Fb

Ja

b

c

d

a

ARRAA

KSSVMPM

1

nscilaiiTmcrocssta

0d

Biosensors and Bioelectronics 24 (2009) 3663–3670

Contents lists available at ScienceDirect

Biosensors and Bioelectronics

journa l homepage: www.e lsev ier .com/ locate /b ios

abrication and characterization of a multiwell array SERS chip withiological applications

.L. Abell a,∗, J.D. Driskell b, R.A. Dluhy c, R.A. Tripp b, Y.-P. Zhao d

Nanoscale Science and Engineering Center, Department of Biological and Agricultural Engineering, The University of Georgia, Athens, GA 30602, United StatesNanoscale Science and Engineering Center, Department of Infectious Diseases, The University of Georgia, Athens, GA 30602, United StatesNanoscale Science and Engineering Center, Department of Chemistry, The University of Georgia, Athens, GA 30602, United StatesNanoscale Science and Engineering Center, Department of Physics and Astronomy, The University of Georgia, Athens, GA 30602, United States

r t i c l e i n f o

rticle history:eceived 27 January 2009eceived in revised form 26 May 2009ccepted 28 May 2009vailable online 7 June 2009

a b s t r a c t

Uniform, large surface area substrates for surface-enhanced Raman spectroscopy (SERS) are fabricatedby oblique angle deposition. The SERS-active substrates are patterned by a polymer-molding techniqueto provide a uniform array for high throughput biosensing and multiplexing. Using a conventional SERS-active molecule, 1,2-di(4-pyridyl)ethylene (BPE) ≥98%, we show that this device provides a uniformRaman signal enhancement from well to well with a detection limit of at least 10−8 M of the BPE solution

eywords:ERSilver nanorod arrayiral detectionultiwell array

atterned substrate

or 10−18 mol of BPE. The SERS intensity is also demonstrated to vary logarithmically with the log of BPEconcentration and the apparent sensitivity of the patterned substrate is compared to previous reportsfrom our group on non-patterned substrates. Avian influenza is analyzed to demonstrate the utility ofSERS multiwell patterned substrates for biosensing. The spectra acquired from patterned substrates showbetter reproducibility and less variation compared to the unpatterned substrates according to multivariateanalysis. Our results highlight potential advantages of the patterned substrate.

icrotiter

. Introduction

Biochemical and genetic diagnostic tests often require largeumbers of samples either to test a large number of variables or tocreen large sample populations. The reagents used in these assaysan be very expensive and the sample itself may not be availablen large quantities; therefore, smaller reaction volumes and paral-el testing are essential. Array techniques such as immunoassaysnd DNA microarray chips have proven to be extremely successfuln biological and genetic assays and are thoroughly documentedn the literature and text books (Ekins, 1989; Schena et al., 1996).hese methods although well established do suffer from a fewajor setbacks. Both immunoassays and DNA microarrays require

areful sample handling and substrate preparation to acquire usefulesults. Due to such processes they are relatively time-consumingften requiring at least several hours to incubate samples with theapture molecules (i.e. antibodies or single stranded DNA) so that a

ufficient number can bind for detection. Several substrate-washingteps are then needed to remove excess reagents and to minimizehe background signal. In addition, as with most molecular-labelingssays, these methods require extrinsic reporter molecules that

∗ Corresponding author. Tel.: +1 706 542 6230.E-mail address: jabell@uga.edu (J.L. Abell).

956-5663/$ – see front matter © 2009 Elsevier B.V. All rights reserved.oi:10.1016/j.bios.2009.05.039

© 2009 Elsevier B.V. All rights reserved.

allow only for indirect detection of the molecule of interest. There-fore, the investigators must know before hand what they are lookingfor, at least approximately, in order to design the probes. Thus, aminimal step, time-efficient, and label-free technique with highspecificity and sensitivity would be preferred for array studies.

Surface-enhanced Raman scattering (SERS) is an analyticalmethod based on vibrational spectroscopy, which can give a char-acteristic spectrum of a specific molecule (Fleishmann et al., 1974;Tsang et al., 1979; Dornhaus et al., 1980; Sandra et al., 1980; Leeand Meisel, 1982; Yang et al., 1996). The method uses nanoscopi-cally rough metallic surfaces to greatly enhance Raman scatteringcross-section by molecules adsorbed onto the metal surface. Thecharacteristic Raman spectrum can act as a fingerprint to directlyidentify molecules or bioagents in proximity to the substrate. Thisis a distinct advantage to the limited ability of the aforementionedmethods. Without the need of incubating, a sample can be analyzedmuch more quickly by simply placing on the substrate and scanning.Beyond its experimental ease, SERS also offers a high degree of sen-sitivity and specificity. Detection of single molecules has previouslybeen reported (Kneipp et al., 1997; Nie and Emory, 1997) and the

specificity of the technique with biological samples has allowednot only for species level discrimination of bacteria and virus, butalso strain discrimination within species (Premasiri et al., 2005;Jarvis and Goodacre, 2004; Shanmukh et al., 2006; Shanmukh etal., 2008).

3 Bioele

soaeddaactwtwmv4asttbfttsa

Sssw5laudtesecswmtifdwidRtivoda2

fsret

so that the surface normal was offset by 86◦ relative to the inci-

664 J.L. Abell et al. / Biosensors and

Currently, there are two different SERS array techniques: extrin-ic and intrinsic (Tripp et al., 2008). Similar to the labeling methodf immunoassays and DNA chips, extrinsic detection generally usesntibodies or single stranded DNA to tether molecules of inter-st onto a SERS-active surface with a SERS reporter molecule foretection. For example, Rohr et al. showed that the Raman dye p-imethylamionasobenzene (DAB) could be covalently attached ton antibody, and could then be detected on silver electrodes viasandwich-configured immunoassay (Rohr et al., 1989). Ni et al.

ovalently coupled antibodies specific for various IgG moleculeso SERS-active gold nanoparticle colloids on which Raman labelsere adsorbed. This allowed for Raman detection of the labels

hat were indirectly immobilized onto a substrate surface treatedith IgG (Ni et al., 1999). Jun et al. designed a SERS reportericro-complex using polymer microspheres embedded with sil-

er nanoparticles on which SERS reporter molecules such as-methylbenzenethiol, 2-naphthalenethiol, and benzenethiol weredsorbed. The microbead/nanoparticle/reporter complex was thentabilized within a silica shell followed by coating with a bioac-ive molecule, such as biotin or antibody, that was shown to bindhe respective bioantigens to the surface of the silica coated SERSead (Jun et al., 2007). Although proven to be relatively success-

ul, this extrinsic method relies on indirect detection, similar tohe immunoassay and DNA microarrays discussed above; besideshe advantages of a narrow spectral bandwidth and possible highensitivity, the extrinsic method suffers from the same general dis-dvantages of the current array techniques.

There have been few reports to elicit methods for array-basedERS detection through intrinsic SERS substrates. Shin et al. demon-trated that colloidal silver nanoparticles could be inked onto ailicon polymer stamp and printed onto a gold substrate coatedith an organic self-assembled monolayer to yield an array of

0 �m × 50 �m square-shaped patterns. They then used the Ramanabel rhodamine G6 to demonstrate the SERS-enhancement gener-ted in the micron patterned squares (Shin et al., 2002). However,sing nanoparticles on solid substrates, the SERS enhancement willepend on the monodispersion and aggregation of the nanopar-icles. Furthermore, the patterning method does not provide annclosure to contain a liquid sample from spreading across the sub-trate, leading to potential cross-contamination of samples. Bellt al. developed a method of mixing silver colloidal nanoparti-les with a liquid polymer, preferentially a polymer that has amall Raman cross-section, which was then deposited into theells of a standard 96-well microtiter plate where the poly-er/nanoparticle mixture was allowed to dry. Upon rehydration of

he polymer with an aqueous sample, the polymer swelled provid-ng the analyte molecules access to the embedded nanoparticlesor enhancement (Bell and Spence, 2001). This study admittedlyid have some limitations since only Raman-resonant moleculesere analyzed, which suffered from poor signal reproducibil-

ty. Furthermore, this method may not be suitable for low-leveletection since the polymer, although having a relatively smallaman cross-section, may still interfere with low signal detec-

ion of an analyte. Strop et al. drop coated polymer nanobeadsnto the wells of a 96-well microtiter plate, followed by silverapor deposition onto into the wells to form silver thin filmsver nanospheres (AgFON). The study shows that the substratesemonstrate good reproducibility from well to well, but only rel-tively high concentrations of analyte were used (Stropp et al.,003).

The key to make SERS as a practical array technique is the

abrication of uniform, higher enhancement and larger area SERSubstrates. Throughout the last thirty years, many different fab-ication techniques have been explored in order to make highnhancement SERS substrates, including electrochemical oxida-ion reduction cycles (ORC) (Pemberton, 1991), chemical etching

ctronics 24 (2009) 3663–3670

(Carron et al., 1991), metal island films (Schlegel and Cotton, 1991),electron beam lithography (Kahl et al., 1998), and nanospherelithography (Jensen et al., 1999) to name a few. However, none ofthese techniques have been shown to produce large enough area,with high signal enhancement and uniformity that SERS substrateswould be required for array techniques. They produce either rela-tively low enhancement and poor reproducibility but large surfacearea SERS substrates, such as ORC (Pemberton, 1991), chemicaletching (Carron et al., 1991), metal island films (Schlegel and Cotton,1991); or large enhancement but small area sample, such as electronbeam lithography (Kahl et al., 1998) and nanosphere lithography(Jensen et al., 1999). Recently we have found that SERS substratesfabricated by oblique angle deposition (OAD) can provide extremelyhigh SERS enhancement (Driskell et al., 2008; Chaney et al., 2005).Since OAD is based on a physical vapor deposition technique, it canalso be used to fabricate large area substrates. The OAD fabricatedSERS substrate discussed in the literature consists of a tilted silvernanorod array with nanorod diameters ∼100 nm, a tilt angle of 71◦

with respect to the surface normal, and a rod density of 15–25 per�m2, yielding an enhancement factor of over 108 (Driskell et al.,2008; Chaney et al., 2005). The low temperature nature of OAD alsomakes it easy to integrate with other microfabrication processes.This study proposes a method of fabricating a microwell-arrayedSERS chip on 2.54 cm × 7.62 cm standard glass microscope slideusing OAD to produce a SERS-active surface followed by use ofa well-array patterning mold in which liquid polydimethylsilox-ane (PDMS) is added and cured by low temperature heating. Thus,providing a time efficient and cost effective utilization of SERS sub-strates for multiple, isolated sample analysis.

2. Experimental

2.1. Materials

Standard 2.54 cm × 7.62 cm glass microscope Gold Seal® slides(Becton and Dickinson Company, catalog # 3010) were used as thestarting substrate for the SERS chip. Titanium (99.995% pure) andsilver (≥99.99% pure) were purchased from Kurt J. Lesker. PDMS(Sylgard® 184 Silicone Elastomer Kit) was obtained from Dow Corn-ing. BPE was supplied by Fluka. Methanol ≥99.8% was obtained fromSigma–Aldrich.

2.2. SERS-active substrate fabrication

Glass slides were cleaned in heated piranha solution (4:1, v/vH2SO4:H2O2) for at least 10 min, and then thoroughly rinsed withDI water and blow dried with N2. A custom-built electron beamevaporation chamber, used only in conjunction with titanium andsilver, was employed to deposit titanium and silver onto the glasssubstrates similar to the method described previously (Chaney etal., 2005; Driskell et al., 2008). All deposition rate and thicknessmeasurements were determined by a quartz crystal microbalance(QCM) placed inside the chamber and facing the source material.The depositions were performed when the chamber reaches a basepressure of 10−6 Torr. The surface of the substrate was positionednormal to the incident vapor. A titanium adhesion layer was firstdeposited at a rate of 0.2 nm/s to a total QCM thickness reading of20 nm. A silver thin film was deposited at a rate of 0.3 nm/s to a totalQCM thickness reading of 500 nm. The substrate was then rotated

dent vapor. Silver was then deposited at a constant rate of 0.3 nm/sto a final QCM thickness reading of 2000 nm to form a tilted Agnanorod array. The substrates in the chamber were allowed to coolto room temperature in vacuum before they were removed fromthe chamber.

Bioele

2

aiachhicPwawtra

stfalb35aAp4s

2

ts

2

ewcm

Fip

J.L. Abell et al. / Biosensors and

.3. SERS array fabrication

The SERS array was fabricated using a molding process. The moldssembly was composed of a base plate that holds the substraten place. The well-patterning plate (Fig. 1A) was machined from

single piece of aluminum with forty regularly spaced hollowedylindrical protrusions arrayed in a 4 × 10 manner. Each protrusionas an outer diameter of 4 mm, a 0.5 mm thick radial wall, and aollowed inner diameter of 3 mm and a height of 1 mm. Therefore,

t is expected that each well formed with this mold will be able toontain a 12.6 �l volume of sample. The protrusions prevent theDMS from coating the SERS-active nanorods at the bottom of theell, which could interfere with the SERS response of the nanorod

rray. If the protrusion did not have a recessed central region, thisould cause physical damage to the delicate nanorod array along

he bottom of the well, which would also change the SERS signalesponse. Thus, there is a 0.5 mm thick ring of inactivated substratelong the edge of each patterned well.

The portion of the base plate to be in contact with theubstrate/well-pattering plate was lined with a piece of wax papero help the release of the substrate/PDMS/well-patterning platerom the bottom plate after curing. The PDMS base and the curinggent (10:1, w/w) were mixed together using a glass rod. Before theiquid PDMS was added to the mold/substrate assembly, the assem-ly was first pre-heated in a benchtop oven (Barnstead Int. LabLine510 L-C) at 100 ◦C for 20 min. The PDMS was also preheated formin and then poured into the opening of the substrate/moldssembly and allowed to cure at 100 ◦C for approximately 20 min.fter the substrate had cooled to room temperature, the wellatterning plate was pulled off of the PDMS, leaving a uniform× 10 well PDMS-patterned SERS-active microwell arrayed sub-

trate as shown in Fig. 1B.

.4. SERS chip characterization

The quality of the SERS array has been demonstrated throughhe following methods: (1) leakage test; (2) uniformity test; (3)ensitivity test; and (4) bioagent detection test.

.4.1. Leakage test

One of the main goals of the array SERS chip is to design microw-

lls that can be isolated from each other so that the samplesill be completely contained in a specific well without cross-

ontaminating the adjacent wells. In order to determine whetherinute amounts of analyte were able to leak out of the well under-

ig. 1. (A) Assembly of the well-patterning plate (1) mold with the SERS substrate (2) withs poured into the opening (3) between the substrate and well-patterning plate. The baseatterned substrate after removing from mold assembly with well labeling scheme. The w

ctronics 24 (2009) 3663–3670 3665

neath the PDMS layer and through the silver nanorod layer, 5 �l ofa relatively concentrated BPE in methanol solution (10−4 M) wasadded to well B8 of a patterned SERS substrate (Fig. 1B). BPE wasused because it has a large Raman cross-ection with its majorspectral Raman peaks at 1008, 1200, 1338, 1604 and 1640 cm−1,respectively. BPE also adsorbs irreversibly onto Ag surfaces, and hasbeen previously studied extensively for Ag nanorod SERS substrates(Yang et al., 1996; Chaney et al., 2005; Chu et al., 2007; Driskell etal., 2008). Methanol is used as a solvent for BPE because it evap-orates quickly and has a very good ability to spread on the silvernanorod array surface. Well B8 with BPE solution was then coveredwith a small glass cover slip just large enough to fit over the wellto inhibit evaporation of the methanol solution. Keeping the wellfilled with analyte solution for an extended time allows the sealformed between the PDMS and the silver nanorod film to be morerigorously evaluated. After 1.5 hours the methanol appeared to haveevaporated and the glass coverslip was removed and any remainingmethanol was allowed to dry. The SERS spectrum of well B8 alongwith the spectrum of the adjacent wells (Fig. 1B) was measuredwith an Enwave Fiber Raman probe system with an incident wave-length of 785 nm and an input power of 35 ± 5 mW. Each scan hadan integration time of 10 s and each output spectrum was set to bean average of two collected spectrums.

2.4.2. Uniformity and sensitivity testAll the SERS mapping measurements were carried out on a

Renishaw inVia Raman microscope system equipped with an auto-mated high precision actuation stage. For the SERS patternedsubstrate mapping, each well was measured in three separate spots.The second and third spots were +0.5/+0.5 mm (forward/up) and−0.5/−0.5 mm (backwards/down) relative to center (spot 1). Eachspot was scanned with a 5× objective, 10% power, 785 nm excita-tion wavelength and 5 s acquisition time unless otherwise stated. Aspectral range from 1130 to 1650 cm−1 was collected at each spot.We used the 1200 cm−1 peak intensity, I1200, to quantify the SERSresponse. The Renishaw’s software was programmed to analyze thepeaks using a mix of Gaussian and Lorenzian curves. The substrateanalyzed by the Renishaw system was mapped within 72 h of fabri-cation. For the well-to-well mapping, any spectral acquisitions thatvisually appeared to have a cosmic radiation (instrument noise)

spike at 1200 cm−1 were removed from the analysis and only twopoints for that well were used for a single mapping, but only onespike at 1200 cm−1 was observed during the entire experiment.

The uniformity and sensitivity of the SERS array were char-acterized by measuring the SERS spectra while consecutively

the SERS active surface of the substrate facing the well-patterning plate. The PDMSand top plates of the mold assembly have been left out for clarity. (B) A photo of aells within the dashed box are those analyzed for the leakage test.

3 Bioelectronics 24 (2009) 3663–3670

itmorobieta1

fpwa

2

tstmcs

dtanr1aoss

cudtr(

3

3

cFaiBtila

3

m

666 J.L. Abell et al. / Biosensors and

ncreasing the BPE concentrations. The first mapping demonstrateshe response of the patterned substrate before the addition of any

ethanol or BPE. Then 5 �l of methanol was added to each wellf this same substrate and allowed to dry and again the SERSesponse of the substrate was mapped. This was repeated using 5 �lf 10−8 M BPE added to each well of the same substrate, followedy the addition of 10−7, 10−6, and 10−5 M BPE. Sequentially increas-

ng the concentration of the BPE solutions allowed the wells to bevaluated at multiple concentrations, but due to accumulation ofhe analyte it should be noted that after adding 1 × 10−7, 1 × 10−6,nd 1 × 10−5 M BPE the effective concentration in each well was.1 × 10−7, 1.11 × 10−6, and 1.111 × 10−5 M BPE, respectively.

In order to determine the uniformity across the SERS-active sur-ace of the bottom of a well, a high resolution SERS mapping waserformed on well B5. The well was scanned at over 6000 spotsith a step size of 50 �m. The acquisition time was 2.5 s per spectral

cquisition.

.5. Virus detection

To test the virus detection capability of the SERS multiwell pat-erned substrate, we used avian influenza virus (AIV). An AIV (H5N2train) isolate was propagated in embryonated chicken eggs. Allan-oic fluid was harvested 48 h post infection and the virus titer was

easured to be 1.8 × 106 pfu/mL by plaque assay. Allantoic fluid (AF)ollected from mock-infected (i.e., PBS) embryonated chicken eggserved as a negative control sample.

Influenza-positive and negative allantoic fluid samples wereiluted 100-fold prior to application to the SERS substrate. Allan-oic fluid is a rather complex and concentrated background matrix,nd when applied without dilution, multilayers formed on theanorod substrate, quenching the SERS signal. To compare theesults obtained from patterned SERS substrates to individualcm × 1 cm SERS substrates, 2.0 �L of each of the samples werepplied to two different 1 cm × 1 cm SERS substrates and three wellsn a patterned substrate. The samples were allowed to dry and fivepectra were recorded for each sample spotted on a substrate orpotted in a well.

SERS spectra were acquired using a Bruker Senterra Ramanonfocal microscope system. A 785 nm near-IR diode laser wassed as the excitation source. The laser was focused into 5 �miameter spot using a 20× objective, and the laser power athe sample surface was set at 24 mW. SERS spectra over theange of 430–1830 cm−1 were collected using the high resolution3–5 cm−1) grating (1200 lines/mm) and 10 s exposure.

. Results and discussion

.1. Leakage test

A multiwell assay plate must be able to confine and isolate theontents of each well to prevent cross contamination of samples.or the leakage test, the SERS spectra of well B8 (soaked with BPE)nd the eight adjacent wells (without BPE) are collected as shownn Fig. 2. We can only observe BPE signatures in the well B8 wherePE was directly added and not in any of the eight adjacent wells;herefore the wells are concluded to sufficiently confine the samplen the lateral direction on the SERS-active substrate surface. Theeakage test has also been performed on well C3 and similar resultsre obtained.

.2. Uniformity and sensitivity tests

The well-to-well uniformity and sensitivity are characterized byeasuring the three-point mean SERS I1200 in each well. Fig. 3A–F

Fig. 2. Raman spectra from nine wells centered around well B8. The dashed spec-trum corresponds to well B8 where BPE was added. The other eight spectra weretaken from the other wells directly neighboring B8. The black arrow is pointing atthe 1200 cm−1 BPE peak from well B8.

shows the average I1200 map for 4 × 10 well array for different con-centrations of BPE along with the background signal. Well D10shows a relatively low I1200 at all BPE concentrations compared tothe other wells, and we conclude that the enhancement in that wellhas been inhibited, and therefore this well is considered an outlier.This does not appear to be an anomaly since many of the SERS chipsfabricated in our lab with this method appear to have relatively lowenhancement in one or more of the corner wells. This may be pre-scribed as a fabrication optimization issue and is probably a resultof the corner wells having more thermal contact with the heatedmold during the curing process. The possible solution to eliminatethe corner well problem is to improve the uniformity of the heatingand patterning environment. Further work is currently underwayto address these issues.

For the patterned substrate, before the addition of methanol orBPE, the mean background I1200 for the patterned substrate is 32counts with a standard deviation of ±14 counts or 44%. With theaddition of methanol the mean background peak intensity increasesto 56 counts with a standard deviation of ± 42 counts or 86% withonly one well reporting an I1200 signal over three standard devia-tions above the mean. The I1200 before the addition of BPE appearsto be due to noise from the system and small contaminant peaksnear 1200 cm−1. Clearly, the I1200 (baseline corrected) is increas-ing after the addition of methanol but the exact reason for thisis not completely understood. We believe this may be due to thedissolving of the ambient contamination into the methanol andupon evaporation the contamination is redistributed at the baseof the nanorods where Raman enhancement has been shown tobe greater (unpublished data). The background signal is thereforedefined to be the I1200 of the substrate after methanol but before BPEis added. After the addition of 10−8 M BPE the SERS spectra of BPEis visible in all thirty-nine wells and the mean intensity is found toincrease to 909 counts with a standard deviation of ± 412 counts or45%, with 100% of the wells reporting a signal above three standarddeviations of background excluding the outlier, well D10. The lowerlimit of detection for each well is therefore assumed to be at least10−8 M BPE. The mean I1200 and standard deviation for the remain-ing 10−7, 10−6, and 10−5 M BPE solutions are 5495 ± 2203 (40%),21149 ± 7653 (36%), and 49101 ± 11673 (24%), respectively (Fig. 3G).

We also observe that the SERS signal within a single well varies.There are several possible explanations for such a variation of SERSsignal across the bottom of the well, one possibility is that thenanorod array is not uniform across the entire substrate and the

J.L. Abell et al. / Biosensors and Bioelectronics 24 (2009) 3663–3670 3667

Fig. 3. (A–F) I1200 mapping of each well of the 4 × 10 patterned substrate for the bare substrate, methanol only, 10−8, 10−7, 10−6, and 10−5 M BPE, respectively. Well D10 iso e meag less ino eak int

otatI(hrdowsnwsniv

bservably less intense than the other wells after the addition of BPE. (G) A plot of threy scale corresponds to the average I1200 of each well with black corresponding tof well B5 after the final 10−5 M BPE solution was added. We can see that the I1200 phe grey scale for (H) is not logarithmic.

ther possibility is that the BPE is being unevenly deposited acrosshe nanorod surface at the bottom of the well during the spreadingnd drying process. To investigate this, we performed a high resolu-ion scan of one of the wells. Fig. 3H is a high (50 �m step) resolution1200 mapping of the well B5 after all BPE concentrations were addedi.e. 1.111 × 10−5 M). The area along the edge of the well appears toave a higher signal intensity. This area appears to coincide with theing of solvent that forms along the edges of the well as the wellries, which has a tendency to dry first at the center; therefore,ne may expect to find the BPE more concentrated along the edgehere the well dries last. Thus, we conclude that the non-uniformly

preading or drying of BPE on the surface could contribute sig-ificantly to the SERS signal variations and therefore the location

ithin a well is important factor affecting signal intensity. The SERS

ignal variations are due to four different sources, the instrumentoise, molecule statistics, substrate uniformity, and sample spread-

ng (see Supplementary materials I). The former two are intrinsicariation due to equipment and statistics, and the latter two can be

n I1200 of the entire patterned substrate at each BPE concentration. The logarithmictensity and white corresponding to higher intensity. (H) A high resolution mappingtensity is not completely uniform across the bottom surface of the well. Note that

improved by improving the substrate uniformity in fabrication pro-cess (see Supplementary materials II) and by redesigning the shapeof the well, or surface treatment of the Ag nanorods to improve thespreading.

The SERS intensity at the 1200 cm−1 band versus BPE concen-tration c for the 39 active wells (excluding well D10) are plottedin a log–log scale in Fig. 4A. There appears to be a linear relation-ship between log I1200 and log c, from which we define an apparentsensitivity ˛,

˛ = � log I1200

� log c.

The mean slope of the thirty-nine curves is 0.58 with a standard

deviation of ± 0.07 or 12.1%, which shows a uniform response forall wells. In order to allow for comparison of this apparent sensitiv-ity with other reports from our group, the number of moles of BPEwithin the laser spot area, which for the 5× objective of the Ren-ishaw system is 1265 �m2, was calculated assuming that the BPE

3668 J.L. Abell et al. / Biosensors and Bioele

Fig. 4. (A) log–log plot of 39 wells at each concentration of BPE with linear fits foreach well. The dark black line is a linear fit of the average peak intensities at eachconcentration with a slope of 0.57. (B) The apparent sensitivity of SERS enhancementrelative to mole number of BPE under laser spot. The patterned substrate describedin this report is compared to the apparent sensitivity of the unpatterned OAD silvernsde

sdtvdsilsTeiss

3

hftli

anorod substrates from references (Chu et al., 2007) and (Driskell et al., 2008) tohow that the apparent sensitivity is in the same general range. The dotted andashed line correspond to the linear fits of data from (Chu et al., 2007) and (Driskellt al., 2008), respectively.

olution was deposited uniformly across the bottom of the 4 mmiameter well (including the 0.5 mm SERS-inactive PDMS ring alonghe outer edge of the well). The average I1200 of thirty-nine wellsersus the mole number is plotted in Fig. 4B along with two sets ofata from our previous studies on a 1 cm × 1 cm unpatterned sub-trate (Chu et al., 2007; Driskell et al., 2008). All of the data shownn Fig. 4B demonstrate a linear relationship between log I1200 andog N, where N is the mole number of BPE, and the apparent sen-itivity ˛ for unpatterned substrates is 0.38 and 0.37, respectively.he use of different instruments and experimental parameters forach of these studies make a direct comparison difficult, however,t does appear that the apparent sensitivity ˛ for the patterned sub-trate falls into the same general range as that of the unpatternedubstrates.

.3. Viral SERS detection

The virus detection capability of the multiwell patterned chip

as been evaluated by a strain (H5N2) of AIV and compared to that

rom the flat 1 cm × 1 cm unpatterned substrate. It should be notedhat the samples spread on the 1 cm × 1 cm chips in an uncontrol-able and irreproducible fashion, unlike the patterned substratesn which the sample is confined to a well-defined area. Fig. 5A

ctronics 24 (2009) 3663–3670

and B shows the representative spectra of the AIV samples on bothpatterned and unpatterned substrates. Cursory examination of theSERS spectra reveals three important findings. First, AIV spectracollected on the patterned and unpatterned substrates are similar.Likewise, spectra of AF is also independent of the sensing substrate.Second, less spectral variation for the same sample is found for sam-ples deposited in the wells as compared to the samples deposited onthe 1 cm × 1 cm unpatterned substrates. Third, similar SERS spectrafor the AIV and AF samples are observed (data not shown). This isnot surprising since the AIV-positive sample largely consists of thesame components as AF. However, closer investigation reveals sev-eral significant spectral features attributed to AIV in the spectra ofthe AIV samples but not the AF samples.

Multivariate methods of data analysis aid in the interpretation ofthese spectra and confirmation of the observations alluded to abovedue to the complex nature of the SERS spectra and similarity ofthe spectra for positive and negative samples. Principal componentanalysis (PCA) has been applied to visualize the clustering of thedata according to class (e.g., AF-well, Flu-well, AF-unpatterned, Flu-unpatterned). A plot is generated to view the scores of PC1 versusPC2, as shown in Fig. 5C. Note that the spectra cluster according tothe classes defined above which confirms that SERS spectra allowfor identification of biological samples. For the purpose of this work,however, it is the spread of the data points within each class that isof importance. Note that the spread of scores along PC1 is greaterfor the spectra collected from the chips compared to the patternedwells. The same trend is observed for the scores on PC2.

PCA reduces the dimensionality of the data and while tightlyclustered data in a PC scores plot suggests minimal signal variation,it does not implicitly confirm minimal variation among samples.The information not captured by the PCs could potentially vary sig-nificantly and would not be observed in a PC scores plot. Therefore,the spectral information not included in the PCA model must beevaluated. Q residual quantifies the information not described bythe model. In other words, even if the data cluster tightly in the 2-DPC scores plots, large Q residual values suggest the data are not assimilar as perceived. The Q residual for the AIV and AF spectra isshown in Fig. 5D. This plot reveals much larger Q residual values forthe spectra collected from the unpatterned substrates comparedto the patterned wells. This can be interpreted that more of thespectral information for the spectra collected on the unpatternedsubstrates is not described by the model and greater variance in theSERS signal is present than is observed in the 2-D scores plots above.This analysis definitively confirms that greater spectral variation isobtained from the chips than the patterned wells.

Hierarchical cluster analysis (HCA) has been applied to quan-tify the similarity among the spectra. The results are displayed inthe dendogram (Fig. 6). This dendogram is generated using theEuclidean distance between furthest neighbors in real space, notPC space. Therefore, the total spectral variability is observed. Thedendogram shows that the distance from the furthest neighbor forthe spectra collected in the patterned wells is ∼0.28 and for thespectra collected on unpatterned substrates is ∼0.54–0.70, whichdemonstrates that the spectra from the multiwell substrate haveless variation.

Each of these methods of spectral analysis results in the sameconclusion; less spectral variation within a sample is obtained usinga patterned well. We attribute this phenomenon to sampling andnot differences in the nanorod structures. The multiwell patternedsubstrate facilitates reproducible spotting and drying of the sam-ple in a well-defined area on the substrate. Variations in sample

spreading and drying as is observed for the 1 cm × 1 cm unpatternedsubstrates can lead to slight variations in the SERS spectra. Admit-tedly, however, the slight increase in spectral heterogeneity on theunpatterned substrates may be insignificant for many studies asthese substrates have been successfully employed for the classifica-

J.L. Abell et al. / Biosensors and Bioelectronics 24 (2009) 3663–3670 3669

Fig. 5. Representative AIV Raman spectra from (A) the patterned substrate and (B) the unpatterned and unpatterned substrates for PCA analysis. (D) Q-residual plots for the same

Fp

te

4

mtr

ig. 6. Dendogram of the allantonic control (AF) and influenza (Flu) samples on bothatterned and unpatterned substrates.

ion of several viruses and differentiation of viral strains (Shanmukht al., 2006).

. Conclusion

Using OAD and low temperature polymer mold patterningethod, we have fabricated a robust, uniform, patterned SERS mul-

iwell chip and have evaluated its ability to handle analysis of theepresentative analyte BPE. Our results show that each active well

patterned 1 cm × 1 cm substrate. (C) PC1 vs PC2 scatter plot for AF and AIV treatedsamples from (C).

has very similar SERS behavior using a Raman probe molecule. Thevirus detection capability for the multiwell chip is better than theunpatterned SERS substrate in terms of reproducibility and vari-ations, which is of great advantage to develop the SERS chip as amultiplexing platform. All of these advantages can help the incor-poration of Raman spectroscopy as a mainstream, highly sensitiveanalytical tool allowing for the commercialization and widespreaduse of SERS substrates, which currently do not exist due to theintrinsic difficulties of processing such devices. We believe that thismultiwell SERS chip can decrease the amount of actual SERS sub-strate required for sample analysis thus decreasing cost, allow highsensitivity and specificity for chemical and biochemical detection,aid in establishing a more uniform substrate for better reproducibil-ity for SERS analysis, and allow for more automated and efficientsample analysis and data acquisition.

Acknowledgements

The authors would like to thank Jeonifer Garren for her helpfuldiscussions regarding statistical analysis. JLA and YPZ are supportedby the National Science Foundation under contract no. CMMI-0726770. This project is also partially supported by U.S. ArmyResearch Laboratory Grant No. W911NF-07-2-0065.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, inthe online version, at doi:10.1016/j.bios.2009.05.039.

3 Bioele

R

BCC

CD

DEFJJJ

KK

LN

670 J.L. Abell et al. / Biosensors and

eferences

ell, S.E.J., Spence, S.J., 2001. Analyst 126, 1–3.arron, K.T., Xue, G., Lewis, M.L., 1991. Langmuir 7, 2–4.haney, S.B., Shanmukh, S., Dluhy, R.A., Zhao, Y.-P., 2005. Appl. Phys. Lett. 87, 031908-

1-031908-3.hu, H., Liu, Y., Huang, Y., Zhao, Y.-P., 2007. Opt. Exp. 15, 12230–12239.riskell, J.D., Shanmukh, S., Liu, Y., Chaney, S.B., Tang, S.-J., Zhao, Y.-P., Dluhy, R.A.,

2008. J. Phys. Chem. C 112, 895–901.ornhaus, R., Long, M.B., Benner, R.E., Chang, R.K., 1980. Surf. Sci. 93, 240–262.kins, R.P., 1989. J. Pharm. Biomed. Anal. 7, 155–168.leishmann, M., Hendra, P.J., McQuillan, A.J., 1974. Chem. Phys. Lett. 26, 163–166.arvis, R.M., Goodacre, R., 2004. Anal. Chem. 76, 40–47.ensen, T.R., Schatz, G.C., Van Duyne, R.P., 1999. J. Phys. Chem. B 103, 2394–2401.un, B.-H., Kim, J.-H., Park, H., Kim, J.-S., Yu, K.-N., Lee, S.-M., Choi, H., Kwak, S.-Y., Kim,

Y.-K., Jeong, D.H., Cho, M.-Haing, Lee, Y.-S., 2007. J. Comb. Chem. 9, 237–244.ahl, M., Voges, E., Kostrewa, S., Viets, C., Hill, W., 1998. Sens. Actuators B 51, 285–291.neipp, K., Wang, Y., Kneipp, H., Perelman, L.T., Itzkan, I., Dasari, R.R., Feld, M.S., 1997.

Phys. Rev. Lett. 78, 1667–1670.ee, P.C., Meisel, D., 1982. J. Phys. Chem. 86, 3391–3395.i, J., Lipert, R.J., Dawson, B., Porter, M.D., 1999. Anal. Chem. 71, 4903–4908.

ctronics 24 (2009) 3663–3670

Nie, S., Emory, S.R., 1997. Science 275, 1102–1106.Pemberton, J.E., 1991. In: Abruna, H.D. (Ed.), Electrochemical Interfaces. Modern

Techniques for In-Situ Characterization. VCH Verlag Chemie, Berlin, p. 195.Premasiri, W.R., Moir, D.T., Klempner, M.S., Krieger, N., Jones II, G., Ziegler, L.D., 2005.

J. Phys. Chem. B 109, 312–320.Rohr, T.E., Cotton, T., Fan, N., Tarcha, P.J., 1989. Anal. Biochem. 182, 388–398.Sandra, P.N., Warlaumont, J.M., Demuth, J.E., Tsang, J.C., Christmann, K., Bradley, J.A.,

1980. Phys. Rev. Lett. 45, 1519–1523.Schena, M., Shalon, D., Heller, R., Chai, A., Brown, P.O., Davis, R.W., 1996. Proc. Nat.

Acad. Sci. 93, 10614–10619.Schlegel, V.L., Cotton, T.M., 1991. Anal. Chem. 63, 241–247.Shanmukh, S., Jones, L., Driskell, J.D., Zhao, Y.-P., Dluhy, R.A., Tripp, R.A., 2006. Nano

Lett. 6, 2630–2636.Shanmukh, S., Jones, L., Zhao, Y.-P., Driskell, J.D., Tripp, R.A., Dluhy, R.A., 2008. Anal.

Bioanal. Chem. 390, 1551–1555.

Shin, H.S., Yang, H.J., Jung, Y.M., Kim, S.B., 2002. Vib. Spectrosc. 29, 79–82.Stropp, J., Trachta, G., Brehm, G., Schneider, S., 2003. J. Raman Spectrosc. 34, 26–32.Tripp, R.A., Dluhy, R.A., Zhao, Y.-P., 2008. Nanotoday 3, 39–45.Tsang, J.C., Kirtley, J.R., Bradley, J.A., 1979. Phys. Rev. Lett. 43, 772.Yang, W., Hulteen, J., Schatz, G.C., Van Duyne, R.P., 1996. J. Chem. Phys. 104,

4313–4323.