Abstract

Quantitative detection of various molecules at very low concentrations in complex mixtures has been the main objective in many fields of science and engineering, from the detection of cancer-causing mutagens and early disease markers to environmental pollutants and bioterror agents. Moreover, technologies that can detect these analytes without external labels or modifications are extremely valuable and often preferred. In this regard, surface-enhanced Raman spectroscopy can detect molecular species in complex mixtures on the basis only of their intrinsic and unique vibrational signatures. However, the development of surface-enhanced Raman spectroscopy for this purpose has been challenging so far because of uncontrollable signal heterogeneity and poor reproducibility at low analyte concentrations. Here, as a proof of concept, we show that, using digital (nano)colloid-enhanced Raman spectroscopy, reproducible quantification of a broad range of target molecules at very low concentrations can be routinely achieved with single-molecule counting, limited only by the Poisson noise of the measurement process. As metallic colloidal nanoparticles that enhance these vibrational signatures, including hydroxylamine–reduced-silver colloids, can be fabricated at large scale under routine conditions, we anticipate that digital (nano)colloid-enhanced Raman spectroscopy will become the technology of choice for the reliable and ultrasensitive detection of various analytes, including those of great importance for human health.

Introduction and Methods

In our system, we used a quartz capillary (inner diameter 1 mm) containing 10 μl of a metallic colloid suspension to generate enhanced Raman spectra from target molecules. A scanning probe system (0.3 numerical aperture, 10× objective lens, 638 nm excitation wavelength) was used to acquire the Raman spectra from each sample in a pointwise scanning mode. Silver colloids (20–50 nm in diameter) were used for their dispersion stability during measurements and their excellent Raman enhancement abilities. The most effective colloidal concentration for these measurements was experimentally determined owing to the fact that while the number of hotspots increases with the concentration of the colloids, the background scattering also increases at the same time, leading to a reduction of detectable signals. In contrast to solid substrates, the monodispersed silver colloids exhibit homogeneous distributions throughout the entire suspension, ensuring a nearly uniform probability of the colloid–target interactions throughout the data acquisition chamber. To satisfy the criterion of independence for sequentially acquired measurements, the acquisition is performed with neighbouring voxels spaced sufficiently far apart (here, 10 μm) to minimize the possibility of double sampling. For digitization, each voxel is designated as positive (‘1’) if the signal of the specific vibrational signature is above a preset threshold that is determined by the fluctuations in the noise window, and as negative (‘0’) otherwise.

With this system, we first verified its single-molecule sensitivity and the appropriate range for single-molecule counting with the bi-analyte SERS technique using hydroxylamine-reduced silver (Hya–Ag) colloids (ζ potential: −35.6 mV) at 0.5 nM, which was found to be optimal. Hya–Ag colloids at this concentration remained monodispersed in suspension over time, ensuring a statistically uniform distribution of hotspots throughout the measurement in space and time. The acquisition time was chosen as 0.1 s or 1 s for different target molecules depending on their signal strength to balance the signal-to-noise ratio (thus the reliability of single-molecule identification) and detection efficiency (the number of qualified single-molecule events in a given period). We used the maximum intensity (Is) in the spectral window containing the specific signature for the measured molecule after baseline removal to identify the target molecules. We note that the appropriate concentration range for single-molecule counting is expected to vary over a broad range, owing to the complex nature of the target–hotspot interactions. As shown in Extended Data Fig. 4, for crystal violet molecules, single-molecule events were found to be dominant below 10−9 M, but for 4-nitrobenzenethiol (4-NBT), a concentration below 10−8 M was necessary. Therefore, the appropriate single-molecule range for an intended target molecule must be pre-determined to satisfy the requirements of digital (nano)colloid-enhanced Raman spectroscopy (dCERS).

Key Results and Conclusions

As a final proof of its validity, we applied dCERS to the detection of a fungicide, thiram, that is highly toxic and has been classified as a carcinogen category toxicant by the European Union. More recently, the European Union has advised against the use of thiram for any agricultural purposes. We first established the calibration curve of thiram in pure water from 10−9 M to 10−13 M with the Hya–Ag colloids using 1,360 cm−1 to 1,400 cm−1 as the signature window. Because our main objective is to measure the residual thiram in agricultural products, we used extracts from laboratory-grown bean sprouts as a model system. The sprouts were homogenized, centrifuged and membrane-filtered to produce the extracts. Thiram was then dissolved in the extract at a concentration of 10−9 M (equivalent to 2.4 × 10−4 mg kg−1). Similar to paraquat, measurements of a serial dilution were performed, and the results are shown in Fig. 3d. Once again, significantly lower values were found at the original concentration (10−9 M), most likely also owing to the competition of the materials retained in the extract. But on further dilution, the same linear relationship as the calibration was recovered at concentrations below 10−10 M. Even at 10−13 M, which is about five orders of magnitude lower than the detection limit of liquid chromatography-tandem mass spectrometry, dCERS measurements are robust and reliable, achieving an error of 25% with only 2,400 total voxels acquired. These two examples demonstrate the unparalleled advantage of dCERS in detecting very low concentrations of soluble target molecules at accuracies that can be directly estimated on the basis of accumulated positive counts. Moreover, these experiments also indicate that measurements with serial dilutions are required to sufficiently reduce any influence from unknown background components to enable correspondence with the calibrations. As such, this procedure requires the colloidal system to be consistent and reproducible, which is a substantial challenge with solid chip-based systems.

In conclusion, we have demonstrated that dCERS based on single-molecule counting is an effective and efficient technology for molecular quantifications at ultralow concentrations solely on the basis of intrinsic vibrational signatures of the target molecules as long as they exhibit well-defined, specific spectral signatures as well as adequate interaction with the hotspots. As demonstrated by the proof-of-concept examples with the toxic compounds (thiram and paraquat), the unprecedented sensitivity and predictable accuracy make dCERS a powerful and perhaps even preferred technology for demanding applications, including environmental protection, food safety and mutagen detection, in which the presence of even a trace amount of certain chemicals could already be a serious threat to human health.

Fig. 1: The concept of dCERS.

a, Schematic of the experiments. The SERS spectra are collected with a quartz capillary containing a suspension of metallic colloidal nanoparticles and target molecules. Each acquisition voxel is assigned as ‘1’ (positive) or ‘0’ (negative) according to whether a single-molecule event is present. b, Typical dCERS results obtained with crystal violet. The standard (STD) spectra (top) as well as several examples of single-molecule spectra assigned as positive (i–iii) or negative (iv–vi) are shown. The background (BG) reference spectrum from the null control (bottom) is also shown. The shaded region in these images reflects the signature window for this target molecule. c, dCERS calibration curve for crystal violet. d, Typical dCERS results for single-stranded DNA (12 nt poly A; A12). The spectra shown are as described in b. e, dCERS calibration curve for A12. f–h, The dCERS curves for 4-nitrobenzenethiol (4-NBT) (f), d-glucose (g) and haemoglobin (h) are also shown. RPV, ratio of positive voxels. Error bars were calculated from three independent measurements. Hya–Ag colloidal nanoparticles (about 24 nm in diameter) at 0.5 nM were used for all these measurements. a.u., arbitrary units.

Fig. 2: Reproducibility of dCERS.

a, The dependency of the mean number of positive counts (grey bars) on the concentration of crystal violet under a fixed number of measured voxels per dCERS test (5,400 voxels). The error bars indicate the standard deviations (n = 3). The RSDs (red dots) and the expectation based on Poisson statistics (blue curve) are also shown. b, Dependency of the mean number of positive voxels on the voxel number at 10−10 M crystal violet (n = 3). The RSD and expectation based on Poisson statistics as in a are also shown.

Fig. 3: Quantitative detection of trace amount of chemicals by dCERS.

a,b, Quantification of crystal violet and Nile blue when mixed at three different concentration proportions (denoted by sample numbers 1, 2 and 3). Typical single-molecule spectra (a) corresponding to crystal violet (i–iii) (red, ‘0’ for Nile blue and ‘1’ for crystal violet) and Nile blue (iv–vi) (blue, ‘1’ for Nile blue and ‘0’ for crystal violet) in a mixed sample. The characteristic peaks (shaded with corresponding colour) for crystal violet and Nile blue are identified and used for quantification. RPV versus the concentrations of crystal violet (CV) (top) and Nile blue (NB) (bottom) in the three samples (b), fitted by a linear equation on the log–log scale. The total acquisition voxels were 1,200 for each measurement. c, When the highly toxic herbicide paraquat was added to normal lake water, 10−8 M (red dots) paraquat were readily detected with 5-fold, 10-fold and 100-fold dilution. However, there was notable competition at the original concentration, and the same linear relationship was recovered only on sufficient dilution (compared with calibration denoted by black dots). Inset, chemical structure of paraquat. d, With fungicide thiram dissolved in the bean sprouts extract (10−9 M), dCERS was also effective even when diluted by a 1,000-fold. However, similar to the case with lake water, the competition owing to the residual materials was significant at the original concentration. But accurate measurement was achieved on further dilution. Inset, chemical structure of thiram. All the error bars indicate the standard deviations (n = 3). NP, nanoparticle.

https://www.nature.com/articles/s41586-024-07218-1