2.1. Synthesis of DNA-AuNP Crystals

DNA-AuNP crystals were synthesized adapting procedures from previous reports ¹⁴. First, 20 nm AuNPs (Nanopartz) were functionalized with thiolated DNA strands (Nihon Gene Research Laboratories; DNA #1: 5\('\)-TCAACTATTCCTACCTACAAAAAAAA-SH-3\('\); DNA #2: 5\('\)-TCCACTCATACTCAGCAAAAAAAAAA-SH-3\('\)) in borate buffer by gradually adding 5 M NaCl, followed by centrifugation to remove excess unmodified DNA. For crystallization, this DNA-AuNP mixture and a linker DNA solution (Linker #3: 5\('\)-GTAGGTAGGAATAGTTGAATTCCTT-3\('\); Linker #4: 5\('\)-TTGCTGAGTATGAGTGGAAAAGGAA-3\('\)) containing 1.0 M NaCl were mixed at a \(1:1\) flow rate ratio in a microfluidic device. Droplets were formed using an oil phase composed of 93 wt% isopropyl palmitate and 7 wt% ABIL EM180, yielding a final NaCl concentration of 0.5 M. Approximately 40 μL of the droplet solution was collected and slowly cooled from 65°C to 25°C at a rate of 0.1°C/10 min ¹⁵.

To generate SERS hotspots, retrieved crystals were incubated with 0.75 mM AgNO\(_3\), L-sodium ascorbate, and polyvinylpyrrolidone (all from Sigma-Aldrich) in \(1/15\) M phosphate buffer containing 0.5 M NaCl at 40°C for 12 hours while mixing at 2000 rpm. Finally, the crystals were washed with 0.5 M NaClO\(_4\), contracted by adding 0.5 M AgNO\(_3\) without a reducing agent at room temperature, and subsequently washed with Milli-Q water ¹⁶.

2.2. Raman Microscopy

To evaluate the SERS performance of the prepared crystals, hyperspectral mapping was performed using a home-built confocal Raman microscope (Fig. 1). The excitation source was a 532 nm continuous-wave laser (Sapphire SF NX 532-150 CDRH, Coherent, USA). The laser beam was directed through a 2-axis scanning mirror (6210 Galvano Scanner, Cambridge Technology, USA) for spatial mapping and focused onto the sample using a 20\(\times\) objective lens (NA 0.95; CFI Plan Apochromat LWWD Lambda S 20XC WI, Nikon, Japan). For sample observation, the system incorporated a high-power LED (SOLIS-1D, Thorlabs, USA) and a CMOS camera (CS165MU1/M-Zelux, Thorlabs, USA), coupled via a 490 nm short-pass dichroic mirror (DMSP490R, Thorlabs, USA).

Raman signals were collected in a backscattering geometry through the same objective and scanning mirror. A 532 nm Bragg-notch filter (SureBlock, Coherent, USA) was used to reject the excitation light. The scattered light was passed through a slit, dispersed by a home-built spectrometer equipped with a 600 lines/mm reflective diffraction grating (GR25-0605, Thorlabs, USA), and detected with a CCD camera (DU401A-BVF, Andor Technology, UK). The spectral dispersion was 6.1 cm\(^{-1}\) per pixel. The excitation power at the sample plane was set to 1.0 mW, and the exposure time was fixed at 100 ms per spectrum. Wavenumber calibration was performed using the Raman spectrum of polystyrene beads.

2.3. Sample Preparation for SERS Measurement

To evaluate the SERS performance of the synthesized substrates, Rhodamine 6G (R6G; Tokyo Chemical Industry, Japan) was employed as a probe molecule. R6G was selected as a benchmark molecule for SERS evaluation due to its large resonance Raman cross-section and well-characterized spectral features. R6G was dissolved in deionized water and subsequently mixed with the DNA-AuNP crystal suspension to achieve a final concentration of 1.0 μM. A small volume of this mixture was mounted between a coverslip and a microscope slide. The edges of the coverslip were sealed with nail polish to prevent solvent evaporation and to maintain constant sample conditions during the prolonged mapping process. The prepared slide was then placed on the stage of the home-built confocal Raman microscope (Fig. 1).

2.4. Dataset and Analysis

A total of five Raman hyperspectral images were acquired, containing 190 individual DNA-AuNP particles. These particles were sampled with a bias toward the target diameter of 1–2 μm to focus the analysis on the intended morphology. Mapping was performed with a step size of 200 nm and an estimated laser spot size of 286 nm. The total acquisition time for these five images was approximately 3 hours, averaging approximately 1 min per particle.

All spectral data were processed using a Python-based program. Fluorescence backgrounds were removed from all raw spectra using a rolling-ball algorithm. For intra-particle analysis of hyperspectral data, noise reduction was performed using singular value decomposition prior to the application of a three-component multivariate curve resolution-alternating least squares (MCR-ALS) model, where each pixel was reconstructed by these three components to retrieve pure component spectra and their distribution maps. The data processing workflow was completed within 5 min. For inter-particle analysis, the background-subtracted spectra were directly averaged to evaluate the overall spectral characteristics across multiple particles.

3.1. SERS Spectral Analysis and Peak Assignment

The SERS performance of the DNA-AuNP crystals was evaluated using R6G as a probe molecule (Fig. 2). While the Raman signals in the bulk solution were entirely obscured by intense fluorescence, the DNA-AuNP crystals yielded sharp R6G fingerprint peaks with a high signal-to-noise ratio. The observed peaks at 610 cm\(^{-1}\) and 770 cm\(^{-1}\) correspond to the C–C–C ring in-plane bend and the C–H out-of-plane bend, respectively. Other identified features include C–H in-plane bends at 1126 and 1177 cm\(^{-1}\), and a series of aromatic C–C stretching vibrations at 1364, 1513, 1576, and 1654 cm\(^{-1}\) ¹⁷. The successful acquisition and assignment of these enhanced signals demonstrate that the DNA-AuNP crystals provide surface enhancement. These results validate the appropriateness of the prepared crystals as a functional SERS substrate for the detection of probe molecules.

3.2. Multivariate Analysis of Intra-Particle Signals

Having confirmed the SERS activity of the DNA-AuNP crystals, we next investigated the spatial distribution of these signals within the individual particles to understand the internal hotspot characteristics. The MCR-ALS analysis identified three primary spectral components (Fig. 3(a)). Components 1 and 2 correspond to the fingerprint signatures of adsorbed R6G, featuring characteristic bands at 609, 770, and 1364 cm\(^{-1}\). Component 1 exhibits a higher relative intensity of the aromatic C–C stretching mode at 1364 cm\(^{-1}\) compared to component 2, whereas component 2 shows more prominent bands at 609 cm\(^{-1}\) (the C–C–C ring in-plane bend) and 770 cm\(^{-1}\) (the C–H out-of-plane bend). Notably, these two modes in component 2 correspond to vibrations oriented in the same direction, causing their intensities to fluctuate in synchronization. This simultaneous enhancement serves as a distinctive signature for a horizontal (flat-on) orientation. According to SERS surface selection rules, vibrational modes with a transition dipole moment perpendicular to the metal surface undergo preferential enhancement. These spectral differences suggest distinct adsorption geometries: a vertical (edge-on) or highly tilted orientation for component 1, and a horizontal (flat-on) orientation for component 2. The spatial distribution maps (Fig. 3(b)) reveal that the high-intensity regions of components 1 and 2 are spatially segregated, suggesting that the adsorption orientation is governed by local variations in facet geometry or surface potential.

Component 3 is characterized by a prominent low-frequency vibration at 241 cm\(^{-1}\), which is attributed to the intrinsic metallic lattice vibrations or metal–metal bonds of the DNA-AuNP assembly. Notably, in certain particles such as crystals 2 and 3, the distribution of component 3 spatially overlaps significantly with that of component 1. This spatial correlation suggests that the vertical orientation of R6G is preferentially induced at sites where the metallic lattice signal is dominant. This observation implies that the local environment favoring the reported SERS bands of probe molecules (component 1) is intrinsically linked to the structural characteristics reflected in the low-frequency metallic lattice vibrations (component 3).

3.3. Statistical Analysis of Inter-Particle Signals

To establish a robust framework for the quantitative evaluation of SERS probes, we analyzed the statistical distribution of the enhancement performance across the ensemble of 190 individual DNA-AuNPs. Unlike conventional methods that rely on single-peak intensities, we utilized the scores (coefficients) obtained from MCR-ALS to evaluate the inter-particle variability. This approach allows for a more holistic assessment of signal stability, encompassing both total enhancement and spectral shape consistency.

Figure 4(a) shows the scatter plot of the coefficients for component 1 (C\(_{1}\)) and component 2 (C\(_{2}\)), representing the vertical and horizontal adsorption geometries, respectively. The plot reveals a moderate positive correlation (\(r=0.6791\)), suggesting that while the overall enhancement is governed by the common plasmonic properties of the crystals, the two adsorption modes are subject to distinct stochastic processes. Notably, the coefficient of variation (CV) for C\(_{1}\) (2.264) is significantly higher than that for C\(_{2}\) (0.767). This difference indicates that the vertical orientation, which typically requires stronger field confinement in narrow gaps, is more sensitive to the structural heterogeneity of the hotspots compared to the more stable horizontal orientation.

The high variability in absolute coefficients reflects the inherent challenges in the reproducible synthesis of particle-like SERS substrates. However, the primary value of our platform lies in the ability to quantify these fluctuations accurately. As shown in the representative spectra (Fig. 4(b)), the transition from high-intensity particles (red) to low-intensity ones (purple) involves not only a decrease in signal-to-noise ratio but also subtle shifts in the relative contribution of each MCR component. The CV of the ratio (\(\textrm{C}_{1}/\textrm{C}_{2}\)) was calculated to be 1.907, providing a quantitative measure of “spectral shape consistency” across the population.

These results demonstrate that our evaluation platform can successfully decompose complex SERS signal fluctuations into interpretable physical parameters. While the current DNA-AuNP crystals exhibit significant heterogeneity, this methodology provides a necessary “performance report” for any candidate SERS probe. By quantifying the limits of both intensity and spectral stability, this framework paves the way for the rational optimization of SERS sensors for reliable intracellular applications.