Droplet Microfluidics and Directed Evolution of Enzymes: An Intertwined Journey

Abstract Evolution is essential to the generation of complexity and ultimately life. It relies on the propagation of the properties, traits, and characteristics that allow an organism to survive in a challenging environment. It is evolution that shaped our world over about four billion years by slow and iterative adaptation. While natural evolution based on selection is slow and gradual, directed evolution allows the fast and streamlined optimization of a phenotype under selective conditions. The potential of directed evolution for the discovery and optimization of enzymes is mostly limited by the throughput of the tools and methods available for screening. Over the past twenty years, versatile tools based on droplet microfluidics have been developed to address the need for higher throughput. In this Review, we provide a chronological overview of the intertwined development of microfluidics droplet‐based compartmentalization methods and in vivo directed evolution of enzymes.


Introduction
Enzymes,naturesprivileged catalysts,were optimized for as pecific biological purpose and evolved over thousands of generations by natural selection. Lowering reaction barriers to selectively enable and accelerate certain reactions is akey characteristic of enzymes.B ut as the enzymes natural activities are often insufficient to meet the needs of mankind, artificial selection and screening have gained importance. Starting from the breeding of crops and domestication of animals for sustaining early populations,i tm atured to directed evolution in order to improve natural systems and introduce new-to-nature reactions for life sciences and other applications.
The2 018 Nobel Prize in Chemistry was awarded for efforts in the development of directed evolution to Frances H. Arnold for the directed evolution of enzymes,and George P. Smith and Sir Gregory P. Winter for the phage display of peptides and antibodies.Directed evolution makes it possible to alter and thus potentially improve biological activities by genetic means,a na pproach generally faster and with better control than natural selection. Accordingly,the study of these enzymes is of high importance to scientific advancement and holds great industrial potential, as it allows the evolution of alternative or new reaction pathways in astreamlined fashion. This approach thus provides environmentally friendly pathways to valorize enzymes as an alternative to the more traditional chemistry toolbox. [1] Applying directed evolution consists of three steps:1 )to iteratively mutate (create genetic diversity), 2) screen (optimize for ad esired property), and 3) choose (pick the best performing variant). If the protein of interest is well characterized, focused mutagenesis strategies can be implemented, followed by lower throughput screening. [2] In ap ioneering study,Arnold and co-workers highlighted the potential of directed evolution using subtilisin E. By screening about 4000 colonies,t hey evolved av ariant capable of hydrolyzing ap eptide substrate with 256-fold higher efficiency than wild-type in 60 %d imethylformamide (DMF). [3] Since then, numerous in vitro and in vivo studies based on focused libraries in microtiter plates (MTPs) have been reported. [4] Selected examples include 1) the directed evolution of sortase A to improve its robustness and activity by focused loop engineering and head-to-tail backbone cyclization, [5] 2) the directed evolution of enantiospecific enzymes, [6][7][8] 3) the directed evolution of P450 for various applications, [9][10][11][12] and, more recently,4)the directed evolution of ad en ovo designed retro-aldolase, [13] of am etalloenzyme for enantiospecific ester hydrolysis designed from short peptides, [14] and of am etalloenzyme for olefin metathesis using an expanded nitrobindin variant. [15] Lately,d irected evolution finds also increased use in the biotechnological field:f or example,t he process and enzyme engineering approach applied to galactose oxidase for the biocatalytic transformation of 5-hydroxymethylfurfural (HMF), avaluable building block in the synthesis of materials from renewable resources. [16] Apart from MTP assays,a nother mediumthroughput approach is the use of agar plate based screening assays,w hich was illustrated with the directed evolution of transaminases as biocatalysts for chiral amine synthesis. [17] Enclosing the enzymatic reaction within cells or immobilizing fluorescent products on the cell surface is yet another strategy to increase the throughput and was applied to several systems, such as the evolution of aP450 monooxygenase. [18] If the structure-activity relationships of the protein are poorly understood, more mutants may need to be screened to achieve atargeted phenotype.T his is often achieved through am ore thorough mutagenesis campaign of the protein and leads therefore to an exponential growth in the number of variants to be screened. [19] Let us consider an example whereby four positions are simultaneously randomized. Using conventional screening assays based on MTPs may require over 80 years (roughly 20 PhDs!) of manual screening and appreciable amounts of material such as screening buffers (> 100 L) and costly catalyst solutions (> 1L). [20] In contrast, the same screening using double emulsions and fluorescenceactivated cell sorting (FACS) could be performed in roughly aweek by asingle operator with substantially lower amounts of material ( Figure 1).
Additionally,t oe nable large screening efforts,o ptical readouts such as color,f luorescence,o rl uminescence are essential. In general, the industrially relevant target products lack readily detectable phenotypes.I ns uch cases,s ubstrate analogues with af luorescent, luminescent, or colorimetric readout that correlates with the enzyme activity need to be implemented.
One of the main requirements in directed evolution is linking the activity of atarget enzyme (i.e.the phenotype) to its genetic information (i.e.t he genotype), which is essential for screening,s election and ultimately evolution campaigns. [21,22] To address this challenge,d ifferent strategies, such as compartmentalizing the enzymatic reaction within/on cells or immobilizing fluorescent products on the cell surface, have been explored and are described in detail in other reviews. [23,24] These strategies opened the way to highthroughput analysis methods such as FACS.F ACSd evices have gained increasing interest since their initial development and the first instrument commercialization in the late 1960s-1970s. [25,26] Thed evelopment of microfluidic devices for fluorescence-based particle-or cell-sorting using negative dielectrophoresis (DEP) contributed to the early advancement of such technologies. [27] Other fluorescence-based methods such as fluorescent correlation spectroscopy (FSC) were developed around the same time for single-molecule detection and analysis in solution and were further optimized with applications in evolutionary biology. [28] Ariane Stucki holds aMaster's degree in EngineeringP hysics and aMinor in Biomedical Technologies from Ecole Polytechnique FØdØrale de Lausanne (EPFL). She is currently aPhD student in the Department of Biosystem Science and Engineering, ETH Zürich, in the Bioanalytics Group under the supervision of Prof. Dr.P etra Dittrich. Her projects focus on the development of droplet-based microfluidic tools for applications in chemistry and biology.  . Schematic representationo fthe screening effort per mutated position using an NNK library.AnNNK library at one position has 32 possible codons encoding for the twenty amino acids. This corresponds to ascreening effort of 94 colonies to achieve atheoretical library coverage of 95 %. This effort increases exponentiallyi ftwo or more positions are screened simultaneously.Screening four positions would require about 80 years, consideringthat eight 96-well plates are screened per week. In comparison,s creening the same library in double emulsionsu sing microfluidic tools would require about one week of work. [19] Approaches where the fluorescent product remains in the cell or is immobilized on the cell are compatible with highthroughput FACS but suffer from potential cross-contamination and are incompatible with certain substrates.I nv itro compartmentalization (IVC) in water-in-oil emulsions has emerged as an alternative to preserve the phenotypegenotype linkage. [24] IVC has attracted al ot of interest and has been developed in parallel to the advancement of research on directed evolution over the past 20 years. Surfactant-stabilized single (water-in-oil) or double (waterin-oil-in-water) emulsions (SEs and DEs,r espectively) constitute optimal compartments for directed evolution thanks to their long-term stability over ar ange of physicochemical factors including temperature,p He tc.M oreover,t he formation of such compartments using microfluidic devices yields monodisperse droplets and allows for more controlled encapsulation of reactants.
Directed evolution studies have directly benefitted from the development of droplet microfluidics,a llowing faster screening of larger libraries.I nt urn, the need for more specific and powerful tools for directed evolution has driven research in droplet microfluidics forward. In the last twenty years,e ngineering and biochemistry research groups have worked together to improve existing systems and develop new ones ( Figure 2). In this Review,w ep rovide ac hronological overview of the intertwined development of microfluidics droplet-based compartmentalization methods and in vivo directed evolution of enzymes.

Single Emulsions
Single-emulsion droplets are aqueous compartments surrounded by an oil phase.T he droplets can be stabilized using surfactants,i .e., amphiphilic molecules that arrange them- Figure 2. Milestones in the development of droplet microfluidics (top) and their applicationst odirected evolution (bottom) in the last twenty years. 1) Bulk production of single emulsions (SEs). [29] Directed evolution of aT aq DNA polymerase based on compartmentalized self-replication in SEs produced in bulk. [30] 2) Bulk production of double emulsions (DEs). [31] Directed evolution of E. coli surface-displayed serum paraoxonase 1( PON1) using DEs produced in bulk. [32] 3) On-chip production of SEs. [33,34] Directed evolution of ap hosphotriesterase through the encapsulation of E. coli expressing the enzyme on their surface in SEs produced on-chip. The SEs consist of agellable liquid and form gel beads following agelation step. The beads can be analyzed and sorted by FACS. [35] 4) On-chip sorting of SEs. [36,37] Directed evolution of aretro-aldolase using SEs formed on-chip and subsequent fluorescence-assisted droplet sorting (FADS) on-chip. [38] 5) On-chip formation of DEs followed by FACS sorting. [39,40] Directed evolution of amanganese-independent a-L-threofuranosyl nucleic acid (TNA) polymerase using DEs generated on-chip and subsequent sorting by FACS. [41] (Reprinted with permission (1) of the National Academy of Science USA. Copyright 2001;R eprinted (5) from ref. [41]). selves at the water/oil interface.M ethods of increasing complexity have been developed for the formation of such compartments,a llowing improved control over the droplet size,t he throughput, or the reagents encapsulation. Due to the external oil phase,w ater-in-oil (w/o) droplets are not compatible with commercially available FACS devices.T o overcome this challenge,m ethods for on-chip sorting have been implemented. Most recent devices have sorting throughputs of up to several kHz. [42] 2.1. Technology Advances I: Bulk Emulsification and Strategies for the Encapsulation and Immobilization of Reagents and Reaction Products Different methods are used for the production of w/o compartments.B ulk emulsification allows fast and simple formation of droplets,but has limited encapsulation efficiency and yields polydisperse droplets.B ulk emulsification techniques,s uch as stirring and emulsifier-based methods,w ere described before 1980. [43,29] Later studies focused on the characterization of the physical properties of emulsions produced with custom-made or commercially available homogenizers, [44][45][46] highlighting that droplets of sizes ranging from 0.1 to 100 mmi nd iameter can be produced.
Whole cells or genetic material can be encapsulated in w/o droplets.T he cell encapsulation follows the Poisson distribution and single-cell compartmentalization can be achieved by adjusting the dilution of the cell-containing solution. [47] In the early 1990s,emulsions could be produced at high throughput but were incompatible with analytical tools with similar throughput. To circumvent this challenge,other droplet-based strategies were developed to screen active variants with FACS devices.One of the first techniques to emerge consisted in coencapsulating an in vitro transcription and translation (ivTT) mixture with single microbeads,e ach displaying the gene encoding the protein of interest in w/o emulsions (Figure 3B). [48] In this study,antibodies bound to the streptavidincoated microbeads could immobilize the translated proteins. Upon translation, the emulsions were ruptured to retrieve the microbeads and, subsequently,i ncubated with horseradish peroxidase (HRP) which bound to the proteins of interest via al igand. In as econd step,t he beads were incubated with hydrogen peroxide and fluorescein tyramide,l eading to the fluorescent labeling of the bead. FACS sorting of the microbeads enabled the identification of ap rotein with high affinity towards the ligand used in the screen.
Another strategy enabling the use of FACS consists in generating droplets with agellable liquid in which genes and either encoded enzymes or whole cells can be encapsulated. Through acooling step,the droplets are converted into FACScompatible gel beads,i mmobilizing and compartmentalizing the genetic material ( Figure 3C). [49] Therelative permeability of gel beads favors the constant intake of growth medium or the addition of certain substrates at al ater time point, while retaining the cell microcolonies.Using this technique,Weaver et al. encapsulated mammalian, bacterial, and fungal single cells in agarose beads with diameters of 20 to 90 mm ( Figure 3D). [50] After an incubation step in the growth medium and astaining step with fluorescent markers for biomass,the cell colonies were analyzed by FACS.Inarelated study,Sahar et al. analyzed the properties of the encapsulated bacterial colonies. [51] Among others,t hey characterized the intracellular esterase activity of a P. aeruginosa cell population. This was achieved through the addition of afluorogenic substrate, 6-carboxyfluorescein-diacetate,t ot he gel beads followed by an incubation step.They additionally described the activity of the secreted enzyme elastase by encapsulating its fluorescently labeled substrate casein during droplet formation and determined the decrease in fluorescence caused by the leakage of the product out of the bead.

Applications I: In Vitro
Thef irst study using in vitro compartmentalization for applications in molecular evolution resulted from ac ollaboration between the Griffiths and Tawfik groups. [52] In this study,i nv itro compartmentalization (IVC) of as ingle gene encoding either aD NA-methyltransferase HaeIII or ad ihydrofolate reductase (DHFR) followed by ivTT led to the enrichment of an enzyme for DNAm ethylation ( Figure 4). M.HaeIII genes encoding HaeIII, and folA genes encoding DHFR, both containing as ite designed for methylation/ restriction by M.HaeIII, were encapsulated and tested for methylation efficiency. If M.HaeIII was present, the gene was methylated and was thus not digested in the subsequent digestion step with the endonuclease HaeIII. On the other hand, if DHFR was present, the gene was not methylated and was therefore digested by the HaeIII endonuclease.
Methylated HaeIII sites resistant to digestion were amplified using PCR and analyzed on an agarose gel. Model enrichment of al ibrary starting with 0.1 %M .HaeIII led to a5 00-fold enrichment in as ingle cycle.T he same approach was used in af ollow-up study to improve the sequence specificity.Amore active species was selected from ar andom mutagenesis library at three positions with % 3.3 10 7 variants.Remarkably,over only two rounds of screening, 11 variants with up to % 19-fold improvement were identified. All identified hits bore two mutations,w hereas the third position proved to be crucial for the methyltransferase activity and did not tolerate any other mutation. [53] With the aim of bringing the technology to the next level, single emulsions were used to evolve ribozymes for ab imolecular Diels-Alder reaction. In al arger evolution campaign consisting of four rounds of IVC,r ibozymes catalyzing the intermolecular Diels-Alder reaction between 9-anthracenylmethyl hexaethylene glycol (AHEG, 1)and biotin-maleimide (2)w ith multiple turnovers were evolved (Scheme 1). After four rounds of evolution, variants with ac atalytic efficiency k cat /(K m1 K m2 ) = 5.3 10 5 M À2 s À1 were identified. These artificial enzymes display efficiencies that are comparable to catalytic Diels-Alderase antibodies. [54] Using ac ustom-built homogenizer,P aegel and Joyce evolved RNAe nzymes with ligase activity,s electing enzymes that could resist inhibition by neomycin. Alibrary of 10 11 variants was evolved over five rounds to obtain mutants with better tolerance to neomycin and generally higher K m values. [55]

Applications II:InV ivo
Meanwhile,t he first in vivo applications using single emulsions were reported. Thes creening approach used for the first studies was based on compartmentalized selfreplication (CSR) ( Figure 5A). Thed irected evolution of TaqD NA polymerase was carried out in polydisperse emulsions generated by stirring. With this approach, Ghadessy et al. identified aT aq DNAp olymerase variant with elevenfold increased thermostability and av ariant with over 130-fold increased resistance to the inhibitor heparin (Figure 5B). [30] Other similar approaches of CSR in single emulsions involved the directed evolution of the same Taq polymerase for broader substrate scope and faster-cycling mutants (35-90-fold higher affinity for the primer,t wofold increase in extension rate). [57] To expand the technology to non-polymerase type enzymes,cooperative CSR was applied to evolve an ucleoside diphosphate kinase (NDK). NDK converted dNDPs to dNTPs which, in turn, could be used by apolymerase to replicate the genetic material. In this manner, only genes encoding active NDK were replicated, thus affording as traightforward approach to evolve simple cascade reactions. [30] Thes ystems described above rely mostly on self-replication. Retaining af luorescent signal on the encapsulated species itself is an essential feature to allow for FACS sorting. This was illustrated with yeast cells encapsulated in droplets. Alibrary of yeast cells with surface-displayed glucose oxidase (GOx) and horse radish peroxidase (HRP) was encapsulated and screened for the conversion of glucose to gluconolactone  (5) incubation with HRP, which binds to the proteins of interest. As econd incubation step with hydrogen peroxide and fluorescein tyramide labels the beads fluorescently and permits FACS sorting for the identification of ap rotein with high affinity towards the ligand used in the screen. [48] C) Micrograph of the encapsulation and growth of different microbial cells in gel microbeads: E. coli (1), S. cerevisiae (2), and M. xanthus (3). Scale bars:2 0mm. [49] D) Fluorescence-based biomassq uantification of yeast cells trapped in gel microbeads. [50] (Images reprinted with permission from (B) Wiley-WCH Verlag GmbH & Co KGaA, (C) the AmericanS ociety for Microbiology,( D) Springer Nature Limited.).
( Figure 6). Thehydrogen peroxide byproduct of this reaction was reduced by HRP,l eading to the generation of af luorescein tyramide radical which, in turn, reacted with atyrosine residue on the surface of the yeast cell. In this manner,t he yeast cells retained the fluorescent information and could be sorted after rupturing the emulsions.F rom al ibrary containing 10 5 variants,resulting from error-prone polymerase chain reactions (epPCR), av ariant with five mutations and a2 .7fold improvement in k cat was identified. [58] Similarly,GOx was evolved for different conditions,r esulting in twofold improved thermal stability compared to wild type (t 1/2 % 20 min at 60 8 8C) as well as afourfold and 5.8-fold improvement in k cat at pH 5.5 and pH 7.4 respectively. [59] Recently,G Ox was coupled to the yeast-enhanced green fluorescent protein (yGFP) to afford ac himera allowing the simultaneous detection of the protein expression level and the activity of the same enzyme.T his system led to a2 .5-fold enrichment of expressed, active variants and a2 .3-fold increase in V max in just one round of screening. [60]

Applications III:Encapsulated Microbeads
Ap rominent example involving microbeads consists of the directed evolution of an extremely efficient phosphotriesterase (PTE) using streptavidin-coated microbeads. [35] In this study,G riffiths and Tawfik used polystyrene microbeads displaying single genes anchored via ab iotin-streptavidin linkage.W ithin the w/o emulsions,m ultiple copies of PTE were produced by ivTT and anchored to the bead using an antibody.T he emulsions were then ruptured and the beads were re-encapsulated to add asoluble biotin-tagged substrate. Thec atalysis was performed inside the emulsions and the biotin-tagged product was retained on the bead. Subsequent rupture of the emulsions and labeling with afluorescent antiproduct antibody facilitated the sorting of active species. Relying on this approach, the authors identified avariant with  (6) either characterizedand/or encapsulatedf or another round of evolution. [52] (Reprinted with permission from Springer Nature Limited.). Scheme 1. The intermolecular Diels-Alderreaction between biotinmaleimide (1)and AHEG (2). The gene to be evolved is covalently attached to 2 and 3,b ut only active catalysts give products 3.The Diels-Alderproducts 3 are subsequently captured by streptavidincoated magnetic beads, allowing their downstream PCR enrichment. [56]  a k cat of 10 5 s À1 after six rounds of directed evolution from al ibrary of 3.4 10 7 variants.T his corresponds to a6 3-fold improvement over the wild type enzyme.This work paved the way for similar approaches such as the enrichment of an oxygen-tolerant [Fe-Fe] hydrogenase Ifrom C. pasteurianum (CpI) for the reduction of the fluorogenic compound C 12resazurin [61] and the mock enrichment of aw ild type HRP immobilized on microbeads. [62] Notable directed evolution efforts include the directed evolution of 1) a trans-acting Bartel class Il igase with up to 90-fold rate enhancement [63] and 2) asortase Afrom Staphylococcus aureus with a114-fold enhancement in k cat /K m . [64] Recently,P anke and co-workers used alginate beads for the directed evolution of the broad-spectrum amino acid racemase from Pichia pastoris (PpAAR) for the racemization of d-ornithine,aninteresting target for industrial applications. Starting from al ibrary with 1.2 10 7 variants,t hey observed an up to 2.7-fold k cat /K M improvement over wild type after three rounds of directed evolution. [65]

Technology Advances II:Microfluidics-Based Droplet Formation and Reduction of Cross Talk between Droplets
Then eed for higher droplet monodispersity and better control over the formation process led to the development of the first microfluidic chips for droplet production. [66,67] Compared to bulk emulsion methods,m icrofluidics-based methods require the fabrication of the device and its operation but allow for high-throughput encapsulation in monodispersed compartments.T he first study by Thorsen et al. introduced adevice with aT-shaped junction to generate emulsions at af requency of 20-80 Hz ( Figure 7B). [33] Similarly,A nna et al. displayed the controlled formation of droplets in aflow-focusing channel geometry for the production of emulsions with droplet diameters as small as 10 mm ( Figure 7C). [34] Similar channel geometries were used later to analyze reagent mixing inside droplets,spontaneous merging of droplets of different sizes,reagent addition to droplets,and droplet splitting. [68,69] Ford irected evolution, each droplet ideally contains one cell. However,c ell encapsulation using microfluidic devices follows the Poisson distribution, resulting in am ajority of empty droplets at low cell concentrations,t hus lowering the effective throughput. Yet, unlike bulk methods,s everal studies have highlighted the possibility of overcoming Poissonsd istribution limitations on am icrofluidic device.U sing particular channel geometries and hydrodynamic effects at high flow rates to order cells,v arious groups succeeded in yielding up to 80 %s ingle-cell-containing droplets (Figure 7D). [70,71] Another critical aspect of the compatibility of droplet microfluidics with directed evolution resides in the ability of the droplets to retain the substrate and product of the enzymatic reaction of interest. In their study,C ourtois et al. investigated the leakage of fluorescein-based substrates from droplets into the oil and succeeded in improving the retention to more than 18 hours by addition of bovine serum albumin (BSA). Thea uthors illustrated the versatility of their system by characterizing the enzymatic activity of alkaline phosphatase expressed by E. coli cells and distinguishing empty droplets from cell-containing droplets. [72] Ther etention of substrate and product in w/o emulsions can also be addressed with the use of gel beads as described earlier, which can be created as well by droplet microfluidics ( Figure 7E). [73]

Applications I: In Vitro and In Vivo
As early as 2013, Scanlon et al. presented the application of hydrogel emulsions produced on chip for the discovery of natural product based antibiotics.Arecombinant antibioticproducing microbe (Saccharomyces cerevisiae or E. coli)w as co-encapsulated with the pathogen (S. aureus)a nd af luorescent label for dead cells. [74] After incubation, the emulsions were ruptured and the cells were sorted using FACS,allowing the identification of yeast or bacteria with bactericidal properties.I namodel sort with ar atio 1:10 000 of positive control yeast (secreting the bacteriolytic enzyme lysostaphin and constitutively expressing yEGFP) and negative control yeast (ineffective against S. aureus, non-fluorescent), the The catalytic reaction leads to stained yeast cells which, after rupturing the emulsions, are amenable to FACS sorting. B) GOx converts glucose into gluconolactone (1) and the byproduct H 2 O 2 is reduced by the HRP to produce af luorescein tyramide radical 5 (2). The radical then reacts with atyrosine residue 6 on the surface of the yeast cell, leading to astained yeast cell 8 (3). [58] authors reported ac omplete enrichment over three sorting rounds.A pplying as imilar methodology, E. coli cells encapsulated in hydrogel emulsions were screened for pBAD promoter activity. [75] In this study,alibrary of single E. coli cells expressing GFP were encapsulated in hydrogel emulsions using amicrofluidic device.Depending on the promoter sequence,d ifferences in GFP expression allowed FACS sorting based on GFP fluorescence.A fter sorting and enzymatic digestion of the agarose,t he microcolonies were plated on agar plates and analyzed to find an averaged 1.25fold improvement in protein expression in one round of screening.Inarecent study,F ischlechner et al. described the formation of gel beads with apolyelectrolyte shell. This shell led to the retention of significantly smaller molecules,w ith am olecular weight cutoff 200-fold lower than gel beads reported previously.T hey reported the co-encapsulation of single E. coli cells expressing avariant of aphosphotriesterase and afluorogenic substrate.Subsequent lysis of the E. coli in the gel beads released the active enzyme catalyzing the hydrolysis of aphosphotriester to yield afluorescent product. Theb eads retained the fluorescent product and the beads containing the most active variants were selected using FACS at rates > 10 7 Hz. Avariant with almost twentyfold higher k cat / K M could be identified in as ingle round. [73] Similar approaches have been used to evolve production hosts for industrially relevant enzymes:i mproved overexpression and 1.3-fold higher secreted amounts of xylanase by P. pastoris in gel microdroplets was recently reported by Ma et al. [76]

Technology Advances III:O n-Chip Observation, Manipulation, and Sorting of Droplets
Most of the technological developments described so far were optimized for one-step processes and reactions.H owever,m any reactions require multiple steps where the addition of new reagents is required or the different reaction conditions are not compatible with each other. With the development of droplet-based microfluidics,more options for generation, fusion, control, and analysis of droplets have emerged ( Figure 8A). [68,77] An important advancement of droplet-producing microfluidic devices consists in the integration of electrodes generating an electric field across microfluidic channels.I n an early study,C habert et al. investigated the electrocoalescence of w/o pair droplets as atool for reagent addition. Using AC fields,they succeeded in displacing and merging droplets with adiameter of 600 mmunder static and flow conditions. [78] Another study established the high-throughput electrocoalescence of pair droplets in aP DMS microfluidic chip ( Figure 8B). [79] Twof low streams with droplets of different diameters (13-50 mm) merged into as ingle channel with downstream electrodes generating an electric field. As the droplet velocity is size-dependent, the size mismatch allowed the smaller and larger droplets to form pairs upon contact and led to the subsequent electrocoalescence as the pair passed through the electric field. Them ethod was illustrated by determining the k cat of an enzymatic reaction through the A) The use of microfluidic chips with channels forming either aT -junctiono raflow-focusing junction allows the formation of highly monodispersedroplets at high throughput. Specific channel geometries can be used to improve droplet mixing, splitting, or merging.B )Micrograph of the first T-junction design for droplet production at 20-80 Hz. [33] C) Micrograph of the first PDMS chip with flow-focusing junction for the formation of droplets as small as 10 mmindiameter. [34] D) Micrograph of the comparison between random bead encapsulation in droplets with improved bead encapsulation using cell alignment resulting from hydrodynamic interactions. Droplets containing two particles or more are highlighted by asquare while droplets encapsulating single particles are highlighted by acircle. Scale bars:1 00 mm. [70] E) Microfluidic production of droplets with ag ellable liquid for the formation of gel beads with asemipermeable polyelectrolyte shell (gel-shell beads, GSBs). The shell can be ruptured under basic conditions. [73] (Images reprinted with permission from (B) American Physical Society,(C) AIP Publishing LLC, (D) The Royal Society of Chemistry,(E) Springer Nature Limited.). encapsulation of b-galactosidase and its fluorogenic substrate, resorufin-b-D-galactopyranoside,inpair droplets.Pioneering studies led to the development of controlled reagent injection to droplets with higher throughput. Abate et al. proposed the use of picoinjectors to add reagents to droplets at frequencies of several thousand Hertz ( Figure 8C). [80] Recent developments for precise reagent delivery inside w/o droplets still involve electric fields and more complex systems such as the rupture of triple emulsions [83] or the use of at hree-phase flow. [84] Yet, the use of electrodes is not restricted to droplet merging but also allows the displacement of droplets. [85,37] In an innovative study,Ahn, Kerbage et al. reported on adroplet sorter based on the use of dielectrophoretic forces to direct droplets towards either side of amicrofluidic junction. [86] In parallel to the development of droplet manipulation, several groups focused on on-chip fluorescence detection for reaction monitoring in droplets. [87] In an early study,Dittrich et al. encapsulated ivTT mixture and red-shifted GFP-encoding (rsGFP) genes in w/o droplets and monitored the protein expression on-chip using fluorescence spectroscopy. [88] The combination of sensitive fluorescence detection and dielectrophoretic sorting led to the development of the first fluorescence-activated droplet sorting (FADS) device. [36] In this joint effort of the Weitz and Griffiths groups, E. coli cells expressing either b-galactosidase or an inactive variant were co-encapsulated with af luorogenic substrate.T he groups sorted the droplets based on enzymatic activity at ar ate of 300 Hz with al ow error rate.I nalater study by the same authors,asystem with three chips for droplet production, reagent addition via pair droplet fusion, and fluorescence intensity based sorting was used for the kinetic monitoring of in vitro translated laccase. [89] Decoupling the processes made it possible to handle droplets at different rates,7 000 Hz for droplet production and 3000 Hz for droplet merging.
Asimilar decoupled process was used to enrich an active variant of in vitro translated b-galactosidase. [90] There,t he droplets containing active variants were merged on-chip with an aqueous stream for easier retrieval of the genes.Similarly, Svahn and co-workers reported on the enrichment of ayeast strain based on its enzyme production using FADS.T hey achieved an enrichment close to the theoretical maximum and identified aclone with twofold increase in amylase production after as ingle round of screening ( Figure 8D). [81] In as imilar study,O stafe et al. used FADS to enrich cellulase-producing  [79] (C) Micrographofthe addition of reagent to pre-formed droplets using picoinjection. [80] (D) Micrographs of droplet production on-chip, followed by the droplet reinjection in asecond chip, using oil as spacer,a nd the droplet sorting by dielectrophoresis based on fluorescence detection. [81] (E) Micrograph of asorting junction designed for 5-ways sorting of droplets. Scale bar: 200 mm. [82] (Images reprinted with permissionf rom (B) AIP PublishingLLC, (C) National Academy of Science USA, (D) The Royal Society of Chemistry.I mage reprinted (E) from ref. [82]).
yeasts from an inactive cell population with an enrichment factor of up to 300-fold. [91] Recent improvements of FADS devices involve multiway sorting:F renzel et al. proposed achip allowing for droplet sorting in four outlets at amaximal throughput of 2-3 Hz. [92] Al ater study by Caen et al. described af ive-way sorting system with an almost 100-fold higher throughput ( Figure 8E). [82] Recently,amicrofluidic chip for the sorting of w/o droplets based on fluorescence lifetime was reported by Hasan et al. [93] Most of the examples described above for dielectrophoresis-based sorting rely on the use of fluorogenic substrates.H owever,r ecent studies propose label-free techniques for on-chip sorting.A lternatives such as interfacial tension based sorting to distinguish between live and dead cells [94] or intelligent image-based sorting capable of analyzing images and taking sorting decisions in real-time [95,96] offer interesting prospects.

Applications I: In Vitro
An initial effort involving am ultistep process was highlighted using the example of aC otA sporulation protein, al accase from Bacillus subtilis catalyzing the oxidation of various aromatic compounds using molecular oxygen as oxidant. Since the laccase assay was incompatible with the in vitro protein expression system, sequential addition of reagents at different time points was required. [89] Following this strategy,F allah-Araghi et al. developed acompletely in vitro platform for the screening of active lacZ genes encoding the enzyme b-galactosidase starting from single genes.I nt heir study,g enes were encapsulated before on-chip electro-coalescence and sorting. In am odel sort of active lacZ genes vs.i nactive lacZmut with ar atio of 1:100, they reached a5 02-fold enrichment in as ingle round of screening. [90] Building on these examples,Goto et al. reported ad evice for the encapsulation and sorting of nanoliter droplets and applied the method on the model screening of an isocitrate dehydrogenase (IDH) from Streptococcus mutans. Starting from al ibrary of 10 3 variants and two rounds of screening,t hey isolated av ariant with about threefold higher activity than wild type. [97]

Applications II:InV ivo
Te chnological progress led to up to 1000-fold faster screening and am illion-fold decrease in reagent costs as exemplified by the joint efforts of Abate,Baret, Griffiths,and Weitz. In their seminal study,t hey applied on-chip droplet generation and dielectrophoretic sorting for the highthroughput screening of HRP displayed on the surface of yeast cells.A fter sorting,t he droplets were ruptured, thus making the most active yeast cells readily available for the next round of mutagenesis and sorting.W ith this screening platform, they screened libraries with up to 10 7 variants and achieved an overall % sevenfold improvement in catalytic efficiency over nine rounds.T he highest catalytic efficiency reached was ca. 2.5 10 7 M À1 s À1 ,t hus approaching diffusionlimited efficiency (i.e.10 8 M À1 s À1 ). [98] Similar setups involving yeast encapsulated in single emulsions were used, for example,f or the evolution of thermostable xylanase with improved activities (up to 4.7-fold) [99] or the improvement of yeast cells as production hosts (twofold increase in a-amylase production). [81] One of the first examples expanding the repertoire to E. coli was reported in 2015 by Abate and co-workers.T hey mapped protein sequence-function relationships by combining microfluidics with next-generation sequencing, and analyzing both sorted and unsorted populations.S tarting from alibrary of 6 10 7 variants,they enhanced glycosidase activity at higher temperatures in asingle round of mutagenesis.T he deep mutational scanning revealed regions which might be crucial for glycosidase activity,b ut also highlighted known patterns with mutational tolerance which were in accordance with examples from the enzyme family. [100] Directed evolution is especially versatile if the initial catalytic activities are low,a sf or example for de novo designed biocatalysts.I na nother study,H ilvert and coworkers applied am icrofluidics-based screening coupled to FADS to evolve ar etro-aldolase by amine catalysis. [101] A previously designed retro-aldolase capable of cleaving ac arbon-carbon bond in an on-natural substrate ((AE)methodol) via an enzyme-bound Schiff base intermediate (11)s howed modest catalytic efficiency (k cat /K M = 0.19 M À1 s À1 )a nd enantioselectivity (ee = 33 %f or (S)-methodol). [13] It was used as the starting point for the directed evolution campaign and the authors were able to significantly improve the catalytic activity. E. coli cells expressing the protein of interest in the cytoplasm were encapsulated in w/o emulsions with lysis buffer to release the enzyme in the droplet and af luorogenic,c harged methodol derivative (9) (Figure 9). Six focused libraries with up to five simultaneously mutated residues were screened and avariant with almost 80fold increase in k cat was identified in as ingle round of screening.Strikingly,the same mutations were identified in an earlier study on the same enzyme.This previous study,relying on am edium-throughput screening campaign using MTPs, required five rounds of directed evolution to install these mutations.T he best performing mutant of the microfluidicsbased study was identified after only two rounds of evolution. It included ten mutations and exhibited a7 3-fold increase in k cat /K M and tenfold preference for (S)-methodol. [101] The kinetics of the catalytic system were further optimized to give > 10 9 rate enhancement, thus approaching Class Ia ldolase activities (natural enzymes catalyzing reversible carboncarbon bond-forming reactions) and accommodating aw ider substrate scope. [102] Thes ame group reported the isolation of an active cyclohexylamine oxidase (CHAO) identified from as ingle screening round of al ibrary with 10 7 variants.T hey remodeled the active site of CHAO, achieving up to 960-fold increase in catalytic efficiency thus approaching the wild type levels of activity for an on-natural substrate. [103]

Double Emulsions
Water-in-oil-in-water (w/o/w) double emulsions can overcome some of the challenges remaining with single emulsions, such as limited stability (e.g.,s hrinkage of droplets) and the Angewandte Chemie Reviews need for rapid on-chip analytical methods.A lthough their creation requires an additional emulsification step, [104] the handling of w/o/w double emulsions on-and off-chip offers intriguing advantages;i np articular, the compatibility with commercially available FACS sorting devices is noteworthy. [105,106]

Technology Advances I: Bulk Emulsification and Development of Compatible Assays
Much like for single emulsions,t he first methods developed for the formation of double emulsions were based on bulk emulsification using stirring or emulsifiers. [107] Double emulsions formed using these strategies are polydisperse and the inner aqueous phase compartment usually consists of multiple compartments (Figure 10 B). In an early study, Ficheux et al. scrutinized the stability vs.c oalescence of the inner aqueous compartments of double emulsions.T heir research highlighted that the double emulsionss tability is mostly affected by the surfactant type (water-o ro il-soluble) and concentration, and can vary on at imescale that ranges from minutes to months. [31] Following these findings,o ther research groups investigated the stability of double emulsions with different surfactants and composition of outer aqueous phases. [108,109] At wo-step emulsification process using aC ou-ette mixer for the formation of quasi-monodisperse double emulsions with multiple aqueous compartments was proposed by Goubault and co-workers. [110] More recent studies involving double emulsions generated in bulk focus on improving the compatibility of double emulsions with different screening methods.I nt heir study, Prodanovic et al. proposed af luorescent cascade assay in double emulsions for sorting enzyme libraries by FACS.T he assay allowed the screening of aglucose oxidase gene library with 10 4 mutants based on the hydrogen peroxide production with a5 0-200-fold enrichment factor. [111] To overcome the limitations imposed by the polydispersity of bulk double emulsions,M ae tal. improved the production method by using membrane extrusion, leading to the generation of more uniform double emulsions.T he advantages of this method were illustrated by enriching apopulation of E. coli cells with esterase activity more than 300-fold ( Figure 10 Ca nd Figure 10 D). Them ethod was further applied to the directed evolution of at hermophilic esterase AFEST,r esulting in at wofold improvement in catalytic activity as well as the identification of several mutants with k cat /K m values approaching diffusion-limited efficiency. [112] 3.1.

Applications:I nVitro and In Vivo
Thef irst application of double emulsions to directed evolution involved the model enrichment of the abovementioned FolA/M.HaeIII system. Positive w/o/w droplets containing FolA and the fluorescence marker FITC-BSA and negative w/o/w droplets containing M.HaeIII and BSA were produced separately and mixed in different ratios before sorting with ac ommercial FACS device.T he positive and negative droplets were mixed in a1:100 ratio and within one round of sorting a % 30-fold enrichment was observed. [113] Using the same technique,Griffiths and co-workers reported the first completely in vitro directed evolution campaign using double emulsions for the evolution of Ebg,aprotein of unknown function. Starting with negligible activity,t hey screened al ibrary of 2 10 6 members over four rounds of directed evolution and identified variants with b-galactosidase activity with at least 300-fold higher k cat /K M values compared to wild type Ebg. [114] Thef irst study involving in vivo directed evolution in double emulsions was based on E. coli surface-displayed serum paraoxonase 1( PON1). PON1 is am ammalian enzyme capable of hydrolyzing ab road range of substrates,i np articular the homocysteine thiolactone,a nd thereby eliminating toxic metabolites.I n at wo-step process using ah omogenizer,s ingle emulsions containing E. coli with surface-displayed PON1 were produced. Thesubstrate (13)and athiol-detecting dye (15)were then added via the oil phase,and asubsequent emulsification step led to the generation of double emulsions.P ON1 was evolved for the hydrolysis of thiobutyrlactones (TBLs, 13), ag enerally poor substrate of PON1 (k cat /K M = 75 M À1 s À1 ) (Figure 11 A). Starting from al ibrary of 10 6 mutants,t hree cycles of screening led to avariant with up to one hundredfold higher TBLase activity (k cat /K M = 10 4 M À1 s À1 ) ( Figure 11 B). [32] Av ariant of the same enzyme,r ePON1, was further investigated as atarget against nerve agents based on  (1). The cells are lysed in the droplets making the expressed retro-aldolase (orange) readily available and converting the aldol substrate (red) into af luorescent product (green) (2). Finally,the droplets are sorted on-chip by activating the sorting electrodes when the fluorescence signal exceeds acertain threshold( 3). B) Retro-aldolase-catalyzedc leavage of acharged methodol derivative (9)via an enzyme-bound Schiff base intermediate ( 11)yields af luorescent naphthaldehyde derivative (12) and acetone. The positive charge on the substrate/product ensures their retention in the droplets. [38] organophosphates.B ya pplying random and targeted mutagenesis,c oupled to high-throughput FACS screening and MTP assays,m utants capable of hydrolyzing cyclosarin with k cat /K M % 10 7 M À1 s À1 were identified. These findings were also applied to prophylactic studies involving mice,w here the identified hits exhibited considerable protection against al ethal dose of ac yclosarin derivative. [115] Further applications include the directed evolution of 1) b-glucosidase leading to at wofold increase in lactose specificity and catalytic turnover rates, [116] 2) the development of am odel protease with 1.6-fold increased resistance towards the inhibitor antipain dihydrochloride, [117] and 3) the screening of acellulase mutant library with the identification of variants with over 13-fold increased specific activity compared to wild type. [118] Prodanovic et al. highlighted the versatility of this screening platform by expanding this technology to in vivo encapsulated yeast in combination with av anadium bromoperoxidase coupled fluorescence assay (ViPer) to detect H 2 O 2 (Scheme 2). In the assay,H 2 O 2 was used by the bromoperoxidase to produce hypobromide,w hich reacted with af luorogenic probe to release fluorescent coumarin. Using this approach, a200-fold enrichment of active GOx was identified in as ingle screening round starting from al ibrary of % 10 4 variants. [111] Similarly,c ellulase activity was evolved to achieve a1 2-fold enrichment of the active variant in asingle round. [119] Figure 10. Bulk emulsification for the formation of double-emulsion droplets. A) Double-emulsion droplets formed by bulk emulsification are polydisperse and can contain multiple inner aqueous phase compartments. An additionalf iltration step can improve the sample homogeneity and lead to more reliable FACS sorting. B) Micrographofdouble-emulsion droplets resulting from bulk emulsification. [109] C) Micrograph of doubleemulsion droplets after membranee xtrusion. [112] D) Micrographofdouble emulsions encapsulating either E. coli cells containing aplasmid for the expression of esterase on the cell surface or negative E. coli cells. The use of af luorescein-based fluorogenic substrate allows the identification and sorting of double emulsionsw ith cells displaying esterase activity as shown in the FACS plot. [112] (Image reprinted with permission from (B) Elsevier.Images reprinted from (C) and (D) ref. [112]).

Technology Advances II:On-Chip Formation and Stability
Optimization for High-Throughput Sorting Producing double emulsions on-chip greatly improved the monodispersity and the control over the number of inner aqueous phase compartments. [120] Furthermore,i tg reatly improved the droplet sorting efficiencya nd throughput. However,t he production of double-emulsion droplets onchip is more challenging than the formation of single emulsions as it requires different surface-wetting properties for each emulsification step.T he microfluidic junctions need to be hydrophobic for the first w/o emulsion and hydrophilic for the second w/o/w emulsion. Va rious strategies have been investigated to address this challenge,such as decoupling the two emulsification steps,u sing coating solutions,o rb uilding the chip from different materials.I napioneer study, Okushima and co-workers proposed several design options allowing for either 1) decoupled emulsification steps in two quartz and Pyrex glass chips or 2) double emulsification on one single Pyrex glass chip.F or both designs,t he double emulsions were produced using T-junctions.T he requirements for different channel surface properties at the junctions were satisfied by coating the first junction hydrophobically with asilane-coupling agent. Using these devices,the authors produced monodisperse double emulsions of about 100 mmin diameter at arate of 22 Hz (Figure 12 B). [121,122] Another study reported the formation of double emulsions at higher throughput using glass microcapillaries.D roplets 10-50 mm in diameter were produced at rates ranging from 100-5000 Hz. Thea uthors further highlighted the potential of their device by controlling the size of the inner water droplet and oil shell of the double emulsion. [123] Then eed for rapid prototyping and simple microfabrication led the field towards the use of PDMS-based microfluidic devices.Inafirst study on PDMS surface modification using plasma polymerization, the authors achieved selective hydrophilic coating and subsequent formation of double emulsions with aT -junction. [39] Theformation of double emulsions with controlled oil shell thickness was reported by Abate et al. [40] A PDMS chip with two consecutive flow-focusing junctions was selectively coated using af low-confinement technique. [124] Similar devices with as tep structure at the second flowfocusing junction were developed to facilitate the second emulsification. [125,126] Due to the critical nature of the coating and the precision required for the wettability patterning, different strategies for coating or decoupling the two emulsification steps were developed and are described in detail elsewhere. [127,128] Double emulsions formed on-chip initially found applications in cell culturing and in vitro protein expressions (Figure 12 C). [47,129,130] Notably,Zhang and co-workers encapsulated E. coli in monodisperse double emulsions produced on two decoupled PDMS chips.They studied bacterial growth and protein expression by addition of the inducer in the outer aqueous phase and utilizing diffusion across the oil layer. [131] Theability to use FACS on double emulsions constituted an essential advance for improved compatibility of double emulsions in the context of directed-evolution studies (Figure 12 D). Efforts were therefore invested in studying the deformation of double emulsions in FACS devices and in identifying suitable surfactants to ensure droplet stability. [132,133] Figure 11. A) PON1 hydrolyzes gTBL (13)t othe corresponding thiol 14. N-(4-(7-diethylamino-4-methylcoumarin-3-yl)phenyl)maleimide (CPM, 15)r eacts with the free thiol to form afluorescent product (16). B) Developmentoft he fluorescence intensity over three rounds of enrichment. TBLase activity was determined in the crude lysate of the selected pool and normalizedt othe activity of wild type PON1 (wt). [32] (Figure reprinted with permission from (B) Elsevier.).

Applications:I nVitro and In Vivo
These technological advances led to the development of ap latform for single-cell and enzymatic activity-screening. Te rekhov and Smirnov et al. combined FACS sorting of double emulsions with downstream next-generation sequencing and liquid chromatography-mass spectrometry (LC-MS) analysis of secretome and proteome.I nt heir comprehensive study,t he authors used at wo-step on-chip emulsification process to perform enzyme screenings with different organisms.They succeeded in sorting active yeast cells displaying an enzyme on their membrane from an on-active population using af luorogenic substrate.S everal mixing ratios-up to 1:10 5 -were investigated, and the authors achieved maximum enrichments for the low dilutions and significant enrichment for the highest dilution. They further illustrated the potential of their platform for distinguishing between different enzymatic activities and between different levels of enzymatic activity.F inally,t he cell-to-cell interaction between different organisms was investigated using yeast and bacterial cells. [134] First studies highlighting the power of double emulsions include the efficient enrichment of active wild-type arylsulfatase from al ow-activity mutant. Enrichment factors of 800and 2500-fold, starting from populations of 0.1 %and 0.01 % active cells,r espectively,h ave been reported. [135] Using af luorescent reporter system which gave ap ositive signal upon full-length amplification of the template DNAb yt he target polymerase,L arsen et al. expanded polymerase func-tion to non-natural genetic polymers.A fter establishing the approach by enriching am odel engineered polymerase % 1200-fold, the screening method was applied to evolve am anganese-independent a-l-threofuranosyl nucleic acid (TNA) polymerase.I nm erely one round of selection, they identified am anganese-independentT NA polymerase with higher fidelity and % 14-fold improved activity. [41]

Latest Developments and Label-Free Methods
Recent progress on the microfluidic/technology side focus on optimizing existing tools and methods,a iming at am ore straightforward use and more reproducible results.N otably, Sukovitch, Kim, and co-authors proposed am ethod to simplify double-emulsion production while conserving the monodispersity.They coupled single-emulsion production onchip with asecond bulk emulsification step,circumventing the complex coating process required for double-emulsion chips. [136] New ways of delivering reagents inside single or double emulsions,mainly by adapting the surfactant type and concentration, have been characterized by several research groups. [137][138][139][140][141] In parallel, substantial advances have been achieved in expanding the screening capabilities of microfluidic platforms.I nagroundbreaking study,M aa nd coworkers reported adual-channel microfluidic droplet screening system (DMDS). This system enables the simultaneous sorting of w/o droplets according to two properties of atarget  [122] C) DE formation and encapsulation of cells for the growth of multicellular spheroids. [129] D) Micrographso f DEs containing discrete concentrations of fluorescentdye, and FACS plot displaying the discrimination between the different DE populations. [132] (Images reprinted with permission from (B) the Royal Society of Chemistry,(C) Springer Nature Limited. Image adapted (D) from ref. [132]). enzyme using two different fluorogenic substrates (Figure 13 A). Thee fficiency of the platform for the screening of complex enzymatic properties was illustrated with the directed evolution of ah ighly enantioselective esterase from Archeoglobus fulgidus (AFEST) (Figure 13 B).
After five rounds of evolution, av ariant with 700-fold improvement in enantioselectivity for (S)-profens was obtained. [142] In ar ecent study,B rower et al. introduced ac omprehensive FACS-based method to sort and isolate double-emulsion droplets produced on-chip.T heir method allows for the encapsulation of av ariety of mammalian cells and sorting at throughputs > 10 kHz while maintaining the w/ o/w droplets integrity,f ollowed by retrieval of genetic material. [143,144] Although fluorescence detection is still the gold standard for assaying enzymatic activities in droplets,r eports investigating other techniques have recently gained attention. These techniques give access to enzyme characteristics without requiring the use of af luorogenic substrate.Afirst example reported by Gielen et al. introduced am icrofluidic device for absorbance-activated droplet sorting (AADS). With this device,t he authors evolved ap henylalanine dehydrogenase over two rounds of screening and found av ariant with 4.5-fold increased activity and > 10 8 8Cincreased thermostability. [145] Similarly,p assive sorting strategies,s uch as sorting by interfacial tension, will allow novel types of assays,w here changes in droplet content translate into different droplet properties. [94] One of the most promising and widely applicable alternatives to fluorescence-based readouts is mass spectrometry (MS), which allows label-free multiplexed characterization of several analytes.I nt he past years,s everal groups have illustrated the compatibility of droplet microfluidics with electrospray ionization (ESI)-MS [146,147] or matrix-assisted laser desorption/ionization (MALDI)-MS for the analysis of enzyme secretion of yeast cells. [148] Notably,t he Kennedy group has reported the coupling of w/o droplets with highthroughput MS for the in vitro screening of enzyme inhibitors and activators. [149,150] With their methods,d roplets can be directly injected in the ESI-MS at at hroughput of almost 1Hz. In af ollow-up study,t he authors increased their Figure 13. A) Schematic representation of the dual-channel microfluidic droplet screening (DMDS). B) (S)-Ibuprofen and (R)-ibuprofen modified with two different fluorophores, and the enzymatic reaction yielding two different fluorescents ignals. Substrate 20 is used as the selection substrate and substrate 22 as the counter-selection substrate. To improve the enantioselectivity of AFEST towards (S)-ibuprofen, variants with increased fluorescence signal for dye 1but lower fluorescence signal for dye 2w ere selected using on-chip sorting. [142] (Image adapted from ref. [142]).

Figure 14.
A) The transaminase activity of ATA117 is screened by evaluating the transformation of the non-native ATAs ubstrate (24)a nd pyruvate (25)t othe ATAp roduct (26)a nd alanine (27)a fter ivTT. B) Schematic representation of the mass-activated droplet sorting (MADS) device. Nanoliter-sized droplets are injected in the bottom left region ("injection") and are split asymmetrically ("splitting"). While the larger droplet travels directly to the mass spectrometer,the smaller droplet flows through the delay line. The smaller droplet is sorted using adielectrophoretic sorter ("sorting") according to the sorting decision made using the MS-signal. C) Forthe MADS device to function, three different samples are analyzed in parallel. Inactive and therefore uncolored droplets are recognized by acamera by pattern tracing (red). Marker droplets for synchronization contain food color and are detected by the camera (blue). The signals are synchronized with the MS signal of the marker-ion( orange). After synchronization, the MS signal of the target ion (green) is used to make asorting decision. [151] (Image reprinted with permission from Wiley-WCHV erlag GmbH & Co KGaA.). platform throughput more than threefold. Them ethod was applied to the screening of transaminase libraries and further highlighted the compatibility of their system with in vitro translation-transcription of proteins (Figure 14 A).
Them ost recent advance in this field concerns the screening of enzymatic reactions with an innovative method, termed MADS (mass-activated droplet sorting). MADS combines MS analysis with FADS and benefits both from the high sensitivity of MS and from the possibility of collecting the sample allowed by FADS.T he MADS device allows for droplet production and splitting of each droplet in two.O ne fraction is analyzed by ESI-MS while the second fraction follows adelay channel leading to aF ADS electrode (Figure 14 B). TheE SI-MS results allow the active sorting of the second fraction with athroughput of 0.7 Hz (Figure 14 C). Thea uthors applied their methods to the activity-based screening of am odel transaminase library expressed in vitro. [151]

Outlook
Moving away from model enrichments and display of platform capabilities,many research groups are now working on improving enzymes with industrial or medical relevance. [65,74,76,150,152,153] Besides the evolution of natural enzymes,t he toolkit of directed evolution is expanding to artificial enzymes to introduce non-natural reactivities urgently needed in the pharmaceutical industry, [152] de novo designed enzymes to understand and reengineer enzyme active sites,a nd, more recently,m achine-learning-assisted directed evolution. [154] Furthermore,asfluorogenic substrates cannot always be synthesized, novel strategies for fast, but non-fluorescence-based detection will be critical for future developments.
Thea dvances in droplet microfluidics over the past 20 years have permitted decisive steps towards the discovery of enzymes with new or improved functionalities.T he higher throughput and facilitated sorting enabled the directed evolution of libraries of increasing sizes at as ignificantly reduced time and material consumption. Droplet microfluidics paved the way to automated workflows and faster screenings,a nd enabled the use of conventional FACS, accessible in most biology institutes.C lose collaborations between engineering groups and chemistry or biology research groups showed as ynergetic effect by allowing successful large campaigns.T oc ontinue on this prosperous avenue,m icrofluidics systems must be further simplified to enable robust operation of microfluidic devices by nonexperts.C heap,c ommercially available microchips will further lower the hurdles to exchange standard tools for microfluidic systems. [155] With more and more groups working on directed evolution using microfluidic systems,t he spectrum of applications and assays will broaden. We believe that the joint effort from these two fields holds great promise,a nd we are looking forward to the new innovative developments that will emerge from collaborations between engineers and biochemists in the future!