Bioengeneering and Nanosystem - Extended Abstract

Droplet Microuidic Systems for Directed Evolution: Aiming at

IntegrationAndré Pereira Rodrigues de Faria Examination Commitee:Supervisors: Chairperson: Gabriel António Amaro MonteiroJoão Pedro Estrela Rodrigues Conde Supervisor: Dr. João Pedro Estrela Rodrigues CondeLígia Maria de Oliveira Martins Member of the Commitee: Dr. Luís Humberto Viseu Melo

(Dated: November 2018)

Directed evolution came to change the world. And its proof is the Nobel Prize in Chemistry this year.Genome variations may create huge libraries that ask for high-throughput screening systems. This workcontributes to the development of an integrated droplet based microuidic system. Several modules arerequired, namely: droplet generation, detection and sorting. Here two dierent sorting platforms were studiedfor a ow focusing droplet generator: pneumatic and dielectrophoresis (DEP) based. Planar and three-dimensional electrode's geometries were compared. Two novel 3D electrodes designs were implemented withsuccess. A circuit to power the electrodes with high voltage (700 Vpp) and high frequency (till hundredsof kHz) was built. Droplet detection using a-Si:H thin lm photoconductors was implemented, also withcustom made transimpedance amplier circuit. Image analysis script was developed to count and measuredroplets from microscope recording. Electrical sorting and detection by photoconductors was implemented,with system delays characterized, but still some challenges were left open.

Keywords: Droplet microuidics, DEP, Photosensors, Integration, Directed Evolution.

I. STATE OF THE ART

Directed evolution has proved to be an eective strat-egy for improving or altering the activity of biomoleculesfor research, therapeutic and industrial applications1.Genetic diversity is the start base for natural and lab-oratory evolution. Dierent strategies to approach mu-tagenesis are chosen depending on the application andlevel of knowledge on the protein behavior. Rational

Design is used when information about protein stru-ture and function is well known, just by changing oneamino-acid at a specic location may improve a activity,binding or other desired spec2. Random mutagenesis

when information available about how the protein activ-ity will change with a variant is hard to predict.Random approach creates huge libraries of variants thathave to be screened. A mere peptide with 10 amino acidsyields 21 aminoacids10 positions ∼ 1013 dierent combi-nations. Droplet microuidics works with volumes fromnL to pL (thousand times lower that with microplates),at kHz rate, although one order slower than the state-of-art FACS3) oers more exibility and compartmentalizesall the cell's products inside each droplet, creating mi-croreactor for each genotype and respective phenotype.

To sort droplets dierent strategies can be used: elec-trical, magnetic, pneumatic, acoustic and thermal. Twooustand, electric for its fast response time and through-put, and pneumatic for its great biocompatibilirty. Mag-netic control has a big drawback due to the need for mag-netic materials, either on uid or in droplets, causingpotential biocompatibility problems4. Thermal methodsexplore the advantage of thermocappilary eects on mi-croscale, either by heating up the liquid reducing its hy-draulic resistance or using a laser beam to block dropletspassage4. Acustic control have good biocompatibility

and can steer liquid plugs5? . Its major drawback is thecomplexity on implementation4.AC electrical eld can be used in droplet sorting taking

advantage of dielectrophoresis (DEP), no charged parti-cle is needed. The force is caused by the induction ofan electrical dipole, m that will be actuated by a spacialnon-uniform eld. The force is given by

F = m · ∇E (1)

For a spherical particle of radius r, with electrical con-ductivity sigmap and electrical permittivity ε∗p, inside amedium with electrical conductivity sigmam and elec-trical permittivity ε∗m the conventional dielectrophoresisforce based on classical Maxwell theory is given by Equa-tion 26.

FDEP = 2πr3εmRe[fCM ]∇E2rms (2)

The force shows dependence on the particle size, r3 term,with the medium and particle materials at a given fre-quency, εmRe[fCM ], and with the gradient of the squaredElectric eld, ∇E2

rms, if AC is the root mean squaredeld. Re[fCM ] is the real part of the Clausius-Mossotifunction given by Equation 3.

fCM (ω) =ε∗p − ε∗mε∗p + 2ε∗m

(3)

Where the complex permittivities, dependent of electriceld angular frequency ω, are described by Equation 4.

ε∗ = ε− j σω

(4)

When designing the electrodes, strong gradient elds aredesired. DEP force was rst reported for droplet mi-crouidic sorting in 20067 using metal electrodes. But

Bioengeneering and Nanosystem - Extended Abstract 2

similar to previous examples, strategy of using microu-idic channels with liquid metal can be used -avoidingalignment issues, and eld loss through height8.

Pneumatic actuation could be performed by impos-ing an external pressure, where a uid will be pumpedinto/out the channels deecting the droplets9,10; or bythe implementation of membrane valves, taking in theadvantage of PDMS elasticity1113. Lateral valve fab-rication has some constriction on the design (becausechannels cannot cross each others) but avoids complex-ity in the fabrication process due to the requirement ofmultilayers4.

Detection is a core element of the integrated system. Itwill inuence the throughput ability of droplet process-ing, as to sort droplets they will have bo screened rstby a detection method. Impedance based sensors canbe use to detect droplets, they are easy to fabricate butrequires electrodes in direct contact with the solution,making reusability more dicult. Also detection fre-quency has some limitations, ranging from Hz till almost100 Hz14. Optical detection overcomes speed limita-tion.Absorbance, chemiluminescence or uorescence canbe used to detect droplets. High speed cameras can becouplet to microscope, covering droplet frequency range(from Hz to tens of kHz) and not requiring direct con-tact with droplet system. They have a big disadvantagein cost14. Thin lm photosensors can be easily minia-turized to droplets dimension and can be arrayed givingspacial resolution15. They do not require direct contactwith uids but alignment is required.

II. MATERIALS AND METHODS

All the microuidic structures used in his work weremade out of PDMS (1:10 PDMS:curing agent) by softlithography methods. Planar electrodes were made of145 ± 2 nm layer of TiW (ρ = 253.37 µΩcm) on aglass substrate, glued and wired bonding to a PCB.The tree-dimensional electrodes consists on two mi-crouidic channels lled with a conductive uid metal,Gallium-Indium eutectic, EGa75.5In24.5 by weight (ρ ∼29.4 · 10−6 W · cm, mp = 15.7 C? ) purchased fromMerck. For the continuous phase was used a commer-cial uorocarbon solution, C9H5F15O, of HFE-7500 En-gineering Fluid (3MTMNovecTM) with added 2 wt. %008-FluoroSurfactant (RAN Biotechnologies).For the dis-persed phase generic green colorant (Globo) was used,absorbance peaks at 412 nm and 629 nm. To pumpthe uid into the microuidic channels two NE-300 sy-ringe pumps (New Era Pump Systems) were used. LE-ICA microscope camera DFC300FX at 60 fps was usedto record droplet's videos. For Pneumatic actuation, acustom system with solenoid valves actuated by On-Orocker switches was used. For the experiments involvingphotosensors, were used 50 µm × 50 µm a-Si:H Photo-conductors made by group colleague Denis Santos16.

III. RESULTS AND DISCUSSION

A microuidic system for directed evolution need asymbiosis of several modules, from cell mutation andencapsulation till sorting and recovery. GuilhermeAndrade17 had already made progresses with dropletgenerator, comparing two dierent geometries, choos-ing ow-focusing over T-junction for its symmetry (oilphase always wets the walls) and faster droplet formation(droplets form closer to the generator). Therefore in thiswork, droplet generation was assumed as guarantee -justsome experiments were made to transfer and get experi-ence. The main focus during the work was at the inte-gration, therefore experiments were conducted to nd oneor two conditions that work and not to fully characterizeeach module. Figure 1 shows the dierent modules usedand developed on this work. (module 1) Droplet gener-ator followed the ow focusing geometry, few modica-tions were made just to allow resolution on higher chan-nels; (module 3)Sorting system by dielectrophoresis wasdeveloped. First were developed planar electrodes, for itsadvantage of allowing multiple alignments with microu-idic bifurcations (also developed, module 2) followed bya nal design with two geometries of 3D electrodes madeout of liquid metal. Pneumatic sorting was also explored.After, (module 4) detection by absorbance with pho-toconductors was implemented and studied. Electricalsorting and detection by Photoconductors modules wereintegrated (module 5), presenting some challenges yet tobe solved.

A. Droplet Sorting by Dielectrophoresis

The intensity of the force depends on the particle size,medium and particle electrical conductivities and permit-tivities, and electrode geometry (see Equation 2). Therst component studies was the dependence with mediumand particle, translated by the Clausius-Mossoti func-tion3. Dierent combinations of particle/medium wereevaluated, conrming that the system water droplet/hfe7500 oil has positive DEP at kHz range. Voltages neededfor dielectrophoresis in microuidic are around hundredof Volts till kV range, with frequencies of few hundredkHz1819. At INESC-MN there wasn't any power supplyable to combine both voltage and frequency, so I builtone. As we have a DC supply (hp 6209B 0-320 V; 0-0.1A) the strategy was to design a circuit that alternate thevoltage creating a square wave with 680 Vpp. Only therelative voltage between electrodes mattered, so H-bridgeconguration suited well. Figure 2 shows circuit's maxi-mum voltage at each frequency. RMS value doesn't scalethe peak-to-peak voltage values because after 50 kHz ca-pacitive eects start to be visible, and the wave is notsquared anymore.Planar electrodes were chosen rst due to enabling dif-

ferent alignments to the microuidic channels. The de-sign process started with COMSOL simulations to eval-

Bioengeneering and Nanosystem - Extended Abstract 3

FIG. 1. Overall view of project.

FIG. 2. Frequency response for the high voltage AC circuitbuilt. Maximum peak-to-peak voltage for a DC componentlower than | ± 10|V ; and respective root mean square, rms,values.

uate ∇|E2| in vectorial and in magnitude. Three dif-ferent geometries were studied. In a general way, all ofthem achieved the same eld magnitude, what changedwas the spacial distribution of eld's gradient. It was in-tended to study two dierent phenomena for passive de-ection, sorting by hydraulic resistance (droplet as uid)and by momentum (droplet as particle). Five dierentbifurcations were combined with the planar electrodesgeometries. Experiments demonstrated sorting, enablingto infer the gradient magnitude required for droplet de-ection. Figure 3 is one example where the voltage in-crease aected the coalescence. Without applied voltagedroplets sort evenly, and for 664.00 Vpp droplets were

trapped and when other arrives, the collision coalescesboth. Figure 4 shows the moment where one droplet wasalready trapped, when other was being trapped; theycollided, coalesced and then released. For the geometrieswhere electrodes touched the droplets (Figure 4 is oneexample) droplet trapping was more severe, and no con-ditions were found for sorting without coalescence. Theseexperiments gave information about microuidic bifurca-tions, which should become longer, taking by principlemainly hydraulic resistance, as by momentum requiresspecic ow rates. In some cases where electrodes stayedin contact with solution, electrolysis was observed (evenin oil) and if in contact with droplets electrodes dissolvein seconds.

Three-dimensional electrodes were also built adaptingthe planar electrodes design. Deection was not achievedbut, for 100 µm channels droplet deceleration (Figure 5)eect was stronger than for 40 µm channels, which ismore related with droplet's diameter (∼ 80 µm).

The new electrode and microuidic design was an iter-ative process with 14 dierent designs. The rational wasto take advantage of the perpendicular gradient and min-imize the component that is against droplet ow -whichcauses stop and coalescence. Bifurcation has now 771µm long instead of 176 µm, and 300 µm width instead ofa cone shape, this was design with the intention to givedroplets time to deect for the streamlines that will owfor the sorter branch. Bifurcation angles were kept simi-lar to the previous one (150), but the default branch is10 closer to horizontal and the sort one more 10, giv-ing some extra momentum sorting eect. At bifurcation,branches have already dierent widths and the hydraulicresistance dierence is also created by their length ratios,1:2. To avoid the cases were droplets stop near to theelectrode because ow is low and the eld is high, it wasplaced the electrode right next to the opening chamber.

Bioengeneering and Nanosystem - Extended Abstract 4

a) OFF

b) 336.00 Vpp

c) 664.00 Vpp

FIG. 3. Sorting experiment for low droplet frequency gener-ation.a) without applied eld droplets sort evenly. b) lowervoltages deect droplet without coalescence. b) higher volt-ages stop the droplet at junction promoting coalescence beforesorting. φd = 0.06 µl/min; φc = 3.00 µl/min; f = 300 kHz

For the ow rates than may not work, the second designhave the electrode just next to the bifurcation havingmaximum gradient just before bifurcation. These eectwere conrmed. Figure 6 shows the dierent trajecto-ries for a single droplets with each electrode alignment.Was conrmed that the electrode near to the bifurcationrequires less voltages to guarantee deection at same con-ditions, but creates coalescence if droplets touched each

FIG. 4. Sorting experiment with low droplet frequencygeneration. φd = 0.06 µl/min; φc = 3.00 µl/min,127.00 Vpp (AC) 0.70 V (DC) f = 300 kHz.

FIG. 5. Sorting experiment with 100 µm height 3D elec-trodes. 675.00 Vpp (AC) 0.50 V (DC) f = 200 kHz;φd = 0.15 µl/min; φc = 7.50 µl/min.

other. The electrode alignment at the bifurcation's open-ing, although required more voltage, droplets were sortedwithout coalescence even without space in between. Forheight = 40 µm sorting was not achieved, just smalldroplet's deection was observed.

B. Droplet Sorting by Pneumatic Actuation

Pneumatic sorting experiments came by inspiration.They were performed by actuating the channels thatwere designed for the liquid metal electrodes, with com-pressed water. The valves were actuated manually tond condition that work in continuum. For integration,delay time had to be characterized. Sorting conditionswere evaluated using a Matlab script developed to rec-ognize droplets inside each chamber and do automaticcounting. Figure 7 shows the variation of droplets in-side each chamber with time for a specic ow rate,φd = 0.1 µl/min; φc = 10.00 µl/min, and pressure 0.36MPa. Three combinations were possible for channel di-

Bioengeneering and Nanosystem - Extended Abstract 5

a) OFF

b) 500.00 Vpp

c) 670.00 Vpp

FIG. 6. Comparison between the trajectories for two relativepositions of the electrodes with - b) and c) - and without -a)-applied eld. height = 100 µm; φd = 0.06 µl/min; φc =4.00 µl/min.

latation: top channel, bottom channel and both. Be-tween each combination both channels were de actuated,restoring the deformation to zero. The blue an greenrectangles represent the areas that droplets inside werecounted, all the yellow circles were false positives thatweren't counted. In the graph, blue line corresponds tothe counting inside blue rectangle, the same for greenline and rectangle. It was assured that both rectangleshad same area. Figure 7 tells information about the sta-bility of measurements. By repetition of the last con-

dition three times, was possible to see that the sortingpercentages are coherent. Some oscillation occurs forthe case of both electrodes o, just meaning that theow rate was not optimum for the bifurcation geometry.More conditions were tested for pneumatic sorting andbest conditions are summarized on Table I. Percentageof sorting refers to the relative amount o droplets thatwent for the bottom electrode, with the best of threepossible combinations of channel deformation. Percent-age of droplet with sorter o refers to the percentageof droplets that went by the down channel with valvesturned o. It was clear that 40 µm height channelsperformed better, having total eciency for status onand o. Three dierent ow rates were tested intend-ing to check for high low and medium droplet frequency,droplets with radius ranging 29.7 µm to 53.92 µm weresuccessfully sorted. For the bigger height, 67.7±0.3 mumseveral eects were analysed. The eect of pressure (andconsequently deformation) was compared using a owrate of φd = 0.1 µl/min; φc = 10.00 µl/min; increas-ing from the minimum eective pressure , 0.2 MPa, toalmost the maximum available, 0.36 MPa, sorting per-formance increases from 78.8 to 99.6 %. Default sort-ing was really improved by raising the continuous phaseow rate, from 29.5 % and 49.1 % for 0 %. For simi-lar total ow rates, conditions 1st (7.1 µl/min) and 4th(6 µl/min), smaller droplets sort better (just 58.9 % ofthe 53.92 µm droplets were sorted in comparison withthe 82 % of 41.92 µm droplets with more 0.6 MPa ofvalve pressure). The deection mechanism may seem abit counter intuitive, by actuating the bottom valve weare constricting the bottom channel, increasing its hy-draulic resistance, thus expecting droplets to ow by theupper channel. A possible explanation for the downwardsdeection, is that the valve expansion is not sucient toincrease hydraulic resistance on the bottom branch butthe strangulation created increases uid velocity at thebeginning of the branch, lowering the pressure and suck-ing the droplet. Also should be noted that disperse phasehave lower density than continuous phase (1000 kgm−3

vs 1614 kgm−3), and shear gradient lift force may domi-nating wall lift force20,21.

C. Droplet Detection with a-Si:H Photoconductors

Aiming at integration, squared 50 µm side photocon-ductors from Denis Santos16 were used. First was usedthe sensors with microscope conrmation, after was stud-ied how to lower the noise and nally was studied thevariation of signal with dilution of the colourant.To conrm that the signal measured was from the

droplets, signal acquisition was done under the micro-scope. Figures 8-a) and b) were chose like a ngerprint,on the top is a video frame with the microuidic channelaligned with the sensor, with one droplet already passed,one plug covering the sensor and more three coming; onthe bottom is the respective photosensor signal. It clear

Bioengeneering and Nanosystem - Extended Abstract 6

FIG. 7. Variation of sorting rate with time, for dierent valve combinations. Percentages are related to the bottom channel.Height = 67.7± 1.3 µm, φd = 0.1 µl/min; φc = 10.00 µl/min; Valve actuation Pressure 0.36 MPa.

.

TABLE I. Comparison for the best conditions found for pneumatic droplet sorting.

height (µm) φd (µl/min) φc (µl/min) Press. (MPa) Rad. (µm) Sort. On (%) Cond. Sort. O (%)

67.7± 0.30.1

7 0.3 41.92± 0.50 82.0both

29.5

100.2

38.47± 0.4178.8 0.4

0.3699.6 0

2 4 53.92± 0.75 58.9 bottom 49.1

40.4± 0.1

0.06 30.36

29.70± 0.15100 bottom 00.07 4 34.80± 1.48

0.17 9.2 32.37± 0.67

identies the rst droplet followed by all the plugs andother droplet after some time. The signal didn't drop tillzero because droplets are green and light is white. Thesetup was changed to reduce noise and for excitation lightto be at the maximum of abortion of green dye, 629 nm.By increasing the bias voltage by 10 and reducing thegain by 10, SNR changed from 2 to 36. Also, analysingin frequency for the higher gain is possible to see thepeaks for 50 Hz noise and other frequencies that are notmultiples of the droplet frequency (noise therefore).

In a real directed evolution experiment, some of thedroplets will have more concentration of a certain ana-lyte which will cause bigger signal. In order to have anestimative of the sensibility of the setup, dierent dyedilutions were measured by the droplet absorbance. Ab-

sorbance was calculated from Equation 5.

A = log10

(φiφt

)(5)

Where the incident light signal, φi, was the ampli-tude of the peak created by the chopper (completelyblocks light), and the transmitted light φt, was thesignalchopper − signaldroplet. Figure 9 shows absorbancevariation with dye's dilution, ranging from 1:0 till 1:1000,measured by the a-Si:H photoconductors and simultane-ously the microscope image. Also the green line rep-resents the signal from droplet's water in oil interface.Droplets do not present all the same size although thesystem was stable when photos were acquired, indicatingthat the dye's concentration is inuencing the uid prop-erties.This should not be a problem when using cells,

Bioengeneering and Nanosystem - Extended Abstract 7

a)

b)

FIG. 8. b) Photoconductor signal for a droplet followed by5 plugs and then other droplet, as seen on the microscopeimage, a) 5V bias; 1010 V/I gain.

because droplets form before the reaction occur. Thefour lower concentrations have a linear behaviour thatwas well characterized by a linear t (y = m · x + b,χ2 = 0.98), with m = 0.67 ± 0.39 and b = 0.15 ± 0.05.The uncertainty is big due to the noise. It's interesting tosee that for low concentrations, the eect of droplet/oilinterface was noticeable, the middle of droplet's signalis a valley that corresponds to the actual droplet signal,the two extremes are the signal caused by the dropletinterface.

D. Integration of a-Si:H Photoconductors withDielectrophoresis Actuation

Similar to the pneumatic study, here is also importantto have an idea about the delays between each module, topredict the sensor to electrode distance and to improveone of the modules if needed. Figure 10 shows the dier-ent modules: signal actuates de photoconductor22 and isamplied (rst delay), when it reach the threshold volt-age microcontroller sets the trigger pin -switches from0 to 3.3 V- (second delay), Ac source will receive thetrigger and start a burst of n cycles (third delay), this

FIG. 9. Variation of the relative signal caused by droplet withthe one from chopper (zero light) with concentration of dye.φd = 0.5 µl/min; φc = 3.00 µl/min; 30 V bias; 1010 V/Igain.

FIG. 10. Various modules evolved between the light blockingby the droplet and the actuation on the electrodes.

will generate the high voltage signal that actuates theelectrodes (fourth delay). Table II compares each mod-ule delay for often used conditions, 200 kHz and 3 Vpp.The amplication delay is very signicant (ms range) andshould be the rst to be taken in consideration, followedby signal reading and setting the trigger (∼ 50 µs). Theorder delays are at hundreds of ns range. The delaybetween microcontroller and Ac source is twice the Acsource to high voltage signal delay, but still in hundredsof ns range. the Turning of delay ∼ 6 µs should be paidattention if droplets are really close to each other, po-tently causing the following droplets to also deect. De-tection and sorting were integrated. Was chosen to startfrom the better setup to understand the limitations. A

Bioengeneering and Nanosystem - Extended Abstract 8

TABLE II. Resume of delays for the dierent modules.

light - TIA TIA - µcontr. µcontr. - AC source*µcontr. - High Volt. Signal

ON OFF7.00± 1.71 ms 51.23± 6.13 µs 391.75± 2.88 ns 620.70± 4.60 ns 6.39± 0.20 µs

microuidic structure was aligned with the photosensor,the light beam with photosensor, electrodes lled withliquid metal. Chopper as light signal at low frequency(7 Hz), picometer as bias supply with 30 V. When thesignal crosses above 0.7 V an 100 cycles with 695 Vpp 200kHz burst is produced causing a voltage drop as shown inFigure 11-a). To understand what this drop was, manualtrigger was setted and hight voltage was applied for 2 sand then turned o. Signal osets almost -6 V (Figure11-b)). Although triaxial cables were used to shield thesignal from photoconductor till Pre-Amplier (109 V/Igain), photoconductors it self, their tracks, wire bonding,pcb connectors were all exposed to the electrodes signal,which were right on top of the photosensor. To try tounderstand what was going on, both photoconductor'sterminals were measured with and without applied eld.At ground, noise jumped from few mVpp to 220 mVpp;and at the positive supply from 50 mVpp to 250 mVpp.The noise was conrmed to be from the electrodes by do-ing a frequency analysis, comparing electrodes switch onand o, appearing the frequency of burst actuation andit's harmonics when it was turned on. One should noticethat 30 V bias created 1.5 V signal, and the summed noiseamplitude is at most 0.5 V, which per si cannot cause a-6 V drop. This voltage drop corresponds to 6 nA of in-duced current, pointing that electromagnetic interferencewas occurring. In fact there is a closed common groundtrack all around photoconductor's substrate, which maywork as a closed line for induction. Some more time wasneeded to understand and debug the system.

IV. CONCLUSION AND OUTLOOK

The aim for this dissertation was to conceive an inte-grated microuidic system for droplet detection and de-ection applied for Directed evolution, major steps weredone in that direction. The work started by a state-of-art review for Droplet's microuidic applied to Di-rected Evolution, identifying the system's core modules:droplet generation and encapsulation, detection and sort-ing. Soft lithography, metal deposition and etch amongother techniques were learned enabling semi-autonomousdevice fabrication. Droplet generation was already stud-ied, serving as a starting point for the core of this work:Sorting, Detection and Integration. The key achieve-ments can be summarized:

• Flow-focusing droplet generator was used as astarting point for system development.

• A circuit to generate high voltage at dierent fre-quencies (50-700 Vpp, 1 Hz - 1MHz) was designed

a)

b)

FIG. 11. a) Photoconductor signal with time, microontrolleraquires the signal and trigger when V > Vth producing a 10cycle burst with 695 Vpp 200 kHz. b) Distortion of Photo-conductor signal when electrodes are actuated and return tonormality when turned o.

and built. It was indispensable to generate the nec-essaries AC voltages for actuation by DEP.

• Three dierent TiW planar electrodes, one 3D elec-trode and ve microuidic structures were sim-ulated, fabricated and tested. Allowing multi-ple alignments between electrodes and bifurcations,giving valuable information about optimal gradienteld magnitude, bifurcation geometry and chan-nel height for droplet sorting. Default sorting byhydraulic resistance performed better than by mo-mentum. Electrodes in contact with solution de-grade faster and usable voltage range is more lim-

Bioengeneering and Nanosystem - Extended Abstract 9

ited.

• Two completely novel 3D electrode/bifurcationgeometries were designed, simulated and tested,achieving sorting without coalescence. One geom-etry performs better for higher ow rates, and re-quires less voltages; the other allows sorting with-out coalescence for droplets at low velocity andwithout space in between.

• Sorting by pneumatic actuation was implementedwith success. Performance was evaluated in func-tion of channel height, droplets radius and owrate. Valve's actuation delay were also character-ized, showing that response time of solenoid valveswas the limitation factor.

• Image analysis script was developed to allowdroplet counting and size measurement. Still re-quiring some characterization to proper validateand use the script "blindly".

• 50 µm side a-Si:H thin lm photoconductors wereused for droplet's optical detection by absorbance.They Oer the advantage of faster measurements(till at least 500 Hz, limited by the setup), cus-tomized to t microuidic channels and droplet di-mensions.

• Preliminary study of light intensity variation withdroplet dye's concentration was performed using 50µm side a-Si:H thin lm photoconductors.

• Transimpedance Amplier circuit with 200 dB gainwas designed and implemented to allow photo-conductor's signal acquisition and processing on aportable platform.

• System integration with detection and electricalsorting was implemented. Delays between signalacquisition and electric actuation were character-ized. Still needing some solution to avoid signaldistortion caused by high voltage burst required forDEP actuation.

At this point each module is working. Still some workis left to be done, my future work suggestions are:

• Firstly remove the signal distortion caused by highvoltage burst for DEP. This may inuence thesetup, or even the actuation method therefrom be-ing the rst problem to approach. Photoconductorstracks, wire bonding and PCB connector, weren'tshielded. The use of optic ber23 to allow photo-sensor shielding from electrodes without having toput away the photoconductor from the light signal,may solve the problem. Acoustic sorting could bethought as an alternative to electrical sorting if in-compatibility is conrmed -but i don't think that'sthe case.

• Droplet generator should be proper characterizedwith microscope and photosensor in parallel to eval-uate droplet's size and frequency, also giving infor-mation about real droplet speed inside the chan-nels, which will be helpful when designed photo-sensors's array.

• A customized setup should be developed to studysorting. Mainly to assure reproducibility, some-times disturbed by outlets tubes's relative heightvariation. A box with several heights should bebuilt to study the eect on passive deection, andto guarantee always the same conditions.

• To perform detection and sorting integration,burst's duration and timming will play a key role.The study would benet if some image acquisitionis performed at same time, because gives informa-tion about droplet position when the actuation isoccurring, and not just a binary response for sortingsuccess, as the integration of another photosensorwould give.

• High Voltage AC Circuit may be miniaturized usinglow power RF oscillators to generate the signal24.

REFERENCES

1Liu Packer, Michael S. and David R. Methods for the directedevolution of proteins. Nature Reviews Genetics, 16(379), 2015.

2Manish Kumar Tiwari, Ranjitha Singh, Raushan Kumar Singh,In-Won Kim, and Jung-Kul Lee. Computational approaches forrational design of proteins with novel functionalities. Compu-

tational and Structural Biotechnology Journal, 2(3):e201204002,2012.

3Amir Aharoni, Karena Thieme, Cecilia P. C. Chiu, Sabrina Bu-chini, Luke L. Lairson, Hongming Chen, Natalie C. J. Stry-nadka, Warren W. Wakarchuk, and Stephen G. Withers. High-throughput screening methodology for the directed evolution ofglycosyltransferases. Nature Methods, 3:609 EP , Jul 2006. Ar-ticle.

4Heng-Dong Xi, Hao Zheng, Wei Guo, Alfonso M. Ganan-Calvo,Ye Ai, Chia-Wen Tsao, Jun Zhou, Weihua Li, Yanyi Huang, Nam-Trung Nguyen, and Say Hwa Tan. Active droplet sorting in mi-crouidics: a review. Lab Chip, 17:751771, 2017.

5Liquid plugs are dened as large droplets in contact with all fourwall of a microchannel.

6Cheng Qian, Haibo Huang, Liguo Chen, Xiangpeng Li, ZunbiaoGe, Tao Chen, Zhan Yang, and Lining Sun. Dielectrophoresisfor bioparticle manipulation. International Journal of Molecular

Sciences, 15(10):1828118309, Oct 2014.7Jingmei Li, Zhou Liu, Haibo Huang, and Ho Cheung Shum.Shielding electric elds to prevent coalescence of emulsions inmicrouidic channels using a 3d metallic coil. Micromachines,6(10):14591468, 2015.

8Jeremy J. Agresti, Eugene Antipov, Adam R. Abate, KeunhoAhn, Amy C. Rowat, Jean-Christophe Baret, Manuel Marquez,Alexander M. Klibanov, Andrew D. Griths, and David A.Weitz. Ultrahigh-throughput screening in drop-based microu-idics for directed evolution. Proceedings of the National Academyof Sciences, 107(9):40044009, 2010.

9Liang Wu, Pu Chen, Yingsong Dong, Xiaojun Feng, and Bi-FengLiu. Encapsulation of single cells on a microuidic device in-

Bioengeneering and Nanosystem - Extended Abstract 10

tegrating droplet generation with uorescence-activated dropletsorting. Biomedical Microdevices, 15(3):553560, Jun.

10Zhenning Cao, Fangyuan Chen, Ning Bao, Huacheng He,Peisheng Xu, Saikat Jana, Sunghwan Jung, Hongzhen Lian, andChang Lu. Droplet sorting based on the number of encapsulatedparticles using a solenoid valve. Lab Chip, 13:171178, 2013.

11A. R. Abate and D. A. Weitz. Single-layer membrane valvesfor elastomeric microuidic devices. Applied Physics Letters,92(24):243509, 2008.

12Dong Hyun Yoon, Junichi Ito, Tetsushi Sekiguchi, and ShuichiShoji. Active and precise control of microdroplet division usinghorizontal pneumatic valves in bifurcating microchannel. 4:197205, 06 2013.

13Adam R. Abate, Jeremy J. Agresti, and David A. Weitz. Mi-crouidic sorting with high-speed single-layer membrane valves.Applied Physics Letters, 96(20):203509, 2010.

14Ying Zhu and Qun Fang. Analytical detection techniques fordroplet microuidics a review. Analytica Chimica Acta, 787:24 35, 2013.

15D. R. Santos, R. R. G. Soares, V. Chu, and J. P. Conde. Per-formance of hydrogenated amorphous silicon thin lm photo-sensors at ultra-low light levels: Towards attomole sensitivitiesin lab-on-chip biosensing applications. IEEE Sensors Journal,17(21):68956903, Nov 2017.

16Denis Roda dos Santos. Integrated biosensors for Lab-on-a-Chip

platforms. PhD thesis, Instituto Superior Técnico, 2018.17Guilherme Araújo de Andrade. Droplet Microuidics Systems

for Dyrected Evolution. Instituto Superior Técnico, Master dis-sertation, 2017.

18Keunho Ahn, Charles Kerbage, Tom P. Hunt, R Westervelt, Dar-ren Link, and DAWeitz. Dielectrophoretic manipulation of dropsfor high-speed microuidic sorting devices. 88:024104024104, 012006.

19After 100kHz double layer eects at electrodes surface start todisapear25.

20Hamed Amini, Wonhee Lee, and Dino Di Carlo. Inertial mi-crouidic physics. Lab Chip, 14:27392761, 2014.

21Matas, Jp., Morris, Jf., and Guazzelli, E. Lateral forces on asphere. Oil & Gas Science and Technology - Rev. IFP, 59(1):5970, 2004.

22Here could been a delay that was not characterized.23N. . Nguyen, S. Lassemono, and F. A. Chollet. Optical detec-tion for droplet size control in microuidic droplet-based analysissystems. In The 13th International Conference on Solid-State

Sensors, Actuators and Microsystems, 2005. Digest of Technical

Papers. TRANSDUCERS '05., volume 2, pages 15571560 Vol.2, June 2005.

24Y.-Y. Jau, F. M. Benito, H. Partner, and P. D. D. Schwindt. Lowpower high-performance radio frequency oscillator for driving iontraps. Review of Scientic Instruments, 82(2):023118, 2011.

25Peter R. C. Gascoyne and Jody Vykoukal. Particle separation bydielectrophoresis. Electrophoresis, 23(13):19731983, Jul 2002.12210248[pmid].