Inhibiting Analyte Theft in Surface-Enhanced Raman Spectroscopy Substrates: Subnanomolar Quantitative Drug Detection

Quantitative applications of surface-enhanced Raman spectroscopy (SERS) often rely on surface partition layers grafted to SERS substrates to collect and trap-solvated analytes that would not otherwise adsorb onto metals. Such binding layers drastically broaden the scope of analytes that can be probed. However, excess binding sites introduced by this partition layer also trap analytes outside the plasmonic “hotspots”. We show that by eliminating these binding sites, limits of detection (LODs) can effectively be lowered by more than an order of magnitude. We highlight the effectiveness of this approach by demonstrating quantitative detection of controlled drugs down to subnanomolar concentrations in aqueous media. Such LODs are low enough to screen, for example, urine at clinically relevant levels. These findings provide unique insights into the binding behavior of analytes, which are essential when designing high-performance SERS substrates.


ESI Section 1:
Following Le Ru et al, the SERS contributions as a function of analyte distribution can be extracted from numerical simulations assuming a uniform angular distribution of binding events around the spheres of a dimer, described below: Figure S1: Estimated SERS contributions from a dimer of two 80.nm nanoparticles (NPs) based on numerical simulations. a) FDTD simulation of a dimer of two 80 nm nanoparticles using a 1.nm gap showing field enhancements up to | / |=500. b) Black trace is the extracted field enhancement around the NP in a as a function of θ, blue trace is the field-enhancement trace smoothed to mitigate the effects of finite resolution in the FDTD simulation. Distribution plot in red shows the angular probability of 10 7 random positions taken on the spheres with respect to the hot-spot. c) Probability distribution of different SERS enhancement factors ( ) based on the distribution presented in b, showing that based on the FDTD model and a uniform distribution of molecules, 0.1% of the analytes already contribute 23% of the total signal, which is in line with what has previously been shown by Le Ru et al. 1 ESI Section 2: Binding effects are expected to play a significant role when the ratio of the number of binding sites to the total number of analytes available in the system ratio starts to approach 1: ratio lim �� 1, (1) with ratio being defined as and analytes and bound are calculated using the following expressions: In eq. (3-4) V is the total volume, M analytes the concentration of analytes, N A Avogadro's constant, C AuNPs the concentration of nanoparticles, r the radius of the nanoparticles, φ the packing density of binding sites per area, K occupancy a factor to describe the number of sites that are occupied by an analyte, and A binding site the area taken up by a binding site.
With a AuNP concentration of AuNPs = 2.6 • 10 13 particles per litre, a radius of = 40 nm, a packing fraction of = 0.5 and an occupancy of occupancy = 0.50, ratio reaches 1 at roughly 1.µM. This means for any lower concentrations the analyte 'theft' of the substrate has to be taken into account.
ESI Section 3: DFT calculations were performed to predict the binding energetics for each THC@CB[n], n = 5, 6, 7, and 8 complex. The Gibbs free energy of binding is calculated as: and the enthalpy gain is analogously obtained as: where ( ) denotes the counterpoise correction, and ∆ / ( ) and ∆ ( ) are the binding Gibbs free energy and enthalpy, respectively prior to the correction for the basis superposition error (BSSE). The results for each of the complexes are given in Table S1, with ( ) and ( ) denoting gas phase (vacuum) and liquid with implicit water molecules as the simulated environment respectively. a The two complexes slightly differ in the THC conformation, but have virtually the same depth of penetration into the CB[7] cavity. b Gas phase potential energy in kcal•mol −1 at the B3LYP/6-31G*+GD3BJ level of theory. c The frequency cutoff applied in Grimme's quasi-harmonic approximation was set to 100.cm  Table S1. Calculated binding Gibbs free energies and enthalpies (in kcal•mol −1 ) at the B3LYP/6-31G*+GD3BJ level of theory in SMD implicit water for each of the THC@CB[n] complexes, using the rigid-rotor harmonic oscillator approximation (RRHO), as well as the quasi-harmonic approximation (QH) following Grimme 2 where the translational entropy was also corrected.  Figure S2a). To remove the contribution of comp II to the loading plot of comp I when no analyte is present, we use: which gives the black comp I curve in Figure S3b (top), and where I and II are the loading plots of comp I and comp II respectively, and I (0%) and II (0%) the component scores for the sample with 0% analyte. The rotation performed for the comp scores follows: Hence for comp II specifically, the score corresponding to analyte concentration =0% is forced to be 0. The resulting transformed scores are shown in Figure S3c (black), with comp I nearly constant and comp II following the analyte concentration closely. Close examination of the other loading plots now reveals for comp.III a resemblance to methanol. 3 Comp IV shows a relation to the analyte concentration and shows a peak in the region of 1600.cm -1 , this we tentatively attribute to hydrogen bonding or other forms of supra-molecular interaction, as discussed in [3]. Comp V shows a double peak around 1000.cm -1 which is very similar to styrene, a polymer commonly used in laboratory consumable such as pipette tips, multi-well plates and cuvettes, and likely a source of contamination though of minor significance compared to the signal from the analytes ( Figure S2c). To normalise the PCA scores and extract a spectral response, the scores were multiplied by the loading plot normalised for power (P) and spectral integration time (T), using: = · � |comp II λ | · =λ ESI section 5: comp II and analyte identification The obtained comp.II from each analyte concentration series can be used as a fingerprinting technique to identify the types of component present in mixtures. Using a simple Pearson correlation between the raw unprocessed spectra and the comp II for each analyte vs concentration shows the analyte can be clearly distinguished above concentrations higher than 10 -8 M ( Figure S4). Figure S4: Pearson correlations between raw unprocessed spectra and different comp II spectra. The plots show each components is clearly distinguishable from its sister compounds once the analyte concentration reaches 10 -8 M or higher.
In addition to fingerprinting correlation, a comparison can be made between the comp II spectra and their respective bulk powder Raman spectra ( Figure S5), offset for clarity. Figure S5: Comparisons between comp II (SERS) and bulk powder Raman spectra for each of the synthetic cannabinoids measured. Distinct peaks from the powder spectra can be clearly recognised in the SERS spectra, while some smaller peaks have shifted or change intensity ratios. A notable difference is the increased width of the SERS lines with respect to the Raman lines resulting in peaks with narrow separations showing up as shoulders in the SERS spectra.
For the majority of the peaks from each compound a good correlation can be made between the PCA comp II (SERS) and the powder Raman spectra. There are however, a couple of notable differences with slight peaks shifts (expected in SERS) and ratio changes in peak intensities. In particular in the region 1100.cm -1 and 1250.cm -1 of analyte (4) the peak ratios for the smaller peaks differ significantly between the comp II and the Raman.
ESI Section 6: Reproducibility and noise The strength of this self-assembled SERS aggregation and measurement technique lies in its extremely high reproducibility. Small variations around 1000.cm -1 arise from trace amounts of styrene contamination leached from the plastic well plates ( Figure S6a). This allows for the background (CB[n]AuNP aggregates without analyte) to be reliably subtracted from the raw data to isolate spectral changes, see Figure 7 of the main text. The level of noise in our spectra was estimated by taking a background region of the spectrum without peaks (above 1600 cm -1 ), fitting a sigmoidal curve, and taking the residuals as noise ( Figure S6b).
To estimate the noise threshold, the standard deviation of the residuals was used giving 0.03.cts·mW  2.53±0.001 The Hill-Langmuir fits for each of the components returned a D value around 10 -7 , and for 3 out of 4 a Hill coefficient between =0.3 and 0.6. However, a clearly different Hill coefficient of =2.53 is found for compound (5) suggesting a different binding behaviour from the other compounds. This shows that understanding the difference in binding behaviour can help bring down the limits of detection. Elucidating the origin of this change in and how to modify this is subject to ongoing research.