High-Content Optical Codes for Protecting Rapid Diagnostic Testsfrom CounterfeitingOnur Gokce,* Cristina Mercandetti, and Emmanuel Delamarche*

IBM Research, Zurich, 8803 Ruschlikon Switzerland

*S Supporting Information

ABSTRACT: Warnings and reports on counterfeit diagnosticdevices are released several times a year by regulators and publichealth agencies. Unfortunately, mishandling, altering, andcounterfeiting point-of-care diagnostics (POCDs) and rapiddiagnostic tests (RDTs) is lucrative, relatively simple and canlead to devastating consequences. Here, we demonstrate how toimplement optical security codes in silicon- and nitrocellulose-based flow paths for device authentication using a smartphone.The codes are created by inkjet spotting inks directly onnitrocellulose or on micropillars. Codes containing up to 32elements per mm2 and 8 colors can encode as many as 1045

combinations. Codes on silicon micropillars can be erased bysetting a continuous flow path across the entire array of codeelements or for nitrocellulose by simply wicking a liquid across the code. Static or labile code elements can further be formed onnitrocellulose to create a hidden code using poly(ethylene glycol) (PEG) or glycerol additives to the inks. More advanced codeshaving a specific deletion sequence can also be created in silicon microfluidic devices using an array of passive routing nodes,which activate in a particular, programmable sequence. Such codes are simple to fabricate, easy to view, and efficient in codinginformation; they can be ideally used in combination with information on a package to protect diagnostic devices fromcounterfeiting.

Infectious diseases claimed 8.23 million lives in 2016, mostlyfrom the developing world.1,2 An estimated 1.2 million

deaths from malaria, tuberculosis, syphilis, and pneumoniacould have been prevented with adequate diagnostics,3 butunfortunately, 70% of Africa and 40% of Asia did not haveaccess to sufficient infrastructure (electricity, clean water,physical infrastructure, skilled staff) to run clinical tests.4 Thisis why POCDs, and in particular low-cost, simple-to-fabricate,and simple-to-operate RDTs had a significant impact on globalhealth in detecting diseases in resource-limited settings,5,6

mapping infectious diseases,7,8 reducing presumptive treat-ments,9,10 and screening blood before donation and trans-fusion.11−13 Not surprisingly, the adoption of RDTs isincreasing worldwide with the help of internationally fundedprograms.5,14

Unfortunately, RDTs also attracted the attention of criminalswho started to inject poor-quality products into the market. Asan example, the World Health Organization (WHO) reportedthat counterfeit RDTs for visceral leishmaniasis started tocirculate in the Indian subcontinent soon after an eliminationinitiative against visceral leishmaniasis was started there.15 TheWHO estimates that more than 8% of the medical devices incirculation in 2010 were counterfeit. This figure is expected toincrease,16 in part because of the growing importance of “onlinepharmacies”.17,18

Counterfeit RDTs are typically nonfunctional productsmimicking the look of genuine tests. They can also consist of

genuine tests with falsified labels and expiration dates. Forexample, pregnancy tests were labeled as HIV tests and sold toblood banks in West Bengal,19 expired HIV tests were resold inKinshasa with altered expiration dates,20 and fake eboladiagnostic kits appeared in Lagos during the 2014 outbreakbefore genuine ebola RDTs ever existed.21 Counterfeit RDTskill individuals and contribute to the spreading of infectiousdiseases due to misdiagnosis. They also foster mistrust inhealth-care workers, who may stop relying on valid testresults.22 Today’s fragmented and complex global supply chainsfurther challenge the viability of RDTs, which frequentlyexperience mishandling during distribution and storage. Theiravailability can be sporadic, sometimes with a limited supplyand at other times in excess, resulting in tests that expire instorage.23

These alarming facts call for a widespread adoption of strongand low-cost security features for authenticating and trackingRDTs. Currently, the two main strategies used in the healthcareindustry against counterfeiting are (1) “security printing”,which consists of methods to create hard-to-replicate securityfeatures such as holograms or tags with special inks, and (2)“digital tagging”, where the information stored in digital tags

Received: February 20, 2018Accepted: May 16, 2018

Article

pubs.acs.org/acCite This: Anal. Chem. XXXX, XXX, XXX−XXX

© XXXX American Chemical Society A DOI: 10.1021/acs.analchem.8b00826Anal. Chem. XXXX, XXX, XXX−XXX

provides the security.24,25 Features produced by securityprinting tend to be copied over time. Holograms, for example,can be reproduced at low quality without a lay person noticingit.26 Hard-to-fabricate security tags, such as photonic andplasmonic micro/nanostructures,27,28 can be integrated intomicrofabricated RDTs, but these tags typically requireexpensive and time-consuming fabrication processes, whichmakes it challenging to label each product with a unique tag.Digital tagging is a promising approach for securing RDTsbecause a large security code (a couple of kilobytes) that isunique for each product can be stored on low-cost RFID chipsor simply printed as QR codes. These codes can be read andcrosschecked using the Internet or a cellular network.25 The EUdirective on falsified medicines will require all drugs to belabeled on their packages with unique identification tags byFebruary 2019.29 The FDA global unique device identificationdatabase will similarly require all medical products to beindividually labeled by September 2018.30 However, securityfeatures on a package secure the package, not necessarily theproduct it contains. Scavenged packages from used productshave been used to repack counterfeit medical products.31

Additionally, security features on packages are exposed and canbe copied anywhere in the supply chain.A stronger security can be achieved by embedding a digital

security code in the functional component of a product, whichcannot be transferred to counterfeits. This approach wasrecently demonstrated by different groups using differentmethods. Scherr et al. patterned control antibodies on anitrocellulose membrane to create a QR code that appearedwhen the diagnostic test was run.32 This method howeverrequired a large area for the QR code (12.5 × 12.5 mm2), didnot allow to authenticate a device before using it, andantibodies used as code generating molecules might beexpensive. In a similar method, Park et al. patterned DNAprobes on a DNA microarray chip to produce a QR code uponhybridization with complementary fluorescent probes.33 Han etal. placed fluorescent polymer microtags carrying QR codesinside individual drug capsules,34 but the drug capsule had to bedestroyed in order to access the code. Ciftlik et al. imprinted

fluorescent QR codes in a parylene-C bonding layer inmicrofluidic chips using UV irradiation.35 This technology hasa high precision capability but is limited to chips packagedusing parylene-C.Our approach on securing RDTs is based on introducing

unique and compact optical security codes into microfluidicdevices or on nitrocellulose flow paths. We specifically focus onimplementing static and dynamic codes where reading the codebefore or during a test provides sufficient complexity whencomplemented with a QR code on a package. The optical codesare easy to write with an inkjet spotter using low-cost dyes andeasy to decode using a smartphone equipped with a simple clip-on macro lens. The strategy can allow authenticating RDTswith or without network connectivity (Figure 1). Withintermittent or continuous connectivity, devices can be trackedthrough the supply chain, their use can be logged or incidentsmay be reported to regulators.

■ EXPERIMENTAL SECTION

Materials and Software. Water deionized using aMillipore system was used for all inks. Inks contained either10% PEG (molecular weight: 3000, Sigma-Aldrich) or 20%(w:v) glycerol (Sigma-Aldrich) in addition to amaranth(providing a magenta-like color, Sigma-Aldrich), brilliant blueFCF (providing a cyan-like color, Sigma-Aldrich), tartrazine(yellow, Sigma-Aldrich), or brilliant black BN (Sigma-Aldrich)dyes. Inks for the codes on Si micropillars contained 20 mgmL−1 amaranth or 10 mg mL−1 brilliant blue FCF. Inks fornitrocellulose membrane (BioTrace NT, pore size: 0.2 μm, PallLife Sciences) contained 10 mg mL−1 amaranth, 3.3 mg mL−1

brilliant blue FCF, or 20 mg mL−1 tartrazine. The ink depositedon the 3D microfluidic device for multiple deletion stepscontained 1.0 mg mL−1 brilliant black BN. The Si microfluidicchips were filled with an aqueous 0.1% (w:v) Tween-20(Sigma-Aldrich) solution. The nitrocellulose membrane waswetted with an aqueous 0.1% sodium dodecyl sulfate (SDS,Fluka) solution in experiments involving the deletion of codeelements.

Figure 1. An optical security code on a POCD device used in combination with security information on the package of the device may lead to securesupply chains for tracing and authentication of POCDs across various network conditions in time and space. Such codes can be produced at amanufacturing level and stored to be accessible by end-users such as doctors or patients using for example a smartphone.

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.8b00826Anal. Chem. XXXX, XXX, XXX−XXX

B

Images were captured using a digital camera (Leica MC170HD) attached to a stereoscopic microscope (Leica MX16), andprocessed using Fiji36 for brightness and contrast. Data analysisand graphs were done using Matlab.The protocol for validating optical codes was implemented

(1) for the client application in JavaScript to run on the webbrowsers of smartphones and (2) for the server in Python usingFlask microframework running on a cloud service (Bluemix).Fabrication. The microfluidic chips were fabricated on Si

wafers having 600 nm-thick thermally grown SiO2 oxide layerusing standard photolithography and deep reactive-ion etching(DRIE, Figure 2c). Briefly, the microchannels were patterned

on the oxide layer using a 1.2-μm-thick positive-tonephotoresist layer (AZ 6612) and glass/chromium masks. Theoxide layer was etched using buffered hydrofluoric acid (BHF)after the removal of photoresist residues by exposing thesurface to a mild plasma. Subsequently, the photoresist layerwas stripped in a plasma asher and a new layer of photoresistwas applied to pattern the micropillars. Twenty μm of theexposed Si substrate was vertically etched using DRIE (AlcatelAMS 200). After stripping the photoresist, the substrate wasfurther etched for 20 μm using DRIE and the oxide layer as amask. Finally, the processed wafers were diced (ESEC 8803Dicing Saw). The quality of the fabrication was verified usingscanning electron microscopy (SEM, Zeiss Leo 1550). Afterwriting the codes on the micropillars, the chips were laminatedusing a negative dry-film resist (DF-1050, Engineered MaterialsSystems, Inc.) at 45 °C on a hot plate.The fabrication of the 3D microfluidic chips was similar

except for an additional photolithography and DRIE process tofabricate the bottom microfluidic layer at the back side of 400-μm-thick wafers. The microfluidic layers were formed byetching 20 μm using DRIE. The vias were opened by etchingthrough the wafer. After fabrication, the chips were silanized for2 min using 0.1% (v:v) tricholoro(octyl)silane (Sigma-Aldrich)

in heptane (Sigma-Aldrich). The chips were sealed with 3 mm-thick layers of polydimethylsiloxane (PDMS, Dow CorningSylgard 184) prior to use. A syringe pump (Kent ScientificGenie) was used to fill these devices with water at a rate of 1.5μL min−1.

Inkjet Spotting. The reagents were deposited using aNanoPlotter 2.1 inkjet spotter equipped with a NanoTip-Jpiezoelectric pipetting tip (GeSiM, Dresden). The spotter wascontrolled using a custom spotting program. Prior to spottinginks, the exact position of the Si chips on the tray of the inkjetspotter needed to be determined for precise targeting of the inkonto Si micropillars. This was done by spotting test spots ontwo grids placed at the opposite ends of the Si chip andmeasuring the lateral offset of the test spots. The height of thetarget surfaces was measured with the tactile Z-sensor suppliedwith the spotter. Spotting was done at a distance of 0.25 mmfrom the Si surface, or 0.5 mm from the nitrocellulosemembrane. The piezoelectric tips were actuated using 60 to 90V amplitude and 20 to 50 μs pulse width. The deposition ratewas 100 Hz for spotting on Si and 1000 Hz for thenitrocellulose membrane.The volume of dispensed droplets was characterized by

measuring their diameter using a stroboscopic image and foundto be 386 ± 64 pL. Single droplets of ink containing PEG weredeposited on Si micropillars. Inks containing PEG or glycerolwere deposited in 20 or 5 droplets respectively on thenitrocellulose membrane to generate the code elements withprimary colors. Thirty or 8 additional droplets of a second inkcontaining PEG or glycerol were spotted over the primarycolors to generate the code elements with secondary colors(green, purple, red).The error rate associated with placing code elements on

micropillars was characterized by spotting single droplets ofamaranth ink containing PEG on an array of micropillars andsubsequently counting uncoated ones.

Analysis of Code Properties. In dynamic codes that arepartially deleted (e.g., Figures 4 and 5), the remaining codecomprises a subset of the initial code elements. Therefore, for adynamic code with n code elements of which k of them are set,the remaining code can have k code elements of which l ofthem can be set. The total number of codes, Spar(n), that can bewritten with this scheme is

∑ ∑== =

⎜ ⎟⎛⎝⎜⎜⎛⎝

⎞⎠

⎛⎝⎜

⎞⎠⎟⎞⎠⎟⎟S n

nk

kl

( )k

n

l

k

par0 0 (1)

which simplifies to Spar(n)=3n. Therefore, codes with partial

deletion have an equivalent capacity of log2 Spar(n) = n log2 3bits.Deleting the code in multiple steps (e.g., Figures 5 and 6)

adds log2 D(b) extra bits to the capacity of the code, whereD(b) is the number of different ways to delete the code in bdeletion steps. D(b) can be represented in a recursive function

∑= + −=

− ⎛⎝⎜

⎞⎠⎟D b

bd

D b d( ) 2 ( )d

b

1

1

(2)

for b > 0, where D(0) = 0 and d is the number of nodesactivated simultaneously in one deletion step. The total numberof nodes that need to be activated to achieve D(b) differentdeletions is

Figure 2. Implementation of static optical security codes onmicrostructures, which can be patterned inside microfluidic chips.(a) Droplets of ink are inkjet spotted on micropillars to create singlecode elements after drying, as seen in the SEM image. (b) Microscopeimage showing a multicolor code composed of 14 × 14 elements anddisplaying various items (encoded numbers, text, checker pattern) and2 alignment marks. (c) Fabrication steps for creating the code on Simicropillars using standard optical lithography, etching, inkjet spotting,and lamination techniques.

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.8b00826Anal. Chem. XXXX, XXX, XXX−XXX

C

∑= + − + −=

− ⎛⎝⎜

⎛⎝⎜

⎞⎠⎟

⎞⎠⎟S b b d

bd

D b d S b d( ) ( ) ( )d

b

node1

1

node(3)

for b ≥ 0. Then, the total number of spots to write all of thepossible codes with n code elements and b deletion steps is

∑= +=

⎜ ⎟⎛⎝⎜

⎛⎝

⎞⎠

⎞⎠⎟S n b k

nk

D b S b( , ) ( ) ( )k

n

spot0

node(4)

Assuming that patterning each spot takes tspot amount of timeto move the inkjet head and to deposit the dye, the averagetime to write a code is the time it takes to write all possiblecodes divided by the total number of writable codes

=·

∑ =⎜ ⎟⎛⎝

⎞⎠

t n bt S n b

nk

D b( , )

( , )

( )kn

codespot spot

0 (5)

Dynamic codes with multistep deletion implemented asshown in Figure 6 require extra area for the array of nodes thathas a size of b × (b + 1) nodes. The total footprint of thesedynamic codes, Atotal(n,b), can be approximated as

= + +A n b nA b b A( , ) ( )total bit2

node (6)

where Abit is the area occupied by 1 code element and Anode isby 1 node.

■ RESULTS AND DISCUSSIONOptical Codes Formed with an Array of Micropillars in

Silicon. A 2D array of code elements, which can be seen by eyeor using a simple optical reader, is an efficient and widespreadapproach to providing information for the authentication ofproducts.37 We chose to form such code elements by inkjetspotting dyes on an array of micropillars (Figure 2). Inkjetspotters are broadly used for integrating biological reagents todiagnostic devices and can easily dispense aqueous solutions athigh-throughput, with a lateral positioning accuracy in themicrometer range, and with droplets of picoliters in volume.38

Spread of the spotted ink on planar surfaces limits theminimum separation between adjacent code elements, andconsequently the maximum spatial density of an optical codethat can be written. Spotting the ink inside depressedmicrostructures, for example, microwells, circumvents thisspreading; however, we observed that the dye accumulated atthe corners inside the microwells upon drying,39 and led tononuniform code elements. Alternatively, when the ink wasspotted on top of micropillars, i.e. raised microstructures,contact line pinning at the periphery of the top surfaceprevented the ink from spreading,40 and yielded well-definedcircular code elements. PEG included in the ink as a dryingagent suppressed the coffee ring effect,41,42 and furtherimproved the optical quality of the codes (Figure 2a). Weused micropillars with a diameter of 150 μm, which were ableto hold a single droplet of ink (∼400 pL). The pillars werespaced at least 25 μm apart to tolerate possible failures inpinning of the ink. These settings translated into a code havingup to 32 elements per mm2 and a patterning error rate of 0.2%.Such low error rates can be corrected by including additionalcode elements and using algorithms for error-correction such asthe Reed-Solomon algorithm.43

Code elements were arranged to store binary data or human-readable alphanumeric characters (Figure 2b). A securityprinting strategy can also be utilized to raise the manufacturing

barrier against counterfeiters. This strategy can use materials forexample with special spectral properties such as upconversionnanoparticles,44 lanthanide-organic frameworks,45 phase-changenanoparticles,46 or solid-state tunable-fluorescence hetero(n)-rotaxanes.47

Micropillars are easily fabricated in a Si substratesimultaneously with microfluidic structures using conventionallithography and etching techniques. A 2-mask fabricationprocess can produce pillars of intermediate height to allowinteraction between the code elements and a liquid filling themicrofluidic structures (Figure 2c). Writing a code can inprinciple be done during the step used for integrating reagentsinto RDTs.

Dynamic Optical Security Codes Inside Microchan-nels. Reuse of single-use medical products after sterilizationand reprocessing is surprisingly very common in hospitalsworldwide and presents ethical and hygienic issues.48,49 In asimilar way, counterfeiters may scavenge, clean, repackage andsell used RDTs again. An optical security code as demonstratedabove is static and cannot readily indicate if a device has alreadybeen used. However, a dynamic code that is deleted with thesample may warn a user against the reuse of a device would thecode not be present.We implemented dynamic codes by depositing dyes onto

micropillars placed inside Si microchannels (Figure 3a and b).

The code can be placed in a flow path parallel to the testchannel or downstream to the test region to avoid potentialcross-reactivity of the code with the test. The native oxide ofthe Si layer and the hydrophilic DFR sealing layer make themicrochannel capillary active. It took less than 10 s for anaqueous solution to fill the microchannel spontaneously anderase the code by dissolving the dye in the example displayed inFigure 3c. This code only occupies 4 mm2 of area andnecessitates as little as 110 nL of sample to be completelyerased. Such dynamic codes have a slightly lower spatial densitythan static codes due to the extra space required by the wallsused to guide the liquid across each code element. The code inFigure 3c specifically includes 20-μm-wide walls that are 15 μm

Figure 3. Dynamic optical security code that interacts with a liquidfilling a Si microfluidic channel. (a) 3D illustration of 20-μm-highmicropillars placed in a meandering, 40-μm-deep, 180-μm-wide flowpath and (b) the SEM image showing part of such a code where somemicropillars were coated with a layer of dye (amaranth, darker pillars).(c) Time series of microscope images showing a dynamic code with 10× 10 elements deleted by an aqueous solution.

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.8b00826Anal. Chem. XXXX, XXX, XXX−XXX

D

away from the pillars, which results in a density of 25 codeelements per mm2.Exploiting the interaction between a liquid sample and code

elements brings many options for improving security of thedevices. Water-soluble and insoluble code elements can becombined to create codes with permanent and erasable codeelements that yield a different code pattern after the flow ofsample. An invisible and therefore copy-proof code can appeartransiently after a reaction with the liquid sample and fade awayto prevent reuse.50 Test results can also be encoded in dynamiccodes.51

Partial Deletion of Dynamic Optical Security Codeson Nitrocellulose Membranes. The majority of RDTscurrently used worldwide employ the principle of “lateral flowassays” and a nitrocellulose substrate.6 These tests areparticularly easy to counterfeit. Taking this into consideration,we developed dynamic optical codes for nitrocellulose, whichcan be partially deleted as an additional feature. Such codes notonly enable the detection of used tests, but also increase theamount of information that can be stored. Figure 4a and Video

S-1 show the transformation of an 8-color code on nitro-cellulose within 5 min as an aqueous solution wicks themembrane. This duration is well within the typical timerequired by biochemical assays.We found that code elements patterned using inks containing

long polymers such as PEG-3000 are retained during migrationof liquids on the nitrocellulose. Labile code elements wereproduced by replacing the PEG with glycerol. Using additivesinstead of water-soluble and nonsoluble dyes is simpler forimplementing partially deleted codes because (1) the same dyescan be used for both types of code elements, which makes theiroptical properties indistinguishable, (2) both inks have similarviscosity and surface tension, and produce similar patterns, and(3) aqueous solutions are easier to dispense with inkjet spottingthan organic solvent based solutions.It was possible to produce codes on a nitrocellulose

membrane with as many as 100 code elements per mm2 (100μm diameter, 100 μm pitch, Figure S-1). Such codes are in linewith the optical resolution of commercially available opticalreaders for POCD devices.52,53 We also simplified some codesto make them readable with smartphones. Specifically, asmartphone camera equipped with a low-cost clip-on macrolens giving a 2× magnification was able to resolve optical codescomposed of 500-μm-diameter elements spaced by 500 μm (4code elements per mm2, Figure 4b).

Reading and Decoding Optical Security Codes. Opticalsecurity codes on a reflective surface such as Si or white supportsuch as a nitrocellulose membrane have good contrast and caneasily be captured with a smartphone camera. We developed asoftware for scanning optical codes, decoding them, andverifying the authenticity of tests. The optical security codeswere detected using a method based on template matching.Briefly, asymmetric alignment marks were searched within thescan area (red rectangle seen on the phone display in Figure4b) to detect the position, size, and rotation of the code. Thealignment marks also provided reference points for calibrationof the colors to account for variations with ambient light. Thevalue of each code element was subsequently extracted using itsposition and color. This process took around 50 ms on average,allowing for a hand-held operation.The capacity of the codes might be limited by the area

available on the device. An efficient way of using compactoptical codes is to couple the code to additional information ona package. Such information can be stored in a high-capacityQR-code and relate to the type of test, its expiration date,expected geographical area of use, etc. Here, the QR-codecontains the address of a server, a shipment identifier, andencrypted data containing the expected value of the opticalsecurity code (Figure 4c). Once the QR-code is scanned using asmartphone, the phone connects to the server to receive adecryption key, which is used to decipher the encrypted part ofthe QR-code. The server also sends information about the testspecifications and operating instructions. A test is validated ifthe decrypted security code from the QR-code matches theread optical security code. This strategy ensures that a testbelongs to the intended package, hence essential informationsuch as the type of the test and the expiration date istrustworthy. The server can supply information related to ashipment so that authentication of a first test can enableauthentication of additional tests without further need forconnectivity.

Information Capacity of Optical Security Codes. Figure5a recapitulates different types of codes presented in this work.

Figure 4. A dynamic optical code written on a nitrocellulosemembrane in a mock-up RDT and reading of the code using asmartphone. (a) Time series of microscope images showing the partialdeletion of a code by an aqueous solution wicking the nitrocellulosemembrane. This code has 32 elements (4 × 8 matrix) and covers a 3 ×5 mm2 area including the alignment marks. (b) Scanning of the codeusing a smartphone equipped with a macro lens providing a 2×magnification. (c) Photographic image of the mock-up RDT having aQR code on the package, which contains a unique product identifier,and the optical security code on the nitrocellulose membrane slightlydownstream to potential test and control areas.

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.8b00826Anal. Chem. XXXX, XXX, XXX−XXX

E

Static codes do not evolve during a test whereas dynamic codesinteract with a sample and change. This change can be a partialor complete deletion of the code. Moreover, code elementsmay be deleted in multiple steps to add further complexity tothe codes. When using a single color for static codes, 1 codeelement corresponds to 1 bit of information (Figure 5b, solidblack line). Incorporating Ncolor colors to the code elementsincreases the capacity of the code element by log2(Ncolor) (for 8colors see Figure 5b, solid blue line). Dynamic codes with 1-step complete deletion have the same capacity as the staticcodes.Dynamic codes that are partially deleted in one step have

1.585 times more capacity compared to static codes with thesame number of code elements (Figure 5b, dashed black line,eq 1). Dynamic codes with multiple colors gain extra capacitysimilarly to static codes (Figure 5b, dashed blue line). Thedynamic code on the nitrocellulose membrane in Figure 4astores 152 bits of information (at least 1045 combinations) in 32code elements. The WHO estimated that 314 million RDTs formalaria were used in 2014.54 A dynamic code having as few as24 single-color elements and 1 partial deletion step can easilyprovide a specific code for 1000 times this number of RDTs.Last example in Figure 5a illustrates a dynamic code that is

deleted in 4 steps, where one row of the code is affected at eachdeletion step. In general, the order of deleted rows can change,multiple rows can be deleted simultaneously, or some rows mayremain unchanged. These possibilities add a time-dimension tothe transformation of the optical security codes. Differentcombinations to delete a code add 7.2 extra bits of informationwith 4 deletion steps, and 20.1 extra bits with 8 deletion steps(Figure 5b, dashed lines, eq 2). An implementation for creatingdynamic codes with multiple deletion steps is introducedbelow.Implementation of Multiple Deletion Steps. A micro-

fluidic network analogous to the crossbar switches inelectronics can be implemented using a 3D architecture,where a top microfluidic layer, which contains chambers forcode elements and an array of routing nodes, is connected to abottom microfluidic layer through vias (Figure 6a). Eachrouting node has a capillary stop valve,55 therefore a liquidcannot enter the node unless it is activated by spotting a

Figure 5. Different types of optical security codes and theircorresponding information capacity. (a) Illustration of the types ofoptical security codes included in this work. The brackets on themultistep partial deletion example indicate the row of the code that ispartially deleted at a particular step. (b) The information capacity ofcodes shown in panel (a), plotted as a function of the number of codeelements, colors, and deletion steps for removing some of theelements. Black lines indicate codes having 1 color and that are static(solid line) or have from 1 to 8 partial deletion steps (dashed anddotted lines). The capacity of codes including 8 colors are indicatedusing blue lines.

Figure 6. Implementation of an optical code that can be deleted in 4steps by a liquid in a 3D microfluidic architecture. (a) 3Drepresentation of the microfluidic flow path showing the top andbottom layers and the connecting vias. (b) The top view of the 3Darchitecture detailing the microfluidic design and its components. Thetop microfluidic layer contains chambers with the code elements, anarray of routing nodes, and 1 inlet and 4 possible outlet microchannelsfor the liquid. The bottom microfluidic layer contains flow diodes.Arrows indicate the flow direction of the liquid during 2 consecutivedeletion steps (first step, with black, and the second with orangearrows). The inset details components of the node. (c) Microscopyimages showing the deletion of code elements by water passingsequentially through activated routing nodes. This architecture ismicrofabricated in Si. Blue rectangles in the first frame indicate thenodes that are activated by spotting black dye inside them. (d) Thegain in information capacity when using multistep codes as a functionof the number of deletion steps, b, compared to static codes having the

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.8b00826Anal. Chem. XXXX, XXX, XXX−XXX

F

wetting agent next to its valve. If activated, the liquid can passthrough the node and proceed to the downstream chamberwhere it can delete code elements. The flow diodes56 at thebottom microfluidic layer force the liquid to travel only todownstream chambers. Activation of multiple nodes set thesequence of the chambers to be filled (Figure 6b). Figure 6cand Video S-2 show a liquid filling such an architecture anderasing the code in 4 steps. In this example, the microchannelsare 100 μm wide and 20 μm deep. The width of the capillaryvalves is 10 μm and the constrictions of the flow diodes are 25μm. The vias have a diameter of 50 μm and a length of 360 μm.The dimensions of the spotting areas for node activation are200 × 100 μm2 and can easily be targeted with the inkjetspotter.Multiple-step deletion of codes using this implementation

has interesting outcomes. First, extra information is stored inhow the code evolves, which augments the capacity of thecodes especially of those with few elements (Figure 6d, eq 2).Second, this extra information can be extracted only byobserving the transformation of the code during the use of atest. Therefore, the array of nodes can be completely concealedfrom the user and/or transparent reagents can be used foractivating the nodes. This adds a covert security feature to theoptical codes and complicates copying attempts by adversaries.The array of nodes uses additional chip area but this penaltydiminishes for codes having high-capacity (Figure 6e, eq 6).Additionally, the mean time to write such codes decreasesbecause spotting reagents on the array of nodes gives morecombinations than when the same number of spots would beused in binary coding (Figure 6f, eq 5).

■ CONCLUSIONCounterfeiting is a huge activity, worth hundreds of billions ofdollars and spanning nearly all sectors of industry.57 It is muchmore widespread than known to the public and is gettingprogressively worse. Counterfeiters are very creative anddetermined, whereas end-users are generally not able todiscriminate fake products. This situation can have particularlydramatic consequences when counterfeit diagnostic devices areused. For this reason, it may be strategic for the largecommunities working on many new diagnostic devices andbiosensors to consider adding security features to their nextgeneration of analytical devices.

■ ASSOCIATED CONTENT*S Supporting InformationThe Supporting Information is available free of charge on theACS Publications website at DOI: 10.1021/acs.anal-chem.8b00826.

Characterization of the printing resolution of the inkjetspotter with the amaranth ink on the nitrocellulosemembrane (PDF)

Video of an 8-color dynamic code on the nitrocellulosemembrane that is partially deleted in a single step (AVI)Video of a liquid filling the 3D architecture and deletingthe code in 4 steps (AVI)

■ AUTHOR INFORMATIONCorresponding Authors*E-mail: ogo@zurich.ibm.com. Tel: +41-44-724 8935.*E-mail: emd@zurich.ibm.com. Tel: +41-44-724 8283.ORCIDEmmanuel Delamarche: 0000-0002-8753-8895NotesThe authors declare no competing financial interest.

■ ACKNOWLEDGMENTSWe thank Yuksel Temiz and Ute Drechsler for their help inchip fabrication, Diana Davila Pineda for the SEM images,Andreas Kind for providing insight on counterfeiting andsecurity solutions, and Walter Riess for his continuous support.We acknowledge partial funding from the IBM ResearchFrontiers Institute.

■ REFERENCES(1) Naghavi, M.; et al. Lancet 2017, 390, 1151−1210.(2) Dye, C. Philos. Trans. R. Soc., B 2014, 369, 20130426.(3) Drain, P. K.; Hyle, E. P.; Noubary, F.; Freedberg, K. A.; Wilson,D.; Bishai, W. R.; Rodriguez, W.; Bassett, I. V. Lancet Infect. Dis. 2014,14, 239−249.(4) Girosi, F.; Olmsted, S. S.; Keeler, E.; Hay Burgess, D. C.; Lim, Y.-W.; Aledort, J. E.; Rafael, M. E.; Ricci, K. A.; Boer, R.; Hilborne, L.;Derose, K. P.; Shea, M. V.; Beighley, C. M.; Dahl, C. A.; Wasserman, J.Nature 2006, 444 (Suppl 1), 3−8.(5) Yager, P.; Domingo, G. J.; Gerdes, J. Annu. Rev. Biomed. Eng.2008, 10, 107−144.(6) Mabey, D.; Peeling, R. W.; Ustianowski, A.; Perkins, M. D. Nat.Rev. Microbiol. 2004, 2, 231−240.(7) Sousa-Figueiredo, J. C.; Stanton, M. C.; Katokele, S.; Arinaitwe,M.; Adriko, M.; Balfour, L.; Reiff, M.; Lancaster, W.; Noden, B. H.;Bock, R.; Stothard, J. R. PLoS Neglected Trop. Dis. 2015, 9,No. e0003831.(8) Jorgensen, P.; Nambanya, S.; Gopinath, D.; Hongvanthong, B.;Luangphengsouk, K.; Bell, D.; Phompida, S.; Phetsouvanh, R. Malar. J.2010, 9, 59.(9) Thiam, S.; Thior, M.; Faye, B.; Ndiop, M.; Diouf, M. L.; Diouf,M. B.; Diallo, I.; Fall, F. B.; Ndiaye, J. L.; Albertini, A.; Lee, E.;Jorgensen, P.; Gaye, O.; Bell, D. PLoS One 2011, 6, No. e18419.(10) Hopkins, H.; Bebell, L.; Kambale, W.; Dokomajilar, C.;Rosenthal, P. J.; Dorsey, G. J. Infect. Dis. 2008, 197, 510−518.(11) Pruett, C. R.; Vermeulen, M.; Zacharias, P.; Ingram, C.; TayouTagny, C.; Bloch, E. M. Transfus Med. Rev. 2015, 29, 35−44.(12) Owusu-Ofori, S.; Temple, J.; Sarkodie, F.; Anokwa, M.;Candotti, D.; Allain, J.-P. Transfusion 2005, 45, 133−140.(13) Allain, J.-P.; Lee, H. Expert Rev. Mol. Diagn. 2005, 5, 31−41.(14) Zhao, J.; Lama, M.; Korenromp, E.; Aylward, P.; Shargie, E.;Filler, S.; Komatsu, R.; Atun, R. PLoS One 2012, 7, No. e43549.(15) World Health Organization; UNICEF/UNDP/World Bank/WHO Special Programme for Research and Training in TropicalDiseases. In The use of visceral leishmaniasis rapid diagnostic tests; 2008;pp 3.(16) Glass, B. Res. Rep. Trop Med. 2014, 5, 11−22.(17) Clark, F. Lancet 2015, 386, 1327−1328.(18) Lavorgna, A. Eur. J. Criminol 2015, 12, 226−241.(19) Two accused over “fake” HIV tests. BBC News, Oct. 30, 2006.http://news.bbc.co.uk/2/hi/south_asia/6099064.stm (accessed Feb.05, 2018).

Figure 6. continued

same amount of code elements. (e) The factor of lost chip area (Abit =200 × 200 μm2, Anode = 400 × 500 μm2) and (f) the gain in codewriting speed (tspot = 1 s) when using multistep codes compared tostatic codes containing the same amount of information. n is thenumber of code elements, and b = n shows the upper boundary whereeach code element can be deleted individually.

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.8b00826Anal. Chem. XXXX, XXX, XXX−XXX

G

(20) Mori, M.; Ravinetto, R.; Jacobs, J. Trop. Med. Int. Health 2011,16, 1439−1449.(21) Akoni, O. Ebola: Lagos alerts on sale of fake EVD cassettes, testkits. Vanguard, 2018. https://www.vanguardngr.com/2014/09/ebola-lagos-alerts-sale-fake-evd-cassettes-test-kits/ (accessed February 5,2018).(22) Rasti, R.; Nanjebe, D.; Karlstrom, J.; Muchunguzi, C.; Mwanga-Amumpaire, J.; Gantelius, J.; Mårtensson, A.; Rivas, L.; Galban, F.;Reutersward, P.; Andersson Svahn, H.; Alvesson, H. M.; Boum, Y.;Alfven, T. PLoS One 2017, 12, No. e0182005.(23) Boadu, N. Y.; Amuasi, J.; Ansong, D.; Einsiedel, E.; Menon, D.;Yanow, S. K. Malar. J. 2016, 15, 126.(24) Bansal, D.; Malla, S.; Gudala, K.; Tiwari, P. Sci. Pharm. 2013, 81,1−13.(25) Mackey, T. K.; Nayyar, G. Expert Opin. Drug Saf. 2017, 16,587−602.(26) Newton, P. N.; Fernandez, F. M.; Plancon, A.; Mildenhall, D.C.; Green, M. D.; Ziyong, L.; Christophel, E. M.; Phanouvong, S.;Howells, S.; McIntosh, E.; Laurin, P.; Blum, N.; Hampton, C. Y.;Faure, K.; Nyadong, L.; Soong, C. W. R.; Santoso, B.; Zhiguang, W.;Newton, J.; Palmer, K. PLoS Med. 2008, 5, No. e32.(27) Liu, Y.; Lee, Y. H.; Lee, M. R.; Yang, Y.; Ling, X. Y. ACSPhotonics 2017, 4, 2529−2536.(28) Yetisen, A. K.; Butt, H.; Mikulchyk, T.; Ahmed, R.; Montelongo,Y.; Humar, M.; Jiang, N.; Martin, S.; Naydenova, I.; Yun, S. H. Adv.Opt. Mater. 2016, 4, 1589−1600.(29) Commission delegated regulation (EU). Official Journal of theEuropean Union 2016, L32/1−27.(30) U.S. Food & Drug Administration. Global Unique DeviceIdentification Database (GUDID). https://www.fda.gov/M e d i c a l D e v i c e s / D e v i c e R e g u l a t i o n a n d G u i d a n c e /UniqueDeviceIdentification/GlobalUDIDatabaseGUDID/default.htm(accessed Feb. 05, 2018).(31) Davison, M. In Pharmaceutical Anti-Counterfeiting: Combatingthe Real Danger from Fake Drugs; Wiley, 2011; pp 40−41.(32) Scherr, T. F.; Gupta, S.; Wright, D. W.; Haselton, F. R. Lab Chip2017, 17, 1314−1322.(33) Park, D.-H.; Han, C. J.; Shul, Y.-G.; Choy, J.-H. Sci. Rep. 2015, 4,4879.(34) Han, S.; Bae, H. J.; Kim, J.; Shin, S.; Choi, S.-E.; Lee, S. H.;Kwon, S.; Park, W. Adv. Mater. 2012, 24, 5924−5929.(35) Ciftlik, A. T.; Dupouy, D. G.; Gijs, M. A. M. Lab Chip 2013, 13,1482−1488.(36) Schindelin, J.; Arganda-Carreras, I.; Frise, E.; Kaynig, V.;Longair, M.; Pietzsch, T.; Preibisch, S.; Rueden, C.; Saalfeld, S.;Schmid, B.; Tinevez, J.-Y.; White, D. J.; Hartenstein, V.; Eliceiri, K.;Tomancak, P.; Cardona, A. Nat. Methods 2012, 9, 676−682.(37) Kato, H.; Tan, K. T. IEEE Pervasive Comput 2007, 6, 76−85.(38) Ru, C.; Luo, J.; Xie, S.; Sun, Y. J. Micromech. Microeng. 2014, 24,053001.(39) Seemann, R.; Brinkmann, M.; Kramer, E. J.; Lange, F. F.;Lipowsky, R. Proc. Natl. Acad. Sci. U. S. A. 2005, 102, 1848−1852.(40) Oliver, J. F.; Huh, C.; Mason, S. G. J. Colloid Interface Sci. 1977,59, 568−581.(41) Deegan, R. D.; Bakajin, O.; Dupont, T. F.; Huber, G.; Nagel, S.R.; Witten, T. A. Nature 1997, 389, 827−829.(42) Seo, C.; Jang, D.; Chae, J.; Shin, S. Sci. Rep. 2017, 7, 500.(43) Reed, I. S.; Solomon, G. J. Soc. Ind. Appl. Math. 1960, 8, 300−304.(44) You, M.; Zhong, J.; Hong, Y.; Duan, Z.; Lin, M.; Xu, F.Nanoscale 2015, 7, 4423−4431.(45) Da Luz, L. L.; Milani, R.; Felix, J. F.; Ribeiro, I. R. B.; Talhavini,M.; Neto, B. A. D.; Chojnacki, J.; Rodrigues, M. O.; Junior, S. A. ACSAppl. Mater. Interfaces 2015, 7, 27115−27123.(46) Duong, B.; Liu, H.; Li, C.; Deng, W.; Ma, L.; Su, M. ACS Appl.Mater. Interfaces 2014, 6, 8909−8912.(47) Hou, X.; Ke, C.; Bruns, C. J.; McGonigal, P. R.; Pettman, R. B.;Stoddart, J. F. Nat. Commun. 2015, 6, 6884.(48) Moszczynski, A. J. Med. Ethics 2009, 35, 87−90.

(49) Popp, W.; Rasslan, O.; Unahalekhaka, A.; Brenner, P.;Fischnaller, E.; Fathy, M.; Goldman, C.; Gillespie, E. Int. J. Hyg.Environ. Health 2010, 213, 302−307.(50) Zhang, Y.; Lyu, F.; Ge, J.; Liu, Z. Chem. Commun. 2014, 50,12919−12922.(51) Yang, M.; Zhang, W.; Zheng, W.; Cao, F.; Jiang, X. Lab Chip2017, 17, 3874−3882.(52) Chin, C. D.; Linder, V.; Sia, S. K. Lab Chip 2012, 12, 2118−2134.(53) Xu, X.; Akay, A.; Wei, H.; Wang, S.; Pingguan-Murphy, B.;Erlandsson, B.-E.; Li, X.; Lee, W.; Hu, J.; Wang, L.; Xu, F. Proc. IEEE2015, 103, 236−247.(54) Incardona, S.; Serra-Casas, E.; Champouillon, N.; Nsanzabana,C.; Cunningham, J.; Gonzalez, I. J. Malar. J. 2017, 16, 196.(55) Zimmermann, M.; Hunziker, P.; Delamarche, E. Microfluid.Nanofluid. 2008, 5, 395−402.(56) Man, P. F.; Mastrangelo, C. H.; Burns, M. A.; Burke, D. T. InProceedings MEMS 98. Eleventh Annual International Workshop onMicro Electro Mechanical Systems. An Investigation of Micro Structures,Sensors, Actuators, Machines and Systems; IEEE, 1998; pp 45−50.(57) Chaudhry, P.; Zimmerman, A. In Protecting Your IntellectualProperty Rights; Springer, 2013; pp 7−31.

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.8b00826Anal. Chem. XXXX, XXX, XXX−XXX

H