Prospects of Surface-Enhanced Raman Spectroscopy for Biomarker Monitoring toward Precision Medicine.

Future precision medicine will be undoubtedly sustained by the detection of validated biomarkers that enable a precise classification of patients based on their predicted disease risk, prognosis, and response to a specific treatment. Up to now, genomics, transcriptomics, and immunohistochemistry have been the main clinically amenable tools at hand for identifying key diagnostic, prognostic, and predictive biomarkers. However, other molecular strategies, including metabolomics, are still in their infancy and require the development of new biomarker detection technologies, toward routine implementation into clinical diagnosis. In this context, surface-enhanced Raman scattering (SERS) spectroscopy has been recognized as a promising technology for clinical monitoring thanks to its high sensitivity and label-free operation, which should help accelerate the discovery of biomarkers and their corresponding screening in a simpler, faster, and less-expensive manner. Many studies have demonstrated the excellent performance of SERS in biomedical applications. However, such studies have also revealed several variables that should be considered for accurate SERS monitoring, in particular, when the signal is collected from biological sources (tissues, cells or biofluids). This Perspective is aimed at piecing together the puzzle of SERS in biomarker monitoring, with a view on future challenges and implications. We address the most relevant requirements of plasmonic substrates for biomedical applications, as well as the implementation of tools from artificial intelligence or biotechnology to guide the development of highly versatile sensors.

The discovery of surface-enhanced Raman scattering (SERS) and its subsequent development and application is relatively recent. The initial evidence of the SERS effect was accidentally observed less than 50 years ago, when Fleischmann and co-workers registered the Raman signal of pyridine on rough silver electrodes and obtained intensity values that exceeded by far those previously considered as standard.[1−3] Such unexpectedly high intensities were subsequently investigated and rapidly identified as a potential solution to the main hurdle in Raman spectroscopy: poor sensitivity hindering molecular detection at low concentrations.[4,5] Although Raman spectroscopy provides characteristic information on the chemical nature of the illuminated molecules, the low probability of the involved spectroscopic events largely hindered its potential applications; it has been estimated that only one in every 108 photons undergoes an inelastic scattering upon light–matter interaction.[6−8] Therefore, the Raman effect is considered to be a very inefficient process, and the acquired intensities are typically lower than those in other light–matter interaction events, for example, when the energy of the scattered photon is conserved (Rayleigh scattering) or real excited states are involved, as in fluorescence emission.[9,10]

Understanding the basis of SERS to achieve suitable Raman signal enhancement has thus become one of the major subjects of study in the field.[11] Although significant progress has been made toward mechanistic understanding, it remains under intense investigation.[12−14] The main mechanism underlying SERS has been primarily defined as the result of an electromagnetic effect, occurring between the analyte and the excitation of plasmons at a metal surface, as well as a chemical enhancement based on the transfer of electrons between the substrate and the target molecules.[15,16] This effect has also been observed for metal particles with sizes significantly smaller than the wavelength of incoming light, resulting in the oscillation of conduction electrons in resonance with the electromagnetic radiation.[17] The generation of so-called localized surface plasmon resonances (LSPR) affects the local electric field, amplifying its intensity around the nanoparticles (NPs). It is widely accepted that such an enhancement of the local electric field alters the polarizability of adsorbed molecules, which in turns leads to higher frequencies of inelastic scattering events and increased Raman signals.[18,19]

The ability to greatly enhance the Raman signal of molecules adsorbed onto metal surfaces, while preserving the rich vibrational information, has raised further interest in SERS technology toward its integration into functional sensors.[20,21] Recent advances in SERS have thus been closely connected to the development of purposely devised plasmonic structures toward optimal amplification of the Raman signal, up to 11 orders of magnitude.[22,23] The fabrication of tailored substrates, typically comprising nanostructured metals such as gold or silver, has allowed the detection of extremely low molecular concentrations within diverse environments.[24,25] SERS thus comprises an extensive area of research ranging from theoretical studies of the mechanisms underlying signal enhancement, to the development of plasmonic substrates and devices for real-world applications. The wealth of progress in SERS technology has laid the foundation toward its future expansion into different disciplines (e.g., biomedicine, food safety, or environmental monitoring).[26−29] Notwithstanding, additional developments are still required to achieve a successful and reliable implementation, in which all of the variables that influence the SERS performance are properly considered.

In this Perspective, we set the focus on applications addressing biomedical problems, which have emerged as a promising direction and therefore have fueled research on the interplay between biomolecules and plasmonic components.[30,31] Such an interest is explained on the need for further development in sensors technology, intended to respond to the growing demand for fast monitoring of biomarkers needed to underpin personalized medicine.[32,33] Hence, cutting-edge SERS studies seek to demonstrate the suitability of this technique for routine screening of human health, including authorized clinical trials. We provide illustrative examples of the use of SERS in biomedical applications, with a particular emphasis on identifying those directions that may drive widespread integration into biomedical sensors, such as the optimization of plasmonic substrates for ongoing applications. Special attention will be paid to those characteristics and requirements needed for collecting accurate and reliable SERS spectra under complex biological conditions. We finally introduce novel strategies resulting from the combination of SERS with state-of-the-art tools from artificial intelligence or biotechnology, thereby creating platforms with a wider versatility and the ability to address future exciting opportunities and challenges lying ahead.

SERS and Biomedicine, Fated to Meet Each Other

The detection of biomarkers has recently become a pivotal stage in the characterization and diagnosis of human disorders, so that a personalized treatment can be selected according to measured features. Prognostic biomarkers may include proteins, nucleic acids, extracellular vesicles, and even whole prokaryotic or eukaryotic cells. In particular, the detection of metabolites, produced as the result of biochemical processes in living organisms, has been found to provide meaningful information on the pathophysiology of diseases. To this end, multiple strategies have been devised, not only to detect such biomarkers in different biofluids, but also to image their distribution within tissues, including in vivo, ex vivo, and in vitro models.[34,35]

Conventionally, laboratory assays for biomarker monitoring have been carried out by means of techniques that entail tissue-destructive procedures, such as immunohistochemistry or liquid chromatography coupled to mass spectrometry (LC-MS).[36] In addition, especially for the detection of small molecules (including metabolites), colorimetric techniques involving the administration of either chemical functional groups or enzymes that can react with the molecule of interest, provide high sensitivity and specificity while not requiring operation by specifically trained personnel.[37] However, even though a wide variety of components have been identified in biological fluids, both techniques cannot be deployed to assess different samples without irreversible damage; their ability to capture dynamics and heterogeneity profiles at the point of interest (for example, across tissue biopsies) is therefore limited. Other alternatives such as nuclear magnetic resonance (NMR) spectroscopy, which do not require elaborate sample preparation, are prone to low sensitivity and complex interpretation, in addition to bulky instrumentation and low portability.[38] In this context, optical methods, which employ light to noninvasively probe matter, offer a valuable alternative to unravel the presence of specific biomarkers with high spatiotemporal resolution, as well as representing a cheaper and faster approach.[39,40] Among all available optical techniques, fluorescence microscopy arguably represents the most popular, commercially available technology to advance toward noninvasive monitoring of biological samples.[41,42] However, as autofluorescence from biological molecules is typically weak and poorly specific, the development of fluorescent probes has been accelerated as a means to monitor different biological events.[43] This strong reliance on fluorescent dyes, which may additionally suffer from photobleaching and be degraded inside cells with cytotoxic effects, hinders the application of this technique to extended monitoring times and imposes a labeling step. Label-free optical techniques with high specificity and sensitivity are thus in high demand, and therefore, nontraditional modalities are being explored for biological analysis.[44]

Beyond fluorescence, other label-free optical techniques that have attracted significant attention in the context of biomarker monitoring include mid-infrared (mid-IR) and Raman scattering spectroscopies. Whereas mid-IR imaging/spectroscopy can be used to record vibrational information from histology samples,[45,46] the strong absorption of mid-IR radiation by water compromises its applications in routine biomarker monitoring. On the contrary, the inelastic Raman scattering cross section is very low for water. Hence, this technique can in principle be applied to record the vibrational fingerprints of metabolites in environments with a high water content (e.g., biofluids or intracellular compartments).[47,48] Unfortunately, the Raman cross sections for most metabolites are also very low, and therefore, the usual limits of detection in Raman spectroscopy are above 1 mM, posing a major drawback against monitoring of biomolecules at lower concentrations. Consequently, approaches toward increasing the intensity in Raman measurements have been implemented (for example, resonance Raman and tip-enhanced Raman).[49] Arguably, surface-enhanced Raman scattering has been the most successful of these techniques, with plenty of examples of successful application in the biomedical field. As a result, we can evoke a future where SERS may drive a new narrative in the diagnosis of human disorders.

Label-free SERS applications, also known as direct SERS, have been primarily used for the detection of small molecules, ranging from relevant metabolites in bacteria or eukaryotic cells to diverse neurotransmitters and drugs.[50−55] Although macromolecules can also be detected by SERS, larger biomolecules (usually around 10–2 μm) are formed by combinations of the same building blocks, thus registering very similar SERS spectra among members of the same class (e.g., in proteins or DNA).[56−59] As a consequence, specific labels must be incorporated for accurate SERS identification. The resulting indirect approaches, in which the detected SERS signal comes from a probe or label, typically involve traditional immunoassays or oligo-microarrays and allow the detection of both proteins and nucleic acids.[60,61] Regarding indirect sensors, SERS probes must comprise, in their simplest version, a noble metal (typically Au or Ag) nanoparticle and a monolayer of reporter molecules acting as fingerprint labels.[62] The background produced by biological fluids can be systematically reduced by carefully designing the nanoprobe (e.g., by including a protecting outer shell) and including washing steps. Thus, only few modifications are required on conventional assays, for instance, a SERS probe instead of a fluorescent dye, and therefore, indirect sensors are closer to being integrated into commercial diagnostic tools.

In contrast to macromolecules, the characteristic fingerprint of small metabolites can be recorded and distinctly recognized by SERS, as long as these species can get in contact with the plasmonic substrate.[63] Each chemical bond has a characteristic vibrational frequency, which determines a Raman scattering wavenumber, so that SERS barcodes can be defined for specific metabolites, thereby facilitating their identification in complex mixtures.[23,64] Thus, small molecules of interest are more likely to be accurately traced in the absence of a tag, reporter, or indicator molecule. The number of molecules ultimately adsorbed onto the plasmonic structure is otherwise controlled by their surface affinity, which in combination with the identity of the sample and its Raman scattering cross-section, determine the SERS spectral fingerprint.[65] However, it should also be taken into account that the high complexity of biological samples, containing a wide and dynamic range of biomolecules that interfere with each other as well as interacting with the plasmonic substrate, may compromise the prediction of SERS spectra resulting from the contribution of individual components (Figure ). In other words, quantification of analytes, especially in complex biological mixtures, poses a great challenge, as the composition of the analytes adsorbed on the metallic structure might differ significantly from those observed in solution.[66,67]

Figure 1

SERS spectra of biological samples are defined by a combination of four main parameters, as indicated in the scheme. The nature of the interrogated biomolecules and the plasmonic substrate control the interplay between both components, via their binding affinities. Biomolecules present within the irradiated area of the plasmonic substrate, as well as their intrinsic Raman cross sections, will eventually determine the recorded signal. Biomolecules that adsorb weakly onto the metallic surface, or which feature low Raman cross sections, are unlikely to be detected (pink dots in the scheme). Optimization of plasmonic substrates and measurement settings are commonly required for an efficient response of SERS platforms in various label-free scenarios.

Additionally, unlike indirect measurements, the sensitivity of label-free experiments cannot be modulated by either varying the type or the number of reporter molecules (e.g., increasing dye concentration or other attempts with different reporter molecules). On this account, the Raman intensity in label-free approaches can only be boosted by optimizing the plasmonic nanostructures for different scenarios, apart from modifying instrument settings, common for all types of Raman measurements.[68] Overall, optimization of plasmonic substrates to the specific running application is an early stage strategy that can significantly improve SERS performance for monitoring of biomolecules.

SERS Substrates, the Main Pillar for Technological Advancement

Based on their ability to concentrate the electric field within a nanoscale volume, a wide variety of metal nanoparticles and nanostructured films have been explored as plasmonic substrates. Such plasmonic nanostructures may differ in composition, morphology, size, and spatial arrangement, pursuing higher enhancement factors and a fine-tuning of their LSPRs.[69,70] Although many efforts are still being devoted to this line of research through the fabrication of increasingly more sophisticated sensors, additional features are required for biomedical applications, especially when in situ measurements are targeted. We propose that the most relevant requirements for biomedical applications are the following: (i) efficiency must be maintained within biological media, along extended exposure times; (ii) the sensor must be fully integrated at the point of interest, while avoiding the presence of external cytotoxic agents that could harm living tissues; and (iii) the possibility of controlling binding affinities between biomolecules and the surface of the plasmonic substrate. While bearing these concepts in mind, the development of plasmonic substrates should pursue a sensitivity matching the specific problem at play, typically the physiological concentration range of the biomolecules of interest.

Substrate Fabrication and SERS Enhancement Optimization

Fabrication methods of efficient plasmonic substrates are typically classified within two categories (Figure ). On one hand, nanostructures that are directly built on the surface of solid materials by top-down approaches, such as electron beam lithography.[25,71,72] On the other hand, bottom-up approaches comprising nanoparticles that can be applied either in a colloidal dispersion or as various types of supported substrates.[73−75] Representative examples from these categories usually exhibit significant differences regarding homogeneity of the geometrical structure, instrumentation, and know-how required for fabrication/synthesis, as well as the potential scale-up.[76−81]

Figure 2

Schematic representation of top-down and bottom-up fabrication of plasmonic substrates: top-down strategies yield metal nanostructures with high resolution; bottom-up strategies provide the simplicity and scale-up possibilities of colloidal nanoparticles. The scalability of bottom-up substrates, obtained by self-assembly of individual nanoparticles, makes them appealing as sensors in biological applications. As the main responsible component for SERS signal enhancement, optimization of plasmonic substrates is constantly being reported. General improvements are mainly oriented to increasing the enhancement factor, that is, generating more intense electric fields around nanoparticles or to better control hotspot formation and plasmonic tunability. Substrates aimed for biological applications have additional requirements (stability, biocompatibility, etc.) due to their constant interaction with biological media. Current efforts attempt to maximize substrate stability and full sensor integration, while minimizing the undesired perturbations induced during the acquisition of SERS spectra. Controlling the binding affinity between nanoparticles and biomolecules may also facilitate detection at low concentrations, even when other biomolecules are present at orders-of-magnitude higher concentrations.

Top-down approaches can produce high-resolution nanostructures on demand, but upscaling these processes is arduous and hinders their final integration into biosensors. Still, compelling examples of application of such top-down structures have endowed valuable biological information in some scenarios.[82−84] For example, densely packed nanometer-sized pillars have been used as SERS substrates with meaningful biological applications.[85] Upon exposure to a liquid sample and subsequent evaporation, metal-coated nanopillars can form clusters due to collective leaning of the pillars, thereby creating hotspots with high electric field enhancement. A reasonable uniformity in the arrangement of plasmonic structures renders such substrates attractive toward the detection of various biomolecules, such as beta amyloids in Alzheimer’s disease or small oligonucleotides.[86,87]

Notwithstanding, most applications have been developed by using substrates comprising colloidal nanoparticles, that is, prepared through the bottom-up approach. In addition to their lower production cost, these substrates can be readily manufactured for worldwide commercialization. The assembly of metal nanoparticles to build plasmonic substrates typically gives rise to plasmon coupling/hybridization effects, which are responsible for the greatest SERS enhancement factors.[88,89] Extremely high electric fields are confined within tiny interparticle distances in such structures; even if such hotspots represent a small fraction of the irradiated surface, they contribute most significantly to the recorded SERS signal.[90−93] Methods as simple as drop casting or precipitation of colloidal dispersions enable the production of plasmonic substrates with hotspots that can efficiently amplify the Raman signals from analyte molecules.[94,95] However, the nature and efficiency of hotspots are strongly dependent on the specific arrangement of the individual NPs, interparticle spacing and orientation. In turn, small changes or perturbations in near-field enhancement (E/E0) would drastically alter SERS intensity, which scales as its fourth power (E4/E04).[96−98] Unfortunately, the source of SERS sensitivity may also be the predominant cause of poor reproducibility. Sensors based on random nanoparticle aggregation typically exhibit a poor performance in terms of repeatability and reproducibility, meaning that better standardization and benchmarking protocols are needed to assess the quality of SERS substrates, in particular, when considering large-scale production.[13] Hence, the development of procedures aiming to define ordered nanoparticle architectures has acquired a strong relevance. In this context, self-assembly processes can be used to drive dispersed colloidal systems into organized structures or patterns without additional guidance.[99] By tailoring the functionality and affinity between nanoparticles and solvent, the self-assembly of individual components in solution can be accurately controlled.[100,101] On the other hand, nanostructured templates can also be used to guide the self-assembly of nanoparticles into predefined structures, ranging from thin films (2D) to colloidal crystals or supercrystals (3D).[102−104] Custom-made molds of different materials, such as polydimethylsiloxane (PDMS) or poly(methyl methacrylate) (PMMA), have been used to trap nanoparticles inside the wells of the template, thereby forming organized nanostructures that replicate the mold.[105] As a result, not only a higher reproducibility in the SERS signal can be obtained, but also an exquisite definition of the optical responses toward achieving maximum enhancement.[106] Strategies combining top-down methods to fabricate nanostructured templates with colloidal NPs can maintain the advantages of both techniques to an extent that they could represent an interesting option, even for large-scale commercialization. Still, the challenge resides in fabricating assemblies that display a sufficient homogeneity over macroscopic areas while reducing intersample variability. Eliminating routine activation requirements for these sensors (such as oxygen plasma or UV–ozone cleaning steps, which may damage plasmonic surfaces) would also contribute to obtaining more reproducible SERS substrates.

In the same direction, purposely matching the plasmon resonances of the substrate to the incoming photons (laser) wavelength is of special interest for biomedical applications. SERS spectra should be collected upon irradiation with an excitation wavelength that is harmless to cells and, ideally, that can propagate through tissue, typically within the first biological transparency window, between 680 and 920 nm.[107,108] Modulation of plasmon resonances has been primarily realized by varying either the chemical nature or the morphology of the nanostructured plasmonic component (the NP).[109−111] Other strategies can be used to dynamically dictate plasmon resonances based on the application of external electrical,[112] magnetic,[113] thermal,[114] or light stimuli.[115] In a recent example, template-assisted self-assembly was used to obtain hierarchical nanostructured substrates, comprising square arrays of hexagonally packed AuNP clusters. These highly ordered substrates generate intense lattice plasmon resonances, which can be engineered by tuning geometrical parameters, in particular, the lattice parameter or separation distance between AuNP clusters.[116] This effect was deployed to extend the spectral window, from the visible to the near-IR, without variation of the nanoparticle building blocks (Figure a).[67,117] In a variation of the same method, the plasmonic substrates were transferred onto elastomeric PDMS-based supports, thereby allowing real-time modification of the lattice plasmon resonances by extension or contraction of the substrate upon application of macroscopic strain (Figure b).[118]

Figure 3

(a) Engineering plasmonic nanoparticle arrays to tune lattice plasmon modes: (i) SEM images of AuNP arrays with a variation of the distance between NP clusters. Scale bars are all 1 μm; (ii) Extinction spectra of the same AuNP cluster arrays displayed in (i). Reproduced with permission from ref (116). Copyright 2019 ACS. (b) The optical response of AuNP arrays can be dynamically tuned by applying extension or contraction forces onto flexible plasmonic substrates. Reproduced with permission from ref (118). Copyright Wiley-VCH 2021. (c) TEM images of silica-coated AuNPs and self-assembled core–satellite superstructures. Silica shells enhance the stability and allow SERS monitoring of molecules in the proximity. Reproduced with permission from ref (126). Copyright Wiley-VCH 2011. (d) Hydrogel polymerization in a colloidal dispersion results in polymeric nanocomposites that can be adapted to different shapes while enhancing the stability of the embedded nanoparticles. Reproduced with permission from ref (139). Copyright Nature 2016.

Optimization of Substrates for Biological Applications

Stability and Reproducibility in Biological Media

When plasmonic nanostructures are immersed in biological media, their physicochemical properties may dramatically change over time.[119] A combination of multiple factors affects the stability of the system through degradation of surface functionalities, adsorption of proteins and other (bio)molecules, and even morphological changes.[120,121] These unwanted effects lead to changes in the plasmonic response and are prone to providing misrepresentative and irreproducible results. This phenomenon particularly hinders the use of nanoparticles in suspension as SERS sensors in biomedicine. Uncontrolled aggregation of nanoparticles occurs frequently within biological environments, resulting in the formation of clusters with largely variable SERS signal enhancements.[122] Several studies have demonstrated that biological media can also induce the release of ions from metallic particles, which in turn would modify their biocompatibility. Hence, it should be stressed that significant differences can be found in SERS platforms, between their ideal behavior in pure water and their practical performance in biological media, even if compared with commonly used isotonic buffers (e.g., phosphate-buffered saline (PBS) solution).[123,124] The stability and performance of substrates should be carefully evaluated in the selected biological environment under relevant conditions as an intermediate step prior to the real-world, uncontrolled sample. Different approaches have been explored to enhance the biological stability of Au NPs, for example, through coatings with silica or polymer shells (Figure c).[125,126] However, enhanced stability may come at the expense of compromising the interaction of nanoparticles with the target molecules, thus, hindering the SERS monitoring performance.[127] On the other hand, the routine approach of immobilizing nanoparticles on solid supports (mainly glass, silicon, or quartz, but even paper) is still sustaining innovation, with novel strategies based on sophisticated functionalization, to strongly retain the nanoparticles.[128−137] Recent approaches have succeeded in tailoring polymer nanocomposites to function as SERS substrates displaying highly tunable features. Plasmonic nanocomposites are formed by embedding plasmonic nanoparticles in polymeric materials, which further assist in preserving the intrinsic properties of the NPs in complex environments (Figure d).[137−139] The polymer would act by enhancing the robustness of the whole sensor over time while preventing aggregation of the embedded nanoparticles.[140] Such polymer composites must feature a sufficient porosity to allow the diffusion of target biomolecules toward the embedded plasmonic nanoparticles.[141] The combination of robust nanoparticle stability and high accessibility renders polymer nanocomposites useful for in situ measurements, potentially fostering a new generation of SERS sensors.[142]

Integration in Biological Systems

Key features of direct SERS include its noninvasive character and the reduced/absence of sample preparation requirements, which enhance the potential of implementing measurements at the point of care (POC), for example, in the clinic or in cell cultures and artificial organs. However, conventional substrates are rigid and thus barely adaptable to be used in the context of arbitrarily shaped surfaces, which dramatically limits their full integration into real-world scenarios. In this context, emerging SERS substrates formed by soft, flexible, transparent materials, open new avenues toward exploiting a rapid screening of analytes within the complexity of real systems.[111] Specifically, flexible substrates allow intimate contact with surfaces that are barely accessible to rigid platforms to an extent that the sensor could provide real-time information on nearby perturbations in a noninvasive manner. Multiple strategies have been explored to adhere metal nanoparticles on different flexible supporting materials, including polymers,[143] paper,[144] graphene oxide,[145] and nanowires.[146] In a recent example, a paper-based substrate in which NPs were adsorbed onto cellulose fibers was used to collect tears directly from a human eye and monitor their composition by SERS.[147] Of particular interest are applications of flexible plasmonic materials as wearable devices, also known as smart tattoos, to monitor biomolecules directly in the body while maintaining their plasmonic activity under various deformations (Figure a). Such flexible SERS substrates can be attached onto the skin or other body surfaces (e.g., the eyeballs) to uncover the presence of trace molecules in sweat and other biofluids (Figure b).[148−150] By engineering plasmonic tattoos with microneedle structures, one could even register intradermal information by SERS.[151,152] It should be stressed that an accurate control over substrate thickness and high transparency is required for in situ measurements, because laser radiation must penetrate through the support layer before reaching the sample. Recent outcomes underpin such devices as powerful tools to bridge the daunting gap between personalized therapy and real-time tracking of meaningful molecules inside the body.

Figure 4

On body SERS: (a) Sketch depicting the operation of smart tattoos to monitor metabolites in sweat. The flexible character of the sensor enables implanting on human skin, so the sensor is in contact with secreted sweat and metabolites present therein. Adapted with permission from ref (148). Copyright 2021 AAAS. (b) Schematic illustration of a contact lens combined with a plasmonic nanostructure for integration on an eyeball via transfer printing for in situ detection. SERS spectra are compared before and after dropping a glucose solution over the contact lens sensor, thereby registering changes in glucose concentration. Adapted with permission from ref (150). Copyright 2016 Wiley-VCH. (c, d) On-cell sensors to monitor intracellular environments by plasmonic nanopipettes, with similar shapes as that in the SEM image (scale bar 1 μm), which can pierce cell membranes with minimum invasion (c), or by internalized nanoparticles, which typically accumulate into vesicles such as lysosomes. For the example in (d), SERS images were reconstructed from two SERS spectral windows at two different times; the red color is the average Raman intensity from endogenous molecules and the green color is the average SERS intensity from 1960 to 2010 cm–1, assigned to the traced lysosomal drug, which displays alkyne peaks. Adapted with permission from refs[157 and 161]. Copyright ACS 2020 and 2016, respectively. (e) 3D-printed plasmonic scaffolds can be used as both platforms for cell culture (left: cells labeled with GFP grow across the scaffold, observed in reflection) and the sensors to register extracellular perturbations with spatial and temporal resolution; the evolution of the SERS signal of a photoactivated drug (methylene blue) is registered in the scaffold over time. Adapted with permission from ref (162). Copyright ACS 2021.

Other plasmonic substrates have also been devised to evaluate cell activity, adapting their design to the challenges imposed by the interaction with cells and other biological components. Initial attempts involved the internalization of nanoparticles by cells, followed by the acquisition of SERS spectra.[153−155] Unfortunately, this strategy often results in considerably altered cell behavior, as well as low control on nanoparticle stability and location within the ever-changing intracellular milieu. Recent studies have explored the specific interactions between certain nanoparticles and cells to achieve more uniformly enhanced SERS signals at defined locations, for example, at cell membranes[156] or in lysosomes[157] (Figure d). However, NP internalization itself is a complex process due to a strong dependence on variable aspects, such as the state of the cells or the specific cell line.[158] Therefore, as intracellular particles are the source of signal enhancement, significant differences are typically encountered in SERS spectra from different cells. To overcome this source of inaccuracy, the development of plasmonic nanotips or nanopipettes proved essential to interrogate intracellular compartments with minimum cell damage, while controlling the nature of the plasmonic components. Optimization of the geometry in plasmonic tips, so that nanoparticles are adhered on the surface, facilitates cell membrane penetration toward intracellular compartments in living cells (Figure c).[159,160] The position of the plasmonic component is thus externally directed by an operator, thereby reducing signal variability. This setup configuration was also used to monitor metabolites in extracellular media, capturing the formation of chemical gradients near cells.[161]

The integration of efficient Raman signal enhancers inside cellular environments is still in its infancy, which impairs the use of SERS to capture in real-time those events that influence cell responses and phenotypes. SERS-active scaffolds obtained by 3D-printing of hydrogel inks containing plasmonic NPs have been recently reported as efficient platforms for SERS monitoring in 3D (Figure e).[162] Such biocompatible inks can be 3D-printed to fine-tune the geometric features of plasmonic scaffolds, which warrants their complete integration within cellular networks at milli-/microscale resolution. A particular benefit offered by this technology is that the plasmonic structure sustaining 3D cell growth can simultaneously function as the SERS sensor, thereby facilitating the acquisition of information in the near vicinity of cells. This approach is foreseen to provide devices where different cell types, or even human explants (e.g., organoids grown from patients’ tumor samples) could be monitored over extended periods of time, with high spatial resolution. As a result, the effect of drugs could be tested on these 3D-printed devices, so that SERS analysis would reveal the response of cultured cells against each treatment.

Control over Biomolecular Fouling

Notwithstanding the ability of SERS toward the direct detection of different analytes at the point of interest, various steps of sample enrichment and isolation may be required prior to collecting SERS measurements, for example, in applications based on the spectroscopic features of extracellular vesicles or exosomes.[163−165] Such prerequisites (e.g., centrifugation or chromatography purification) are typically needed to remove, or significantly lessen, those components in biological media that may end up masking the presence of target analytes.[166] The adsorption of biomolecules at high concentrations will likely block the access of target analytes to plasmonic hotspots. Such a competitive adsorption drastically impairs the enhancement of the Raman signal and, as a result, reduces assay sensitivity and specificity. It is well-known that native proteins in biofluids are prone to binding onto bare nanoparticles, forming a so-called protein corona that increases background noise and may even prevent the detection of smaller-sized biomolecules.[167] Numerous strategies have been developed to modulate the adsorption of biomolecules onto metal surfaces, according to their size and charge, and even to regulate such interactions upon controlled exposure of the substrate. At this point, it is important to emphasize that SERS is a surface technique: a SERS spectrum consistently results from the interaction between sample and plasmonic substrate so that, for the same biofluid sample, substrates with different characteristics, for example, built from nanoparticles stabilized with different ligands (citrate, CTAB, etc.), may yield different spectra.

A common approach to prevent competitive binding while promoting the interaction of target analyte molecules with the sensor involves tailoring the surface chemistry of the plasmonic substrate. Many studies have reported the conjugation of plasmonic surfaces with different molecular probes that specifically react with target analytes, which are then detected through the vibrational changes derived from their interaction with pretagged Raman reporters.[168−172] Although such sensor configurations successfully repress signal overlap, they are limited to the established target-binding entities and can hardly be applied to multiplex detection, a valuable feature of SERS sensing (Figure a). A broadly applied strategy involves the use of self-assembled monolayers (SAM) to promote the binding of analytes with distinct affinities to SERS sensors and to minimize nonspecific fouling.[173−175] SAM-functionalized substrates lead to specific physicochemical interactions between plasmonic substrates and different sample constituents. The nature of the selected SAM can be tailored to either improve specificity, the SAM may enhance binding of a small subset of molecules or even larger entities, such as exosomes,[174,176] or to increase multiplexing.[177] For the latter approach, SAMs providing interactions with low specificity (i.e., biasing the approach of molecules with different properties to the proximity of the SERS substrate) can be used to diversify the SERS signatures that can be detected in complex environments (Figure b).

Figure 5

(a) Schematic representation of the modification of plasmonic surfaces to prevent fouling by proteins present in biological media while containing (i) probe molecules that interact directly with the target analyte and induce changes in SERS spectra or (ii) molecules that promote the adsorption of a selected group of targets with similar characteristics. Adapted with permission from ref (174). Copyright Nature 2016. (b) Arrays of plasmonic surfaces for label-free SERS with different self-assembled monolayers, which are depicted as brushes on the metallic support with different colors, showing that different sets of metabolites would interact with the metallic surface for each deposited SAM. A range of molecular interactions take place within complex biological media at each sensor unit, where mildly selective SERS enhancement of the constituents yields multiplex spectral data sets. Such an effect occurs in the example (right-hand panel) corresponding to the cell lysates of the Hs578T fibroblast, for which a different SERS spectrum is recorded from each functionalized Au-nanopillar array. This strategy enables the diversification of the detectable SERS signatures in complex environments and an increase in data dimensionality. Reproduced with permission from ref (177). Copyright Nature 2020. (c) Schematic view of the direct detection of small, charged molecules using oppositely charged hydrogels containing nanoparticles. Large proteins and like-charged molecules are excluded from the microgel while concentrating oppositely charged molecules, thereby facilitating SERS monitoring. Adapted with permission from ref (182). Copyright 2018 Wiley-VCH. (d) Creation of holes on a sheath polymer layer, previously deposited on plasmonic superlattices, by a photothermal effect; adsorption of molecules occurs only at the new measurement window, thereby preventing SERS memory effects during in situ measurements. The SERS spectra confirm that the vibrational fingerprint of molecules in solution (4-mercaptobenzoic acid in this case) is only registered after opening a measurement window by sufficient laser irradiation. Adapted with permission from ref (186). Copyright ACS 2021.

Other strategies exploit the coverage of plasmonic substrates with sheath layers made of porous materials, which would exclude the diffusion of large molecules toward the plasmonic nanostructure. In these systems, contamination of the plasmonic surface with macromolecules can be avoided by using, for example, mesoporous silica coatings, which block the transport of molecules larger than their pore size.[105,178,179] Alternatively, gold-coated TiO2 macroporous inverse opal structures were reported to yield a specific enrichment of extracellular vesicles within a wide size range (30–150 nm), which could be subsequently screened by SERS.[180] In addition to size-dependent impediments, hydrogel networks can additionally facilitate the selective penetration of molecules by electrostatic interactions while preventing like-charged small molecules to reach the plasmonic component (Figure c).[181,182] An interesting case is provided by molecularly imprinted polymers (MIPs), which can be devised to display a predetermined selectivity for a given analyte.[183−185] In this configuration, the MIP layer renders the SERS substrate capable of capturing target molecules on its surface with high selectivity, thereby mimicking the interaction between bioreceptors and antibodies.

Finally, it has been recently reported that the irreversible adsorption of (analyte) molecules onto plasmonic substrates may interfere in subsequent measurements, which is known as the SERS-memory effect and represents a source of uncertainty in real-time measurements.[186] To resolve this issue, micron-thick, thermolabile polymer layers were deposited on top of plasmonic substrates, providing a physical barrier against molecular diffusion and adsorption. Laser irradiation at high fluence induces plasmonic heating of the underlying nanoparticles, so the sheathing layer degrades under high local temperature and opens a measurement window at the selected measurement time and location. Depending on the nature of the polymer layer, this process can be applied in a reversible (poly-N-isopropylacrylamide, pNIPAm)[187] or an irreversible (poly(lactic-glycolyc acid), PLGA)[186] manner. As soon as a micrometer-sized window is created in the polymer layer, the molecules in solution can reach the plasmonic surface and can be registered by SERS (Figure d). This process can be repeated as many times as required for monitoring the event of interest with no interference from earlier measurements. In biological systems, valuable information can be obtained about the evolution of the cellular state.

Hitting the Target with the Aid of Artificial Intelligence

Although SERS has been applied to monitor biomarkers for different disorders (e.g., neurological, infectious, and genetic diseases),[188−191] the direct addition of complex biological media onto a plasmonic substrate may give rise to SERS spectra from which the presence of individual biomarkers cannot be identified. Most studies are typically based on the screening of biomarkers that display both high affinity toward metal surfaces and high Raman cross sections (e.g., metabolites containing aromatic moieties). When such molecules are present in the probe solution, they can easily dominate the SERS signal, masking the presence of other analytes while their characteristic fingerprint is directly peeked in the spectra. This screening, even at very low concentrations, is highly advantageous, therefore, reaping the reward of fast SERS monitoring. This is likely the underlying reason why similar biomolecules are reported in most ongoing applications; a myriad of studies have demonstrated that SERS can become extremely effective for monitoring nitrogenous bases (mainly purine derivatives or the nicotinamide ring),[51,192−194] aromatic amino acids or metabolites (tryptophan, phenylalanine, tyrosine, and their derivative products),[193−195] and thiolated biomolecules (glutathione, cystine, among others).[65,196] In contrast, the identification of other analytes with different properties has not been achieved in similar terms so far. In this scenario of impaired signal enhancement, classical methods for SERS spectral analysis, in which one can establish a direct correlation between vibrational peaks and the presence of individual metabolites, are no longer sufficient. Therefore, chemometrics methods have emerged as a reliable alternative to conventional approaches. In this context, multivariate and artificial intelligence (machine learning in particular) algorithms can be devised to substantially improve data processing, for example, for calibration and classification of multicomponent samples and identification of interferences in complex biochemical systems.[197]

A wide range of multivariate statistical and machine learning methods are currently available to decipher the optically rich and complex information contained in SERS spectra.[198] For instance, some groups have tackled quantification challenges of biomarkers by using a so-called SERS digital procedure,[82,199] which considers the number of SERS event counts rather than SERS intensities and couples them with machine learning.[200] Although these methods have been used to identify the presence of specific biomarkers, they are primarily employed for clinical diagnosis. This means that they seek to determine whether a patient is prone to suffer from a specific pathology, using SERS analysis of selected biofluids, and even including additional predictions such as the stage of the disease and its origin.[201−206]

Multiple strategies, from “simple” exploratory analyses through machine learning methods, have been exploited for pattern-recognition and classification of SERS spectra.[207,208] Unsupervised methods, such as principal component analysis (PCA), are routinely used to visualize variations in the data by compressing the dimension of the SERS spectra, with minimal loss or reduction of uncorrelated variables.[209] By applying PCA, one may discover hidden patterns in the original data with minimal intervention (see Figure a).[210−213] However, as long as labeled data are available, which is often the case in normal and altered conditions (e.g., classification of healthy vs cancer cells), a supervised algorithm would be more reliable in SERS analysis. Indeed, the separation of classes is not a trivial task because even spectra from the same class may vary significantly, for example, one single metabolite might exhibit different spectra depending on its molecular orientation on the SERS substrate. On the other hand, vibrational features from various classes are likely to overlap, thereby hampering the visualization of classes by simple unsupervised learning (Figure b).[214] Sophisticated machine learning methods, such as deep learning, have been widely adopted by the SERS community because of being less cumbersome for the nonspecialist and many user-friendly and open-source frameworks and libraries are easily accessible, unlike other strategies based on shallow artificial neural networks and ensemble methods, which require extensive feature extraction procedures to avoid overfitting.[215]

Figure 6

(a) Unsupervised analysis of SERS spectra obtained from three bacterial strains of L. monocytogenes, revealing the existence of different clusters and their proximity. Strains represented by blue and red dots express a similar set of genes, while the green strain does not. Reproduced with permission from ref (209). Copyright 2018 Springer. (b) General scheme of a supervised, machine-learning-based approach to analyze data sets of SERS spectra. Training data are initially fed into the system to create the network. At this point, each data point is labeled according to its metabolites or condition. Subsequently, unknown data are classified by the trained machine-learning algorithm. Reproduced with permission from ref (214). Copyright 2019 ACS. (c) Variational autoencoders (VAE) latent space depicting the SERS spectra of bacteria after a rapid antimicrobial susceptibility test. The method achieves high accuracy in discriminating susceptible vs resistant to antibiotic cultures. The black arrows indicate the data-informed approach that allows the user to improve the classification performance of the model. Adapted with permission from ref (217). Copyright 2020 ACS.

Several studies have shown the ability of deep learning to outperform the shallow machine learning methods by accounting for the multiple SERS responses to the same conditions, for example, due to intrinsic interpatient variability, with no need for feature extraction or reduced-variance methods, as was shown in the analysis of spectra obtained from exosomes of different cancer patients.[216] In this direction, a generative machine learning method termed variational autoencoders (VAE) was applied to study antibiotic susceptibility (Figure c).[217] The analysis was able to discriminate the bacterial metabolome after antibiotic treatment from the untreated patient with an accuracy greater than 95%. Additionally, important vibrational features were identified in the output from the trained VAE model, which were associated with a metabolite data set and subsequently used to improve the generalization of the model. Arguably, this data-informed approach, also known as transfer learning,[218] appears as a promising tool toward the classification of other biological events in complex metabolite systems. Overall, these pioneering works demonstrate the potential of deep-learning methods to produce highly predictive models with rich vibrational information contained in the Raman spectra. The combination of data-informed analysis and SERS will enable researchers to go beyond perfecting machine learning models, while saving time and reducing costs related to data acquisition, training long data sets and sophisticated GPUs (graphics processing unit) and CPUs (central processing unit) to generate robust models. Still, new methodologies are needed to provide access to high-throughput approaches for SERS screening in precision medicine, which will be relevant toward clinical translation.[219] An example in this direction comprised the combination of a portable SERS spectrometer with deep-learning as a point-of-care automatic device, providing early diagnosis of multiple diseases in a single run.[220]

Thus far, we have covered potential applications of the combination of SERS and data processing toward extracting meaningful information from complex biological systems. Additional considerations must be taken for the democratization of machine learning-based methods in SERS biomonitoring; that is, it is still challenging to reproduce measurements in different laboratories. In this regard, a large-scale multi-institution study was carried out in an attempt to compare and validate SERS for biomedicine under varying laboratory environments (e.g., applying different SERS substrates and instruments with different configurations).[221] Updated protocols are still required to warrant the quality of biosensors combining SERS and machine learning methods, similar to those previously proposed for Raman analysis.[222] The broad application of these protocols should drive the implementation of shared key steps in data processing, such as the efficient removal of artifacts without corrupting the nature of the original data, as well as the use of validated data processing protocols, such as data augmentation, toward producing more predictive models.[223,224]

Updating SERS Sensors for the Near Future

Research efforts over the past decades have been primarily focused on extending the possibilities offered by the direct SERS detection of complex molecules in solution, for example, biofluids. An alternative approach is the application of SERS to the detection of metabolites in the gas state. This strategy leverages the relatively simple composition of gases compared to biofluids and, therefore, the smaller number of molecules competing with the targets. Therefore, gas-state SERS sensors may significantly facilitate metabolite detection and rapid diagnosis of various severe diseases. Reported approaches range from the simple deposition of metal nanoparticles on a gas filter to complex systems in which NPs are carefully combined with other elements. As a simple example of application, AuNPs were loaded onto a flat filter by vacuum filtration to detect common foodborne bacteria in pork samples: Escherichia coli, Staphylococcus aureus, and Pseudomonas aeruginosa. The SERS spectra acquired after 1 day of exposure proved the detection of gaseous metabolites with a sensitivity of 10 nM, comparable to mass spectrometry and much lower than earlier reports on gas chromatography.[225] Other approaches implemented more sophisticated methods, intended to enhance the adsorption of gas molecules onto solid SERS substrates, e.g., by covering the substrates with polymer layers carrying different functionalities or involving microfluidic devices.[226,227] More complex systems comprising metal–organic frameworks (MOFs), such as ZIF-8 to coat gold particles, are of particular interest. Characteristic gaseous aldehydes excreted by lung cancer patients could be screened thanks to their enhanced accumulation on SERS-active surfaces by the ZIF-8 layer (Figure a).[228]

Figure 7

(a) Schematic representation of the detection of volatile organic compounds secreted by patients (left) and effect of a ZIF-8 MOF layer on the enhanced molecular adsorption onto SERS surfaces (right). Reproduced with permission from ref (228). Copyright 2018 Wiley-VCH. (b) Sketch showing the induced activity of a CRISPR-Cas12a complex upon binding to its target DNA. CRISPR-Cas12a activation causes the release of the SERS probes from graphene oxide-Au nanoflower arrays, resulting in a decrease of the SERS signal. Reproduced with permission from ref (236). Copyright ACS 2021.

Another recent and exciting sensing strategy involved the combination of SERS with biotechnology tools. The introduction of biorecognition elements on SERS platforms offers selective affinity to target analytes that largely exceeds those obtained by other methodologies. So far, most of these strategies have been based on the use of SERS tags or reporters (i.e., indirect sensing) and the integration of conventional biotransducers, such as enzymes or antibodies, with high specificity and affinity for their targets.[229,230] Although such SERS sensors could compete with fluorescence/colorimetric methods and even reach higher sensitivity, the experience accumulated over the years with such routine technologies, as well as the additional equipment requirements for implementing SERS measurements, have delayed their replacement in the clinic. Therefore, research has been devoted toward providing faster, single step, more reproducible, and robust SERS measurements of multiple biomarkers simultaneously.[231] In this context, a promising complement for SERS biosensors are oligonucleotides. These small size (10–20 kDa) biomolecules can be easily conjugated to metal nanoparticles with controlled orientation and high density on the nanoparticle surface. As an example, aptamers (single-stranded oligonucleotides that fold into well-defined architectures) can be designed to attach on the NPs surface and specifically interact with predefined biomolecules, facilitating acquisition of their SERS signal.[232−235]

Overall, both SERS and biotechnology represent cutting-edge areas of research, and hence, opportunities for their joint integration into disruptive sensors are likely to sprout in the future. The combination of SERS and CRISPR-associated nucleases has been recently reported as a powerful tool for the rapid detection of infectious DNAs. One of the few reported approaches comprises a viral biosensor with a CRISPR-Cas12a-assisted SERS analytical system. As depicted in Figure b, the binding of a CRISPR-Cas12a complex with its corresponding target is followed by cleavage of a ssDNA connecting graphene oxide-Au nanoflowers to a Raman reporter, with a significant decrease in the SERS signal. Viral DNA could be detected by this methodology at extremely low concentrations, down to 1 aM, within only 20 min.[236] The combination of SERS and CRISPR systems has not only been tested for viral DNA detection, but also for the detection of multidrug-resistant bacteria.[237]

Conclusions

The development of new technologies for precision medicine will guide the evolution of biomedicine, eventually defining the way it will be practiced in the future. We anticipate that the tools and methods presented throughout this Perspective will carry over into the future, to be applied to next-generation diagnostic technologies based on SERS. However, despite of the knowledge gained during recent years, further adjustments are still needed, including the combination of recently reported advances (e.g., to control biomolecular fouling and related to integration of biological systems) within a single sensor or integration and interpretation of complex spectral signatures (artificial intelligence), which will result in multifaceted SERS sensors for upcoming challenges.