The effect of DNA bases permutation on surface enhanced Raman scattering spectrum
TThe effect of DNA bases permutation on surface enhanced Ramanscattering spectrum
Shimon Rubin ∗ , Phuong H.L. Nguyen ∗ , Yeshaiahu Fainman ∗ Equal contributionDepartment of Electrical and Computer Engineering, University of California,San Diego, 9500 Gilman Dr., La Jolla, California 92023, USA
Abstract
Surface enhanced Raman scattering (SERS) process results in a tremendous increase of Ramanscattering cross section of molecules adsorbed to plasmonic metals and influenced by numerousphysico-chemical factors such as geometry and optical properties of the metal surface, orientationof chemisorbed molecules and chemical environment. While SERS holds promise for singlemolecule sensitivity and optical sensing of DNA sequences, more detailed understanding ofthe rich physico-chemical interplay between various factors is needed to enhance predictivepower of existing and future SERS-based DNA sensing platforms. In this work we report onexperimental results indicating that SERS spectra of adsorbed single-stranded DNA (ssDNA)isomers depend on the order on which individual bases appear in the 3-base long ssDNA dueto intra-molecular interaction between DNA bases. Furthermore, we experimentally demonstratethat the effect holds under more general conditions when the molecules don’t experience chemicalenhancement due to resonant charge transfer effect and also under standard Raman scatteringwithout electromagnetic or chemical enhancements. Our numerical simulations qualitativelysupport the experimental findings and indicate that base permutation results in modification ofboth Raman and chemically enhanced Raman spectra. a r X i v : . [ phy s i c s . c h e m - ph ] F e b NTRODUCTION
Surface enhanced Raman scattering (SERS) [1–3] is an intriguing effect which results inseveral orders amplification of Raman scattering cross section of molecules chemisorbedor physiosorbed to a rough metal surface. In particular, SERS enhancement factors (EFs),which characterize increase in molecular Raman scattering relative to under non-SERScondition [4], mostly stem from conjointly operating electromagnetic enhancement (EM)and chemical enhancement (CE) effects. While EM effect stems from local increase ofelectromagnetic fields in so-called hot spots which contribute to both larger inducedRaman dipole and intensified radiation of the vibrating dipole through excitation ofplasmon resonances in the metal nanostructure (operating as plasmonic nanoantenna[5, 6]), CE effect [7–9] stems from changes of the Raman polarizability and of themolecular electronic structure due to formation of chemical bonds between the relevantmolecule and the metal and therefore selectively operates on the formed metal-moleculecomplex depending on its’ specific properties [5, 10]. In particular, CE effect may beaccompanied by either resonant charge transfer between the adsorbed molecule and themetal substrate [11] facilitated by formation of charge-transfer states, or non-resonant(static) ground-state charge redistribution due to overlap of metal and molecularorbitals [12] and surface binding (see [13, 14] and references within) leading to staticfar-from-resonance changes of the molecular polarizability.Due to the high sensitivity of the SERS process inherent to Raman-based vibrationalspectroscopy with narrow spectral lines, leading (under some conditions) to singlemolecule detection [15, 16] as well as single base sensitivity [17, 18] in simple DNAsequences, SERS-based DNA sensing holds promise for numerous future bio-relatedapplications such as medical diagnosis, bio-analysis, and environmental monitoringwhere composition and sequencing analysis of the information-rich DNA moleculesserve as one of the central tools. Furthermore, the significant progress in more recentyears allowing individual nucleobases resolution [19] achieved by employing the closelyrelated tip-enhanced Raman spectroscopy (TERS) [20–23], and exploitation of CE effect[24] to achieve higher specificity in ssDNA sensing, both indicate potential applicationsof SERS/TERS for DNA sensing. The latter include DNA characterization with areduced number of thermocycling steps in polymerase chain reaction (PCR) amplification225], mutation detection due to molecular conformation change [26] or employingbranched DNA technique which relies on the signal amplification of the probe thatbinds to a specific nucleotide sequence [27], which are typically more sensitive thanconventional fluorescent labeling (see for instance a recent work [28]). Nevertheless, theextreme sensitivity of SERS with respect to numerous factors such as substrate quality,dielectric environment, as well as unknown molecule orientation, occasionally leadsto controversial results on gold [29] and silver [30] substrates, indicating that furtherresearch is needed to understand the basic physical mechanisms of plasmon-DNAinteraction and how to leverage them in order to enhance present capabilities, eventuallyleading to bio-medical applications which require higher precision than achieved today[31]. In particular, considering increasingly complex DNA sequences, where eachnucleotide in turn is composed of additional sub-units, is expected to introduce novelmechanisms that affect the corresponding Raman and SERS spectra; it is reasonableto assume that incorporating these effects might improve the accuracy of futureplasmonic-based detection schemes.In this work we experimentally and numerically study, for the first time to the bestof our knowledge, the effect of intra-molecular interaction between DNA bases on theresultant SERS and Raman spectra, by considering simple ssDNA sequences whichadmit identical number of bases appearing at different order along the sequence. Takingadvantage of the controlled distance between DNA bases in ssDNA molecules allowsus to investigate how the position of DNA bases and their nearest neighbors affect thefrequency, RA (Raman activity) and vibrational pattern of the corresponding normalmodes. Fig.1 presents schematic description of the key elements in our study whichincludes silver/gold nanorod array serving as a substrate for adsorbed ssDNA 3-merstrands, allowing to extract changes of the SERS spectra due to nucleobases permutation,as well as different units/groups in a single DNA base which are activated dependingon their position in ssDNA molecule. In particular, we experimentally measure SERSspectra of two groups of isomers classified according to the total number of nucleotidesof a given type; group ( ) comprised of ACA and CAA sequences which admit twoadenines (A) and a single cytosine (C), and group ( ) comprised of CAC and CCAwhich admit a single A and two C’s. We then analyze the corresponding SERS spectraand compare them by constructing the corresponding statistical correlation matrices3 IG. 1. Schematic description of the underlying concept of our work. (a) Key elements of theexperimental setup presenting a silver/gold nanorod array substrate of mean period Λ and height h , hosting ssDNA 3-mers allowing to compare SERS spectra of ACA with CAA and of CAC withCCA (only ACA and CAA are presented in the scheme). The permutation of DNA bases in thesequences leads to a change in the interaction between the DNA bases, schematically representedas different bonds between the bases, and to enhancement of different molecular vibrationalmodes with distinct dominant frequency (b) and different kinetic energy distribution among thedifferent atom groups (c); relative disks’ size indicate the corresponding kinetic energy stored inthe relevant atoms at the dominant mode. which provides a quantitative metric to characterize the changes of the spectra due topermutation of DNA bases. In addition to the experiments where ssDNA moleculesadsorbed directly to metal nanorod substrate array which facilitates both EM and CEeffects, we also consider Raman spectra of ssDNA molecules in bulk solution withoutCE and EM effects as well as ssDNA molecules adsorbed to gold and silver nanorodsubstrate arrays covered with a -nm thickness of Al O dielectric film. The latter is thinenough to facilitate EM effect but eliminate the CE effect, allowing to study the differencebetween the various spectra of ssDNA molecules without the charge transfer resonanteffect [24]. Furthermore, we employ density functional theory (DFT) which became awidespread method for calculating electronic excitations and related optical propertiesof molecules [32] (including Raman polarizability), to study the permutation effect onRaman spectra of ssDNA 3-mers with and without the CE effect. The latter is achieved by4ntroducing Ag silver nano-clusters [33] bond to each one of the nucleotides, allowingto study the effect of static non-resonant (i.e. far from resonance) CE effect on Ramanspectra and kinetic energy distribution due to electrostatic charge redistribution. Forsimplicity, our computational model studies single orientation of DNA bases relative tothe metal, and therefore it is not used for quantitative comparison against experimentalresults. In fact, predicting the binding sites of ssDNA molecules with the metal andthe resultant orientation of these molecules is still somewhat controversial; e.g. adeninebinding orientation [29] or orientation dependence due to interaction with magnesiumcations [34] (which are used in our experiments), is expected to lead to considerablevariation in the corresponding spectra.This work is structured as follows. First, we present DFT simulation results whichcompare Raman spectra of ssDNA 3-mers under nucleobases permutation with andwithout non-resonant CE effect. We then present experimental results of the ssDNA3-mers with an increasing number of enhancement effects; first we present Raman spectraacquired in bulk solution without EM and CE effects, then we present SERS spectra onnanorod array substrates with only EM effect, and finally SERS spectra when both EMand CE effects are present. We analyze the numerical and experimental data, determineprominent trends, and distinguish the spectra of permuted ssDNA 3-mers based on thecorrelation values of the elements in the corresponding statistical correlation matrix. RESULTSSimulation results
Fig.2 and Fig.3 present DFT simulation results of Raman spectra of ACA, CAA(group 1) and CCA, CAC (group 2) sequences, as well as the same sequences whereeach DNA base is also bound to Ag nano-particle. All calculated Raman intensityresults are normalized to unity, and Raman intensities of the cases with metal particlesadmit values higher by approximately factor of five or smaller. The broadening of thepeaks is manually introduced by Lorentzian functions with bandwidth full-width athalf maxima (FWHM) of cm − , where − cm − is usually employed to describevarious broadening effects [4]. Fig.2(c,f) and Fig.3(c,f) present the corresponding 2D5 IG. 2. DFT simulation results comparing between Raman spectra of ACA and CAA isomers(group 1) with and without bonded three Ag nanoparticles. (a,b) Present Raman spectra of ACAand of ACA-Ag molecules with dominant modes at . cm − and . cm − , respectively,whereas (d,e) present Raman spectra of CAA and CAA-Ag molecules with dominant modes at . cm − and . cm − , respectively. (c,f) Relative kinetic energy distribution among thedifferent atoms at the dominant normal modes with red disks and green disks correspondingto the cases with and without silver nanoparticles, respectively. The disks with black contoursin ACA and CAA molecules admit kinetic energy values larger by a factors . and . ,respectively. (cid:48) and 5 (cid:48) . The red/green disks representthe relative kinetic energy of each atom with/without CE effect at the correspondingdominant vibrational mode; solid contour lines around few of the atoms indicate that therelevant atom carries larger kinetic energy as described in figures’ captions. We followthe common convention and write the sequences from (cid:48) -end to (cid:48) -end [35], where forinstance, CAA stands for (cid:48) − CAA − (cid:48) .Our simulation results indicate that the modes can be classified according to thegroups that store the kinetic energy of the vibrating atoms, and that the normalfrequencies of these groups depend both on their position in the sequence as well ason the closest neighbor DNA bases. For instance, the frequencies and the RAs of the ringbreathing modes (RBMs) of adenine or cytosine bases, which involve relatively smallnumber of atoms, depend on their location in the sequence and are affected by the DNAbases permutation as we discuss in the following. Furthermore, changing the number ofDNA bases leads to frequency splitting of the normal modes due to partial degeneracylifting which stems from normal modes dependence on the position along the sequence.Fig.2(a,b,d,e) present Raman spectra of ACA and CAA molecules, group (1), showingintensification, decline and shift of the normal modes, where red and green colorsindicate the cases with and without Ag nanoparticles, respectively. To demonstrate basictrends and patterns we analyze the Raman spectra in the following four spectral regions:(1) − cm − , (2) − cm − , (3) − cm − and (4) − cm − ,and use the spectra presented in Fig.2 in conjunction with Table.1 summarizing modes’frequencies, RAs, as well as the corresponding Raman active group in the sequence. Notethat in some cases the dominant peak is composed of several peak of comparable Ramanintensity, and therefore the value of the maximal frequency presented in Fig.2(a,b,d,e)and Fig.3(a,b,d,e) is different than the frequency of the corresponding dominant modesanalyzed in Fig.2(c,f) and Fig.3(c,f). Spectral region (1): is dominant for ACA and CAA molecules due to normal modeswhich involve 3 (cid:48) sugar, central (cen) sugar and the single C base. For ACA the dominant7requency at . cm − , is mostly due to . cm − (RA = 39 . ˚A /amu) and . cm − (RA = 30 . ˚A /amu) modes. Both admit kinetic energy concentrated at 3 (cid:48) sugarand C with different out-of plane and ring breathing amplitude vibrations (see fig.2(c)for kinetic energy distribution), whereas kinetic energy of . cm − mode is alsoconcentrated at central sugar group. Interestingly, as we will see in the following, normalmodes with kinetic energy more concentrated mostly at one group are more prone tosplitting, i.e. appearance of similar modes with closely separated frequencies at differentgroups.In CAA molecule, which is obtained from ACA by permutation of (cid:48) A with C, a slightred shift is observed at the peak vibrational Raman intensity to . cm − , mostly dueto red shift of the . cm − mode to . cm − (RA = 37 . ˚A /amu) and to thedecrease of RA of the . cm − mode. Bonding to Ag nanoparticles results in a redshift of all modes in both ACA and CAA molecules, slight decrease of some modes RA(e.g. . cm − ) and more significant decrease of . cm − RA mode in CAA.
Spectral region (2): describes asymmetric RBM vibrations of the single C base andthe attached sugar molecule. Permutation of (cid:48) A with C introduces a relatively minorchange in the vibrational frequency from . cm − to . cm − and to a slightincrease in the corresponding RA (from . ˚A /amu to . ˚A /amu). Bonding to Ag nanoparticles leads to a blue-shift of the resonant frequency and to a decrease in the RAvalues. Spectral region (3): describes two normal modes of two different A bases, withpractically identical atomic displacements, but with different frequencies due to differentposition in the sequence. In particular, ACA admits second dominant peak at frequency . cm − which is a contribution of the following stretching-bending modes (whichlead to change in bonds’ length and angles): (cid:48) A at . cm − (RA = 59 . ˚A /amu)and similar mode of (cid:48) A at . cm − (RA = 50 . ˚A /amu). From Table.1 below welearn that under permutation these modes are concentrated at A bases, but with slightlydifferent values of frequency and RA. Under bonding to Ag nanoparticles all modes areintensified, slightly red shifted, and the vibrational modes become concentrated also inthe adjacent sugar groups. Spectral region (4): presents vibrational modes of two A bases around cm − withtwo slightly different frequencies. Permutation of (cid:48) A with C in ACA molecule leads to8light intensification of RA and red shift of the vibrational frequencies, whereas bondingto Ag nanoparticles leads to more prominent intensification of the RA values and redshift of practically all modes.Due to the chiral nature of DNA molecule, AAC sequence, obtained by permuting Cbase with 3 (cid:48) A base in CAA sequence, is expected to admit a different Raman spectra.Indeed in Supplementary Information we describe additional DFT simulation resultsindicating a red shift of cytosine active mode in the second spectral region comparedto ACA and CAA sequences, and also lower RA values of adenine modes in the fourthspectral region when bound to Ag nanoparticles.Consider now group (2) with sequences CAC and CCA, which are formed by replacingone of the A bases in group (1) by a C basis. Following our discussion above, we expectnormal modes splitting for the two C bases and less modes involving a single A base.Fig.3(a,b,d,e) present Raman spectra of CAC and CCA molecules, and similarly to thediscussion above for group (1), we consider the following four spectral regions: (1) − cm − , (2) − cm − , (3) − cm − and (4) − cm − . We useFig.3 results in conjunction with Table.2 summarizing modes’ frequencies, RAs, as wellas the active group in the sequences. Spectral region (1): describes mode splitting due to a pair of C bases which introducesadditional normal modes. In particular, the modes . cm − in ACA sequence and . cm − in CAA sequence split into the pair . , . cm − in CAC sequenceand into the pair . , . cm − in the CCA sequence (see Table.2). Furthermore, thepair . cm − and . cm − , which involves several groups in CAA sequence, joinsinto another mode which involves several molecular groups at frequency . cm − .Bonding to Ag nanoparticles leads to a drop in RA, especially for the . cm − modein CAC sequence. Spectral region (2): describes mode splitting of the . cm − and . cm − C-based modes in ACA and CAA sequences (with single C), respectively, to the C-basedpaired modes . cm − , . cm − and . cm − , . cm − in CAC andCCA sequences (with two C’s), respectively. Bonding to Ag nanopartices leads todecrease of all RAs except for the . cm − mode in CAC sequence which increasesfrom . ˚A /amu to . ˚A /amu. Spectral region (3): describes the activity of a single Raman active A-based normal9
IG. 3. DFT simulation results comparing Raman spectra between CAC and CCA isomers(group 2) with and without three Ag nanoparticles. (a,b) Present Raman spectra of CAC andof CAC-Ag at dominant modes . cm − and . cm − , respectively, whereas (d,e)present Raman spectra of CCA and CCA-Ag with dominant modes at . cm − and . cm − , respectively. (c,f) Relative kinetic energy distribution among the different atoms of thedominant modes with red disks and green disks corresponding to the cases with and withoutsilver nanoparticles, respectively. The disks with black contours around red and green disks inCAC molecule indicate atoms with kinetic energy . and . larger, respectively, whereas inCCA molecules admit kinetic energy values larger by a factors . and . , respectively. ame Frequency [cm − ] Raman Activity [ ˚A /amu] Mode locationACA . .
34; 775 .
52; 779 . . .
37; 1376 . .
24; 1542 .
54 2 . .
91; 39 .
88; 30 . . .
78; 50 . .
11; 44 . CC; 3 (cid:48) sugar, cen sugar, C; C, (cid:48) sugarC, cen sugar (cid:48) A; (cid:48) A (cid:48) A; (cid:48) ACAA .
82; 778 . .
98; 776 . . .
52; 1376 . .
86; 1542 .
66 6 .
47; 18 . .
36; 17 . . .
40; 54 . .
33; 47 . C; CC, 3 (cid:48) sugar, 3 (cid:48)
PHB; C, 3 (cid:48) sugar, 3 (cid:48)
PHBC, cen sugar3 (cid:48)
A; cen Acen A; (cid:48) AACA-Ag .
65; 790 . . .
57; 1371 . .
92; 1543 .
55 23 .
76; 38 . . .
06; 110 . .
83; 106 . C; CC, cen sugar (cid:48) A, (cid:48) sugar; (cid:48) A, (cid:48) sugar (cid:48) A; (cid:48) ACAA-Ag .
75; 781 . . .
45; 1368 . .
41; 1543 .
41 58 .
46; 4 . . .
96; 201 . .
31; 147 . (cid:48) sugar; CC, cen sugarcen A, cen sugar; 3 (cid:48) A, 3 (cid:48) sugarcen A; 3 (cid:48)
ATABLE 1. DFT simulation results presenting the effect of C-3 (cid:48)
A permutation as well as basesbonding to Ag particles on the frequency and the Raman activity of the normal modes in the fourdifferent spectral regions: − cm − , − cm − , − cm − and − cm − . mode, as opposed to a pair of similar A-based modes in group (1). Under permutationfrom CAC to CCA the resonant frequency is red-shifted from . cm − to . cm − with slight increase in RA as further described in Table.2. Bonding to Ag nanoparticles leads to red shift and intensification of this mode in both CAC and CCAsequences (see Table.2). 11 pectral region (4): Under permutation of 3 (cid:48)
C with A, RA of A mode in CAC sequenceincreases from . ˚A /amu to . ˚A /amu in CCA sequence, and to decrease ofRA values of the C-based pair modes. Bonding to Ag nanoparticles leads to strongintensification of both modes in the CCA-Ag sequence. In particular, RA of cen C modein CCA sequence with RA = 29 . ˚A /amu increase to RA = 282 . ˚A /amu in CCA-Ag sequence.Due to chiral nature of DNA molecule ACC sequence, obtained by permuting A basewith 5 (cid:48) C base in CAC sequence, is expected to admit a different Raman spectra. InSupplementary Information we describe additional DFT simulation results which allowto compare ACC to other sequences. These include blue shift of the cytosine modes infourth spectral region, . cm − and . cm − in CAC sequence to . cm − and . cm − in ACC sequence. Furthermore, from Table.2 and Table.S1 we learnthat cytosine modes, . cm − and . cm − , in CCA-Ag sequences experienceblue shift to . cm − and . cm − in ACC-Ag sequences accompanied with adecrease in RA.Importantly, as indicated in Fig.S1, each one of the ssDNA molecules ACA, CAA,CAC, and CCA admit different spatial conformation. The different conformationswithin each one of the groups (1) and (2), due to permutation modified intramolecularinteraction, leading to different distances between the atoms and eventually tomodification of normal modes and Raman/SERS spectra. We then assume that theoptimized geometry holds also for the case when the sequences are bond to metalnanolusters, somewhat mimicking behavior of real molecules acquiring a given shapeprior to metal adsorption. Metal atoms further modify polarizability properties of thessDNA molecules leading to modification of the SERS/Raman signal and the kineticenergy distribution of the dominant modes as indicated above. Experimental results
Fig.4 and Figs.5-7, respectively, present experimental Raman and SERS measurementresults under exciting laser of wavelength nm, where each measurement is conductedseparately on each one of the following molecules ACA, CAA, CAC, CCA; the firstand the second pairs are referred below as group (1) and group (2), respectively. First,12 ame Frequency [cm − ] Raman Activity [ ˚A /amu] Mode locationCAC .
33; 769 . .
42; 773 . . .
62; 1242 . . .
30; 1572 .
78; 1573 .
49 7 .
25; 45 . .
34; 1 . . .
46; 17 . . .
28; 36 .
30; 32 . (cid:48) C; 3 (cid:48) C3 (cid:48) C; 5 (cid:48) C3 (cid:48) sugar, cen sugar, cen PHB, 3 (cid:48) PHB, 3 (cid:48) sugar5 (cid:48)
C, 5 (cid:48) sugar; 3 (cid:48)
C, 3 (cid:48) sugarAA; (cid:48) C; (cid:48) CCCA .
82; 766 . . .
20; 779 . .
04; 1231 . . .
30; 1572 .
84; 1573 .
99 4 .
62; 5 . . .
87; 32 . .
34; 49 . . .
77; 33 .
69; 29 . cen C; 5 (cid:48) Ccen C, 3 (cid:48) sugar, 3 (cid:48)
PHB, 5 (cid:48) C5 (cid:48) C; cen C, 3 (cid:48) sugar, 3 (cid:48)
PHBcen C; 5 (cid:48)
CAA; 5 (cid:48)
C; cen CCAC-Ag .
73; 769 . .
58; 782 . .
54; 1250 . . .
52; 1565 .
53; 1568 .
45 2 .
62; 5 . .
42; 34 . .
48; 39 . . .
22; 28 .
48; 46 . (cid:48) C; 5 (cid:48) C5 (cid:48) C; 3 (cid:48)
C, cen sugar3 (cid:48)
C, 3 (cid:48) sugar; 5 (cid:48)
C, 5 (cid:48) sugarA, cen sugarA; (cid:48) C; (cid:48) CCCA-Ag .
95; 769 . .
26; 789 . .
27; 1238 . . .
09; 1568 .
30; 1569 .
83 1 .
61; 2 . .
66; 45 . .
89; 30 . . .
58; 282 .
20; 69 . cen C; 5 (cid:48) Ccen C; cen C, PHB, 5 (cid:48)
Ccen C; 5 (cid:48)
CA, 3 (cid:48) sugarA; cen C; 5 (cid:48)
CTABLE 2. DFT simulation results presenting the effect of C-3 (cid:48)
A permutation and Ag nanoparticles bonding to the molecules, on the normal modes’ frequencies and Raman activitiesin the four spectral regions: − cm − , − cm − , − cm − and − cm − . IG. 4. Experimental results presenting mean Raman spectra and correlation matrices of measurements of ACA, CAA, CAC and CCA, ssDNA molecules in bulk solution. (a) Mean Ramanspectra of ACA (blue) and CAA (red), and (b) presents the corresponding statistical correlationmatrix. (c) Mean Raman spectra of CAC (blue) and CCA (red), and (d) presents the correspondingstatistical correlation matrix. we consider Raman signal of ssDNA molecules in IDTE bulk solution, i.e. withoutmetal substrate and therefore without the accompanying EM and CE effects. Theorder of magnitude of the EM enhancement factor, as defined according to [7, eq.4],which compares the total SERS signal from a given surface element per estimatednumber of molecules adsorbed to the illuminated surface area, to the total Raman signalfrom a given spatial illuminated volume (defined by beam waist and Rayleigh length)per number of molecules in that volume, is ∼ . Fig.4(a,b) and Fig.4(c,d) presentcomparison between Raman spectra of group (1) and between group (2) sequences,respectively, where for clarity of presentation range bars are presented at 12 points14 IG. 5. Experimental results presenting the effect of permutation of DNA bases, adsorbed tonanorod gold substrate covered with nm thick Al O film, on the mean SERS spectra andcorrelation matrices without the CE effect. The experimental set is comprised of , , and measurements of ACA, CAA, CAC and CCA, ssDNA molecules, respectively. (a) Mean SERSspectra of ACA (blue) and CAA (red), and (b) presents the corresponding correlation matrix. (c)Mean SERS spectra of CAC (blue) and CCA (red) and (d) presents the corresponding correlationmatrix. along the profile and indicate 95% confidence intervals on the mean. Importantly,here and elsewhere unless otherwise specified, the experimental data was subject tomultiplicative-scattering correction (MSC) in order to filter out various noises, such asmultiplicative light scattering [36]. In fact, unless ssDNA molecules directly adsorbed tothe metal, MSC is necessary to observe the emerging square regions in the correlationmatrices Fig.4(b,d) (as well as in Fig.5(b,d) and in Fig.6(b,d)). The latter are defined as15he ratio of the covariance of the two variates to the variance product of the independentvariates [37], admit close to unity values in the self-correlation blocks (e.g. ρ ( ACA , ACA ) , ρ ( CAA , CAA ) , and lower values in the cross-correlation blocks (e.g. ρ ( ACA , CAA ) ),signaling a difference in Raman spectra due to nucleobases permutation.Next we consider the case where ssDNA molecules are adsorbed to nm thick Al O dielectric film deposited on top of oblique-angled deposition (OAD) [38] fabricated goldnanorod arrays. In so doing we introduce EM effect due to gold nanorod array buteliminate the resonant charge transfer CE effect. Figs.5(a,c) present mean SERS spectraof ACA and CAA sequences, respectively, accompanied by the corresponding statisticalcorrelation matrices Fig.5(b,d). The latter indicate a close to unity correlation in thediagonal blocks and smaller correlation in the off-diagonal block which is more visiblebetween group (2) sequences. In order to eliminate charge transfer CE effect, we performsimilar experiments with ssDNA adsorbed to nm thick Al O dielectric film depositedon top of silver OAD nanorod arrays. Fig.6 presents the corresponding experimentalresults where Figs.6(a,c) illustrate the mean spectra of group (1) and (2) ssDNA, andFigs.6(b,d) present the correlation matrices indicating the differences between the twogroups. Similarly to the case of gold nanorod arrays coated with Al O , presentedin Fig.5 above the observed differences between group (1) molecules are smaller thanbetween group (2) molecules; the results in Fig.6(a,b) lead to very close signal and thesequences ACA, CAA adsorbed to silver nanorod array coated with Al O are practicallyindistinguishable. Interestingly, the pattern of smaller differences between group (1)molecules than between group (2) molecules is also consistent with the computationalspectra which can be seen by visually comparing the mean spectra of group (1) molecules,Fig.2(a) vs Fig.2(d), and Fig.2(b) vs Fig.2(e); group (2) molecules, Fig.3(a) vs Fig.3(d), andFig.3(b) vs Fig.3(e) due to more pronounced changes of the (cid:48) C relative to (cid:48) A due to A-Cpermutation of the nearest nucleobase.Next we consider the case where ssDNA molecules are adsorbed to silver nanorodarrays which support both CE and EM effects. Figs.7(a,c) present SERS spectra ofthe four ssDNA sequences with similar range bars as presented in figures above.Fig.7(b,d) present the corresponding statistical correlation matrices indicating close tounity correlation in the diagonal blocks and smaller correlation in the off-diagonal blocks.In fact, in this case MSC wasn’t implemented and signal processing wasn’t necessary16
IG. 6. Experimental results presenting the effect of permutation of DNA bases, adsorbed tonanorod silver substrate covered with nm thick Al O film, on the mean SERS spectra andcorrelation matrices without the CE effect. The experimental set is comprised of , , and measurements of ACA, CAA, CAC and CCA, ssDNA molecules, respectively. (a) Mean SERSspectra of ACA (blue) and CAA (red), and (b) presents the corresponding correlation matrix. (c)Mean SERS spectra of CAC (blue) and CCA (red) and (d) presents the corresponding correlationmatrix. to observe the difference between the permuted cases. Interestingly, unlike the casepresented in Fig.5 and Fig.6 above, where differences between group (1) sequencesare less visible compared to differences between group (2) sequences, in this case thedifferences between group (2) sequences is more visible compared to group (1) sequences.We attribute this difference to more prominent CE effect of adenine bases reported in the17 IG. 7. Experimental simulation results presenting mean SERS spectra of measurements ofeach ssDNA molecule adsorbed to nanorod silver substrate and the corresponding statisticalcorrelation matrices between the different measurements. (a) Mean SERS spectra of ACA (blue)and CAA (red), and (b) presents the corresponding statistical correlation matrix. Minimal valuesin the first and second diagonal block of the statistical correlation matrix are . and . ,whereas the off-diagonal blocks admit lowest values . . (c) Mean SERS spectra of CAC (blue)and CCA (red). (d) Minimal values in the first and second diagonal block of the diagonal matrixare . and . , whereas the off-diagonal blocks admit lowest values . . recent work [41]. 18 ethods DFT simulation setup and governing equations
In this work we employ time-dependent functional density theory (DFT) by usingGaussian software [39] at San Diego Supercomputer Center (SDSC) at the Universityof California, San Diego [40]) by implementing a built-in Becke’s three-parameters hybridfunctionals and Lee-Yang-Parr (B3LYP) electron correlation; − G ∗ basis set is usedfor H, C, P, O, N atoms and LANL2DZ basis set is used for Ag atoms to achievebetter accuracy. All simulated ssDNA molecules were electrically neutral singlets, whereadditional protons bonded to the oxygen atoms in the phosphate backbone screened thenegative charge of the phosphate group.The Raman cross section of mode k , ( dσ/d Ω) k , and the corresponding Raman intensity, I , are given by (cid:18) dσd Ω (cid:19) k = f ( ν L − ν k ) ν k B k ( T ) S k , (1a) I = (cid:88) k (cid:18) dσd Ω (cid:19) k L ( ν − ν k ) . (1b)Here, B k ( T ) and L ( ν − ν k ) are the temperature dependent correction and the Lorentzianfunction centered at ν k both given by B k ( T ) = 1 / (cid:0) − e − ν k /ν T (cid:1) , ν T ≡ hc/ ( k B T ) (2a) L ( ν − ν k ) = 1 πg g g + ( ν − ν k ) , (2b)respectively. Here, h , c , k B , T are the corresponding Planck’s constant, speed of light,Boltzmann constant, room temperature, respectively, whereas g is the half-width athalf height (HWHH) parameter of the Lorentzian which represents broadening effects.Typically, for room temperature values ( . K and ν T = 207 . cm − ) the temperaturedependent factor introduces correction below for frequencies above . cm − . Substrate and sample preparation
Silver and gold nanorod substrate arrays were fabricated using a single-step OADmethod [38] by employing Denton Discovery Sputter system. Nanorods formation19s governed by the so-called “shadowing effect”; when a material flux of depositedparticles is made incident on the target surface, initially formed nuclei structures preventthe deposition of later arriving particles in regions situated behind them (see [38] andreferences within), thus leading to growth of only initial nuclei. In our case the substrateis oriented at zenithal deposition angle o leading to a nanorod tilt angle of o relativeto the substrate normal. Fig.S2 in the Supplementary Material section describes SEMimages of a typical nanorod array used in our experiments.The metal nanorod arrays, which were used to eliminate charge trasfer CE effect,were covered with nm Al O thick film using Atomic Layer Deposition (ALD); see [41]for fabrication details and also sample preparation steps. Briefly, the ssDNA solutionswere prepared by diluting the DNA stock solution in 4-(2-hydroxyethyl)-1- piperazineethanesulfonic acid (HEPES) and MgCl . We then drop-casted this ssDNA solution ontothe metal nanorod substrate, let it dry overnight, rinsed with deionized water in order toremove excess of crystallized salt and unbound ssDNA molecules, and then blow-dried.For acquisition of Raman spectra of ssDNA molecules suspended in bulk solutionwe employed high throughput mm diameter Wilmad NMR tubes filled with ssDNAsequences in IDTE buffer (pH . , mM Tris-HCl/ . mM EDTA) normalized to aconcentration of µ M. SERS and Raman measurement and data processing
SERS and Raman measurements were acquired with Renishaw inVia Ramanspectrometer by employing mW continuous nm excitation wavelength; acquisitiontime of s per spectrum. The objective magnification used was X with N A = 0 . .The grating type used was l/mm at nm, and the grating setting was set toa static regime with spectrum range extending between cm − to cm − withspectral resolution around cm − . The mapping setting of the spectrometer was usedto acquire measurements from a total substrate area of dimensions × µm ; thisarea is divided into × square units, where the dimension of each area unit is × µm and each acquired spectrum measurement is taken from a different square area.Raman measurements of bulk solution (i.e. without metal substrate) were performed byemploying mW continuous nm excitation wavelength; objective magnification X N A = 0 . ; grating l/mm.After data acquisition, SERS/Raman spectra undergo the following signal processingsteps which include: baseline correction, cosmic ray removal and MSC. The first twowere implemented by employing built-in algorithms in Renishaw WiRE 4 software withGaussian smoothing window of length 20, whereas MSC was implemented in MATLABcodes. These procedures lead to small artifact regions with negative values of Ramanintensity, which do not affect the derived statistical correlation matrices.The statistical analysis was performed by employing built-in ’corrcoef’ function inMatlab (version 9.7.0.1261785 (R2019b), Natick, Massachusetts, The MathWorks Inc.)which calculates Pearson’s correlation matrix ρ ( A, B ) of two random variables A and B , defined as the ratio of the corresponding covariance matrix (cov), divided by thecorresponding standard deviations σ A,B , explicitly given by ρ ( A, B ) ≡ cov ( A, B ) σ A σ B ; cov ( A, B ) ≡ N N (cid:88) i =1 ( A i − µ A )( B i − µ B ) . (3)Here, cov reflects the amount of correlation between the magnitudes and directionality(i.e. positive or negative) of random variables deviation from the correspondingmean values µ A,B , normalized by standard deviations σ A,B . In our case A and B designate several tens of SERS/Raman spectra which belong to group (1) and group (2),respectively, and N = 1021 is the size of each spectra. DISCUSSION
In this work we experimentally and numerically studied the effect of spatialarrangements of DNA bases in ssDNA 3-mers on SERS and Raman spectra. Ourexperimental results, where CE and EM effects were gradually introduced into thesystem, clearly indicate a difference between the acquired SERS and Raman spectra ofsequences which differ under permutation of the DNA bases, and our DFT simulationsprovide qualitative evidence for dependence of normal mode frequency of DNA base onposition along the sequence, with and without CE effect. In particular, DFT simulationswithout metal indicate that frequency dependence and mode splitting are intrinsic21roperty due to complex structure of the molecule, being composed of basic units withpermutation dependent molecular conformations; complex interaction with the metalmay intensify this difference as it allows higher variability such as different orientationrelative to the metal surface. Furthermore, we expect that CE effect should play muchmore significant role in highlighting differences between different permuted moleculesthan EM effect; the plasmonic mode, which is responsible for the EM enhancementfactor, isn’t expected to be affected by refractive index changes due to sub-nanometermolecular permutations. Nevertheless, more realistic simulations are needed to providequantitative agreement by incorporating additional effects such as CE due to resonantcharge transfer between the adsorbed molecule and the metal substrate, as well asadditional dielectric layer between the molecule and the metal. While in this work weconsidered only 3-base long sequences to simplify the system and potentially enhance theeffect, we expect this work to facilitate future theoretical studies, as well as experimentalTERS-based sensing platforms, which admit sufficiently high spatial resolution to enableexcitation of just few bases [19, 42–44] to study sensing of longer ssDNA moleculeswhere permuted nucleobases are embedded in a longer DNA sequences. In particular,studying SERS/Raman spectra as a function of intermolecular [45] or nearest neighborinteraction may potentially enable more accurate SERS-based sensing platforms andallow lower read-out error needed to meet bio-applications [31], support DNA memoryapplications [46] and allow novel tool to control Raman/SERS response of complexmolecules. Furthermore, our DFT simulation, presenting mode splitting and normalmodes dependence on the position along the phosphate backbone, may lead to anovel structure dependent mechanism for inhomogeneous broadening of the peaks incomplex molecules such as ssDNA sequences, which may be of importance from bothfundamental and application perspectives.
FUNDING
This work was supported by the Defense Advanced Research Projects Agency(DARPA) DSO, NLM, and NAC Programs, the Office of Naval Research (ONR),the National Science Foundation (NSF) grants CBET-1704085, DMR- 1707641, NSFECCS-180789, NSF ECCS-190184, NSF ECCS-2023730, the Army Research Office22ARO), the San Diego Nanotechnology Infrastructure (SDNI) supported by the NSFNational Nanotechnology Coordinated Infrastructure (grant ECCS-2025752), and theASML-Cymer Corporation. [1] M. Fleischmann, P. J. Hendra, and A. J. McQuillan. “Raman spectra of pyridine adsorbed ata silver electrode,” Chem. Phys. Lett., 26:163 166, (1974).[2] D. L. Jeanmaire and R. P. Van Duyne. Surface Raman spectroelectrochemistry Part I.Heterocyclic, aromatic, and aliphatic amines adsorbed on the anodized silver electrode. J.Electroanal. Chem., 84:1 20, (1977).[3] M. G. Albrecht and J. A. Creighton, “Anomalously intense Ramanspectra of pyridine at asilver electrode,” J. Am. Chem. Soc., 99: 5215 5217, (1977).[4] E. Le Ru and P. Etchegoin,
Principles of Surface-Enhanced Raman Spectroscopy: and relatedplasmonic effects,
Elsevier, (2008).[5] M. Moskovitz,
Surface-Enhanced Raman Scattering—Physics and Applications , Vol. 103,Springer, Heidelberg, (2006).[6] M. I. Stockmann,
Surface-Enhanced Raman Scattering–Physics and Applications , Vol. 103,Springer, Heidelberg, 47–65, (2006).[7] E. C. Le Ru, E. Blackie, M. Meyer, and P. G. Etchegoin. “Surface enhanced Raman scatteringenhancement factors: a comprehensive study,” The Journal of Physical Chemistry C 111
38 (8) (5)
13] S.-Y. Ding, En-Ming You, Zhong-Qun Tian, and M. Moskovits. ”Electromagnetic theories ofsurface-enhanced Raman spectroscopy.” Chemical Society Reviews 46 (13) (5303) (15)5149-5154, (2015).[18] L. Guerrini, ˇZ. Krpeti´c, Danny van Lierop, Ramon A. Alvarez-Puebla, and D. Graham,“Direct Surface-Enhanced Raman Scattering Analysis of DNA Duplexes,” AngewandteChemie International Edition 54 (4) 1144-1148, (2015).[19] Z. He, Z. Han, M. Kizer, R. J. Linhardt, X. Wang, A. M. Sinyukov, J. Wang et al.,“Tip-enhanced raman imaging of single-stranded DNA with single base resolution,” Journalof the American Chemical Society 141 (2) 753-757, (2018).[20] R. M. St ¨ockle, Y. D. Suh, V. Deckert and R. Zenobi, Chem. Phys. Lett., 318, 131–136 (2000).[21] M. S. Anderson, Appl. Phys. Lett., 76, 3130–3132 (2000).[22] N. Hayazawa, Y. Inouye, Z. Sekkat and S. Kawata, Opt. Commun., 183, 333–336 (2000).[23] B. Pettinger, G. Picardi, R. Schuster and G. Ertl, Electrochemistry, 68, 942–949 (2000).[24] L. M. Freeman, L. Pang, and Y. Fainman, “Maximizing the electromagnetic and chemicalresonances of surface-enhanced Raman scattering for nucleic acids,” Acs Nano α -fetoprotein for early diagnosis of livercancer.” ACS applied materials & interfaces 10.41 34869 (2018).
28] X. Fu, J. Wen, J. Li, H. Lin, Y. Liu, X. Zhuang, C. Tian, and L. Chen, ”Highly sensitivedetection of prostate cancer specific PCA3 mimic DNA using SERS-based competitive lateralflow assay.” Nanoscale 11(33), 15530, (2019).[29] S. G. Harroun, “The Controversial Orientation of Adenine on Gold and Silver,”ChemPhysChem
Time-dependent density-functional theory: concepts and applications,
OUP Oxford,(2011).[33] R. Huang, et al. ”Structural and charge sensitivity of surface-enhanced Raman spectroscopyof adenine on silver surface: a quantum chemical study,” J. Phys. Chem. C
Molecular biologyof the gene,
Cold Spring Harbor Laboratory Press, Cold Spring Harbor, New York, (2014).[36] T. Isaksson, and T Næs, “The effect of multiplicative scatter correction (MSC) and linearityimprovement in NIR spectroscopy,” Applied Spectroscopy, , 1273-1284, (1988).[37] J. D. Gibbons and S. Chakraborti,
Nonparametric Statistical Inference: Revised and Expanded ,CRC press, (2014).[38] A. Barranco, “Perspectives on oblique angle deposition of thin films: from fundamentals todevices,” Prog. Mater. Sci.
41] P. H. L. Nguyen, B. Hong, S. Rubin, and Y. Fainman, “Machine Learning for CompositionAnalysis of ssDNA using Chemical Enhancement in SERS”, Biomedical Optics Express 11(9), 5092-5121.[42] A. Downes, D. Salter, and A. Elfick. “Finite element simulations of tip-enhanced Raman andfluorescence spectroscopy,” The Journal of Physical Chemistry B 110.13 6692-6698, (2006).[43] S. Trautmann, J. Aizpurua, I. G ¨otz, A. Undisz, J. Dellith, H. Schneidewind, M. Rettenmayr,and V. Deckert, “A classical description of subnanometer resolution by atomic features inmetallic structures.” Nanoscale
19 (8)
Funding:
This work was supported in part by the National Science Foundation(NSF), the Office of Naval Research (ONR), the Semiconductor Research Corporation(SRC), the Army Research Office (ARO), DARPA, Cymer Corporation, and the San DiegoNanotechnology Infrastructure (SDNI) supported by the NSF National NanotechnologyCoordinated Infrastructure (grant ECCS-1542148).
Competing interests:
The authors declare no competing financial interests.
Acknowledgments:
This work used the Extreme Science and Engineering DiscoveryEnvironment (XSEDE), which is supported by National Science Foundation grantnumber ACI-1548562. 26 upplementary Material for:“The effect of DNA bases permutation on surface enhanced Raman scatteringspectrum”
Shimon Rubin ∗ , Phuong H.L. Nguyen ∗ , Yeshaiahu Fainman ∗ Equal contributionDepartment of Electrical and Computer Engineering, University of California, San Diego, 9500 Gilman Dr., La Jolla, California 92023, USA
S.1. DOMINANT MODES OF AAC AND ACC MOLECULES
DNA molecules are inherently chiral and therefore in addition to ACA, CAA, andCAC, CCA sequences we also consider vibrational modes of AAC and ACC, whichare expected to introduce changes in the Raman spectra. Table.S1 and Table.S2 below,describe Raman modes in the spectral regions corresponding to regions specified inTable.1 and Table.2, respectively, with less modes presented in the first spectral region.
Name Frequency [cm − ] Raman Activity [ ˚A /amu] Mode locationAAC .
87; 775 . . .
36; 1376 . .
47; 1541 .
96; 1576 .
51 19 .
85; 55 . . .
36; 47 . .
32; 36 .
80; 37 .
31; 3 (cid:48) sugar, (cid:48) PHB, C; CC, (cid:48) sugarcen A, cen sugar; (cid:48) sugar (cid:48) A; cen A; CAAC-Ag .
36; 772 . . .
59; 1372 . .
26; 1545 .
89; 1570 .
77 83 .
76; 30 . . .
15; 108 . .
58; 106 .
19; 191 . C; (cid:48) sugar, (cid:48) PHB, cen sugarC, (cid:48) sugar (cid:48) A, (cid:48) sugar; cen A, cen sugar (cid:48) A; cen A; CTABLE S1. DFT simulation results presenting frequencies, RAs and the corresponding Ramanactive molecular groups in AAC and AAC-Ag molecules which belong to group (1). Fig.S1 presents the effect of partial bonding of Ag nanoparticles on the emerging SERSspectra. 1 ame Frequency [cm − ] Raman Activity [ ˚A /amu] Mode locationACC .
46; 779 . .
42; 1242 . . .
21; 1573 .
07; 1576 .
45 36 .
05; 22 . .
48; 19 . . .
43; 33 .
87; 34 .
46 3 (cid:48)
C; cen C, (cid:48) Ccen C, cen sugar; (cid:48) C, 3 (cid:48) sugarA, (cid:48) sugarA; cen C; (cid:48) CACC-Ag .
32; 791 . . . .
19; 1570 .
39; 1572 .
39 70 .
31; 47 . . . .
88; 90 .
51; 48 . cen C; 3 (cid:48) Ccen C, cen sugarAA; cen C; 3 (cid:48)
CTABLE S2. DFT simulation results presenting frequencies, RAs and the corresponding Ramanactive molecular groups in ACC and ACC-Ag molecules which belong to group (2). .2. SPATIAL CONFORMATIONS OF ACA, CAA, CAC, AND CCA MOLECULES The following image presents computational results of the optimized geometry forgroup (1) and group (2) ssDNA molecules.
FIG. S1. DFT results presenting optimized geometry of group (1) and group (2) ssDNA moleculesin the first and second rows, respectively. Red - oxygen (O), blue - nitrogen (N), grey - carbon(C), white - hydrogen (H), orange - phosphorous (P). In all molecules the orientation (coordinatesystem) was fixed by fixing the orientation of the four O’s which are bond to P at (cid:48) end. Inparticular P is chosen as origin; x-axis is parallel to a pair of O’s where each O is connected to asingle P and H atoms; y-axis is antiparallel to the P-O bond where O is bond to C; z-axis (towardsthe reader) is along the P=0 bond. S.3. ADDITIONAL DETAILS ABOUT FABRICATED NANOROD ARRAY
Fig. S2 presents SEM image of a typical OAD-fabricated nanorod array. The averageheight of the nanorods, is approximately h = 200 nm, whereas the average diameter of3he rods is approximately Λ = 50 nm for gold (and
Λ = 100 nm for silver)
FIG. S2. SEM image showing typical morphology of the OAD-fabricated gold nanorod array.
Based on the width of the error bars, which represent standard deviation due tomeasurements acquired at 25 different spatial locations on the substrate, relative to themean peak height, we conclude that the mean signal variation across all spectral domainsis approximately 6%.Based on our recent work [41], where identical nanorod arrays were employed,we bring here the corresponding values of the CE factors. In particular, these areapproximately ∼ − for gold (depending on the spectral region) and ∼ for silver. Thelatter were obtained by comparing between SERS spectra of molecules adsorbed directlyto the metal nanorod array to similar array coated with a nm thick Al O film, whicheliminates potential resonant charge transfer effect. Experimental measurement of totalenhancement factor, due to both EM and CE effects, requires determination of surfaceconcentration of adsorbed ssDNA molecules to the nanorod array after the rinsing step,which is beyond the scope of this work. S.4. ESTIMATION FOR THE EM ENHANCEMENT FACTOR
We define the EM enhancement factor (EF) for each one of the 3-mers employed in thiswork as EF = ¯ I SERS / ¯ I Raman , (S1)where ¯ I SERS and ¯ I Raman represent ”signal per molecule”, i.e. the integral of SERS(or Raman) spectra ( I SERS , I
Raman ) over the relevant spectral domain divided by the4orresponding number of molecules ( N SERS , N
Raman ) in the illuminated region and alsoby the illumination power ( P SERS , P
Raman ) , allowing to write the equation above as EF = I SERS N SERS · P SERS (cid:30) I Raman N Raman · P Raman . (S2)Based on our experimental spectra, the ratios of the signals I SERS /I Raman for thesequences ACA, CAA, CAC, CCA are: . , . , . , . , respectively; i.e. verysimilar and differ at most by a factor approximately two. Next, P SERS = 10 mW, P Raman = 100 mW, leading to P SERS /P Raman = 10 . Finally, we determine the number ofactive molecules in each one of the cases. Assuming the 3-base molecules cover uniformlythe × µ m sampling area, and the in-plane area of the molecule islength × width = (0 . nm × bases ) × . nm = 1 . · − µ m , (S3)yields the following number of molecules in the sampling area, N SERS = 25 / (1 . · − ) =1 . · . To determine N Raman , defined by, N Raman = illumination volume × copies per volume , (S4)we estimate illumination volume as πω L R = 2 . · − l. Here, ω = λ/ ( π · N A ) = 785 nm / ( π · .
12) = 2082 . nm is the radius of the Gaussian beam waist, λ is the wavelength of illumination laser in the free space, and NA is the correspondingnumerical aperture; L R = πω /λ = 17352 . nm is the Rayleigh length. The number ofmolecules per volume is a product of molar concentration, µ M = 10 − m/l, withAvogadro number, N A = 6 . · copies/m, leading to N Raman = 1 . · copies.Inserting, these parameters into the definition of the EF, yields the following similarvalues, EF = 2 . · , . · , . · , . ·2