Preparations and characterizations of the trimer arraysFigure 1a, b illustrates the working principles of the plasmonic trimers. For general plasmonic structures (Fig. 1a), the ‘hotspots’, strong SERS occurs at the gap position between two plasmonic nanoparticles, whereas the molecules are randomly adsorbed on the surface, leading to the fact that most of the molecules cannot be detected in this configuration. While for the plasmonic heterogeneous trimers (Fig. 1b), the probe molecules were mostly adsorbed on the trap particle due to the differences in chemical affinity. Therefore, the probability that a molecule can experience the amplification from the ‘hotspots’ is increased, leading to higher SERS sensitivity.Fig. 1: Schematic diagrams illustrating the preparations of the plasmonic trimer arrays.Differences in working principles between the homogeneous trimers (a) and heterogeneous trimers (b). c porous AAO nanomask transferred on a substrate. d Fabrication of the plasmonic dimers by angle-resolved shadow depositions. e optional atomic layer deposition coating after the shadow depositions. f Further modulation of the pore diameter of AAO by sputtering. g Deposition of the trap particle by final vertical e-beam deposition. h Optical photos of the fabricated samples on silicon wafer and quartz substrates. The highlighted areas in (a) and (b) represent the “hotspots” region.Figure 1c–h outlines the preparation of trimer arrays. The porous anodic aluminum oxide (AAO) membranes with a membrane thickness of 150 ± 10 nm and pore diameter of 80 ± 2 nm were prepared and transferred onto substrates as masks for e-beam evaporations (Fig. 1c). The methodologies describing the fabrication and manipulation of the AAO membranes with desired membrane thickness and pore diameter are detailed in refs. 43,44. The dimer particles with a gap distance of ≈ 20 nm were firstly fabricated by two-step angle-resolved shadow depositions (Fig. 1d). The deposition angles were controlled by tilting the samples to the desired angle, calculated based on the pore thickness and diameter. The gap distance between the dimers can be continuously modulated by adjusting the deposition angles (Supplementary Fig. 1). This is followed by optional atomic layer deposition (ALD) of 1.5 nm thick Al2O3 to form a dielectric layer on the surface on the dimers (Fig. 1e). The diameter of the AAO pores was further adjusted by magnetron sputtering (Fig. 1f), employing the same sputtering material as that will be used in the following vertical deposition. Different to e-beam evaporation, the sputtering gasified the target atoms with no directionality, and therefore, the sputtering material was mostly deposited onto the top surface of the AAO. The sputtering offers the possibility to deposit the trap particle at the gap region between the dimers in the following vertical evaporation step (Fig. 1g), and the diameter of the trap particle can be well tuned by adjusting the sputtering parameters (Supplementary Fig. 2). Afterwards, the AAO membranes were peeled off with adhesive tape, and the plasmonic trimer arrays were exposed. The prepared sample is shown in Fig. 1h, and the dimension and shape of the sample area can be controlled by cutting the AAO membranes to a suitable size. This strategy enables large-area and mass fabrication of functional trimers with aligned orientations, small gap distances, and freedom from surface chemicals, all of which are favorable for SERS studies.This fabrication strategy allows on-demand construction of trimer structures. The materials of both the dimer particle and trap particle can be metals or oxides, and the optional ALD process can be used to further adjust the functionality. To demonstrate the universality, we have prepared different trimers including homogeneous Au-Au-Au trimers and heterogeneous Au@Al2O3-Au-Au@Al2O3, Au-Ag-Au, and Au-TiOx-Au trimers (Supplementary Fig. 3). In particular, we employed the Au@Al2O3-Au-Au@Al2O3 trimers as a typical model to illustrate how to improve the SERS performance by inducing the target molecules to the trap particle.Figure 2a presents the scanning electron microscopy (SEM) image of the Au@Al2O3-Au-Au@Al2O3 trimers (for large-area SEM image, see Supplementary Fig. 4), suggesting that Au is deposited at the gap region between the Au@Al2O3 dimers. This is further identified by the high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) in Fig. 2b. The elemental mappings in Fig. 2c, d unveil that Al only exists around the dimers. The transmission electron microscopy (TEM) images in Fig. 2e, f reveal that the ellipsoidal dimer particles were coated with an Al2O3 layer with a thickness of ≈ 1.5 nm, whereas the trap particle was not.Fig. 2: Characterizations of the Au@Al2O3-Au-Au@Al2O3 trimers.a SEM image of the Au@Al2O3-Au-Au@Al2O3 trimers. HAADF-STEM image (b) and corresponding Au (c) and Al (d) elemental mappings of the Au@Al2O3-Au-Au@Al2O3 trimers, along with TEM images for the areas marked in red (e) and blue (f). The scale bars of the elemental mapping and TEM images are 20 nm and 2 nm, respectively. Experimental (g) and simulated (h) polarized extinction spectra (horizontal: blue; perpendicular: bronze) of the Au@Al2O3-Au-Au@Al2O3 trimer arrays, respectively. i Local EM field distributions around the Au@Al2O3-Au-Au@Al2O3 trimers under parallel and perpendicular polarized excitations at 785 nm, and color scale corresponding to the change in electric field intensity. The scale bars are 20 nm. The SEM characterizations were independently repeated for dozens of times for samples from different batches, and the results were similar. The HAADF, elemental mappings and TEM were independently repeated for three times for samples from different batches, and the results were similar. Source data are provided as a Source Data file.The plasmonic modes of the trimer arrays were characterized by polarized UV-vis spectroscopy. The results in Fig. 2g suggest that the modes corresponding to the long axis of the Au@Al2O3 ellipsoidal particles were located at approximately 585 nm under perpendicular excitation (θ = 90°), and the modes of the trap particle were immersed in this case. On the other hand, a broadening and redshift in the spectrum were observed under horizontal polarization (θ = 0°), due to the hybridized coupling of different modes of the trimers. The observations are basically consistent with the simulations by finite-element method (FEM) as shown in Fig. 2h. The narrowing of spectra in simulations can be explained by the differences of surface roughness and uniformity in particle size. In the simulations, a shoulder peak at 540 nm, which probably arises from the short axis mode of the Au@Al2O3 ellipsoidal particle, was observed under horizontal excitation (θ = 0°). The assignment was confirmed by evaluating the experimental and simulated plasmonic spectra of the Au-Au-Au trimers, where the Au@Al2O3 dimers were replaced by bare Au dimers (see Supplementary Fig. 5).Figure 2i plots the electromagnetic (EM) fields around the Au@Al2O3-Au-Au@Al2O3 trimers calculated by FEM simulations. The results reveal that the ‘hotspots’ are located at the regions around the trap particle. Obvious polarization-dependent enhancement was observed with a maximum of E/E0 = 87.2 under horizontal excitation. Considering that the enhancement factor can be approximated by the fourth power of E/E0, we can estimate that the average EM enhancement around the trimer structures is on the order of 105 through simple calculation45 (maximum 5.8 × 107), representing a 102-fold increase compared with monomer arrays46,47.SERS performance and single-molecule detectionFigure 3 illustrates the SERS performance of the Au@Al2O3-Au-Au@Al2O3 trimers. The Au-Au-Au trimers, which were similar to Au@Al2O3-Au-Au@Al2O3 in configuration but without the Al2O3 coating, were prepared for comparison. 4-Mercaptopyridine (4-Mpy), which strongly adsorbs on gold by the thiol group, was employed as a Raman probe to illustrate the chemical selectivity. As shown in Fig. 3a, the SERS signal was significantly improved on the Au@Al2O3-Au-Au@Al2O3 trimers. The improvement would be more significant with lower molecular concentrations (see Supplementary Fig. 6). It can be explained by the differences in chemical affinities between the molecules and different substrate materials. For the Au@Al2O3-Au-Au@Al2O3 trimers, the 4-Mpy molecules primarily adsorbs on the Au traps, ensuring the overlapping between the Raman probes and ‘hotspots’ in position, and thus substantially improve the SERS sensitivity. While for the Au-Au-Au trimers, 4-Mpy random adsorbs on the trimers, and the probability that a molecule is located at the hotspots is decreased. Additionally, comparative Raman mappings of 10−11 M of 4-Mpy were performed to illustrate the differences in SERS performance between the two trimer configurations, as evidenced in Supplementary Fig. 7. The impact of ALD thickness on SERS performance was also discussed in Supplementary Fig. 8.Fig. 3: SERS performance of the Au@Al2O3-Au-Au@Al2O3 trimer arrays.a SERS spectra of 10-9 M 4-Mpy from the Au@Al2O3-Au-Au@Al2O3 and Au-Au-Au trimers. b SERS spectra of 10−6 M toluene from the Au@Al2O3-Au-Au@Al2O3 and Au-Au-Au trimers. c Calculated adsorption energy of 4-Mpy and toluene, respectively, on Au(111) and γ-Al2O3(100) surfaces. Au, Al and O atoms are represented in gold, gray and red colors. d–f SERS spectra of 10−14 M 4-Mpy, 10−14 M 4-MBT and 10−14 M 4-ATP from the Au@Al2O3-Au-Au@Al2O3 trimer arrays, respectively (integral time 10 s). Source data are provided as a Source Data file.As a contrast, toluene, which is structurally similar to 4-Mpy but without the thiol group, was investigated for comparison. The results in Fig. 3b suggest that there are no distinct differences in SERS intensities between the two cases. This is because toluene adsorbs on gold and Al2O3 with similar binding affinities via a methyl group. It should be noted that the preparations for toluene and 4-Mpy were different due to their differences in binding affinities, and the real differences in SERS sensitivity would be much more significant (see Methods section, Raman preparations). Besides, Raman characterizations of 4-Mpy and toluene at different concentrations are available in Supplementary Fig. 6.The adsorption energies of 4-Mpy and toluene on Au(111) and γ-Al2O3(100) were calculated by density functional theory (DFT) simulations, as depicted in Fig. 3c. The gold nanoparticles fabricated by e-beam evaporation preferentially expose Au(111) facets48, which is confirmed by the X-ray diffraction (XRD) and high-resolution TEM results in Supplementary Fig. 9. On the other hand, γ-Al2O3(100) was employed for the simulation due to its similarity to the amorphous and partially hydroxylated alumina produced in the ALD process49,50,51. Details are available in the Methods section, and the full simulation results are available in Supplementary Fig. 10.The simulations reveal a strong binding affinity between 4-Mpy and Au(111) of −1.054 eV, while a significantly weaker binding affinity was observed for γ-Al2O3 (−0.455 eV). The results suggest that 4-Mpy molecules would be preferably induced to the Au trap particle in the Au@Al2O3-Au-Au@Al2O3 trimer configuration, while they would be randomly adsorbed on the surface of the Au-Au-Au trimers, illustrating the differences in SERS intensities in Fig. 3a. In comparison, the binding affinities between toluene and Au(111) and γ-Al2O3(100) are −0.380 eV and −0.307 eV, respectively, indicating that toluene weakly adsorbs on Au(111) and Al2O3 by physical adsorption with similar bonding energies. Consequently, no distinct differences in the SERS intensities in Fig. 3b is observed for toluene. Additionally, the plasmonic modes of the Au@Al2O3-Au-Au@Al2O3 trimers are relative closer to the excitation laser wavelength at 785 nm (Fig. 2 and Supplementary Fig. 5), which might be a secondary reason for the differences in SERS intensities in Fig. 3a.Figure 3d–f presents the SERS detection limit spectra of 4-Mpy, 4-methylbenzenethiol (4-MBT) and 4-aminothiophenol (4-ATP) at a concentration of 10−14 M on the Au@Al2O3-Au-Au@Al2O3 trimers, whereas their detection limits on the Au-Au-Au trimers were determined to be 10−11 M (full data in Supplementary Figs. 6, 11). Additionally, improved signal homogeneity was observed for the Au@Al2O3-Au-Au@Al2O3 trimers, suggesting the efficiency against the heterogeneous distribution of ‘hotspots’ (Supplementary Fig. 12). The DFT simulations of different adsorption sites for these molecules are available in Supplementary Fig. 10. The experimental enhancement factor was calculated to be ≈ 1010 for 4-Mpy (see Methods section, calculation of SERS enhancement factors). Moreover, similar comparative SERS measurements of adenine, thiram, and benzeneselenol (BSe), which have a high affinity towards the gold trap particles, have also been performed to validate the functionality of the Au@Al2O3-Au-Au@Al2O3 trimers, as included in Supplementary Fig. 13.Figure 4 illustrates the single-molecule SERS detection measurements on the Au@Al2O3-Au-Au@Al2O3 trimer arrays by the bi-analyted method. This method employs two analytes to sort out single-molecule SERS events by judging whether the event contains an individual analyte or two analytes52, and the details about the statistical method in this experiment are available in ref.53. We selected 4-MBT and 4-Mpy, which are similar in chemical structures, as the Raman probes. The concentrations were set to be 10−13 M, and the Raman intensities at 1078 cm−1 (4-MBT) and 1098 cm−1 (4-Mpy) were analyzed to evaluate the SERS performance. Figure 4a plots the statistical map from 1640 measurement points, where pure 4-MBT events, pure 4-Mpy events and mixed events are marked with different colors. Here most of the events were blank due to the low molecular concentrations, which is favorable to improve the validity of single-molecule SERS53,54. Fig. 4b presents the typical SERS spectra of single-molecule 4-MBT events, single-molecule 4-Mpy events, mixed events and null events. Figure 4c records the statistic results from the map, including 52 pure 4-Mpy events, 48 pure 4-MBT events, and 9 mix events possessing Raman characters from both molecules. Rigorous statistical histogram of the single-molecule frequency by modified principal-component analysis (MPCA) algorithm was plotted in Fig. 4d (details see Supplementary Fig. 14 and Methods section, computational steps for MPCA algorithm). The edge of the histogram indicates a high probability of single-molecule events. It can be observed that the contributions of p ≈ 1 (4-Mpy event) and p ≈ 0 (4-MBT event) are very close because of their similarity in Raman scattering cross sections. Moreover, single-molecule characteristics including signal blinking and spectrum wandering were investigated (see Supplementary Fig. 15), further identifying the validity of the single-molecule SERS events.Fig. 4: Bi-analyte single-molecule SERS investigations with the Au@Al2O3-Au-Au@Al2O3 trimers.a 2D statistical map of the single-molecule 4-MBT events, 4-Mpy events and mixed events by the bi-analyte method (40 × 41, step 1.5 μm, integral time 500 ms). b Typical SERS spectra of single-molecule 4-MBT events (purple), single-molecule 4-Mpy events (brown), mixed events (blue) and null events (pink). c Statistical histograms of pure 4-Mpy events (brown), mixed events (bule) and pure 4-MBT events (purple). d Probability histogram of the pure 4-Mpy events obtained by MPCA algorithm. Source data are provided as a Source Data file.Pathological diagnosis and subtyping of lung tumorsFor lung cancer patients, the balances of oxygen free radical breakdown, leading to the lipid peroxidation of polyunsaturated fatty acids in cancer cell membranes and organelle membranes, and consequently extra production of aldehydes and other volatile organic compounds (VOCs)55,56,57. As a result, patients with lung cancer may exhibit significantly higher levels of aldehydes in their lung tissues58,59,60. While SERS measurement of aldehyde molecules has been showcased with exhaled breath condensate from patients61,62, conducting direct SERS analysis with human tissues remains challenging due to the limited quantity of aldehydes within cells. In-depth investigations, including methods for assessing the accuracy of SERS results with tissue biopsy data, exploring the relationship between TNM stage and SERS outcomes, and investigating the possibility of tumor subtyping by SERS, remains to be answered.Leveraging the sensitivity of the Au@Al2O3-Au-Au@Al2O3 trimers, SERS investigations with human tissues were conducted to demonstrate their potential for the early diagnosis and subtyping of lung tumors, as depicted in Fig. 5. We prepared 4-ATP-modified Au@Al2O3-Au-Au@Al2O3 samples for the detection of aldehyde molecules. A monolayer of 4-ATP was adsorbed on the gold trap particle via its thiol group during preparations. The amino group located at the opposite end of 4-ATP reacts with aldehydes to form a C=N bond for Raman identification55,63, as illustrated in Fig. 5a. Typical Raman spectra from the 4-ATP-modified sample before and after benzaldehyde (BA) collection were displayed in Fig. 5b, where the mode at 1623 cm−1 suggesting the C=N vibration mode can be clearly discerned after the collection64,65. The SERS optimizations for the experiment are available in Supplementary Fig. 16, with a detection limit of 10−9 M for BA, as shown in Supplementary Fig. 17. Additionally, SERS detection of other aldehydes are presented in Supplementary Fig. 18. The anti-interference performance of the Au@Al2O3-Au-Au@Al2O3 trimer samples were displayed in Fig. 5c, suggesting the ability in specific detection of aldehyde molecules.Fig. 5: Pathological diagnosis of lung tumors with the Au@Al2O3-Au-Au@Al2O3 trimer arrays.a Schematic diagram of the reaction between aldehydes and 4-ATP on gold. b SERS spectra of 4-ATP on Au@Al2O3-Au-Au@Al2O3 trimers before (violet) and after (orange) reacting with discharged gas containing BA molecules. The C=N vibration mode suggesting the reaction is marked with a light green strip. c Anti-interference test of the Au@Al2O3-Au-Au@Al2O3 trimer samples exposed to BA, acetone, methanol, ethanol, acetonitrile, ethyl acetate, hexyl hydride and water. d, e Optical images of the ad-ca and sq-ca lung tumor tissues, respectively, after HE staining under a microscope. The scale bars are 50 μm. This experiment was independently repeated for all the tumor samples, and the results were similar. f Typical SERS spectra from the samples containing fresh ad-ca (red) or sq-ca cancer (violet) tissues, and blank sample (brown). g, h Raman mapping at 1623 cm−1 from the ad-ca and sq-ca samples (1600 points), respectively (step 1.5 μm, integral time 10 s), which color scales representing Raman intensities. Source data are provided as a Source Data file.Further investigations were performed with fresh lung cancer tissue of different subtypes, as illustrated in Fig. 5d–h. Lung tumors are generally divided into adenocarcinoma (ad-ca), squamous carcinoma (sq-ca), and other minor subtypes58. We obtained ten fresh lung tumor tissues from patients by surgical operation. The tissues were cut into dimensions of approximately 1 × 1 × 0.5 cm3 for SERS investigations. The typical pathological characterizations of cancer cells from the two major subtypes after hematoxylin-eosin (HE) staining are shown in Fig. 5d, e. The 4-ATP modified trimer samples were immersed in buffer solution containing tumor tissues (Supplementary Fig. 19). We collected 1600 Raman spectra from each sample. The averaged SERS spectra from an ad-ca sample and a sq-ca sample are compared in Fig. 5f, suggesting that the signal intensity for ad-ca was much stronger than that for sq-ca. Corresponding Raman intensity maps are plotted in Fig. 5g, h, revealing that a majority of the measurement points was positive for ad-ca, while only a few events were positive for sq-ca.We analyzed the SERS results from the ten patients, including six ad-ca, three sq-ca, and one benign case (Supplementary Fig. 20), and compared the results with their TNM stage (Supplementary Table 1). Some preliminary conclusions were drawn: 1) SERS is sensitive in distinguishing different lung tumor subtypes, suggesting variations in the body content of aldehyde molecules among them. Specifically, SERS is sensitive to ad-ca but not to sq-ca or benign cases. 2) There is no evidence suggesting a correlation between the SERS outcomes and the original tumor size or lymph node with the same volume of tumor tissue. For sq-ca patients, SERS diagnosis is appliable to all the patients with TNM stages from IA1 to IIB. Still, further studies with a larger dataset are required to refine these preliminary observations. Similar Raman measurement of ad-ca has also been also performed on the Au-Au-Au trimers for comparison, as revealed in Supplementary Fig. 21.
