Direct Observation of Single DNA Structural Alterations at Low Forces with Surface-Enhanced Raman Scattering

Abstract

DNA experiences numerous mechanical events, necessitating single-molecule force spectroscopy techniques to provide insight into DNA mechanics as a whole system. Inherent Brownian motion limits current force spectroscopy methods from observing possible bond level structural changes. We combine optical trapping and surface-enhanced Raman scattering to establish a direct relationship between DNA’s extension and structure in the low force, entropic regime. A DNA molecule is trapped close to a surface-enhanced Raman scattering substrate to facilitate a detectable Raman signal. DNA Raman modes shift in response to applied force, indicating phosphodiester mechanical alterations. Molecular dynamic simulations confirm the local structural alterations and the Raman sensitive band identified experimentally. The combined Raman and force spectroscopy technique, to our knowledge, is a novel methodology that can be generalized to all single-molecule studies.

Introduction

The DNA molecule acts as a platform for a host of critical functions such as transcription, replication, and other molecular motor-driven processes, where the DNA strand undergoes numerous mechanical events that are primarily supported by the polymer-like phosphate backbone. As a passive substrate, DNA maintains mechanical compliance to allow interactions with proteins. Some of these interactions occur at low forces where the mechanical load on DNA is balanced by lowered entropy via unfolding. Thus, understanding DNA structural responses in this entropic regime is essential to elucidating overall DNA function.

The advancement of single-molecule force spectroscopy (SMFS) has connected physiological function to molecular-level processes, for example, with DNA (1,2). A typical SMFS experiment provides force measurements, which are related to models that idealize the molecule mechanically (3). Advancements in instrumentation have pushed length scale resolution limits leading to dynamical studies of active processes (4). However, these experiments are either limited to measuring a few averaged parameters or require advanced particle tracking with fluorescence probes (5). The use of fluorescence imaging is generally applicable to strong topological changes (6) occurring in high-force regimes. More importantly, light interacting directly with chemical bonds would provide a more ideal measure of the DNA structure.

Raman spectroscopy is well suited to provide important insights to SMFS, because it outputs the highest level of chemical structure information with minimal external interference due to its fundamental scattering process at all optical wavelengths. Raman studies of DNA have been present for two decades and have produced a database of Raman peaks that characterize the various components of the DNA structure (7,8). Raman signal can be amplified by metal nanostructures through plasmonic effects (9), which make possible single DNA molecule detection (10–12). The drawback is that the DNA or its constituent samples are measured as an ensemble or anchored to hard surfaces; far from an ideal physiological state. We have previously demonstrated a methodology to overcome some of these issues by optically trapping single DNA molecules (13) with silver nanoparticles nonspecifically bound to the phosphate backbone, which made possible the Raman detection of single molecule DNA in its natural aqueous environment.

In this work, we demonstrate, to our knowledge, a novel combination of optical tweezers and surface-enhanced Raman scattering (SERS) to study DNA structural responses from an applied load in the low force, entropic regime. A single DNA molecule is optically stretched close to an external SERS substrate while its Raman spectrum is simultaneously measured. We identify a correlation between the phosphate backbone structure and molecular extension. The results of this novel, to our knowledge, experimental technique are confirmed with state of the art theoretical modeling, which combines molecular dynamics (MD) with mixed quantum mechanics/molecular mechanics (QM/MM), to compute Raman modes for DNA structures modeled at different extensions. The overall result demonstrates a structural response in a regime where mechanical load is thought to only be countered by entropy changes. We also highlight an innovative methodology for directly observing single-molecule chemical structure changes in response to controllable forces.

Methods

Dual optical trap and Raman spectroscopy

The experimental setup (Fig. 1) is a platform that combines optical trapping with Raman spectroscopy. A specimen held by a dual trap is optically excited by a third beam and its Raman spectrum is recorded in the backscattered direction. The dual optical trap was constructed with a 985 nm single-mode fiber laser beam (Avanex, Route de Nozay, France) passed through an interferometer (100 mW power for each beam) with one mirror motor driven (Newport, Irvine, CA, motorized optical mount, 8807) and controlled by a computer. This mirror was optically conjugated to the back focal plane of the bottom objective (Nikon, Japan, CFI PL FL, 100×, NA 1.30). An additional 635 nm low-noise optical beam (0.5 mW) (Coherent, Santa Clara, CA, ultralow-noise diode laser LabLaser635) was passed to the sample, coaxial to the nonmotor-driven trapping beam for position detection of the trapped microsphere. The forward scattered light of the detection beam was collected by a top objective (Edmund Optics, Barrington, NJ, 40×, NA 0.65) and analyzed by a quadrant photodiode (Newport, model 2921). The Brownian motion of the bead was measured to calibrate the trap using established procedures (14) and to give the bead position for the force-extension curves. The resulting signals were then transferred through an analog-to-digital conversion card (National Instruments) and recorded with custom software in the Labview environment. A 532 nm beam was used for the Raman excitation (OZ Optics, Ottawa, Canda, OZ-2000, 20 mW) with a power density of $1.5\times {10}^6$ W/cm² at the focus. The dependence of Raman spectra on excitation polarization direction was not tested. The backscattered light was collected by the bottom objective and passed through a holographic notch filter (Semrock, Lake Forest, IL, 532 nm RazorEdge Dichroic laser-flat beamsplitter) before entering the confocal system with a 150 μm pinhole (Thorlabs, Newton, NJ, P150S). Raman spectra were recorded with a spectrometer (Andor, Belfast, UK, SR163, 1200 lines/mm) equipped with a charge-coupled device camera (Andor, DV401A-BV) at a spectral resolution of 3 cm⁻¹. For the measurement, the DNA-bead construct was aligned such that the Raman excitation beam passed between the beads.

Figure 1.

Schematic of the combined dual optical trap and Raman spectroscopy setup where a single DNA molecule is stretched by moving one of the attached beads with the optical trap while a laser excites the SERS substrate and DNA molecule (top left inset). All lasers used are coupled through single mode fibers before entering the system. The 985 nm trapping beams are split by an interferometer made with two prism beam splitters. Both beams are passed by a dichroic mirror to the bottom objective (O1). A 635 nm beam also passes through O1 and is used for position detection of the trapped microsphere. The forward scattered light of the 635 nm beam is collected by a top objective (O2) and analyzed by a quadrant photodiode with a band-pass filter (Thorlabs, FL635-10) that blocks the trapping and Raman excitation beams. The 532 nm Raman excitation beam is reflected by a 45° Notch filter and propagates colinear with the trapping and detection beams through O1. The backscattered light is collected by O1 and passed through the Notch filter to a confocal spectroscopy system that consists of two lenses and a pinhole in front of a spectrophotometer with a charge-coupled device camera. The bottom right inset is an electron microscope image of the SERS bead (1 μm scale bar) and a bright field live image of the optically trapped beads with a DNA molecule (not visible in image) approaching the SERS bead for the experiment.

None of the auxiliary beams (i.e., the trapping and position detection beams) added background signals to the Raman spectra. These beams are shifted by half the molecule length (2 μm) relative to the Raman excitation focus (see Fig. 1) thus propagating outside of the confocal volume of the spectroscopic system. In addition, the power of the detection beam is significantly weaker than the Raman excitation, whereas the trapping beam wavelength is outside the spectral window of the spectrometer that is tuned to wavelengths close to 532 nm. No Raman peaks were observed from two trapped beads without DNA attached between them.

DNA Raman spectrum versus extension measurement

The experiment (see Fig. 1 (top left inset)) consisted of DNA-bead constructs, double-stranded λ-phage 12 kbp DNA (4.25 μm contour length) molecules anchored to silica microspheres via established methods (15), manipulated with the dual optical trap in phosphate buffered saline (pH 7.4) within a custom-built fluid chamber. The Raman signal of the molecule was enhanced by bringing it close to a SERS substrate, 5 μm silica beads with silver nanoparticles (70 nm average diameter) attached to their surface (16), which was previously deposited onto the glass coverslip surface of the fluid chamber. Briefly, the probes were prepared by attaching citrate-reduced silver colloidal nanoparticles (17) to the surface of the silica beads. The particles were anchored to the surface via a self-assembled alkyloxy silane monolayer. Adjusting parameters, such as silane concentration and the agitation speed of the reaction vial that contained the beads and chemicals, produced a stable silane monolayer with silver nanoparticles attached at <20% packing density on the bead surface, confirmed by scanning electron microscopy imaging (Fig. 1 (bottom right inset)).

For the measurements, a full force versus extension curve was first measured to confirm the presence of a single DNA molecule by fitting to a worm-like chain (WLC) model and confirming the single molecule persistence length (13). The molecule was then extended to ∼85% of its contour length before approached in the z direction toward the upper surface of the SERS probe by lowering the bottom objective. This was the initial extension for the measurements and was essential to bring the molecule close to the SERS probe without having it contact the trapped beads. The Raman signal was continuously detected at 1 s acquisition times in live mode and the DNA z displacement was stopped upon appearance of Raman peaks and adjusted in the x-y directions by slightly shifting the movable trapped bead to maximize peak intensities. The DNA molecule was then extended through two additional 100 nm increments with a single Raman spectrum at 1 s acquisition time recorded at each step. The position of the bead in the fixed trap was continuously recorded to monitor the applied force. The distance between the molecule and SERS probe could not be directly measured or held perfectly constant due to the inherent Brownian motion of the molecule. However, this motion was minimized by the taut state of the molecule throughout the experiment.

MD and QM/MM modeling

MD modeling was performed with the density functional theory (DFT) normal mode frequencies to confirm the experimentally observed Raman modes and their sensitivity to molecular extension. The modeled DNA three-dimensional structures, used in previous studies (18), were 30 basepairs long with only cytosine-guanine basepairs, which minimized the number of theoretical modes that are not measurable in the experiment. The initial structure of these simulations was 97 Å in end-to-end length. Further extensions were made, taken from the pinned simulations, resulting in 120, 135, and 155 Å lengths, corresponding to scaled forces of 69, 99, and 106 pN, respectively (18). Structures were prepared by removing most waters but leaving a covering 7 Å deep from all DNA surface atoms and K⁺ ions in the solution. All QM/MM calculations were prepared with Maestro version 9.2 (Schrödinger, LLC, New York, NY) and carried out with Qsite using the MO6 DFT functional (due to a good reproduction of stacking interactions) (20) and the OPLS 2005 force field (21), which is standard for Maestro and Qsite. Due to the large amount of calculations and our interest in frequency shifts only, the basis set was limited to 6-31G(d,p). All MM degrees of freedom were frozen throughout to maintain the structure at a room temperature conformation using the original MD force field (AMBER-BSC0 (22)). A five-step optimization was carried out on every system to release bad contacts within the QM region. The vibrational analysis, conducted within Qsite, calculated frequencies and vectors from a Hessian of second derivatives of the total energy. Total computation time was ∼7 days on 15 Intel itanium processors (for each normal mode calculation) on an Altix shared memory computer. The theoretical structure extensions were higher than in experiment due to the long computational time necessary to resolve smaller extensions (forces). For each modeled structure, normal mode frequencies were calculated for three adjacent segments on each chain to observe any variations along the strand. Each segment contained two basepairs with hydrogen caps inserted directly above and below surrounding sugars for a total of 134 atoms (Fig. 2).

Figure 2.

Image of DNA molecular model, highlighting the individual segments that were included in the MD normal mode frequency computation. Each segment contained two basepairs with an adjoining phosphodiester chain and three pairs of segments (6 total) between the two chains and were calculated to observe any variations along the strand.

For the initial structure, the Raman active modes in this list of normal mode frequencies were determined by computing Raman activities in a reduced structure containing only two bases with one joining phosphate. Obtaining Raman activities for the full modeled structures was too computationally intensive for this study. Gaussian03 (23) was used to calculate Raman activity of vibrational modes using the M06 DFT functional. The result was a list of frequencies with corresponding Raman activity values. Raman active frequencies in the region of experimental interest were tracked in the full QM/MM systems to see which modes changed with DNA extension. All computed Raman frequencies were lower than in the experimental measurements. For example, the O-P-O stretch mode with the highest Raman activity was at 794 cm⁻¹, whereas this mode always appeared above 800 cm⁻¹ in the experiment.

Results and Discussion

Raman signatures of a single DNA molecule

Fig. 3 a is a representative series of Raman spectra from a single DNA molecule at three different extensions where a difference in some peak positions between spectra is immediately clear. Each trace was background subtracted using multiplicative scatter correction and smoothed with a spline technique (24). The plotted region contains stretch vibrational modes of the phosphodiester network (C-O-P-O-C) and C-N bonds in the cytosine base (8), with bands at lower and higher frequencies containing bending and scissor modes, respectively, that are typically inconsistent with SERS. The peak positions measured here are similar to what was observed in our previous work (13) where silver nanoparticles were attached to the DNA molecule, demonstrating consistency between two different silver nanoparticle-based SERS systems. The band assignments used here are based on literature that used bond symmetry arguments to determine normal mode frequency regions for the DNA components (8); a direct derivation from the bond orientations of the molecule that is independent of any electromagnetic field enhancement. Moving the DNA molecule away from the SERS bead and Raman excitation focus led to disappearance of the signal, thus eliminating band assignment to spurious peaks that could be due to the surface chemistry of the SERS bead. As seen in Fig. 3 a, the Raman peak position shifts only occur for the phosphodiester stretch mode, and only in an increasing direction with extension, whereas all other peaks stay in the same position or experience slight random movement. Plotting the Raman peak position versus extension for the phosphodiester and cytosine base mode for this set of traces further exemplifies the difference in sensitivity to molecular stretching between the two bands (Fig. 3 b, inset). The persistence length (53 nm) from the WLC fit to the force extension curve in Fig. 3 b confirms the presence of a single DNA molecule and indicates the applied force and extension range where this Raman band sensitivity is observed.

Figure 3.

A representative set of Raman spectra (a) from a single DNA molecule at three different extensions: 1), 3700 nm, 2), 3800 nm, and 3), 3900 nm. The O-P-O Raman band in the 800–900 cm⁻¹ range undergoes an upward shift in position, whereas the C-N vibration (1280 cm⁻¹) remains constant as the molecule is extended. A single molecule between the trapped beads is confirmed by measuring a force versus extension curve (b) and fitting experimental data (solid squares) to a WLC model (3900 nm) (solid line) to validate the single DNA persistence length (53 nm). The peak position versus extension for the two highlighted Raman bands from the three traces in (a) are plotted with their respective location in the WLC curve indicated (inset (b)).

Raman spectra from measurements of five DNA molecules at all extensions are grouped together and cross-correlated via a two-dimensional (2D) correlation analysis (25) to identify peak movement in the sample set (Fig. 4). 2D correlation analysis is an objective method for identifying trends in a set of spectra that respond to a single effect, in this case the extension of the molecule. The outputs are in-phase (synchronous) and out-of-phase (asynchronous) maps where the former represents Raman peak position and intensity variations and the latter the presence of a sequential trend of these variations relative to the molecule extension. Peak position trends with molecule extension will emerge as hotspots in both maps. Features between 800 and 900 cm⁻¹ band, where stretch modes of the O-P-O unit lie, are present in both maps, signifying a displacement of this band with extension. The doublet spot is due to a shift greater than the width of the peaks. The C-N stretch mode of the cytosine ring, 1280 cm⁻¹, only has an in-phase feature due to an intensity change with no position shift in the peak. Generally, intensity changes of SERS spectra are difficult to interpret because the enhancement level can vary due to differences in the relative positions of the SERS substrate, molecule, and incoming excitation light (26). Debris in the solution that could come from the SERS beads or DNA sample could also float into the excitation path causing brief flashes of an intense peak during the acquisition. For these reasons, we chose to ignore a few occurrences of a large peak at 1200 cm⁻¹ that caused lines of spots in the maps at this position, because it does not match a particular DNA mode, its intensity was orders of magnitude greater than the next most intense peak, and it appeared in only a few of the spectra.

Figure 4.

2D correlation maps (synchronous (a) and asynchronous (b)) from Raman spectra of five different molecules that were extended. The signatures in both maps (boxes) together indicate a shift in peak position of the O-P-O Raman band (800–900 cm⁻¹), whereas the C-N cytosine base vibration (1280 cm⁻¹) remains constant during the molecule extensions. The large spots along the 1200 cm⁻¹ axes are due to a large random peak that appeared at this position in two of the spectra, which is common with SERS.

The unique feature of this SERS arrangement is the ability to maintain the molecule free of attachments allowing for a more true measure of the structure. Based on geometrical optics, the Raman excitation focus diameter is ∼400 nm, however, the SERS effect occurs only within the decay length of the evanescent field from the silver nanoparticles, which has been reported to be in the range of 50–100 nm (27), equivalent to ∼100 bp. Although it is difficult to estimate the exact distance between molecule and SERS substrate, the DNA molecule must be within this decay length to observe Raman signal, which is a product of multiple basepairs segments simultaneously optically excited. Maintaining the molecule in extended states minimized Brownian motion fluctuations that could cause the DNA to come in contact with the SERS bead, which would alter enhancement levels and potentially hinder extension-based effects. As a result, Raman signals remained stable during periods of minutes without observing degradation of the molecule or SERS bead, and spectra from the set of DNA molecules was consistent.

One differing characteristic in the Raman spectra was the adenine peak, previously observed to be dominant in single-stranded DNA (28) and from single adenine molecules in silver colloid solutions (11), but detected at smaller intensities (730 and 1325 cm⁻¹) here. One reason for this discrepancy could be the less understood interaction between silver and the double strand, which may cause shielding of the other bases from the silver nanoparticle plasmonic fields. Indeed, the adenine peak is the most responsive to metal nanoparticles adsorbed to uncoiled segments of DNA (29). The molecule is not in contact with the metal surface as it is in other studies, thus, a proper measure of the SERS cross section or intensity of each component is not possible in this configuration, where the 730 cm⁻¹ peak of adenine is traditionally dominant. Adenine signatures, and other bases, are most likely present in the Raman spectra, but the same intensity ratios previously observed with molecules adsorbed to metal surfaces cannot be repeated. The lack of adsorption produces an averaging of the base peaks instead, where the C-N stretch modes of the bases (1100–1700 cm⁻¹) seem to be stronger than the base ring breathing modes (<800 cm⁻¹). With adenine having the highest SERS cross section, this peak intensity stands to experience the most dramatic reduction in the absence of adsorption to the metal, as observed here.

Bond orientations alter with molecule extension

Although intensities are solely dependent on the interaction of the chemical bond with light, band positions also depend on bond symmetry and polarizability, which is a direct consequence of the conformation of the bond. Thus, Raman band position changes are expected to correlate with mechanical alteration of the local structure (30). We conclude that the 800 cm⁻¹ phosphodiester stretch mode is sensitive to molecule extension, whereas the cytosine mode (1280 cm⁻¹) is unaffected. A similar mechanical sensitivity of the O-P-O stretch mode was previously observed, where an upward shift of the band is present between looser wound A-form DNA and tighter wound B-form structures (8).

Fig. 5 a is a plot of Raman peak position versus extension for all measurements. The plots contain the phosphodiester and cytosine base stretch modes that had distinct spots in the correlation plots with points from all of the tested molecules plotted together, because of the known high mechanical uniformity between DNA molecules. Thus, the majority of error will lie in the Raman peak positions. A small spread in the extension points is due to uncertainty in the starting extension, because each molecule must be stretched to a finite initial length to be approached to the micron-sized SERS substrate. The effect is immediately clear: the phosphodiester mode shifts whereas the cytosine mode remains unaffected during the DNA extension. Assuming a linear relationship, the most striking parameter is the sensitivity of the phosphodiester Raman band to the force extension, 0.11 cm⁻¹/nm.

Figure 5.

Experimental (a) and theoretical (b) plots of Raman peak position versus DNA extension for the O-P-O phosphate backbone and C-N cytosine base stretch modes. The points from all five molecules and six calculated regions are plotted together for the experimental and theoretical plots, respectively. The data sets are fit to a straight line and the slopes are included for the experimental data. The representative experimental Raman trace for the initial extension from Fig. 3a is plotted in full range (c) and shares the same x axis with the theoretical plot of the initial modeled structure (d) where frequencies and intensities are taken from the full QM/MM and reduced structure calculation, respectively. All plots show agreement between experiment and theory.

The MD modeling confirmed the measured Raman modes and allowed visualization of the modes’ atomic motions. The Raman active modes were determined by computing the frequencies and Raman activities of a reduced structure, which consisted of a single phosphodiester network between a guanine and cytosine base. The phosphate vibration with the highest Raman activity in the vicinity of 800 cm⁻¹ prominently involves a single O-P vibration but includes motion throughout the phosphodiester chain (C-O-P-O-C), whereas the 1280 cm⁻¹ cytosine peak comes from a C-N vibration in the cytosine ring. Fig. 6 a depicts the identified motion of the phosphate mode. Identifying the motion in the reduced system allowed us to locate the corresponding band in the full DNA structure computation and track it through the different extensions, because the frequencies in the two systems did not match up perfectly. The computed modes for each region (Fig. 2) of the full modeled DNA structure are plotted versus extension in Fig. 5 b. Once again, the phosphate stretch mode shows sensitivity to DNA extension while the cytosine base remains unaffected. Further agreement between experiment and theory is evident when examining the singular traces. Fig. 5 c is the representative trace for the initial extension from Fig. 3 a plotted in full range and compared to the calculated modes of the initial structure in Fig. 5 d where the peak positions are taken from the full QM/MM calculation and Raman intensities from the reduced structure. The measured Raman bands can be confidently assigned to DNA components based on the consistency between experiment and theory and agreement with previous studies in the literature.

Figure 6.

Theoretical snapshots (a) of the O-P-O vibrational mode responsible for the 800 cm⁻¹ Raman peak taken from the reduced structure calculation used to determine the Raman activities. An overlay (b) of the calculated DNA structures focused at a single region, with the phosphorus atom (arrow) as the common origin: 97 Å (blue), 120 Å (red), 135 Å (green), 155 Å (purple).

The Raman frequency of a vibration can change due to an alteration of the mechanical and electromagnetic components of the bond. The increase in the phosphate mode frequency is a result of an alteration in geometry and energetics of the system as the molecule is extended. Fig. 6 b is an overlay of the initial and extended modeled structures with a single calculated region of the DNA chain isolated from the full modeled structure and the phosphorus atom as the common origin. The orientation of the O-P-O unit remains constant and the O-P-O bond angle shows no correlation with extension. However, there is a drastic change with the orientations of the carbons and adjacent sugars relative to the O-P-O unit. Nearest neighbors reconfigurations can affect the bond’s polarizability and its interaction with light via Raman scattering by changing the symmetry of the local region. From an energetics standpoint, the region experiences a mechanical perturbation as the structure is extended. The total end effect of the above is a shift in the bond’s Raman peak position.

The DNA structural alterations observed here at small forces should aide in understanding DNA-protein interactions in the entropic regime. Although DNA can be modeled as a semiflexible polymer, the result shows that chemical bond orientations contribute to mechanical loading at all scales.

When probed at longer length scales, such as with single molecule stretching, DNA acts as an entropic spring, following the WLC model with a persistence length that spans multiple basepairs. However, studies that have directly probed the short-length scale bending distributions of DNA have suggested that the probability of these occurrences is much higher than predicted by the WLC model (31). This implies that there are underlying elasticity contributions at length scales that are much shorter than the DNA persistence length. Our experiments and modeling confirms this idea by demonstrating a correlation between extension and bond orientation where the Raman peaks are the markers. We start measuring at 85% of the contour length, due to practical reasons as described in the main text, which avoids the portion of conformational restriction due to tangling.

The technique, to our knowledge, is a novel combination of known methodologies that provides a new addition to the SMFS field. By directly measuring at the bond level with Raman scattering, these structural shifts can be observed at low forces without the need for complex particle tracking. Most importantly, with the constant advancement of SERS substrates, the stability and amplification level will continue to improve while maintaining the DNA strands free of optical probes and leaving them in a more ideal physiological condition.