (iii) the microfluidic device should receive minimal fluid inputs to reduce setup time. The MCD has a two-layer architecture (dilution layer and cell injection layer) with a total of four fluid inputs and one (waste) output (Figure 1 and Figure 1—figure supplement 2). Each having two inputs, the dilution layer and cell injection layer are designed to be regulated by a pressure-driven flow controller operating at a pressure, pin,1-2100mbar and pin,3-450mbar, respectively, while the lone output is at atmospheric pressure (pout=0). The dilution layer receives a base concentration chemostimulus solution (concentration, C0; pin,1) and a buffer solution (C=0;pin,2), and the cell injection layer receives a bacterial suspension in buffer (pin,3) and a second buffer solution (C=0;pin,4). The serial dilution (Sugiura et al., 2010) process sequentially combines the chemostimulus and buffer fluids to produce separate microchannel streams having chemostimulus concentrations of Ci=10-iC0 (i[0,4]) and C5=0 (control). The resulting six diluted chemostimulus solutions are merged in separate observation channels with a flow-stratified bacterial suspension and second buffer stream, which eventually forms the chemostimulus gradient in the chemotaxis assay. The three solutions are designed to symmetrically occupy the following fractional widths of the observation channel (total width, W): w1=4W/9,w2=W/9, and w3=4W/9. The observation channel width (W=1mm) and height (H=90μm) were chosen to set the chemostimulus gradient strength based on a physically relevant length scale (Ahmed et al., 2010; Stehnach et al., 2021; Mattingly et al., 2021) and to ensure that the upper and lower microchannel walls do not impede bacterial motility, respectively. These dimensions are in line with conventional microfluidic chemotaxis assays (Ahmed et al., 2010), and as a consequence of the microfabrication process, the chosen H sets the height of the dilution layer channels excluding the micromixer. The initial flow rate of the three streams in the observation region (Figure 1d) was designed to be 20,5,20μlmin-1 for the chemostimulus (Qout), cell (QCIL,1), and buffer solution (QCIL,2), respectively (Figure 1—figure supplement 2). These flow rates maintain the stratification of the cell suspension with a 4:1:4 ratio, while ensuring that the bacterial cells are not damaged by the flow (Figure 1—figure supplement 3). Beyond these design requirements, the geometries - and thus hydraulic resistances – of several components are set independently, including: micromixer channels RM, bridge channels RB, and observation channels R4,4 (Figure 1—figure supplement 2 and Appendix 1). Based on these design requirements, hydraulic circuit theory was used to determine the required resistances of each microchannel in the MCD network, and subsequently the microchannel geometries (Oh et al., 2012). A complete list of the microchannel resistances and dimensions for the final design is provided in Appendix 1.

Herringbone micromixer design

Mixing performance

Request a detailed protocol

For the serial dilution process to perform as designed, effective mixing of the chemical solution and buffer are critical. Here, we use a well-established herringbone micromixer geometry (Stroock et al., 2002), where a series of ridges on the upper wall of an otherwise rectangular microchannel (Figure 1—figure supplement 1a and b) drive a three-dimensional flow to enhance mixing (Stroock et al., 2002; Ottino and Wiggins, 2004). A separate microchannel - having the same cross section geometry as the MCD design - was fabricated to independently quantify mixing performance and to select the necessary mixer length. The test micromixer channel was 41.3  mm long with 29 mixing cycles (comprised of two sets of six alternating herringbone ridges each). Two aqueous solutions of fluorescein salt (Sigma; concentrations, C50=50μM and C10=10μM) (Petrášek and Schwille, 2008) were injected individually into the MCD, and calibration images of dye intensity were captured after each herringbone mixer cycle (Figure 1—figure supplement 1), corresponding to the maximum (I50) and minimum (I10) dye concentrations, respectively. The region within 20 μm of the microchannel walls was excluded from analysis due to reflection and refraction effects (Stroock et al., 2002). Subsequently, the two solutions were flowed side-by-side with images (Ii) recorded in the same locations as above and normalized as follows:

I=IiI10I50I10,

where denote spatial averaging. The degree of mixing is defined as (Stroock et al., 2002), DOM=(II)2, where values of 0.5 and 0 indicate fully non-mixed and mixed solutions, respectively. This measurement (Figure 1—figure supplement 1c) was repeated for both the designed flow rate for the MCD (QM=22μlmin-1) and for a second higher flow rate (222μlmin-1). Based on standard metrics (Stroock et al., 2002), the two solutions are considered mixed when DOM0.05 (i.e. 90% complete mixing). For both flow rates, this criterion is met after 9 complete herringbone ridge cycles, and a final design with 26 herringbone cycles was chosen for the MCD. The independence of mixing efficiency on flow rate, combined with a safety factor of approximately three for the number of herringbone cycles, ensures that the serial dilution portion of the MCD will perform accurately for a wide range of chemostimulants and flow speeds.

Micromixer hydraulic resistance

Request a detailed protocol

To complete the design of the MCD, it was necessary to determine the hydraulic resistance of the herringbone micromixer RM which was measured empirically using a parallel microfluidic device (Choi et al., 2010). Briefly, a microfluidic device was fabricated with two parallel channels connected by shared inlets and outlets. The parallel channels had identical rectangular geometries except one had the herringbone ridges replicating the micromixer channel section RM (Figure 1—figure supplement 4). Two solutions, DI water and 1μlml-1 tracer particle suspension (0.25 μm radius; 2% solid; carboxylated FluoroSpheres, Life Technologies), were flowed through the device using glass syringes (2.5  ml; Hamilton) mounted on two separate syringe pumps (Harvard Apparatus). The particle solution was visualized using fluorescence microscopy, and the flow rates of the two pumps were adjusted such that the two streams divided equally into the parallel channels. The micromixer hydraulic resistance was determined from the resulting flow rate ratio and the known (analytical) resistance of the non-mixer channel (Oh et al., 2012; RM=0.0043mPasμm-3; Figure 1—figure supplements 2 and 4c), and the results were corroborated by COMSOL Multiphysics simulations (not shown; Stehnach, 2022).

Microfabrication and assembly

Request a detailed protocol

Microfluidic channel molds were fabricated using standard single and two-layer photolithography (Anderson et al., 2000) to transfer the final channel designs from a photomask (Artnet Pro, formally CAD/Art Services, Inc) onto a silicon wafer (100  mm diameter; University Wafer), which was spin-coated with photoresist (SU-8; Kayaku Advanced Materials). The single assay chemotaxis devices and MCD cell injection layer were made using SU-8 2050 and 2025, respectively, and multilayer devices (micromixer validation channels, MCD dilution layer) were made using SU8-3050 and SU8-2025 for the main rectangular channels and herringbone ridges, respectively. The ridges of the micromixers (Stroock et al., 2002) were applied by halting the first-layer photolithography after the first post-exposure bake (PEB), spin-coating a second layer of SU-8 photoresist onto the wafer, then completing the remainder of the photolithography processes as usual (Anderson et al., 2000). The ridges of the herringbone micromixer (Stroock et al., 2002) extend over the main channel by 10μm on both sides to account for potential misalignment during the multilayer photolithography (Figure 1—figure supplements 14). The final channel heights for the fabricated MCD dilution layer (Figure 1d and Figure 1—figure supplement 4a–c) were 90-94.5μm, 37-38.5μm for the main channel and herringbone ridges, respectively, while the cell injection layer (Figure 1e) was 71-73μm high (Bruker’s DekTak).

The MCD was fabricated using two-layer soft lithography (McDonald et al., 2000) with polydimethylsiloxane (PDMS; Sylgard 184) at a 10:1 (elastomer:curing agent) ratio. All channel wells were punched using a 1.5  mm diameter biopsy punch (Integra). The dilution layer mold was first silanized through vapor deposition (Sidorova et al., 2009) in a vacuum desiccator with 1–2 drops of tridecafluoro-1,1,2,2-tetrahydrooctyl trichlorosilane (Gelest Inc) to help release the cast PDMS. Post-silanization, PDMS was poured onto both the cell injection layer and dilution layer molds and degassed in a vacuum chamber (1hour) prior to curing (65°C for 1hour). The resulting PDMS dilution layer channel was first plasma bonded onto a standard thickness, double wide glass slide (75 mm × 50 mm×1  mm; Fisherbrand) using a plasma oven (Plasma Etch Inc), and subsequently heated on a hot plate at 110°C for one hour to promote covalent bonding (McDonald et al., 2000; Figure 1—figure supplement 4d). Next, the cell injection layer was plasma bonded on top of the dilution layer, with care taken to ensure the alignment of the fluid wells connecting the two layers (Figure 1—figure supplement 4e). The assembled device was baked again on a hot plate at 110°c for 1 hr. The PDMS-PDMS bond was found to be sufficiently strong for the relatively low pressure applications of the MCD (Eddings et al., 2008). Before injecting any fluids into the MCD, the microchannels should be inspected to insure no debris is blocking a channel. Debris (e.g. dust introduced during the fabrication process) that clogs or partially clogs the microchannels could negatively impact performance by changing the hydraulic resistances of the individual channel (Figure 1—figure supplement 2 and Appendix 1). Particulates can potentially be removed by flushing the device, but as with many microfluidic applications, large obstructions may render the device unusable. All other devices (e.g. single assay chemotaxis devices and micromixer validation channels) were fabricated using single-layer soft lithography, where an individual PDMS device was molded and subsequently bonded to a standard microscope slide using the procedures described above.

MCD dilution, flow, and gradient generation performance

The performance of the fabricated MCD was validated using epifluorescence microscopy (Nikon Ti-E) with an aqueous fluorescent dye (fluorescein sodium salt, Sigma) in various concentrations (described below). Images of the dye distribution were captured at the midplane of the channels with a sCMOS camera (Zyla 5.5; Andor Technology). Fluorescein was chosen due to its similar diffusion coefficient with the chemostimulant serine (Altindal et al., 2011). Minor deviations in the performance of the MCD from the original circuit design (Figure 1—figure supplement 1) are likely due to variations in the fabricated channel mold heights (Appendix 1). Such variations impact the hydraulic resistances (Oh et al., 2012) and symmetry of the cell solution (Figure 1—figure supplement 3).

Serial dilution

Request a detailed protocol

The primary function of the MCD dilution layer is to sequentially dilute the input chemical solution (C0) with buffer to generate four logarithmically decreasing concentrations (C1-4) for each of the observation channels (plus one control, C5=0). The dilution performance was quantified by injecting a solution with known fluorescein concentration (C0=1mM). The diluted concentration field, Ci(x,y), at each of the observation channels was then determined from the local measured dye intensity, Ii(x,y), which are linearly proportional, Ci(x,y)Ii(x,y). Fluorescence images were recorded upstream of the inlet before the three fluids made contact in each observation channel (Figure 1a) using a 20× (0.45 NA) objective. Due to the strong 10-fold dilution, pairs of images were acquired for adjacent channels with optimized exposure times to account for the finite dynamic range of the camera. The mean normalized concentrations provided for each observation region from the serial dilution process were reconstituted from the measured image intensity as follows:

CiC0=n=1iIn(x,y)In1(x,y),n[1,4],

where the angled brackets indicate the spatial average. The resulting serial dilution followed the expected logarithmic (10-fold) dilution Ci/C0=10-i for i[0,4] for which the system was designed (Figure 1j). This measurement was performed for three different sets of applied driving pressures, which yielded nearly identical results and illustrated the robustness of the serial dilution process.

Stratification symmetry

Request a detailed protocol

The symmetry of the stratified chemostimulus and buffer distributions in the observation channel is critical to prevent bias in the chemotaxis measurements. As minor errors in the manufacturing process can alter this symmetry, the applied pressure for the cell injection layer (pin,3,4; Figure 1—figure supplement 2) was tuned until the widths of the chemical, cell, and buffer streams in each observation channel were 4:1:4 ratio, respectively. Tuning was visualized by flowing a fluorescein solution (C0=100μM) in both the chemical and buffer inlets of the dilution layer as well as the buffer inlet of the cell injection layer (Figure 1d and e). The ratio of applied pressures (pin,1-2/pin,3-4) between the dilution and cell injection layer remained the same for all chemotaxis assays (pin,1-2/pin,3-4=10/7), which was slightly lower than the designed value (pin,1-2/pin,3-4=2). Tuning is only required for the first device fabricated from a particular set of molds, after which the calibration and tuning applies to all subsequent devices fabricated from the same mold set due to the robust nature of soft lithography. If significant variations in the stratification symmetry occur in the observation regions, the MCD will not function properly, where possible causes include: (i) mis-alignment of the two PDMS layers and/or (ii) debris blocking or impeding the flow. In the latter case, if inspection of each microchannel and flushing (see: Experiment replicates and device reusability) is ineffective, a new device is recommended.

Chemostimulus gradient consistency

Request a detailed protocol

Beyond ensuring the symmetry of the chemostimulus and buffer stratification, the time evolution of the resultant chemostimulus gradient must be consistent across each of the observation channels to accurately compare bacterial chemotactic responses. The chemical gradient evolution (Figure 1k and Figure 1—figure supplement 3c and d) was measured by first flowing a fluorescein solution (C0=100μM) through both the chemical and buffer inlets of the dilution layer and DI water through both inlets of the cell injection layer. Having independently verified the performance of the serial dilution process, this approach produces identical base concentrations for all observation channels, Ci=C0, and thus, enables easy comparison of the resulting concentration gradients in each channel. Upon halting the flow, an image was recorded (10×, 0.3 NA objective) in each observation channel every 5  s for approximately 9  min. The time evolution of the (normalized) spatial fluorescence intensity was measured to visualize the chemical gradient. The resulting concentration profiles were found to be highly consistent across the various observation regions and for different driving pressures (Figure 1—figure supplement 3c and d).

Cell culturing

Request a detailed protocol

Vibrio alginolyticus (YM4; wild-type) from -80°C stock were grown overnight in Marine 2216 media (Difco) by incubating at 30°C and shaking at 600 revolutions per minute (RPM). The overnight culture was diluted 100-fold into fresh pre-warmed 2216 media and grown for three hours (30°C, shaking at 600 RPM) to O.D.0.2. 7  ml of culture was then washed and resuspended (1,500 RCF for 5 min) in 4  ml of artificial seawater (ASW).

Psuedoalteromonas haloplanktis (ATCC 700530) from -80°C stock were grown overnight in Marine 2216 (Difco) media by incubating at room temperature and shaking at 100 RPM (Stocker et al., 2008).

Escherichia coli (MG1655) from -80°C LB stock were grown overnight in Terrific Broth (TB, Sigma Aldrich) by incubating at 34°C and shaking at 220 RPM (Stocker et al., 2008). The overnight culture was diluted 100-fold into fresh pre-warmed TB media, and grown for approximately three hours (34°C, shaking at 220 RPM) to O.D.0.5. 8  ml of culture was then washed three times and resuspended (4000 RCF for 5 min) in 4  ml of motility buffer (10  mM potassium phosphate, 0.1  mM EDTA, 10  mM NaCl, pH 7, filter sterilized 0.2μm). A total of 16  ml of culture was washed twice and resuspended (1200 RCF for 5 min) in 6  ml of ASW.

Media and chemostimulants

Request a detailed protocol

Artificial seawater (ASW) was prepared following the NCMA ESAW Medium recipe, which was adapted from Harrison et al., 1980 and modified (Berges et al., 2001). ASW was used as the buffer and the chemical solvent for chemotaxis assays for both V. alginolyticus and P. haloplanktis, while motility buffer (see above) was used for chemotaxis assays with E. coli. Chemostimulus materials were purchased from Sigma Aldrich for use in the chemotaxis experiments, specifically: serine (S4500), phenol (P1037), and leucine (L7875).

Microfluidic chemotaxis assays

Request a detailed protocol

Prior to use, the MCD was pre-treated by flowing a 0.5% (w/v) bovine serum albumen solution (BSA; Sigma Aldrich) to reduce cell adhesion to the microchannel surfaces. The device was flushed for over 10  min prior to first use with the cell, chemostimulus, and buffer suspensions. For chemotaxis assays, fluid flow was driven by a single pressure controller (Elveflow OB1; 1 mbar =100  Pa): pin,1-2=200mbar (dilution layer) and pin,3-4=140mbar (cell injection layer). Pressures were scaled down to 100 mbar and 70 mbar, respectively, for P. haloplanktis experiments. Between each chemotaxis assay, the fluid inputs were flowed for a minimum of 2  min to stratify cell, chemostimulus, and buffer streams in the observation channels. Upon halting the flow, a monotonic concentration profile was established in each observation channel due to the diffusion of the chemostimulus (Figure 1 and Figure 1—figure supplement 3). The spatio-temporal evolution of the bacterial distribution was determined by imaging the cells using phase-contrast microscopy (10×, 0.3 NA objective; Nikon Ti-E) with a sCMOS camera (Zyla 5.5, Andor Technology) for approximately 10 min. An automated computer-controlled stage was used to cyclically move the microscope field of view to each observation channel 75 times, producing an effective imaging period of 8  s for each observation channel. Each experiment was technically replicated at least three times with the same culture and repeated on different days with freshly grown cells. For validation of the MCD, a conventional single assay (SA) microfluidic device (Figure 2b–f) was designed with a similar geometry to the individual MCD observation channels. Specifically, the SA devices had three inlets (width, 0.5  mm) which merged in a single observation channel. The devices were fabricated in a single layer using soft lithography (see above), and they were pre-treated with a 0.5% (w/v) BSA solution and flushed with ASW prior to experiments. The SA chemotaxis devices were used to validate the MCD for the well-established chemotactic behavior of V. alginolyticus to the chemoattractant serine (Sigma). The three inlets of the single assay device (Figure 2b) carried the chemoattractant dissolved in ASW, V. alginolyticus suspended in ASW, and ASW, respectively. The three solutions were flow stratified for a minimum of 2  min using a syringe pump (Harvard Apparatus), whereby flow rates were controlled by syringe size to maintain a 4:1:4 ratio of the stream widths. In an identical manner to the MCD, a chemostimulus gradient develops in the channel via diffusion, and the chemotactic response of the cell population was observed over time. Imaging was performed with phase-contrast microscopy (4×, 0.13 NA objective; Nikon Ti-E) at 1 fps over the course of approximately 10  min using a CMOS camera (Blackfly S, Teledyne FLIR). Sample size in each replicate chemotaxis experiment ranged from 7000 to 16,000 measured cell positions, dependent on the strain used for each particular assay.

Experiment replicates and device reusability

Request a detailed protocol

After completing the initial 10  min filming period corresponding to the first measurement of a given organism/chemostimulus pairing, technical replicates were achieved in both the MCD and SA devices by restarting the flow. Stratification was maintained for at least 2  min to ensure consistent initial conditions among replicates. Next, the flow was halted and recording of bacteria positions commenced. This process was repeated for all subsequent technical replicates. A new MCD was fabricated for each unique organism/chemostimulus combination tested to prevent any cross-contamination between assays. Between biological replicates for a given organism/chemostimulus pairing, the MCD was cleaned by first flowing ethanol and then deionized water through all of the inlets. The device was then dried by flowing clean compressed air through the device and placing it under vacuum. Finally, the MCD was pre-treated again with a 0.5% (w/v) BSA solution and flushed with ASW prior to additional experiments. With appropriate cleaning, the same MCD can safely be used for multiple biological replicates of the same organism/chemostimulus combination. It is recommended to use a new device outside biological replicates of a particular organism/chemostimulus combination, consistent with the majority of PDMS applications (Toepke and Beebe, 2006). For SA experiments, a new device was used for each biological replicate in Figure 2e, although SA devices can also be easily cleaned and reused in the same manner as the MCD.

TEM imaging

Request a detailed protocol

For each species, initial cultures were grown following the previously described protocols (without any initial washing/resuspending), before the following final cell suspensions were prepared: (i) 4  ml of V. alginolyticus culture washed and resuspended (1,500 relative centrifugal force (RCF) for 5 min) in 1  ml of fresh 2216 media, (ii) 1  ml of P. haloplanktis culture washed and resuspended (1200 RCF for 5  min) in 1  ml of fresh 2216 media, then diluted 10× in DDW (double distilled water), and (iii) 8  ml of E. coli culture washed and resuspended (4000 RCF for 5 min) in 1 ml of DI water, then diluted 10× in DDW. For each species, 4 μl of cell suspension was applied to a glow discharged copper mesh carbon coated grid and allowed to adsorb to the grid for 30 s. The grids were briefly washed in DDW, followed by staining with 1% Aqueous Uranyl Acetate, and allowed to dry fully before imaging. Grids were imaged using a FEI Morgagni transmission electron microscope (FEI, Hillsboro, OR) operating at 80 kV and equipped with a CMOS camera (Nanosprint5, AMT).

Appendix 1

Multiplexed chemotaxis device (MCD) microchannel resistances and dimensions

Hydraulic circuit analysis (Materials and methods) was used to determine the hydraulic resistance and thus geometry of the individual microchannels comprising the MCD channel network (Figure 1d and e). The location of each channel in the dilution layer is indicated in the device layout in Figure 1d and in the hydraulic circuit in Figure 1—figure supplement 2a. Subscripts “CIL” indicate channels in the cell injection layer shown in Figure 1e and Figure 1—figure supplement 2b. The resistance for the observation region (R4,4) varies in the microchannel width due to the hydraulic resistance varying between the channel sections before and after the chemical and buffer solution meet (Figure 1d), and the micromixer (RM) has nonuniform channel height due to the herringbone ridges (Materials and methods; see also Figure 1—figure supplement 1).

Appendix 1—table 1
Multiplexed chemotaxis device (MCD) microchannel resistance and dimensions (corresponding to Figure 1—figure supplement 2f).
ChannelResistance(mPamPa.s. μm-3)Height(μm)Width(μm)Length(mm)
R0,20.0082909019.06
R1,10.0095909022.08
R2,10.0147909034.06
R3,10.0201909046.43
R4,10.0277909064.13
R5,10.0300909069.45
R0,40.0201909046.50
R1,40.0149909034.54
R2,40.0098909022.60
R3,40.0046909010.76
RB0.00359015019.93
RM0.0043-20040.7
R4,47.7 × 10−490--
RCIL,10.0575757564.24
RCIL,20.0142757515.89

Data availability

Data files used during the current study, including all CAD designs and drawings, and code to reproduce the analysis reported are publicly available at BCO-DMO (https://www.bco-dmo.org/dataset/885701). The algorithm codes are described in the Materials and methods.

The following data sets were generated

References

    1. Cha SH
    (2007)
    Comprehensive survey on distance/similarity measures between probability density functions
    Mathematical Models and Methods in Applied Sciences 1:300–307.
  1. Book
    1. Kirby BJ
    (2010)
    Micro- and Nanoscale Fluid Mechanics: Transport in Microfluidic Devices
    Cambridge: Cambridge University Press.
  2. Book
    1. Nguyen N
    2. Wereley S
    3. Shaegh S
    (2019)
    Fundamentals and Applications of Microfluidics (Third Edition)
    Artech House.
    1. Ottino JM
    2. Wiggins S
    (2004) Introduction: mixing in microfluidics
    Philosophical Transactions. Series A, Mathematical, Physical, and Engineering Sciences 362:923–935.
    https://doi.org/10.1098/rsta.2003.1355
  3. Book
    1. Pozrikidis C
    (2001) Channel, tube, and film flows
    In: Pozrikidis C, editors. In Fluid Dynamics. Springer US. pp. 306–363.
    https://doi.org/10.1007/978-1-4757-3323-5
  4. Book
    1. Stehnach MR
    (2022)
    Microbial transport in inhomogeneous environments
    Tufts University.
  5. Book
    1. Weinbauer M
    2. Chen F
    3. Wilhelm S
    (2011)
    Virus-mediated redistribution and partitioning of carbon in the global oceans
    In: Weinbauer M, editors. Microbial Carbon Pump in the Oceans. Science/AAAS. pp. 54–56.

Decision letter

  1. Raymond E Goldstein
    Reviewing Editor; University of Cambridge, United Kingdom
  2. Meredith C Schuman
    Senior Editor; University of Zurich, Switzerland

Our editorial process produces two outputs: (i) public reviews designed to be posted alongside the preprint for the benefit of readers; (ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "Multiplexed microfluidic screening of bacterial chemotaxis" for consideration by eLife. Your article has been reviewed by 2 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Meredith Schuman as the Senior Editor. We regret the very long time it has taken to complete the review process – we had to consult a number of reviewers before obtaining the reviews below.

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

General matters:

1) We urge the authors to make available the original channel designs they used for these 2-layer devices, e.g. 3D CAD files or similar, to make it as easy as possible for another user to replicate/print this design.

2) Please also make available any datasets and codes associated with this project, including any custom codes for calibrating the channels.

3) In the discussions, expand on the vision of what exactly these devices might be useful for, beyond the type of study demonstrated in Figure 4, perhaps beyond bacterial chemotaxis (e.g. what about eukaryotic chemotaxis) and with some discussion of design limitations.

Specific issues:

1. In several of the examples presented, e.g. Figure 2a, g, it seems that at long times (>5 minutes) a significant negative chemotactic response follows from the initially positive chemotactic response (not just delayed accumulation), where does this come from? Is this a known effect?

2. Related to the above comment, it seems that a major advantage of the MCD is its ability to sustain precise chemical gradients over a spectrum of concentrations, so these results could provide new insights into the dynamics of bacterial chemotaxis and accumulation, as demonstrated in the curves of β(t) vs time in figures 2,3. Perhaps Figure 4 could be supplemented with some more sophisticated analysis of the dynamics of the response – how does β vary in time at the different concentrations, not just the behaviour of max-β.

3. A few more words of justification on the choice of species/chemical would be helpful – why select those combinations? The response of VA to leucine is not unexpected, given observations in other motile species. It would have been more interesting to see the MCD applied to a greater number of species/chemostimulant than these standard combinations (for comparison, the scope of Raina et al., 2022, was much more extensive). Alternatively, could there be some attempt to connect – at a mechanistic or molecular level – these detailed measurements of how the cells responded to chemoattractants to their species-specific motility strategies/number of flagella, etc?

4. What determines the width of the central band of immotile cells that is to be ignored in the analysis? Was this always the same width across different species? Is anything gained by removing the band from the analysis? What if this procedure introduces some bias by removing motile bacteria that are transitioning from a positively chemotactic towards a negatively chemotactic state?

5. The concluding paragraphs seem to be somewhat self-congratulatory, while there are many potential applications of the device, the paper only demonstrated one scenario where it concretely outperforms traditional assays in terms of efficiency, the usefulness and relevance of the MCD for 'ecosystem and human health investigations' are grand claims that remain to be tested.

6. For balance, there should also be some statements outlining some of the limitations of the device, how accessible is this design for a lab ill-equipped to build multilayer devices compared to the traditional single assay designs, what about the propensity for clogging, etc.

7. Line 56: change principle to principal.

8. Reconsider the use of "stratified", "stratification" etc. Using "fluid stratification" to describe chemical gradients does not seem like the most straightforward word choice, however, is this is a stylistic choice?

9. It is noted that these devices are "reusable"? Obviously, one might be apprehensive about reusing a device that has been exposed to phenol or other toxins, or other bacterial species that might affect motility of the focal strain. Were the devices actually reused in this study? If so, maybe you can explain how they were "cleaned" between uses? e.g. by injecting ethanol or..?? This might be useful for users of these devices.

10. Figure 2I – please describe how the constant gradients strengths were obtained using the device. It is not clear if these measurements were obtained using a multiplexed device operated in a different manner or..?? The description provided in Figure 1 —figure supplement 3 is not clear to us.

11. Figure 3. – all of the cells in the central 250 µm wide band are not necessarily "non-motile" (a fraction of these are certainly motile given the random nature of bacterial motility). The same in Line 208 of the manuscript. Also, it would be more straightforward to list the bacterial strain and chemoaffector on the figure itself to avoid having to dig so deeply into the caption (especially since you show an EM image of each strain).

12. Line 281: it is not clear what you mean by "alluded"?

13. Line 323: rather than "aging cells" it would be more accurate to say something like "metabolic changes in cell cultures that occur over time".

14. Line 328: how might the MCD design be amenable to cell retrieval (presumably you mean isolating chemotactic cells from non-chemotactic ones), given that the channels have a common outlet?

15. Line 570: it is noted that "Each experiment was technically repeated at least three times with the same culture and repeated on different days with freshly grown cells." However, it is not clear what the error bars and the shaded regions around line graphs in the various figures actually indicate. Are these indicating variability across technical repeats or biological repeats or..?

16. Figure 1 —figure supplement 3. It is noted that: "To achieve symmetric flow the applied pressure was tuned." This tuning was required because of the small variations in channel height and tubing lengths. Is such tuning required to ensure that the desired dilutions are accurate? It seems that such tuning could only be accomplished with dye in a separate "calibration" experiment. As noted in the public review, it is not clear whether these devices can produce accurate dilutions "off the shelf" or whether the applied pressures of each device need to be calibrated in separate experiments with dye.

17. Figure 2 —figure supplement 2: sentence after (b) is lacking a verb.

https://doi.org/10.7554/eLife.85348.sa1

Author response

Essential revisions:

General matters:

1) We urge the authors to make available the original channel designs they used for these 2-layer devices, e.g. 3D CAD files or similar, to make it as easy as possible for another user to replicate/print this design.

The original channel designs have been submitted to an existing dataset in the BCO-DMO database and will be publicly available imminently. A citation (line #733) for this dataset has been included in the data availability statement for ease of access.

2) Please also make available any datasets and codes associated with this project, including any custom codes for calibrating the channels.

All codes and supporting material, including custom codes for channel calibrations, have been submitted to the BCO-DMO database and will be publicly available imminently. A citation (line #733) for this dataset has been included in the code availability statement for ease of access.

3) In the discussions, expand on the vision of what exactly these devices might be useful for, beyond the type of study demonstrated in Figure 4, perhaps beyond bacterial chemotaxis (e.g. what about eukaryotic chemotaxis) and with some discussion of design limitations.

We thank the Reviewers for this feedback – we agree that our original Discussion was limited on other potential research avenues to which this device might be applied, as well as limitations of the device itself. As mentioned in our original Discussion, the current design will accommodate eukaryotic cells and could be scaled up for larger organisms, but we have now expanded our revised Discussion (along with incorporating comments from the Public Reviews) to include how the MCD could be adapted to study other conditions such as slow steady flow for surface-attached microorganisms for example. Furthermore, a new paragraph has been added to the Discussion to comment on some of the limitations of the MCD (again, incorporating other comments from Specific Issues raised in these reviews) and how such limitations could be mitigated for potential future users (line #348).

Specific issues:

1. In several of the examples presented, e.g. Figure 2a, g, it seems that at long times (>5 minutes) a significant negative chemotactic response follows from the initially positive chemotactic response (not just delayed accumulation), where does this come from? Is this a known effect?

Yes, for particularly high initial concentrations, the direction of chemotactic bias changes in the channel, which was briefly noted in the Results section (“Validation of MCD performance….”) of our original manuscript. Specifically, this is a consequence of the serine consumption of V. alginolyticus – as the cells migrate up the gradient they are also consuming the available serine. The high degree and rapid nature of their accumulation in the high concentration region results in a rapid consumption of serine. This local depletion of serine combined with the diffusion of remaining serine across the channel can result in an inversion of the chemical gradient and thus a reversal of the cell accumulation at intermediate times. This phenomenon is not observed at lower initial concentrations, because after any initial consumption, the serine concentrations (and thus gradients) are likely below the detection limits of V. alginolyticus. However, regardless of the mechanism, we observe highly consistent behaviour between the single assay (SA) device and the MCD (Figure 2e-j). From a purely microfluidic perspective, the performance of the two devices is highly consistent with one another under these particular conditions (i.e. apart from any nuances of the biological response of the cells). Based on the Reviewer’s comment, we now recognize that our description of this behavior in the original manuscript was too brief and insufficient for the reader. In response, we have now expanded on this description in the Results section (line #168; “Validation of MCD performance….”) to improve the clarity surrounding this point.

2. Related to the above comment, it seems that a major advantage of the MCD is its ability to sustain precise chemical gradients over a spectrum of concentrations, so these results could provide new insights into the dynamics of bacterial chemotaxis and accumulation, as demonstrated in the curves of β(t) vs time in figures 2,3. Perhaps Figure 4 could be supplemented with some more sophisticated analysis of the dynamics of the response – how does β vary in time at the different concentrations, not just the behaviour of max-β.

We thank the Reviewer for this feedback, and agree there is scope for more sophisticated analysis of the accumulation dynamics from assays such as the MCD. While we emphasize that the primary goal of the present work is to demonstrate the performance of the device, in response to this feedback we have carried out several additional analyses into the accumulation dynamics captured by β(t). These results are now included in an updated Figure 4 (top, page 9) and associated text (line #289).

To supplement our original analysis, we examined the rate of response by calculating the maximal value of dβ/dt (prior to peak accumulation), which is indicative of the rate of accumulation and the chemotactic velocity of the bacteria (Son et al. 2016). The rate of response is a particularly useful metric to separate the strong observed responses of different species. In particular, at high concentrations (200 uM), V. alginolyticus and P. haloplanktis exhibit a similar degree of accumulation (βmax) to serine along with V.

alginolyticus to leucine. In contrast, the maximal rate of accumulation max(dβ/dt) for the rapid-swimming P. Haloplanktis is nearly twice as high as V. alginolyticus at 200 uM. Separately, for chemotaxis of V. alginolyticus to leucine, these bacteria exhibit a maximum degree of accumulation at 49 μm but accumulate fastest at lower chemostimulant concentrations of 3.6 uM.

To address this feedback, we have added a new Figure 4 (b) (top, page 9) showing the maximal response rate (i.e. max(dβ/dt), prior to the maximum cell accumulation) along with associated text in the Results section (line #289; “Multiplexed microfluidic device …”). As pointed out by the Reviewer, this additional analysis provides a broader perspective on the responses summarised by β(t). This approach demonstrates an advantage of the MCD in comparison to other high-throughput assays in being able to extract cell motility information from such single-cell assays in addition to simple population accumulation. We envision that future applications will take advantage of the flexibility afforded by the MCD and will adapt and improve upon the analytical approaches presented here.

3. A few more words of justification on the choice of species/chemical would be helpful – why select those combinations? The response of VA to leucine is not unexpected, given observations in other motile species. It would have been more interesting to see the MCD applied to a greater number of species/chemostimulant than these standard combinations (for comparison, the scope of Raina et al., 2022, was much more extensive). Alternatively, could there be some attempt to connect – at a mechanistic or molecular level – these detailed measurements of how the cells responded to chemoattractants to their species-specific motility strategies/number of flagella, etc?

The choices of species/chemical were selected due to their prominence in chemotaxis literature – this is briefly summarised in the beginning of the Results section “Multiplexed microfluidic device supports high-throughput chemotaxis screening for novel stimuli and various microorganism”. The chemotaxis of the organisms in the present study to model attractants such as serine has been well-documented and is extremely robust. Between the three organisms, we cover two main branches of bacterial motility: the classical run-tumble of E. coli, and the run-reverse-flick of both P. haloplanktis and V. alginolyticus. Hence, these species/chemicals were chosen primarily for benchmarking the MCD against a single gradient generation assay (SA) due to the (i) robustness, (ii) reliability, and (iii) reproducibility of the known chemotactic response. This approach aided in eliminating (as much as possible) other sources of uncertainty between the two devices.

The Reviewer is indeed correct – the MCD does open up interesting applications including (but not limited to) a larger range of species/chemostimulant and motility-dependent studies. We appreciate the Reviewer pointing out Raina et al. 2022, which was considerably more extensive in its application. In their work, the authors applied the in-situ chemotaxis assay (ISCA) to a large multi-year field study to study the chemotactic responses of prokaryotes towards phytoplankton-derived dissolved organic matter (DOM) exudates. However, we emphasise that a fairer comparison would perhaps be the original publication (Lambert et al., Nat. Microbiol. (2017); cited in main text) of the ISCA device, without which the study in Raina et al. 2022 would not have been possible. In Lambert et al. 2017, they first establish a method, then demonstrate the effectiveness of the ISCA device in a laboratory setting. To accomplish this, they rely on the known chemotactic response of Vibrio coralliilyticus, before undertaking a first in-situ deployment. The work by Lambert et al. 2017, thus opened the door to a range of studies, which extensively utilise the ISCA device, including Raina et al. 2022.

In our current work, we are proposing the MCD as a complementary approach to such devices as the ISCA – i.e. to provide a high-throughput microfluidic alternative that can both rapidly screen across concentrations whilst retaining the ability to extract detailed information from both the population and single-cell levels. We do not doubt that future studies will be able to apply the MCD to a more extensive range of conditions and combinations now that the device has been established. However, our ability to make compelling, broad insights through comparative studies among the organisms tested – based on motility patterns or flagellation – is unfortunately beyond the scope of the present work. We feel that such comparisons would be largely speculative and divergent from our primary goal of establishing the performance of the MCD.

To address the above points based on this Reviewer comment, we have added text and supporting references to the end of the aforementioned Results section (line #297; “Multiplexed microfluidic device supports …”) and Discussion section highlighting the complementary nature of the MCD in comparison to existing techniques such as the ISCA (line #297).

4. What determines the width of the central band of immotile cells that is to be ignored in the analysis? Was this always the same width across different species? Is anything gained by removing the band from the analysis? What if this procedure introduces some bias by removing motile bacteria that are transitioning from a positively chemotactic towards a negatively chemotactic state?

The initial (t=0 min) band of cells in the centre of the device includes all cells – both motile and non-motile (Figure 2e,g). The width of this initial band is set by the predetermined flow rates in the three branches (Figures 1a and 2b,c) leading into the observation region of the device (Figure 1 —figure supplement 3). While the fraction of motile cells is different across species, the initial spatial distribution of cells across species was constant in the device – provided that the flow rates were sufficient to maintain stratification. We agree this point was not clear in the original manuscript, so we have now added an additional figure supplement (specifically Figure 3 Figure Supplement 1) which shows the cell distributions at the beginning (t=0 min; solid lines) and end (t=10 min; dashed lines) of each experiment for all observation channels in the MCD. At time t=0 (a), the cell positions from all experiments consistently overlap in the expected zone in the centre of the channel. At time t=10 min (b), the motile cells have dispersed with the exception of the sub-population of non-motile cells, which remain in the centre of the channel. This feature is more prominent, for example, with P. haloplanktis than it is with V. alginolyticus. The vertical dashed black lines indicate the 250 μm wide “excluded” region of cells corresponding to the central band. The spatial distribution of non-motile cells was consistent across all experiments and comparable to the theoretical width (111um) based on the relative flow rate of the cell injection stream. Based on this high degree of consistency, we chose a fixed exclusion zone of 250 μm in the channel centre to encapsulate the majority of non-motile cells for all experiments.

We emphasise that the cells in the central band are not entirely ignored and that our choice of exclusion region has no bearing on the outcome of our analysis. The benefit of excluding this region from the conditional probability plots (P(y|t); Figures 2e,g and 3a,c,e,g) is that it better highlights the chemotactic nature of the cells: Large numbers of non-motile cells simply skew the visualisation by drawing the reader’s attention to strong features in the plots, which represent cells that are not contributing to the population level response. The calculation of β(t) only depends upon cells that approach within 200 μm of the upper/lower boundaries (Figure 2c), which is a typical method of studying cell accumulation (Seymour et al. 2010, Stocker et al. 2008). Thus, the instantaneous cell locations in the central band and those used in the calculation of β(t) are mutually exclusive, with the former having no bearing on β(t). We further note that any cells initially in the central band, which leave that region are of course counted in the analysis if they migrate toward the upper/lower boundaries.

We now clarify all of these points in relevant sections of the revised manuscript (Figure 3 caption (page 7), Results: Validation of MCD performance against conventional chemotaxis assays). See also our response to Specific Issue #11 below.

5. The concluding paragraphs seem to be somewhat self-congratulatory, while there are many potential applications of the device, the paper only demonstrated one scenario where it concretely outperforms traditional assays in terms of efficiency, the usefulness and relevance of the MCD for 'ecosystem and human health investigations' are grand claims that remain to be tested.

We agree with the Reviewer that we do lean heavily into applications outside of the tested scope of the MCD. We have toned down our language in the concluding paragraphs and added literature support where appropriate to better place the MCD in context of potential future applications. Additionally, we have added some discussion of limitations of the MCD in this section per Specific Point #6 (see response below).

6. For balance, there should also be some statements outlining some of the limitations of the device, how accessible is this design for a lab ill-equipped to build multilayer devices compared to the traditional single assay designs, what about the propensity for clogging, etc.

We thank the Reviewer for this feedback, and we agree that our original Discussion did not sufficiently balance the limitations of the device against its advantages. To address this, we have added a new paragraph to the Discussion section (line #348; also mentioned in our reply to General Point #3) to summarise some limitations of the device, and how these might be mitigated by potential future users of the MCD.

To the Reviewers’ specific points: We expect that most laboratories capable of conventional single layer microfluidics will also have the capability to produce multilayer devices with multilayer lithography, where the key piece of equipment is a mask aligner. In our experience, neither conventional single assay devices nor the MCD are prone to clogging by cells during normal operation. However, partial or full blockages of channels may occur during the fabrication process or if contaminants are present in the cell media or buffer (i.e. dust). Thus, normal precautions should be taken as for any microfluidics experiments, including inspecting devices after fabrication and filtering media to the extent possible. We now comment directly on both of these points and additional potential issues in both the Discussion (line #363) and Methods (“line #496”) sections of the revised manuscript.

7. Line 56: change principle to principal.

This typo has been corrected, and the manuscript has been proofed for further grammatical errors.

8. Reconsider the use of "stratified", "stratification" etc. Using "fluid stratification" to describe chemical gradients does not seem like the most straightforward word choice, however, is this is a stylistic choice?

We thank the reviewers for pointing out the potential confusion with the phrasing. We emphasize that the word “stratification” is used to describe only the initial state of the gradient, which is set by the very strong initial fluid flow. In other words, the initial fluid flow actively maintains three distinct regions having abrupt interfaces within the observation channel. We feel that this intended use is consistent with the general definition of the word stratum, meaning a layer or series of layers (i.e. not meant to be taken as “density stratified” in the traditional fluid mechanical sense). Where the word “stratified” is first mentioned in both the main text and Figure 1, we clarified the wording in hopes to indicate the word stratified explains the pattern formed by the 3 fluids flowing. For example, the first paragraph of the Results section now reads (line #106):

“The individual assays in the observation region are based on laminar flow patterning and established stop-flow diffusion methods: Rapid, parallel flow of chemostimulus (concentration, $C_i$) and buffer ($C=0$) solutions maintain initially stratified fluid regions. Upon stopping the flow, a chemostimulus gradient forms via diffusion (Figure 1a,b).”

9. It is noted that these devices are "reusable"? Obviously, one might be apprehensive about reusing a device that has been exposed to phenol or other toxins, or other bacterial species that might affect motility of the focal strain. Were the devices actually reused in this study? If so, maybe you can explain how they were "cleaned" between uses? e.g. by injecting ethanol or..?? This might be useful for users of these devices.

The Reviewers are correct that users should proceed with caution when using any device that previously contained incompatible materials. However, the MCD devices and most other PDMS microfluidic devices (e.g. the single gradient generator devices also employed here) can be reused in a number of ways: First, as demonstrated in the present study, technical replicates can be easily achieved by simply restarting the flow, allowing the fluid streams to re-stratify, then stopping the flow to begin a new replicate experiment. Second, if the chemoaffectors/organisms are identical, devices can be reused between different biological samples by cleaning the device, which involves flowing ethanol, then DI water sequentially through all of the inlets. Devices are then dried by flowing clean compressed air through the devices and finally placed under vacuum. Ethanol readily sterilizes the surfaces and helps to desorb cells and chemoaffectors from the surfaces. It is highly miscible with water and highly volatile, allowing it to be efficiently washed and evaporated from the device, respectively.

We agree this reusability is not clear in the manuscript and is useful information for future users of the MCD (as well as being an advantage of such microfluidic designs). Hence, we have added a subsection to the Methods (line #618; “Experiment replicates and device reusability”) summarising how replicates were achieved and the cleaning protocol for the device. Additionally, we added the following statements for future users to consider:

“…caution should be exercised to ensure no bias is introduced. It is recommended to use a new device outside biological replicates of a particular organism/chemostimulus combination, consistent with the majority of PDMS applications.”

10. Figure 2I – please describe how the constant gradients strengths were obtained using the device. It is not clear if these measurements were obtained using a multiplexed device operated in a different manner or..?? The description provided in Figure 1 —figure supplement 3 is not clear to us.

We apologise for any confusion around Figure 2l (page 6). We have revised this figure caption for clarity, which now reads:

“No gradient (k) conditions ($C_i=0$) were obtained by injecting buffer into the chemical inlet (setting $C_0=0$). Fixed gradient (l) conditions (Ci=200 μM of serine) were obtained by injecting C0=200 μM of serine into both the chemical and buffer inlets of the dilution layer.”

A similar description is also provided in the caption for Figure 2 —figure supplement 1, from which Figure 2l was derived.

Additionally, we have revised the caption for Figure 1—figure supplement 3 (line #992) for clarity.

11. Figure 3. – all of the cells in the central 250 µm wide band are not necessarily "non-motile" (a fraction of these are certainly motile given the random nature of bacterial motility). The same in Line 208 of the manuscript. Also, it would be more straightforward to list the bacterial strain and chemoaffector on the figure itself to avoid having to dig so deeply into the caption (especially since you show an EM image of each strain).

The Reviewer is correct in the split between motile/non-motile sub-populations in the central region of the channel: Initially (t=0 min), all cells including both motile and non-motile cells are located in the central band. As the motile bacteria migrate relative to the chemoaffector and swim randomly, they mostly disperse from the central region. In this work we refer to non-motile cells as those which have not moved (aside from diffusion) over the course of an experiment. See also our response to Specific Point #4 and the new figure addition of Figure 3 —figure supplement 1, which shows a consistent sub-population of cells that do not move from their starting positions for the full duration of each experiment. Undoubtedly, this band still contains some motile cells. However, we emphasise that this band is omitted simply for clarity in the kymographs of P(y|t). As such, this has no impact on our subsequent analysis, as β(t) by definition is computed using cell subpopulations far outside of that central band.

To further improve the clarity around this 250 μm wide central band, we have revised the Figure 3 (page 7) caption to read:

“Central 250 μm wide band, which often contains a significant subpopulation of non-motile cells, is omitted from P(y|t) for clarity with no impact on β(t).”

We have also added text to the manuscript (“Validation of MCD performance….”) that explicitly reads (line #210):

“The persistent central band of cells at later times is due to a sub-population of bacteria which remain non-motile over the course of each assay and are thus localised near y=0 (Figure 2e,g; Figure 3—figure supplement 1). Because any nonmotile or motile bacteria in this region do not impact the calculation of β(t), the central band is omitted from future kymographs for visualization purposes.

Finally, we have updated Figure 3 to include the strain names and chemoaffectors as titles on the accumulation plots (Figures 3bdfh), as well as the new Figure 3 —figure supplement 1 (line #885).

12. Line 281: it is not clear what you mean by "alluded"?

The word “alluded” has been omitted, and the sentence has been revised to read “… a feature that is consistent with the chemotactic sensitivity of E. coli to serine …”.

13. Line 323: rather than "aging cells" it would be more accurate to say something like "metabolic changes in cell cultures that occur over time".

The phrase “aging cells” has been revised to “changes in cell cultures that may occur over time”.

14. Line 328: how might the MCD design be amenable to cell retrieval (presumably you mean isolating chemotactic cells from non-chemotactic ones), given that the channels have a common outlet?

Our choice of a single common outlet for this particular iteration of the MCD design stemmed from our focus on demonstrating the primary mode of operation of the device. However, the device design could be easily altered to accommodate separate outlets for each of the individual channels. Moreover, sophisticated existing chemotactic cell-sorting designs may be implemented for each channel to separate and collect subpopulations of cells with different chemotactic performance (e.g. Mao et al. 2003). We agree that this aspect was not clearly articulated in the manuscript, and thus we have updated portions of the text in the Discussion to highlight this (line #331), including supporting citations.

15. Line 570: it is noted that "Each experiment was technically repeated at least three times with the same culture and repeated on different days with freshly grown cells." However, it is not clear what the error bars and the shaded regions around line graphs in the various figures actually indicate. Are these indicating variability across technical repeats or biological repeats or..???

We apologise for this lack of clarity, and we have now revised the description of the error bars/shading in the appropriate figure captions. Variability is reported in most instances across biological repeats.

16. Figure 1 —figure supplement 3. It is noted that: "To achieve symmetric flow the applied pressure was tuned." This tuning was required because of the small variations in channel height and tubing lengths. Is such tuning required to ensure that the desired dilutions are accurate? It seems that such tuning could only be accomplished with dye in a separate "calibration" experiment. As noted in the public review, it is not clear whether these devices can produce accurate dilutions "off the shelf" or whether the applied pressures of each device need to be calibrated in separate experiments with dye.

We thank the reviewer for the comment. Foremost, we emphasise that the accuracy of the dilution is set by the design and fabrication of the dilution layer alone (Figure 1c,d), and it is independent of the calibration mentioned here. The thickness of the dilution layer channels is determined essentially by a single lithography process (Methods). The relative resistances of the dilution layer channels (Appendix 1; Figure 1—figure supplement 2) – and thus their dilution ratios – are largely unaffected by minor uncertainty in the absolute channel height from lithography.

However, differences in the fabricated height of the dilution layer versus the cell injection layer can lead to asymmetric flow of cells in the observation region (again, independent of the dilution process). The calibration to which the Reviewer refers is a simple process to ensure that the initial distribution of cells is centred in the observation channel during an assay (thus avoiding any bias in the cell distribution). Indeed, the device only requires a single calibration to determine the relative driving pressures required by the dilution and cell injection layers to achieve centering of the cells. Furthermore, this calibration need not be achieved with fluorescence microscopy or any specialised imaging: When setting up the experiment, observation of the cells (or tracer particles or dye) in the central channel are sufficient to perform pressure adjustment for calibration to ensure the cells start in the middle of the microchannel. In the present work, we used fluorescence imaging for demonstration purposes and ensure that the calibration was clear (Figure 1 —figure supplement 3). To clarify this issue, we have added the following text to emphasize the nature of the calibration and that it only needs to be performed once for the first device fabricated from a given set of molds:

– Results (“Multiplexed microfluidic device….”; line # 158):

“Due to variations in the replica mold fabrication process, a one-time tuning of the applied pressure ratio (p_{in,1-2} / p_{in,3-4}) for the dilution and cell injection layers is required to ensure symmetric flow in the observation channel (Methods).”

– Discussion (line #320):

“The MCD performs reliably for a wide range of applied pressures, and only requires calibration once, as long as the PDMS microchannels are cast from the same molds (Methods).”

– Methods (“MCD dilution, flow, and …”; line #532):

“The symmetry of the stratified chemostimulus and buffer distributions in the observation channel is critical to prevent bias in the chemotaxis measurements. As minor errors in the manufacturing process can alter this symmetry, the applied pressure to the cell injection layer ($p_{in},3,4$; Figure1—figure supplement 2) was tuned until the widths of the chemical, cell, and buffer streams in each observation channel were 4:1:4 ratio, respectively. Tuning was visualized by flowing a fluorescein solution (C_0 = 100 uM) in both the chemical and buffer inlets of the dilution layer as well as the buffer inlet of the cell injection layer (Figure 1d,e). The ratio of applied pressures (pin;1-2_pin;3-4) between the dilution and cell injection layer remained the same for all chemotaxis assays (pin;1-2_pin;3-4 = 10_7) which was slightly lower than the designed value (pin;1-2_pin;3-4 = 2). Tuning is only required for the first device fabricated from a particular set of molds, after which the calibration and tuning applies to all subsequent devices fabricated from the same mold set due to the robust nature of soft lithography. If significant fluid width variation across the observation regions is observed, this means the MCD is not functioning properly.”

17. Figure 2 —figure supplement 2: sentence after (b) is lacking a verb.

There is not a Figure 2 —figure supplement 2 in the original manuscript, so we believe this comment refers to either Figure 1 —figure supplement 2 or Figure 2 —figure supplement 1. We have reviewed the text in both captions and made grammatical corrections as noted by the Reviewers.

https://doi.org/10.7554/eLife.85348.sa2

Article and author information

Author details

  1. Michael R Stehnach

    Department of Mechanical Engineering, Tufts University, Medford, United States
    Present address
    Department of Physics, Brandeis University, Waltham, United States
    Contribution
    Conceptualization, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing – original draft, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-0302-4641
  2. Richard J Henshaw

    Department of Mechanical Engineering, Tufts University, Medford, United States
    Present address
    Department of Civil, Environmental and Geomatic Engineering, Institute of Environmental Engineering, ETH Zürich, Zürich, Switzerland
    Contribution
    Conceptualization, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing – original draft, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-7053-9341
  3. Sheri A Floge

    Department of Biology, Wake Forest University, Winston-Salem, United States
    Contribution
    Conceptualization, Supervision, Funding acquisition, 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:" 0000-0002-1996-3347
  4. Jeffrey S Guasto

    Department of Mechanical Engineering, Tufts University, Medford, United States
    Contribution
    Conceptualization, Supervision, Funding acquisition, Investigation, Methodology, Writing – original draft, Project administration, Writing – review and editing
    For correspondence
    Jeffrey.Guasto@tufts.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-8737-8767

Funding

National Science Foundation (1829827)

  • Michael R Stehnach
  • Richard J Henshaw
  • Jeffrey S Guasto

National Science Foundation (1554095)

  • Michael R Stehnach
  • Richard J Henshaw
  • Jeffrey S Guasto

National Science Foundation (1701392)

  • Michael R Stehnach
  • Richard J Henshaw
  • Jeffrey S Guasto

National Science Foundation (1829905)

  • Sheri A Floge

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

Acknowledgements

We thank J Vlahakis of the Tufts Micro and Nano Fabrication Facility for assistance in the device fabrication, JB. Noble for preliminary work on the microfluidic device design, and R Stocker and M Salek for helpful discussions. TEM samples were prepared and imaged by the Brandeis Electron Microscope Facility. This work was funded by NSF Awards OCE-1829827, CAREER-1554095, and CBET-1701392 (to JSG), and OCE-1829905 (to SAF).

Senior Editor

  1. Meredith C Schuman, University of Zurich, Switzerland

Reviewing Editor

  1. Raymond E Goldstein, University of Cambridge, United Kingdom

Version history

  1. Preprint posted: December 2, 2022 (view preprint)
  2. Received: December 4, 2022
  3. Accepted: June 15, 2023
  4. Version of Record published: July 24, 2023 (version 1)

Copyright

© 2023, Stehnach 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

  • 649
    Page views
  • 102
    Downloads
  • 0
    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. Michael R Stehnach
  2. Richard J Henshaw
  3. Sheri A Floge
  4. Jeffrey S Guasto
(2023)
Multiplexed microfluidic screening of bacterial chemotaxis
eLife 12:e85348.
https://doi.org/10.7554/eLife.85348

Share this article

https://doi.org/10.7554/eLife.85348

Further reading

    1. Ecology
    2. Plant Biology
    Jamie Mitchel Waterman, Tristan Michael Cofer ... Matthias Erb
    Research Article

    Volatiles emitted by herbivore-attacked plants (senders) can enhance defenses in neighboring plants (receivers), however, the temporal dynamics of this phenomenon remain poorly studied. Using a custom-built, high-throughput proton transfer reaction time-of-flight mass spectrometry (PTR-ToF-MS) system, we explored temporal patterns of volatile transfer and responses between herbivore-attacked and undamaged maize plants. We found that continuous exposure to natural blends of herbivore-induced volatiles results in clocked temporal response patterns in neighboring plants, characterized by an induced terpene burst at the onset of the second day of exposure. This delayed burst is not explained by terpene accumulation during the night, but coincides with delayed jasmonate accumulation in receiver plants. The delayed burst occurs independent of day:night light transitions and cannot be fully explained by sender volatile dynamics. Instead, it is the result of a stress memory from volatile exposure during the first day and secondary exposure to bioactive volatiles on the second day. Our study reveals that prolonged exposure to natural blends of stress-induced volatiles results in a response that integrates priming and direct induction into a distinct and predictable temporal response pattern. This provides an answer to the long-standing question of whether stress volatiles predominantly induce or prime plant defenses in neighboring plants, by revealing that they can do both in sequence.

    1. Ecology
    Congnan Sun, Yoel Hassin ... Yossi Yovel
    Research Article

    Covid-19 lockdowns provided ecologists with a rare opportunity to examine how animals behave when humans are absent. Indeed many studies reported various effects of lockdowns on animal activity, especially in urban areas and other human-dominated habitats. We explored how Covid-19 lockdowns in Israel have influenced bird activity in an urban environment by using continuous acoustic recordings to monitor three common bird species that differ in their level of adaptation to the urban ecosystem: (1) the hooded crow, an urban exploiter, which depends heavily on anthropogenic resources; (2) the rose-ringed parakeet, an invasive alien species that has adapted to exploit human resources; and (3) the graceful prinia, an urban adapter, which is relatively shy of humans and can be found in urban habitats with shrubs and prairies. Acoustic recordings provided continuous monitoring of bird activity without an effect of the observer on the animal. We performed dense sampling of a 1.3 square km area in northern Tel-Aviv by placing 17 recorders for more than a month in different micro-habitats within this region including roads, residential areas and urban parks. We monitored both lockdown and no-lockdown periods. We portray a complex dynamic system where the activity of specific bird species depended on many environmental parameters and decreases or increases in a habitat-dependent manner during lockdown. Specifically, urban exploiter species decreased their activity in most urban habitats during lockdown, while human adapter species increased their activity during lockdown especially in parks where humans were absent. Our results also demonstrate the value of different habitats within urban environments for animal activity, specifically highlighting the importance of urban parks. These species- and habitat-specific changes in activity might explain the contradicting results reported by others who have not performed a habitat specific analysis.