A novel ternary heterostructure with dramatic SERS activity for evaluation of PD-L1 expression at the single-cell level

See allHide authors and affiliations

Science Advances  02 Nov 2018:
Vol. 4, no. 11, eaau3494
DOI: 10.1126/sciadv.aau3494


Surface-enhanced Raman scattering (SERS) probes based on a charge transfer (CT) process with high stability and reproducibility are powerful tools under open-air conditions. However, the key problem ahead of practical usage of CT-based SERS technology is how to effectively improve sensitivity. Here, a novel ternary heterostructure SERS substrate, Fe3O4@GO@TiO2, with a significant enhancement factor of 8.08 × 106 was first synthesized. We found the remarkable enhanced effect of SERS signal to be attributed to the resonance effect of CuPc, CT between GO and TiO2, and enrichment from a porous TiO2 shell. In addition, we developed a robust SERS probe with good recyclability under visible light illumination on Fe3O4@GO@TiO2 nanocomposites toward ultrasensitive detection of cancer cells down to three cells. We have now successfully applied this probe for in situ quantification and imaging of programmed cell death receptor ligand 1 (PD-L1) on triple-negative breast cancer cell surface at the single-cell level and for monitoring the expression variation of PD-L1 during drug treatment.


Surface-enhanced Raman scattering (SERS) has been a promising analytical technique for the detection of trace species because of its remarkable advantages, namely, it is sensitive, it is nondestructive, it has a quick response time, and it provides unique information on the species (13). Studies on noble metal nanostructures, such as Au, Ag, and Cu, indicate that the tremendous electromagnetic field enhancement at the hotspots greatly increases the sensitivity of SERS (4). However, the SERS effect based on electromagnetic field enhancement has no selectivity for Raman reporters, resulting in complicated signal outputs generated by unexpected interactions (5). Recently, another mechanism, chemical enhancement mechanism (CM), which is related to the formation of charge transfer (CT) states between the adsorbed molecules and the nanoparticles (NPs) (6), has attracted more attention. Compared with SERS, which is based on electromagnetic field mechanism, this unique method always demonstrates high molecule selectivity. Most substrates based on CM show stable and reproducible SERS signals without agglomeration, which is different from those obtained from noble metals. Moreover, the CM-based substrate exhibits good biocompatibility without cytotoxic outcomes. Thus, various new materials have been developed to investigate CT in plasma-free SERS, such as graphite oxide (GO) and semiconductor materials (7). However, only 10 to 100 enhancements have been achieved by these materials (8, 9), limiting their potential applications for reliable quantitative detection in biological and biomedical analysis and diagnosis. Accordingly, low enhancement factor (EF) values prevent further applications of CM-based SERS substrates.

Here, a ternary heterostructure (Fig. 1), magnetic Fe3O4@GO@TiO2 (denoted as MGT), was developed (10), working as a plasmon-free SERS substrate with high molecular selectivity and stability, as well as a remarkable EF value of up to 8.08 × 106. This notable EF value was found to be mainly attributed to a high-efficiency interfacial CT process between GO and TiO2, which was verified by theoretical calculation and time-resolved diffuse reflectance (TDR) spectroscopic measurements. The present SERS substrate with high sensitivity and selectivity was developed to determine trace biological species, such as programmed cell death receptor ligand 1 (PD-L1, a model molecule), at the single-cell level. PD-L1 and its predominant target, programmed death 1 (PD-1), are two crucial immune checkpoint molecules tightly associated with immune resistance (11), which promote immune evasion and tumor recurrence and metastasis (12). PD-L1 is rarely expressed in regular epithelial tissues but is aberrantly expressed in many cancers, including breast (13), melanoma (14), and renal cell carcinoma (15). Immunotherapy using anti–PD-1 or anti–PD-L1 antibody (Ab) to block the interaction between PD-1 and PD-L1 has shown positive antitumor effects in clinical studies (16). Therefore, the quantitative characterization of PD-L1 expression on the cell surface has become an extremely important subject directly related to the development of diagnostic tools to guide PD-L1–based personalized treatment. A variety of analytical tools for PD-L1 monitoring have been reported during the past few years, such as immunohistochemical detection, mass spectrometry (MS), radiological imaging, flow cytometry, and positron emission tomography (1721). To give a better understanding of the structural and functional roles that cells play in biological systems, it is necessary to develop a robust tool for the analysis and quantification of the components and metabolites in different individual cells (22). Developing a robust preliminary testing approach that provides reliable and quantitative diagnostic results in real time and in situ PD-L1 monitoring at the single-cell level remains a challenge. Triple-negative breast cancer cells (TNBCs) are types of cancer cells that always exhibit immune evasion and tumor recurrence and metastasis triggered by PD-L1 (12). In this experiment, we successfully used an MGT-based SERS probe to visualize and compare the expression and distribution of PD-L1 on three different kinds of TNBC surfaces at the single-cell level, which helped us to understand how PD-L1 behaves within different breast cancer genomic subtypes.

Fig. 1 The design and illustration of MGT.

Schematic illustration of the synthesis process and enhancement mechanism of the MGT substrate.


Synthesis and characterization of MGT

MGT nanocomposites were prepared through a layer-by-layer coating process (23). Magnetic Fe3O4 NPs were prepared through modified solvothermal reaction by reduction of FeCl3 (23). As shown in transmission electron microscopy (TEM) images (fig. S1A), the as-prepared Fe3O4 NPs were highly uniform with diameters of 120 ± 5 nm. We observed a well-resolved lattice fringe with an interplanar distance of 0.26 nm for Fe3O4, which corresponds to the (311) plane of magnetite (fig. S1B). In addition, hysteresis loops of Fe3O4 particles in fig. S1C indicate high-performance magnetic properties without remanence or coercivity at 300 K. We coated Fe3O4 NPs with GO through electrostatic adsorption with the linkage of polydimethyldiallylammonium chloride (PDDA) to produce magnetic Fe3O4@GO (MG). Scanning electron (SEM) micrographs (Fig. 2A) show that Fe3O4 NPs were effectively coated with GO. TEM images show that the irregular shapes of the nanocomposites were highly dispersed with diameters of 150 to 250 nm (Fig. 2B). Then, we prepared the uniform porous TiO2 shell via a versatile kinetics-controlled coating method to produce MGT nanocomposites. We observed the ternary heterostructure of MGT (size, 200 to 400 nm) coated by the aggregated TiO2 NPs (~8 nm) and with uniform pores (~5 nm; Fig. 2, C and D). N2 sorption isotherms (Fig. 2E) of the nanocomposites after calcination show typical type IV curves with distinct hysteresis loops close to the H1 type. The Brunauer-Emmett-Teller surface area and the pore volume of the as-prepared nanostructures were calculated to be as high as 105 m2 g−1 and 0.16 cm3 g−1, respectively. The corresponding pore size distribution curves (Fig. 2F) derived from the adsorption branches of the isotherms using the Barrett-Joyner-Halenda method show a sample pore size of ~4.3 nm, which was similar to that obtained from the SEM image. A high-resolution TEM (HR-TEM) image [Fig. 2G (III)] shows that the TiO2 NPs were crystallized with a well-resolved lattice fringe with an interplanar distance of 0.24 nm, which corresponds to the (103) plane of anatase (24). We carried out energy dispersive spectroscopy (EDS) to characterize the elemental composition and the structural relationship of the nanostructures from every step obtained (Fig. 2G). It was clear that Fe was located at the center of nanocomposites, while the shell contained C and Ti, which were well separated from each other. X-ray diffraction (XRD) fitted well to those of Fe3O4, GO, and TiO2 bulk phases (Fig. 2H). The diffraction peaks of Fe3O4 were ascribed to the (220), (311), and (440) planes. After Fe3O4 was coated with GO and TiO2, two additional diffraction peaks were observed at 21.2° and 43.3°, which should be ascribed to GO. Several new diffraction peaks appeared at 25.3°, 37.8°, 38.6°, 48.1°, 53.9°, and 55.1°, assigned to the (101), (004), (112), (200), (105), and (211) planes of anatase-phase TiO2, respectively (25, 26), confirming the successful preparation of MGT. We further used Raman and Fourier transform infrared (FTIR) spectroscopy to characterize the as-prepared MGT. As depicted in Fig. 2I, three distinct peaks at 397, 518, and 640 cm−1 in the Raman spectrum of MGT were observed, indicating the presence of anatase-phase TiO2 (26, 27). In the FTIR spectra of MGT (Fig. 2J), two additional infrared bands assigned to Ti–O–C at 1680 cm−1 and Ti–O–H stretching vibration at 3260 cm−1 were observed (28). Both Raman and FTIR results indicated that the MGT nanocomposites were successfully synthesized. The stepwise assembly of MGT was also monitored by microeletrophoresis measurements whose results were expressed as ζ-potentials (fig. S1D). Because of the superparamagnetic properties of the magnetic cores of MGT (Fig. 2K), the MGT nanocomposites were conveniently separated from the reaction solution after the photodegradation treatment upon application of an extended magnetic field. Moreover, it is known that TiO2 shows absorption ranging from 200 to 350 nm. On the other hand, absorbance of MGT nanocomposites extended to the visible region (400 to 600 nm) because of the conjugation of TiO2 with visible GO (fig. S1E) (26). Figure S2 shows the Raman spectra of copper phthalocyanine (CuPc) before and after illumination under visible light (460 nm < λ < 700 nm, 1000 W) on MGT nanocomposites. The intensity of the Raman spectra of CuPc sharply decreased after illumination by visible light for 90 min, revealing that CuPc was significantly photodegraded (26). In contrast, we obtained negligible changes for MT NPs (fig. S2, B and C). Accordingly, after five cycles of enrichment followed by illumination, the SERS activity of MGT remained more than 98% (fig. S2D), confirming the good recyclability of MGT. Considering the photocatalytic activity of MGT under the visible light region, we investigated the Raman signal reproducibility of MGT in fig. S3. After irradiating 15 times with an exposure time of 1 s, the Raman intensity of CuPc remained approximately 97%, indicating the high stability and reproducibility of MGT substrates. Besides, we measured the surface wettability of MGT through contact angle. As shown in fig. S1F, MGT showed hydrophilic properties with an MGT contact angle of 41° ± 2°, which is suitable for further bio-applications.

Fig. 2 Characterization of MGT nanocomposites.

(A) SEM image of the MG substrate. (B) TEM image of the MG substrate. Inset: Size distribution of MG by dynamic light scattering measurement. a.u., arbitrary units. (C) SEM image of MGT nanocomposites. (D) TEM image of MGT nanocomposites. Inset: Size distribution of MGT by dynamic light scattering measurement. (E) N2 sorption isotherms and (F) corresponding pore size distribution curves of the MGT. (G) (I) Bright-field EDS images, (II) high-magnification image of (I) (marked with a green square), (III) high-resolution image of a select region in (II) (marked with a red square), and (IV to VIII) low-/high-magnification EDS images of MGT. (H) XRD patterns of (I) Fe3O4 NPs, (II) MG, (III) MGT, and (IV to VI) standard patterns of (IV) Fe3O4, (V) GO, and (VI) porous TiO2 NPs. (I) Raman spectra and (J) FTIR spectra of (I) Fe3O4 NPs, (II) GO, (III) MG, (IV) Fe3O4@TiO2 (MT), and (V) MGT. (K) The magnetic hysteresis loops at 300 K of the MGT. Inset: Photograph of the magnetic separation process of the MGT from the solution. emu, electromagnetic units.

Mechanism of Raman enhancement in MGT

Enhancement with CM usually related the matching degree of energy level and molecular symmetry between molecules and substrates (26). Figure 3 shows the Raman spectra of four typical symmetric molecules of D2h [catechol (CC) and 3,4,9,10-perylene-tetracarboxylicacid-dianhydride (PTCDA)], C3 [crystal violet (CV)], and D4h (CuPc) on Fe3O4 NPs, MT, MG, and MGT under 633-nm laser excitation. Compared with those on Fe3O4 NPs and MT, obvious enhancement was obtained on both MG and MGT for all four symmetric molecules, due to the symmetric matching between molecules of CC, PTCDA, CV, CuPc, and GO planes (29). On the other hand, we observed negligible enhanced effects for asymmetric molecules (oxalic acid, salicylic acid, and dopamine) on each substrate (fig. S4, A to C). Furthermore, π-π interactions generated by the structural similarity between molecular structures with parallel connected rings of CuPc and the hexagonal lattice surface structure of GO generally reduce the distance between CuPc molecules and MG or MGT substrates. The strong interaction between CuPc and MGT enhances the coupling between CuPc and MGT, resulting in a more significant enhancement effect for CuPc molecules than that for CV or CC on the same substrates (26).

Fig. 3 Enhancement effect of MGT.

Raman spectra of (A) CC, (B) PTCDA, (C) CV, and (D) CuPc on (I) Fe3O4 NPs (M), (II) MT, (III) MG, and (IV) MGT substrates under excitation of a 633-nm laser.

Remarkably, the intensity of the Raman peak at 1530 cm−1, corresponding to the in-plane ring symmetric nonmetal bound N–C stretch of CuPc, was used as a reference peak for further research (table S1) (30). It was found that MGT had better enhancement for CuPc compared with MG, an effect that could be ascribed mainly to the addition of a semiconductor, TiO2, to the GO-containing platform. The EF value of MGT for CuPc molecules was calculated to be as high as 8.08 × 106 (for calculation details, see Materials and Methods), which was much higher than that of graphene alone (47.3) (29) and the highest EF value (6.62 × 105) of plasmon-free materials reported to date (31). Given that the wavelength of the excitation laser used for Raman measurement in this work was 633 nm, which is close to the ultraviolet-visible (UV-vis) absorption peak of CuPc (fig. S4D), the high EF value may be attributed to the SERS effect of CuPc (32). After the excitation laser wavelength was adjusted to either 532 or 780 nm, we observed relatively weak absorption for CuPc molecules. Compared with absorption under 633-nm excitation, an obvious decline (~44%) in the Raman intensity of CuPc was observed under 532-nm irradiation, while no obvious Raman signals were obtained under 780 nm (fig. S4E). The results indicate that resonance Raman effect plays an important role in Raman enhancement.

We also observed obvious Raman enhancement of PTCDA (a molecule that shows no obvious absorption in 630- to 800-nm range) on MGT (Fig. 3B), which is lower than that for CuPc (about 68% Raman intensity for CuPc on MGT). In this case, the significant Raman enhancement was ascribed to the CT process that occurred between MGT substrates and the Raman reporter, due to the formation of Ti–O–C bonds between TiO2 and GO. We measured x-ray photoelectron spectroscopy (XPS) to clarify the interaction between GO and TiO2 (Fig. 4, A to D). Figure 4A shows the XPS survey spectrum of MGT, indicating that MGT is mainly composed of C, O, Fe, and Ti without a trace of contamination. The high-resolution XPS of the C 1s spectrum (Fig. 4B) of MG is ascribed to C–C (nonoxygenated ring C, 284.6 eV), C–O (285.6 eV), and carbonyl O=C (carboxylic, 287.8 eV) species (26, 33). Compared with MG, all these peaks of MGT (especially that of the O=C species) showed higher binding energies. Figure 4C shows O 1s core level spectra. Three obvious peaks at 529.4, 530.0, and 531.1 eV were observed, which should be attributed to the existence of O2−, −OH, and oxygen vacancies (Ov), respectively (34, 35). The high ratio of Ov indicates large numbers of Ov in MG, which could couple with Ti3+ easily (26). After MG was coated with TiO2 NPs, the peak intensity of O2− increased, while that of Ov decreased. In addition, we observed two peaks centered at 459.3 and 465.0 eV in the Ti 2p spectrum of MGT (Fig. 4D), which were consistent with the binding energy of Ti in the Ti–O–C bond (26). These obvious changes indicate the existence of interaction between TiO2 and GO, resulting in an efficient CT between Raman reporters and SERS substrates.

Fig. 4 Raman enhancement mechanism of MGT.

(A) XPS survey spectra of MG and MGT nanocomposites. High-resolution XPS spectra of (B) C 1s, (C) O 1s, and (D) Ti 3p of MG and MGT nanocomposites. TDR spectroscopic measurement for (E) MG, (F) MGT, and (G) CuPc after irradiation with a 633-nm laser flash. OD, optical density. (H) Time profiles of normalized transient absorption at 750 nm. (I) Two-dimensional integrated density difference plotted along the slab’s perpendicular axis. The bottom dashed line at 16 indicates the position of graphene, and the upper dashed line at 28 denotes that of the molecule. The red line indicates the charge difference on the GO and the green line denotes that on the TiO2-supported GO. (J) Conclusion of calculated results and the mechanism in MGT platforms. (K) Raman spectra of CuPc on Fe3O4, TiO2 (T), MT, MG, MGT with compact TiO2 shell (cMGT), Fe3O4@TiO2@GO (MTG), and MGT with the porous TiO2 shell (MGT) substrates under a 633-nm laser (5 mW) with an exposure time of 0.5 s.

We then investigated the interaction and CT process between GO and TiO2 through TDR spectroscopic measurements. Figure 4E shows an absorption band of MG centered at 750 nm, excited by 633 nm, and assigned to trapped electrons. The electrons decayed quickly owing to the fast charge recombination of electrons and holes generated in MG. Similarly, a broad absorption, assigned to the electrons in MGT and MGT-CuPc, was observed in the 750- to 800-nm region (Fig. 4, F and G). However, both decay lifetimes in MGT and MGT-CuPc (1424.41 and 3112.41 ps, respectively) were much longer than that in MG (246.97 ps), suggesting that the Ti–O–C coordinate bond at the interface acts as an electron trap site, leading to efficient CT. The time profiles of the transient absorption at 750 nm for MG, MGT, and MGT-CuPc were fitted by two-exponential functions (Fig. 4H). In contrast to the lifetime of MG, the time constants for MGT and MGT-CuPc at 750 nm indicate a much longer lifetime. The surface trap states generally increase the lifetime of electrons in semiconductors. Thus, TDR results provided direct evidence for CT from MGT to CuPc, resulting in much enhanced SERS activity using MGT.

To further confirm the TDR results, we investigated the electronic structure of MGT using density functional theory (DFT) calculation by examining a new electronic state, which was constructed in the interface. As illustrated in fig. S4F, we calculated both CuPc and MGT as a system to obtain the CT process, instead of isolated molecules. We investigated the coupling of CuPc and MGT as well as the consequent CT at equilibrium using DFT calculations. The DFT results (Fig. 4I) showed that much more obvious electron transfer occurred from MGT to CuPc than from MG to CuPc, which gave theoretical evidence for the efficient CT between MGT and CuPc molecules. Moreover, to further elucidate the CT process, we determined the Fermi level, the conductive band (CB) and valence band (VB) of MGT, and the highest occupied molecular orbital (HOMO) and the lowest unoccupied molecular orbital (LUMO) levels of different organic molecules using DFT calculation, which are shown in Fig. 4J. The Fermi level of GO was −4.6 eV from the vacuum level, while the HOMO and the LUMO levels of organic molecules were −7.9 and −3.4 eV for CC, −6.8 and −4.7 eV for PTCDA, −6.0 and −4.1 eV for CV, and −5.2 and −3.07 eV for CuPc, respectively. In general, Raman enhancement effect is derived from a robust CT between the substrate and the molecules (6). In the CuPc-MG system, the energy barrier from the Fermi level of GO to LUMO of CuPc is 1.53 eV. With the introduction of TiO2 to the MG system, charge redistribution occurs around the interface between TiO2 and molecules, which leads to the alignment of the energy levels between TiO2 and GO. In this case, the narrow energy barrier is 0.9 eV from the Fermi level of GO to CB of TiO2 and 0.83 eV from CB of TiO2 to LUMO of CuPc, respectively, which are much smaller than that in the MG system. Thus, compared with the CuPc-MG system, CT occurs much easier, and more obvious CT selective enhancement was observed in the SERS spectra of the MGT-CuPc system. Different from CuPc, the LUMO levels of CV (−4.10 eV) and PTCDA (−4.70 eV) were lower than CB of TiO2 (−3.70 eV; Fig. 4J), which results in a rare contribution to CT between the molecules and MGT with the introduction of TiO2, preventing a further enhancement of Raman intensity. The results of theoretical calculation also showed that there was an efficient CT between the molecules and the substrate, leading to SERS enhancement with a high EF value, which agrees well with that proposed by TDR measurement above.

Moreover, the SERS effect of M, T, MT, MG, MTG, cMGT, and MGT substrates to CuPc molecules was compared to further confirm the contribution of the enrichment of CuPc on MGT by porous TiO2 to the SERS effect (Fig. 4K). We observed no significant enhancement on M, T, and MT, which indicated that GO mainly contributed to the SERS effect. On the other hand, compared with MGT, MG and MTG also have multilayered GO nanosheets. However, the EF values of cMGT (4.67 × 105), MTG (1.06 × 106), and MG (3.05 × 105) were lower than that of MGT for CuPc (8.08 × 106; Table 1). Significantly, the enhancement effect of cMGT is much lower than that of MGT (about 51% Raman intensity for CuPc on MGT), which was attributed to cMGT preventing the interaction between GO and CuPc. Consequently, compared with MTG, cMGT, and MG, the significant SERS effect of MGT should be ascribed to the enrichment of CuPc on the porous TiO2 shell for MGT. Consequently, the remarkable enhanced EF toward CuPc molecules obtained at MGT substrates should be attributed to our designed ternary heterostructure: (i) the selective SERS effect of GO toward CuPc, (ii) the resonance effect of CuPc under 633-nm laser excitation, and (iii) CT between GO and TiO2 that remarkably enhances the Raman signal of CuPc and the enrichment of CuPc molecules by the porous TiO2 shell of MGT.

Table 1 EF value of CuPc on M, T, MT, MG, cMGT, MTG, and MGT substrates.
View this table:

Immunoassay of nanoprobes and evaluation of PD-L1 on TNBC cell surface

The MGT substrate with high reproducibility, sensitivity, and selectivity was then developed as an immunoassay SERS probe (MGT-Abs-CuPc) through the immobilization of the PD-L1 Abs and CuPc for PD-L1 analysis on TNBC surfaces (Fig. 5A). As shown in fig. S5A, upon the introduction of CuPc molecules to MGT, the UV-vis spectrum exhibits two peaks at 620 and 690 nm, which correspond to the adsorption of CuPc. After the immobilization of Abs to MGT, we observed an additional peak at ~260 nm, which should be attributed to adsorption of Abs. All these three additional peaks show the successful fabrication of the MGT-Abs-CuPc nanoprobe. To further confirm Abs conjugation to the nanostructures, we performed SDS–polyacrylamide gel electrophoresis (PAGE) imaging of MGT, Abs, and Abs-modified MGT (MGT-Abs). SDS-PAGE is a powerful analytical technique for protein identification in complex biological samples and has been widely used in biological studies for decades. As shown in fig. S5B, MGT-Abs [fig. S5B (III)] was higher than both MGT [fig. S5B (I)] and Abs [fig. S5B (II)], indicating the different molecular weights: MGT-Abs > MGT > Abs. SDS-PAGE imaging in fig. S5B (I to III) definitively showed the successful conjugation between the nanostructure and Abs. In addition, we also carried out FTIR spectroscopy (fig. S5C) to investigate the conjugation. Compared to MGT, MGT-Abs showed an additional infrared band assigned to the C–N stretching vibration at 1400 cm−1 and the C=O stretching vibration at 1620 cm−1. These two additional peaks were attributed to the formation of an amide bond at the interface between MGT and Abs. Both SDS-PAGE images and FTIR results revealed that Abs has been successfully conjugated to the nanostructures. We carried out confocal imaging to verify the recognition ability of MGT-Abs to the PD-L1 protein. We conducted three parallel experiments, which are shown in fig. S5 (D to G). No fluorescence signal was obtained after the coincubation of MGT and Cy5-labeled PD-L1. Moreover, before incubating with Cy5-labeled PD-L1, MGT-Abs also showed weak fluorescence. However, after the incubation with Cy5-labeled PD-L1, confocal images exhibited a bright fluorescence signal, indicating the high affinity and recognition ability of MGT-Abs to the PD-L1 protein. We incubated the developed probe with high SERS activity with TNBCs to quantify the PD-L1 expression on the surface of different TNBCs (Fig. 5A) using Raman spectroscopy. Temperature and incubation time are two important factors that determine the overall SERS signals. As shown in fig. S5 (H and I), the maximum SERS response was achieved at 37°C and 60 min. The amount of CuPc and Abs immobilized on MGT was another important experimental parameter. As the concentration of CuPc and Abs progressively increased, we observed the Raman peak at 1530 cm−1 to first increase and then decrease after the concentration reached 10 mM for CuPc and 100 μM for Abs, indicating that the optimal concentration of CuPc is 10 mM, while that of Abs is 100 μM (fig. S5, J and K). Moreover, we examined the cytotoxicity of the MGT-Abs-CuPc probe with a 3-(4,5-dimethylthiazol-2-yl)-2-diphenyltetrazolium bromide assay. As a result, the variability of HCC38, MDA-MB-231, and MCF-7 cells remained approximately 98% (15 μM, 120 min; fig. S6), indicating the low cytotoxicity of the probe. After optimizing experimental conditions and assessing biocompatibility, we conducted three parallel experiments to evaluate the PD-L1 expression levels on three different types of TNBC (HCC38, MCF-7, and MDA-MB-231) surfaces. HCC38 cells were first used as model cells. As shown in Fig. 5B, Raman intensity increased with an increasing number of cells captured on the silicon foil. The intensity of the Raman peak at 1530 cm−1 was used to reflect the amount of the MGT-Abs-CuPc probe attached to the cell surface. The intensity of the Raman peak (I) was proportional to the logarithmic value of captured cancer cells (N) in the range from 5 × 102 to 5 × 105 cells (R = 0.990, n = 7; Fig. 5C), which was consistent with the kinetics mechanism of the antigen-Abs reaction in the Abs excess zone (fig. S7A). The linear regression is shown in Eq. 1

Fig. 5 Evaluation of PD-L1 expression on cell face.

(A) Site-specific recognition and detection of PD-L1 on TNBC cells. (B) Raman spectra of MGT-Abs-CuPc obtained with HCC38 cells at various concentrations. (C) Plot of Raman scattering intensity versus logarithm of cell numbers. Each data point represents the average value from six replicate SERS spectra (SD, n = 6). Error bars represent SDs. (D) Plot of I versus the effect of PD-L1 concentration (m) on the decrease of Raman intensity at 1530 cm−1I) of HCC38 cells on the cytosensor. (E) Raman spectra of MGT-Abs-CuPc obtained with MDA-MB-231 cells at various concentrations. (F) Plot of Raman scattering intensity versus logarithm of cell numbers. Each data point represents the average value from six replicate SERS spectra (SD, n = 6). Error bars represent SDs. (G) Effect of PD-L1 concentration (m) on the decrease of Raman intensity at 1530 cm−1I) of MDA-MB-231 cells on the cytosensor. (H) Raman spectra of MGT-Abs-CuPc obtained with MCF-7 cells at various concentrations. (I) Plot of Raman scattering intensity versus logarithm of cell numbers. Each data point represents the average value from six replicate SERS spectra (SD, n = 6). Error bars represent SDs. (J) Effect of PD-L1 concentration (m) on the decrease of Raman intensity at 1530 cm−1I) of MCF-7 cells on the cytosensor. (K) The expression of PD-L1 on the cell surface. (L) Raman scanning mapping images obtained from HCC38, MDA-MB-231, and MCF-7 incubated with 0, 4, 10, 16, and 20 ng of IFN-γ for 48 hours. (M) Raman intensity of the peak at 1530 cm−1 obtained from HCC38, MDA-MB-231, and MCF-7 without stimulation. (N to P) Relationship between the Raman intensity and IFN-γ concentration for (N) HCC38, (O) MDA-MB-231, and (P) MCF-7 cells.

The HCC38 cell detection limit was calculated to be ~3 cells at 3σ. The overall intensity of Raman peaks at 1530 cm−e was determined by both the number of captured HCC38 cells and the number of PD-L1 on each cell. To quantify the PD-L1 expressed on cell surfaces, we used a binding competition method (36) to partially block the PD-L1 binding sites on the nanoprobe surfaces. Because the PD-L1–blocked Abs on nanoprobe surfaces could not conjugate to PD-L1 on cell surfaces, the peak intensities obtained from cytosensors incubated with PD-L1–blocked nanoprobes were lower than those of the cytosensors incubated with nonblocked nanoprobes. The decrease in intensity (ΔI) versus the amount of PD-L1 used to block the PD-L1 Abs on nanoprobe surfaces (m) showed a linear relationship ranging 0.1 to 1.0 fmol (R = 0.999, n = 5; Fig. 5D)


The average number of PD-L1 per captured HCC38 cell was calculated to be 1729 using Eqs. 1 and 2.

To further confirm and calculate the exact amount of PD-L1 on various cell surface types, we evaluated PD-L1 on MDA-MB-231 and MCF-7 cell surfaces with the same method as HCC38. The intensity (I) obtained on the cytosensors showed a linear relation to the logarithmic value of captured cancer cells (N) in the range from 5 × 102 to 106 cells (R = 0.992, n = 9) for MDA-MB-231, which is shown in Fig. 5 (E and F). The linear regression equation is


After partially blocking the Abs sites on the nanoprobes with PD-L1, the decrease in intensity (ΔI) versus the amount of PD-L1 (m) showed a linear relation in the range 1.0 to 5.0 fmol (R = 0.999, n = 5; Fig. 5G)


The average number of PD-L1 sites on each captured MDA-MB-231 cell was calculated to be ~2973, which was more than twofold that on each HCC38 cell. This result suggests that the MDA-MB-231 cells may be more sensitive to the PD-L1 Abs treatment than the HCC38 cells. Comparably, the intensity obtained on the cytosensors (I) for MCF-7 is much lower, and it shows a linear relation to the logarithmic value of captured cancer cells (N) in the narrow range from 5 × 103 to 5 × 104 cells (R = 0.993, n = 5). As shown in Fig. 5 (H and I), the linear regression equation is


In addition, the decrease in intensity (ΔI) does not respond to the change in the amount of PD-L1 (m; Fig. 5J). It indicated that MCF-7 cells express few PD-L1 on the surface (Fig. 5, E to J). The average number of PD-L1 on each captured MDA-MB-231 cell was calculated to be ~2973, which was more than twofold that on each HCC38 cell. Conversely, negligible expression of PD-L1 was shown on the MCF-7 cell surface. Consequently, the expression of PD-L1 on three types of TNBCs is summarized in Fig. 5K: MDA-MB-231 > HCC38 >> MCF-7. This result suggests that the MDA-MB-231 cells might be more sensitive to the PD-L1 Abs treatment than HCC38 and MCF-7 cells. Moreover, compared with MCF-7, which belongs to the subtype Luminal in the American Type Culture Collection breast cancer cell line, HCC38 and MDA-MB-231 exhibit higher PD-L1 expression on cell surfaces. Both HCC38 and MDA-MB-231 belong to Basal B, not Luminal. Therefore, the quantitative characterization of PD-L1 expression on different subtypes of cell surfaces has become an extremely important subject directly related to the development of diagnostic tools for personalized treatment. The high expression of PD-L1 on cell surfaces of HCC38 and MDA-MB-231 indicated that Basal B subtype TNBCs always express more PD-L1 on cell surfaces, which means that this type of TNBC may be more sensitive to PD-L1 treatment than Luminal TNBCs. The expression of PD-L1 in MDA-MB-231 is slightly higher than that in HCC38. Thus, MDA-MB-231 cells may be more sensitive to PD-L1 treatment. In addition, within the same subtype of TNBCs, different cell lines express different amounts of PD-L1 and exhibit different responses to the stimulation of interferon-γ (IFN-γ), which may relate to the different activities of signal pathways in different cell lines, such as JAK/STAT, MEK/ERK, and PI3K/AKT signaling. It has been reported that high PD-L1 basal cell lines have lower expression of IRF2BP2 and higher STAT1 levels, compared with low PD-L1–expressing cell lines (37). A direct comparison for the PD-L1 expression among three types of TNBC cell lines provided direct evidence for further evaluation of the mechanism about the cell signal pathway.

We further investigated the PD-L1 expression on TNBC surfaces using Raman mapping. We collected a topographic image of 50 μm by 50 μm from the integrated intensity of the peak at 1530 cm−1. The first column of Fig. 5L shows the Raman mapping images of the PD-L1 expression level of three types of TNBCs. The color-false images revealed that the expression level of PD-L1 on three types of TNBCs was MDA-MB-231 > HCC38 >> MCF-7. Both the calculated results and Raman mapping images were consistent with those of the commercial enzyme-linked immunosorbent assay (ELISA) kit (fig. S7B). The Raman imaging technique mentioned above was further successfully used to monitor dynamic PD-L1 expression on three different TNBCs in response to IFN-γ, which was a type of biomolecule reported to upgrade the expression of PD-L1 on cell surfaces (38). Three types of IFN-γ–treated TNBCs progressively increased with the concentrations of IFN-γ (Fig. 5, M to P). Besides concentration-dependent mapping, to provide additional evidence that our probe selectively targets PD-L1 on cell surfaces with high sensitivity, we further conducted time-dependent Raman mapping experiments for IFN stimulation. Figure S8 shows the time-dependent responses for IFN-γ during the course of the experiment. The Raman peak obtained from IFN-γ–treated cells increased as the IFN-γ incubation time increased and reached a plateau after 48 hours. Both the histogram and Raman mapping images indicated that IFN-γ treatment significantly stimulated the expression of PD-L1 on cell surfaces within 48 hours. All these results confirmed the high sensitivity and selectivity of the MGT-Abs-CuPc probe for Raman bioanalysis and imaging.


In this work, a novel ternary, plasmon-free MGT SERS probe platform has been first successfully developed for the sensitive and selective detection of various types of TNBCs and for the evaluation of PD-L1 expression at the single-cell level. First of all, a ternary nanostructure has been designed and synthesized layer by layer and used as a plasmon-free SERS substrate. The EF value of MGT for CuPc has been estimated to be 8.08 × 106, which is one order of magnitude higher than the highest value reported by far. Both experimental results and theoretical DFT calculation have revealed that introduction of TiO2 to the MG nanocomposites induces a significant enhancement of the Raman intensity of individual molecules with high selectivity, which should be attributed to the resonance Raman effect of CuPc, CT between GO and TiO2 in MGT nanocomposites, and the enrichment of porous TiO2. The developed SERS substrate has been designed as a nanoprobe providing remarkable properties for quantification of PD-L1 expression on TNBC surfaces with a detection limit down to ~3 cells. Besides, this SERS probe has been used to measure in situ the expression of PD-L1 on TNBC surfaces at the single-cell level and to monitor the expression variation and distribution of PD-L1 during drug treatment through Raman mapping images. Thus, a probe with ultrahigh EF based on CM has been provided, which can also act as a robust and rapid cytosensor for cancer cells and membrane protein with selectivity, sensitivity, and accuracy. In addition, MGT nanocomposites have demonstrated excellent recyclability under visible light. This versatile SERS cytosensing platform has been a promising tool for detection of PD-L1 expression with great clinical value for diagnosis and treatment of human TNBCs, which can be expanded for detection of other trace cancer biomarkers and for reliable diagnosis of cancer and other diseases.


MGT substrate characterization

The structures and compositions of MGT were characterized by SEM, TEM, XRD, FTIR, XPS, and Raman spectroscopy. TEM images were obtained using a JEM2100 transmission electron microscope operated at an accelerating voltage of 200 kV. HR-TEM images were taken using a JEM-2100F transmission electron microscope with an accelerating voltage of 200 kV. SEM measurements were performed using a field emission SEM (FEI Helios G4 UC) operated under UC (UniColore) mode at 1 kV + 1000 V UC. XRD patterns were measured by a D/max2550VB3+/PC x-ray diffractometer using Cu (40 kV, 100 mA). FTIR spectra were recorded on a Nicolet iS10 infrared spectrometer (Thermo Fisher Scientific, USA). XPS data were performed on an XPS (AXIS Ultra DLD, Japan) equipped with an Al Kα (1486.6-eV photons). SERS was performed on a Thermo Fisher Scientific DXR Raman microscope with a 10× [numerical aperture (NA), 0.4] microscope and a 10-mW laser power for SERS measurements.

Synthesis of Fe3O4 NPs

The Fe3O4 NPs were synthesized according to the method reported by Zhao and colleagues (39). FeCl3·6H2O (3.25 g) was dissolved in ethylene glycol (100 ml). Then, 1.3 g of trisodium citrate and 6.0 g of sodium acetate (NaAc) were added. With magnetic stirring for about 30 min, the resulting yellow solution was next transferred to a Teflon-lined stainless steel autoclave (100 ml in capacity). The autoclave was heated at 200°C for 10 hours and then cooled to room temperature. The black products were washed three times with deionized (DI) water and ethanol, respectively. As-prepared Fe3O4 NPs were stored in 5 ml of ethanol without O2 at 4°C for 2 weeks.

Synthesis of GO

GO was synthesized using the Staudenmaier method and developed by other authors (40). Briefly, 5.0 g of graphite was added to a mixture solution containing 87.5 ml of H2SO4 (98%) and 45 ml of fuming HNO3 in an ice-water bath. KClO3 (55 g) was slowly added into the mixture (>15 min). Because of the release of chlorine dioxide gas, all procedures were carefully performed in a fume hood to reduce the risk of explosion. After stirring at room temperature for 96 hours, the mixture was poured into 4.0 liters of water and filtered to obtain the GO.

Synthesis of MGT

First, because the surface of Fe3O4 NPs was rich in carboxyl, the surface charge of Fe3O4 NPs needed to be changed from negative to positive by PDDA. In a typical synthesis, about 5 mg of the prepared Fe3O4 NPs were dispersed in 15 ml of water and mixed with 5 ml of 1% PDDA solution. Then, the mixture was treated with ultrasonication for 30 min at room temperature. After that, 5 ml of the as-prepared GO was added to the mixture. With another 30 min of ultrasonic treatment, the MG nanocomposites were synthesized with a high yield. Last, the products were washed three times with DI water and ethanol with magnetic separation, respectively. The final precipitate was dispersed in 5 ml of ethanol for further use.

The porous TiO2 shell was prepared via a versatile kinetics-controlled coating method reported by Zhao and colleagues. Two milliliters of MG ethanol solution was dispersed in 20 ml of absolute ethanol and mixed with 60 μl of ammonia solution (28%). Then, the mixture was subjected to ultrasound for 15 min. After that, the solution was immersed into a 45°C water bath, and 0.15 ml of titanium tetrabutoxide (TBOT) was added dropwise to the mixture within 5 min. Next, the reaction proceeded for 24 hours with continuous mechanical stirring. The products were washed three times with water and ethanol and dispersed in ethanol. The solution was transferred into a crucible and dried at 100°C overnight. Last, the powder was calcined at 500°C in air for 2 hours.

Synthesis of MTG and cMGT

Typically, MT was synthesized through the same method reported by Zhao and colleagues. Fe3O4 NPs were dispersed in 20 ml of absolute ethanol and mixed with 60 μl of ammonia solution (28%). Then, the mixture was subjected to ultrasound for 15 min. After that, the solution was immersed into a 45°C water bath, and 0.15 ml of TBOT was added dropwise to the mixture within 5 min. Next, the reaction proceeded for 24 hours with continuous mechanical stirring. The products were washed three times with water and ethanol and dispersed in ethanol. The solution was transferred into a crucible and dried at 100°C overnight. The powder was then calcined at 500°C in air for 2 hours. After that, and using the same method as above, PDDA was used to change the surface charge of MT from negative to positive. Last, the GO and MT solutions were mixed together and subjected to ultrasound for 15 min to form the MTG NPs.

Besides, cMGT was synthesized by a similar process. Typically, MG was dispersed in a mixture of 25 ml of ethanol and 7 ml of acetonitrile under vigorous stirring. After the addition of 0.2 ml of ammonia aqueous solution, 0.3 ml of TBOT in 3 ml of ethanol and 1 ml of acetonitrile was injected into the mixture. After stirring for 3 hours, the particles were centrifuged and washed with water and dried in an oven at 80°C for 6 hours.

Recyclability of MGT

CuPc solution (10 mM) was mixed with MGT substrates with continuous stirring for 2 hours. After centrifugation at 8000 rpm for 5 min, the precipitate was collected and dispersed in 5 ml of DI water. The suspensions were illuminated under visible light (460 nm < λ < 700 nm, 1000 W) at different times. Then, the suspensions were separated through centrifugation at 8000 rpm for 5 min or magnetic separation and dispersed in 1 ml of ethanol. Last, 100-μl suspensions were dropped onto silicon foils and dried at room temperature. The silicon foils were measured by Raman spectroscopy under a 633-nm laser.

Raman detection

Salicylic acid, oxalic acid, CC pyrocatechol, dopamine, PTCDA, CV, and CuPc were chosen as different signal molecules. Certain amounts of molecules were dissolved in DI water (for PTCDA, in 0.1 M NaOH solution) to form a uniform solution with 10 mM, and then, 1 ml of this solution was mixed with 5 ml of a solution containing Raman substrate (MGT). The mixture was stirred for an hour. The resulting solution was centrifuged and dispersed in 2 ml of DI water. The final solution was dropped onto the Si/SiO2 substrate and dried at room temperature. Last, a 633-nm laser was used to detect each sample by Raman spectroscopy. The exposure time was 0.5 s, and laser energy was 5 mW.

Synthesis of MGT-Abs-CuPc nanoprobes

CuPc-Abs-MGT nanoprobes were synthesized by immobilized PD-L1 Abs and CuPc on the MGT surface. Briefly, 10 mg of 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide (EDC) was added into 5 ml of MGT solution in water and stirred for 1 hour. Then, 5 mg of N-hydroxysuccinimide and 20 mg of PD-L1 Abs were added to the solution containing MGT and EDC. After reacting with Abs for 2 hours, nanoprobes were separated with centrifugation at 5000 rpm for 10 min. The precipitate was then transferred into 10 ml of phosphate-buffered saline (PBS) [0.1 M (pH 7.4)], and 100 μl of CuPc solution (10 mM) was mixed with Abs-modified MGT solution and shaken for 1 hour. The final precipitate was collected through centrifugation at 5000 rpm for 10 min.

Confocal imaging of MGT-Abs-CuPc and PD-L1 antigen

To validate the recognition ability of SERS nanoprobes in this work, CuPc-Abs-MGT and MGT were incubated with Cy5-labeled PD-L1 antigen in PBS for 1 hour. After incubation, confocal images were collected through a Leica SP8 confocal laser scanning microscope with a 63× oil lens.

Modification of the silicon probe

To capture the HCC38 cell from the solution, silicon foils were modified with SYL3C aptamer (5′-NH2-CACTACAGAGGTTGCGTCTGTCCCACGTTGTCATGGGGGGTTGGCCTG-3′) as a capture platform. First, silicon foils were soaked in 15 ml of ethanol containing 1 ml of 3-aminopropyltriethoxysilane and shaken at 37°C for 4 hours. After that, NH2-modified silicon foils were washed twice with DI water and ethanol, respectively. Then, NH2-modified silicon foils were transferred to the solution with 2.5% glutaric dialdehyde for further incubation. Two hours later, the silicon foils were separated from the solution and washed three times with DI water. Last, 10 μl of 10 μM aptamer solution was dropped onto pretreated silicon foils and incubated for 2 hours at 37°C. As-prepared aptamer-modified foils were used for capture cells immediately.

Cell culture and capture

The HCC38 cells were cultured in a flask in RPMI 1640 medium (Gibco, Grand Island, NY) supplemented with 10% fetal bovine serum (ScienCell) in an incubator (5% CO2 at 37°C). Glucose (4.5 g liter−1), 5 mM MgCl2, transfer RNA (0.1 mg ml−1), and bovine serum albumin (1 mg ml−1) were dissolved in PBS with CaCl2 and MgCl2 to obtain a homogeneous solution (binding buffer). After that, the cells at the logarithmic growth phase were centrifuged at 800 rpm for 3 min to separate them from the medium and then dispersed in the binding buffer to form a cell suspension. Cell numbers were obtained by a cell counter (Nsum). About 100 μl of cell suspension was dropped onto the aptamer-modified silicon probe and incubated at 37°C for 60 min. The cells could be captured by the specific binding between the aptamer and HCC38 cells. The captured cells on silicon foils (N) were calculated by counting the number of cells after capture (Nres) and calculating using the equation N = NsumNres.

PD-L1 Raman scanning immunoassay

The silicon foils that have captured the cell were incubated with 10 μl of the CuPc-Abs-MGT nanoprobes at 37°C for 60 min. To minimize the background, they were washed with PBS buffer (pH 7.4) to remove the free nanoprobe in the solution. Then, after the recognition approach and washing process, the silicon foils were subjected to Raman spectroscopy to collect the sample spectra (each sample was collected 10 times at different points). For the Raman mapping mode, a 50 μm by 50 μm area was selected for imaging. We collected 32 pixels by 32 pixels in 2 hours.

Influence of PD-L1 expression by IFN-γ

Cells in logarithmic phase were selected for this experiment. Different amounts of IFN-γ were injected to the cell media and further incubated for 48 hours. Then, cells were captured and detected with the same process as before.

Calculation of EF

The EF for each nanomaterial was calculated by the following equation

EF =(ISERS/Nads)/(Ibulk/Nbulk)(6)

MGT: Nads and Nbulk represent the number of CuPc molecules in the SERS sample and the normal Raman sample, respectively. ISERS and Ibulk are the same vibration peak of the CuPc molecule on one MGT and the normal Raman spectrum from the solid sample, respectively.

For the calculation of Nbulk, to ensure that CuPc molecules on silicon foils contribute to the Raman signal equally, a concentration-dependent experiment was carried out and shown in fig. S9A. A good linear relationship between the concentration of CuPc and the Raman intensity at 1530 cm−1 was observed, ranging from 1 to 12 mM (fig. S9B), indicating the equal contribution of each CuPc molecule on silicon foils to the Raman signal.

ISERS and Ibulk were obtained on the peak intensity of the CuPc molecule at 1530 cm−1 at 10 different points on MGT substrates (fig. S9, C and D). The average intensity of CuPc at 1530 cm−1 (ISERS = 21,000 and Ibulk = 200) was used for the calculation of the EF value (fig. S9E).

In the experiment, 100 μl of aqueous CuPc solution (10 mM) was dried onto the Si wafer (0.5 cm by 0.5 cm), and Nbulk can be estimated asNbulk=100μl×102M×6.02×1023mol1×2.93μm2/0.25cm2(7)where d is the diameter of the light spot (d = 1.22λ/NA), λ is the incident wavelength (633 nm), and NA of the objective lens is 0.4; thus, the laser spot size [π(d/2)2] is about 2.93 μm2. Nbulk was estimated to be 7.055 × 1010.

Nads was determined by the laser spot illuminating the sample and the density of the CuPc molecule saturate adsorbed on the surface of MGT. Through the UV-vis spectra in fig. S9 (F and G), a linear relationship between the concentration of CuPc (C) and the absorbance (A) was found, ranging from 10 nM to 1 μM


After that, 100 μl of 10 mM CuPc was incubated with 1 g of MGT nanocomposites for 1 hour. Then, the MGT nanocomposites were collected through magnetic separation. The UV-vis spectrum was ascertained to determine the amount of CuPc absorbed on MGT (fig. S9H).

Thus, Cads = (0.0033 − 0.000711)/2/10−3 × 10−7 = 1.3 × 10−7M. MGT is closely arranged on a silicon chip and maximizes the number of MGT under the spots.


Similarly, liquid chromatography–MS was also used as a quantitative method for the number of CuPc absorbed on MGT (Nads-ms). As shown in fig. S9 (I to K), an external standard method was used for the quantification. The ion peak at m/z (mass/charge ratio) 360 was selected as the reference peak and made a linear relationship between peak intensity (I) and CuPc concentration (C). Figure S9J shows the linear relationship range from 10−7 to 10−5 M. The equation is


Considering the intensity of the MS peak of CuPc-MGT, the actual concentration of CuPc (fig. S9K) absorbed on MGT can be calculated


Nads-ms = 1.51 × 10−7 × 100 μl × 6.02 × 1023 × 2.93 μm2/0.25 cm2 = 1.06 × 106, which is a similar result to Nads-UV through UV. Substituting the values of the above variables into Eq. 6, EF could be calculated to be around 8.08 × 106.

MTG: The numbers of molecules on the graphene area Nads could be calculated to be 5.12 × 105 through UV-vis spectra of MG-CuPc as reported. Therefore, the calculation of the SERS EF can be obtained through the same expression. Considering ISERS = 15,200 and Ibulk = 200, the EF was calculated to be ~1.06 × 106.

MG: The process of calculating MG was similar to that of MTG, considering ISERS = 10,400 and Ibulk = 200. The EF was calculated to be 3.05 × 105.

Theoretical calculation

DFT calculation was performed by using the CP2K package. The Perdew-Burke-Ernzerhof functional with Grimme D3 correction was used to describe the system. Unrestricted Kohn-Sham DFT was used as the electronic structure method in the framework of the Gaussian and plane-wave method. The Goedecker-Teter-Hutter (GTH) pseudopotentials and DZVP-MOLOPT-GTH basis sets were used to describe the molecules. A plane-wave energy cutoff of 500 rydbergs was used.

The simulation was carried out in a 31.04 Å by 23.14 Å by 31.70 Å cubic box. A four-layer anatase TiO2 (101) surface, which contains 216 Ti atoms and 432 O atoms, was used in the simulation. A single layer of graphene was covered over the anatase (101) surface.

On the basis of our experiment results, where a Ti–O–C bond was observed, the oxidized graphene adsorbed on the TiO2 surface was modeled. For comparison, we also modeled the oxidized graphene. The molecule adsorbed on the oxidized graphene surface and the TiO2-supported graphene was first optimized.

The charge density difference is defined asΔρ=ρmol/surρmolρsur(8)where ρmol/sur, ρmol, and ρsur are the electron density of the molecule adsorbed on the surface and the individual electron density of the molecule and the surface. Results show that when graphene was supported by the TiO2 surface, the CT was more pronounced.

Correction (13 June 2019): Two funding sources were excluded in the original publication. Funding from the National Natural Science Foundation of China and the Innovation Program of Shanghai Municipal Education Commission have been included in the funding statement in the Acknowledgments section.


Supplementary material for this article is available at

Fig. S1. Characterization of nanostructures.

Fig. S2. Recyclability of MGT.

Fig. S3. Photostability of CuPc of MGT nanocomposites.

Fig. S4. Mechanism of Raman enhancement of MGT.

Fig. S5. Characterization and experimental optimization of MGT-Abs-CuPc.

Fig. S6. Cytotoxicity of the MGT-Abs-CuPc probe.

Fig. S7. Kinetics mechanism of antigen-Abs reaction and ELISA data of PD-L1 expression.

Fig. S8. Raman scanning mapping images of three different types of breast cancer cells incubated at various times.

Fig. S9. Calculation of EF values of MGT.

Table S1. Selected vibrational modes of CuPc.

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.


Acknowledgments: Funding: We appreciate the financial support from the National Natural Science Foundation of China (21635003, 21605049, and 21827814), innovation Program of Shanghai Municipal Education Commission (201701070005E00020), and the Program of Shanghai Sailing Program (16YF1402700). Author contributions: E.F. synthesized and characterized MGT and completed the explanation of the mechanism as well as the quantification and imaging of PD-L1 on the cell surface. T.Z and Y.T. designed and directed the project. X.H. and J.C. obtained and explained the TDR measurement. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors.

Stay Connected to Science Advances

Navigate This Article