Microfluidic droplet platform for ultrahigh-throughputsingle-cell screening of biodiversityStanislav S. Terekhova,b,1, Ivan V. Smirnova,c,1, Anastasiya V. Stepanovaa, Tatyana V. Bobika,c, Yuliana A. Mokrushinaa,Natalia A. Ponomarenkoa,2, Alexey A. Belogurov Jr.a,c, Maria P. Rubtsovab,d, Olga V. Kartsevac, Marina O. Gomzikovac,Alexey A. Moskovtseve, Anton S. Bukatinf, Michael V. Dubinaf, Elena S. Kostryukovag,h, Vladislav V. Babenkog,Maria T. Vakhitovah, Alexander I. Manolovg, Maja V. Malakhovag, Maria A. Kornienkog, Alexander V. Tyakhtg,h,Anna A. Vanyushkinag, Elena N. Ilinag, Patrick Massonc,i, Alexander G. Gabibova,b,c,3, and Sidney Altmanj,3

aShemyakin-Ovchinnikov Institute of Bioorganic Chemistry, Russian Academy of Sciences, 117997 Moscow, Russian Federation; bDepartment of Chemistry,Lomonosov Moscow State University, 119991 Moscow, Russian Federation; cKazan Federal University, 420008 Kazan, Russian Federation; dSkolkovoInstitute of Science and Technology, 143026 Skolkovo, Russian Federation; eInstitute of General Pathology and Pathophysiology, 125315 Moscow, RussianFederation; fSt. Petersburg Academic University, 194021 St. Petersburg, Russian Federation; gResearch and Clinical Centre of Physical-Chemical Medicine,119435 Moscow, Russian Federation; hMoscow Institute of Physics and Technology, 141701 Dolgoprudny, Russian Federation; iFrench Academy ofPharmacy, 75270 Paris, France; and jDepartment of Molecular, Cellular and Developmental Biology, Yale University, New Haven, CT 06520

Contributed by Sidney Altman, January 7, 2017 (sent for review September 15, 2016; reviewed by Robert S. Phillips and Israel Silman)

Ultrahigh-throughput screening (uHTS) techniques can identifyunique functionality from millions of variants. To mimic the naturalselection mechanisms that occur by compartmentalization in vivo, wedeveloped a technique based on single-cell encapsulation in dropletsof a monodisperse microfluidic double water-in-oil-in-water emulsion(MDE). Biocompatible MDE enables in-droplet cultivation of differentliving species. The combination of droplet-generating machinery withFACS followed by next-generation sequencing and liquid chromatog-raphy-mass spectrometry analysis of the secretomes of encapsulatedorganisms yielded detailed genotype/phenotype descriptions. Thisplatform was probed with uHTS for biocatalysts anchored to yeastwith enrichment close to the theoretically calculated limit and cell-to-cell interactions. MDE–FACS allowed the identification of humanbutyrylcholinesterase mutants that undergo self-reactivation afterinhibition by the organophosphorus agent paraoxon. The versatilityof the platform allowed the identification of bacteria, including slow-growing oral microbiota species that suppress the growth of a com-mon pathogen, Staphylococcus aureus, and predicted which generawere associated with inhibitory activity.

ultrahigh-throughput screening | microfluidic encapsulation |butyrylcholinesterase | Staphylococcus aureus | cell–cell interactions

The ultrahigh-throughput (1, 2) technique of screening (uHTS)in a double emulsion was applied in directed enzyme evolu-

tion (3, 4) to investigate the idea that a universal genotype–phenotype linkage was provided by compartmentalization. Arti-ficial compartments of double emulsions were produced with highpolydispersity by shear stress (5, 6), which significantly decreasedthe portion of uniform droplets and thereby reduced the sensitivityand the maximal sorting rate. By contrast, sophisticated customsorters demonstrated the screening of precise monodisperse drop-lets of water-in-oil emulsions generated by microfluidic technology(1, 7). However, it is not always convenient to use custom devices,and the use of oil as a continuous phase limits the sorting rate.Alternatively, compartmentalization in microfluidic double emul-sion (MDE) enables uHTS of >10,000 events/s using commerciallyavailable cell sorters (8, 9). Furthermore, biocompatible oil andwater phases provide viability and proliferation of Escherichia colicells (10) inside the microenvironment of a double emulsion.Here, we propose an MDE–FACS platform that combines the

benefits of previously reported systems based on compartmen-talization in MDE and FACS selection together with modernomics (Fig. 1). We succeeded in assembling this platform usingcommercially available parts, which included straightforwardmicrofluidics (Fig. S1) for MDE generation, multiparametric FACSfor uHTS, next-generation sequencing (NGS) for bioinformaticpredictions, and mass spectrometry for proteome and secretome

analysis. We demonstrated this idea with several single-cell meth-ods (Fig. S2), including the selection of different biocatalytic ac-tivities, screening enzymes with different levels of the same activity,de novo creation of enzymes with artificial activity (Fig. 2), andinvestigation of bacterial cell-to-cell interactions (Fig. 3).

Results and DiscussionEnzyme screening was performed with two types of yeast (Fig.2A): “active” cells displaying a membrane-anchored target en-zyme and producing the RFP reporter mCherry (Fig. S3 A andB) and control cells with an “inactive” protein [a fragment of anantibody (Fab) lacking catalytic activity] without any fluorescentmarkers. A mixture of active and inactive cells was encapsulatedwith a specific green fluorogenic substrate in MDE droplets. Theincubation of active cells with the substrate inside droplet com-partments resulted in the generation of a fluorescent product.Active and inactive cells were distinguished by green fluores-cence using FACS. Droplets with inactive cells had low greenfluorescence and low red fluorescence, whereas the active cellshad high red and green fluorescence. Fig. 2B shows an example

Significance

Biocompatible microfluidic double water-in-oil-in-water emul-sion (MDE) enables in-droplet cultivation of different livingspecies. The combination of droplet-generating machinery withFACS followed by next-generation sequencing and liquid chro-matography-mass spectrometry analysis of the secretomes ofencapsulated organisms yielded detailed genotype/phenotypedescriptions. The MDE–FACS platform we developed enabledhighly sensitive single-cell selection of predesigned activity andexploration of pairwise interactions between target and effectorcells without interference from other microbiota species.

Author contributions: S.S.T., I.V.S., N.A.P., M.V.D., A.G.G., and S.A. designed research;S.S.T., I.V.S., A.V.S., T.V.B., Y.A.M., O.V.K., V.V.B., M.T.V., M.V.M., M.A.K., and A.A.V.performed research; Y.A.M., M.P.R., M.O.G., A.A.M., A.S.B., M.V.D., and E.N.I. contributednew reagents/analytic tools; S.S.T., I.V.S., A.V.S., N.A.P., A.A.B., E.S.K., A.I.M., A.V.T., A.A.V.,E.N.I., P.M., A.G.G., and S.A. analyzed data; and S.S.T., I.V.S., A.G.G., and S.A. wrotethe paper.

Reviewers: R.S.P., University of Georgia; and I.S., Weizmann Institute of Science.

The authors declare no conflict of interest.

Freely available online through the PNAS open access option.1S.S.T. and I.V.S. contributed equally to this work.2Deceased December 16, 2016.3To whom correspondence may be addressed. Email: gabibov@mx.ibch.ru or sidney.altman@yale.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1621226114/-/DCSupplemental.

2550–2555 | PNAS | March 7, 2017 | vol. 114 | no. 10 www.pnas.org/cgi/doi/10.1073/pnas.1621226114

Dow

nloa

ded

by g

uest

on

Aug

ust 1

, 202

0

of an enzymatic reaction in droplets with the encapsulatedmixture of active and inactive cells (1:10). The efficiency ofbiocatalyst selection was quantified by mixing active and inactivecells in ratios varying from 1:10–1:105 within one screeninground (Fig. 2C). For low dilutions (1:10 and 1:100), we managedto achieve the theoretical enrichment maximum. Moreover, en-richment efficiency was more than 3.5 × 104, corresponding to

35% of active cells, even at an extremely high dilution (1:105).Thus this approach may be useful for the specific selection ofexceptionally rare events.To test whether the MDE–FACS platform could be used to

select different enzymatic activities from the biocatalyst bouillon,we mixed separate individual clones displaying phosphodiester-ase (DNase I), protease (bovine enteropeptidase, EK), and

Library

Machinery

Trash

Activators

SequencingKinetics

LC-MSPlating and cultivation

Microfluidic compartmentalization in biocompatible double W/O/W emulsion

Activated machinery produces fluorescent substance

Selection of specificactivators using FACS

Genotypic, phenotypic, structural andfunctional analysis of activators

Fig. 1. Principal scheme of the MDE–FACS technique. Compartmentalization of a library of species with specific fluorescence-generating machinery in MDEusing emulsification in microfluidic chips enabled single-cell probing of a targeted function. Only specific phenotypes (indicated by the yellow star) fit andactivated the machinery in droplets, thus leading to the production of fluorescent substances. Because the droplet volume was uniform, similar concen-trations and conditions resulted in a narrow fluorescence distribution that corresponded to the same activity. The fluorescent signal was registered by aconventional cell sorter that separated the variants that activated the machinery (activators) from those that did not (trash). Unculturable species wereanalyzed by the direct sequencing of MDE. If the selected activators could be cultivated in vitro, they were regenerated from the droplets, grown, andanalyzed through the combination of classical methods (kinetics, LC-MS, sequencing, and others). W/O/W, double water-in-oil-in-water emulsion.

Pro

duct

RFP

Active protein(enzyme)

Inactive protein(antibody Fab fragment)

Accumulation offluorescent product

No reaction

107-108

individual reactions

Selection usingFACS

Active

Inactive

G

A

C E

F HLib G3 G2 G1

0

50

100

150

BChE

activ

i ty, %

102 103 104

Nor

mal

ized

cou

nt

Product, RFU

G1G2G3

Fab Lib BChE

Activity, % of WT BChEcl.3

55

cl.8

5.4

cl.13

1.1

Fab

0

102 103 104

Nor

mal

ized

cou

nt HighMediumLow

Product, RFU

Gate

B

D

1

10

100

DNase EK BChE

DNase EK BChE Fab

Theoretical maximum

1

10

100

1000

**Low Medium High

Theoretical maximum

Enric

hmen

t

0

50

100

150

200

250

300

0 5 10

Res

oruf

in,R

FU

Time, h

BChEWT

cl 14

POX-R, µM2

0.71.226

Ki k1 k2E OP+ E OP... E-OP E OP-OH+

H2OG

cl.3 cl.8 cl.13Fab

Enr

ichm

ent

0

20

40

60

80

100

Act

ive

afte

rscr

eeni

ng,%

1

10

102

103

104

105

Enrichment Portion of active

Theoretical maximum

10 102 103 104 105

Dilution of active with inactive

Enr

ichm

ent

Fig. 2. Screening of biocatalysts anchored to the yeast surface using MDE–FACS. (A) Compartmentalization of active (RFP-positive) and inactive (non-fluorescent) yeast cells with fluorogenic substrate. After the mixture of active and inactive cells was encapsulated, the fluorescent product accumulated solelyinside the droplets with active cells, which were selected using FACS. (B) Visualization of biochemical reaction by merging the signals of green fluorescence(reaction product), red fluorescence (reporter protein), and the visible light image. (Scale bar, 100 μm.) (C) The plot represents enrichment efficiencydepending on the dilution of active cells with inactive cells after one round of MDE–FACS. (D) Enrichment efficiency of MDE–FACS for active cells displayingdifferent biocatalysts (DNase I, EK, and BChE) and inactive cells (Fab) mixed in a 1:1:1:100 ratio. (E) Selection of BChE mutants with different levels of activityfrom the BChE library before selection (Lib) and after the selection using gates G1–G3. (Inset) Overlay of the MDE–FACS spectra of droplets with Fab, BChElibrary, and WT BChE. The gates used for sorting are indicated. Median ± interquartile ranges are shown for each group. P < 0.0001 for gates G1–G3 vs. library(Mann–Whitney–Wilcoxon test). (F) Distribution of droplets with encapsulated BChE mutants and Fab by fluorescence of the reaction product. The upperpanel shows activity of the BChE mutants in comparison with WT BChE. Gates have distinct fluorescence levels: high (the median of cl.3), medium (cl.8), andlow (cl.13). (G) Enrichment efficiency of BChE mutants mixed with Fab in a 1:1:1:1,000 ratio. The analysis was performed using high, medium, and low gates asdefined in F. Asterisks indicate an undetectable level of enrichment. (H) Interaction of BChE (E) with OP. The cl.14 mutant rather than WT BChE displayscatalytic hydrolysis of POX-R. Ki, inhibition constant; k1, phosphylation constant; k2, self-reactivation constant. The BChE concentration was 0.45 μM. All datapoints plotted are derived from at least two measurements; error bars denote SD (n = 3 technical replicates). RFU, relative fluorescence units.

Terekhov et al. PNAS | March 7, 2017 | vol. 114 | no. 10 | 2551

APP

LIED

BIOLO

GICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Aug

ust 1

, 202

0

esterase (human butyrylcholinesterase, BChE) activities, and aFab fragment in a 1:1:1:100 ratio. A set of fluorescence-generatingsystems (Fig. S3 C and D) for sensitive discrimination of

different substrate specificities was designed. The results obtainedafter one round of MDE–FACS (Fig. 2D) indicated that selectionwas both efficient and specific for all enzymes. The experimental

GNo enrichment

Gre

en

Red

Gre

en

Far-Red

D

S.aureusS.aureusS.venezuelae

S.aureusE.coli

C

B

A

GEnrichment

TargetTarget

Killer

Target

Mate

S.aureusS.aureusS.venezuelae

S.aureusE.coli

E

Initial load of target cells (sCy5)

Viable targets after coculturing (GFP)

Total count of viable cells after coculturing (Calcein Violet AM)

Effector

Target pathogenproducing GFP

sCy5-NHSstaining

Viable

Viable labledtarget

Coencapsulation of app. 1 effector and 10 targets

Dead

Source of differentbacterial species

Dead target Viable target

Desirable No survival The most expected

1

2

G1

sCy5

Dro

plet

s

G2

GFP

Dro

plet

s G3

Calcein Violet

Dro

plet

s

3

F

2 4 8 16

Por

tion

of c

lone

s,%

0

20

40

60

80

100

PlatesDroplets

Inhibiting dilution

H

102

103

104

102 103 104

Reads before selection

4 - Pseudomonas

10 - Streptococcus

1 - Propionibacterium

3 - Sphingomonas

5 - Escherichia6 - Janthinobacterium7 - Corynebacterium8 - Serratia9 - Enterobacter

2 - Stenotrophomonas

Rea

dsaf

ters

elec

tion

0

1

2

3

4

Enric

hmen

t

Genera

1

12

2

3

4

3455

6

67

78

89

9

10

10

StaphylococcusKillers similar to Streptococcus Killers similar to Pseudomonas

No enrichment

GV

isib

le li

ght

Targ

et

Effe

ctor

Ove

rlay

S.aureusS.venezuelae

S.aureusE.coli

I II III IV

G1 G2 G3

Fig. 3. Screening of bacteria inhibiting S. aureus growth using MDE–FACS. (A–D) The concept for screening of antibiotic producers in MDE. (A) TargetS. aureus cells with a GFP reporter were coencapsulated with either killer S. venezuelae cells producing red fluorescent metabolites or mate E. coli cells with afar-red fluorescent reporter. (B) Coculturing of S. aureus with S. venezuelae in contrast to E. coli inhibited S. aureus growth and yielded distinct combinationsof fluorescent signals. (C and D) Selection of droplets with the lowest level of green fluorescence resulted in the enrichment of killers (C) rather than mates (D).(E) Visualization of droplets containing S. aureus cocultured with S. venezuelae (Left) or E. coli (Right). (Scale bar, 100 μm.) (F) Selection scheme for bacteriainhibiting S. aureus growth from human oral microbiota. The target S. aureus cells were labeled with the red fluorescent dye sCy5 and were further en-capsulated with oral microbiota in MDE. Coculturing the target cells with the effectors provided four different scenarios. (1) From left to right: (i) the effectorkills the target; (ii) both the target and the effector are dead; (iii) both the target and the effector are alive; and (iv) the target kills the effector. (2) Theaddition of Calcein Violet AM to the emulsion after coculturing resulted in the accumulation of blue fluorescent product only inside the droplets with livebacteria. (3) Gating of droplets with sCy5high, GFPlow, and Calcein Violethigh fluorescence resulted in desirable droplets with proliferated effector and inhibitedtarget cells. (G) Prediction of bacterial genera from human oral microbiota inhibiting S. aureus growth using 16S-ribosome sequencing. (Inset) Estimated en-richment of bacterial genera as indicated; subpopulations indicated by line differ with P < 0.01. (H) Comparative analysis of the inhibition activity of bacteriaselected by conventional plate analysis (plates) and MDE–FACS (droplets). Data represent mean ± SD (n = 3 biological replicates from one experiment).

2552 | www.pnas.org/cgi/doi/10.1073/pnas.1621226114 Terekhov et al.

Dow

nloa

ded

by g

uest

on

Aug

ust 1

, 202

0

enrichment efficiency (74 ± 5) was close to the theoretical maxi-mum (78.7), and nonspecific enrichment was not observed.To focus on another biotechnologically relevant issue, the

isolation of clones with different levels of the same activity, wecreated a BChE library (Fig. S4) mutated in the acyl-bindingloop (284–TPLSV–288). This loop is highly variable amongmammalian BChEs and represents an important region formutagenesis aimed at altering the enzyme catalytic properties.Mutations dramatically reduced the average activity of the li-brary to ∼1% of WT BChE. Three different gates with de-creasing fluorescence levels (G1 > G2 > G3) were used forselection in the MDE–FACS mode (Fig. 2E). Individual yeastclones were regenerated from droplets before and after selectionusing G1–G3 to illustrate qualitatively that the gate fluorescencecorrelated with the activity of the selected BChE mutants. Todetermine quantitatively the selection efficiency of enzymes withdifferent levels of the same activity (Fig. 2F), we mixed repre-sentative BChE mutants (cl.3 with high activity, cl.8 with mediumactivity, and cl.13 with low activity) selected using G1–G3 withthe control yeasts in ratios of 1:1:1:1 and 1:1:1:1,000. The frac-tion of BChE mutants with different levels of BChE activity afterthe gate selection process with high, medium, and low fluores-cence levels indicated efficient, specific enrichment of eachBChE mutant in the 1:1:1:1 mixture (Fig. S5). Cl.3 and cl.8 alsowere selected efficiently from the more diluted 1:1:1:1,000 mix-ture (Fig. 2G); specific enrichment in cl.13 was lower, indicatinglimited selection efficiency of mutants with activity close tobackground. In both mixtures, cl.3 was not selected in gates withmedium or low fluorescence levels, suggesting that the selectionof droplets with less fluorescence resulted in efficient exclusionof the most active mutant. As a part of the negative selection,this strategy enabled us to identify mutations that abolishedenzymatic activity.The MDE–FACS platform allowed us to identify BChE mu-

tants resistant to irreversible inhibition by organophosphate(OP) nerve agents using a BChE library preinhibited with OPs(Fig. S6A): paraoxon (POX) or a coumarinyl analog of soman[GDC, 3-chloro-4-methyl-2-oxo-2H-chromen-7-yl (3,3-dime-thylbutan-2-yl) methylphosphonate]. BChE mutants displayingresistance to POX (cl.14 and cl.15) and GDC (cl.14 and cl.19)were isolated after one round of screening. The interaction be-tween WT BChE and OP leads to fast (high bimolecular reactionconstant k1/Ki) and irreversible (k2 ∼0) inactivation of WT BChEby phosphylation with OP (Fig. 2H) (11). However, severalBChE mutants (cl.14 and cl.15) showed residual BChE activityeven after prolonged incubation with POX (Fig. S6B). We at-tribute this property to self-reactivation (k2 > 0), i.e., hydrolysisof the enzyme–OP adduct, which was demonstrated using theresorufin analog POX-R (Fig. 2H). The k1/Ki and k2 values (Fig.S6C) indicated that the selection of BChE mutants resistant to OPinactivation occurred via two distinct pathways: by decreasing thereactivity with OP (low k1/Ki values obtained for cl.19) and theemergence of self-reactivation (e.g., paraoxonase activity observedfor cl.14). A combination of both pathways also was observed forcl.15. The selected BChE mutants were of high interest becausethey revealed a previously unreported target (the BChE acyl-binding loop) for the creation of catalytic bioscavengers based onhuman BChE, which are used extensively for protection againstnerve agent poisoning and postexposure treatment (12, 13).The scope of the MDE–FACS platform may be expanded to

highlight the peculiarities of bacterial coexistence, cell-to-cellinteraction, and signaling, as demonstrated using a simple model(Fig. 3 A–D). The concept was proved by the coencapsulation ofcell pairs (target/killer and target/mate) with a well-studied co-existence (14) into MDE. A common pathogen, Staphylococcusaureus, was used as a target, and an antibiotic producer, Strep-tomyces venezuelae, which inhibits the growth of S. aureus, wasthe model killer (Fig. 3A); E. coli was used as the mates, because

it does not influence the growth of S. aureus. All the bacteria haddistinct fluorescent reporters for monitoring their presence indroplets: S. aureus produced GFP, S. venezuelae produced redfluorescent prodiginines (15, 16), and E. coli produced the far-red fluorescent protein Katushka2S (17). The inhibition ofS. aureus and the growth of S. venezuelae corresponded to dropletswith low green and high red fluorescence in cocultivation (Fig. 3B and E). Simultaneously, cocultivation of S. aureus and E. coliled to the growth of both bacteria in droplets with high green andfar-red fluorescence (Fig. 3 B and E). Thus killers and matescould be distinguished by the fluorescence of their compartments,and selection of the droplets with the least green fluorescenceresulted in the enrichment of killers (Fig. 3C) rather than mates (Fig.3D). We observed that the enrichment efficiency was limited and wasdramatically dependent on the S. venezuelae dilution (Fig. S7). Theselimitations result from similarity between “desirable” droplets withlow green fluorescence from S. venezuelae inhibition and “empty”droplets during negative selection using a single GFP reporter.Despite the high pathogenic potential of S. aureus (18), it is

rarely associated with dentoalveolar infection (19). We hypoth-esized that the suppression of S. aureus infection relates to anatural inhibitory activity of some unknown effectors from hu-man oral microbiota. We tested this possibility by identifyingS. aureus killers from the oral microbiota at the level of bacterialclones and their genomes and secretomes. Accordingly, the ini-tial scheme of bacterial screening was modified to use a combi-nation of fluorescent signals for both positive and negativeselection to enable the efficient isolation of S. aureus killers. Thefollowing additional fluorescent reporters were used: sCy5 toevaluate the initial S. aureus load and Calcein Violet to estimatethe total number of viable cells in each individual droplet aftercocultivation (Fig. 3F). The gating of droplets with simultaneoussCy5high, GFPlow, and Calcein Violethigh fluorescence (Fig. 3 F,3) yielded desirable droplets with a high initial S. aureus load, alow number of viable S. aureus cells, and a high number of un-known effector cells after cocultivation. The selected dropletswere used for (i) NGS and bioinformatic analysis of the bacteria,especially unculturable bacteria, and (ii) proteomic and secre-tomic analysis of culturable bacteria regenerated by agar plating.The MDE–FACS platform provided an essential link between

the genotype and its functionality for a diverse population at thesingle-cell level, a challenging problem in microbiome analysis.Isolation of droplets containing an inhibited target with subsequentNGS and bioinformatic analysis allows the prediction of bothculturable and, more critically, unculturable killers, thus revealingpotential effectors that are not identifiable using classical micro-biology approaches. The prediction of S. aureus inhibitors by 16SrRNA sequencing uncovered distinct populations of bacterialgenera characterized by different enrichment efficiencies (Fig. 3G).Propionibacterium, Stenotrophomonas, Sphingomonas, Pseudomo-nas, and Escherichia were specifically and efficiently enriched inselected droplets, and Corynebacterium, Janthinobacterium, Ser-ratia, Enterobacter, and Streptococcus were enriched with lowerefficiency. Thus, the technology we developed allows the pre-diction of potential cell coexistence and allows the microbiometo be subdivided into discrete functional subpopulations. Whole-genome sequencing confirmed the dramatic enrichment of slow-growing Propionibacterium acnes. The Streptococcus mitis group(S. pneumoniae, S. mitis, S. oralis, and S. pseudopneumoniae),Prevotella dentalis (slow-growing bacteria), Staphylococcus epi-dermidis (the well-known S. aureus effector), and Pseudomonasaeruginosa were considerably amplified after selection.This platform allowed us to verify predictions for culturable

bacterial species. Fig. 3H shows substantial improvement of theselection procedure for inhibitory clones compared with a con-ventional agar plating screening assay, allowing us to select bac-terial clones with enhanced S. aureus inhibition. Mass spectrometryenabled us to perform taxonomical classification and secretome

Terekhov et al. PNAS | March 7, 2017 | vol. 114 | no. 10 | 2553

APP

LIED

BIOLO

GICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Aug

ust 1

, 202

0

analysis of culturable S. aureus killers. Most (>90%) belonged tothe Streptococcus genus, and 64% were classified as Streptococcusoralis. Growth media obtained from S. oralis strains were charac-terized by a high (up to 16-fold) inhibiting dilution, althoughS. oralis did not abolish S. aureus growth in diluted liquid coculture.P. aeruginosa, an uncommon bacteria in the analyzed microbiotasamples (Fig. 3G), was selected exclusively using MDE–FACS butnot with the conventional screening assay. P. aeruginosa displayedoutstanding eradication of S. aureus at a dilution of >1:106 in co-culture (Fig. S8). Liquid chromatography-mass spectrometry (LC-MS) was used to analyze the secretome of P. aeruginosa for me-tabolites inhibiting S. aureus. Pyocyanin, phenazine-1-carboxylicacid, and heptyl-4-hydroxyquinoline were the principle activecompounds in P. aeruginosa, displaying pronounced synergetic in-hibition of S. aureus growth (Figs. S8 and S9). It is noteworthy thatPseudomonas and Streptococcus predicted by 16S rRNA sequenc-ing (Fig. 3G) were selected using MDE–FACS platform as cul-turable bacteria that inhibited S. aureus growth.

ConclusionsThe platform used for the selection of biocatalysts and identifica-tion of cell-to-cell interactions allowed the screening of enormous(>108) repertoires of DNA-encoded information. Theoretically,the rate-limiting step for the application of this platform is theFACS-assisted droplet-screening technology (20), which is limitedto several tens of thousands of events per second. The character-istics of the MDE–FACS platform—specifically, sensitive detectioncombined with specific separation—allow its use in practical researchon drug design, including the selection of novel desired activity, directisolation of probiotics, antibiotic producers, and prediction of uncul-turable effectors. However, the analysis of metabolites or secretomesseems to be the main limiting factor in MDE–FACS automation. TheMDE–FACS platform enabled highly sensitive single-cell selection ofpredesigned activity and the exploration of pairwise interactions be-tween target and effector cells without interference from othermicrobiota species. The unique ability this technique to accomplish theseparation of living species in biocompatible conditions has greatpotential applications in the field of synthetic biology.

Materials and MethodsFabrication of Microfluidic Chips. Microfluidic chips with flow-focusing geom-etry (Fig. S1 B and C) were produced using standard soft lithographic tech-niques (21, 22), i.e., SU-8/Si masters were used to make polydimethylsiloxane(PDMS) slides with 60 × 70 μm and 20 × 22 μm (width × height) orifices. Inletsand outlets were made using a 1.2-mm Harris Uni-Core biopsy punch. Slideswere bound to PDMS or glass slides using an Atto plasma cleaner (Diener)after a 1-min treatment [maximum power, p(O2) = 0.7 mbar]. Immediatelyafter bonding, chips were treated with 1% (wt/vol) polyvinyl alcohol (23)Mowiol 23–88 (Kuraray Specialties Europe) for hydrophilic chips or 0.5% (wt/vol)trichloro(octadecyl)silane (Sigma-Aldrich) in mineral oil for hydrophobic chips[Aquapel (PPG Industries) was used for fluorocarbon oil]. After a 1-min treat-ment, surface modifiers were removed thoroughly by vacuum, and the chipswere heated for 3 min in a drying oven at 120 °C. After modification, hydro-phobic chips could be stored for more than 3 y, and hydrophilic chips retainedtheir properties for at least 1 y of storage at room temperature.

Fluidics Assembly and Generation of Double Emulsion. Our fluidics were basedon anOB1-MkII piezoelectric pressure controller (Elveflow) that independentlyoperated with four liquid streams: inner water 1 (IW1) for cell suspension;inner water 2 (IW2) for substrate solution; oil (O); and other water (OW).Actual flow rates were measured using microfluidic flow sensors (Elveflow)automatically operated in flow-control mode using an inner feedback loop.The IW1 reservoir and the corresponding flow sensor were arranged verticallyto minimize cell sedimentation. The inner water streams were mixed beforeusing the hydrophobic chip with MicroTee (P-890; IDEX) and tubing with adead volume of ∼150 nL. Subsequently, the joint inner water stream wassequentially emulsified in hydrophobic and hydrophilic chips using O [lightmineral oil (Sigma-Aldrich) with 3% (wt/vol) Abil EM 180 (Evonic) or fluoro-carbon oil (Novec 7500) with 2% (wt/vol) Pico-Surf 2 (Dolomite)] and the OWphase [2% (wt/vol) Pluronic F127 (Sigma-Aldrich), 0.1% Mowiol 23-88, and50 mM potassium phosphate buffer, pH 7.4]. The characteristic flow rates

were 6:6:4:200 μL/min for the IW1:IW2:O:OW phase for the 60-μm chips and4:4:2:50 μL/min for the 20-μm chips.

Yeast Display of Model Enzymes. Recombinant BChE, human EK, and humanDNase I were produced in Pichia pastorisGS115 (Invitrogen) using the modifiedexpression vector pPICZ-mCherry-F2A-HSAss-AfeI/PvuI-HA-SAG1 based onpPICZαA (Invitrogen). This vector contained the fluorescent reporter proteinmCherry, self-processing F2A peptide, the human serum albumin (HSA) signalsequence, AfeI and PvuI cloning sites, a-agglutinin anchor subunit Aga1p fromSaccharomyces cerevisiae, and an HA epitope. The fragments encoding BChE,EK, and DNase I sequences were PCR-amplified from pFUSE PRAD-F2A-BChE(24), pHENm/L-HEP/C122S (25), and p6E-rhDNaseI, which was kindly providedby Pharmsynthez PJSC, and were cloned into vector pPICZ-mCherry-F2A-HSAss-AfeI/PvuI-HA-SAG1 digested with AfeI and PvuI. Primers used for amplificationare presented in SI Materials and Methods.

Encapsulation of Yeasts and Enzymatic Reactions in Droplets. Yeasts producinganchored enzymes or a Fab fragment were grown overnight in liquid cultureusing yeast extract/peptone/dextrose (YPD) medium and were induced bygrowth in buffered methanol-complex medium (BMMY). Subsequently, theyeasts were washed twice with 50 mM potassium phosphate buffer (pH 7.4)and resuspended in appropriate buffer [1 mM MnCl2, 0.1 mM CaCl2, 20 mMTris·HCl (pH 7.4) for DNase I; 0.2 mM CaCl2, 20 mM Tris·HCl (pH 7.4) for EK;and 50 mM potassium phosphate buffer (pH 7.4) for BChE]. Next, the yeastswere filtered using a 20-μm solvent filter A-313 (IDEX), diluted with buffer toreach λ = 0.5 (OD600 = 0.3 and OD600 = 3 for the 60-μm and 20-μm chips,respectively), and mixed in an 1:10–1:105 enzyme:antibody ratio. Finally,they were compartmentalized in drops of double emulsions with an ap-propriate substrate in a corresponding buffer [1 μM FAM: AAAAAAAC-CCCCCCATATAGCGCGTTTTTTT-RTQ1 (Syntol) for DNase I; 10 μM SensoLyteRh110 substrate (AnaSpec) for EK, and 30 μM butyrylthiocholine iodide(Sigma-Aldrich) with 30 μM 3-(7-Hydroxy-2-oxo-2H-chromen-3-ylcarbamoyl)acrylic acid methyl ester (Millipore) for BChE].

Encapsulation of S. aureus and S. venezuelae in Droplets. We used S. aureusconstitutively producing GFP that was kindly provided by Andrei Shkoporovof the Department of Microbiology and Virology, Russian National ResearchMedical University, Moscow. The S. venezuelae Ehrlich Aс-505 strain wasobtained from the Russian Collection of Microorganisms. The E. coli JW5503strain (kindly provided by Hironori Niki, National Institute of Genetics, Mis-hima, Shizuoka 411-8540, Japan) was transformed with a pKatushka2S-Bvector (Evrogen) that enabled inducible expression of the far-red fluorescentprotein Katushka2S. Bacteria were cultured using 2YT medium (16 g/L tryptone,10 g/L yeast extract, 5.0 g/L NaCl) supplemented with 2% glucose (for Strepto-myces) or 100 μg/mL ampicillin (for E. coli) in shaking flasks at 37 °C (S. aureus andE. coli) or 28 °C (S. venezuelae) at 250 rpm. S. aureus and E. coli were grown for4–6 h until they reached a logarithmic phase of growth; S. venezuelaewas grownfor 2 d. Subsequently, liquid cultures were washed three times with cultivationmedium [0.16% tryptone, 0.1% yeast extract, 0.05% NaCl, 0.5% glycerol, 0.67%yeast nitrogen base (YNB) and 1 mM isopropyl β-D-1-thiogalactopyranoside(IPTG)]. After washing, cultures were filtered using 40-μm cell strainers (GreinerBio-One) and were diluted to reach (i) OD600 = 0.1 (λ = 10) for S. aureus;(ii) OD600 = 0.03 (λ = 2) for E. coli; and (iii) OD600 = 0.3 (λ = 1) for S. venezuelae.S. venezuelae and E. coli cells were encapsulated pairwise with S. aureus indouble-emulsion droplets using 60-μm chips. Similarly, effector cells (S. venezuelaeand E. coli) were diluted by target cells in a 1:10 or 1:100 effector:target ratioand were encapsulated with S. aureus in double-emulsion droplets. Afterencapsulation, the droplets were incubated overnight for 2 d at 25 °C.

Droplet Visualization. Microfluidic double-emulsion droplets with encapsu-lated yeast or bacterial cells were loaded into a hemocytometer and werevisualized using an Eclipse Ti inverted fluorescence microscope (Nikon) withstandard FITC and Texas Red filters.

Sorting and Regeneration from Droplets. Droplets were sorted using FACSAriaIII (BD). Initial gating of intact double-emulsion droplets was performed usingscattering and background fluorescence of the substrate or culture medium.Sorting of positive events was performed using a 530/30 nm emission filter (or450/50 nm for BChE). Yeast and bacteria clones were regenerated by agarplating after sorting in three replicates. Individual yeast colonies with dis-played enzymes were analyzed by fluorescence from the mCherry reporterusing VersaDoc (Bio-Rad). The number of E. coli colonies was calculated bymeasurement of the far-red fluorescence of reporter Katushka2S. Thenumber of S. venezuelae colonies was calculated manually based on specificmorphology. The degree of enrichment was defined as the ratio of the

2554 | www.pnas.org/cgi/doi/10.1073/pnas.1621226114 Terekhov et al.

Dow

nloa

ded

by g

uest

on

Aug

ust 1

, 202

0

percentage of positive colonies after and before sorting in three replicates.Maximal theoretical enrichment was calculated as a dilution multiplied bythe maximal theoretical purity of sorting estimated using a Poisson distri-bution (Fig. S2).

Creation of the BChE Library. Recombinant BChE and its library were producedin the methylotrophic yeast Pichia pastoris GS115 (Invitrogen) using themodified expression vector pPic9k-α-SfiI-FLAG-anchor based on pPIC9k(Invitrogen). The degenerated primers (SI Materials and Methods) were usedto amplify regions flanking the mutated 284–TPLSV–288 loop of WT humanBChE and were cloned into expression vector pPic9k-α-Lib_BChE-FLAG-anchorusing SfiI. Vector was linearized by PmeI and transformed into P. pastorisGS115 as previously described (26). The diversity of the library was ∼9 × 105.

Selection of BChE Mutants with Resistance Against OP Inactivation. Yeastsproducing a library of anchored BChE mutants were grown, induced, andwashed as previously described. Next, a cell suspension was incubated withOPs [0.2 mM POX (Sigma-Aldrich) or 1 mM GDC] for 0.5 h at 25 °C, washedwith 50 mM potassium phosphate buffer (pH 7.4) supplemented with 2 μMof the respective OP, filtered as previously described, and encapsulatedwith BChE substrate, including 100 μM butyrylthiocholine and 100 μM 3-(7-Hydroxy-2-oxo-2H-chromen-3-ylcarbamoyl)acrylic acid methyl ester. After 3 h,the most fluorescent droplets were screened using FACS. The regeneratedyeast colonies were analyzed for BChE activity in a 384-well plate. The mostactive clones after OP inactivation (cl.14 and cl.15 for POX and cl.19 for GDC)were selected and sequenced to determine substitutions in the 284–TPLSV–288sequence of WT human BChE (Fig. S4). The selection of BChE mutants withaltered activity is presented in SI Materials and Methods.

Production of BChE Mutants and Kinetic Measurements. Recombinant humanBChEwas produced in FreeStyle 293-F cells (Thermo Fisher Scientific) using themodified expression vector pFUSE-BChE-6xHis based on pFUSE-hIgG1-Fc(Invivogen) containing truncated BChE W541H6Δ (27). Genomic DNA wasextracted from clones with improved resistance against irreversible OP in-hibition (28) and was used as a PCR template with primer pair i/iv (SI Materialsand Methods). This PCR product was used to replace the insert between theNheI and NcoI sites in pFUSE-BuChE-6xHis. Transfection of FreeStyle 293-F cellswas performed using 293fectin Transfection Reagent (Thermo Fisher Scien-tific). WT BChE and its mutants were purified using TALONMetal Affinity Resin(Clontech). Purified BChE mutants were used to determine the bimolecular

inhibition constant (k1/Ki) according to the Kitz–Wilson method (29). The re-sidual BChE activity was determined according to Ellman’s method (SI Mate-rials and Methods) (30). The activity was measured using a Varioskan Flashmultimode reader (Thermo Fisher Scientific) at λEx/λEm = 570/585 nm (for POX-R)and λEx/λEm = 360/450 nm (for GDC). All data were derived from threetechnical replicates.

Selection of S. aureus Killers from Oral Microbiota Using Droplets. Samples oforal microbiota were obtained by scraping the gingivae and sublingual foldof three independent healthy donors. The study was approved by the LocalEthics Committee of the Scientific Research Institute of Physical-ChemicalMedicine (SRI-PCM). All donors provided written informed consent. Scrapedbacteria were resuspended in medium for coculturing. Target S. aureus cellsproducing the GFP reporter at a logarithmic phase were stained with 5 μMsulfo-Cyanine5 NHS (Lumiprobe) for 1 h in PBS and were washed three timeswith medium for coculturing. S. aureus were coencapsulated with an oralmicrobiota suspension using 60 μm chips. After overnight incubation at35 °C, Calcein Violet AM (Thermo Fisher Scientific) was added to the dropletemulsion to a final concentration of 10 μM. After 30 min of incubation,droplets with simultaneous sCy5high, GFPlow and Calcein Violethigh fluorescence(using 660/20-, 530/30-, and 450/50-nm filters) were sorted. For detailed in-formation see SI Materials and Methods. Bacterial colonies that regeneratedafter agar plating and demonstrated inhibition of S. aureus growth wereanalyzed by mass spectrometry. Bacterial cells were spotted on a sample spotof a MALDI target plate (MSP 96 target, ground steel; Bruker Daltonics) andwere overlaid with 2 μL of matrix solution HCCA (saturated solution of α-4-cyano-hydroxycinnamic acid; Bruker Daltonics) in 50% acetonitrile (Sigma-Aldrich) and 2.5% trifluoroacetic acid solution (Sigma-Aldrich). Mass spectraprofiles were acquired using a Microflex spectrometer (Bruker Daltonics). Themolecular ions were measured automatically in linear positive ion mode withinstrument parameters optimized for a range of 2,000–20,000 m/z. The soft-ware packages flexControl 3.0 (Bruker Daltonics) and flexAnalysis 3.0 (BrukerDaltonics) were used for mass spectra recording and processing. Spectraidentification and analysis were carried out using the MALDI Biotyper 3.0(Bruker Daltonics). Identification was performed by comparing the obtainedspectra with the MALDI Biotyper 3.0 library (version 3.2.1.1). All additionalmethods are described in SI Materials and Methods.

ACKNOWLEDGMENTS. This work was supported by Grant RFMEFI60414X0069from the Ministry of Education and Science of Russia.

1. Agresti JJ, et al. (2010) Ultrahigh-throughput screening in drop-based microfluidicsfor directed evolution. Proc Natl Acad Sci USA 107(9):4004–4009.

2. Zhang H, et al. (2016) Autocrine-based selection of drugs that target ion channelsfrom combinatorial venom peptide libraries. Angewandte Chemie InternationalEdition 55(32):9306–9310.

3. Gupta RD, et al. (2011) Directed evolution of hydrolases for prevention of G-typenerve agent intoxication. Nat Chem Biol 7(2):120–125.

4. Aharoni A, Amitai G, Bernath K, Magdassi S, Tawfik DS (2005) High-throughputscreening of enzyme libraries: Thiolactonases evolved by fluorescence-activatedsorting of single cells in emulsion compartments. Chem Biol 12(12):1281–1289.

5. Tawfik DS, Griffiths AD (1998) Man-made cell-like compartments for molecular evo-lution. Nat Biotechnol 16(7):652–656.

6. Miller OJ, et al. (2006) Directed evolution by in vitro compartmentalization. NatMethods 3(7):561–570.

7. Najah M, et al. (2014) Droplet-based microfluidics platform for ultra-high-throughputbioprospecting of cellulolytic microorganisms. Chem Biol 21(12):1722–1732.

8. Yan J, et al. (2013) Monodisperse water-in-oil-in-water (W/O/W) double emulsiondroplets as uniform compartments for high-throughput analysis via flow cytometry.Micromachines (Basel) 4(4):402–413.

9. Zinchenko A, et al. (2014) One in a million: Flow cytometric sorting of single cell-lysateassays in monodisperse picolitre double emulsion droplets for directed evolution.Anal Chem 86(5):2526–2533.

10. Zhang Y, et al. (2013) A programmable microenvironment for cellular studies viamicrofluidics-generated double emulsions. Biomaterials 34(19):4564–4572.

11. Masson P, Lockridge O (2010) Butyrylcholinesterase for protection from organo-phosphorus poisons: Catalytic complexities and hysteretic behavior. Arch BiochemBiophys 494(2):107–120.

12. Geyer BC, et al. (2010) Plant-derived human butyrylcholinesterase, but not an or-ganophosphorous-compound hydrolyzing variant thereof, protects rodents againstnerve agents. Proc Natl Acad Sci USA 107(47):20251–20256.

13. Ilyushin DG, et al. (2013) Chemical polysialylation of human recombinant butyr-ylcholinesterase delivers a long-acting bioscavenger for nerve agents in vivo. ProcNatl Acad Sci USA 110(4):1243–1248.

14. Kelsic ED, Zhao J, Vetsigian K, Kishony R (2015) Counteraction of antibiotic productionand degradation stabilizes microbial communities. Nature 521(7553):516–519.

15. Tenconi E, Guichard P, Motte P, Matagne A, Rigali S (2013) Use of red autofluorescencefor monitoring prodiginine biosynthesis. J Microbiol Methods 93(2):138–143.

16. Lopez-Garcia MT, Rioseras B, Yague P, Alvarez JR, Manteca A (2014) Cell immobili-zation of Streptomyces coelicolor: Effect on differentiation and actinorhodin pro-duction. Int Microbiol 17(2):75–80.

17. Luker KE, et al. (2015) Comparative study reveals better far-red fluorescent proteinfor whole body imaging. Sci Rep 5:10332.

18. Sawyer AJ, et al. (2013) A peptide-morpholino oligomer conjugate targeting Staph-ylococcus aureus gyrA mRNA improves healing in an infected mouse cutaneouswound model. Int J Pharm 453(2):651–655.

19. Smith AJ, Jackson MS, Bagg J (2001) The ecology of Staphylococcus species in the oralcavity. J Med Microbiol 50(11):940–946.

20. Shim JU, et al. (2013) Ultrarapid generation of femtoliter microfluidic droplets forsingle-molecule-counting immunoassays. ACS Nano 7(7):5955–5964.

21. McDonald JC, et al. (2000) Fabrication of microfluidic systems in poly(dimethylsilox-ane). Electrophoresis 21(1):27–40.

22. Mazutis L, et al. (2013) Single-cell analysis and sorting using droplet-based micro-fluidics. Nat Protoc 8(5):870–891.

23. Teh S-Y, Khnouf R, Fan H, Lee AP (2011) Stable, biocompatible lipid vesicle generation bysolvent extraction-based droplet microfluidics. Biomicrofluidics 5(4):044113.

24. Terekhov S, et al. (2015) A novel expression cassette delivers efficient production ofexclusively tetrameric human butyrylcholinesterase with improved pharmacokineticsfor protection against organophosphate poisoning. Biochimie 118:51–59.

25. Gasparian ME, et al. (2013) Heterogeneous catalysis on the phage surface: Display ofactive human enteropeptidase. Biochimie 95(11):2076–2081.

26. Wu S, Letchworth GJ (2004) High efficiency transformation by electroporation ofPichia pastoris pretreated with lithium acetate and dithiothreitol. Biotechniques36(1):152–154.

27. Ralph EC, Xiang L, Cashman JR, Zhang J (2011) His-tag truncated butyrylcholinester-ase as a useful construct for in vitro characterization of wild-type and variant bu-tyrylcholinesterases. Protein Expr Purif 80(1):22–27.

28. Lõoke M, Kristjuhan K, Kristjuhan A (2011) Extraction of genomic DNA from yeasts forPCR-based applications. Biotechniques 50(5):325–328.

29. Kitz R, Wilson IB (1962) Esters of methanesulfonic acid as irreversible inhibitors ofacetylcholinesterase. J Biol Chem 237(10):3245–3249.

30. Ellman GL, Courtney KD, Andres V, Jr, Feather-Stone RM (1961) A new and rapidcolorimetric determination of acetylcholinesterase activity. Biochem Pharmacol 7(2):88–95.

Terekhov et al. PNAS | March 7, 2017 | vol. 114 | no. 10 | 2555

APP

LIED

BIOLO

GICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Aug

ust 1

, 202

0