Biological SERS-active sensor platform based on flexible silk fibroin film and gold nanoislands

: In contrast to conventional surface-enhanced Raman scattering (SERS) platforms implemented on non-biological substrates, silk fibroin has the unique advantages of long-term biosafety and controllable biodegradability for in vitro and in vivo biomedical applications, as well as flexibility and process-compatibility. In this study, a silk fibroin film was developed to fabricate a flexible SERS sensor template with nanogap-rich gold nanoislands. The proposed biological SERS platform presents fairly good enhancements in detection performance such as detection limit, sensitivity, and signal-to-noise ratio. In particular, the sensitivity improvement was by more than 10 times compared to that of the counterpart sample, and an excellent spatial reproducibility of 2.8% was achieved. In addition, the near-field calculation results were consistent with the experimental results, and the effect of surface roughness of the silk substrate was investigated in a quantitative way. It is believed that biological SERS-active sensors could provide the potential for highly sensitive, cost-effective, and easily customizable nanophotonic platforms that


Introduction
Raman spectroscopy is a non-destructive detection technique that can provide detailed information about the molecular structure of target material through an interaction between light and chemical bonds within the material [1].We can efficiently identify the complex biochemical structures of biomolecules because of their unique Raman signals.However, because weak Raman signals are often overwhelmed by the fluorescence background, many researchers have incorporated metallic nanostructures on the substrate to amplify the signals, this is called surface-enhanced Raman scattering (SERS) effect [2][3][4][5].By utilizing localized surface plasmon resonance (LSPR) as a result of coupling between light and metallic nanostructure, it has been reported that extremely high enhancement in Raman scattering could be achieved [6][7][8][9][10].When metallic nanostructures are conjugated with biomolecular ligands, they can be manipulated to target cells and tissues with high affinity and bind to multiple diagnostic and therapeutic agents [11,12].Recent advances in SERS techniques have led to the development of self-assembled plasmonic nanoparticles for drug delivery [13], label-free monitoring of drug metabolism in living cells [14], and nanoparticle tags for in vivo tumor targeting [15].
As a next-generation SERS sensor, a wearable or implantable platform for personalized diagnosis is essential to monitor important biomolecules inside the body.Such in vitro and in vivo SERS approaches should be supported using flexible substrates.Soft and flexible materials can conform to complex and dynamic surfaces in biological systems and fill the gaps where rigid substrates cannot be used.For example, polymer substrates such as polyethylene terephthalate and polyethylene naphthalate with high transparency and low water permeability have been developed [16,17].Polyimide and parylene with conformability in biological media have been available at high processing temperatures [18,19].However, these non-biological materials are not directly suitable for wearable or implantable sensing applications due to the limitation in biocompatibility and biodegradability.In addition, realization of a variety of substrate forms is not feasible.
Recently, silk has gained much attention for biomedical applications owing to its unique optical, mechanical, and physiochemical characteristics [20][21][22].In the native form, a glue-like layer of sericin covers two singular filaments of fibroin composed of light and heavy chains.The heavy chain of the fibroin protein consists of an antiparallel β-sheet, a repetitive amino acid sequence that generates crystalline domains for ordered structures.As glycine increases, β-sheets bind more tightly and molecular interactions enhance the rigidity and tensile strength of the silk in bulk, imparting a mechanical tunability and robustness [21].When silk protein is annealed into films, most of the secondary structures are crystallized.This high crystallinity can improve the optical transparency by more than 90% over a wide wavelength range from 300 nm to 2 µm [23].Owing to its advantages of long-term biocompatibility, controllable biodegradability, and no inflammatory response in vivo [22], silk fibroin has been considered as an excellent candidate to serve as a biological supporting and packaging material in various fields of neural prosthetics, tissue engineering, and drug delivery [24][25][26][27].
Unfortunately, none has yet demonstrated that a biological SERS substrate can be used as a versatile nanomaterial platform with highly sensitive, cost-effective, and easily customizable abilities.Further, the softness and flexibility of the silk fibroin enables intimate contact with curved surfaces of the skin, tissues, or organs.Transparent and process-compatible silk fibroin facilitates its integration with functional materials, providing great convenience for micro/nanofabrication processes [28,29].Silk fibroin has a slow and controllable degradation rate and the ability to be fabricated into multiple forms of fibers, films and gels [30,31], and has, therefore, emerged as an ideal material of choice for the design and integration of high-performance biological sensing platforms.
In this study, for the first time, we propose a SERS-active detection platform incorporating a biological silk template and gold nanoislands (GNIs).We developed a simple but effective process of synthesizing silk fibroin film and obtained a free-standing silk substrate of a large area.It should be emphasized that the silk-based platform was smooth, flexible, transparent, and durable, which could be advantageous for biophotonic devices.Moreover, the constructed nanoislands allowed a rich nanogap structure and good spatial uniformity, which is essential for sensitive and spatially reproducible SERS detection.To investigate the influence of the proposed silk platform on the SERS characteristics, we performed various Raman scattering studies numerically and experimentally, and compared the results for individual SERS substrates.

Fabrication process of flexible silk substrate
We fabricated a biological SERS sensor by creating GNIs on a soft and flexible substrate of silk fibroin, which is composed of a repetitive amino acid sequence (Gly-Ser-Gly-Ala-Gly-Ala) n (Fig. 1(a)).The fabrication procedure is presented in Fig. 1(b).B. mori cocoons were cut into small pieces and boiled in 0.02 M sodium carbonate at 95 °C for 30 min once and 0.01 M sodium oleate solution at 95 °C for 30 min twice to fully remove the residual sericin from the silk fiber.During the degumming process, sodium carbonate quickly removes sericin and a prolonged boiling may affect the quality of silk fibroin.In contrast, sodium oleate slowly removes sericin but the loss of silk fibroin is not significant [32].We repeated the second degumming step twice to ensure the removal of residual sericin and to preserve the structure of the silk fibroin.Note that sericin should be removed to prevent unwanted immune responses due to the potential harmful interplay between fibroin and sericin [22].The cocoons were then rinsed in deionized (DI) water for 30 min and dried at room temperature overnight.The degummed silk fibroin was dissolved in a 9.3 M LiBr solution at 60 °C for 4 h, followed by filtering the solution dissolved in silk fibroin using a cloth membrane filter for artifact removal.Dialysis was then conducted for 2.5 days to extract the pure silk fibroin solution from the filtered solution.The filtered solution was injected into a dialysis membrane tube (12-14 kDa MWCO, Fisher Scientific) and then dialyzed against DI water.In this dialysis process, DI water was periodically replaced and maintained at 5 °C.The solution was centrifuged at 7,000 rpm for 20 min to remove impurities.To fabricate a flexible silk film, the pure silk solution was dropped onto petri dish and then dried at room temperature for 24 h.Finally, 2 nm-thick gold was deposited onto silk substrate at a deposition rate of 0.2 Å/s via electron beam evaporator for GNI formation.While the results are not presented here, a GNI thickness of 2 nm was determined to be the optimum with the largest SERS peak intensity when the thickness was varied.The plasmonic silk substrate was optically transparent, mechanically flexible, and robust (Fig. 1(c)).The device showed good adhesion between the silk film and GNIs because no damage of crack or breakage was observed on bending or stretching.

SERS experiments
To investigate the SERS characteristics, Raman signals of 4-ABT for four types of SERS substrates were prepared and immersed for 15 min in various concentrations of the 4-ABT solution.After 4-ABT immobilization, the substrates were rinsed with ethyl alcohol and DI water for 5 min to remove non-immobilized 4-ABT.Raman spectra of 4-ABT were obtained at five different points in each given sample with an acquisition time of 50s at 100× magnification.The experimental setup for the Raman spectra measurements consisted of a microscope (BX43, Olympus), a continuous wave laser of λ = 785 nm (I0785MM0350MF, Innovative Photonic Solutions), a spectrometer (SR-303i-A, Andor Technology) and a low dark current deep-depletion CCD detector (iVac, Andor Technology).

Numerical simulation using the finite-element method (FEM)
For normal time-varying fields, Maxwell-Ampere's and Faraday's equations can be written in the following forms using the constitutive relations of D = εE and B = µH as well as a current J = σE, where D, ε, E, B, µ, H, J, and σ are the electric displacement or electric flux density, permittivity, electric field intensity, magnetic flux density, permeability, magnetic field intensity, current density, and conductivity, respectively.
Assuming sinusoidal excitation and linear media, the two laws can be combined into a time harmonic equation for the electric field.
Using the relation ε = n 2 , where n is the refractive index, the above equation is transformed into the alternate form: where k 0 is the free-space wave number.The equation was formulated using FEM and numerically solved using Wave Optics Module of COMSOL Multiphysics (version 5.3a).We computed the transverse magnetic mode when a plane wave at λ = 785 nm had a normal incidence of unit amplitude at the interfacial boundary between the crystalized silk substrate and plasmonic nanostructures.The computational domain was flanked by the Floquet periodic boundary condition.

Results and discussion
Field-emission scanning electron microscope (FE-SEM) images were acquired to investigate the morphology of the GNIs on the silk substrate (Fig. 2(a)).For comparison with the results on a highly smooth substrate, GNIs were also constructed on a silicon wafer using an electron beam evaporator (Fig. 2(b)).From the SEM images, we found that the GNIs had a well-distributed pattern over a large surface area on silk and silicon substrates.The average values of width d w and gap d g of GNIs on a silk film were determined to be d w = 24.4nm and d g = 10.1 nm, which were consistent with the GNIs on a silicon wafer (Fig. 2(c)).Atomic force microscopy (AFM) data showed that a bare silk substrate had a nanoscale roughness compared to that of a smooth silicon wafer (Figs.2(d) and 2(e)).Two geometric parameters, the distance between valleys d w_AFM and height of valley d h_AFM were extracted from the AFM results.The averaged roughness obtained from probability distribution of d h_AFM was 3.0 nm for a bare silk film and 0.2 nm for a silicon wafer (Fig. 2(f)).It is interesting to note that such a contrast in surface roughness was not critical for the formation of GNIs because the GNI patterns on both substrates were statistically equivalent, as shown in Fig. 2(c).Next, we measured the scattering characteristics of four types of SERS substrates; i) GNIs on a silk substrate (Sample A), ii) GNIs on a silicon wafer (Sample B), iii) a 40-nm thick gold film on a silk substrate (Sample C), and iv) a 40-nm thick gold film on a silicon wafer (Sample D).In the SERS experiments, 4-ABT molecule, also known as para-aminothiophenol, was chosen as the Raman probe.4-ABT can be specifically immobilized on a gold surface via thiol groups.4-ABT is also used as a linker to tightly combine ligands with nanoparticles because of its special characteristic of double functional groups, which plays an important role in Raman immune detection [33].After the SERS substrates were immersed in the 4-ABT solution for 15 min to induce chemical binding, SERS spectra were measured at five different sites on the individual SERS substrates (Fig. 3(a)).The Raman peak at 1076 cm −1 of 4-ABT, which is associated with the a 1 -type vibrational mode was selected as the primary signal.For a concentration of 10 µM, we observed average primary SERS peaks of 22,213 ± 307 for Sample A and 1555 ± 65 for Sample B (Figs. 3(b) and 3(c)).Obviously, silk substrates led to a significant signal enhancement in the presence of GNIs and the intensity of the representative peak at 1076 cm −1 for Sample A was about 14 times larger than that of Sample B.
SERS enhancement factor (EF) was defined as EF = (I SERS /N SERS )/(I BULK /N BULK ), where I SERS and I BULK are the SERS and Raman peak intensities at 1076 cm −1 for 4-ABT, respectively, and N SERS and N BULK are the numbers of 4-ABT molecules on the SERS substrate and 4-ABT powders within a laser spot volume [34].The Raman peak intensity I BULK at 1076 cm −1 for the 4-ABT powder was 1476.Using the conditions of a laser spot size of 10 µm, penetration depth of 2 µm, and 4-ABT molecule size of 0.2 nm, the EF values were 5.49 × 10 4 for Sample A and 3.84 × 10 3 for Sample B, respectively.The SERS EF values obtained were in good agreement with previous results obtained for aggregated metallic nanostructures [35].Subsequently, we analyzed the detection limit and sensitivity by varying the concentration of 4-ABT in the range of 200 nM to 10 µM.For Sample A, we found the primary SERS peak at a concentration as low as 500 nM (Fig. 3(b)), whereas the peak of Sample B was observable at a concentration of 2 µM (Fig. 3(c)), presenting an improvement in the detection limit by 4 times.We also calculated the slope between the average SERS peak intensity, I SERS_AVG and the concentration of 4-ABT, C 4−ABT using linear regression analysis to quantify the sensitivity S, i.e. S = ∆I SERS_AVG /∆C 4−ABT (Fig. 3(d)).The sensitivity value of S = 18,521 for Sample A was about 10 times larger than that of S = 1786 for Sample B, while R 2 values were determined to be 0.91 and 0.92, respectively.The signal to noise ratio (SNR) was defined as SNR = I SERS_AVG / σ, where σ is the standard deviation.For very low concentrations at the detection limit, the SNR was determined to be 11.4 for Sample A compared to 5.65 for Sample B, thereby presenting 2 times the SNR enhancement.It is noteworthy that Sample A showed a higher sensitivity and a wider dynamic range of target concentration, compared with the performance of Sample B. On the other hand, Raman signals were not found for Samples C and D (Figs. 3(e) and 3(f)) because non-localized propagating plasmon modes by a thin gold film could not contribute to an enhancement of Raman signals to detect 4-ABT molecules.We speculated that such a notable enhancement in Sample A could be attributed to the electromagnetic field enhancement by plasmonic nanostructures on the silk substrate, especially in the vicinity of the GNIs.These localized surface plasmons greatly amplify the subsequent Raman intensities of the target analytes.
To study the underlying physics of localized field amplification and plasmonic field-analyte interactions, we performed an FEM simulation.FEM models were prepared for Sample A and Sample B with perfectly flat surfaces.Refractive indices (n, k) of silicon, gold, and silk at λ = 785 nm were chosen to be (3.7060,0.0074), (0.2313, 4.3933) and (1.5405, 0), respectively [36][37][38].As a quantitative metric of field-analyte interactions, we defined an overlap integral (OI) (Fig. 4(a)), which is the integration of the electric field amplitude within a 0.2-nm thick binding layer of 4-ABT [39].The shape of the GNI was assumed to be a rectangle with a thickness of 2 nm, width of 20 nm and gap distance between GNIs of 10 nm.The FEM results showed that the electric field amplitude at the corner of the GNIs of Sample A was much stronger than that of Sample B (Fig. 4(b)).The OI value of Sample A was nine times higher than that of the EF and detection sensitivity, as shown in Fig. 4, providing an estimate of the detection performance (Fig. 4(c)).Moreover, the plasmonic field-analyte interaction was maximized at d g = 10 nm; fortunately, this theoretical optimum gap distance coincided with the GNIs experimentally realized on the silk substrate (Fig. 4(d)).However, in practice, the real surface profile of a substrate tends to be rough to some degree, which is caused by inherent imperfections in the fabrication processes.Since surface roughness could affect SERS characteristics, we reflected the averaged roughness of 3.0 nm for the silk substrate into FEM simulation, while a silicon substrate with a roughness of 0.2 nm was considered as an ideally flat surface.To understand the effects of the surface roughness and the nanogap between gold nanoislands on field enhancement and field-analyte interactions simultaneously, we assumed 2D FEM models of gold nanoislands on the silk substrate whose interfacial boundaries were modulated using AFM data in Fig. 2(d).The total number of non-flat silk surfaces was 104, and the individual simulation results were compared to the cases with a flat surface.The overall surface dimension of 1 µm was large enough for the given geometric values, and thus a number of statistical variations were included in the constructed surface profile, indicating that the use of a single rough surface can provide valid insights into the averaged roughness effect.
Among the three FEM models, Sample A, which had a non-flat surface, showed the highest field enhancement compared to the other flat substrates (Figs.5(a) and 5(b)).At the regions of interest marked with a star icon, we found notable field enhancement at the corners and in the middle of the GNIs for the silk substrates.Because GNIs introduced to excite LSPR modes are a few tens of nanometers in size, the SERS effects can be influenced by nanoscale changes in the surface profile.Interestingly, for all configurations of plasmonic silk substrates with and without a surface roughness, the OI values were typically greater for Sample A with a non-flat surface (Fig. 5(c)).For a more explicit comparison, we calculated the electric field amplitudes at all corners of GNIs.The difference in OI values between non-flat and flat silk substrates was found as a biased value of 0.14 (Fig. 5(d)).Considering the field amplitude at all corners of the GNIs, a higher localized field amplitude was dominant for Sample A with a non-flat surface and statistically, 80% of the corners of the GNIs exhibited greater field-analyte interactions for the non-flat cases (Fig. 5(e)).Based on the correlation between OI and SERS characteristics, FEM analyses demonstrated that flexible silk templates could provide higher detection performance in the presence of surface roughness.
Finally, the 2D Raman mapping data were measured at a square of 100 × 100 µm 2 to investigate the spatial reproducibility of the SERS signals.Figures 6(a) and 6(b) showed the SERS mapping images for the primary SERS peaks of Samples A and B at a concentration of 10 µM for 100 different locations.The mapping images and graphs in Figs.6(c), 6(d), and 6e exhibited that the plasmonic silk substrate had a relative standard deviation (RSD) of 2.8%, which was better than the silicon wafer substrate (RSD = 6.3%).Although the absolute deviation was larger for the silk substrate due to the surface roughness, the normalized RSD value to evaluate the repeatability in SERS signals could be improved by Raman signal enhancement.

Conclusion
In this paper, we reported the fabrication and characterization of a novel SERS sensor platform consisting of a flexible silk fibroin film and nanogap-rich GNIs.We successfully developed a simple but effective process of synthesizing silk fibroin films and constructed the large-area GNIs with good spatial uniformity onto the free-standing silk substrate.When the target molecules were adsorbed onto the GNI pattern, the SERS intensity was enhanced by approximately 14 times compared to a silicon wafer with a smooth surface, and fairly good performance in terms of detection limit and signal-to-noise ratio was achieved.From theoretical analyses, it was found that such enhancement was associated with two factors: an improved field-analyte interaction in the vicinity of GNIs on the silk substrate and the effect of nanoscale roughness of the silk substrate by promoting a strong light scattering.The 2D Raman mapping data also confirmed that the proposed SERS sensor provided an excellent spatial reproducibility of less than 3%.Our biological SERS-active platform is expected to have great potential for the highly sensitive and quantitative detection of a variety of in vitro and in vivo biomedical applications.

Fig. 1 .
Fig. 1.Plasmonic silk fibroin substrate for SERS application.(a) Schematic illustration of a biological and flexible substrate for wearable healthcare device.SERS platform consists of two functional parts, gold nanoislands that enhance an interaction between target analytes and localized surface plasmons and a silk substrate with biocompatibility and flexibility.(b) Procedures for fabricating plasmonic silk substrate.(c) Photography of the fabricated device.

Fig. 2 .
Fig. 2. Surface characteristics of the fabricated samples.(a) FE-SEM images of GNIs on a silk substrate (Sample A) and (b) on a silicon substrate (Sample B).(c) Distributions of width d w and gap distance d g of GNIs deposited on Sample A (top) and on Sample B (bottom).AFM images of (d) a bare silk substrate and (e) a silicon wafer.(f) Probability distribution of distance between valleys d w_AFM (top) and height of valley d h_AFM (bottom) for silk and silicon substrates.

Fig. 3 .
Fig. 3. SERS experiments of 4-ABT analytes.(a) Schematic illustration of chemical binding of 4-ABT on SERS substrate and measurement setup.SERS measurements of 4-ABT (b) for Sample A and (c) for Sample B. (d) Linear regression analysis of the intensity at 1076 cm −1 to quantify the sensitivity.(e) SERS measurements of 4-ABT at concentrations of 10 µM for 40-nm thick gold film on a silk substrate (Sample C) and (f) at concentrations of 10 µM for 40-nm thick gold film on a silicon wafer (Sample D).

Fig. 4 .
Fig. 4. Quantitative analyses on a field-analyte interaction.(a) Boundaries, domains, and definitions for FEM simulation.(b) Normalized electric field amplitude |E| for GNIs on a flat silk substrate (top) and for GNIs on a flat silicon substrate (bottom).For GNI model of 2-nm thickness and 20-nm width, (c) characteristics of OI as a function of refractive index n of the dielectric substrate and (d) characteristics of OI as a function of gap distance d g for Sample A (blue line) and Sample B (red line).

Fig. 5 .
Fig. 5. Effect of surface roughness of a flexible silk substrate.(a) FEM data for Sample A with and without a surface roughness (top and middle), and for Sample B with a flat silicon substrate (bottom).(b) Field amplitude profiles along the top surface of individual samples.(c) Comparison of OI values for 104 configurations of non-flat and flat silk substrates.(d) Difference in OI value between non-flat and flat silk substrates.The ensemble averaged difference of a red line is 0.14.(e) Comparison of field amplitudes at the corners of GNIs on non-flat (red) and flat (blue) silk templates.80% of GNI corners showed a higher field amplitude in the presence of surface roughness.

Fig. 6 .
Fig. 6.Spatial reproducibility test by SERS mapping data.SERS mapping images for the primary SERS peak, I SERS at 10 µM (a) for Sample A and (b) for Sample B. Deviations of SERS intensities (c) for Sample A and (d) Sample B. The gray lines indicate 95% confidence intervals of the expected deviation value.(e) Statistics of normalized deviations of SERS intensities for Sample A (top) and Sample B (bottom).