Skip to main content
NCBI home page
ACS AuthorChoice logoLink to ACS AuthorChoice
. 2022 Nov 30;14(49):54550–54557. doi: 10.1021/acsami.2c16446

Highly Adsorptive Au-TiO2 Nanocomposites for the SERS Face Mask Allow the Machine-Learning-Based Quantitative Assay of SARS-CoV-2 in Artificial Breath Aerosols

Charles S H Hwang †,∥, Sangyeon Lee †, Sejin Lee †,∥, Hanjin Kim †, Taejoon Kang ‡,§, Doheon Lee †,*, Ki-Hun Jeong †,∥,*
PMCID: PMC9718102  PMID: 36448483

Abstract

graphic file with name am2c16446_0006.jpg

Human respiratory aerosols contain diverse potential biomarkers for early disease diagnosis. Here, we report the direct and label-free detection of SARS-CoV-2 in respiratory aerosols using a highly adsorptive Au-TiO2 nanocomposite SERS face mask and an ablation-assisted autoencoder. The Au-TiO2 SERS face mask continuously preconcentrates and efficiently captures the oronasal aerosols, which substantially enhances the SERS signal intensities by 47% compared to simple Au nanoislands. The ultrasensitive Au-TiO2 nanocomposites also demonstrate the successful detection of SARS-CoV-2 spike proteins in artificial respiratory aerosols at a 100 pM concentration level. The deep learning-based autoencoder, followed by the partial ablation of nondiscriminant SERS features of spike proteins, allows a quantitative assay of the 101–104 pfu/mL SARS-CoV-2 lysates (comparable to 19–29 PCR cyclic threshold from COVID-19 patients) in aerosols with an accuracy of over 98%. The Au-TiO2 SERS face mask provides a platform for breath biopsy for the detection of various biomarkers in respiratory aerosols.

Keywords: SARS-CoV-2, surface-enhanced Raman spectroscopy, breath biopsy, machine-learning, plasmonics, nanocomposite

Introduction

Respiratory airborne particulates from human breath contain vital information on the current health status of an individual.1,2 Different forms of such particulates, i.e., aerosols and volatile organic compounds (VOCs), contain potential biomarkers for a wide spectrum of diseases including viruses, asthma, cancers, and neurodegenerative disorders.3−8 In particular, respiratory aerosols from human breath have attracted significant interest with the recent global outbreak of the 2019 coronavirus disease (COVID-19) pandemic caused by severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2).9−11 With the first report in the late December of 2019 in Wuhan, China, there are over 500 million confirmed cases globally as of April 2022.12 Both the World Health Organization and the United States Centers for Disease Control and Prevention have confirmed that COVID-19 is principally transmitted through respiratory aerosols.12,13 Combined with the high proportion of asymptomatic patients reaching up to 40% of the total confirmed patients, the reproduction number of COVID-19 has been reported to be 4–20 times greater than that of seasonal flu.14 However, the current gold standard for the diagnosis of such large-scale COVID-19 patients is still limited to nucleic acid detection by amplification,15,16 where the examination process takes up to several hours under the supervision of highly qualified medical personnel. COVID-19 face masks have recently been demonstrated that diagnose the collected viruses on a cellulose matrix;17,18 yet, these also require time-consuming post-extraction processes for a polymerase chain reaction. In contrast, a comprehensive and direct examination of breath aerosols has a high potential for the rapid diagnosis of COVID-19 but has been relatively unexplored for such diagnostic applications.

Recently, surface-enhanced Raman spectroscopy (SERS) has demonstrated the detection and analysis of aerosols for diverse air monitoring applications by utilizing the electromagnetic “hotspots” of plasmonic nanostructures.4,19−22 Unlike conventional aerosol detection methods such as physical impactor and evaporative light scattering detection, SERS serves as a cost-effective alternative for rapid, sensitive, and quantitative analysis of chemical fingerprints within the aerosols. However, plasmonic SERS substrates often show some technical limitations in detecting respiratory breath aerosols due to a low Weber number of an aerosol, i.e., the ratio of the aerosol’s momentum to the surface tension.23 In other words, the fast emission velocity of such aerosols (2–10 m/s)24 substantially deters effective adsorption onto a substrate. While conventional SERS substrates with assorted nanostructures25−29 and nanoparticles30,31 display highly packed geometries for enhanced electromagnetic hotspots, they often result in low surface energy that is inefficient for aerosol adsorption. As a result, hotspot-rich SERS substrates with high surface energies are still in need of efficient capture and facile preconcentration of respiratory aerosols.

Results and Discussion

Here, we report a SERS face mask for the label-free detection of the aerosolized SARS-COV-2 virus using gold-titanium dioxide (Au-TiO2) nanocomposites. The face mask features a highly adsorptive Au-TiO2 nanocomposite SERS chip on the inner cloth of a filtering facepiece respirator (Figure 1a). Human breath aerosols are oronasally released during respiratory action such as breathing, coughing, sneezing, or speaking and are continuously preconcentrated onto the SERS chip inside the respirator. Au-TiO2 nanoislands on a quartz substrate exhibit high aerosol adsorption and dense electromagnetic hotspots, allowing the rapid, facile, and quantitative SERS detection of the SARS-CoV-2 virus (Figure 1b). An autoencoder neural network is further employed for the accurate classification of the SARS-CoV-2 virus at various concentrations. The aerodynamic behavior of aerosols impacting a surface mainly depends on the surface energy (Figure 1c). For instance, respiratory aerosols with an average velocity of 2–10 m/s24 simply rebound upon impact with conventional Au nanoislands due to low surface energy.23 In contrast, an ultrathin film of TiO2 nanoclustered in the Volmer–Weber mode on the Au nanoislands significantly increases the surface energy (contact angle, θ = 20–30°), allowing the efficient adsorption of the respiratory aerosols. The Au-TiO2 SERS chip with an active sensing area of 28 mm2 was diced and attached to a commercial KF94 respirator with a round medical plaster bandage (Figure 1d).

Figure 1.

Figure 1

Open in a new tab

Schematic illustration of SARS-CoV-2 detection from respiratory breath aerosols using the Au-TiO2 SERS chip on a face mask. (a) Oronasal aerosols emitted upon respiratory actions are easily and directly collected onto the SERS chip. (b) Machine-learned SERS detection of SARS-CoV-2 aerosols. The measured SERS signals are quantitatively classified using an ablation-assisted autoencoder. (c) Aerodynamic behavior of aerosols impacting on a solid substrate with different surface energies. Au-TiO2 nanoislands with high surface energies efficiently adsorb the low-volume and high-velocity respiratory aerosols. (d) Optical image of the SERS face mask as an application example. The Au-TiO2 SERS chip was attached to the inner cloth of a KF94 mask with a medical plaster.

The enhanced surface energy of the Au-TiO2 nanocomposites is crucial for the highly sensitive SERS detection of respiratory aerosols. The nanofabrication of Au-TiO2 nanocomposites exploits a facile two-step process including repeated thermal dewetting of Au thin films and thermal evaporation of TiO2 (Figure 2a). The hotspot-rich Au nanoislands were fabricated at repeated dewetting of 6 + 6 nm thick Au thin films, showing a 21% increase in average size, 72% decrease in the interparticle distance, and 5-fold E-field enhancement (Figure. S1), compared to single dewetting of a 12 nm thick Au thin film. See the Methods section for the repeated thermal dewetting of Au nanoislands, geometric characterization of Au nanoislands, and FDTD calculation of E-field enhancements. A 2 nm thick TiO2 film was evaporated in the Volmer–Weber mode onto the as-fabricated Au nanoislands to form Au-TiO2 nanocomposites. The geometry and the composition of Au-TiO2 nanocomposites were characterized using a field-emission transmission electron microscope (300 keV FE-TEM, Tecnai G2 F30 S-TWIN, FEI) equipped with energy-dispersive X-ray (EDX) (Figures 2b and S2). The ultrathin TiO2 film forms nanoclusters on Au nanoislands, while TiO2 of larger thickness leads to uniform thin-film formation.32,33 The contact angle of a water droplet was then measured to compare the surface energy of Au-TiO2 substrates with different TiO2 thicknesses ranging from 0, 2, 4, 6, 8, and 10 nm (Figure 2c). While simple Au nanoislands without TiO2 show a contact angle of 60°, even a slight addition of 2 nm thick TiO2 significantly reduces the contact angle to 29°. The measured SERS signals of nebulized 10–7 M rhodamine 6 g (R6G) also show that the Au-TiO2 substrate efficiently captures aerosols from the increased surface energy. The E-field intensities of the Au-TiO2 nanocomposites were calculated using the FDTD method and compared with the observed SERS peak intensity of nebulized R6G aerosols at 1360 cm–1 (Figures 2d and S3). See the Methods section for the FDTD calculation of E-field enhancements. The relative E-field intensity ratio (E4/EAu4) rapidly decays to 0.67, 0.49, 0.43, 0.39, and 0.36 for increasing TiO2 thickness of 2, 4, 6, 8, and 10 nm because the TiO2 thin films behave as a dielectric gap spacer.34 However, the SERS intensity ratios at 1360 cm–1 are observed as 1.47, 0.81, 0.74, 0.58, and 0.51 with respect to that of simple Au. This phenomenon is explained by the highly adsorptive Au-TiO2 substrate efficiently capturing more aerosols that compensates for the SERS intensity decay. In particular, the 2 nm TiO2 thin film evaporated as nanoclusters on Au allows direct contact of the target analytes in the aerosol to the plasmonic hotspots, exhibiting substantially enhanced SERS intensity. The preconcentration of aerosols on the face mask was further investigated by measuring the SERS signals of 10–7 M R6G for increasing nebulizing times (Figure 2e). The low-volume aerosols dry quickly, allowing for the continuous accumulation of target analytes within the aerosols. The SERS intensity at 1360 cm–1 linearly increases as the preconcentration time of the aerosols changes from 5, 10, 15, and 20 s. Note that the collection of aerosols nebulized during five seconds is equivalent to a four-hour preconcentration of aerosols from respiration.35

Figure 2.

Figure 2

Open in a new tab

Highly adsorptive Au-TiO2 nanocomposites substrate for aerosol preconcentration and enhanced SERS signals. (a) Nanofabrication process employs repeated solid-state dewetting of Au thin films for high-density EM hotspot generation. A thin layer of TiO2 is then thermally evaporated to obtain Au-TiO2 nanocomposites. (b) FE-TEM images of the fabricated Au-TiO2 nanoislands for 2 nm (left) and 10 nm (right) thin films of TiO2. (c) Contact angles for various thicknesses of TiO2 on Au nanoislands. The 2 nm thick TiO2 thin film substantially increases the surface energy. (d) Comparison of measured SERS intensity of R6G aerosols and calculated E-field intensities for different thicknesses of TiO2. The Au-TiO2 substrate with 2 nm TiO2 exhibits 47% increased SERS intensity of R6G aerosols at 1360 cm–1 due to highly adsorptive surfaces efficiently adsorbing nebulized R6G aerosols. (e) Preconcentration of 100 nM R6G aerosols using a nebulizer. The SERS intensity of R6G at 1360 cm–1 linearly increases with respect to the preconcentration times.

Characteristic SERS peaks of SARS-CoV-2 spike proteins in artificial respiratory aerosols were characterized using the Au-TiO2 substrates. The SARS-CoV-2 is surrounded by protruding spike proteins, which serve as a primary biomarker for immunoassays in the receptor recognition and membrane fusion process.36 Several characteristic SERS peaks in 1 μM SARS-CoV-2 spike proteins are observed including CH2 rocking of phenylalanine at 651, 772, and 996 cm–1, CH2 rocking of tryptophan at 852 cm–1, NH3 rocking of histidine at 1177 cm–1, CH2 wagging of L-arginine at 1267 cm–1, C-N stretching and amide III band at 1372 cm–1, and CH2 deformation and NH bending of tryptophan and phenylalanine at 1523 and 1587 cm–1 (Figure 3a).37 Note that the SERS signals of SARS-CoV-2 spike proteins show high selectivity to other viral proteins.38,39 Also see Figure S4 for SERS signals of the spike proteins with different concentrations ranging from 1 μM to 100 pM. Next, the SARS-CoV-2 spike proteins with concentrations ranging from 100 nM to 100 pM were mixed in an artificial respiratory solution and nebulized onto the SERS substrate for the emulation of the human oronasal emission. The chemical constituents,40 concentrations, and observed SERS peaks of the artificial respiratory aerosols (ARA) are respectively summarized in Figures 3b and S5. The ARA samples were prepared by adding spike proteins of different concentrations to the stock artificial respiratory solution. See Methods for the emulation of oronasal aerosols and artificial respiratory solutions. Strong SERS bands are observed in ARA at the 1400–1600 cm–1 range, resulting from various chemicals such as potassium citrate, lactic acid urea, etc. As a result, the SERS signals of the spike protein in ARA (Figure 3c) are partially different from that of Figure 3a, particularly at the 1400–1600 cm–1 range, due to signal interference and different adsorption kinetics in complex mixtures caused by ARA.41 The characteristic SERS peaks of the SARS-CoV-2 spike proteins in Figure 3a were then quantitatively compared by calculating the SERS peak intensity ratios depending on the spike protein concentration, with respect to ARA as the reference (Figure 3d). While the signal intensity ratio of nondiscriminant peaks (1523 and 1587 cm–1) remains relatively constant, all other characteristic peaks are highly discriminant and increase for a higher concentration of the SARS-CoV-2 spike protein in ARA.

Figure 3.

Figure 3

Open in a new tab

SERS measurements of SARS-CoV-2 spike proteins in artificial respiratory aerosols. (a) SERS signals of SARS-CoV-2 spike protein aerosols adsorbed on Au-TiO2 nanocomposite substrates. The characteristic Raman peaks of the spike proteins are clearly observable at 651, 772, 853, 927, 996, 1177, 1267, 1372, 1523, and 1587 cm–1. (b) Composition of artificial respiratory aerosols (ARA) and measured SERS peaks. Raman bands of the spike proteins at 1523 and 1587 cm–1 overlap with those of artificial respiratory aerosols. (c) SERS signals at various concentrations of the spike proteins in ARA. (d) SERS signal intensity ratios at various characteristic Raman bands of the spike proteins were calculated using ARA as the reference. Some Raman bands, i.e., 1523 and 1587 cm–1 Raman bands, are shadowed by the reference signals, while most others provide quantitative fingerprints of SARS-CoV-2 spike proteins in ARA.

An ablation-assisted autoencoder with a logistic regression model was further employed for a quantitative assay of SARS-Cov-2 lysate in aerosols. The autoencoder algorithm allows characteristic feature extraction by reducing the dimensions of input signals into low-dimensional features in the latent space.42,43 The autoencoder-based dimensionality reduction model was trained in a supervised manner to minimize the total loss function including latent loss, interclass, intraclass, and distances from centroids44 (Figure 4a). See the Methods section for training the autoencoder model. The initial SERS signals of SARS-CoV-2 lysate aerosols in artificial respiratory solutions were acquired for different concentrations ranging from 0, 101, 102, 103, and 104 pfu/mL at five different positions on the Au-TiO2 substrates (Figures 4a and S6). The SERS signals mapped onto the two-dimensional (2D) latent space via the autoencoder reveal that the individual classes are well-clustered and diagonally aligned depending on the lysate concentration (Figure 4b). This allows the logistic regression model to be used as a prediction model for the highly accurate quantitative assay of target molecules in the complex solution, i.e., artificial respiratory aerosols, that supersede the conventional chemometrics such as principal component analysis. The ablation of the nondiscriminant Raman features of spike proteins in ARA further substantially increases the classification accuracy of the prediction model (Figure 4c). The two nondiscriminant SERS features of the spike proteins in ARA, i.e., 1532 and 1587 cm–1 bands, were ablated based on Figure 3. Data sets with the random ablation of the same length of vectors were also generated for comparison. Each of the data sets with nondiscriminatory ablation, random ablation, and without ablation was then trained using the autoencoder-based dimensionality reduction model, followed by logistic regression models. The receiver operating characteristic (ROC) curve explicitly demonstrates that the ablation of the 1532 and 1587 cm–1 SERS peaks provides 7.6% higher classification accuracy compared to nonablated SERS signals. The confusion matrix of the ablation-assisted autoencoder model indicates that the SARS-CoV-2 lysates adsorbed on the Au-TiO2 nanocomposite SERS substrate exhibit over 98% classification accuracy for concentrations ranging from 101–104 pfu/mL (Figure 4d). Note that typical COVID-19 patients demonstrate an average cyclic threshold value of 27 during PCR testing,45 which corresponds to 102–103 pfu/mL concentration of the SARS-CoV-2 virus.46 As a result, the highly accurate assay of the SARS-CoV-2 lysates down to 101 pfu/mL infers the successful diagnosis of COVID-19 from respiratory aerosols.

Figure 4.

Figure 4

Open in a new tab

Deep learning-based quantitative SERS assay for label-free SARS-CoV-2 lysates in ARA. (a) Ablation-assisted autoencoding. The nasopharyngeal SARS-CoV-2 lysates were nebulized with ARA for initial SERS measurements, followed by the ablation of the low-importance SERS features of spike proteins in ARA. Then, an autoencoder was used to quantitatively classify the SERS signals of the viral lysates (101–104 pfu/mL) in ARA. The 2D latent space scores were optimized with hyperparameters including latent loss (L0), intraclass (L1), interclass (L2), and distances from centroids (L3). (b) Two-dimensional latent scores of SERS signals of SARS-CoV-2 lysates with different concentrations. The classified clusters show a strong diagonal alignment with SARS-CoV-2 concentrations. (c) Area under the receiver operating characteristic curve (ROC) for the ablation-assisted autoencoder algorithm. (d) Confusion matrix of the autoencoder algorithm exhibiting the average prediction accuracy of over 98% from the label-free SERS detection of SARS-CoV-2 lysates.

Conclusions

To conclude, this work has successfully demonstrated a highly adsorptive Au-TiO2 nanocomposite-based SERS face mask for the SERS detection of SARS-CoV-2 in artificial breath aerosols. Nanoclustered TiO2 on Au nanoislands efficiently adsorbs respiratory aerosols and shows 47% enhanced SERS signals compared to conventional Au nanoislands. The SERS chip inside a face mask further exhibits preconcentration of low-volume respiratory aerosols. The autoencoder prediction model with a modified loss function demonstrates the quantitative assay of SARS-CoV-2 lysates with 98% accuracy by ablating nondiscriminant SERS peaks of spike proteins. This Au-TiO2 SERS face mask provides a rapid, robust, and facile screening method for the pre-emptive diagnosis of COVID-19 and a biosensing platform for breath biopsy, detecting various disease biomarkers in respiratory aerosols.

Methods

Repeated Thermal Dewetting of Au Nanoislands

First, a 6 nm thick Au thin film was thermally evaporated onto a 4-inch quartz wafer at a constant rate of 0.5 Å/s. The Au thin film was then thermally dewetted inside a box furnace (Lindberg/Blue M, Moldatherm Box Furnace) for one hour at 700 °C to form large-area Au nanoislands with high adhesion to glass. The target temperature was steadily increased from the room temperature with ramp-up and ramp-down rates of 20 and 5 °C/min, respectively. The above procedure was repeated once more to form closely packed and enlarged Au nanoislands for highly uniform and substantially increased SERS signals.47

Geometric Characterization of Au Nanoislands

The SEM images were directly converted into binary images. The radius and interparticle distances of the Au nanoislands after single and repeated dewetting were calculated with ImageJ software, assuming that Au nanoislands are periodic arrays with uniform sizes.

FDTD Calculation of E-Field Intensities

The three-dimensional finite-difference time-domain (FDTD) simulation was performed to calculate the E-field enhancement of the nanoislands by directly importing the two-dimensional SEM images. The Au nanoislands were assumed to have a thickness of 58 nm on the dielectric surface (n = 1.4) with a uniform thin-film TiO2 layer on top of the nanoislands with respect to the target thickness. The TiO2 layer was disabled for the E-field comparison between single and repeated thermal dewetting. The E-field intensity was monitored at a 633 nm wavelength region positioned at 20 nm above the substrate. The average E-field intensity for the region of interest was utilized for the calculation of the relative E-field intensity (Enanocomposite4/EAu). (Release: 2019a r6, Version 8.21.1933)

Emulation of Oronasal Aerosols and Artificial Respiratory Solutions

The oronasal emission of human respiratory aerosols was emulated using a commercial nebulizer (Teledyne CETAC Technologies) (Figure 2d). The carrier gas at a 200 sccm flow rate was controlled using a mass flow controller (MKS Instruments), while the Au-TiO2 nanocomposite SERS substrate was positioned 25 cm away from the nebulizer’s end in a fume hood. The average size of the nebulized aerosol is <10 μm. The nebulizing time was 10 seconds. The artificial respiratory solution was prepared to emulate respiratory aerosols from human breath. SARS-CoV-2 spike proteins with respective concentrations were diluted in a stock solution with final concentrations of 27.3 μM NaCl, 4.1 μM NH4NO3, 4.6 μM KH2PO4, 1.6 μM KCl, 1.0 μM C6H5K3O7, 0.1 μM C5H4N4O3, 3.3 μM CH4N2O, and 1.2 μM C3H6O3 in deionized water.

SARS-CoV-2 Lysate Preparation

The SARS-CoV-2 (BetaCoV/Korea/KCDC03/2020) was provided by the National Culture Collection for pathogens, which is operated by the Korea National Institute of Health. The virus was carefully cultured in a biosafety level 3 laboratory at the Korea Research Institute of Bioscience and Biotechnology (KRIBB) and heated with the lysis buffer.48 The stock SARS-CoV-2 lysates were then diluted to concentrations ranging from 101–104 pfu/mL and mixed in artificial respiratory solution prior to nebulizing. The control sample in Figure 4 refers to an artificial respiratory solution mixed with the culture media and lysis buffer without any viral lysates.

SERS Signal Measurement of Nebulized R6G Aerosols

The 10–7 M R6G was nebulized for 10 s, and the SERS signals were measured using a benchtop spectrometer equipped with a CCD camera and 50× objective lens (MicroSpec 2300i, Princeton Instruments) under excitation of 5 mW 633 nm HeNe laser. The acquisition time was 1 s.

Training the Autoencoder Model

First, five SERS spectra measured at different positions on the SERS substrate were obtained for different concentrations of aerosolized SARS-CoV-2 lysates in an artificial respiratory solution (control, 101, 102, 103, 104 pfu/mL). The measured SERS spectra with a vector length of 1096, containing the signal intensities for the Raman shift range of 600–1580 cm–1, were used for the quantitative analysis of SARS-CoV-2 lysates. The number of measured SERS data was augmented to improve the performance and the robustness of the prediction model and to prevent overfitting by adding random noise and applying the synthetic minority oversampling technique (SMOTE) algorithm provided by the Python package.49 Twenty synthetic data with random noise were generated from five measured SERS data sets, followed by generating 80 additional synthetic data from the SMOTE algorithm with a shrinkage value of 1.8 for each concentration class of SARS-CoV-2 lysates. The amplitude of noise for a wavenumber was randomly determined between ±1/5 of the original SERS value. One hundred vectors were acquired for each SARS-CoV-2 lysate concentration. The total 500 vectors, resulting from five different concentrations, were finally utilized for the autoencoder training. The model consists of a symmetric encoder and decoder with 1069 input nodes and two fully-connected layers, each consisting of 256 and 36 nodes and two latent nodes. The batch size and epoch were manually set as 32 and 40, respectively, with a 1e–4 learning rate and 1e–4 weight decay. Other hyperparameters for the linearizing autoencoder, such as weights of loss, are heuristically determined through the experiment. An entirely different data set of 500 test data was generated for validation and evaluation after the training—475 synthetic vectors were generated using the aforementioned protocol of generating synthetic training data and combined with 25 measured SERS data. Fifty data points on both sides of the nondiscriminant SERS bands of the SARS-CoV-2 spike were set to zero during the training of the ablation-assisted autoencoder. The same number of wavenumbers at random points was set to zero as a control set. The concentrations of the SARS-CoV-2 lysate data were classified by the logistic regression of the two-dimensional latent features from the autoencoder. The ROC curve is calculated during the classification and macro-averaged for every label.

Acknowledgments

This work was supported by ETRI (Electronics and Telecommunications Research Institute) internal funds (21YR2500), the Korea Medical Device Development fund grant funded by the Korean government (KMDF_PR_20200901_0074), the National Research Foundation of Korea (NRF) grant funded by the Korean government (2021R1A2B5B03002428, 2021M3E5E3080379), the Bio-Synergy Research Project (NRF-2012M3A9C4048758) of the Ministry of Science and ICT, and the KRIBB Research Initiative Program (1711134081).

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsami.2c16446.

  • SERS EF comparison between single and repeatedly dewetted Au nanoislands (Figure S1); EDX elemental analysis of Au-TiO2 nanocomposites (Figure S2); SERS intensity of nebulized R6G aerosols on Au-TiO2 nanocomposites (Figure S3); SERS spectrum of SARS-CoV-2 spike proteins for various concentrations (Figure S4); SERS spectra of artificial respiratory aerosols (Figure S5); and SERS spectra of nebulized SARS-CoV-2 lysates in artificial respiratory aerosols for concentrations ranging from 0 to 104 pfu/mL (Figure S6). The data that support the findings of this study are available upon reasonable request (PDF)

Author Contributions

⊥ C.S.H.H. and S.L. contributed equally to this work. The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript.

This work was supported by the Electronics and Telecommunications Research Institute, the Korea Research Institute of Bioscience & Biotechnology, the Korea Medical Device Development, and the National Research Foundation of Korea.

The authors declare no competing financial interest.

Supplementary Material

am2c16446_si_001.pdf (654.3KB, pdf)

References

  1. Pereira J.; Porto-Figueira P.; Cavaco C.; Taunk K.; Rapole S.; Dhakne R.; Nagarajaram H.; Camara J. S. Breath Analysis as a Potential and Non-Invasive Frontier in Disease Diagnosis: An Overview. Metabolites 2015, 5, 3–55. 10.3390/metabo5010003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Chan L. W.; Anahtar M. N.; Ong T. H.; Hern K. E.; Kunz R. R.; Bhatia S. N. Engineering Synthetic Breath Biomarkers for Respiratory Disease. Nat. Nanotechnol. 2020, 15, 792–800. 10.1038/s41565-020-0723-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Kharitonov S. A.; Barnes P. J. Biomarkers of Some Pulmonary Diseases in Exhaled Breath. Biomarkers 2002, 7, 1–32. 10.1080/13547500110104233. [DOI] [PubMed] [Google Scholar]
  4. Chen Y.; Zhang Y.; Pan F.; Liu J.; Wang K.; Zhang C.; Cheng S.; Lu L.; Zhang W.; Zhang Z.; Zhi X.; Zhang Q.; Alfranca G.; de la Fuente J. M.; Chen D.; Cui D. Breath Analysis Based on Surface-Enhanced Raman Scattering Sensors Distinguishes Early and Advanced Gastric Cancer Patients from Healthy Persons. ACS Nano 2016, 10, 8169–8179. 10.1021/acsnano.6b01441. [DOI] [PubMed] [Google Scholar]
  5. Pleil J. D. Breath Biomarkers in Toxicology. Arch. Toxicol. 2016, 90, 2669–2682. 10.1007/s00204-016-1817-5. [DOI] [PubMed] [Google Scholar]
  6. Amal H.; Ding L.; Liu B. B.; Tisch U.; Xu Z. Q.; Shi D. Y.; Zhao Y.; Chen J.; Sun R. X.; Liu H.; Ye S. L.; Tang Z. Y.; Haick H. The Scent Fingerprint of Hepatocarcinoma: In-Vitro Metastasis Prediction with Volatile Organic Compounds (VOCs). Int. J. Nanomed. 2012, 7, 4135–4146. 10.2147/IJN.S32680. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Shirasu M.; Touhara K. The Scent of Disease: Volatile Organic Compounds of The Human Body Related to Disease and Disorder. J. Biochem. 2011, 150, 257–266. 10.1093/jb/mvr090. [DOI] [PubMed] [Google Scholar]
  8. Mazzatenta A.; Pokorski M.; Sartucci F.; Domenici L.; Di Giulio C. Volatile Organic Compounds (Vocs) Fingerprint of Alzheimer’s Disease. Respir. Physiol. Neurobiol. 2015, 209, 81–84. 10.1016/j.resp.2014.10.001. [DOI] [PubMed] [Google Scholar]
  9. van Doremalen N.; Bushmaker T.; Morris D. H.; Holbrook M. G.; Gamble A.; Williamson B. N.; Tamin A.; Harcourt J. L.; Thornburg N. J.; Gerber S. I.; Lloyd-Smith J. O.; de Wit E.; Munster V. J. Aerosol and Surface Stability of SARS-CoV-2 as Compared with SARS-CoV-1. N. Engl. J. Med. 2020, 382, 1564–1567. 10.1056/NEJMc2004973. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Tang S.; Mao Y.; Jones R. M.; Tan Q.; Ji J. S.; Li N.; Shen J.; Lv Y.; Pan L.; Ding P.; Wang X.; Wang Y.; MacIntyre C. R.; Shi X. Aerosol Transmission of SARS-Cov-2? Evidence, Prevention and Control. Environ. Int. 2020, 144, 106039 10.1016/j.envint.2020.106039. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Lamote K.; Janssens E.; Schillebeeckx E.; Lapperre T. S.; De Winter B. Y.; van Meerbeeck J. P. The Scent of COVID-19: Viral (Semi-)Volatiles As Fast Diagnostic Biomarkers?. J. Breath Res. 2020, 14, 042001 10.1088/1752-7163/aba105. [DOI] [PubMed] [Google Scholar]
  12. WHO . https://covid19.who.int/.
  13. CDC https://www.cdc.gov/.
  14. Ito K.; Piantham C.; Nishiura H. Relative instantaneous reproduction number of Omicron SARS-CoV-2 variant with respect to the Delta variant in Denmark. J. Med. Virol. 2022, 94, 2265–2268. 10.1002/jmv.27560. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Corman V. M.; Landt O.; Kaiser M.; Molenkamp R.; Meijer A.; Chu D. K.; Bleicker T.; Brunink S.; Schneider J.; Schmidt M. L.; Mulders D. G.; Haagmans B. L.; van der Veer B.; van den Brink S.; Wijsman L.; Goderski G.; Romette J. L.; Ellis J.; Zambon M.; Peiris M.; Goossens H.; Reusken C.; Koopmans M. P.; Drosten C. Detection of 2019 Novel Coronavirus (2019-nCoV) by Real-time RT-PCR. Eurosurveillance 2020, 25, 2000045 10.2807/1560-7917.ES.2020.25.3.2000045. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Tahamtan A.; Ardebili A. Real-time RT-PCR in COVID-19 Detection: Issues Affecting the Results. Expert Rev. Mol. Diagn. 2020, 20, 453–454. 10.1080/14737159.2020.1757437. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Kanaujia R.; Biswal M.; Angrup A.; Ray P. Inhale, Then Exhale: Start Afresh to Diagnose Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-Cov-2) by Non-Invasive Face-Mask Sampling Technique. Clin. Microbiol. Infect. 2020, 26, 1701–1702. 10.1016/j.cmi.2020.06.034. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Daniels J.; Wadekar S.; DeCubellis K.; Jackson G. W.; Chiu A. S.; Pagneux Q.; Saada H.; Engelmann I.; Ogiez J.; Loze-Warot D.; et al. A Mask-based Diagnostic Platform for Point-Of-Care Screening of Covid-19. Biosens. Bioelectron. 2021, 192, 113486 10.1016/j.bios.2021.113486. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Dong X.; Ohnoutek L.; Yang Y.; Feng Y.; Wang T.; Tahir M. A.; Valev V. K.; Zhang L. Cu/Ag Sphere Segment Void Array as Efficient Surface Enhanced Raman Spectroscopy Substrate for Detecting Individual Atmospheric Aerosol. Anal. Chem. 2019, 91, 13647–13657. 10.1021/acs.analchem.9b02840. [DOI] [PubMed] [Google Scholar]
  20. Craig R. L.; Bondy A. L.; Ault A. P. Surface Enhanced Raman Spectroscopy Enables Observations of Previously Undetectable Secondary Organic Aerosol Components at the Individual Particle Level. Anal. Chem. 2015, 87, 7510–7514. 10.1021/acs.analchem.5b01507. [DOI] [PubMed] [Google Scholar]
  21. Sivaprakasam V.; Hart M. B.; Eversole J. D. Surface Enhanced Raman Spectroscopy of Individual Suspended Aerosol Particles. J. Phys. Chem. C 2017, 121, 22326–22334. 10.1021/acs.jpcc.7b05310. [DOI] [Google Scholar]
  22. Phan-Quang G. C.; Yang N.; Lee H. K.; Sim H. Y. F.; Koh C. S. L.; Kao Y. C.; Wong Z. C.; Tan E. K. M.; Miao Y. E.; Fan W.; Liu T.; Phang I. Y.; Ling X. Y. Tracking Airborne Molecules from Afar: Three-Dimensional Metal-Organic Framework-Surface-Enhanced Raman Scattering Platform for Stand-Off and Real-Time Atmospheric Monitoring. ACS Nano 2019, 13, 12090–12099. 10.1021/acsnano.9b06486. [DOI] [PubMed] [Google Scholar]
  23. Caviezel D.; Narayanan C.; Lakehal D. Adherence and Bouncing of Liquid Droplets Impacting on Dry Surfaces. Microfluid. Nanofluid. 2008, 5, 469–478. 10.1007/s10404-007-0248-2. [DOI] [Google Scholar]
  24. Mhetre M. R.; Abhyankar H. K. Human Exhaled Air Energy Harvesting with Specific Reference to PVDF Film. Eng. Sci. Technol. 2017, 20, 332–339. 10.1016/j.jestch.2016.06.012. [DOI] [Google Scholar]
  25. Lee W. S.; Ahn J.; Jung S.; Lee J.; Kang T.; Jeong J. Biomimetic Nanopillar-based Biosensor for Label-Free Detection of Influenza A Virus. BioChip J. 2021, 15, 260–267. 10.1007/s13206-021-00027-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Potejanasak P.; Duangchan S. Gold Nanoisland Agglomeration upon The Substrate Assisted Chemical Etching Based on Thermal Annealing Process. Crystals 2020, 10, 533 10.3390/cryst10060533. [DOI] [Google Scholar]
  27. Telecka A.; Li T.; Ndoni S.; Taboryski R. Nanotextured Si Surfaces Derived from Block-Copolymer Self-Assembly with Superhydrophobic, Superhydrophilic, or Superamphiphobic Properties. RSC Adv. 2018, 8, 4204–4213. 10.1039/C8RA00414E. [DOI] [Google Scholar]
  28. Kang M.; Kim J. J.; Oh Y. J.; Park S. G.; Jeong K. H. A Deformable Nanoplasmonic Membrane Reveals Universal Correlations Between Plasmon Resonance and Surface Enhanced Raman Scattering. Adv. Mater. 2014, 26, 4510–4514. 10.1002/adma.201305950. [DOI] [PubMed] [Google Scholar]
  29. Oh Y. J.; Jeong K. H. Glass Nanopillar Arrays with Nanogap-Rich Silver Nanoislands for Highly Intense Surface Enhanced Raman Scattering. Adv. Mater. 2012, 24, 2234–2237. 10.1002/adma.201104696. [DOI] [PubMed] [Google Scholar]
  30. Meng X.; Dyer J.; Huo Y.; Jiang C. Greater SERS Activity of Ligand-Stabilized Gold Nanostars with Sharp Branches. Langmuir 2020, 36, 3558–3564. 10.1021/acs.langmuir.0c00079. [DOI] [PubMed] [Google Scholar]
  31. Gambinossi F.; Mylon S. E.; Ferri J. K. Aggregation Kinetics and Colloidal Stability of Functionalized Nanoparticles. Adv. Colloid Interface Sci. 2015, 222, 332–349. 10.1016/j.cis.2014.07.015. [DOI] [PubMed] [Google Scholar]
  32. El-Atab N.; Chowdhury F.; Ulusoy T. G.; Ghobadi A.; Nazirzadeh A.; Okyay A. K.; Nayfeh A. Similar to 3-nm ZnO Nanoislands Deposition and Application in Charge Trapping Memory Grown by Single ALD Step. Sci. Rep. 2016, 6, 38712. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Vogt K. W.; Kohl P. A.; Carter W. B.; Bell R. A.; Bottomley L. A. Characterization of Thin Titanium-Oxide Adhesion Layers on Gold - Resistivity, Morphology, and Composition. Surf. Sci. 1994, 301, 203–213. 10.1016/0039-6028(94)91300-5. [DOI] [Google Scholar]
  34. Lee S. J.; Guan Z.; Xu H.; Moskovits M. Surface-enhanced Raman Spectroscopy and Nanogeometry: The Plasmonic Origin of SERS. J. Phys. Chem. C 2007, 111, 17985–17988. 10.1021/jp077422g. [DOI] [Google Scholar]
  35. Bake B.; Larsson P.; Ljungkvist G.; Ljungstrom E.; Olin A. C. Exhaled Particles and Small Airways. Respir. Res. 2019, 20, 8 10.1186/s12931-019-0970-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Liu D.; Wu F.; Cen Y.; Ye L.; Shi X. Y.; Huang Y. L.; Fang S. S.; Ma L. Comparative Research on Nucleocapsid and Spike Glycoprotein as The Rapid Immunodetection Targets of COVID-19 and Establishment of Immunoassay Strips. Mol. Immunol. 2021, 131, 6–12. 10.1016/j.molimm.2021.01.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Daoudi K.; Ramachandran K.; Alawadhi H.; Boukherroub R.; Dogheche E.; El Khakani M. A.; Gaidi M. Ultra-sensitive and Fast Optical Detection of The Spike Protein of The SARS-Cov-2 Using Agnps/Sinws Nanohybrid Based Sensors. Surf. Interfaces 2021, 27, 101454 10.1016/j.surfin.2021.101454. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Zhang M.; Li X.; Pan J.; Zhang Y.; Zhang L.; Wang C.; Yan X.; Liu X.; Lu G. Ultrasensitive Detection of SARS-Cov-2 Spike Protein in Untreated Saliva Using SERS-Based Biosensor. Biosens. Bioelectron. 2021, 190, 113421 10.1016/j.bios.2021.113421. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Pezzotti G.; Boschetto F.; Ohgitani E.; Fujita Y.; Shin-Ya M.; Adachi T.; Yamamoto T.; Kanamura N.; Marin E.; Zhu W.; Nishimura I.; Mazda O. Raman Molecular Fingerprints of SARS-CoV-2 British Variant and the Concept of Raman Barcode. Adv. Sci. 2022, 9, 2103287 10.1002/advs.202103287. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Gal J. Y.; Fovet Y.; Adib-Yadzi M. About a Synthetic Saliva for In Vitro Studies. Talanta 2001, 53, 1103–1115. 10.1016/S0039-9140(00)00618-4. [DOI] [PubMed] [Google Scholar]
  41. Zou S.; Hou M.; Li J.; Ma L.; Zhang Z. Semi-Quantitative Analysis of Multiple Chemical Mixtures in Solution at Trace Level by Surface-Enhanced Raman Scattering. Sci. Rep. 2017, 7, 6186 10.1038/s41598-017-06543-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Ciloglu F. U.; Caliskan A.; Saridag A. M.; Kilic I. H.; Tokmakci M.; Kahraman M.; Aydin O. Drug-Resistant Staphylococcus Aureus Bacteria Detection by Combining Surface-Enhanced Raman Spectroscopy (SERS) and Deep Learning Techniques. Sci. Rep. 2021, 11, 18444 10.1038/s41598-021-97882-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Thrift W. J.; Ronaghi S.; Samad M.; Wei H.; Nguyen D. G.; Cabuslay A. S.; Groome C. E.; Santiago P. J.; Baldi P.; Hochbaum A. I.; Ragan R. Deep Learning Analysis of Vibrational Spectra of Bacterial Lysate For Rapid Antimicrobial Susceptibility Testing. ACS Nano 2020, 14, 15336–15348. 10.1021/acsnano.0c05693. [DOI] [PubMed] [Google Scholar]
  44. Lee S.; Kim H.; Lee D.. Linearization Autoencoder: an Autoencoder-based Regression Model with Latent Space Linearization bioRxiv 2022.
  45. Azzi L.; Carcano G.; Gianfagna F.; Grossi P.; Dalla Gasperina D.; Genoni A.; Fasano M.; Sessa F.; Tettamanti L.; Carinci F.; Maurino V.; Rossi A.; Tagliabue A.; Baj A. Saliva is a Reliable Tool to Detect SARS-Cov-2. J. Infect. 2020, 81, E45–E50. 10.1016/j.jinf.2020.04.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Chung Y. S.; Lee N. J.; Woo S. H.; Kim J. M.; Kim H. M.; Jo H. J.; Park Y. E.; Han M. G. Validation of Real-time RT-PCR For Detection of SARS-Cov-2 in The Early Stages of The COVID-19 Outbreak in The Republic of Korea. Sci. Rep. 2021, 11, 14817 10.1038/s41598-021-94196-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Kang M.; Park S. G.; Jeong K. H. Repeated Solid-state Dewetting of Thin Gold Films for Nanogap-rich Plasmonic Nanoislands. Sci. Rep. 2015, 5, 14790 10.1038/srep14790. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Cha H.; Kim H.; Joung Y.; Kang H.; Moon J.; Jang H.; Park S.; Kwon H. J.; Lee I. C.; Kim S.; Yong D.; Yoon S. W.; Park S. G.; Guk K.; Lim E. K.; Park H. G.; Choo J.; Jung J.; Kang T. Surface-enhanced Raman Scattering-based Immunoassay for Severe Acute Respiratory Syndrome Coronavirus 2. Biosens. Bioelectron. 2022, 202, 114008 10.1016/j.bios.2022.114008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Lemaitre G.; Nogueira F.; Aridas C. K.. Imbalanced-learn: A Python Toolbox to Tackle the Curse of Imbalanced Datasets in Machine Learning arXiv 2017, p 18. 10.48550/arXiv.1609.06570. [DOI]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

am2c16446_si_001.pdf (654.3KB, pdf)

Articles from ACS Applied Materials & Interfaces are provided here courtesy of American Chemical Society

RESOURCES