Surface-enhanced Raman scattering nanotags for bioimaging

The well-known phrase “seeing is believing” reveals the eternal driving force for humans to develop advanced imaging technologies for the exploration of the universe. As a special way to “see” the micro- and nano-scale biological objects, bioimaging has garnered long-lasting attention in the field of biology and medicine. In vitro bioimaging on the cellular level investigates complex structures and processes that can reveal cell fate and functions. In vivo bioimaging delivers subtle physiological and pathological information on animal models or patients, serving as a translational bridge between fundamental studies and clinical applications.

Optical imaging exhibits advantages with the capability of real-time image acquisition and high spatial resolution, achieved through radiation propagation in tissues and interaction with exogenous optical probes. The commonly studied optical imaging methodology includes fluorescence and Raman scattering. However, fluorescence is restricted by the broad linewidth and autofluorescence background interferences, leading to poor multiplexing ability and specificity. As for the fluorescent optical probes, fluorophores and organic dyes usually lack enough photostability, and quantum dots (QDs) do not fully satisfy biosafety requirements. In contrast, Raman scattering (or Raman spectroscopy) based nanotags have shown promises as optical probes for biological analysis and imaging.

Surface-enhanced Raman scattering (SERS) enables several orders of enhancement of intrinsic Raman scattering by adsorbing molecules on the surface of rough plasmonic nanostructures.¹ With the sensitivity of single-molecule detection level, SERS is, therefore, regarded as one of the most promising weapons in biomedical imaging and detection.^2,3 In recent years, the development of SERS nanotags has garnered much attention. The main components of SERS nanotags typically include the metallic nanoparticles (NPs) as the enhancement substrate and adsorbed Raman reporters as the signal source.³ The surface coating and biofunctionalization further bring better biocompatibility, particle stability, and targeting efficiency. So far, the advances in commercial microscopic and portable Raman spectrometers have made it possible to acquire sensitive and reproducible Raman signals. These developments together provide a platform for the significant progress of SERS nanotags. For bioimaging, SERS nanotags have further presented advantages in fundamental in vitro studies and biomedical intraoperative detections, such as good photostability, encoding capability, and specificity.⁴ 

This tutorial introduces the fabrication, spectral analysis, and bioimaging applications of SERS nanotags. Starting with the principles of SERS imaging, we illustrate the advantages and imaging strategies of SERS in direct and indirect ways in Sec. II. The design and fabrication process of SERS nanotags is further presented, followed by the topics in novel SERS nanotags and Raman reporters in Sec. III. Section IV discusses Raman instrumentation and analysis, to show the Raman scanning, data collection, and spectral analysis method in typical SERS imaging applications. We then focus on the recent progress of using SERS nanotags for biomedical imaging ranging from in vitro cellular imaging to in vivo tumor and multimodality imaging. At last, we provide perspectives projecting into the possible obstacles and promises of SERS bioimaging in future translational studies. More information on SERS (such as enhanced Raman spectroscopy, SERS sensor applications, trade-offs and future developments, etc.) can be found from other related reviews.^5–7 

Raman scattering opened a new era for research and analysis since it exhibits the fingerprint nature of molecular vibrations, similar to infrared absorption spectroscopy but offering complementary information and opportunities for different applications (e.g., working in water). Derived from the techniques of Raman spectroscopy and plasmonics, SERS appears as a promising analytical tool for bioimaging applications.

Raman spectroscopy is the inelastic part of light scattering generated when incident light interacts with molecules. It is distinguished from the elastic (Rayleigh) scattering (Fig. 1). Raman scattering shows changed frequency between incident (ω) and scattered photons (ω_υ), corresponding to the energy uptake or give-off generated by the interaction of the photon with molecules, and the latter in this process change their internal energy (typically due to vibrational transitions). The red-shifted frequency (ω – ω_υ) is defined as the Stokes shift and corresponds to a photon energy loss in the scattering process, while the blue-shifted frequency (ω + ω_υ) as the anti-Stokes shift and corresponds to a photon energy gain. In most cases, the Stokes shift dominates Raman scattering because most molecules at room temperature are in the ground electronic and vibrational state due to the Boltzmann distribution.

The main obstacle in Raman spectroscopy applications has long been its inherent small cross section (usually only 10⁻²⁸ cm² sr⁻¹), which is much smaller than that of fluorescence (typically about 10⁻¹⁶ cm² sr⁻¹).^9,10 This brings challenges to obtain Raman signals with satisfying intensities. To circumvent this problem, SERS was developed by combining of plasmonics and Raman scattering, bringing enormously enhanced Raman signals.^11,12

SERS is widely recognized to be generated by two mechanisms: electromagnetic (EM) enhancement and chemical enhancement (CE). For the metallic nanostructure, the quasi-free electron clouds are pulled back and forth harmonically with the incident EM field against the Coulombic force between the electron and the nuclei [Fig. 2(a)], leading to an enhancement of the localized EM field. The plasmonic nanostructures with sharp tips,^13–15 narrow gaps,^16,17,179 and aggregates [Figs. 2(b) and 3(a)] generally provide more enhanced Raman scattering² and their localized surface plasmon resonance (LSPR) wavelength^18,19 can be tuned by the factors, e.g., NP composition, size, alloy proportion,^20,21,180 etc. Typically, LSPR-matched/on-resonance excitation [Fig. 3(b)] can arouse more intense SERS scattering but bring about side effects such as photoinduced degradation, so practical concerns are indispensable in order to balance the enhancement with molecular photostability.^22–25 Nevertheless, on-resonance heating effect can be turned into a competitive advantage when applied to photothermal therapy (PTT).^26–28 EM provides a more salient contribution to the enhancement with a high enhancement factor (EF) from 10⁶ to 10¹⁴,¹² while CE acts as a compensate enhancement (about 10²–10⁴). Though not fully understood so far, the mostly recognized process of CE involves (i) the increased polarizability of the molecules involved in the process, (ii) molecular electronic excitation promoted by the photon excitation, or (iii) a special condition in which the orbitals involved in the electronic excitation are localized in different parts of the system (corresponding to charge transfer transitions), i.e., the electron initially localized in a metal orbital is promoted into a molecular orbital localized on the organic molecule.² Common practices such as using fluorophores as Raman reporters excited by a laser wavelength close to the maximum absorbance of the fluorophores can bring about a notable signal improvement.^29–31,179

SERS shows favorable properties as an analytical tool in bioimaging.^32,33 It possesses capabilities of multiplexing owing to the narrow bandwidths (typically ∼0.1 nm) of the relevant signal compared with that of the fluorescent emission (20–80 nm). In addition, one laser is capable to excite Raman features of multiple molecules, and incident wavelengths ranging from visible to near-infrared (NIR) are all applicable. Therein, NIR excitation can minimize the disturbance of autofluorescence from biosamples³ and alleviate the phototoxicity^34,35 to improve the imaging contrast and the long-term viability of living subjects. Furthermore, SERS nanotags tend to have better photostability than typical fluorophores,^25,36 which benefits time-lapse SERS bioimaging.

The SERS bioimaging strategies can be divided into two categories: direct and indirect imaging. The direct strategy (also known as the label-free strategy) is to use SERS substrate to directly enhance the structural and locational information of target biomolecules. Kang et al. developed a technique called “Targeted Plasmonically Enhanced Single Cell Raman Spectroscopy” to image and monitor the molecular changes of targeted cellular components during the full cell cycle.³⁷ They used peptide-functionalized gold (Au) nanospheres to target the cell nucleus. The characteristic peaks from cellular biomolecules allowed Raman imaging to show the position of the nucleus. Also, by tracking the dynamic changes of Raman bands attributed to different molecules (e.g., guanine, adenine, protein), the structural and molecular information of a single cell over the entire cycle was revealed. Similarly, Ando et al. demonstrated the usage of SERS imaging along with the dark-field scattering to record the particle trajectory inside a cell and analyze the biochemical information along the cellular transport pathway.³⁸ They located the Au NPs via the dark-field microscope and collected the corresponding SERS spectra. By adopting three Raman bands CH₂ and CH₃ vibrations in lipids and proteins at 1457 cm⁻¹, Amide II in most proteins at 1541 cm⁻¹, and the phosphate vibration at 977 cm⁻¹, a triple-color molecular Raman mapping of cellular pathways over time was constructed at a spatial resolution of 65 nm and a temporal resolution of 50 ms. This work demonstrated the prospects of direct SERS imaging in studying dynamic biological functions.

The indirect strategy (also known as the labeled strategy) relies on the Raman signals from exogenous agents to mark, identify, locate, or track the targeted analytes. The key to this strategy is the application of imaging contrast agents, SERS nanotags, which will be elaborated in Sec. III. Created by chemically binding specific Raman reporters to the surface of metallic NPs, SERS nanotags exhibit strong characteristic signals and could be used for optical labeling. They allow bioimaging and detections with high specificity, high sensitivity, and excellent multiplexing capability.

Generally, the direct SERS method is more appropriate for the identification of analytes endowed with rich aromatic rings and unsaturated bonds but can lead to rather weak Raman signals when applied for analytes without these groups.³⁹ In addition, it faces significant challenges in complex and heterogeneous biological environments due to protein non-specific adsorption on NP surface, strongly affecting the adsorption of targeted analytes and circulation time of NPs. In contrast, the indirect method using SERS nanotags is more applicable to target a variety of molecules and cellular organelles, showing high advantages and a wider application ranging from fundamental studies to clinical trials.

The development of SERS nanotags is a significant advance in the field of biological imaging and detections. So far, a variety of SERS nanotags with fascinating plasmonic structures and suitable Raman reporters have been explored with the goal of higher SERS enhancement, better specificity, improved uniformity of hot spot distribution, and the capability against background interferences.

SERS nanotags are typically comprised of metal substrates, Raman reporters, protection layers, and targeting molecules (Fig. 4). Each part has its own functions and contributes significantly to the overall SERS performance of tags. The metal substrate acts as a Raman signal amplifier due to strong enhancement of EM fields produced upon laser irradiation. The Raman reporter, adsorbed on or close to the substrate, generates the SERS fingerprint signatures. Then, a protection layer is usually coated onto SERS nanotags for better stability and biocompatibility, and easier surface modifications. Further binding of targeting molecules to SERS nanotags can endow the NPs with the ability to specifically recognize and bind to the targeted analytes. Each synthetic step is guided by individual design principles, which should all be taken into consideration for optimal Raman performance.

SERS substrates play a vital role in tag design. Their chemical composition, geometry, and size distribution can dramatically influence the Raman performance. Au^36,40 and silver (Ag)^41,42 nanostructures are the most widely used SERS platforms owing to their easy synthesis through wet chemistry protocols and strong plasmonic properties.³ Compared to Ag NPs, Au NPs have many advantages including easily controlled size distribution, good chemical stability, and biocompatibility, although the Ag NPs usually exhibit higher SERS enhancements.^43,44 As for the particle morphology, nanospheres are the most common choice for SERS substrate. However, nanospheres can provide limited enhancement ability, and are not suited to biological-friendly NIR laser excitation due to their typical LSPR in the range of 400–600 nm.⁴⁵ An alternative to nanospheres is the nanorod (NR) with two LSPR bands, a weak transverse band in the visible region and a strong longitudinal band in the longer wavelength region.⁴⁶ The longitudinal LSPR can be tuned from the visible to the NIR region by changing the aspect ratio of NRs,⁴⁷ thus possibly suitable for a broad range of excitation wavelengths. Nanostar particles, which consist of a spherical core and protruding tips, have also been extensively investigated. They can generate highly enhanced and localized EM hot spots at their sharp tips due to the lightning rod effect, thus providing remarkable Raman enhancement.⁴⁸ It should be noted that protective coatings such as thiolated compounds are usually applied to nanostars to maintain the stability of the arms.^49,50 In addition, the control over the lengths and numbers of branches and their spatial arrangement around the core is rather challenging for the synthesis of nanostars, especially for the Ag ones.^51,52 Another SERS substrate is the metallic core–shell structure NP with a nanogap in between. It can produce uniform and intense hot spots in the gap region and result in stable and reproducible signals owing to the shell protection.^36,53 Besides geometry, the particle size should also be taken into consideration for tag design. The small NPs may not sustain strong plasmons, while too large NPs can lead to radiation damping effects that decrease the enhancement factor.⁵⁴ Also, particle size is considered as one of the factors that tends to affect the cellular uptake efficiency and final intracellular localization of NPs,^55–58 which can then exert influence on imaging effects.

Raman reporters provide characteristic Raman signals for identification. Several principles should be taken into account in selecting reporters: (1) the molecules should have the ability to adsorb on SERS substrates because CE requires chemical bonding and EM enhancement is strongly distance-dependent; (2) reporters with large Raman cross sections are preferred for stronger Raman signals; (3) reporters must produce signals with unique spectral profiles and are preferred to exhibit at least one characteristic Raman band that does not overlap with the others when applied for multiplexed labeling; (4) surface-enhanced resonance Raman scattering (SERRS) effect occurs when the excitation laser overlaps with the optical absorption wavelength of molecules,⁵⁹ which brings further enhancement. At present, aromatic molecules with thiol groups are usually chosen as Raman reporters, such as 4-aminothiophenol (4-ATP),¹⁷ methylbenzenethiol (MBT),⁶⁰ 1,4-benzenediol (1,4-BDT),^14,61,62 4-nitrobenzenethiol (4-NBT),^36,63 and 4-mercaptobenzoic acid (4-MBA).⁶⁴ These molecules have high Raman cross sections and characteristic signals, are free from fluorescence interference, and can covalently bind to metallic substrates. In addition, a special group of molecules termed silent-region (SR) Raman reporters exhibit characteristic Raman peaks in the biology-silent window (1800–2800 cm⁻¹). Since bio-tissues usually have abundant spectral information in the fingerprint region (700–1800 cm⁻¹), SR molecules can thus avoid background disturbance from bio-samples.⁶⁵ Representative SR molecules include Prussian blue (PB),⁶⁶ 4-mercaptobenzonitrile,⁴⁰ and so on. More details about SR molecules can be found in Sec. III D. Apart from small molecules, organic fluorescent dyes have also been adopted as Raman reporters^42,67 and their absorption peaks typically range from the visible to NIR region (the most frequently used excitation laser range). This may bring ultrahigh Raman intensity due to the SERRS effect. Their issues lie in their fluorescence background and poor binding affinity to metal NPs that result in unstable and limited experimental Raman signals. As for the addition method of Raman reporter molecules, it is crucial to consider if they are water soluble or hydrophobic ones. The water-soluble molecules can be directly added into aqueous NPs. While the hydrophobic ones could be first dissolved in organic solvents such as ethanol, dimethyl formamide (DMF), or dimethylsulfoxide (DMSO), they can then be transferred in small amount to the NP colloids. The NPs coated by hydrophobic molecules can be encapsulated by amphiphilic layers, which render the core hydrophilic for biological applications. For the encapsulation, a non-exhaustive list of the polymer or co-polymer of polyvinyl pyrrolidone (PVP), poly(maleic anhydride alt−1-tetradecene)-bis(aminohexyl)amine, poly (vinyl alcohol)-polystyrene can be considered.^68,69

Surface coating is generally applied to SERS nanotags to improve their stability and biocompatibility, since NPs in the absence of a protection layer would rapidly undergo coalescence and precipitation in physiological conditions. Also, the surface coating can alleviate the issue of NPs being covered by proteins in biological environments where large amounts of proteins exist.⁷⁰ The commonly used surface-coating materials include bovine serum albumin (BSA), polymers, and silica shells.^3,71 BSA adsorbs onto metal surfaces via weak interaction and forms a biocompatible protective shell after being mixed with NPs.^72,73 Nevertheless, it fails to provide sufficient stability for SERS nanotags suspended in media, buffer, or strongly acidic or basic solutions due to its delicate interaction with metal surfaces.⁷⁰ The thiolated polyethylene glycol (SH-PEG) is a polymeric coating that can be firmly conjugated to SERS nanotags, showing lots of merits including low toxicity, adjustable coating layer thickness, retention of particle morphology even after endocytosis, excellent in vivo biodistribution, and pharmacokinetic properties.^74–78 In addition, rational selection of PEG molecules can provide pendant groups amenable to further surface modifications, e.g., HS-PEG-NH₂ enabling further functionalization with RGD (Arginine-Glycine-Aspartic Acid) peptides.⁷⁸ A newly developed polymer coating for SERS nanotags is polydopamine (PDA).^42,79,80 Polymerized from dopamine monomers, PDA is able to tightly attach to NP surfaces via a simple and rapid chemical process.⁸⁰ The PDA layer could be tuned very thin in thickness (a few nanometers), improving biocompatibility of NPs and providing further conjugating sites for targeting ligands. The most exciting part for this technique is that virtually every Raman reporter, with or without an anchoring group, can be incorporated onto NP surface during dopamine polymerization,⁸⁰ thus significantly broadening the selection of Raman reporters. Silica coating is another widely used encapsulation method, endowing SERS nanotags with high stability, low nonspecific binding, excellent biocompatibility, and the ability of further surface modification.³ Silica coating was first reported more than one decade ago, and has now been commercialized for quite some time.^81,82 Silica shells are typically grown according to the modified Stöber method,⁸³ where tetraethyl orthosilicate (TEOS) is hydrolyzed in ammonia and then grows onto Raman reporter-decorated NPs to form a shell. Additionally, the synthesis of mesoporous silica around SERS nanotags has also been explored in cases where surfactants such as cetyltrimethylammonium chloride (CTAC) exist, adding flexibility to SERS coating.^36,84

Targeting molecules allow NPs to bind to certain biological targets with high specificity and affinity. Active targeting can be realized using peptides,⁸⁵ antibodies,⁸⁶ and aptamers,⁸⁷ which can recognize and bind to specific biomarkers, such as proteins overexpressed by tumor cells. Antibodies have achieved widespread usage, taking advantage of mature fabrication methods and general applicability for various biomarkers. However, conformational change of antibodies can occur upon their binding to NPs, consequently altering the protein function and decreasing their affinity for targets.⁷⁰ Peptides and aptamers show better stability and reduced sensitivity to conformational changes, but they are limited to some extent by lack of known oligonucleotide sequences in targeting biomarkers for diseases like breast and prostate cancer.⁷⁰ As a result, the targeting moieties should be specifically determined under different circumstances. The linkage of targeting molecules to NPs can be realized by several means. First, sulfhydryl group-containing molecules (e.g., thiolated aptamers) can directly bind to metallic NPs.⁸⁸ Second, surface coating on SERS nanotags could provide binding sites. For example, in the presence of coupling reagents such as 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide (EDC) and N-hydroxysuccinimides (NHS), carboxylic acid groups on surface-coating molecules are activated for the reaction with the amine groups in the antibody.⁸⁹ Similarly, the biotin-functionalized nanotags can be easily conjugated with streptavidin-tagged biomolecules,⁹⁰ and the silica shell coating can be modified with a variety of organic molecules via silane chemistry for further targeting functionalization.⁹¹ Third, positively charged antibodies could adsorb onto negative NPs via electrostatic interaction.⁹² 

Gap-enhanced Raman tag (GERT) is a category of the core–shell SERS nanotags with interior gaps [Fig. 5(a)]. They exhibit multiple advantages, including excellent and uniform Raman enhancement, high photostability, flexibility in structural construction, Raman reporters, surface protection and functionalization, exhibiting extraordinary potential in biomedical fields.⁹³ 

In terms of the initial consideration of forming a gap between two metallic layers, the nano space in between generates an intense EM field. This gap-enhanced effect has been proved to bring about 3 orders of magnitude higher Raman intensity compared to the plasmonic spherical NP.^14,62 So far, heterogeneous core and shell structures have been explored to provide more hotspots, e.g., nanostar core,¹⁴ nanotriangle core,²⁸ nanocube core,¹⁵ spiky shell,¹⁶ petal-like shell,⁴⁰ etc. [Fig. 5(a)]. For example, 20-fold enhancement by incorporating nanostar,¹⁴ or nanotriangle core,²⁸ and an EF as high as 5 × 10⁹ of GERTs with petal-like shells can be realized.⁴⁰ EFs at the sharp tips or narrow corners of these structures can be even much higher,^15,16 which can be calculated by numerical simulations. The core size and the shell thickness (or their relative ratio) also play an important role in determining the EF, opening for the unlimited optimization of core–shell coupling EM field.^94–96 Bimetallic core–shell NP is another common way to promote higher Raman intensities by introducing Ag.^94–96 Besides, the plasmonic resonance changes accordingly as the nanostructure of the GERTs, being entitled to the tunability to fulfill the demand of different incident laser wavelengths and after all, different clinical scenarios.

The embedded insulation layer, which is usually Raman-active, can be generally categorized by three types: aromatic molecules with thiol groups,^14,61,62 fluorescent dye-coded DNA,^97–99 and polymers with Raman-active parts.¹⁶ The nanogap size and the growth process of the shell can be influenced by factors such as molecular lengths and densities.^100,181,182 The flexibility of selecting the embedded Raman reporters opens the possibilities in encoding and quantifications. To expand the encoding capacity, the developed strategies include substituting different Raman reporters in the gap,^26,101 using multiple Raman reporters of different ratios in one gap,¹⁰² and constructing multiple layers with different molecules.⁶² Combined with surface functionalization, different nanotags can be modified with biocompatible coatings or specific antibodies to recognize the target cells, realizing cell imaging,^26,42 distinguishment,¹⁰³ and encoding.¹⁰⁴ 

GERTs can also serve as nanotags with internal standards for quantification to minimize the inaccuracy of signal fluctuations.^26,105 The special core–shell structure of GERTs not only separates the embedded Raman reporters and surface targets to avoid adsorption competition, but also protects the Raman reporters from being affected by NP aggregation, ensuring good signal reproducibility and uniformity.^15,105,183 Lin et al. utilized self-assembly Au@Ag nanocube GERTs with internal standard molecule 4-MBA for aspartame (APM) quantification [Fig. 5(b)].¹⁵ The corrected SERS intensity of APM to 4-MBA (red curve) shows improved R² than the intensity of APM only (blue curve), showing the potential of GERTs for quantitative SERS examination.

Other relevant literature reported the photostability and photothermal effects of GERTs. Jin et al. developed the NR GERTs for off-resonance and on-resonance excitation [Fig. 5(c)].²⁶ The off-resonance incidence can be used to improve photostability and reduce harmful thermal effect,^25,36 while the on-resonance incidence to facilitate photothermal therapy (PTT),^26,28 or smart drug release system.¹⁰⁷ By virtue of the layer-by-layer synthesis, multiple functions are endowed to GERTs, e.g., Raman-fluorescence bimodal imaging by embedding fluorophores as Raman reporters.¹⁰⁸ Therefore, GERTs show great potential as the multifunctional and multimodal bioimaging contrast agents. More information on the fabrication, optical properties, and bio-applications of GERTs can be seen from Ref. 93.

The word “satellite” is vivid and exactly depicts the nanostructure that multiple smaller NPs assemble around the core NP. Typically, three key points are involved in this construction: the core NP, the satellite NPs, and the “glue” molecules that connect the former two parts.¹⁰⁹ The main consideration for designing a satellite structure lies in generating more hot spots in the gaps or corners via the adjacency of core and satellites or one satellite and another, because the EM enhancement by a bare metallic sphere is very limited. Au or Ag NPs are commonly used as satellites assembling around the core (either metallic or nonmetallic, e.g., Au, Ag, SiO₂, and Fe₃O₄). The structure of core-satellites has evolved along the route of improving the Raman enhancement capability, such as heterogeneous or rough core surface to bring about a large number of more intense hot spots. For example, Feng et al. utilized Au NRs as the core to obtain a higher enhancement factor, thus improving the limit of detection (LOD).¹¹⁰ Furthermore, the size of the core, the satellites and the gaps,¹¹¹ the density of satellites,¹¹² as well as the metal types¹¹⁰ all determine the plasmonic resonance, so it is usually critical to cohere the incident laser wavelength with the LSPR peak for better EM enhancement. In addition, a magnetic core such as Fe₃O₄ is able to enhance the SERS signal via magnetic amplification strategy, together with its advantages of high dispersibility, biocompatibility, and tunability. Qi et al. developed the multifunctional magnetic–plasmonic assembly nanoprobes using Fe₃O₄ core and Au satellites.¹¹³ These NPs can be electrically stimulated to release activated caspase-3 and induce cell apoptosis, thus showing biomedical prospects in enzyme detection in the process of diagnosis and treatment.¹¹³ 

Owing to a large number of hot spots in plasmonic core-satellite NPs, they exhibit enormously enhanced signals and are, therefore, applied in SERS detection. For example, portable SERS lateral flow assay used core-satellite NPs as SERS nanotags to achieve absolute quantification and simultaneous multiplexing detection.¹¹⁴ SERS-fluorescence hybrid nanotags were reported by using 4-MBA and rhodamine B isothiocyanate (RhB) as Raman reporters for switchable multiplexing by three laser wavelengths.¹¹⁵ The specialty of core-satellite SERS nanotags is endowed by the “glue” molecules connecting the core and the satellites. The imagining dynamic releasing process of the “satellites” when functioning, honored to be “smart”! Such a tactic achieved by glue molecules shares some similarity with Förster resonance energy transfer (FRET) tags,¹¹⁶ generating intense SERS signals of Raman reporters adsorbed on satellites of initial assemblies, but experiencing a signal loss linearly related to the release of satellites. The separation of satellite can be launched by the biorecognition of the targets by glue aptamer molecules, thus leading to the dissociation from the core. Figure 6 shows the DNA-glued core-satellite for indirect/labeled detection and quantification of aflatoxinB1 (AFB1) or Mucin-1 aptamers.^13,110 For both designs, the satellite NPs with target aptamers are originally binding on the core via complementary oligonucleotides in the condition without targeted molecules, while the aptamers sense and bind the targets in the presence of target molecules, finally resulting in the dissociation of Ag NP satellites from the core, accompanied with the reduced SERS intensity. The detection with a wide dynamic range and low LOD has been achieved by using core-satellite SERS nanotags in multiple fields such as food examination,^13,117,118 clinical test,^110,114,118 bio-substance localization,^113,119 and biomonitoring.^113,120 Similar to core-satellite structures, the newly emerged “click” SERS also took advantage of the splicing nanotags.¹¹⁸ Briefly, the controllable splicing of the SERS-active NP assemblies was derived from biomarker targeting (e.g., DNA hybridization), which further mediated the generation of hotspots and SERS signal.¹¹⁸ The combined signals render dual judgments of the target object by assessing the Raman position and intensity with different Raman reporters. This “click” SERS strategy has been applied for the selective homogeneous sensing of intracellular capspase-3.¹²¹ 

In addition, core-satellite NPs can be utilized to perform direct (label-free) detection by virtue of the rich hot spots on the surface. Typically, during the process of free diffusion in the liquid mixture of NPs and analytes, analytes can be detected when they come into the hot spots on the NPs.¹¹⁷ 

The endogenous background interference would lead to the inevitable yield of noise and a relatively poor signal-to-background ratio. To reduce the interference from the bio-environment, a special set of Raman reporters named bio-orthogonal molecules were reported.⁶⁵ They have characteristic peaks in the bio-silent region of 1800–2800 cm⁻¹, where tissues have barely any Raman signals.¹²¹ The bio-orthogonal Raman reporters benefit from accurate distinguishability, localization, detection, and quantification.¹²² The most commonly utilized chemical groups such as alkyne, nitrile, and some isotopes substitutions (such as ²H and ¹³C) are considered background-free, bio-orthogonal, and, particularly, highly multiplexable facilitated by the tunability of the Raman bands via chemical moieties,¹²² ion replacement,¹²³ isotopes,¹²⁴ etc. [Fig. 7(a)].^125,126 Recently, many research studies have cast on the delicate selection and modification of SR Raman reporters and combined them with other EM and chemical enhancement to realize their full potentials in clinical practice and theoretical studies.

PB molecules with rich cyanide bridges (—C≡N—) and broad absorption peaks (500–900 nm) outstands for a strong and sharp single vibrational peak at 2156 cm⁻¹ and Raman resonance enhancement. Yin et al. synthesized PB-assembled Au NPs and showed its ability in immunoassay to test human IgG antibodies and in cancer cell targeting imaging with folic acid functionalization.⁶⁶ He et al. co-modified 4-NBT with —C≡N groups to 4-MBA molecules, and used them to conjugate with sialic acids (SAs) in order to quantify the SA expression level on cancer cells regarding to the drug influence.¹²⁷ 

The Raman multiplexing can be realized by using different kinds of SR molecules with different chemical bonds. For example, Wu et al. developed SERS nanotags modified with benzazido (2010 cm⁻¹), benzalkyne (2142 cm⁻¹), and 1,4-diphenylbuta-1,3-diyn-1-yl moieties (2209 cm⁻¹), respectively.¹²⁸ Li et al. also used SERS nanotags containing Raman reporters with either an alkyne or nitrile groups for profiling multiple biomarkers expressed in cancer cells without signal overlapping.¹²⁹ Furthermore, the derivatives from one farther structure and isotopes can expand the encoding capacity. Zou et al. reported a ratiometric control of C12 and C13 to facilely shift the Raman 2D-band ranging from 2600 to 2706 cm⁻¹, designed for the pattern recognition of targeted cancer cells.¹²⁴ Based on previously mentioned PB, Prussian blue analogs (PBAs) generated by metal ions replacement (i.e., Pb²⁺, Co²⁺, Cu²⁺) advanced a supermultiplex bacteria barcoding model to an encoding capacity of 2ⁿ – 1.¹²³ The concept of Raman palette was proposed by Chen et al. based on different substituent groups linked on the father structure (benzene rings or terminal alkynes).¹³⁰ Similarly, Hu et al. engineered the polyyne-based materials for optical super-multiplexing. They have achieved 20 distinct Raman frequencies through combining the rational design of conjugation length, bond-selective isotope doping and end-capping substitution of polyyne, leading to a theoretical capacity as high as 3 × 10¹³, which is sufficient to barcode all cells in the human body (about 10¹³ cells).¹³¹ 

Raman chips have also been explored with SR-featured SERS nanotags. For example, Zou et al. designed a nanostructure consisting of a large number of Au NPs decorated on multilayered graphitic magnetic nanocapsules to quantify the analyte.¹³² The nanocapsules demonstrated a unique Raman band from the graphitic component, which was localized in the SR, making them an ideal internal standard for quantitative Raman analysis without background interferences. Kong et al. applied a triosmium carbonyl cluster−boronic acid conjugate as the secondary carbohydrate probe in a SERS-based assay for quantitative SERS in medical detection, where the CO stretching vibrations of the metal carbonyl were served as the SR labels.¹³³ Combined with QDs, the SERS-fluorescence joint tags produced by Wang et al. put the high-throughput encoding capability for biomolecular recognition and quantification on an immunoassay chip into a reality [Fig. 7(b)].¹⁰¹ Overall, SR molecules have held great promises toward the quantitative and multiplexing biomedicine and cell imaging.

The advances in commercial microscopic Raman spectrometers and portable Raman systems allow the acquisition of the sensitive and reproducible Raman signals. The developments in SERS nanotags, advanced scanning mode, and spectral analysis method together promote the continuous growth of SERS imaging applications.

Based on various applications, Raman instruments for signal acquisition can be classified into several categories.¹³⁴ Herein, we mainly focus on the routinely used confocal Raman spectrometer and portable Raman system. They differ partly in configurations and exhibit respective advantages.

The confocal Raman system is mainly composed of five parts: narrow-band laser, automated sample stage, optical components, detector based on two-dimensional (2D) array charge coupled device (CCD), and computer control unit that controls, displays, and processes the measured data.¹³⁵ The optical components transmit the excitation laser and collect the scattered Raman photons, mainly including objective lens, mirror, Rayleigh filter, pinhole, diffraction grating, and so on. The confocal Raman spectrometer demonstrates a high detection sensitivity owing to the high-performance CCD (especially electron-multiplying CCD, EMCCD) and a high spectral resolution (down to less than 1 cm⁻¹) due to fine gratings and CCD with more pixels.⁴⁰ Aided by the automated sample stage and 2D array CCD, X–Y mapping can be accomplished via the confocal Raman system. The diameter of laser spot can be focused below sub-micrometer range with high numerical aperture objectives, thus enhancing the image resolution of Raman mappings. In addition, the pinhole can prevent the entrance of scattered light away from the focal plane into the detector, hence improving the mapping resolution along Z-axis.¹³⁶ Moreover, the confocal Raman instruments can be equipped with several laser wavelengths simultaneously (e.g., 325, 532, 638, 785, and 1064 nm) and powerful affiliated software for data manipulation.

Despite the above merits, the confocal Raman system suffers from problems of high cost and inconvenience. The development of fiber optics technologies has greatly boosted the applications of portable Raman readers for cheaper and more flexible measurements.^{135,137–139} The configurations of portable Raman are simplified mainly consisting of a laser, a detector based on linear-image CCD, a fiber optic probe, and a control unit.¹³⁵ This kind of Raman instrument takes the advantage of portability, especially suitable for rapid and real-time detection. However, automatic Raman mappings cannot be performed using a portable spectrometer at present due to the use of less performant CCD and the lack of a mechanical sample stage. Also, its laser spot is much larger (hundreds of micrometers to a few millimeters) due to its simplified optical focusing system (which could be possibly beneficial to illuminate a wide area of interest in rapid clinical diagnosis). Nevertheless, the portable Raman system is limited by reduced spatial resolution and light throughput due to the space between fiber cores.¹⁴⁰ Also, they are equipped with limited laser wavelengths (usually no more than two).

Based on their respective advantages, these two types of Raman instruments are appropriate for different application situations. The confocal Raman system is widely used in labs for fundamental research studies, which requires enhanced sensitivity, high spatial/spectral resolution, and high-quality imaging capability.^{36,40,61,141,142} In contrast, the portable Raman spectrometer is more applicable in field test, substance identification, real-time detection, rapid diagnostics, and intraoperative surgery guidance.^{114,137–139,143}

The data acquisition of Raman imaging can be characterized as two categories: scanning-based imaging and wide-field imaging.¹⁴⁰ The most often applied is the pixel-by-pixel point scanning [Fig. 8(a)], which can be implemented on basic Raman microscopes by moving mechanical stage.¹⁴⁴ It provides good spectral resolution by collecting a highly resolved spectrum at each point of pixels on the sample but suffers from the relatively long acquisition time. As a special example of point-scan, Horiba spectrometer Duoscan^TM mode uses a combination of two galvo mirrors that make the laser beam scan across a mapping area. In this case, a “macro laser spot” for a large area with a scanning step size down to 50 nm could be generated. The imaging speed can be greatly accelerated through the manipulation of the laser spot instead of mechanical movements of the sample stage.

The line scanning [Fig. 8(b)] can be applied to accelerate the imaging process with sufficient temporal resolution. By illuminating with a linear laser beam rather than an intense spot, line scanning obtains spectra from a line of points and potentially prevent laser-induced sample damage with an enlarged laser spot. Okada et al. have reported the slit-scanning technique using line-shaped illumination, to conduct label-free observation of dynamic cytochrome c translocation in apoptotic cells with a temporal resolution of less than 10 min.¹⁰⁶ Similarly, He et al. developed a fast line-scan Raman imaging method, which was realized by widening the slit and laser beam and scanning the sample with a large scan step. The imaging quality was further improved by a deep learning model that was trained to transform low-resolution images acquired at a high speed into high-resolution ones.¹⁴⁵ 

The wide-field imaging illuminates a region and detects the emitted photons within an area by using a 2D detector array.¹⁴⁰ The wavelength scan is conducted by acquiring images at one wavelength channel each time [Fig. 8(c)]. Spectral resolution around 1–5 nm can be reached by using monochromators, filter wheels, or electronic tunable filters such as liquid crystal tunable filters (LCTFs) and acousto-optic tunable filters (AOTFs).¹⁴⁰ Another wide-field imaging is the multifocal imaging mode [Fig. 8(d)], by capturing a snapshot to acquire both spatial and spectral information.¹⁴⁶ It applies linear or 2D beam patterns but arranges excitation beam arrays throughout an area with the use of optical elements or prism arrays, etc.¹⁴⁴ It reduces the required number of spectra in Raman imaging by taking advantage of spatial correlation within samples, effectively shortening the overall acquisition time.¹⁴⁷ Nevertheless, a balance between spatial and spectral resolution is required.

Both confocal Raman system and fiber optics based portable Raman reader can be operated in either scanning or wide-field imaging modes. Even though the wide-field approach was believed to be able to image a large area with a shorter acquisition time, this technique has not been widely used. The reason is that it has to take a longer integration time to obtain an image with strong-enough signal. Therefore, brighter SERS nanotags are still in need to work together with the wide-field scanning method to further improve the image speed.

The collected SERS spectra are endowed with substantial spectral information related to the position and intensity of Raman bands and can reflect the spatial distribution of bio-substances through imaging. However, the raw data usually contain unwanted interferential information or represent complicated mixtures of multi-components that are required to be separated. Therefore, it is fundamental to perform data analysis and processing to acquire reliable results.

Raman spectra obtained using CCD detectors can sometimes be disturbed by the cosmic ray resulting in sharp and intense spikes overwhelming Raman bands.¹⁴⁸ Thus, spike removal should be conducted first before other analyzing procedures. Most Raman systems are equipped with spike removal routines. Next comes the outlier detection, which aims to investigate anomalous spectral behaviors. Outliers can either arise from an extreme sample (a sample that belongs to the test set but owns an extreme property value) or a measurement error. The former is characterized by a high Hotelling's T2 and generally does not skew the model (built afterward if needed) in a high degree, although eliminating the sample is usually a better choice; the latter is characterized by high Q residuals and should be excluded from the dataset.¹⁴⁸ Preprocessing is then applied to rid or mitigate the signal not from the analyte, which usually includes the following steps: (1) smoothing, which aims to remove random noise while preserving meaningful spectral information. The Savitzky–Golay algorithm is the most popular smoothing method, which is based on a polynomial equation fitted in a least squares sense within a predefined interval of spectral points;¹⁴⁹ (2) baseline correction, the goal of which is to eliminate background signals such as fluorescence interference. The frequently used approaches of baseline correction include polynomial baseline correction, rubber-band baseline correction, Whittaker filter, asymmetric least squares, and automatic weighted least squares;¹⁴⁸ and (3) normalization, which aims at making different SERS spectra comparable when they are collected with inconsistent parameters (e.g., concentration, laser power, acquisition time). It should be noted that this procedure is performed only when needed because a fraction of useful information may be hidden during normalization. The preprocessing steps must be conducted in a logical order to ensure that the next step does not mask the signal of interest highlighted with the previous preprocessing.¹⁵⁰ 

After obtaining a series of processed SERS spectra, the analyst should adopt appropriate methods to transform them into an image. The univariate imaging is a simple and direct way to demonstrate the location and relative concentration of analytes in the sample by displaying the intensity or integral area of a specific Raman peak at each pixel.¹⁴⁴ When dealing with complex biological samples containing multi-components, advanced multivariate methods are usually applied to demultiplex the mixed spectra, identify differences between spectra, classify spectra, and so on. Various multivariate techniques can be combined and optimized according to specific experimental purposes: (1) demultiplexing algorithms, e.g., least squares method which can separate the SERS spectra of mixed components. It assumes that the raw spectrum is a linear superposition of the pure spectra of each analyte, and the weight of each analyte is determined through an optimization process.¹⁰¹ Thus, the matrix of Raman spectra of each analyte can be acquired, and the corresponding image can then be plotted. (2) Feature extraction and selection, which can extract significant spectral features and reduce the dataset to a smaller set of variables responsible for later classification. For instance, principal component analysis decomposes the spectral data into a few orthogonal principal components responsible for the majority of the variance within the original dataset.¹⁴⁸ (3) Classification, aiming to classify SERS spectra into different groups after which the image can be plotted if various colors are assigned to the groups. For example, linear discriminant analysis (LDA) makes classification based on a Mahalanobis distance calculation between the samples of each group.¹⁵¹ Particularly, the recent emergence of artificial intelligence (AI) technologies provides new avenues for spectrum classification.^152,183 For the fingerprint SERS spectrum, machine learning data processing is highly useful in extracting the spectrum of known and unknown physical quantities if trained using enough data.¹⁵³ For example, we have reported the work to minimize the fluorescence background interference and some partially overlapped peaks by using multivariate curve resolution (MCR) methods, highly improving the limit of SERS detection with strong Raman background.^152,183 More studies on AI-assisted SERS imaging are thus anticipated to benefit the feature extraction and spectrum classification to further enhance Raman encoding capacity.¹⁰¹ 

SERS bioimaging differentiates itself for the unique light-tissue interaction and complex bio-environment revelation. The biofunctionalized SERS nanotags are designed to attach to cells or tissues through bio-recognition; then, by reading the Raman spectra, the location of nanotags could be monitored to trace the tissues of interest. Surface biofunctionalization brings better biocompatibility, particle stability, specificity, and targeting efficiency. High brightness and photostability of SERS nanotags enable sensitive and real-time imaging. Their characteristic fingerprint spectra with narrow bandwidth can be well distinguished from that of biological tissues, allowing high specific and multiplexed imaging without auto-fluorescence background interferences. Also, SERS nanotags can be excited with adjustable laser wavelengths, offering high flexibility in different application situations. So far, SERS imaging has been widely studied for in vitro cell imaging as well as in vivo tumor imaging, etc.

In vitro cell SERS imaging has gained much attention since they allow the observation of cellular processes and functions, playing an important role in the understanding of the interactions between cells and NPs [Fig. 9(a)]. To achieve that, biofunctionalization of as-synthesized SERS nanotags is conducted through the targeting ligand conjugations on the nanotag surface, allowing SERS nanotags to identify cancer cells via the bio-recognition between targeting ligands and the biomarkers on the cell membrane. Sometimes, the unspecific tagging of SERS nanotags to cancer cells was also applied. For instance, we have previously adopted the Au NP SERS nanotags functionalized with thiol polyethylene glycol acid to conduct imaging on Hela cells.^36,42 For the imaging of fixed cells, they were incubated with SERS nanotags and then fixed for SERS measurements.³⁶ The ultra-stable imaging was achieved under 30-min laser irradiation with a power density of 10⁵ W/cm², demonstrating a common and efficient strategy for sensitive SERS imaging at the cellular level.³⁶ As for the live-cell imaging, the cells were firstly cultured in a quartz dish for adhesion; after that, the culture medium was incubated with SERS nanotags of an appropriate concentration. The dish was immediately placed in a live-cell incubator, which was integrated with an inverted confocal Raman microscope with an airtight chamber for imaging.⁴² The tagging of SERS nanotags to cell surface enabled the cellular uptake of NPs, which could be dynamically tracked via time-lapse SERS imaging.⁴² 

Subcellular imaging is an important way to investigate the complex internal structures and processes of cells, in which the molecular composition and the specific intracellular chemical interactions at each location determine the cell functions. The typical way for subcellular imaging is to design SERS nanotags with organelle-targeting ability to bind to different parts of cells. With the fingerprint spectrum, SERS is advantageous in the high-throughput screening of biomarkers. The concentration of the biomarkers could be calculated through data processing of individual Raman spectrum in quantitative multiplexing cell imaging. For instance, Kang et al. developed Au SERS nanotags encoded with different NIR-sensitive Raman reporters.¹⁵⁶ The SERS nanotags were modified with three types of targeting ligands including cell-penetrating peptide, mitochondria-targeting peptide, and nucleus-targeting peptide so that they could selectively target specific intracellular organelles [Fig. 9(b)]. By separating their individual Raman spectrum, the location of each type of nanotags in living cells can be obtained, either around the cytoplasm or in the nucleus. They further achieved multiplexing live-cell imaging of a single human oral cancer cell (HSC-3) within 30 s (10 ms/pixel).¹⁵⁶ Furthermore, Chen et al. combined label-free and labeled (with nanotags) SERS imaging techniques to investigate cell apoptosis. Similar to Kang et al., they have first designed three types of SERS nanotags with either membrane or nucleus-targeting ability. Further, they designed two types of Au NPs without Raman dyes to bind to the membrane or nucleus of the same cell [Fig. 9(b), top panel]. After verifying the localization of labeled SERS nanotags in cells, the label-free Au NPs were adopted to simultaneously trace the biomolecular information from these locations. Through this strategy, multi-targeting SERS imaging of a HeLa cell was presented. They also studied the time-dependent changes of cell nuclei as well as membrane receptor proteins during 48-h apoptosis in live cells [Fig. 9(b), bottom panel].¹⁵⁴ This strategy allows for 2D/3D SERS imaging with a high spatial resolution (typically 30 × 30 pixels on a single cell), which is of great significance to image complex physiological processes in cells.

Another category of subcellular imaging is the visualization of the local chemical changes in the cell microenvironment. The sensing of intracellular pH has garnered much attention, since intracellular pH homeostasis plays a vital role in the cell functions, such as phagocytosis, apoptosis, and endocytosis.¹⁵⁷ The basic principle to sense pH is to use Raman reporters with pH-dependent spectra, such as 4-MBA, 4-ATP, 4-mercaptopyridine (4-MPy), and 2-aminobenzenethiol (ABT). 4-MBA has carboxyl groups that can be protonated at low pH. By monitoring intensity ratios of 1423 (COO- stretching mode) to 1076 cm⁻¹ (benzene ring vibration mode) of the MBA-coated SERS nanotags, the intracellular acidification during NP endocytosis could be visualized.³ Zhang et al. applied 4-MBA modified SERS nanotags for pH imaging to investigate the endocytosis process of SERS NPs from accumulation in late endosomes to the lysosome, and they realized the high-resolution 3D visualization of intracellular pH changes during NP endocytosis at different time points (2 and 8 h), opening new possibility toward the dynamic SERS imaging of NPs endocytosis [Fig. 9(c)].¹⁵⁵ For 4-MPy molecule, the pyridine has pH-dependent ionization behaviors. The intensity ratio of Raman mode at 1095 (X-sensitive mode) and 1003 cm⁻¹ (benzene ring breathing mode) can be used as a function of pH.¹⁵⁷ Xu et al. have designed 4-MPy modified nanotags as pH nanosensors, whose response were in the pH range from 4 to 9.¹⁵⁷ Functionalized with organelle-targeted peptide, the nanotags can be selectively transported to the nucleus, mitochondria, and lysosome. They further applied these nanosensors on a tumor cell line (human liver hepatocellular carcinoma) and a normal cell line (mouse embryonic liver cells), and they found that the averaged pH values of the cancer cells are lower than that of the normal ones. Their technique has revealed the subcellular pH values with high spatial resolution and pH sensitivity, providing an appropriate tool for fundamental studies of intracellular microenvironments.

The in vivo bioimaging and detection of tumor lesions using SERS nanotags [Fig. 10(a)] started from the pioneered study in 2008 by Qian et al., who used PEG-coated Au SERS nanotags to attach to xenograft tumor models by the epidermal growth factor Receptor (EGFR) biomarkers on human cancer cells, realizing the in vivo tumor detection in a mice model.⁷⁴ Following this pioneered work, plenty of progress has been made in recent decades by the rational design of SERS nanotags, such as Au nanospheres, nanostars, NRs, and GERTs,^3,93,158,159 broadening the SERS applications in the field of engineering and medicine. Andreou et al. presented the ability of SERS nanotags as a molecular imaging modality for precise visualization of liver tumor margins and detection of otherwise invisible microscopic lesions.¹⁶⁰ They synthesized silica-encapsulated SERS NPs and injected them intravenously into the tumor-bearing mice. SERS imaging of in vivo and excised liver showed an over 40-fold higher accumulation of NPs in healthy liver parenchyma than that in liver tumors, thus enabling precise discrimination of healthy from cancerous tissues; microscopic cancerous lesions that were detected neither with magnetic resonance imaging (MRI) nor with visual inspection or palpation could also be outlined via SERS imaging. In contrast, the identification of tumors with fluorescent dyes (e.g., indocyanine green, ICG) demonstrated cases of false positive and false-negative results. These findings suggest that SERS imaging holds promise for improved resection of tumors. Similarly, we adopted GERTs for the prostate tumor-targeted Raman imaging on the mice model [Fig. 10(b)].⁶¹ After 20 h of the nanotag injection via intravenous administration, the biodistribution of nanotags in orthotopic prostate tumor models can be well revealed by Raman imaging with 785 nm excitation. Particularly, after achieving maximal surgical resection of observable tumors, Raman imaging shows the ability as surgical guidance to eliminate residual tumor lesions or small intrusions of the tumor into the surrounding tissues. This work suggested the efficiency of SERS nanotags for a precise medicine.

The combination of SERS with deep Raman technique is another promising topic in recent years.^161,162 It is believed to circumvent the restriction of common penetration depth and enable tumor imaging at depth with high sensitivity and specificity. One of the deep Raman techniques is spatially offset Raman spectroscopy (SORS), which acquires signals from the subsurface layer by collecting light from areas that are spatially distanced from the illumination spot.¹⁶³ Nicolson et al. for the first time reported non-invasive in vivo imaging of glioblastoma multiforme (GBM) tumors in mice with the use of spatially offset SERS [Fig. 10(c)].¹⁶⁴ They applied IR-792 dye modified Au nanostars as resonant SERS nanotags, which were further functionalized with RGD and injected via tail vein of GBM bearing mice. With a relatively low laser density (13.8 mW/mm²), the deep-seated GBM was detected and profiled through the intact skull by Raman spectra. The deep SERS imaging is not only applicable for a brain tumor but also suitable for a wide range of diseases that SERS nanotags can actively target. The clinical translation toward the application of SORS-based Raman imaging is thus anticipated.

As the first station for tumor cells to spread along the lymphatic system, sentinel lymph nodes (SLNs) are critical to determining lymphatic metastasis of tumor sites with the presence or absence of cancer cells there. Therefore, SLN mapping appears as a key process in SLN biopsy for routine cancer staging and surgery. The detection and imaging of SLNs has been reported using SERS nanotags.¹⁶⁵ Iacono et al. demonstrated the potential of SERS in SLN detection by using a SERS-active Au NP with NIR dye modifications (Au@IR-pHPMA).¹⁶⁶ They injected the SERS nanotags dispersed in MES buffer in the paw of the right forelimb of the non-tumor-bearing nude mice. After 2 h, SERS imaging was obtained both on the injection site and on the SLN. At 24 h, they observed that SERS nanotags had drained almost completely from the lymphatic ducts into the lymph node with only slight signal remained in the forelimb. Also, we have reported GERTs as SLN imaging tracer [Fig. 11(a)] and achieved high-contrast wide-field in vivo SLN imaging (3.2 × 2.8 cm²) within 52 s using a laser power of 370 μW.⁴⁰ Besides that, the dynamic migration of nanotags into the SLN was monitored by measuring the real-time SERS spectra.⁴⁰ GERTs turned out to be an excellent SLN imaging tracer, exhibiting ultra-bright, highly photostable and of long retention time (24 h) in the lymph node.^36,40,141 Particularly, these ultrabright SERS nanotags allow the utilization of portable Raman spectrometer to detect the SLNs,¹⁴¹ opening new opportunities for rapid clinical lymph node identification.

In recent years, the integration of SERS with other imaging modalities has garnered much attention. The invention of multimodal SERS nanotags combined with fluorescence, photoacoustic (PA) imaging, positron emission tomography (PET), CT, and MRI is beneficial to combine the strengths of additional imaging modalities. For example, MRI plays an important role in early detection and preoperative staging evaluation of tumors. The SERS-MRI nanotags thus exhibit the ability of wide-field visualization and deeper tissue penetration for preoperative MRI, as well as the high sensitivity and good spatial resolution for intraoperative Raman imaging.

We have recently reported SERS nanotags for CT/MRI/SERS triple-modality biomedical imaging [Fig. 11(b)].¹⁶⁷ These nanotags were synthesized using gadolinium (Gd)-loaded Au NPs with a petal-like rough surface. Au NPs have been explored as CT imaging agents because of their enhanced x-ray attenuation property; small molecular Gd contrast agents show high T1 relativity for MRI. Therefore, these nanotags demonstrated strongly enhanced CT and MR imaging capability and superior Raman intensity. The results showed that CT and MRI achieved the optimal enhancement effect on mice xenograft tumor at 2 h post-injection of nanotags, and the in vivo tumor Raman imaging can still be detected at 6 h after injection. This well satisfied the practical clinical requirements of preoperative MRI and CT scanning and intraoperative Raman detection.

Fluorophore-embedded SERS nanotags have been reported to emit both fluorescence and Raman lights at the same time when excited by one laser wavelength, through the strategy of controlling the distance between the fluorescent molecules and the metal surface to avoid the quenching of fluorescence.^31,108,168 Pal et al. invented fluorescence-SERS bimodal nanotags based on a type of core–shell Au NRs.¹⁰⁸ The fluorophores of high Raman cross sections were embedded in NRs and served as Raman reporters. Their dual-mode imaging using these NRs has achieved the visualization of tumor sites on mice models of subcutaneous ovarian cancer or glioblastoma. The work unveils the possibility of dual-mode SERS-fluorescence cancer imaging with the advantages of both methodologies, i.e., high-speed and accurate localization.

The spatial resolution of optical microscopy was thought to be fundamentally limited, as described by Abbe’s theory, for more than a century. Recent breakthroughs in fluorescence imaging techniques have pushed the resolution of optical microscopy down to the molecular level.¹⁶⁹ Specifically, stochastic optical reconstruction microscopy (STORM) takes advantage of photo-switchable fluorescence dyes that can be modulated between the “bright” and “dark” states, to localize fluorescence centroids. Under a similar principle, super-resolution SERS imaging has also been reported, which relies on the natural SERS “blinking” effect to provide a temporally fluctuating signal for localization algorithms. So far, super-resolution SERS was reported to provide a resolution of sub-5 nm to image the single-molecule hot spots and to reveal the difference between SERS and luminescence centroids.^170,171 Compared to the state-of-art fluorescent super-resolution imaging that requires sophisticated sample preparation and dye selection, SERS holds superiority with its easy operation and label-free detection capability by using plasmonic substrates to directly enhance the target signal.

Olson et al. demonstrated SERS-STORM combined methodology for super-resolution chemical imaging.¹⁷² As shown in Figs. 12(a-i) and 12(a-ii), bacteria cells were adsorbed on a rough Ag substrate surface to interact with the plasmonic hot spots. They used a “snapshot” strategy by imaging through a diffraction grating, so STORM algorithms can be modified to extract a full SERS spectrum, thereby capturing spectral as well as spatial content simultaneously.¹⁷³ Images on E. coli bacteria cells [Figs. 12(a-iii)] taken with this strategy match well with scanning electron microscope images. This technique reaches a high spatial resolution of <50 nm and allows direct imaging of the cell/substrate interface of thick samples in turbid or opaque liquids without additional nanofabrication. This work holds promises in identifying unique chemical signatures of various cells and could be applied to other biological structures of interest.

de Albuquerque et al. applied super-resolution SERS imaging to track membrane receptors interacting with peptide-functionalized Au NSs.¹⁷⁴ As shown in Fig. 12(b), the cells were first incubated with Au NS probes on Ag island coated coverslips, then fixed for imaging. The peptide interaction with the αvβ3 integrin receptor led to successful NPs targeting to cells, introducing bright and fluctuating SERS signals that could be analyzed with localization microscopy algorithms. A series of wide-field SERS images were collected using 100 ms per frame. SERS-STORM methodology was further applied to process these wide-field SERS images to localize AuNSs interacting with the cell, showing an improvement in the resolution compared to the original wide-field images [Fig. 12(b), top panel]. By comparing the SERS images using targeted or nontargeted NPs, they illustrated specific binding events with a localization precision of ∼6 nm. This work demonstrated super-resolution SERS imaging to probe membrane receptor interactions in cells, providing chemical information and spatial resolution with potential for diverse applications in life science.

In this tutorial, we have introduced the basic principles of SERS and SERS nanotags, as well as the recent progress in SERS nanotag based bioimaging applications and high-resolution SERS imaging techniques, providing a whole picture for the potential of SERS imaging technology in the biomedical scenario. Centering around the keys points in designing the appropriate clinical SERS nanotags, including nanostructures, Raman reporters, surface coating, and biofunctionalization, we elaborated three subtopics, i.e., GERTs, core-satellite NPs, and SR Raman reporters, to give a comprehensive overview of the design of novel SERS nanotags.

Despite the current dramatic progress of biomedical SERS, there still leaves blank for further fulfillment. First, the enhancement factor is still far from satisfaction, especially for in vivo direct imaging, a scenario that is more complicated and with analytes unable to be stably confined on the plasmonic surface. Second, for the in vivo imaging, the effectiveness of transferring NPs to the intended tumors is unfortunately low. Although research studies have succeeded in relatively better models with animal experiments, there exhibits a much lower efficiency in the real clinical tests in the human body,^175,176 together with the recently-claimed failure of long-held enhanced permeability and retention (EPR) effect, which further challenges the clinical translation of SERS-based NPs as nearly all the other nanomaterials.¹⁷⁷ 

The biggest concern remains the relatively low imaging speed that hinders the wide application for in vitro and in vivo situations. For example, it typically takes tens of minutes to acquire a wide-area cell image using the conventional point-scan mode (e.g., assuming 1 s/pixel for an area of 30 × 30 pixels). The current strategies for high-speed SERS bioimaging mainly include the use of brighter SERS nanotags and advanced imaging approaches, to shorten the acquisition time. A high-speed confocal Raman microscopy system can be realized by the galvano mirrors instead of mechanical stage movements to manipulate the laser spot on samples.¹⁷⁸ With the use of intra-gapped GERTs, this custom-built Raman system was successfully applied to accomplish the high-resolution (50 × 50 pixels) and high-speed (30 s, 10 ms/pixel) single live-cell SERS imaging.¹⁵⁶ Similarly, we have adopted GERTs and Horiba DuoScan mode to achieve single-cell imaging within 6 s and multiplexing cell imaging within 43 s, using a 370 μW laser power and 0.7 ms acquisition time per pixel [Fig. 9(b)].⁴⁰ Besides the above efforts, machine learning technologies provide new avenues for SERS spectral analysis to improve the imaging process. He et al. reported the improved data quality by a trained deep convolution network, which statistically learns to transform low-resolution images acquired at a high speed into high-resolution ones.¹⁴⁵ Their imaging speed can be five times faster than the conventional line-scan Raman imaging without sacrificing spectral and spatial resolution. These works provide insights about future directions to improve imaging speed toward real applications. Nevertheless, the current speed still lags far behind the widely used fluorescence imaging. Further efforts are expected toward the developments of ultra-bright SERS nanotags and advanced spectrometer with the wide-field acquisition.

The future study of SERS in vivo applications can be expected in two categories. The first one is the design of multifunctional SERS nanotags with combined therapeutic effects, including photodynamic therapy, photothermal therapy, and chemotherapeutic delivery. We have reported the utilization of cisplatin-loaded SERS nanotags designed for chemo-photothermal synergistic therapy.¹⁴² With the guidance of Raman imaging, the widely disseminated small tumors of ovarian cancers, which are hardly distinguishable by naked eyes, could be successfully identified and then treated by photothermal and chemotherapy in one treatment. Similar work was conducted using Au SERS nanotags for intraoperative real-time detection and eradication of residual microtumors on the mice prostate tumor model.⁶¹ The nanotag-treated group achieved complete tumor-free survival for 100% of the animals after 39 days. These works demonstrated the promising diagnostic and therapeutic potential of multifunctional SERS nanotags.

The second category would be the biosafety and biotoxicity study. Although a wide variety of SERS nanotags have been developed, the most threatening impediment to its clinical translation still lies in the biosafety of inert NPs. We studied the in vivo biodistribution and toxicity of silica-coated Au SERS nanotags, by monitoring long-term histological changes in several major organs (heart, liver, spleen, lung, and kidney) at 1, 2, 3, 4 weeks, and 2 months after intravenous injection of a dosage (0.2 ml in saline solution, [Au] = 0.05 M).¹⁶⁷ No obvious signs of abnormality such as morphological or inflammation changes were observed compared to the control group for up to 60 days, proving the good long-term biocompatibility. Generally, Au is considered biocompatible, but the modification of molecules or ligands on particle surfaces may induce additional toxicity. For example, CTAB-stabilized Au NPs may induce high levels of toxicity to cultured cells. The sophisticated knowledge of the pharmacokinetics and toxicity related to metallic, inorganic, and polymeric nanomaterials is still under investigation.

To sum up, SERS nanotags have brought new opportunities for spectroscopic biomedical imaging and detections. We have seen the glow of biomedical SERS in every corner that holds promises toward SERS-based intraoperative imaging and navigation. Though most of the current studies are still confined to animal models, we anticipate a bright future for the regulatory approval of inert NPs and the real clinical translation.

This work was supported by the National Natural Science Foundation of China (Nos. 81871401 and 81901786), the China Postdoctoral Science Foundation (Nos. 2018M640395 and 2019T120343), the Science and Technology Commission of Shanghai Municipality (No. 19441905300), the Shanghai Jiao Tong University (No. YG2019QNA28), and the Shanghai Key Laboratory of Gynecologic Oncology.

Data sharing is not applicable to this article as no new data were created or analyzed in this study.