Flexible and Transparent Substrates Based on Gold Nanoparticles and TiO2 for in Situ Bioanalysis by Surface-Enhanced Raman Spectroscopy

Flexible and transparent substrates are emerging as low cost and easy-to-operate support for surface-enhanced Raman spectroscopy (SERS). In particular, in situ SERS detection approach for surface characterization in transmission modality can be efficiently employed for non-invasive analysis of non-planar surfaces. Here we propose a new methodology to fabricate a homogenous, transparent, and flexible SERS membrane by the assistance of a thin TiO2 porous layer deposited on the PDMS surface, which supports the uniform loading of gold nanoparticles over large area. The substrate was first characterized for homogeneity, sensitivity and repeatability using a model molecule for SERS, i.e., 7-mercapto-4-methylcoumarin. Satisfactory intra-substrate uniformity and inter-substrates repeatability was achieved, showing an RSD of 10%, and an analytical sensitivity down to 10 nM was determined with an EF of 3.4 × 105 ± 0.4 × 105. Furthermore, SERS detection of pyrimethanil (PMT), a commonly employed pesticide in crops for human consumption, was performed in situ, exploiting the optical transparency of the device, using both model surfaces and non-flat bio-samples. PMT contamination at the phytochemical concentration levels corresponding to commonly used infield doses was successfully detected on the surface of the yellow Ficus benjiamina leaves, supporting the use of this substrate for food safety in-field application.


Characterization of synthetized AuNPs
The synthetized AuNPs were characterized for shape and dimensions using scanning electron microscopy, and the characteristic localized surface plasmon resonance frequency wavelength was measured by UV-vis absorption spectroscopy.

Optimization of SERS tape preparation procedure
The Design of Experiments approach was used to determine the best conditions to obtain the desired features of the film. Three factors are considered in the optimization of the PDMS substrate preparation, i.e. the thickness (0.2 mm -2 mm), the curing time (3 min -15 min) and the microwave power (300 W -700 W). The monitored responses were mechanical resistance and adhesion; these two features were evaluated on arbitrary performance scales. Marks ranging from 0 to 5 for mechanical resistance and from 0 to 10 for adhesion were assigned to each experiment. The exploited DoE method was D-optimal carried out using the Modde 7 Umetrics® software. 23 experiments were designed with different combinations of the varying parameters, three replicates for the central point were included in the experiment list to enhance the model stability. After the collection of experiment responses, multiple linear regression (MLR) coefficients were calculated and response curves were elaborated to identify the optimal conditions for PDMS preparation. The experimental sheet with the corresponding results is reported in Table S2 in supplementary information.  Figure S4. Good models are obtained as demonstrated by the symmetric random dispersion of the experimental values around the diagonal, which represents the perfect superimposition of real and calculated values. No evident outlier are identified.  The centered and scaled coefficients plot with confidence intervals are presented in Figure S5. The size of the coefficients represents the change in the response when a factor varies of one unit, while the other factors are kept at their averages. The coefficient is significant when the associated confidence bar does not cross the zero. For both responses the most relevant factors are power and time of curing, which provide opposite effects on mechanical resistance and stickiness of the obtained PDMS tape. The longer and more energetic was the curing, the more the tape was resistant. Conversely, very sticky PDMS was obtained with shorter and less energetic curing, as it was expected. The DoE and subsequent modeling is very useful to define the best compromise parameters to satisfy contemporarily the two desired features by defining the region in the experimental space where performance indexes of both responses fulfill the expectations. The response curve plots for both responses are shown in Figure S6 a, b, c, b, and allow to identify the experimental conditions that provide desired results. The sweet spot plot is represented in Figure S3e, as the red-colored region in which all the responses show satisfactory levels. For this specific application, i.e. the fabrication of a flexible SERS tape to be laid down on testing surfaces, it should by sticky enough to adhere on the testing surface, but sufficiently consistent to allow easy handling with tweezers. In the arbitrary appreciation scale for adhesion from 0 (completely inconsistent and gluey) to 10 (perfectly solid and plastic), the desired optimum correspond to 5, whereas the mechanical properties should be maximized, and the desired value was 5. The superimposition of the response curves allows the identification of a sweet spot where both requirements are fulfilled, in particular the final protocol for an optimal PDMS tape preparation consists in drop casting an amount of liquid PDMS in a petri dish adequate to obtain 1 mm thick PDMS layer, curing was carried out at maximum power (700 W) for 15 minutes. In this way reproducible sticky but consistent tapes were obtained.
A similar approach was used to determine the best TiO2 deposition conditions. In this case the evaluated factors were TiO2 paste concentration (5% -20%) and dropped volume (5 µ L -15 µ L). The operative conditions were kept constant in accordance with doctor blade published methodology (51, 52). Also in this case arbitrary appreciation scales are defined and performance indexes are attributed to different experimental results to evaluate the responses. Subsequently, a MRL model was calculated. The monitored responses were flexibility, i.e. absence of fractures on the dried films, adhesion, and diffusion, i.e. AuNPs homogenous distribution. The list of experiments is reported in Table S3 with the associated responses; replicates are included to test reproducibility and strengthen the model.  Figure S7) provided a further proof of models goodness. The coefficients bar plots ( Figure S8) attested that the considered factors, i.e. volume and concentration of titanium dioxide paste, play an opposite role as it was expected.  The experimental design was set to identify the so-called sweet spot, which is the set of experimental conditions that lead to optimal results. Figure S9. visual appearance of TiO2 covered PDMS tape corresponding to undesired and desired experimental conditions. (a, c) thin and flexible TiO2 layer obtained by depositing 8 µ l 10% w/w paste before and after AuNPs deposition respectively; (b, d) thicker TiO2 layer obtained by depositing 15 µ l 20 % w/w paste, before and after AuNPs deposition respectively.
Thanks to the MLR model it was possible to define the proper concentration and amount of TiO2 paste to be spread on the tape to meet the desired requirements, as shown in Figure S10, where the response curves and the sweet spot plot is reported. According to the experimental results, the selected preparation conditions for the titanium dioxide film deposition were a 10% w/w concentration, of which 8 µ L are dropped and rapidly spread with a clean glass stick.  7 mercapto-4-methylcumarin was selected as a Raman reporter because of its relatively large Raman cross section and its strong chemical affinity for gold due to the sulfhydryl functional group (R-SH) Raman spectrum of MMC in solid state is shown in Figure S9 with the characteristic peaks assignment. The band assignments were obtained by a combination of a computational procedure with vibrational information reported in literature (1, 2, 3). Geometry optimization of model MMC structures and consequent calculations of vibrational spectra were carried out with DFT method using Gaussian 03 program (Gaussian 03, Revision B.05, References cited in http://www.gaussian.com). Full geometry optimizations were carried out without symmetry constraints. The computations were performed with the Lee, Yang and Parr correlation functional (LYP) (4) combined with the Becke's non-local three-parameter hybrid exchange functional. Vibrational information coming from the computational procedure were compared with the experimental Raman spectrum of MMC and the main bands in the spectrum were assigned using the Handbook of Infrared and Raman Spectroscopy (5). Figure S12. Raman spectrum of 7-mercapto-4-methylcoumarin (MMC) in solid state with the characteristic peaks assignments.

Substrate homogeneity and response repeatability
The uniformity of the SERS response over an active substrate is a crucial aspect in view of a real application; in order to evaluate the variability of the signal across the substrate, the relative standard deviation (RSD) was evaluated. The RSD is defined as the percent ratio of the standard deviation to the mean, and it is widely adopted in the SERS community to assess the spatial homogeneity of a substrate. Lower RSD values show remarkable homogeneity (6) and very low values have been recently reported (7). However, a standard protocol to evaluate spatial homogeneity of SERS substrates is still missing. Sometimes the RSD is calculated by considering the average of tens to thousands punctual spectra (8,9), whereas it may be obtained from multiple scanning areas on the substrate (10). Moreover, the considered area onto which the analysis was carried out is not clearly declared, leading to non-comparable results (11,12). For the purpose of defining the SERS substrates homogeneity for analytical and bioanalytical applications over large area, the analysis should be extended to greater portions of the substrate and the analyzed area should be stated unequivocally. First, the intra-map homogeneity, which is described by the RSD calculated on all the spectra composing the Raman map, was calculated. In other words, it represents the variability from pixel to pixel within one single map. As long as the scanned area is enlarged, the response variability increases, leading to a higher RSD value. On the other hand, repeatability of the measurements can be obtained by increasing the spot size of analysis: in this way, local differences are averaged. In punctual confocal Raman, the spot size is a constant which depends on the section of the focalized laser on the investigated surface, which is 2.7 μm for a 780 nm laser and a 0.25 NA 10× objective. However, a spot size enlargement can be practically obtained collecting spectra on a wider area and averaging all of them. In this way, the point-to-point intensity differences can be overcome. The resulting mean spectrum is representative for the whole mapped area. The intra and inter maps RSDs of the three replicates are reported in Table 1.
The minimum area that guarantees adequate repeatability was investigated. The RSD of repeated measurements of equal areas on the same SERS substrate, i.e. the inter-maps RSD, is considered to evaluate this. For this scope, the calculation of the RSDs was performed by considering the SERS spectra acquired over three analogue samples. Three model surfaces covered by a MMC monolayer are covered with a SERS tape and measured using SERS mapping. For each sample the SERS map is repeated considering progressively increasing areas as reported in Table S4 and Figure  S13. It can be noticed that as long as the scanned area is larger, the variation of the results obtained for three repeated measurements gets lower and lower. An area of 0.25 mm 2 is demonstrated to be optimal to guarantee good measurement repeatability since 10% RSD is obtained, however adequate measurement repeatability is reached from 0.1 mm 2 on. Moreover, it can be noticed that for scanned areas higher than 0.1 mm 2 the accuracy of the results is also improved, to narrow investigated areas can lead to inaccurate intensity determinations since the investigation may be not representative enough.

Enhancement factor
To establish the values for IRaman a 0.01 M ethanol solution of MMC was poured into a well. With a 20× LWD objective and using an excitation wavelength of 780 nm and a power of 8 mW (20 s exposure time), 5.2 counts/s (IRaman) were measured at the 1595 cm -1 peak. Using the interaction volume of 68 μm 3 , the number of molecules responsible for the Raman signal (NRaman) was estimated to be 4 × 10 9 .
To establish the values for ISERS, a SERS tape coated with AuNPs was incubated in 4 ml of MMC 10 -4 M for 4 hours, then abundantly rinsed and dried. 3 Raman maps were collected on 6 different samples using identical conditions as for IRef a signal intensity of 2200 ± 400 counts/s was measured for the 1595 cm -1 peak. The number of molecules inside the laser spot was estimated by assuming a monolayer of MMC all over the AuNPs surface. Knowing the AuNPs concentration (2.7 x 10 -10 mol/l) and the diameter of one spheroidal NP (116 ± 11 nm), hypothesizing a uniform distribution of AuNPs on the surface (as proved by SEM imaging and SERS homogeneity tests) and a laser spot-size with diameter 1.9 μm an effective active area available for SERS was (2.7 ± 0.5 µm 2 ). This gives an estimate of NSERS = 4.2 × 10 6 ± 0.3 × 10 6 ). Hence the calculated enhancement factor comes to 3.4 x 10 5 ± 0.4 × 10 5 . The analytical enhancement factor was also evaluated in the "use configuration". The EF was obtained by applying the previously described equation.

EF=ISERSCRaman/IRamanCSERS
The intensity of the MMC peak at 1595 cm −1 was used from the spectra and the control test flat spectrum in Figure 3 to get the ISERS and IRaman, respectively. Since no Raman signal of the MMC was collected without the SERS tape, due to the low sensitivity of the traditional Raman in the detection of a monolayer of organic molecules, the noise of the control spectrum, intended as the standard deviation of the random oscillation of the baseline in correspondence of the non-present peak, i.e. spectral region 1750 cm −1 -1550 cm −1 , was considered at the IRaman term. Assuming that number of MMC molecules was the same for SERS and Raman measurements the terms CSERS and CRaman are removed. The average EF calculated out of three determinations is 2.43 × 10 2 with a RSD of 9.7%.