1.

Introduction

The detection of genetic markers has become ubiquitous in modern scientific research as a tool for examining the genetic relationships that underpin traits involving heredity, biological variability, disease susceptibility, and pharmacological efficacy, to name a few. Because DNA sequence variations such as single-nucleotide polymorphisms (SNPs) have been associated with approximately 90% of polymorphisms in the human genome,¹ researchers often use SNPs as markers for a variety of genetic diseases² and cancers.³

The importance of improving the speed, specificity, and cost efficiency of methodologies used to collect genetic information directly correlates with the importance of the information itself. Although polymorphisms as genetic markers are traditionally probed using time-intensive Southern blot hybridizations and polymerase chain reaction (PCR)–related methodologies,⁴ more recent innovations using DNA functionalized with gold nanoparticles (AuNPs) have made preliminary steps in detecting terminal DNA mismatches.^{5, 6, 7, 8, 9, 10} In 2004, it was noted that single-stranded DNA oligonucleotides (oligos) can be functionalized to gold nanoparticles in separate solutions that each have an inherent pink color.⁵ When the nanoparticle solutions of DNA with mismatched terminal bases are mixed, the gold nanoparticles precipitate, and the solution loses its pink color. Unfortunately, researchers were not able to use this technique with mismatched bases closer to the middle of the sequence, and the test DNA had to be modified to attach the nanoparticles.⁵

Raman spectroscopy is another sensitive detection technique that has the potential of reducing the need for PCR and laborious biochemical techniques. Since the early 1970s, researchers have been using Raman to explore nearly every aspect of biology and biomedicine,^{7, 8, 10, 11} including DNA.¹² While innately weak Raman signals, fluorescence, and large background noise can make biological Raman applications difficult, advances in laser and filter technology and the advent of surface-enhanced Raman spectroscopy (SERS) have improved Raman capabilities. Further, combinations of Raman spectroscopy with aspects of nanotechnology provide even greater utility.¹³ Combining gold nanoparticle–based SERS with DNA opens several avenues of pure DNA analysis and subsequent gene detection methodologies.^{14, 15, 16, 17} Sun have developed a way to detect breast cancer genes using the hybridization of sample DNA with probing DNA that has been modified with gold nanoparticles and specific Raman labels. Using this technique, Sun were able to achieve detection sensitivities of down to $1\mathrm{fM}$ (Ref. 13).

Previous results from our lab used SERS to detect, with high levels of statistical confidence, differences between samples of complementary oligos and those formed from oligos with a midsequence base mismatch.¹⁸ However, a major drawback to the practical and clinical applicability of this technology is the need to add thiol modifications to each different oligo intended for analysis. Not only is it expensive to add thiol modifications to numerous test strands, the variability associated with the surface chemistry involved in functionalizing a large variety of oligonucleotides with gold nanoparticles gives rise to aggregation, precipitation, poor adsorption, and other deleterious interactions with the gold nanoparticles. We have found that using a single modified detection oligo to test for multiple mutations reduces sample-to-sample variability, decreases the amount of preparation time, and decreases the need for costly reagents and modified DNA material.

In this paper, we refine and expand previous techniques to further investigate the technology’s potential for genetic marker detection and to make the technology more applicable to biological studies. Namely, we have eliminated the need to directly functionalize the test oligo with gold nanoparticles, and we have expanded the capabilities of this technology to detect and explore two different types of polymorphic regions (PRs) in addition to lone SNP detection. The enhanced ability to test for multiple mutations along a native, unmodified gene broadens the applicability of this technique to practical genetic marker detection methodologies.

2.

Materials and Methods

The overall idea is to capture single-stranded “test” oligonucleotides between closely spaced gold features. As depicted in Fig. 1 , a DNA test strand can be probed for abnormal base mutations by hybridizing one highly conserved region of the strand to a gold surface-bound “capture” oligo and another region to a gold nanoparticle-bound “detection” oligo, thus creating a SERS environment around the test oligo. This was accomplished in a three-step aqueous hybridization procedure using preparations similar to those previously developed^{13, 18} with several variations. Test samples were composed of the four types of DNA duplexes visualized in Fig. 1 and defined in Table 1 —those formed from complementary oligos, those formed from oligos with a single midsequence base mismatch (SNP), those formed from oligos that have three delocalized mismatches (DTM), and those that have a polymorphic region of three adjacent mismatches (ATM). Resulting Raman spectra were compared for statistically significant intensity and peak position differences at four characteristic peaks as described later. The overall method involves nine main steps, as follows:

Fig. 1

Visualization of the hybridization scheme used to create an SERS environment around DNA duplexes. Single-stranded oligos are hybridized as either completely matched, with a single mismatch (SNP), with an adjacent triple mismatch (ATM), or with a delocalized triple mismatch (DTM). Gray boxes highlight the mismatches for clarity.

Fig. 4

Raman spectra highlighting the peak position shifts that occur when comparing the average matched sample to the SNP and ATM samples. The DTM sample followed in trend with the other mismatched samples and was therefore excluded for clarity. The highlighted peaks represent (a) down-shifting guanine ring breathing and (b) phosphodiester stretch modes and (c) an up-shifting guanine/sugar stretch. Peak positions are shown with their respective standard errors, representing the variability between the seven median spectra averaged together to obtain these positions. Levels of significant difference are compared to the matched sample. Constituent peak intensities were baseline subtracted and normalized to the peak of interest before averaging and then smoothed. (See Table 3 for $p$ -values.)

Table 1

Oligonucleotide design for matched and mismatched sequences.

3.

Results

Twenty-four Raman spectra from each test sample were collected and processed as previously reported.¹⁸ To eliminate high and low outliers, subsets of 7 median spectra were selected from each original group of 24. These median subsets were averaged to determine overall intensity for plotting, and the intensities and peak positions of the four characteristic nucleic acid peaks were extracted for statistical analysis. Statistical significance was determined (from $p<0.01$ to $p<0.001$ , $\alpha =0.05$ ) for differences between each of the samples at the four selected peaks. The characteristic peaks primarily included contributions from the two purine-containing nucleic acids, guanine and adenine, due to the increased cross-sectional areas and therefore increased SERS activity of these molecules.²¹ The four selected peaks were a guanine ring breathing at $660{\mathrm{cm}}^{-1}$ (Refs. 22, 23), the phosphodiester backbone peak around $840{\mathrm{cm}}^{-1}$ (Ref. 24), a guanine stretch at $970{\mathrm{cm}}^{-1}$ (Refs. 22, 23, 24), and guanine/denine vibrations at $1174{\mathrm{cm}}^{-1}$ (Refs. 22, 23). Figure 2 more succinctly shows the peak positions and standard error of these key peaks. Note that among all these peaks, the overall Raman intensity of the complementary sample is significantly greater than that of the single-mismatched sample, which is itself greater than the triple-mismatched samples. Significance levels are individually summarized in Table 2 .

Fig. 2

Raman spectra comparing the averaged median intensities of single-stranded oligos (matched, single mismatched, delocalized triple mismatched, adjacent triple mismatched, and completely mismatched) hybridized to form double-stranded duplexes. Standard error bars are included for the key peaks. (See Table 2 for $p$ -values.)

Table 2

Two-tail p -values from t -test results used to test the significance of intensity differences among selected peaks. Asterisks represent corresponding significance levels of the sample comparisons as indicated.

Scanning electron microscopy (SEM) images seen in Fig. 3 qualitatively support the theory that more nucleotides are being hybridized in complementary samples than in single-mismatched and multiple-mismatched samples. The figure shows greater surface area coverage of nanoparticles on the gold spot with complementary oligos and a reduced amount on the spots with mismatched oligos. Additionally, image analysis software (ImageJ) concludes that the percent coverage of nanoparticles in the representative areas is 25.0% for the matched sample, 17.8% for the SNP sample, and 14.1% for the triple-mismatched sample. The relatively small differences in nanoparticle coverage between samples are to be expected given that the binding affinities of mismatched samples are altered only by one to three base mutations as compared to the matched samples. Note that while these images show qualitative support for the Raman data, sample-to-sample variability in nanoparticle surface coverage limits subsequent significance analyses.

Fig. 3

Scanning electron micrographs of representative sample capture surfaces after hybridizing single-stranded oligos and gold nanoparticles to form double-stranded SERS complexes (top row). The bottom row contains corresponding particle count analyses that clarify the different particle densities of each micrograph. These images enable a qualitative analysis of the nanoparticle densities that results from hybridizing with (a) completely matched, (b) SNP, (c) triple-mismatched, and (d) completely mismatched test strands.

Figure 4 represents a magnification of the $500\text{to}1300{\mathrm{cm}}^{-1}$ area in Fig. 2. The three aforementioned peaks, highlighted and magnified further in Figs. 4, 4, 4, show statistically significant peak shifts when comparing the average matched sample to the SNP and ATM samples. The DTM sample followed in trend with the other mismatched samples and was therefore excluded for clarity. Constituent peak heights were normalized to the peak of interest before averaging. It should be noted that while the average position among mismatched samples appears different in Fig. 4, this difference is not statistically significant and therefore will not be treated as meaningful. Only position shifts between the matched samples and the mismatched samples are statistically significant. Peak positions are shown with their respective standard errors and levels of significant difference as compared to the matched sample. Table 3 summarizes the significance level of each shift. Studies verify that these peak shifts are not affected by the surface coverage of nanoparticles.

Table 3

Two-tail p -values from t -test results used to test the significance of peak position differences among selected peaks. Asterisks represent corresponding significance levels of the ATM and SNP samples as compared to the matched sample as indicated.

4.

Discussion

The effort of this paper is to extend the use of Raman spectroscopy to detect genetic markers on unmodified oligonucleotides and to determine the applicability of using this type of approach for exploring DNA hybridization characteristics associated with SNPs and PRs. There are two mechanisms behind the feasibility of these experiments. First, because genetic markers such as SNPs and PRs contain energetically unfavorable nucleotide mismatches in the DNA helix, mismatched oligos are less likely to hybridize as effectively as matched oligos and are more successfully removed with stringent washes. Because the creation of an SERS environment is dependent on the hybridization of oligos, mismatched samples have a decreased overall Raman signal that can be used to differentiate them from matched samples in a direct comparison analysis. Increasing the number of mismatches only amplifies this effect and further decreases the Raman signal. Second, it is proposed that hybridized samples with mismatches are subject to altered intermolecular bonding patterns that affect characteristic vibration modes. Therefore, not only does the overall Raman intensity differ between matched and mismatched samples, the Raman signatures also give insight into the mismatch.

In accordance with the theoretical basis for this experiment and as shown in Figs. 2 and 3, it is to be expected that the SNP samples have a weaker Raman intensity than matched oligos and that multiple mismatches exhibit a weaker signal still. Based on free-energy calculations of single-stranded oligo hybridizations as described and tabulated by Turner, ²⁵ it is predicted that among the multiple-mismatched samples, the DTM will show a higher binding affinity than the ATM. These calculations assume that base-pair mismatches form energetically unfavorable internal loops within the helix and infer that when predicting hybridization stability, both the size of the mismatch and the sequence position of the mismatch are important. In this case, not only would the DTM incur a lower cumulative energy penalty than the ATM but also the mismatches in the DTM are positioned such that they sacrifice fewer energetically favorable matches. Such calculations conclude that the energy difference between the two triple-mismatched samples is only $1.4\mathrm{kcal}$ in favor of DTM hybridization. Although Raman experiments were able to distinguish this difference significantly, as seen in Fig. 2, experimental conditions such as hybridization temperature, reagent concentration, DNA age, and quality of the photolithography must be tightly controlled for successful differentiation. It is proposed that such a small energy difference is very near the lower level of detection for this technique and that while these calculations give general theoretical reinforcement to the observed experimental trends, it should be noted that due to the nature of SERS enhancements, there is little quantifiable correlation between binding energies and Raman intensity. However, between the match, SNP, and triple mismatches in general, experimental Raman results show robust, consistent, and reproducible intensity trends. For comparison, free-energy calculations comparing DTM and SNP show a $5.5\text{-}\mathrm{kcal}$ difference, and those comparing SNP and the complete match hybridization show a $4.8\text{-}\mathrm{kcal}$ difference.

To further improve understanding of the system, observations were also made regarding peak positions in the Raman spectra seen in Figs. 2 and 4. The principal observation is a slight shift in the position of three of the characteristic peaks mentioned earlier—the guanine ring breathing peak at $660{\mathrm{cm}}^{-1}$ (Refs. 22, 23), the phosphodiester backbone peak around $840{\mathrm{cm}}^{-1}$ (Ref. 24), and the guanine stretch at $970{\mathrm{cm}}^{-1}$ (Refs. 22, 23, 24). Because Raman spectroscopy is a fingerprint technique, each peak represents a specific molecular vibration that contains information about its local chemical environment. Therefore, additional information can be gained by observing shifts in peak position as they relate to molecular conformations, structural rearrangements, and other changes in the chemical environment.

Significant differences here can be used to gain insight into chemical changes that occur upon hybridization of mismatched oligos. Figure 4 shows that the phosphodiester peak at $842{\mathrm{cm}}^{-1}$ in the matched sample is conspicuously shifted to a lower wavelength (down-shifted) of around $831{\mathrm{cm}}^{-1}$ for the mismatched samples. It has been observed in previous studies that the phosphodiester band shifts from $835{\mathrm{cm}}^{-1}\text{to}815{\mathrm{cm}}^{-1}$ to near $800{\mathrm{cm}}^{-1}$ as the order and rigidity of the system decreases from B-DNA to A-DNA to single-stranded oligonucleotides, respectively.²⁴ In other words, as the order and rigidity of the bond decreases, it takes less energy from an incident photon to induce the same vibrational mode. Empirically, this can be represented with the following equation:²⁶

Interestingly, Fig. 4 shows that the $967{\mathrm{cm}}^{-1}$ peak associated with the bond stretch between guanine and its sugar is seen to shift to higher wavelengths in the SNP and ATM samples as compared to the matched sample. We speculate that breaking the hydrogen bonding pattern in the bases induces a more widespread redistribution in electron arrangement over the entire nucleotide that relaxes structural bonds near the backbone and strengthens those nearer to the base. In this case, the $967\text{-}{\mathrm{cm}}^{-1}$ peak shifts to approximately $974{\mathrm{cm}}^{-1}$ in the two mismatched samples. In contrast, a vibrational mode not necessarily associated with nucleotide structure, guanine’s imidazole ring-breathing mode related to the peak at $667{\mathrm{cm}}^{-1}$ , down-shifts to $\sim 654{\mathrm{cm}}^{-1}$ in the mismatched samples, again indicating a decreased sense of order within the ring. This final shift can be seen in Fig. 4. Again, there were no meaningful position shifts among the mismatched samples, only between the mismatched samples and the matched sample. Such observations attest to the varied amount of information that can be obtained from these types of Raman studies, extending the technique both as a robust polymorphism detection scheme for genotyping analyses and as a useful tool for analyzing additional minute and fundamental chemical relationships regarding DNA hybridization.

In summary, this work reports a preliminary method capable of detecting and exploring genetic markers composed of midsequence SNPs and multiple base mismatches in unmodified DNA sequences. The SERS approach provides both a robust detection rubric for further analyses using polymorphisms and a useful research tool potentially capable of exploring precise biochemical changes associated with DNA mutations. Future approaches will be developed to enhance the speed and efficiency of this process. Experiments will be attempted to define minimum hybridization times and to create automated microarrays or microfluidic formats that further improve efficiency by using continuous or large batch processing designs.