Ultrahigh Throughput Evolution of Tryptophan Synthase in Droplets via an Aptamer Sensor

Tryptophan synthase catalyzes the synthesis of a wide array of noncanonical amino acids and is an attractive target for directed evolution. Droplet microfluidics offers an ultrahigh throughput approach to directed evolution (up to 107 experiments per day), enabling the search for biocatalysts in wider regions of sequence space with reagent consumption minimized to the picoliter volume (per library member). While the majority of screening campaigns in this format on record relied on an optically active reaction product, a new assay is needed for tryptophan synthase. Tryptophan is not fluorogenic in the visible light spectrum and thus falls outside the scope of conventional droplet microfluidic readouts, which are incompatible with UV light detection at high throughput. Here, we engineer a tryptophan DNA aptamer into a sensor to quantitatively report on tryptophan production in droplets. The utility of the sensor was validated by identifying five-fold improved tryptophan synthases from ∼100,000 protein variants. More generally, this work establishes the use of DNA-aptamer sensors with a fluorogenic read-out in widening the scope of droplet microfluidic evolution.


■ INTRODUCTION
Directed evolution has enabled the search for proteins that are of immediate interest to the pharmaceutical and chemical industries. 1 The success of directed evolution is determined by (i) the starting point in sequence space, (ii) the throughput and library design to probe the surrounding fitness landscape, and (iii) the assay through which fitness is determined. 2 Microtiter plate screening (∼10 3 variants per day) is readily coupled to a direct measurement of the product, either through liquid/gas chromatography and/or mass spectrometry.Cellular or droplet-based ultrahigh throughput screening (uHTS, ∼10 5 −10 7 droplets per day) benefits from an increase in throughput but requires an optical signal of the reaction product. 3he optical signal needs to be recorded at a wavelength above 320 nm to be compatible with uHTS as the lowest reported wavelength in fluorescence-activated droplet sorting (FADS) is 375 nm 4 and commercial fluorescence-activated cell sorting (FACS) systems are generally capped at 320 nm using UV lasers.Therefore, directed evolution in single cells or droplets has seen the most success using fluorogenic proxy molecules: these small molecules are synthesized with a scissile bond connected to a fluorophore.Upon enzymatic cleavage, the leaving group becomes fluorescent and enables sorting by FADS sensitively (e.g., >3000 molecules fluorescein per droplet) and rapidly (>kHz). 5These substrates are, however, hard to synthesize and often do not resemble substrates relevant for industrial biocatalysis, e.g., respective chemical challenges are not reflected in activated model substrates.−9 Nevertheless, the bulk of analytes are currently not amenable to any optical assay and remain beyond the reach of uHTS, although mass-activated, dielectrophoretic, and NMR sorting are emerging as a future alternative 10−12 (albeit at the price of lower throughput).Simple, fast and modular screening approaches in droplets addressing nonsurrogate substrates are highly sought-after and would allow the implementation of a wider spectrum of target reactions in uHTS format for enzyme evolution pipelines.
Engineering protein or oligonucleotide sensors to detect small molecules of interest, i.e., the reaction product, is an alternative strategy to obviate the need for fluorogenic substrates.Protein sensors require engineering a conditional inactive state that can be reversed by binding a molecule of interest. 13For example, a protein sensor was engineered to detect the redox state of NAD, thus enabling the screening of dehydrogenases on nonfluorogenic substrates in droplets by fluorescent activated cell sorting (FACS). 14−17 Numerous methods exist to engineer the resulting binder into a sensor suitable for uHTS: RNA-based aptamers can be engineered into a sensor through fusion with the "Spinach" probe, 14 which adapts its secondary structure when the original aptamer binds the target analyte, rendering it fluorescent. 18,19For DNA-based aptamers, a sensor can be engineered by fluorescently labeling the aptamer while designing a complementary strand with a quencher. 20When the target analyte is present, the complementary strand is displaced in favor of the target analyte, resulting in a fluorescence increase.Recently, the Heemstra group was able to detect unlabeled L-tyrosinamide in droplets using this DNA-based sensor design for enantiopurity analysis, 21 setting the scene for future development of labelfree assays for evaluating and sorting enzyme variants in droplets.
As such, DNA aptamers are convenient to use and require only synthesis of the evolved aptamer with a fluorophore and a complementary strand with a quencher to create a sensor.To utilize DNA aptamers in uHTS, one must (i) find/evolve an aptamer for a small molecule of interest; (ii) ensure that the aptamer does not bind the substrate, i.e., is specific for the product; (iii) manipulate the equilibrium of complementary strand and product binding to create a sensor; (iv) encapsulate the sensor in single or double emulsion droplets with substrates and unique enzyme variants, and sort by FADS or FACS respectively.
Here, we develop an aptamer sensor from an existing DNA aptamer for L-tryptophan (Trp) (Figure 1).We demonstrate the integration of the Trp aptamer sensor for the uHT evolution of the tryptophan synthase β-subunit from Pyrococcus furiosus (from here on referred to as TrpB), which condenses L-serine (Ser) and indole to produce Trp.TrpB has a relaxed substrate scope and has evolved to produce a wide range of noncanonical amino acids (NCAAs) of industrial interest with good yield. 22However, Trp (and its derivatives) cannot be assayed by conventional fluorescence or absorbance-activated droplet sorting, 6 since any intrinsic change in optical signal is too small to be measured at high throughput in microfluidic devices.Assaying ∼100,000 TrpB variants in a single day enabled us to isolate TrpB A9 , which accumulated four mutations in a single round to make it five-fold more catalytically efficient than wild-type TrpB and establish the utility of DNA aptamer sensors for uHTS.

■ RESULTS
From Aptamer to Sensor: Concentration-Dependent Fluorescence Change upon Trp Binding.The starting point for this work was an existing aptamer against the reaction product of TrpB with a K d of 1.8 μM for Trp. 23For the Trp aptamer to work in conjunction with droplet-based screening, a Trp-concentration-dependent fluorescent signal is necessary, i.e., a sensor.To create the aptamer sensor, a fluorescently labeled aptamer is supplemented with a complementary strand (CS) with a quencher, which is released in favor of the analyte (Figure 2A). 20In this case, the secondary structure prediction of the Trp aptamer (Figure 2A) suggests the presence of a stem loop (estimated to exist with >99% likelihood). 24When the Trp aptamer is bound to a CS, the stem loop would be unable to hybridize, disrupting secondary structure formation, which was thought to disable the ability of aptamers to bind Trp.An ideal equilibrium would have the complementary strand with a quencher (CS) bind the fluorescent aptamer at saturation when no target analyte is present, minimizing background, while readily being displaced with increasing concentrations of Trp so that a fluorescent signal emerges.
The equilibrium between the aptamer's affinity for Trp and the aptamer's affinity for the CS is most easily optimized by altering the length of CS. 20 Whereas the affinity for the target analyte is a direct outcome of SELEX, the affinity of the CS can be systematically modified based on the dependence of K D on its complementary length to the aptamer (the longer the complementary region, the more stable the duplex) and the entropic assistance of intramolecularity.A panel of CSs was synthesized, altering the length of the CS between 9 and 15 nucleotides (Figure 2B).Each of the CS contained a 3-prime Black Hole Quencher 2 (BHQ2) that would, after annealing, be in close proximity to the 5-prime cyanine-5 (Cy5) linked to the Trp aptamer.
The aptamer was hybridized with an excess of each CS in separate reactions, after which Trp was titrated to the solution (Figure 2C,D).The aptamer to which no CS was added displayed as expected the highest fluorescence, which was unaffected by the presence of Trp.The shortest CS, CS-9 (complementary strand of nine base pair length), lowered the fluorescence by 50% in the absence of Trp, indicating that 50% of the aptamer was still unbound in solution and 50% of the aptamer was bound.The length of the CS lowered the background systematically, consistent with the idea of making the equilibrium more favorable with increasing base-pairing interactions.CS-15 reduced the background to zero, quenching all of the Cy5 fluorescence and highlighting the efficiency of the quenching method.CS-15, however, created a duplex that was too stable to be affected by the presence of Trp, blocking the Trp binding site and rendering it useless as a sensor.Finally, CS-10 showed a concentration dependence in titration curves (Figure 2C,D).The equilibrium favors binding of CS-10 in the absence of Trp, while CS-10 gets readily displaced in favor of Trp when present.
The background fluorescence of the Trp aptamer:CS10 duplex was further reduced by increasing the concentration of the duplex (to 0.5 μM, Supplementary Figure 2A) and by increasing the stoichiometry of aptamer:CS10 (2.5-fold excess, or 0.5 μM:1.25 μM Trp aptamer:CS10, Supplementary Figure 2B).Under these conditions, we found the CS-10 to have the highest signal-to-noise ratio of all tested aptamer:CS complexes, which was ∼6-fold when 5 mM Trp was present (Supplementary Figure 2C).The K sens of the Trp aptamer sensor was determined to be 3.3 mM [2.8−4.6 mM 95% credible region], which is an order of magnitude weaker than the K d of the original aptamer for Trp (1.8 μM) (Supplementary Figure 2D).The drop in K sens compared to the K d of the original aptamer was also observed in the similarly constructed aptamer sensor for L-tyrosinamide. 20,25hile the K sens could be further improved by increasing the affinity of the aptamer for Trp, mM production of Trp in a droplet is already within reach considering the activity and expression levels of TrpB.Indeed, the expected concentrations of enzyme in a droplet from a single cell are estimated to range between 1 and 10 μM of enzyme, 26 so that no more than 1000 turnovers are necessary to create the mM quantities of Trp− well within reach of evolved TrpB variants. 22pecificity of the Trp Sensor and Compatibility with TrpB in Plates and Droplets.The hybridized Trp sensor was combined with 5 mM of Trp, D-tryptophan, or similar amino acids (Figure 3A).The Trp sensor is specific for Trp and did not dehybridize when similar amino acids such as L- phenylalanine were present.The aptamer sensor (much like the aptamer) is, however, sensitive to D-tryptophan.This is not a problem for TrpB evolution, as TrpB can convert L-serine and the corresponding indole with retention of enantiopurity. 27ext, the compatibility of the Trp sensor with the enzymatic production of Trp was probed.Three variants of TrpB were purified: TrpB K82A , TrpB WT , and TrpB 7E6 .TrpB K82A functions as a catalytically inactive control as the K82 residue connects the pyridoxal phosphate cofactor to the protein backbone and is mutated to an alanine.The previously evolved TrpB 7E6 (in microtiter plates) functions as a high-activity comparison. 28rst, both TrpB WT and TrpB 7E6 were assessed for their compatibility with Trp aptamer buffer (TAB; 10 mM Na 2 HPO 4 , 2 mM KH 2 PO 4 , 2.7 mM KCL, 5 mM MgCl 2 , 500 mM NaCL, pH 7.4), set up to stabilize the secondary structure of the aptamer by high ionic strength and high magnesium concentrations (Supplementary Figure 3).The sensor is sensitive to changes in the buffer components (requiring 1−10 mM Mg 2+ and 100−1000 mM NaCl, Supplementary Figure 4), suggesting that reaction buffer conditions should be tailored for the aptamer sensor. 23In TAB, both variants complete turnover of 5 mM of indole with Ser to produce 5 mM Trp (chosen as a benchmark as it is well above the K sens of the sensor, 3.3 Supplementary Figure 2D) in a matter of ∼6 h.For comparison, in a KP i buffer (in which TrpB 7E6 was evolved), 28 the enzymes require <1 h for complete turnover.
The purified proteins were combined with Ser, indole, and the Trp sensor in the microtiter plate format and left to react for ∼72 h to probe the stability of the sensor under reaction conditions (Figure 3B).In the presence of TrpB WT or TrpB 7E6 a clear increase in fluorescence suggests that the production of Trp can be monitored, compared to a ∼4.5-fold smaller fluorescence change in TrpB K82A containing wells that do not produce Trp.The TrpB K82A control highlights the specificity of the aptamer for Trp over its substrates and any tryptophan on the surface of TrpB, reducing the dynamic range compared to buffer-only conditions only marginally from ∼6 fold to ∼4.5 fold.
Using the same concentrations, TrpB variants, substrates, and Trp sensors were combined in microfluidic droplets and left to incubate for 72 h (Figure 3C).Flow cytometric analysis of the separate populations of double emulsion droplets is shown in Supplementary Figure 5. Again, the Cy5 fluorescence of the sensor increased in droplets where active TrpB was present compared to TrpB K82A , which for TrpB 7E6 , mimicked the dynamic range observed in the microtiter plate-based format.The TrpB WT -containing droplets deviated from the plate-based results, which, although more active than TrpB K82A , did not completely turnover indole (Figure 3B).This could be explained by the mandatory addition of surfactant Tween-20 to stabilize double emulsions or the presence of an oil interface when performing the catalysis in droplets.Nevertheless, the aptamer is both selective and compatible with enzymatic catalysis of Trp by TrpB variants.
Enrichment of Active TrpB in the Plate-Based and Droplet-Based Format with the Trp Sensor.The Trp sensor is compatible with purified TrpB, but for directed evolution, TrpB must be expressed in single wells or single cells to screen a library of enzyme variants.E. coli contains endogenous tryptophanase (TnaA), which can degrade up to 5 mM of exogenous Trp to produce indole. 29To facilitate screening with E. coli, a tryptophanase-deficient cell strain was used, E. coli DE3 (C43) ΔTnaA, previously engineered in the Bernardt group. 30The removal of TnaA was shown not to affect the growth rate of E. coli and was, therefore, well suited for directed evolution experiments. 30irst, we probed the Trp sensor in a microtiter plate-based assay.TrpB 7E6 and negative control (plasmid without insert) were transformed and expressed in a 1 mL culture of E. coli DE3 (C43) ΔTnaA.Next, the fluorescence was measured after sequential addition of Ser, indole, and the aptamer sensor (Supplementary Figure 6).Each of the wells containing cells expressing TrpB 7E6 showed higher fluorescence than those containing cells expressing no TrpB.There was a significant difference between fluorescence measured in wells containing cells with empty vector or cells expressing TrpB 7E6 , respectively (paired t-test, P value = <0.0001,n = 32).As such, the Trp sensor was able to localize TrpB activity expressed from a 1 mL culture.
Given the potential of noncanonical amino acids as precursors to pharmaceuticals, we probed whether the Trp sensor could be used to screen for value-adding, Trp derivatives (Supplementary Figure 7).We found that the sensor is versatile enough to distinguish both 5-fluoro-Ltryptophan and 5-methoxy-L-tryptophan from their respective precursors.Given the more limited solubility of 5-methoxyindole, we used 5-fluoroindole as the substrate for testing the enrichment of the TrpB 7E6 variant in droplets.
Both positive and negative controls were transformed into E. coli DE3 (C43) ΔTnaA again, this time aiming to encapsulate single cells in droplets with 5-fluoroindole as the substrate.Before encapsulation, cells transformed with an empty vector control were mixed with TrpB 7E6 transformed cells in a 100:1 ratio.After incubation, the droplets were sorted via FACS, and the genotype was recovered from both the sorted and unsorted fractions (Figure 4).The recovered genotype was transformed into E. coli DE3 (C43) ΔTnaA, picking individual colonies for a plate-based rescreen using the aptamer sensor (Figure 4).As predicted, the unsorted fraction contained only 2% TrpB 7E6 , closely matching the intended ratio.The sorted fractions, containing the top 0.7% and 0.1% fluorescent droplets, however, contained 36% and 45% TrpB 7E6 variants after sorting (and confirmed by sequencing), which constitutes a ∼40× enrichment of the originally intended ratio of active and empty vector control.Having shown enrichment for active TrpB in both plate-and droplet-based formats, we started evolving TrpB.
Evolution of TrpB.The first round of evolution from the starting point, TrpB WT , was performed in plates using the aptamer sensor, focusing on finding an activating and/or strongly expressing variant that would benefit the sensitivity of screening in droplets.An error-prone library was prepared with 2.2 ± 1.4 mutations per gene.Out of ∼200 variants, the top variant, TrpB B10 was isolated, sequenced, and expressed.The variant contained one mutation, F176L, which increased expression levels of TrpB from 50 to 135 mg/L, while additionally moderately increasing the number of TTNs and catalytic efficiency (Table 1; Supplementary Figure 10).Combined, we deemed the TrpB B10 variant to be well-suited as a template for directed evolution in a droplet-based screen.
An error-prone library of TrpB B10 was prepared with 4.5 ± 2.5 mutations per gene.For the droplet screen, the mutation frequency was increased as the increase in screening throughput was thought to counteract the expected decrease in expected functional variants. 31With this high mutational rate, we set out to enrich for variants that combine 4−5 new, activating, or neutral mutations per round.Single cells harboring unique TrpB variants were encapsulated with 5fluoroindole, Ser, and lysis reagents, incubated, encapsulated again, and sorted (Supplementary Figure 8).TrpB B10 was expected to turn over 5 mM of product for a maximal signal during incubation (Supplementary Figure 3).The top 5.7% of fluorescent double emulsion droplets were sorted from a total of 234,000 droplets, and the genotype was recovered and rescreened in plate-based format.We observed an average activity increase of 32% in the sorted fraction and a 44% increase in activity for the top 25% of rescreened wells (M sorted = 3939, Q3 sorted = 5111, compared to that of the unsorted library M unsorted = 2988, Q3 unsorted = 3548; Supplementary Figure 9).

■ DISCUSSION
Combining microfluidics with the versatility of SELEX to create binders for small molecules enables screening for improved enzymatic production of any product at ultrahigh throughput.A myriad of aptamers for small molecules have been evolved through SELEX, highlighting its versatility. 16,17he most important criterion of an aptamer used in enzyme screening is differential recognition of the product of interest over the substrate(s) from which it originates.To guarantee this specificity, negative selection rounds can be employed to counter-select for substrate binding, as exemplified by an aptamer for L-arginine with a 10,000:1 preference for the Lover the D-enantiomer. 32Here, we chose to use an existing aptamer for Trp, 23 which showed minimal binding of the substrates from which Trp is assembled: indole and Ser.Tryptophan-producing enzymes are interesting targets for directed evolution as derivatives of Tryptophan−noncanonical amino acids are found in 12% of the 200 top-grossing pharmaceuticals, 33,34 and are notoriously difficult to synthesize using conventional synthetic chemistry.
After a specific aptamer for the product of interest is obtained, the aptamer is engineered into a sensor.The Heemstra group has developed a straightforward approach, requiring the synthesis of <10 complementary strands with a quencher, complexing it, and titrating the product of interest to check for signal-to-noise ratio. 20Here, we applied the same strategy initially developed for a L-tyrosinamide aptamer and applied it to a Trp aptamer.We found the strategy to be generalizable: our optimal CS was only one base pair longer than the CS for the L-tyrosinamide aptamer, whereas the optimal stoichiometry and concentration of aptamer to CS were found to be identical.The resulting pipeline simplifies the screening approach compared to the RNA-aptamers-indroplets (RAPID) screening approach. 19First, for DNAaptamer sensors, there is no need for in silico optimization to predict the best aptamer-Spinach hybrids.Second, RNA is inherently less stable and requires IVT and subsequent purification to be produced.DNA aptamers can be readily supplied by commercial manufacturers as fluorescent singlestrand oligonucleotides and annealed with complementary quenching strands to create a stable duplex.Lastly, the use of DMHBI, which complexes with the folded Spinach aptamer for fluorescence, is not required as the aptamer itself is already labeled with fluorescent dye.
Whereas the Heemstra group used the aptamer sensor primarily to differentiate between L-tyrosinamide and Dtyrosinamide, here, we applied the concept to screen for improved enzyme variants in droplets.To make the screening  approach more available to the wider user base, the concept was developed in a double emulsion droplet format compatible with a commercial FACS instrument. 35(Double emulsion droplet sorting may require some optimization, which is well summarized in the Supporting Information of Brower et al. 36 ) We found, with good reproducibility, that the majority of our droplets (80%) were individual double emulsions droplets so that droplet fusions and oil droplets were of minimal concern.Furthermore, FACS allows for gating around the correct droplet population so that a more homogeneous sample is sorted, which provides an advantage over fluorescent sorting of droplets on chips, where larger droplets as a result of fusion cannot easily be practically excluded before sorting.As such, the setup provides experimental risk mitigation, as the double emulsions can be prepared using a setup with three pumps and a camera, after which the double emulsions can be stored in a buffer in standard tubes and sorted on FACS.Thus, allowing the screening of enzyme variants with little reagent consumption, without the need to invest in specialized on-chip sorting devices.Other nonlabeled droplet sorting techniques based on RAMAN or Mass Spectrometry (MS) function independent of aptamer maturation, and are not limited by the buffer sensitivity of aptamer folding. 10,37,38Although RAMAN and MS sorting do not require SELEX, and are therefore intrinsically more flexible, both methods require advanced setups and are currently only possible at low sorting frequencies (sorting 1−10 Hz, i.e., <9 × 10 5 droplets per day), whereas FACS can sort 10 4 −10 5 droplets per second and is therefore only limited by droplet formation (creating 10 3 −10 4 droplets per second).The aptamer sensor has lower sensitivity compared to MADS (mid μM).The k sens of the tryptophan sensor was 3.3 mM with a ∼6-fold dynamic range across experiments (Supplementary Figure 2D), with a lower limit of detection of 1 mM (Supplementary Figure 2C).Due to the solubility of Trp, the dynamic range is limited at 5 mM, where the fluorescent signal was shown to saturate (Supplementary Figure 2C).Increasing the sensitivity can be achieved by selecting higher affinity aptamers during the initial SELEX, which, as a result, can be complexed with a longer CS, so that background fluorescence is reduced and the k sens is lowered.
The Trp aptamer droplet-based uHTS approach was utilized to screen a library with a high mutational rate, in departure from the widespread practice to explore sequence space "one amino acid at a time". 39,40Increasing the number of mutations per gene lowers the chance of finding active variants and, therefore, requires increased throughput to find variants of interest. 31We obtained a TrpB variant that accumulated 4 mutations in a single round, making it 5-fold more active than WT TrpB at 37 °C.The improvement is similar to what was achieved in previous low-throughput mutagenesis studies of TrpB from the thermophile Pyrococcus furiosus.The first study, evolved TrpB for Trp over 3 rounds, which resulted in an 83fold increase in catalytic efficiency at 75 °C (averaging 2 mutations per round, and ∼28-fold activity increase per round). 41Next, TrpB 4D11 was evolved for β-methyl-Trp over 3 rounds, which resulted in a 7-fold improvement in β-methyl-Trp production over 2 rounds (averaging 1 new mutation and 3.5-fold increased activity per round). 42After the introduction of a rationally designed active site mutation, TrpB L161A was evolved for β-ethyl-Trp over 3 rounds, which resulted in a 6fold increase in TTN (averaging 1 mutation and 2-fold activity increase per round). 28Recently, the continuous directed evolution platform OrthoRep was utilized to evolve TrpB from Thermotoga maritima (TmTrpB) in ultrahigh throughput. 43Over 13−20 passages, or 40 days of continuous evolution, an average of 10 mutations per gene were introduced, increasing catalytic activity of TmTrpB by 22fold at 37 °C.The Trp-producing gene of the screening host was knocked out so that the proliferation of cells was directly dependent on the production of Trp by TmTrpB.While uncovering a plethora of new mutations (∼200) in TmTrpB that accelerate the evolution of promiscuous activities, it is unlikely that the strategy employed is adaptable to products that are not essential for cell growth.Even though our dynamic range is small (and practically capped at production of 5 mM of Trp) we envision increasing the stringency of selection by decreasing the incubation time so that increases in k cat are made evident over multiple rounds of mutagenesis with high mutational load.
Interestingly, the most active component mutation in TrpB A9 , N166D had been localized both in the first lowthroughput evolution of TrpB and in the continuous evolution of TmTrpB (corresponding to N167D). 41,43Additionally, the positions of component mutation F176L and E364V were also found as beneficial sites in TmTrpB (corresponding to mutations L177Q and K361E, respectively). 43Most of the mutations uncovered here (I102F, N166D, and F176L) are found in the COMM domain of TrpB. 44Mutations in this domain were shown to increase the activity of TrpB by mimicking allosteric activity by TrpA�resulting in a closed conformation of TrpB, changing the rate-limiting step of the catalytic cycle. 45Indeed, the mutations uncovered are likely to act through a general mechanism outside of the active site, as they not only improve activity on 5-fluoroindole but also increase activities for multiple substrates of TrpB.
Considering the abundance of negative epistasis (estimated to affect ∼50% of pairwise interactions), 46 it is interesting to observe that the component mutations were almost perfectly additive.Although screening with a high mutational rate does not preclude observing activating variants with hitchhiking mutations, which negatively affect overall activity, it can select for combinations of activity-decreasing mutations, which together form a more active enzyme through reciprocal sign epistasis.The synergy of multiple residues cannot be addressed through low-throughput mutagenesis, e.g., by assessments of mutability 47−49 (which will evaluate single positions iteratively by mutational scanning).Instead, simultaneous mutagenesis in multiple positions may give rise to residue combinations that break new ground and can be visualized as novel fitness peaks, even in a rugged landscape. 50aken together, the aptamer strategy employed here opens up possibilities to find fitness peaks that are often clouded through negative epistasis in faraway regions of sequence space.More importantly, we hope that the general availability of DNA aptamers and the ease of screening double emulsions in a flow cytometer will facilitate high-throughput screening at a scale that paves the way for new strategies in directed evolution campaigns, with minimal capital investment and low consumable expenditure due to miniaturized assay volumes.

■ CONCLUSIONS
We developed an assay system by repurposing a DNA aptamer to become a specific sensor for Trp and used this sensor to detect the formation of Trp in droplets.Monitoring Trp levels in droplets allowed us to evolve TrpB, catalyzing the condensation of indole derivatives and Ser to their respective Trp derivatives.This approach exemplifies a potentially general practical strategy for ultrahigh throughput evolution of biocatalysts without relying on an endogenous optical signal of the reaction product: formation of reaction product is made detectable by design of an aptamer sensor, and the accessibility of double emulsions makes droplet screening possible with a flow cytometer, together broadening the applicability of in vitro compartmentalized screening of biocatalysts to include reactions for which currently no ultrahigh throughput method exists.

■ MATERIALS AND METHODS
Cloning.TrpB WT and TrpB 7E6 from Pyrococcus furiosus were ordered as gene strings and next were subcloned to pRSF, a kind gift from J. D. Schnettler.TrpB variants were amplified using Q5 2× MM (NEB): 98 °C for 2 min, 98 °C for 10 s, 67 °C for 15 s, 72 °C for 1 min, and final extension at 72 °C for 5 min.The acceptor plasmid, pRSF, was amplified with Q5 2× MM: 98 °C for 2 min, 98 °C for 10 s, 68 °C for 15 s, 72 °C for 3 min, and final extension at 72 °C for 5 min.The TrpB genes were assembled in pRSF using Gibson assembly.TrpB K82A was prepared using site-directed mutagenesis (NEB), according to the protocol.Protein and DNA sequences are listed in Supplementary Table 1.
Error-prone libraries were prepared using the GeneMorph II Random Mutagenesis Kit (Agilent).The fragments were purified from agarose gels over silica by using agarose agarosedissolving buffer (Zymoclean Gel DNA Recovery, Zymo Research).The library fragments were cloned into the pRSF acceptor vector described above.The Gibson fragment was purified over silica columns and transformed to E. cloni 10F ELITE Electrocompetent cells (Lucigen) and plated on two 140 mm Petri dishes containing LB kan -agar.The next day, the colonies were scraped, and the plasmid was isolated using a Genejet plasmid miniprep kit (Qiagen).This purified plasmid stock was used for transformation BL21 (DE3) competent E. coli (NEB, 2527) and E. coli C43 DE3 ΔTnaA for plate screening experiments, and to E. coli C43 DE3 ΔTnaA for microfluidic experiments.
Protein Expression and Purification.Purified pRSF_TrpB, pRSF_TrpB 7E6 , and pRSF_TrpB K82A plasmids from a single colony were transformed to E. coli BL21 (DE3) and plated on LB-Agar with the appropriate selection marker(s).The next day, the colonies were scraped and used to inoculate 500 mL of LB and grown until an OD 600 of ∼0.4− 0.8.Cells were induced with 1 mM IPTG.Expression was done overnight at 25 °C and 200 rpm.
The eluted product was dialyzed with PD-10 columns according to the manufacturer's protocol (GE Healthcare) to Trp Aptamer buffer (TAB, (2 mM KH 2 PO 4 , 10 mM Na 2 HPO 4 , 2.7 mM KCl, 5 mM MgCl 2 , 500 mM NaCl, pH 7.4)) for all experiments and 50 mM KPi as a control when optimizing buffers.
Proteins were flash-frozen in liquid nitrogen, aliquoted, and stored at −80 °C.Thawed aliquots were only used once and discarded the same day.
Preparation of the Tryptophan Sensor.Tryptophan aptamer and complementary strands were diluted in Trp aptamer buffer (TAB (10 mM Na 2 HPO 4 , 2 mM KH 2 PO 4 , 2.7 mM KCL, 5 mM MgCl 2 , and 500 mM NaCL, pH 7.4)).The aptamer sensor was prepared by generating a 2× solution of 1 μM aptamer with 2.5 μM CS-10.For microfluidic experiments, a 4× solution of 2 μM aptamer with 5 μM CS-10 was prepared.The mixtures were incubated for 10 min at 90 °C, before being cooled down rapidly in a thermocycler to 4 °C after which the sensor was immediately placed on ice for 10 min.The sensor was placed at room temperature for 10 min or longer before use, and never kept overnight.
UV−Vis Spectroscopy.The aptamer sensor was combined 1:1 either with purified chemicals or with TrpB reaction mixtures in a final volume of 80 μL in low-binding microtiter plates (Corning, 3881).The increase in fluorescence was measured over 90 min (25 °C) on a Tecan infinite 200Pro using a 650 nM excitation wavelength and measuring the 700 nm emission wavelength.The bandwidth of excitation/ emission was 9/20 nm, respectively, with 25 flashes per measurement.The increase in fluorescence was typically saturated after 90 min incubation and taken as the end-point measurement reported.
Catalysis by TrpB.All analytical reactions were performed in 2 mL HPLC glass vials.A solution consisting typically of 20 mM L-serine (from 0.5 M stock in ddH 2 O), 20 mM Indole (from 0.5 M stock in DMSO, 4% DMSO final), and 200 μM PLP (from 10 mM stock in ddH 2 O) was combined with purified TrpB (in TAB).Reactions were incubated for 24 h in a 37 °C water bath.
Read-out with the Aptamer Sensor.The mixture was cooled to room temperature and combined 1:1 with the aptamer sensor before measuring for 90 min by UV−vis spectroscopy as described above.
Read-Out with HPLC.The mixture was quenched 1:1 with acetonitrile, and centrifuged at >14,000g for 10 min.The supernatant was analyzed by HPLC.HPLC was performed on an Agilent 1260 infinity II, equipped with a C-18 column (Mackerey-Nagel, 5 μm, ref 760.100.40) using acetonitrile (HPLC grade, Agilent) and ddH 2 O (0.1% (v/v) formic acid (FA)).The program was as follows: 0 min −100% ddH 2 O (0.1% (v/v) FA) 0% acetonitrile.Three min −40% ddH 2 O (0.1% (v/v) FA) 60% acetonitrile.Five min −0% ddH 2 O (0.1% (v/v) FA) 100% acetonitrile.Six min −0% ddH 2 O (0.1% (v/v) FA) 100% acetonitrile.Seven min −100% ddH 2 O (0.1% (v/v) FA) 0% acetonitrile All samples were analyzed at 277 nm, representing the isosbestic point between indole and tryptophan, allowing for the estimation of yield by comparing the area of the substrate peak to the areas of both the substrate and product peak combined. 27ichaelis−Menten Kinetics.Kinetic measurements of TrpB wt and its mutants were performed by monitoring 5fluoro-L-Trp formation in a plate reader (Thermo Scientific Varioskan Lux) over 20 min at 290 nm using Δ E290 =1.89 mM −1 •cm −1 .Measurements were taken every 10 s and slopes were normalized on background absorbance changes.Initial rates were calculated using a L-tryptophan standard curve as a proxy for 5-fluoro-tryptophan formation.All assays were conducted in a quartz 96-well plate at 37 °C using 10 μM TrpB and 20 μM PLP in TAB with a final concentration of 1% DMSO.For serine kinetics, 0−25 mM L-serine and 200 μM 5fluoroindole were used.For indole kinetics, 0−400 μM 5fluoroindole and 25 mM L-serine were used.All measurements were performed in three technical replicates.Data was fitted to the Michaelis−Menten equation using Origin 2018 (Origin Lab).
Plate Screening.Individual E. coli BL21(DE3) or E. coli C43 DE3 ΔTnaA colonies containing either a variant for rescreening or the TrpB wt and TrpB 7E6 controls were picked in 300 μL TB kan and grown overnight at 37 °C, ∼900 rpm in a plate shaker.The next day, 30 μL was taken to inoculate 920 μL of TB kan and grown for 3 h at 37 °C and ∼900 rpm.The plates were inoculated with 50 μL of TB kan containing 20 mM IPTG (1 mM final) and grown overnight at 20 °C.The next day, cells were palleted (4000 rpm, 10 min), and the supernatant was discarded.The pallet containing plates were frozen overnight at −20 °C.The pallets were thawed and lysed with 300 μL Lysis buffer (TAB, 1 mg/mL hen egg white lysozyme (HEWL), 200 μM PLP, 125 U/mL Benzonase) for 30 min at 37 °C.The plates were centrifuged for 30 min (4800 rpm).150 μL supernatant was mixed with 150 μL TAB containing L-serine (10 mM final), indole (5 mM final), and PLP (50 μM final) in 2 mL 96-deep well plates.The reaction mixture was incubated for 16 h, 37 °C, gently shaking.On the next day, the reaction mixture was allowed to cool down, and 30 μL was mixed with 30 μL 2× aptamer sensor stock solution in low-binding microtiter plates (Corning, 2881).The fluorescent values were determined as described under the section UV−vis spectroscopy, taken within one hour of incubation after which the aptamer sensor was thought to be degraded by endogenous nucleases from E. coli and the signal decreased.
Chip Fabrication.The microfluidic devices used for double emulsion generations (Supplementary Figure 1) were fabricated following standard photolithography and soft lithography procedures using high-resolution acetate masks and SU-8 photoresist patterning. 51,52hotolitography.The microfluidic chips were designed using AutoCAD (Autodesk) and printed out on a highresolution film photomask (Micro Lithography Services).The mask designs are published on dropbase (https:// openwetware.org/wiki/Dropbase:_Double-emulsion-02).The master molds of microfluidic devices were fabricated following standard hard lithography protocols.First, 15-μm-high microfluidic structures were patterned on 3 in.silicon wafers (Microchemicals) using high-resolution film masks and SU-8 2015 photoresist (Kayaku Advanced Materials) according to the guidelines of the manufacturer (SU-8 2000, Micro Chem).A MJB4 mask aligner (SU ̈SS MicroTec) was used to UV expose all of the SU-8 spin-coated wafers.The thickness of the structures (corresponding to the depth of channels in the final microfluidic devices) was confirmed by measurement with a Dektat stylus profilometer (Bruker).
Soft Lithography.The poly(dimethyl)siloxane (PDMS) mold was prepared by pouring a mixture of PDMS and curing agent (Sylgard 184 kit, Dow) in a ratio of 10:1 (w/w) over the Si wafer with master mold.The elastomer mixture was then cured at 65 degrees overnight, and the PDMS replica was cut from the Si wafer to release the microfluidic chip.Next, the inlet and outlet holes were punched by using a biopsy punch (1 mm).The chip was washed with dishwashing liquid and water, followed by ethanol (96%).The chip was dried using compressed air and cleaned additionally by the application and removal of Scotch Magic tape (3 M).The glass surface on which the chip was to be bonded was cleaned with Scotch Magig tape.The microfluidic chip and glass surface were both placed in a Femto Diener plasma surface treater.A vacuum was created before the chamber was flushed with oxygen.Next, the chips were treated with plasma.The chamber was quickly ventilated to retrieve the microfluidic chip and glass surface and bonded by rolling the chip on the glass surface.The chip was baked at 65 °C for 10 min.
Hydrophilic and Hydrophobic Treatment of the Microfluidic Chip.To apply the hydrophobic coating to the channels in the chip which creates the single emulsion (chip design https://openwetware.org/wiki/Dropbase:_Double-emulsion-02), the freshly baked microfluidic chips were flushed with 1% (v/v) trichloro (1H, 1H, 2H, 2H)perfluoroacetylsilane in HFE-7500 (3M-Novec).The chips were heated for 10 min at 75 °C before storage at room temperature, sealing of the inlets and outlets with scotch tape.To apply the hydrophobic coating to the channels in the chip which creates the double emulsion (chip design https://openwetware.org/wiki/Dropbase:_ Double-emulsion-02), the freshly backed microfluidic chips w e r e fi r s t fl u s h e d w i t h 0 . 2 w t % p o l y -(diallyldemethylammonium chloride) or pDADMAC in 0.5 M NaCl.After 10 min, the channels were flushed with 0.1 M NaCl, before flushing the chip immediately with 0.2 wt % poly(styrenesulfonate) or PSS in 0.5 M NaCl.After 10 min, the chip was flushed three times with ddH 2 O and kept submerged in ddH 2 O at 4 °C until use.
Microfluidics. Preparation of 2× Cell Solution.Two days prior to the w/o formation, the plasmid library or purified plasmid was transformed to E. coli C43 DE3 ΔTnaA and plated on LB kan -Agar plates.The next days, the cells were scraped with LB kan and diluted to OD 600 of 0.8 in 20 mL of LB kan in small Erlenmeyer flasks.The cells were induced with 1 mM IPTG final, and shaken overnight at 20 °C, 250 rpm.The next day, the OD 600 was measured.The cells were diluted to an OD 600 of 0.84.This assumes an OD 1 to contain 2 × 10 8 cells, and after 2× dilution with the substrate, sensor, and lysis solution in a droplet volume of 6 pL, will result in a final occupancy of λ = 0.5.The cells were pelleted (2000 g, 3 min) and washed 3× times with TAB.
Preparation of the 2× Substrate, Sensor, and Lysis Solution.A 4× aptamer stock solution (2 μM aptamer, 5 μM CS-10) was prepared as described above.A 4× substrate stock solution (40 mM L-Ser, 20 mM indole, 0.2 mM PLP) was prepared in a 2 mL HPLC glass vial and incubated for 2 min at 65 °C until all indole was dissolved after it was cooled down to room temperature.The 4× substrate stock solution was filtered through a 0.22 μM filter and mixed 1:1 with the 4× aptamer stock solution to the final 2× aptamer:substrate stock solution of 800 μL.Finally, 16 μL polymyxin (200 μM in 2× solution from 10 mM solution, 100 μM final in droplets) and 3.2 μL rlysozyme (4 μL/mL in 2× solution, 2 μL final in droplets) were added.
Preparation of Oil Solution.HFE-7500 was filtered through a 0.22 μM filter (Millex) and mixed with RAN 008-FluoroSurfactant (RAN Biotechnologies) to a final concentration of 1% RAN.The solution was again filtered through a 0.22 μM filter (Millex).
Preparation and Flushing of Collection Chambers.The collection chamber was a 0.5 mL Eppendorf tube, which was glued upside down on a glass plate.One mm diameter holes were punched with a biopsy punch both at the top and on the side of the tube, nearer to the bottom.Polyethylene Portex tubing (0.38 × 1.09 mm inner/outer diameter, SLS) or BOLA Tubing, PTFE (0.5 × 1.0 mm inner/outer diameter) were attached with cyanoacrylate glue (PR1500, Scotch-Weld).The collection chambers were flushed with 0.22 μm filtered HFE-7500 (Novec 3M) prior to use and filled with 1% RAN in HFE-7500.Droplets enter the collection chamber via the top, and the emulsion floats on top of the oil solution.Excess oil was discarded through the side channel at the bottom.For double emulsion formation, oil was pushed from the bottom channel, which pushed the emulsion out through the top channel to the microfluidic chip, where the droplets were encapsulated again.
Microfluidic Rig, Tubing, and Syringe Setup.The microfluidic rig was a setup with neMESYS syringe pumps (Cetoni), and a high-speed camera (Phantom Miro eX2), which was mounted on an inverted light microscope (Brunel Microscopes Ltd.).Glass syringes (Hamilton), either 250 or 500 μL were used to contain the cell and substrate solutions.Glass syringe (SGE) of 2500 μL was used for the oil-phase.The tubing consisted either of polyethylene Portex tubing (0.38 × 1.09 mm inner/outer diameter, SLS) or BOLA Tubing, PTFE (0.5 × 1.0 mm inner/outer diameter).For the polyethylene tubing, Hamilton glass syringes were fitted with 26 gauge needles (0.464 mm outer diameter).For the PTFE tubing, Hamilton glass syringes were fitted with 22 gauge needles (0.718 mm outer diameter).For both sorts of tubing, the SGE syringes were fitted with a gauge 25 (0.515 mm outer diameter) needle.
W/o Formation.The oil solution, 2× substrate, sensor, and lysis solution, and the 2× cell solution were pumped with flow rates of 800/30/30 μL/h, respectively.This resulted in droplet sizes of ∼6 pL.Droplets were collected in the collection chamber as described above with the tubing connected to the top of the collection chamber connected to the exit hole on the chip.Droplets were heat treated at 55 °C for 1 h to kill endogenous nuclease activity of E. coli upon cell lysis (pf TrpB is a thermophile and stable at 75 °C).Next, the temperature was lowered to 37 °C, and the droplets were incubated for a maximum of 72 h.
W/o/w Formation.TAB (1.5% Tween-20) and the w/o droplets were pumped with flow rates of 150/50 μL/h.A 250 μL glass syringe (Hamilton) containing 1% RAN (in HFE-7500) was fixed to the lower tubing of the collection chamber, and the droplets were pushed out from the top of the collection chamber to the inner inlet.A small piece of tubing connected to the outlet channel collected the double emulsion droplets and was placed in the middle of a 1.5 mL low binding tube (Eppendorf), which contained 1 mL of TAB (1.5% tween), with the double emulsions settling at the bottom.The double emulsions were kept at 4 °C until they were analyzed or sorted by flow cytometry/FACS, respectively.
Flow Cytometric Analysis and FACS of Double Emulsions.Double emulsions were resuspended with a 200 μL pipet prior to measuring.Flow cytometric analysis was carried out on a CytoFLEX S machine for double emulsions stored in TAB (1.5% tween).Aptamer-Cy5 fluorescence was quantified using 640 nm excitation, with a 660/10 nm bandpass filter.Flow cytometric sorting of double emulsions was performed on FACSAria III or FACSAria II instruments (BD) , with sorting into different low-binding tubes (Eppendorf) containing 100 μL of nuclease-free water according to aptamer fluorescent intensity.Prior to sorting, the double emulsions were often diluted ∼5 times in TAB (1.5% tween).Cy5 fluorescence was quantified using 633 nm excitation, with a 660/20 nm bandpass filter.Importantly, the nozzle size to smoothly accommodate the 22.5 μm double emulsion droplets was 130 μM.The selection threshold was deliberately set close to the level of background noise, to create relatively permissive screening conditions and avoid missing moderately improved catalysts in library screens. 49,53lasmid Recovery.Immediately after sorting, 200 μL of 1H,1H,2H,2H-perfluorooctanol (PFO) (Alfa Aesar) was added to the ∼150 μL double emulsions in nuclease-free water, vortexed, and centrifuged quickly for 10 s.The top layer was extracted and added to a DNA-low binding tube (Eppendorf).To the tube was added 4 μL of UltraPure Salmon Sperm DNA solution (Thermo Fisher) diluted 100× in nuclease-free water (final 2500× dilution) was added.The leftover PFO with small amounts of aqueous phase on top was extracted once with a 100 μL solution of UltraPure Salmon Sperm DNA solution (Thermo Fisher), diluted 2500x in nuclease-free water.To the 200 μL recovered DNA, 1000 μL of DNA binding buffer (Zymo) was added and purified over silica columns (Zymoclean Gel DNA Recovery, Zymo Research), eluting in minimal amounts of nuclease-free water.The resulting purified plasmids were transformed into E. cloni 10F ELITE Electrocompetent cells (Lucigen) and plated on two 140 mm Petri dishes containing LB kan -agar.The next day, the colonies were scraped, and the plasmid was isolated using a Genejet plasmid miniprep kit (Qiagen).This purified plasmid stock was used for transformation to BL21 (DE3) competent E. coli (NEB, 2527) for rescreening in plates.

Figure 1 .
Figure 1.Overview of the screening approach.(a) TrpB-expressing cells are encapsulated in single emulsion droplets with the aptamer sensor, the substrates (Ser, indole), and lysis reagents.(b) One�Lysis reagents will release TrpB in each droplet where a cell is present.Two�TrpB catalysis of Ser and indole to produce Trp.Three�The fluorescent aptamer is initially self-quenched, but once Trp is bound in favor of the quenching complementary strand, the sensor lights up and becomes fluorescent.The Trp concentration measured as a fluorescence signal is a function of the catalytic efficiency, stability, and expression strength of TrpB variants, and screening and selection can be carried out accordingly.(c) Droplets are encapsulated again into double emulsion droplets so that they are compatible with fluorescence-activated cell sorting on a commercial flow cytometer.(d) Genotype from the pool of highly fluorescent droplets is recovered, after which the enriched pool of active variants is rescreened in the microtiter plate-based format for single variants of interest.The chip design is shown in Supplementary Figure 1.

Figure 2 .
Figure 2. From aptamer to aptamer sensor.(A) Predicted loop structure of the aptamer without CS present (right) 24 contains a hybridized stem region.Binding of a CS with a quencher (left) puts the BHQ2 in the proximity of the Cy5 fluorescent group on the aptamer, quenching its fluorescence, and prohibits the formation of the stem loop.When enough Trp is present to tilt the equilibrium, the CS gets displaced in favor of Trp and the Cy5 fluorophore gets unquenched.(B) Trp aptamer and CS sequences with Cy5 and BHQ2 moieties, respectively, added during synthesis.(C) Titration of Trp to (hybridized) solutions of Trp aptamer without any CS (blue circles) and in the presence of CS-9 (red squares), CS-10 (green triangles), CS-11 (purple triangles), CS-12, (orange diamonds), and CS-15 (black circles).The gray area is enlarged in the second panel (on the right) to show the CS-11, CS-12, and CS-15 titration curves.Standard deviations derived from two technical repeats are shown.(D) Aptamer:CS10 duplex has the highest delta fluorescence in the presence of Trp.Experiments were performed in tryptophan aptamer buffer (TAB), pH 7.4 and 25 °C.

Figure 3 .
Figure 3. Specificity and compatibility of the Trp sensor with catalysis.(A) Trp sensor was hybridized and combined with 5 mM of each of the substrates shown in separate reactions, measuring the increase in fluorescence compared to when only TAB was added.(B) 10 mM Ser, 5 mM indole, 20 μM purified TrpB, and the aptamer sensor were combined and left to react at 37 °C for 72 h to assess the stability of the sensor.After 72 h, the mixtures were cooled down to RT and analyzed either by UV−vis (for wells containing the solution) or via flow cytometry (for double emulsions containing the solution).(C) Encapsulation of 20 μM TrpB, with 10 mM Ser and 5 mM indole, with the Trp sensor.The droplets were incubated for 72 h before being encapsulated again and prepared for flow cytometry.All raw data for the double emulsions are shown in Supplementary Figure 4.

Figure 4 .
Figure 4. Enrichment of TrpB 7E6 in double emulsions.Cells expressing either TrpB 7E6 or the vector without the insert were mixed in a 1:100 ratio and encapsulated with 5-fluoroindole, Ser, and lysis reagents.After 72 h of incubation, the droplets were encapsulated again and sorted via FACS.The recovered genotype from the unsorted fractions (a), top 0.7% (b), and top 0.1% (c) was transformed and rescreened in plates for 5-fluoro-Ltryptophan production using the aptamer sensor.The fluorescent intensity of all wells is shown in d.Additionally, the plates were sequenced to confirm the high fluorescent wells contain TrpB 7E6 , which are denoted as purple-colored dots, where no false negatives were observed.Double emulsion droplets were sorted on an ARIA II flow cytometer (see Materials and Methods).

a
Total turnover numbers with 20 mM 5-fluoroindole, 20 mM Ser, and 2 μM TrpB (max TTN = 10,000) after 14 h incubation at 37 °C.b Conversion of 5-fluoroindole with 20 mM 5-fluoroindole, 20 mM Ser, and 20 μM TrpB (max TTN = 1000) after 14 h incubation at 37 °C.c For TrpB V368I a K M could not be determined due to too low activity and too low signal in the assay used.d Catalytic efficiency (k cat /K M ) and changes with TrpB WT are reported for Ser. e Michaelis−Menten kinetics were obtained at 37 °C with 0−400 μM 5-fluoroindole, 0−25 mM Ser, and 10 μM TrpB.Standard deviation is shown from 2 to 3 technical repeats.

Table 1 .
Characterization of the Evolved TrpB Variants e