Plasmonic membrane design and microarray fabrication
We demonstrated bacterial identification using a P-FS, where the nitrocellulose (NC) membrane was modified with a metallic microarray made of gold, and silver. The microarray was generated through desired shapes and sizes of designed patterns, which were transferred onto dry film photoresist (DFP) using UV light radiation during photolithography. The process begins with the lamination of DFP onto a polyethylene terephthalate (PET) substrate at 90°C (Figure 1A, Figure S1), followed by photolithography where the DFP is exposed to UV light through a photomask to create a patterned array. This is then laminated with a NC membrane at 90 °C, resulting in a bonded membrane. Figure 1B shows a detailed view of the 100 µm-sized pattern generated on the membrane through the photolithography process, with a closer look through the SEM revealing intricate fibrous and porous structures of the membrane. As the nitrocellulose was layered with DFP, metal was deposited on the top side of the DFP using a sputtering process to generate the microarray for SERS enhancement. After wetting the NC membrane with water to facilitate subsequent steps, the DFP was removed, leaving the patterned metal array on the NC membrane and resulting in a final product that mirrors the original photomask pattern (Figure 1C, steps 4-7).
A metal layer of two different thicknesses, 10 and 30 nm, was applied to pattern sizes of 50, 100, 150, and 200 µm to optimize hotspot generation in each metal deposition. After coating with metal and removing the DFP, plasmonic arrays of 40,400, 10,200, 4,489, and 2,600 features were generated on the surface of the NC membrane (Figure S2A). Various patterned shapes, including circles, triangles, and hearts, are showcased in Figure 1D, demonstrating the process's versatility and precision. The first column displays the photomask patterns on the DFP features, while subsequent columns show bright field images of patterns on the NC, and the final metal features with SEM images after the removal of DFP at different magnifications, emphasizing the method's capability to create diverse and accurate patterns. Figure 1E illustrates the correspondence between mask feature size, DFP feature size, and metal feature size using different colors to represent each stage (DFP in black, NC in blue, and Metal in brown). The graph in Figure 1F, and Figure S2B plots DFP feature size against mask feature size and metal feature size, showing a linear relationship. This correlation indicates the process's predictability and scalability, with inset images providing visual reference for various feature sizes.
The pore size of the membrane, based on the thickness of metal coating on different patterned structures, was determined using SEM images (Figure 2). These images illustrate the structural characteristics of silver (Ag) and gold (Au) films deposited at various thicknesses and subjected to different thermal conditions. Micrographs were taken showing structural changes at 20°C for 1 hour (left columns) and at 150°C for 1 hour (right columns) for Ag and Au films at thicknesses of 5 nm, 10 nm, 20 nm, and 40 nm.At 5 nm, the film appears smooth and continuous at 20°C (scale bar 300 nm), but shows noticeable agglomeration at 150°C in Ag deposition, while Au exhibits only slight agglomeration, indicating nanoparticle formation. The 10 nm film shows some granular structure at 20°C, with further coalescence upon heating in Ag deposition. For 20 nm thickness, the film at 20°C displays a more distinct granular morphology, while heating leads to significant particle formation. The 40 nm film is relatively rough at 20°C and forms large clusters after heating, indicating extensive agglomeration. In contrast to Ag, Au films maintain a smoother surface under the same conditions. The results reveal that increasing the temperature to 150°C significantly affects the morphology of both Ag and Au films, inducing agglomeration and grain growth, which is more pronounced in thicker films. Silver films exhibit more extensive structural changes with temperature compared to gold films, which tend to maintain a smoother morphology at lower thicknesses.
Figure 2C shows the overall pattern and pore structure of the 50, 100, 150, and 200 µm plasmonic arrays on Ag deposition. The pores on the 50 µm plasmonic array reveal a dense network structure in the top micrograph, while the pores on the 100 µm array show a similar dense network with clearly defined patterned pores and high roughness. In contrast, the 150 µm and 200 µm arrays display a consistent smooth network structure and the largest patterned pores, which, due to their high smoothness, did not generate hotspots for the SERS applications.
Overall, this demonstrates the process's capability to create patterned membranes with various pore sizes, maintaining uniformity and consistency even at different sizes, and highlights the precision and control of the fabrication process.
Performance analysis of a hand-powered plasmonic fidget spinner
We developed a customized hand-powered P-FS that operates without electricity, and by replacing the NC membrane used in our Dx-FS study with an advanced plasmonic membrane, we successfully converted the Dx-FS into the P-FS. Figure 3 presents a comprehensive analysis of a microfluidic device designed for the enrichment and detection of bacterial cells using a plasmonic membrane, providing a schematic layout that indicates the position of each component to facilitate understanding of the fluid flow and interaction between the sample and the plasmonic membrane within the device.
Additionally, a photograph of the device is shown with labeled components, including the sample inlet (1), chamber (2), plasmonic membrane (3), reservoir (4), absorption pad (5), vent (6), and bearing (7) (Figure 3A, 3B, Figure S4). The close-up image highlights the detailed structure of the 14 mm plasmonic membrane, which is crucial for capturing and enriching bacteria from the sample (offset image Figure 3A: white patterned on the blue background, Figure S3). This membrane was fitted onto the P-FS for bacterial enrichment, with the detailed fabrication process provided in the methodology.
Figure 3C illustrates the operational workflow, showing the process of spinning to enrich bacteria, beginning with the device in its initial state, progressing through the spinning phase where the sample is moved towards the plasmonic membrane, and concluding with the post-spin state where the bacteria are concentrated on the membrane. In the pre-spin stage, a buffer was added to the bottom chamber of the plasmonic membrane using fluid-assisted separation technology (FAST), which ensured uniform distribution of bacterial cells on the membrane at the post-spin stage (Figure 3D). FAST created a uniform pressure difference across the membrane by removing air from the chamber. During spinning, centrifugal force drove the sample towards the membrane, facilitating bacterial attachment, and post-spin, the bacteria were concentrated on the membrane, ready for detection.
The kinetic energy generated by hand was applied to the P-FS, converting it into centrifugal force that enabled the device to spin. Performance analysis revealed variability in spinning speeds achieved by different operators, similar to the operator-dependent variability observed in a Dx-FS device (Figure 3E, Figure S5). This variability suggests that the metal patterning on the filter did not affect the performance of filtration speed, possibly due to the nanoscale size of the patterning, which maintained consistent filtration efficiency across different spinning speeds.During spinning, an angular rotational frequency (ω) of up to 311 rad s⁻¹ was generated, depending on the strength of the individual operator, and the filtration efficiency was directly related to the spinning power. We allowed 10 different operators to spin the device and summarized the power output across multiple trials. The ωmax range varied from 78 to 311 rad s⁻¹ across 100 measurements, with an average of 178.2 ± 62.2 rad s⁻¹.
Raman spectroscopy was employed to identify distinct vibrational signatures in molecules by analyzing three dye-modified polystyrene (PS) beads coated with methyl red, rhodamine 6G, and fluorescein, as well as two biomolecule-modified PS beads with L-arginine and L-ascorbic acid. The workflow for the Raman measurement process, as outlined in Figure 4A, includes loading the dye-modified PS bead sample into the device, spinning the device by hand for centrifugal enrichment, conducting Raman measurements to obtain molecular fingerprints, and generating a SERS intensity map. The beads enriched on the membrane were characterized using SERS spectra under specific measurement conditions: a 525 nm laser, 1-second exposure time, and 0.9 mW laser power. This setup demonstrated the stretching and bending vibrations of different bonds within the individual molecules. Comparing the Raman spectra obtained with the Dx-FS and P-FS from the different molecules highlighted a significant improvement in sensitivity (Figures 4B, 4C). The P-FS provided clearer and stronger signals with specific molecular fingerprints for each dye, with key enhanced peaks observed at 1406 cm⁻¹ for methyl red-PS (N=N rings), 1651 cm⁻¹ for rhodamine 6G-PS (C-C), 1632 cm⁻¹ for fluorescein-PS (C-O), 982 cm⁻¹ for L-arginine-PS (N-C), and 1792 cm⁻¹ for L-ascorbic acid-PS (C=O), corresponding to their molecular vibrations (Figure S6). The SERS intensity map demonstrated a significant enhancement—3.8-fold for the gold membrane and 32.7-fold for the silver membrane in the P-FS—compared to the control study using an NC membrane (Figure S7).
Additionally, the study explored the effect of metal type (silver and gold), film thickness (5 nm, 10 nm, 20 nm, and 40 nm), and temperature (20°C and 150°C) on SERS intensity at specific vibrational modes for various dye-modified PS beads, including methyl red (1406 cm⁻¹, ν(N=N rings)), rhodamine 6G (1651 cm⁻¹, ν(C-C rings)), fluorescein (1632 cm⁻¹, ν(C-O rings)), L-arginine (982 cm⁻¹, ν(N-C rings)), and L-ascorbic acid (1792 cm⁻¹, ν(C=O rings)). These results highlight the importance of optimizing film thickness and temperature for effective SERS-based detection, with significant differences observed between Ag and Au films (Figure S8). The optimal SERS signals were achieved with silver at a thickness of 10 nm and a temperature of 150°C. Further analysis of different silver pattern sizes (50 µm, 100 µm, 150 µm, and 200 µm) revealed that the best plasmonic signal was obtained with a 100 µm pattern size, which was selected for further applications (Figure S9). For all the molecules, at their main peak positions—1406 cm⁻¹ for methyl red, 1651 cm⁻¹ for rhodamine 6G, 1632 cm⁻¹ for fluorescein, 982 cm⁻¹ for L-arginine, and 1792 cm⁻¹ for L-ascorbic acid—Raman intensity measurements across 10,000 plasmonic arrays (153.94 mm²) confirmed the presence of the molecules (Figure 4D).
Based on the SERS intensity, all plasmonic arrays (~10,000, 153.94 mm²) present in the membrane were mapped to determine a threshold (blank + 3 × standard deviation) for distinguishing the presence of molecules, which was then used to differentiate between positive ("1") and negative ("0") Raman signals, transforming the data into a binary format. This binary data was subsequently used for digital counting, followed by linear regression analysis to correlate Raman counts with dye concentrations, where higher concentrations yielded higher Raman counts, establishing a linear relationship between signals and analyte concentration (Figure 5A).
Here, we present SERS intensity digital counting for methyl red-PS corresponding to a specific Raman shift (1406 cm⁻¹), with signal counts increasing as concentrations ranged from 0 to 10⁻⁸ M. These Raman counts visually demonstrate the correlation between concentration and detected Raman signals, where higher concentrations result in denser and more numerous signal "1" areas (represented in yellow), while "0" areas, which did not cross the threshold (532.5), are shown in purple (Figure 5B). Figure 5C presents linear regression analyses for various dye-modified PS beads, including methyl red-PS, rhodamine 6G-PS, fluorescein-PS, L-arginine-PS, and L-ascorbic acid-PS, with each graph plotting the number of positive Raman counts against the concentration of dye-PS beads, and insets highlighting the correlations at lower concentrations. The high R² values (0.9739 for methyl red-PS, 0.9576 for rhodamine 6G-PS, 0.9336 for fluorescein-PS, 0.9271 for L-arginine-PS, and 0.9731 for L-ascorbic acid-PS) indicate a strong linear relationship, confirming the reliability and sensitivity of the digital counting method in quantifying low concentrations of dye-modified PS beads. The calculated limit of blank, limit of detection, and limit of quantification for methyl red-PS were 6.3 pM, 12 pM, and 38 pM, (y = 4E+11x + 2.3) respectively; for rhodamine 6G-PS were 4.1 pM, 82 pM, and 250 pM, (y = 4 E+11x + 3) respectively; for fluorescein-PS were 4.7 pM, 9.5 pM, and 28 pM, (y = 4 E+11x + 1.7) respectively; for L-arginine-PS were 4.1 pM, 8.2 pM, and 25 pM, (y = 4 E+11x + 2) respectively; and for L-ascorbic acid-PS were 5.0 pM, 10.0 pM, and 30.5 pM, (y = 4 E+11x + 1.7) respectively. Overall, the process flow, intensity maps, and regression analyses in this figure collectively validate the SERS-based approach as a robust and sensitive technique for detecting and quantifying low concentrations of analytes. The method's strength lies in its ability to visually represent and quantify the relationship between analyte concentration and Raman signal intensity, providing both qualitative and quantitative insights.
Identification of bacteria using Raman spectroscopy
We explored the potential of the P-FS to identify bacterial species by analyzing the unique molecular compositions that define different bacterial phenotypes, which result in distinct variations in their Raman spectra. Each bacterial species exhibits a characteristic set of vibrational modes in its Raman spectrum, reflecting the specific biomolecular makeup of its cell structure, such as proteins, lipids, nucleic acids, and carbohydrates. By harnessing these unique spectral signatures, the P-FS allows for the precise identification of bacterial species based on their distinct Raman fingerprints, enabling differentiation even among closely related bacterial strains. In this study, we employed five different bacterial strains—E. coli 25922, S. aureus 25923, E. coli MG1655, Lactobacillus brevis, and S. mutans 3065—each at a concentration of 10⁵ CFU/mL, for identification using the SERS mechanism (Table S1).
A 1 mL bacterial solution was introduced into individual wells of the P-FS device, where the bacteria were enriched on the plasmonic membrane array during the spinning process. Following enrichment, the bacteria on the plasmonic membrane were characterized using Raman spectroscopy, which provided detailed spectral fingerprints unique to each bacterial species (Figure 6A). The distribution of bacterial cells on the plasmonic array was modeled using a Poisson distribution at this concentration, which predicted that more than 90% of the plasmonic arrays contained between 1 and 9 bacterial cells (Figure S10). This uniform distribution was achieved because FAST created a uniform pressure difference across the membrane during spinning, allowing centrifugal force to effectively drive the sample toward the membrane, facilitating consistent bacterial distribution across the plasmonic arrays.25
For each bacterial species, we mapped 10,000 plasmonic arrays, each measuring 100 × 100 µm², on the plasmonic membrane. The bacteria exhibited characteristic Raman spectra with high signal-to-noise ratios, achieved with an integration time of 30 seconds and a 532 nm laser at a power of 0.99 mW. These spectra provided some shared basic structural components such as DNA (adenine 726 cm-1), cell wall (957 cm-1) and proteins (1003 cm-1 for C-H plane of phenylalanine and C-C aromatic ring breathing) and many distinct vibrational signatures for each bacterial species, enabling their precise identification and differentiation.26 Key vibrational modes unique to each bacterial species were identified and assigned, including 1567 cm⁻¹ for E. coli 25922, 2413 cm⁻¹ for S. aureus 25923, 1494 cm⁻¹ for E. coli MG1655, 1241 cm⁻¹ for S. mutans 3065, and 2579 cm⁻¹ for Lactobacillus brevis. These distinct peaks, corresponding to specific molecular vibrations unique to each bacterial species, allow for accurate quantification and are visually represented in SERS intensity maps that depict the distribution and density of Raman signals, with higher intensity regions highlighting bacterial fingerprints aligned with the specific Raman shifts identified in the spectra (Figure 6B).
After assigning the threshold on intensity of each Raman shift for individual bacteria (498.47 at 1567 cm⁻¹ for E. coli 25922, 513.23 at 2413 cm⁻¹ for S. aureus 25923, 508 at 1494 cm⁻¹ for E. coli MG1655, 521.27 at 1241 cm⁻¹ for S. mutans 3065, and 511 at 2579 cm⁻¹ for Lactobacillus brevis) these intensity maps are transformed into binary data, assigning "1" for the presence and "0" for the absence of signals. The total Raman counts for each bacterial species were recorded, demonstrating the effectiveness of digital counting in quantifying bacterial presence: E. coli 25922 had 363 counts, S. aureus 25923 had 251 counts, E. coli MG1655 had 243 counts, S. mutans 3065 had 199 counts, and Lactobacillus brevis had 280 counts (Figure 6C).
The SERS intensity maps and digital counting results effectively illustrate the ability to detect and quantify bacterial presence. Transforming intensity maps into binary data facilitates straightforward quantification, providing clear and interpretable results. Linear regression analyses in Figure 6D show strong correlations between Raman counts and bacterial concentrations for all tested species, with high R² values confirming the method's accuracy. The calculated limit of blank, limit of detection, and limit of quantification for E. coli 25922 were 15 CFU/mL, 31 CFU/mL, and 94 CFU/mL, respectively; for S. aureus 25923 were 26 CFU/mL, 53 CFU/mL, and 163 CFU/mL; for E. coli MG1655 were 14 CFU/mL, 29 CFU/mL, and 89 CFU/mL; for S. mutans 3065 were 22 CFU/mL, 45 CFU/mL, and 138 CFU/mL; and for Lactobacillus brevis were 20 CFU/mL, 41 CFU/mL, and 124 CFU/mL, respectively. This quantitative capability is essential for applications requiring precise bacterial load measurements.
Next, we wanted to demonstrate the detection capability of the P-FS in complex biological samples, such as urine, where multiple bacterial species may be present. Figure 6E in the image presents digital counting results of urine samples that have been spiked with Lactobacillus brevis and E. coli 25922. The first map on the left serves as a control, showing the SERS intensity map of a blank urine sample with negligible signal, indicating the absence of detectable bacterial fingerprints. The second map illustrates the detection of Lactobacillus brevis in urine, with red dots representing the presence of the bacteria at a Raman shift of 2579 cm⁻¹. The third map shows the detection of E. coli 25922, where green dots indicate bacterial presence at a Raman shift of 1567 cm⁻¹. The rightmost map merges the detection of both Lactobacillus brevis and E. coli 25922 in the same urine sample, with red and green dots clearly identifying each species. This visualization underscores the SERS method's ability to simultaneously detect and distinguish multiple bacterial species within a single sample, demonstrating its specificity and potential for clinical applications, such as diagnosing infections in urine samples.
Outlook
This study presents an innovative method for the detection, identification, and quantification of bacterial species by combining Raman spectroscopy, digital counting, and plasmonic-enhanced devices. The microfluidic device, featuring a plasmonic membrane, effectively enriches bacterial cells through a simple hand-spinning process, enhancing both user-friendliness and broad applicability. The P-FS significantly improves the sensitivity of Raman signal detection, enabling the detection of low concentrations of bacterial cells and other analytes—an essential feature for high-sensitivity applications. Unique Raman fingerprints were identified for different bacterial species, allowing precise identification based on distinct Raman shifts and facilitating accurate microbial analysis.
The digital counting method transforms SERS intensity maps into binary data, enabling straightforward quantification. Linear regression analyses demonstrated strong correlations between Raman counts and bacterial concentrations, confirming the method's accuracy and reliability for quantitative measurements. The study also applied this method to detect bacterial species in complex samples, such as urine, highlighting its practical utility in real-world diagnostic scenarios. The advancements in device design, signal enhancement, and data processing pave the way for numerous future applications in scientific and industrial fields, offering a powerful tool for microbial analysis and beyond. Future developments, particularly integrating SERS in the P-FS as a point-of-care test, hold promise for enhancing its reliability and accessibility in remote and developing areas, positioning it as a revolutionary tool in microbial diagnostics and public health.
