Digital Protocol for Chemical Analysis at Ultralow Concentrations bySurface-Enhanced Raman ScatteringCarlos Diego L. de Albuquerque,†,‡ Regivaldo G. Sobral-Filho,† Ronei J. Poppi,‡

and Alexandre G. Brolo*,†,§

†Department of Chemistry, University of Victoria, Victoria, British Columbia V8P 5C2, Canada‡Institute of Chemistry, University of Campinas (Unicamp), CP 6154, 13084-971 Campinas, Sao Paulo, Brazil§Center for Advanced Materials and Related Technologies (CAMTEC), University of Victoria, Victoria, British Columbia V8W 2Y2,Canada

*S Supporting Information

ABSTRACT: Single molecule surface-enhanced Raman spectros-copy (SM-SERS) has the potential to revolutionize quantitativeanalysis at ultralow concentrations (less than 1 nM). However,there are no established protocols to generalize the application ofthis technique in analytical chemistry. Here, a protocol forquantification at ultralow concentrations using SM-SERS isproposed. The approach aims to take advantage of the stochasticnature of the single-molecule regime to achieved lower limits ofquantification (LOQ). Two emerging contaminants commonlyfound in aquatic environments, enrofloxacin (ENRO) andciprofloxacin (CIPRO), were chosen as nonresonant molecularprobes. The methodology involves a multivariate resolution curvefitting known as non-negative matrix factorization with alternatingleast-squares algorithm (NMF-ALS) to solve spectral overlaps. The key element of the quantification is to realize that, under SM-SERS conditions, the Raman intensity generated by a molecule adsorbed on a “hotspot” can be digitalized. Therefore, thenumber of SERS event counts (rather than SERS intensities) was shown to be proportional to the solution concentration. Thisallowed the determination of both ENRO and CIPRO with high accuracy and precision even at ultralow concentrations regime.The LOQ for both ENRO and CIPRO were achieved at 2.8 pM. The digital SERS protocol, suggested here, is a roadmap for theimplementation of SM-SERS as a routine tool for quantification at ultralow concentrations.

Single-molecule surface-enhanced Raman scattering (SM-SERS) was recognized 20 years ago1,2 as a potential tool to

revolutionize the field of analytical spectroscopy. However,although there are several examples of cleverly engineerednanostructures and approaches that allowed reliable SM-SERSmeasurements,3−11 there still are no reports of an analyticaldetermination that takes full advantage of the phenomenon.Typically, an analytical chemist seeks to obtain a linearcalibration curve that will correlate the measurements to theconcentrations of the analyte in solution. Unfortunately, thereare a few caveats that challenge the linearity of the calibrationcurve and the use of SM-SERS as direct probe in ultralowconcentrations.The SM-SERS phenomenon occurs for molecules adsorbed

in special areas in a nanostructured metal surface, called SERS“hotspots”.12 A SERS hotspot is a subwavelength region underthe area illuminated by the excitation laser that exhibit localenhanced field due to the excitation of surface plasmons.12 Theelectric field strength distribution around a hotspot is highlyinhomogeneous, and large enhancements are observed onlywithin a few nm away from the surface.2 Moreover, in a random

SERS substrate, such as a metal colloid deposited in glass, adistribution of hotspots with different efficiencies is expected.13

In the SM-SERS limit, only molecules adsorbed on highlyefficient hotspots contribute to the overall Raman signal.14,15

The bottom line is that highly efficient hotspots are rare andthe amount of adsorbed analytes is expected to also be small atultralow concentrations. Hence, SM-SERS experiments arecharacterized by fluctuations (either temporal or spatial) inRaman intensities,1,16 due to the low probability of a singlemolecule to find, by chance, an efficient SERS hotspot. Thethreshold for fluctuations in SERS intensities (SM-SERSregime) depend then on several factors, including the densityof hotspots, the distribution of hotspot strengths, the SERScross-section of the molecule, and the number of moleculesilluminated by the probing laser.17

In terms of quantification, the strong intensity fluctuationsobserved in the SM-SERS regime should affect the linearity of

Received: September 27, 2017Accepted: December 13, 2017Published: December 13, 2017

Article

pubs.acs.org/acCite This: Anal. Chem. 2018, 90, 1248−1254

© 2017 American Chemical Society 1248 DOI: 10.1021/acs.analchem.7b03968Anal. Chem. 2018, 90, 1248−1254

the calibration curve. Some groups have attacked this issue by,for instance, selectively direct the species of interest to thehotspot.18,19 Here we proposed a general method that actuallyembraces the stochastic characteristics of the SM-SERSintensity fluctuations. This quantification method can beapplied for any type of SERS-active analyte, even nonresonantmolecules, and for a variety of planar SERS substrates. Theprotocol is based on the concept of digital assays, a well-established approach in biomedical research.20 The analyticalmethod reported here should then provide an avenue foranalytical chemists to finally take advantage of the SM-SERSphenomenon for quantification and fulfill the potential of thetechnique touted since its discovery 20 years ago.

■ EXPERIMENTAL SECTIONChemicals. HPLC degree (≥98%) Enrofloxacin (ENRO)

and ciprofloxacin (CIPRO-C12,N14) were supplied by Sigma-Aldrich. CIPRO-C13,N15 isotope in HCl, 100 μg mL−1 inmethanol, HPLC degree (≥98%), was purchased fromCambridge Isotope Laboratories. Gold(III) chloride trihydrate(HAuCl4.3H2O, 99.9%), sodium citrate tribasic dihydrate(C6H5Na3O7 .2H2O, 99%), and (3-aminopropyl) -trimethoxysilane (APTMS, 97%) were purchased from theSigma-Aldrich (St. Louis, US). The chemical structure ofCIPRO-C12,N14, CIPRO-C13,N15, and ENRO are shown,respectively, in the Supporting Information file (SI file, FigureS3).Synthesis of AuNPs. The synthesis of gold nanoparticles

(AuNPs) was carried out using a procedure describedelsewhere.21 Briefly, 37 μL of gold(III) chloride trihydratewas diluted in a 25 mL volumetric flask and then transferred toa 250 mL Erlenmeyer flask, where the volume was adjusted byadding water up to 100 mL. The solution was heated untilboiling at constant stirring and then 3 mL of citrate sodium(1%) solution was added. A change in color to red-wineindicated the formation of the gold nanoparticles, and theheating was stopped. The AuNPs colloid was characterized byUV−vis and TEM (see SI, Figure S4).Immobilization of AuNPs on Coverslip Glass. Micro-

scope coverslip glass (18 mm (W) × 18 mm (L) × 0.15 mm(T)) were purchased from Fisher Scientific. The coverslipglasses were carefully washed in piranha solution (3:1concentrated H2SO4:30% H2O2) for 1 h (CAUTION! Thisreaction releases large amounts of corrosive fumes) and thenrinsed thoroughly with high amounts of water. Then, thecoverslip glasses were allowed to dry in an oven (∼1 h). Thecleaned coverslip glasses were placed in a coplin jar andcompletely covered by an 30% APTMS solution in toluene.The immersed coverslips were left overnight to ensure uniformmolecular packing at the surface. The coverslip glasses modifiedwith APTMS (glass-APTMS) were then removed from thecoplin jar, thoroughly washed with ethanol (HPLC grade) andthen immersed in (HPLC grade) ethanol overnight. Thisethanol washing/immersion procedure was repeated once a dayfor 5 consecutive days. After the fifth day, the glass-APTMS wasremoved from the ethanol and dried in an oven at 110 °C for 3h. The drying step facilitates cross-linking of the siloxane (Si−O−Si) bonds. Additional details about the glass functionaliza-tion procedure can be found elsewhere.22

The glass-APTMS substrates were immersed in the AuNPssuspension (diluted 1:1 with ultrapure water) for 2 h underconstant stirring. This procedure led to a highly homogeneouscoverage of AuNPs on the glass surface. Finally, the glass-

APTMS-AuNPs substrates were rinsed with water, dried using agentle N2 flow, and stored under air. The homogeneity of theresulting substrate (glass-APTMS-AuNPs) was probed by bothAFM and SERS (see the SI file).

SERS Measurements. SERS mapping measurements,depicted in Figure 1, were collected on a dispersive inVia

Raman confocal microscope system (Renishaw) equipped witha He−Ne laser source at 632.8 nm, a 50× objective (NA =0.75), and a motorized stage control. The measurements werecarried out using the StreamLine (Renishaw) operation mode.The mapped spectral range was from 1086 to 1667 cm−1. ASERS band observed at ∼1390 cm−1 was assigned to thestrongest O−C−O stretching mode,23 which is common for allanalytes (CIPRO-C12,N14, CIPRO-C13,N15, and ENRO). Allspectral acquisition was performed using 50% of the laserpower (around 10 mW at the laser head), 5 s acquisition timeand 1 accumulation.Aqueous analyte solutions, ENRO or CIPRO-C12,N14 or a

CIPRO-C12,N14 and CIPRO-C13,N15 mixture, were added tothe SERS substrate (glass-APTMS-AuNPs) as illustrated inFigure 1. For each experiment, 1 μL of the analyte solution wasdropped onto substrate (the average diameter of the drop onthe slide was ∼1 mm). Then, the drop was dried using a gentleN2 gas flow. The dry sample avoids adsorption dynamics,24

dilution factors,25 and cross contamination since a highnumerical aperture (NA) objective can be used without riskingcontact between the lens and the solution, which could severelycontaminate the results at ultralow concentrations. Notice that,since we are using a dried sample, the edges of the sample stainin the substrate were excluded to avoid bias due to, for instance,coffee ring effects. Our approach shows reasonable homoge-neity only when the middle of the dried sample was probed.The laser was then positioned to the middle of the dried dropusing the motorized stage control. The mapped areas were 50× 50 μm2 with 5 μm step size, totaling 121 SERS spectra(pixel) per map. The process was repeated for each one of theconcentrations in different regions around the middle of thedried sample spot.Data analysis. The raw spectra were stored in.txt format and

directly exported to the Matlab (version 7.12.0) environment.The raw spectra were preprocessed as follows: (1) Outlierspectra were removed; (2) the remaining spectra was smoothed

Figure 1. Scheme illustrating the experimental procedure carried outin this work.

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.7b03968Anal. Chem. 2018, 90, 1248−1254

1249

using a Savitzky-Golay filter;26 and (3) the baseline wascorrected using an asymmetric least-square method.27 Afterpreprocessing, the data were first analyzed by PCA todetermine the number of factors, and then the non-negativematrix factorization with alternating least-squares algorithm(NMF-ALS) resolution method was employed.23,28,29 Finally,the SERS response in each pixel was “digitized” by attributingvalues of “1” or “0” when the SERS response was either “above”or “below” a predetermined threshold. A comprehensivedescription of the data analysis procedure, including theoriginal spectra and the resulting NMF loads, can be foundas Supporting Information.

■ RESULTS AND DISCUSSIONThe SERS substrate chosen for this work was simply Aucolloids21 immobilized on an aminated glass slide (glass-APTMS-AuNPs). This common substrate assures that theprocedure developed here can be widely tested, since it doesnot require high-end fabrication tools. The molecular probesused for the method development, enrofloxacin (ENRO) andciprofloxacin (CIPRO), are nonresonant molecules (they donot have electronic absorptions that coincide with thewavelength of the laser excitation). As one of the most widelyused type of fluoroquinolone antibiotics, ENRO and CIPROare prominent emerging contaminants that can be found aspollutants in surface waters.22−24

The problem of quantification at ultralow concentrations inSM-SERS can be viewed as a typical sampling issue, which iscommon in analytical chemistry, and recently has been framedin the context of SERS-based assays with nanoparticles.27

Sampling errors can be curtailed by increasing the sample size.For instance, increasing the laser illuminated area to severalhundreds of μm2 would guarantee that a larger number ofmolecules adsorbed to efficient hotspots are probed, even atultralow concentrations. However, this approach might sufferfrom lower excitation power density and a decreased efficiencyin the collection optics relative to a high NA objectivemicroscope lens. Raman microscopes are commerciallyavailable, and they allow tight focus in small areas (∼1 μm2,depending on the excitation wavelength). Prolonged laserillumination in such a small area in a SERS substrate can lead tophotodecomposition. Therefore, the best approach to increasesample size is through spatial mapping.28,29 In the case ofspatial mapping, depicted in Figure 1, a diffraction limited laserexcitation spot is scanned (X−Y directions) through the surfaceof the SERS substrate, and a SERS response is recorded fromeach laser position (pixel).SERS maps were then performed in a central 50 × 50 μm2

area on the dried sample spot on the substrate. The SERSintensities from all maps from different concentrations (121data points each) were then evaluated using the non-negativematrix factorization with alternating least-squares algorithm(NMF-ALS).30 The NMF-ALS is a multivariate curveresolution (MCR) method that have already been applied inSERS (see details in the SI).23,31

Evaluation of Distribution of Scores in Ensemble andSM-SERS Conditions. Figure 2 shows normalized NMFscores (equivalent to SERS intensities) histograms obtainedfrom the SERS mappings (details in the SI) at twoconcentrations of ENRO. At high concentrations (278 μM),shown in Figure 2A, every pixel illuminated by the laser in themap was expected to contain a large number of adsorbedspecies (the probability of observing a SERS signal in every

illuminated area was high). Ideally, the molecules of the analytewere homogeneously distributed on the surface and thevariation in scores values in Figure 2A should reflect the spatialdistribution of SERS efficiency expected from a random SERSsubstrate. The red line is the mean score, suggesting that thecentral limit theorem was followed in Figure 2A (“quasi-normal” distribution). A typical spatial variation in SERS scoresobtained from the maps was of the order of 15% RSD. At lowerconcentrations, such as in Figure 2B, the shape of the SERSdistribution histograms changes from a quasi-normal behaviortoward a tailed behavior (noted that a true Gaussian was notobtained in Figure 2A due to the small sample size (number ofpixels)). The tailed distribution observed in Figure 2B is aconsequence of a decreased probability of molecules to findhighly efficient hotspots.1,15,17 In that case, a large number ofpixels yielded small score values (SERS intensities) than the redline (average), while a few pixels produced SERS signal muchlarger than the average.15

Evaluation of Calibration Curves in Ensemble and SM-SERS Conditions. Pseudocalibration curves were plotted usingthe average values of the NMF-ALS scores at 95% confidencelevel and the surface density of the analyte (proportional to the

Figure 2. Histogram showing the frequency of normalized scores from“factor 1”. “Factor 1” corresponds to the spectral signature of theanalyte obtained by PCA. (A) high concentration regime (278 μM or2.13 × 108 molecules/μm2) and (B) ultralow concentration regime(278 pM or 213 molecules/μm2). The red line is the mean value foreach distribution. More details can be found in the SI.

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.7b03968Anal. Chem. 2018, 90, 1248−1254

1250

solution concentrations). Typical pseudocalibration curves areshown in Figure 3. A good linear relationship is observed in

Figure 3A. This is in agreement with several previous reportsthat demonstrated that reliable calibration curves can beobtained by SERS.30 The variation in scores at highconcentrations, Figure 3A, should reflect the spatial variabilityof the SERS efficiency of the substrate (assuming that thesurface coverage is uniform at high concentrations). Moreover,Figure 3A shows that the SERS signal tracks the concentrationswithin a very wide dynamic range. The estimated error in aconcentration determination using this relative simple substratewas ±13%.The extended linearity observed in Figure 3A is lost when the

concentration of ENRO decreased to ultralow levels (less than70.0 nM), as shown in Figure 3B. As the concentrationdecreases, the number of species under the illuminated area (1μm2) also decreases. Since highly enhancing hotspots are rare,14

the probability of multiple molecules occupying hotspots underthe probed area also becomes very small. In extreme conditions,the observed SERS signal from a pixel is believed to be derivedfrom single species adsorbed at a (rare) highly efficient hotspot.The enhancement efficiency vary widely between differenthotspots (and with the position of the molecule within ahotspot1); hence, a small number of SERS events should have avariety of different SERS intensity values, leading to an averagewith very large variations. This is well-reflected in Figure 3B,where the linearity, accuracy, and precision of the calibrationcurve was lost. Notice that other groups observed the same sortof limitation (lack of linearity in the calibration curve) whenstudying ultralow concentrations by SERS.31

Digital Calibration Curve for SM-SERS Experiment.The limitation encountered in Figure 3B can be overcome by“digitizing” the SERS response. The rationale is that, when theSM-SERS regime is achieved, the SERS signal from a particularpixel will most probably originate from a single molecule.Therefore, it is not necessary to consider the Raman intensitybut only attribute the intensity to a single molecule event. It isthen possible to correlate the solution concentration with thenumber of pixels that present a SERS response (digital counts)within a given mapped area (instead of the Raman intensity).Figure 4 shows the digitized version of the SERS maps

obtained at different (low) concentrations of ENRO. Each

Figure 3. Pseudocalibration curves for ENRO. (A) “High concen-tration” regime and (B) “low concentration” regime. The confidencelevel was 95%.

Figure 4. Digital mapping for several concentrations of ENRO under SM-SERS regime. The black squares correspond to positive SM-events(assigned to 1). The white squares correspond to negative SM-events (assigned to 0). The estimate number of molecules per area (μm2) is indicatedin red. The laser illuminated area was about 1 μm2.

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.7b03968Anal. Chem. 2018, 90, 1248−1254

1251

black spot in Figure 4 corresponds to a pixel that yielded aSERS response (scores above a set threshold) that can beassigned to the vibrational signature of ENRO. The white spotsin Figure 4 were assigned to the null events. The estimatednumber of molecules/μm2, calculated from the solutionconcentration, is also indicated in Figure 4. Notice that thedigital counts (black spots) decreased with the analyteconcentration, as expected (Figure 4).Figure 5 presents a revised version of Figure 3B, where the

NMF scores were substituted by the “SERS digital counts”

(“number of black spots”). Note that the digital counts shouldbe equivalent to the number of single molecules identified bySERS during the mapping. A linear digital calibration curve wasthen recovered and plotted in Figure 5 (equivalent data forCIPRO is shown as SI). It is clear from Figure 5 that the linearrange of the digital calibration curve was extended to lowerconcentrations using this procedure, potentially allowingquantification at ultralow concentration levels.The limit of quantification (LOQ) for ENRO estimated from

the data in Figure 5 was 2.8 pM or about 2 molecules per μm2.Proof of SM-SERS Regime: Isotopologue Experiment

for the Digital Calibration Approach. After the demon-stration of the digital SERS calibration (Figure 5), it isappropriated to confirm the main assumption; that is, that theSM-SERS regime was achieved at those ultralow concen-trations. There are several approaches to prove the SM regimein SERS, these included measuring Stokes/anti-Stokes ratio,32

investigating changes in the shape of the SERS intensityhistograms,33 and analyzing the characteristics of the SERSintensity fluctuations.17,34,35 However, the most convincingargument of SM-SERS has arisen from studies with mixtures ofisotopologues (molecules of the same compound, so sameaffinity to the surface, that differ due to isotopic composi-tion).36−38 The isotopologue SERS experiment demonstratedthat the unique vibrational signature of both (isotopologues)analytes rarely appear mixed and most of the individual eventsshow the vibrational fingerprint from either one or the othercompound. This agrees with the idea that single moleculeevents are rare and confirms (by an unique spectroscopicsignature) that the analyte producing the SERS signature isalone at a highly efficient hotspot.

Figure 6 shows SERS mapping obtained from mixtures oftwo isotopologues of CIPRO (CIPRO-C12,N14 and CIPRO-

C13,N15, molecular structures are indicated in the SI) ofdifferent compositions. The results in Figure 6 were obtainedby adding 1 μL of mixtures (at different concentration ratios) ofboth isotopologues (CIPRO-C12,N14 and CIPRO-C13,N15) onglass-APTMS-AuNPs, following the same procedure as inFigures 3 and 4. In contrast to previous isotopologues SERSexperiments, the SERS fluctuations were not observed againsttime (this would not be reliable because the samples were driedand illumination on a fixed spot could lead to severephotodecomposition).32 The spectral signature from eachpixel was analyzed using the NMF-ALS resolution method,which recovered the unique vibrational signature of eachisotopologue. Any pixel that showed exclusively the presence ofCIPRO-C12,N14 is colored “blue” in Figure 6A,C. On the otherhand, pixels with a vibrational signature of the isotopicallysubstituted CIPRO (CIPRO-C13,N15) are colored “red” inFigure 6B,D. The SERS digital counts are higher for bothisotopologues at 140 pM (about 30−31 pixels) than at 14 pM(about 6−10 pixels). This is in agreement with the assumptionthat, at very low concentrations, the probability for a moleculeto find a hotspot is small, and it becomes rarer when theconcentration of specie(s) drastically decreases (140−14 pM).Second, it is important to emphasize that most of the pixelspresent either spectral noise or a signature of one of theanalytes (CIPRO-C12,N14 or CIPRO-C13,N15). Nine out of 61positive events (30 and 31 events of CIPRO-C12,N14 andCIPRO-C12,N14, respectively), ∼13%, are represented in“purple” color in Figure 6A,B. These “purple” events revealedmixtures between the two isotopologues at the same spot at

Figure 5. Digital calibration curve for ENRO at ultralowconcentration.

Figure 6. Digital SERS mapping from isotologue experiments (A) and(B) 140 pM (∼100 molecules/μm2) of CIPRO-C12,N14 (blue box)and (B) CIPRO-C13,N15 (red box); (C) and (D) 14 pM (∼10molecules/μm2) of CIPRO-C12,N14 (blue box) and CIPRO-C13,N15(red box). Purple boxes correspond to simultaneous detection ofCIPRO-C12,N14 and CIPRO-C13,N15.

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.7b03968Anal. Chem. 2018, 90, 1248−1254

1252

140 pM. On the other hand, only one pixel show the presenceof the two compounds simultaneously when the concentrationwas decreased to 14 pM (Figure 6C,D), representing ∼6% ofthe positive events. In summary, the probability of observingsingle molecule events in a particular pixel is higher when theconcentration is reduced from 140 (87%) to 14 pM (94%).These observations are strong indications that the singlemolecule regime was attained and it corroborate previousstudies.38

The results reported here indicate that SM-SERS combinedwith chemometrics can be used to directly quantify analytes atultralow concentrations, i.e., without requirements forpreconcentration. Separation steps might also be excluded,since SERS provided vibrational fingerprint identification.39 It isimportant to emphasize that the procedure described hereassumes SM-SERS conditions and a uniform distribution of theanalyte at the surface. Spatial variation of SERS efficiency of thesubstrate is also an important parameter. Substrates with spatialvariation in SERS intensities smaller than 20% RSD arerequired for this procedure.40,41 SERS substrates with lowspatial variation in intensities are now commercially availableand that should not limit the applicability of the method. Infact, the limiting aspect of the procedure should be related thesize of the sample set or, in the proof of concept presentedhere, the number of SERS digital counts (N) per mapped area.Considering basic counting statistics,42 the best estimate of thecount should be N ± N1/2 and the fractional uncertainty (error,%) of the count should decrease as N increases according:

= ×error(%)1N

100(1)

Therefore, increasing the mapped area would provide moreopportunities (pixels) for single molecule detection, leading tolarger N. Although large area Raman mapping has beennotorious time-consuming, almost all of the new generationcommercial Raman microscopes present options for fastmapping, which should be helpful on the verification andimplementation of the digital SERS procedure proposed here.

■ CONCLUSIONS

A quantification procedure based on SM-SERS statistics wasintroduced. This method allows the direct determination ofultralow concentrations by exploring the stochastic nature ofSM-SERS. The concept consists of depositing a low volume ofan aqueous sample at low concentration in a planar SERS activesubstrate. The surface of the substrate is then spatially mappedwith a laser beam focused on a small area (1 μm2, for instance).The goal is to ensure that the SERS response from eachilluminated spot will most probably be generated by a SM-SERS event. The SERS signal was then “digitized” and thenumber of pixels that provide a SM-SERS response (SERSdigital count) was shown to be proportional to the solutionconcentration. The protocol suggested here provides a uniqueexample of application of the concepts of SM-SERS for highsensitive quantification. The method was developed using arelative common SERS substrate, and emergent contaminantswere used as molecular probes. This work should provide someimportant guidelines for the development of analyticalprotocols that take full advantage of the main characteristicsof SERS, including its unique selectivity and sensitivity, forultralow concentration quantification.

■ ASSOCIATED CONTENT*S Supporting InformationThe Supporting Information is available free of charge on theACS Publications website at DOI: 10.1021/acs.anal-chem.7b03968.

Details on the quantification method; details on theexperimental protocol; substrate characterization; che-mometric methods; and additional data for CIPRO.(PDF)

■ AUTHOR INFORMATIONCorresponding Author*E-mail: agbrolo@uvic.ca. Phone: 1 (250) 721-7167. Fax: 1(250) 721-7147.ORCIDCarlos Diego L. de Albuquerque: 0000-0002-7615-7325Alexandre G. Brolo: 0000-0002-3162-0881Author ContributionsAll authors have given approval to the final version of themanuscript.NotesThe authors declare no competing financial interest.

■ ACKNOWLEDGMENTSThis work was supported by NSERC, CFI, BCKDF andUniversity of Victoria. C.D.L.A. thanks the Brazilian fundingagency CAPES (process PDSE-99999.010880/2014-03) for aninternational visiting student fellowship.

■ REFERENCES(1) Le Ru, E. C.; Etchegoin, P. G. Annu. Rev. Phys. Chem. 2012, 63,65−87.(2) Kneipp, K.; Kneipp, H.; Itzkan, I.; Dasari, R. R.; Feld, M. S. Chem.Rev. 1999, 99, 2957.(3) Chen, H. Y.; Lin, M. H.; Wang, C. Y.; Chang, Y. M.; Gwo, S. J.Am. Chem. Soc. 2015, 137, 13698−13705.(4) Im, H.; Bantz, K. C.; Lindquist, N. C.; Haynes, C. L.; Oh, S. H.Nano Lett. 2010, 10, 2231−2236.(5) Yang, S. K.; Dai, X. M.; Stogin, B. B.; Wong, T. S. Proc. Natl.Acad. Sci. U. S. A. 2016, 113, 268−273.(6) Tabatabaei, M.; Najiminaini, M.; Davieau, K.; Kaminska, B.;Singh, M. R.; Carson, J. J. L.; Lagugne-Labarthet, F. ACS Photonics2015, 2, 752−759.(7) Cabrera, F. C.; Aoki, P. H. B.; Aroca, R. F.; Constantino, C. J. L.;dos Santos, D. S.; Job, A. E. J. Raman Spectrosc. 2012, 43, 474−477.(8) Liu, H. Q.; Li, Q.; Li, M. M.; Ma, S. S.; Liu, D. B. Anal. Chem.2017, 89, 4776−4780.(9) Patra, P. P.; Chikkaraddy, R.; Tripathi, R. P. N.; Dasgupta, A.;Kumar, G. V. P. Nat. Commun. 2014, 5, 4357 DOI: 10.1038/ncomms5357.(10) Simoncelli, S.; Roller, E. M.; Urban, P.; Schreiber, R.;Turberfield, A. J.; Liedl, T.; Lohmuller, T. ACS Nano 2016, 10,9809−9815.(11) Zhang, Y. Q.; Shen, J. F.; Xie, Z. W.; Dou, X. J.; Min, C. J.; Lei,T.; Liu, J.; Zhu, S. W.; Yuan, X. C. Nanoscale 2017, 9, 10694−10700.(12) LeRu, E. C.; Etchegoin, P. G. Principles of Surface-EnhancedRaman Spectroscopy: And Related Plasmonic Effects 2009, 1−663.(13) Le Ru, E. C.; Etchegoin, P. G. J. Chem. Phys. 2009, 130, 181101.(14) Fang, Y.; Seong, N. H.; Dlott, D. D. Science 2008, 321, 388−392.(15) Etchegoin, P. G.; Meyer, M.; Le Ru, E. C. Phys. Chem. Chem.Phys. 2007, 9, 3006−3010.(16) Pieczonka, N. P. W.; Aroca, R. F. Chem. Soc. Rev. 2008, 37,946−954.

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.7b03968Anal. Chem. 2018, 90, 1248−1254

1253

(17) dos Santos, D. P.; Andrade, G. F. S.; Temperini, M. L. A.; Brolo,A. G. J. Phys. Chem. C 2009, 113, 17737−17744.(18) Le Ru, E. C.; Grand, J.; Sow, I.; Somerville, W. R. C.; Etchegoin,P. G.; Treguer-Delapierre, M.; Charron, G.; Felidj, N.; Levi, G.;Aubard, J. Nano Lett. 2011, 11, 5013−5019.(19) Fan, M. K.; Cheng, F. S.; Wang, C.; Gong, Z. J.; Tang, C. Y.;Man, C. Z.; Brolo, A. G. Chem. Commun. 2015, 51, 1965−1968.(20) Zhang, Y.; Noji, H. Anal. Chem. 2017, 89, 92−101.(21) Lee, P. C.; Meisel, D. J. Phys. Chem. 1982, 86, 3391−3395.(22) Brito-Silva, A. M.; Sobral-Filho, R. G.; Barbosa-Silva, R.; deAraujo, C. B.; Galembeck, A.; Brolo, A. G. Langmuir 2013, 29, 4366−4372.(23) Hong, K. Y.; de Albuquerque, C. D. L.; Poppi, R. J.; Brolo, A. G.Anal. Chim. Acta 2017, 982, 148−155.(24) Haran, G. Isr. J. Chem. 2004, 44, 385−390.(25) Darby, B. L.; Le Ru, E. C. J. Am. Chem. Soc. 2014, 136, 10965−10973.(26) Savitzky, A.; Golay, M. J. E. Anal. Chem. 1964, 36, 1627−1639.(27) Eilers, P. H. C. Anal. Chem. 2004, 76, 404−411.(28) Albuquerque, C. D. L.; Poppi, R. J. Anal. Chim. Acta 2015, 879,24−33.(29) Lee, D. D.; Seung, H. S. Nature 1999, 401, 788−791.(30) Schlucker, S. Angew. Chem., Int. Ed. 2014, 53, 4756−4795.(31) Costa, J. C. S.; Cordeiro, D. S.; Sant’Ana, A. C.; Rossi, L. M.;Santos, P. S.; Corio, P. Vib. Spectrosc. 2010, 54, 133−136.(32) dos Santos, D. P.; Temperini, M. L. A.; Brolo, A. G. J. Am. Chem.Soc. 2012, 134, 13492−13500.(33) Bohn, J. E.; Le Ru, E. C.; Etchegoin, P. G. J. Phys. Chem. C 2010,114, 7330−7335.(34) Aoki, P. H. B.; Carreon, E. G. E.; Volpati, D.; Shimabukuro, M.H.; Constantino, C. J. L.; Aroca, R. F.; Oliveira, O. N.; Paulovich, F. V.Appl. Spectrosc. 2013, 67, 563−569.(35) Aroca, R. F.; Constantino, C. J. L. Langmuir 2000, 16, 5425−5429.(36) Dieringer, J. A.; Lettan, R. B.; Scheidt, K. A.; Van Duyne, R. P. J.Am. Chem. Soc. 2007, 129, 16249−16256.(37) Kleinman, S. L.; Ringe, E.; Valley, N.; Wustholz, K. L.; Phillips,E.; Scheidt, K. A.; Schatz, G. C.; Van Duyne, R. P. J. Am. Chem. Soc.2011, 133, 4115−4122.(38) Zrimsek, A. B.; Wong, N. L.; Van Duyne, R. P. J. Phys. Chem. C2016, 120, 5133−5142.(39) Lussier, F.; Brule, T.; Vishwakarma, M.; Das, T.; Spatz, J. P.;Masson, J. F. Nano Lett. 2016, 16, 3866−3871.(40) Fan, M.; Brolo, A. G. Phys. Chem. Chem. Phys. 2009, 11, 7381−7389.(41) Fan, M. K.; Andrade, G. F. S.; Brolo, A. G. Anal. Chim. Acta2011, 693, 7−25.(42) Cherry, S. R.; Sorenson, J. A.; Phelps, M. E. Physics in nuclearmedicine, 4th ed.; Elsevier/Saunders: Philadelphia, 2012; p xvii, 523.

Analytical Chemistry Article

DOI: 10.1021/acs.analchem.7b03968Anal. Chem. 2018, 90, 1248−1254

1254