Robust and tunable performance of a cell-free biosensor encapsulated in lipid vesicles

A transcriptionally regulated fluoride riboswitch can function inside lipid vesicles

We first sought to confirm that a transcriptional riboswitch can function when encapsulated inside lipid vesicles. For the riboswitch, we chose the fluoride responsive riboswitch from Bacillus cereus, which we characterized previously (3). In earlier studies, we showed that this riboswitch can be used to control the expression of several different reporter proteins and fluorescent RNA aptamers in bulk Escherichia coli extract–based cell-free systems (3). In this system, the fluoride riboswitch is encoded within a single DNA template, downstream of a consensus E. coli promoter sequence, and upstream of a reporter coding sequence. In the absence of fluoride, E. coli polymerase transcribes the riboswitch sequence, causing it to fold into a conformation that exposes a transcriptional terminator hairpin and subsequently causes RNA polymerase to stop transcription (38). In the presence of fluoride, fluoride binding to the riboswitch aptamer domain prevents the terminator from folding, allowing transcriptional elongation of the reporter coding sequence.

In this study, we first chose to use a super folder green fluorescent protein (GFP) reporter, as it allows convenient measurement of riboswitch activity. For the cell-free system, we used an E. coli S30 lysate prepared with runoff and dialysis, which has been shown to allow the function of biosensors that require bacterial polymerases (39). Embedding the riboswitch DNA template into the extract system alongside varying concentrations of sodium fluoride (NaF) showed, as expected (3), an increase in GFP fluorescence as fluoride concentrations increased up to 3 mM, followed by a decrease in fluorescence at higher concentrations (Fig. 2A). This decrease is likely caused by fluoride inhibition of the gene expression machinery (40) and is consistent with our previous characterization of this construct (3). Accordingly, we used 3 mM NaF for the remainder of this study to obtain the expected maximum fluorescent output of the system.

We then set out to assess whether the fluoride riboswitch could retain functionality when encapsulated within lipid vesicles. Vesicles were synthesized using a water-in-oil emulsion transfer method (Fig. 2B) (41). In this method, various membrane amphiphiles (e.g., lipids, cholesterol, fatty acids, and diblock copolymers) are dissolved into an oil phase, and an emulsion is formed by vortexing the aqueous cell–free reaction into this mixture. The emulsion is then layered onto a second aqueous layer containing the small-molecule components of the cell-free reaction, and emulsified droplets are centrifuged through the oil-water interface to generate unilamellar vesicles. Previous characterization of vesicles formed using this method has yielded encapsulation efficiencies of ~98%, indicating high levels of entrapment of aqueous components (41). Vesicle synthesis using this approach yields a distribution of different vesicle sizes on the 5- to 50-μm scale, which could affect our quantification of fluorescence (42). To control for this, we also incorporated a protein-conjugated dye, ovalbumin-conjugated Alexa Fluor 647 (OA647), which served as a volume marker and allowed us to detect the vesicle interior regardless of GFP expression level (30, 43). After synthesis, vesicles were incubated under varying conditions at 37°C, and protein expression was assessed using epifluorescent microscopy after 6 hours of incubation (Fig. 2C and fig. S1) (30, 42). Vesicles were imaged using GFP and Cy5.5 channels, and images were analyzed using the NIS-Elements Advanced Research (AR) software program (44), which allowed us to automatically select vesicle interiors using the OA647 marker and report GFP fluorescence in those regions. This protocol allowed us to analyze hundreds of vesicles per sample, maintain the same selection parameters between samples, and minimize the impact of user selection bias in the analysis. In addition, the encapsulated volume marker allowed us to report GFP expression relative to OA647 fluorescence to control for possible variability in vesicle size or loading. Last, to control for variations in levels of extract autofluorescence between vesicle preparations, we analyzed all samples compared to an “extract-only” control population for each preparation. Using these methods, we were able to ensure that our measurements were isolated to intact (nonlysed) vesicles, which retained their protein cargo (fig. S2).

Using the above approach, we encapsulated cell-free reactions with and without fluoride present in the bulk reaction mixture. We chose to use a 2:1 ratio of cholesterol and 1-palmitoyl-2-oleoyl-glycero-3-phosphocholine (POPC) phospholipid as membrane amphiphiles due to their previous use in similar encapsulated expression studies (21, 28–30, 43). Upon coencapsulation of the riboswitch with 3 mM NaF, we observed GFP expression inside vesicles, indicating the riboswitch was in the “ON” state (Fig. 2C). In contrast, in the absence of DNA (extract only) or in the absence of fluoride (0 mM NaF), we observed minimal GFP expression, indicating an “OFF” state (Fig. 2C). This high level of GFP induction inside vesicles by fluoride indicates that membrane encapsulation does not eliminate the ability of the riboswitch to fold properly and does not cause significant nonspecific expression.

We observed that populations of vesicles exhibited variations in GFP fluorescence between individual liposomes after 6 hours of incubation (Fig. 2C), a phenomenon that has been observed in similar studies across multiple encapsulation protocols (21, 24, 30, 42, 45–47). It has been hypothesized that these variations in gene expression may be caused by variability in vesicle loading and/or varied levels of molecular exchange with the surrounding buffer for vesicles of different sizes (24, 45, 46, 48). To report this variability across vesicle populations, we have included metrics of skew for each population result (tables S1 to S3). Even after taking this variability into account, however, induction of GFP expression is clearly observable across the vesicle population, indicating proper riboswitch sensor activity and a robust response to fluoride in encapsulated sensors.

External fluoride can be detected by an encapsulated riboswitch

We next sought to determine whether the encapsulated riboswitch could detect fluoride added to the external solution of preassembled sensor vesicles. To assess this, we prepared vesicles containing cell-free reactions without NaF present in the reaction mixture. We then titrated in NaF into the solution surrounding the vesicles (Fig. 3A) and imaged them following incubation for 6 hours at 37°C. We observed increasing GFP expression with increasing concentrations of NaF up to 3 mM and a slight decrease in average fluorescence at 5 mM, consistent with bulk studies (Fig. 3, B to D). Vesicle populations exhibited increases in both mean GFP/OA647 fluorescence and population skew in response to increasing fluoride, either of which could serve as a metric of fluoride detection (table S1). All fluoride-containing conditions exhibited a significant increase in fluorescence compared to no DNA and no fluoride controls (Fig. 3, B and C, and table S1). When incubated with chloride, a similarly monovalent anion, a slight response to increasing ion concentration was observed; however, these responses were significantly lower than any response to fluoride and did not exhibit any of the highly active vesicles that were observed under all fluoride-containing conditions (fig. S3). These responses were easily distinguishable between fluoride and chloride, indicating sufficient specificity to fluoride, as has been observed previously (3). Together, these results indicate that increasing concentrations of fluoride added to the extravesicular environment can be detected by the encapsulated riboswitch.

This result was somewhat unexpected, as we anticipated that the membrane would be relatively impermeable to charged fluoride ions. The observed magnitude of fluoride permeability may be explained in part by the transient formation of hydrofluoric acid (HF). HF has been shown to exhibit a permeability coefficient that is seven orders of magnitude greater than fluoride anions through lipid/cholesterol bilayers, indicating that HF travels through the membrane much more readily than its anionic F⁻ counterpart (40, 49). We confirmed this effect by encapsulating a pH sensitive dye, 8-hydroxypyrene-1,3,6-trisulfonic acid trisodium salt (HPTS), which reported a slight decrease in pH in the vesicle interior upon the addition of fluoride to the external buffer (fig. S4). This result indicates an increase in proton concentration inside the vesicle as fluoride concentration increases, consistent with cross-membrane transport of HF.

Because we observed that fluoride could pass through the membrane to interact with the encapsulated riboswitch, we wondered whether we could alter the composition of vesicle membranes to modulate membrane permeability and thereby modulate sensitivity of these sensors to external fluoride. Membrane permeability to small molecules depends significantly on membrane composition, as various lipid chain chemistries and contributions from other amphiphilic components can impart an effect on membrane physical properties. Cholesterol, a major component of our original 2:1 cholesterol:POPC lipid composition, is known to decrease membrane permeability by increasing lipid packing and altering membrane fluidity and rigidity (50). Poly(ethylene oxide)-b-poly(butadiene) (PEO-b-PBD) diblock copolymers are similarly known to reduce membrane permeability by increasing membrane viscosity, introducing steric barriers from the polyethylene glycol groups that assemble at the membrane interface, and increasing thickness and chain entanglements within the hydrophobic portions of the membrane (51, 52). In contrast, fatty acids such as oleic acid (OA) have been shown to increase membrane permeability to ionic solutes by incorporating single hydrocarbon chains that have a different shape and amphipathicity than diacyl chains, reducing lipid chain packing and enhancing fluidity of the bilayer (53, 54). Using this series of amphiphilic molecules, we set out to assess the capacity of membrane amphiphiles and the resulting membrane permeability to modulate the performance of an encapsulated cell-free sensor.

To explore the effect of these amphiphiles on membrane permeability to fluoride, we prepared vesicles with (i) pure POPC lipid, (ii) POPC lipid + 10% OA, or (iii) POPC lipid + 10% PEO₁₄-b-PBD₂₂ diblock copolymer [molecular weight = 1.8 kDa; hereafter referred to as 1.8k] components in the lipid/oil mixture, encapsulating cell-free reactions as normal (Fig. 3E). We observed an increase in overall GFP expression in both pure POPC lipid and POPC + 10% OA compositions compared to our original 2:1 cholesterol:POPC lipid composition (for 3 mM external fluoride; 2:1 cholesterol:POPC versus POPC, P < 0.0001; 2:1 cholesterol:POPC versus POPC + 10% OA, P < 0.0001). These results are consistent with the removal of cholesterol and the addition of OA, respectively, both of which are expected to increase membrane permeability. In addition, the inclusion of OA in vesicle membranes led to a reduction of sensor dynamic range, measured via a reduced concentration dependence of GFP expression on NaF concentration, consistent with our hypothesis of increased permeability to any amount of external fluoride (10% OA + 1 mM NaF versus 10% OA + 3 mM NaF, P > 0.9999; 10% OA + 1 mM NaF versus 10% OA + 5 mM NaF, P = 0.8149). In contrast, vesicles containing 10% 1.8k diblock copolymer exhibited reduced responsiveness to increasing external NaF, indicating reduced membrane permeability. Mean POPC vesicle fluorescence peaked at 1 mM NaF, while 10% 1.8k diblock copolymer responses were maximum at 5 mM NaF (table S2). All fluoride-containing conditions for samples containing 10% OA were statistically indistinguishable. Together, these results indicate that exchanging membrane components to control membrane permeability provides a handle to tune the sensitivity of an encapsulated riboswitch to an analyte of interest. Furthermore, the selection of highly permeable amphiphiles does not necessarily improve sensor performance and may instead increase overall signal but limit sensor resolution. A balance between analyte access and desired sensing behavior is likely an important consideration for engineering encapsulated biosensing systems depending on the desired application.

Encapsulation protects sensor components from degradation

Having established that these vesicle sensors can detect external fluoride, we next wanted to explore how they might function in complex samples. One major benefit of membrane encapsulation is the ability to leverage the semipermeable barrier formed by the membrane to contain and protect encapsulated components. Cell-free reactions, particularly those using riboswitches, are highly sensitive to the presence of nucleases and proteases, which can degrade sensor components before a target analyte is encountered (19). Because of their large size, however, enzymes are unable to pass through the vesicle membrane to access encapsulated reactants.

To determine whether the vesicle membrane can sufficiently protect encapsulated reactions from external degradation, we tested various vesicle assemblies in the presence of RNase A (Fig. 4A). We observed that RNase completely eliminated the riboswitch response to NaF both under bulk conditions and when RNase was coencapsulated with the cell-free reaction inside vesicles (Fig. 4, B and C). In contrast, encapsulated sensors maintained the ability to respond to externally added NaF when RNase was present in the external sample (Fig. 4, D and E, and table S3). We noticed a decrease in mean GFP fluorescence at higher external NaF concentrations—a trend that we did not observe in non–RNase-exposed samples. We hypothesized that this difference was due to slightly higher degrees of membrane permeability from the addition of small amounts of glycerol in the RNase buffer (fig. S5), which could cause increased reaction poisoning with high fluoride concentrations. In addition, glycerol has been demonstrated to inhibit cell-free protein synthesis at high concentrations and may be causing slight effects even at the low concentrations used here (55). Nevertheless, all vesicle populations exhibited increased GFP expression in the presence of fluoride, demonstrating simultaneous permeation of fluoride into the vesicle interior and exclusion of RNase A from the cell-free reaction. These results indicate that encapsulation in bilayer membranes can sufficiently exclude RNase from the reaction environment, thereby protecting the cell-free sensor within a semipermeable compartment. The external addition of RNase A, demonstrated here, serves as a proof-of-concept step toward cell-free detection in complex environments, such as biological samples or highly contaminated environmental samples.

Vesicle-based sensors can incorporate a colorimetric readout and detect fluoride in real-world samples

Last, we wondered whether we could extend these results to conditions that would be more relevant for real-world environmental sensing. Although fluorescence is a common readout for many biological assays, GFP fluorescence inside vesicles is difficult to monitor with common equipment, particularly in nonlaboratory settings. In addition, with our vesicle-based construct, GFP fluorescence was too low to measure in vesicle populations in bulk, necessitating more sensitive microscopy analysis methods. To address this limitation, we coupled fluoride detection to an alternative reporter enzyme, catechol (2,3)-dioxygenase (C23DO) (3). In this system, the riboswitch ON state leads to the expression of C23DO, which catalyzes the conversion of its colorless substrate, catechol, to the yellow-colored 2-hydroxymuconate semialdehyde to generate a colorimetric response (Fig. 5A). Under bulk conditions, this construct exhibits a fast and robust response to fluoride, and the colorimetric output generated is clearly distinguishable by eye for both laboratory- and field-collected water samples (3).

To investigate whether this enzymatic reporter could function within our sensor vesicles, we encapsulated cell-free reactions with DNA coding for the riboswitch-C23DO construct and supplemented them with 1 mM catechol (Fig. 5A). We then titrated NaF into the outer solution and monitored color changes in each population of vesicles via changes in absorbance at 385 nm. In contrast to our GFP-based readout, signal amplification from the enzyme-regulated construct allowed us to assess absorbance changes in an entire population of vesicles rather than on a vesicle-by-vesicle basis. RNase A was added to the outer vesicle solution to control for any unencapsulated reactions caused by vesicle lysis and to assess the membrane’s ability to protect encapsulated sensor components. Sensor responses were significantly lower than those observed in bulk without RNase (fig. S6); however, despite this reduction in signal magnitude, sensors exhibited significantly increased absorbance in response to increasing fluoride concentrations (Fig. 5B). Vesicle sensors exhibited sensitivity to fluoride at a range of concentrations; here, readout time is an important consideration, as enzymatic signal amplification can lower the resolution of a dose response curve and even reduce overall signal at long reaction times (fig. S6) (3). Sensor responses to fluoride reach statistical significance at varying times depending on the external concentration of fluoride encountered (fig. S7). This variability in response as a function of both time-to-detection and fluoride concentration can enable these sensors to serve as dose-responsive or binary (yes/no) sensors depending on the time at which the response is assessed (Fig. 5C and fig. S7). Sensors could detect fluoride concentrations at levels consistent with the Environmental Protection Agency (EPA)’s maximum contaminant level for fluoride (~0.21 mM fluoride) and secondary contaminant standard (~0.1 mM) within 3 and 5 hours, respectively (Fig. 5C and fig. S7) (56). Responses were significant but noisy at 0.05 and 0.1 mM fluoride, however, indicating that these sensors would be most useful for detection of fluoride concentrations above the maximum contaminantt limit of 0.2 mM (Fig. 5C). By accounting for both the detection time and the final absorbance of these colorimetric vesicle sensors, health-relevant concentrations of fluoride can ultimately be detected in either a binary or a dose-dependent manner.

Last, we set out to test whether these sensors could be used to monitor fluoride concentrations in real-world samples. We collected water samples from Lake Michigan and the Evanston, IL municipal tap water supply and used each sample to prepare the vesicle outer solution supplemented with the small molecule components of the reaction. We supplemented these outer solutions with either 0 or 3 mM NaF and RNase A, which removes sensor activity from any ruptured vesicles, and added vesicles to the sample. We first assessed the performance of encapsulated GFP sensors and observed increased GFP expression in the presence of 3 mM NaF, with slightly higher levels of GFP expression in both lake and tap water samples compared to those incubated with laboratory-grade Milli-Q water (fig. S8). We next assessed the performance of encapsulated C23DO enzyme sensors and similarly observed increased colorimetric outputs, measured via absorption at 385 nm, in all populations of enzyme-expressing vesicles incubated with 3 mM NaF (Fig. 5D). We did not observe increases in absorbance in the absence of supplemented fluoride, with the exception of municipal tap water samples (fig. S6). Color changes were visible by eye in tubes containing vesicles, and changes in color inside of individual vesicles could be observed on the microscope when imaged through the eyepiece (Fig. 5E). By incorporating a colorimetric detection method as demonstrated here, these sensors could ultimately be used to detect fluoride in samples containing inhospitable components and in settings that prohibit the use of laboratory-based equipment. The results observed here highlight the feasibility of these vesicle-based sensors to detect environmentally relevant small molecules in complex, real-world samples, a step toward encapsulation to generate deployable cell-free sensors.

Chemicals

POPC and cholesterol were purchased from Avanti Polar Lipids Inc. OA, glycerol, sucrose, glucose, HPTS, BioUltra Mineral Oil, phosphate-buffered saline (PBS), bovine serum albumin (BSA), and NaF were purchased from MilliporeSigma. 1.8k PEO-b-PBD polymer was purchased from Polymer Source. OA647, calcein dye, and Hepes buffer were purchased from Thermo Fisher Scientific. RNase A was purchased from New England Biolabs.

Plasmid construction

Plasmids were assembled using Gibson assembly (New England Biolabs, catalog no. E2611S) and purified using a QIAGEN QIAfilter Midiprep Kit (QIAGEN, catalog no. 12143). Coding sequences of the plasmids consist of the crcB fluoride riboswitch from B. cereus regulating either superfolder GFP (pJBL3752) or C23DO (pJBL7025), all expressed under constitutive Anderson promoter J23119. Plasmid sequences are available on Addgene with accession numbers 128809 (pJBL3752) and 128810 (pJBL7025).

Cell-free reaction assembly

Cell-free extract and reactions were prepared according to established protocols (3, 39). Briefly, cell-free reactions were assembled by mixing cell extract, a reaction buffer containing the small molecules required for transcription and translation (nucleoside triphosphates, amino acids, buffering salts, crowding agents, and an energy source), and DNA templates and inducers at a ratio of approximately 30:30:40 (table S4). Sucrose was added to a final concentration of 200 mM to allow vesicle formation through the density-dependent double emulsion technique. Each reaction was prepared on ice to 16.5 μl of final volume in batches of seven. Reactions were prepared with 10 nM pJBL3752 (riboswitch-GFP plasmid) or pJBL7025 (riboswitch-enzyme plasmid) + 1 mM catechol. Reaction master mix was assembled and then added to DNA, inducers, sucrose, and water to a final volume of 16.5 μl per reaction aliquot. For reactions containing volume marker, reaction mix was supplemented with 1.4 μl of OA647. Preparation conditions were kept consistent between reactions, only varying NaF concentration or omitting DNA for extract-only controls.

Encapsulation of cell-free reactions

Encapsulated sensors were prepared via water-in-oil double emulsion methods. Lipid films were prepared by mixing amphiphiles (lipid, cholesterol, fatty acid, or polymers) in chloroform to a final amphiphile concentration of 25 mM at a volume of 200 μl. Films were dried onto the side of a glass vial under nitrogen gas and then placed in a vacuum oven overnight. A total of 200 μl of BioUltra mineral oil was added to lipid films and heated at 80°C for 30 min, followed by 10 s of vortexing to incorporate amphiphiles into the oil. Lipid/oil mixtures were cooled to room temperature and then placed on ice during cell-free reaction assembly. Cell-free reactions were prepared on ice as described above. Reactions were layered on top of lipid/oil mixture and then vortexed for 30 s to form an emulsion. Emulsions were incubated at 4°C for 5 min and then layered onto outer solution containing all small molecules required for transcription and translation, 100 mM Hepes buffer (pH 8), and 200 mM glucose, which is used to create a density gradient with the encapsulated sucrose. Samples were again incubated at 4°C for 5 min and then centrifuged for 15 min at 18,000 relative centrifugal force at 4°C to spin the denser, sucrose-containing vesicles into a pellet. Vesicle pellets were collected by pipette and placed into fresh Eppendorf tubes. Prepared vesicles were then added in 10 μl of aliquots to 20 μl of fresh outer solution supplemented with NaF, certain water samples, and/or RNase A (final concentration, 5 μg/ml). Osmolarity of NaF stock solution was adjusted to match that of the outer solution by adding glucose to minimize osmotic effects on vesicles.

Cell-free protein expression

For bulk assays, unencapsulated reactions were prepared as described above and added to 384-well plates. Protein expression was monitored at 37°C in a SpectraMax i3x plate reader (Molecular Devices). GFP was monitored at an excitation of 485 nm and an emission of 510 nm. Catechol absorbance was monitored at 385 nm.

Encapsulated sensors with a colorimetric readout were monitored at an absorbance of 385 nm using a SpectraMax i3x plate reader at 37°C until expression reached a plateau, about 2.5 hours, after which samples were removed from plates and placed into Eppendorf tubes or microscopy chambers for imaging. Images of tubes and through the microscope eyepiece were taken using an iPhone 8. Absorbance measurements in the plate reader are reported relative to initial absorbance to control for slight differences in vesicle concentration between vesicle preparations.

Encapsulated sensors expressing GFP were incubated in outer solution for 6 hours at 37°C and then imaged on a Nikon Ti2 inverted microscope. Imaging chambers were blocked with BSA for 20 min and then triple rinsed with 766 mosM of PBS. Vesicles were added to equiosmolar PBS and allowed to settle for 5 min before imaging. Images were taken using differential interference contrast, GFP (excitation, 470 nm; emission, 525 nm) and Cy 5.5 (excitation, 650 nm; emission, 720 nm) filters under ×10 magnification, 20% laser intensity, and 1-s exposure. Images were analyzed using Nikon NIS-Elements AR software Advanced Analysis tool (44): Vesicles were selected using the OA647 channel. General analysis protocol was set with the following settings: preprocessing: local contrast; size, 105; power, 50%; threshold minimum, 393; smooth, 1×; clean, 1×; size minimum, 2 μm; return mean, GFP; mean, OA647; max, GFP.

Encapsulated dye assays

Statistical analysis

All graphing and statistical analysis were conducted in GraphPad Prism (60). Populations of vesicles were analyzed using one-way analysis of variance (ANOVA) with Tukey’s multiple comparisons test or two-way ANOVA and Dunnett’s multiple comparisons test and descriptive statistics. Significance is reported as follows: nonsignificant (ns), P > 0.05; *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001; ****P ≤ 0.0001. Symbols and error bars on line plots represent mean and SD, respectively.

This PDF file includes: