Multiplexing potential of NIR resonant and non-resonant Raman reporters for bio-imaging applications

Abstract

Multiplexed imaging, which allows for the interrogation of multiple molecular features simultaneously, is vital for addressing numerous challenges across biomedicine. Optically unique surface-enhanced Raman scattering (SERS) nanoparticles (NPs) have the potential to serve as a vehicle to achieve highly multiplexed imaging in a single acquisition, which is non-destructive, quantitative, and simple to execute. When using laser excitation at 785 nm, which allows for a lower background from biological tissues, near infrared (NIR) dyes can be used as Raman reporters to provide high Raman signal intensity due to the resonance effect. This class of imaging agents are known as surface-enhanced resonance Raman scattering (SERRS) NPs. Investigators have predominantly utilized two classes of Raman reporters in their nanoparticle constructs for use in biomedical applications: NIR-resonant and non-resonant Raman reporters. Herein, we investigate the multiplexing potential of five non-resonant SERS: BPE, 44DP, PTT, PODT, and BMMBP, and five NIR resonant SERRS NP flavors with heptamethine cyanine dyes: DTTC, IR-770, IR-780, IR-792, and IR-797, which have been extensively used for biomedical imaging applications. Although SERRS NPs display high Raman intensities, due to their resonance properties, we observed that non-resonant SERS NP concentrations can be quantitated by the intensity of their unique emissions with higher accuracy. Spectral unmixing of five-plex mixtures revealed that the studied non-resonant SERS NPs maintain their detection limits more robustly as compared to the NIR resonant SERRS NP flavors when introducing more components into a mixture.

Introduction

Surface-enhanced Raman spectroscopy (SERS) has shown great potential for biomedical applications.^1-4 SERS nanoparticles (NPs) rely on a metallic core to increase the intensity – hence, surface-enhanced – of the reporter’s Raman scattering signal.⁵ Each reporter molecule possesses a unique combination of Raman bands, thus creating a unique SERS NP “flavor” characterized by its sharp, narrow peaks resulting from the inelastic scattering of incident light. The characteristic spectral features of these NPs allow for the high sensitivity and multiplexing capabilities that make SERS ideal for molecular imaging and immunoassays, among many other applications.^6-9 SERS NPs can target specific biomarkers once they are conjugated with targeting ligands, such as antibodies, peptides, and aptamers. Most notably, targeted SERS NPs have been used to target various cancer biomarkers, such as cluster of differentiation-47 (CD47) on breast cancer cells,¹⁰ human epidermal growth factor receptor 2 (HER2),¹¹ epidermal growth factor receptor (EGFR),^12,13 estrogen receptor (ER)^14,15 on breast cancer cells, folate receptor (FR) in human ovarian carcinoma,¹⁶ and integrin α_vβ₃ on lung cancer cells.¹⁷

Some research groups have demonstrated single-nanoparticle sensitivity using a variety of nanoparticle structures.^18,19 As for multiplexing with SERS NPs, the narrow characteristic Raman peaks have previously allowed successful spectral unmixing of up to 26 SERS reporter molecules both in vitro and in vivo.¹¹ These multiplexing capabilities make SERS NPs ideal for shedding more light on a patient’s individual cancer by providing important molecular insights, including tumor heterogeneity and interactions with the immune system, which can be used to predict the response to treatments and better personalize a patient’s therapy. In particular, many facets of medicine require the ability to label abnormal cells to make their presence and properties known – surgery for location and resection of tumors, pathology for diagnosis and subtyping of cancer cases, and pharmaceutical research to characterize responses to experimental treatments. In each discipline, the ability to perform labeling and imaging for a large number of cellular targets simultaneously would dramatically increase our ability to capture a patient’s molecular profile and gain important biological insights. Academic and industry researchers interested in understanding the complex nature of a patient’s tumor could simultaneously explore how variables such as specific biomarker expression, tumor microenvironment, immune cell infiltration, and inflammation could have an effect on their newly derived experimental treatments. Researchers are already adopting solutions from the growing “high-plex” imaging market to capture the molecular expression profile of tumors using advanced pathology imaging techniques such as immunofluorescence imaging (IFI) and imaging mass cytometry (IMC) on histological sections.

Surface-enhanced resonance Raman spectroscopy (SERRS) is a growing sub-category of SERS-based Raman imaging.^20-23 SERS uses non-resonant Raman reporter molecules such as trans-1,2-bis(4-pyridyl)ethylene (BPE).²⁴ In contrast, SERRS uses chromophores close in energy to the excitation frequency, thus generating an additional source of enhancement of the Raman signal.^1,25 Some examples of resonant Raman reporter molecules include rhodamine 6G (Rh6G) for 514 nm excitation wavelength,²⁶ malachite green isothiocyanate (MGITC) – for 633 nm,²⁷ and the near-infrared dye 3,3′-diethylthiatricarbocyanine (DTTC) – for 785 nm.^23,28 When using laser excitation at 785 nm, which allows for a lower background from biological tissues, near infrared (NIR) dyes can be used as Raman reporters to provide high Raman signal intensity due to the resonance effect. Thus, for biomedical imaging, NIR dyes, such as DTTC, IR-775, IR-780, IR-792, and IR-797 (for the IUPAC names, see the Experimental section), have been extensively used in Raman imaging of biological tissues. This class of imaging agents are known as SERRS NPs. With SERRS, the visible excitation and emission spectra of resonant Raman reporters overlap with the wavelength of the incident light being used to excite the NPs.

Currently, there is little information regarding the benefits of using non-resonant Raman reporter molecules over those of resonant Raman reporters, or vice versa for multiplexing purposes. Although both have been reported to be sensitive and capable of multiplexing in vivo^8,11,23,29 and ex vivo,^12,13 there is yet to be a direct comparative study exploring their respective multiplexing potential, side by side, in a controlled setting. Herein, we focus on determining the multiplexing capabilities of five non-resonant Raman reporters and five NIR-resonant heptamethine cyanine dyes that have routinely been used for biomedical imaging^{2,8,11,13,23,29} with the intention to increase plexity (number of flavors) for accurate and robust spectral unmixing. Both types of Raman reporters were studied on gold nanospheres with a silica coating. In comparing these resonant and non-resonant Raman reporter types, our findings aim to further understand their respective multiplexing capabilities in order to facilitate further development of both SERS and SERRS NP types, that is specifically important for biomedical applications.

Experimental

Reagents and materials

Gold(iii) chloride hydrate (HAuCl₄·xH₂O, 99.995%), trisodium citrate dihydrate (C₆H₅Na₃O₇·2H₂O, 99.0%), hydroxylamine hydrochloride (98.0%), sodium borohydride (98.0%), (3-aminopropyl)trimethoxysilane (APTMS, 98%), DTTC (3,3′-diethyl-thiatricarbocyanine iodide, 99%), IR-775 chloride (2-[2-[2-chloro-3-[2-(1,3-dihydro-1,3,3-trimethyl-2H-indol-2-ylidene)ethylidene]-1-cyclohexen-1-yl]-ethenyl]-1,3,3-trimethyl-3H-indolium chloride, 90%), IR-780 iodide (2-[2-[2-chloro-3-[(1,3-dihydro-3,3-dimethyl-1-propyl-2H-indol-2-ylidene)ethylidene]-1-cyclohexen-1-yl]ethenyl]-3,3-dimethyl-1-propylindolium iodide, 95%), IR-792 perchlorate (2-[2-[3-[(1,3-dihydro-3,3-dimethyl-1-propyl-2H-indol-2-ylidene)ethylidene]-2-(phenylthio)-1-cyclohexen-1-yl]ethenyl]-3,3-dimethyl-1-propylindolium perchlorate, 99%), IR-797 chloride (2-[2-[2-chloro-3-[2-(1,3-dihydro-1,3,3-trimethyl-2H-indol-2-ylidene)ethylidene]-1-cyclopenten-1-yl]-ethenyl]-1,3,3-trimethyl-3H-indolium chloride, 70%), 4,4′-bis(mercaptomethyl)biphenyl (BMMBP, 97%), 4,4′-dipyridyl (44DP, 98%), sodium silicate aqueous solution (~26.5%) and tetraethyl orthosilicate (TEOS, 99%) were purchased from Sigma-Aldrich. 1,2-Bis(4-pyridyl)ethylene (BPE, 98%) was purchased from TCI; 5-(4-pyridyl)-1,3,4-oxadiazole-2-thiol (PODT, 97%) – from Alfa Aesar; and 5-(4-pyridyl)-1H-1,2,4-triazole-3-thiol (PTT, 98%) – from Acros Organics. Anhydrous DMSO (99.7%, <0.005% water) and ethanol were ordered from Koptec, Acros Organics, and Corning, respectively. Coffee filter paper (Hario V60 for 02 size dripper, Japan) was used as the substrate for Raman imaging of the SERS and SERRS NPs. Deionized water (Milli-Q grade, Millipore) with a resistivity of 18.2 MΩ cm was used throughout the experiment.

SERS and SERRS nanoparticle preparation

A modified method of hydroxylamine seeding of colloidal gold nanoparticles (AuNPs) was used to fabricate spherical AuNPs. Briefly, 3 mL of 30 mg mL⁻¹ HAuCl₄ was added to 450 mL of cold water (4 °C) under vigorous stirring. Then, 0.6 mL of a 0.135 g mL⁻¹ sodium citrate and 0.085 g mL⁻¹ hydroxylamine hydrochloride mixture was added rapidly. After 10 s, 105 μL of 1‰ NaBH₄ was injected rapidly. The colloidal solution was stirred for an additional 10 min.

For the SERS NP preparation, the as-synthesized AuNPs (20 mL; 50 pM) were rendered vitreophilic with the dropwise addition of 300 μL of 0.1 mM APTMS. After 15 min of vigorous stirring, a solution of one of the Raman reporters in ethanol (0.1 mM) was added: 200 μL of BPE, 100 μL of BMMBP, 350 μL of PODT, 100 μL of PTT, or 100 μL of 44DP. After 5 min of stirring, a total volume of 400 μL of 2.16 wt% sodium silicate was added. For the silica shell growth via the Stöber method, 50 mL of ethanol was added to the NP solution. Growth of 30 nm of an additional silica shell was accomplished by the addition of 250 μL of ammonia and 50 μL of TEOS (Fig. 1). The reaction mixture was stirred at room temperature for 24 h.

Fig. 1.

Scheme of SERRS NP preparation.

For preparing SERRS NPs, the as-synthesized citrate-stabilized AuNPs (20 mL; 50 pM) were centrifuged at 5000g for 10 min at 4 °C, and the supernatant was removed. The softly pelletized AuNPs were collected and dialyzed for 5 days with a 7 kDa molecular weight cut-off dialysis cassette. Then, 1 mL of the resulting 0.2 nM AuNP solution was mixed with 3 mL of ethanol, 275 μL of TEOS, 150 μL of H₂O and 10 mL of isopropanol. Depending on the SERRS flavor, 40 μL of 20 mM DTTC, 20 mM IR-775, 20 mM IR-780, 5 mM IR-792, or 5 mM IR-797 in DMF was added. Then, 200 μL of ammonia was introduced into each tube. After 1.5 h of stirring, the NP suspensions were purified by centrifugation at 2000g for 15 min (repeated 5 times).

Nanoparticle characterization

The bare AuNP colloidal solution was characterized using a Cary 60 UV-Vis (Agilent Technologies, USA) spectrophotometer. The maximal absorption band of AuNP solution was observed at 535 nm. The size and concentration were verified by dynamic light scattering (DLS) and nanoparticle-tracking analysis (NTA) using a Zeta Sizer Nano ZS (Malvern Panalytical, UK) and a NanoSight NS300 (Malvern Panalytical, UK), respectively. According to DLS and NTA measurements, the size distribution was 60 ± 4 nm. SERS spectra were obtained to confirm the success of the labeling procedure and SERS activity by using an RA816 Biological Analyser Raman instrument (Renishaw, UK). All SERS spectra were acquired using a 100 mW 785 nm near infrared diode laser using a ×50 objective lens (NA 0.50). We used a silicon wafer for calibration. The size and concentration of Au@SiO₂ were determined by UV-vis spectroscopy and NTA.

DFT calculations

Geometry optimization (Fig. S7-S12^†) and frequency calculations (Fig. S1^†) for isolated Raman reporter molecules were performed using the hybrid GGA B3LYP functional with D3 dispersion correction and the 6-311++G(d,p) basis set²⁴ within the Q-Chem 6.1 package.³⁰ All the calculated Raman cross sections were scaled with a scaling factor of 0.980, and convoluted by a Lorentzian function with the full width half maximum of 10 cm⁻¹.

Spectral correlation and condition number

The spectral correlation between SERS and SERRS spectra was calculated using the following equation:³¹

$$C_{i,j}=\sum _{w_{\text{min}}}^{w_{\max }}{S}_i[w]\cdot S_j[w],$$ (1)

where $C_{i,j}$ is the spectral correlation between flavors $i$ and $j$; $S_i$ and $S_j$ are the discrete normalized Raman spectra of flavors $i$ and $j$, respectively; and $w_{\text{min}}$ and $w_{\max }$ are the minimum and maximum wavelengths of the spectra, respectively.

The condition numbers for combinations of SERS and SERRS spectra with various plexities was calculated using the following equation:²³

$$\kappa (A)=\frac{{\sigma }_{\max }(A)}{{\sigma }_{\text{min}}(A)},$$ (2)

where $\kappa (A)$ is the condition number for the spectral matrix $A$ for each combination of NP flavors, and ${\sigma }_{\text{min}}(A)$ and ${\sigma }_{\max }(A)$ are the maximal and minimal singular values of $A$, respectively.

Raman imaging with SERS and SERRS nanoparticles

Evaluation of sensitivity and multiplexing capabilities of SERS and SERRS NP types was performed by Raman imaging of NPs deposited onto a filter paper substrate. Briefly, we prepared circular pieces of coffee filter paper (d = 7 mm). Then, for each NP concentration and their mixtures, 3 μL of the studied solution were dropcast onto the paper circle and allowed to dry for 20 min at room temperature. For Raman imaging with SERS and SERRS NPs, we used a 2.5% neutral-density filter and 0.1 s acquisition time. All Raman spectra were collected in the 0–2030 cm⁻¹ spectral range with 1.5 cm⁻¹ spectral resolution.

Spectral unmixing of Raman images

Identical laser power and spectral collection times were used to measure the reference spectra and the subsequent Raman imaging. The non-negative least-squares (NNLS) method, also called the linear unmixing method and K-matrix method,^9,32 was used to perform a quantitative analysis of Raman images. Similar to the direct classical least squares (DCLS) method, NNLS finds the linear combination of spectra from the pure components contained in the sample that most closely matches the Raman spectrum of the sample, but with the constraint that the weight values (WVs), or scaling factors, must not be negative. Pure component spectra of all SERS and SERRS NPs were acquired from a pure 3 μl sample aliquoted onto a stainless-steel microscope slide. The WVs derived by the NNLS analysis are proportional to the concentration of the pure components. The NNLS method was chosen because all the Raman spectra of the paper substrate background and NPs were available, and these components have a spectral overlap. For the quantitative analysis, Windows-based Raman Environment (WiRE) software (Renishaw) was used. Before each Raman imaging experiment, pure spectral components were measured from the SERS and SERRS NPs along with the paper substrate background. Spectral demultiplexing of Raman mapping in the 200–2000 cm⁻¹ range without baseline subtraction resulted in interpretable images made up of WVs, or abundances, of the SERS and SERRS NP flavor content.

Statistical analysis

The Raman data were obtained from multiple points within Raman maps, and the mean and the standard deviation (SD) were derived from the Raman maps. Average spectra and spectral SDs were calculated from the Raman spectra (n = 500) using WiRE software. Mean WVs for each NP flavor were derived from the Raman image analysis, acquired from the coffee filter paper substrate (ca. 500 pixels per circle) using ImageJ software (NIH).

Results and discussion

Optical properties of SERS and SERRS nanoparticles

The most attractive advantage of Raman imaging with SESR-active NPs among other imaging techniques is its unprecedented multiplexing potential.³³ Consequently, the evaluation of multiplexing capabilities and spectral signatures of both types of SERS and SERRS NPs is of high importance.^11,34

Starting from an in silico analysis, Raman scattering properties of SERS and SERRS NPs can be predicted by density functional theory (DFT) calculations.²⁴ Herein, as NIR dyes for Raman labeling, we assess DTTC, IR-775, IR-780, IR-792, and IR-792 (Fig. 2), that have been previously used for biomedical imaging, including multiplexing applications.^23,35,36 These NIR dyes are available and stable and are known to coordinate to the gold surface.

Fig. 2.

Optical properties of 10 SERS-active NP flavors. Raman spectra of SERRS NP flavors, each bearing spectral features of DTTC, IR-775, IR-780, IR-792, and IR-797, with respective DFT-calculated Raman spectra of the NIR dyes (in gray). Chemical structures and optical properties (excitation and emission) with the identification of ${\lambda }_{\mathsf{ex}}$ 785 nm used for Raman measurements. Raman spectra of SERS NP flavors, each bearing spectral features of BPE, BMMBP, PODT, PTT, and 44DP, with respective DFT-calculated Raman spectra of the non-resonant Raman reporters (in gray). The DFT spectra were simulated for the Raman reporters attached to a Au₂₀ cluster with B3LYP/6-311++G(d,p) level of theory. For the baseline corrected Raman spectra of SERRS and SERS NP flavors with spectral background subtracted, see ESI Fig. S2-S6.^†

These NIR dyes (Fig. 2), opposite to non-resonant Raman reporters, are physisorbed on the AuNP surface by the nitrogen atom in the pyrrol.²⁸ In general, a Raman band is drastically enhanced on the plasmonic NP if the polarizability change of the associated mode is perpendicular to the gold surface.³⁷ When the Raman reporter molecule approaches the surface of the AuNP face-on, then out-of-plane modes should be most enhanced; if a molecule coordinates to the metallic surface edge-on, then in-plane modes should appear the most enhanced. With the face-on orientation of the Raman reporter to the gold surface, both physisorption and coordination are possible. With edge-on orientation, chemisorption is most likely to take place. For instance, non-resonant Raman reporters like BPE, BMMBP, PODT, PTT, and 44DP demonstrate the strongest enhancement for bands at ca. 1000 cm⁻¹ (in-plane ring breathing), ca. 1200 cm⁻¹ (in-plane C─N stretching and C─H bending in ring), ca. 1600 cm⁻¹ (in-plane C─C stretching and C─H bending in ring) and demonstrate edge-on DFT-simulated orientation.²⁴ This indicates their edge-on coordination to the AuNP surface and supports chemisorption through the nitrogen in pyridine or sulfur in the sulfhydryl group (see ESI Fig. S14^†).²⁴

Whilst the NIR dyes possess strong Raman peaks at ca. 520 cm⁻¹ (out-of-plane N─C─C stretching in pyrrole ring and C─H twisting in benzene ring), ca. 560 cm⁻¹ (out-of-plane N─C─C stretching in pyrrole ring), ca. 800 cm⁻¹ (out-of-plane C─H wagging in benzene ring), and ca. 930 cm⁻¹ (out-of-plane C─H twisting in benzene ring) proving their face-on orientation and, thus, the physisorption playing the most important role in their interaction with the AuNPs (see ESI Tables S1-S5^†).

Moreover, a protective shell is essential to enhance the colloidal stability of the NPs and prevent the dissociation or structural alternation of Raman reporters in complicated biological media which the SERRS NP might encounter during in vivo circulation.³⁸ However, the weak attraction of the NIR dyes to gold causes one of the bottleneck issues in keeping the reporters on the nanoparticle surface during a silica coating process.³⁹ Thus, other smart approaches for coating SERRS NPs should be thoroughly investigated.^39,40

For the prepared SERRS NPs, we assessed the average number of NIR dye molecules per NP based on the absorption of supernatants during the washing steps. The prepared SERRS NPs had 9.9 × 10⁵ molecules of DTTC per NP contrast agent, 7.2 × 10⁶ – of IR-775, 4.5 × 10⁶ – of IR-780, 6.3 × 10⁶ – of IR-792, and 1.9 × 10⁶ – of IR-797, respectively. We observed a significant background signal associated with the Raman spectra of the NIR resonant dyes as opposed to the non-resonant dyes. This background is mostly attributed to fluorescence, which can swamp Raman signals, and can result in poorer reproducibility and stability of Raman signals.⁴¹ As shown in Fig. 2, absorption wavelengths of the NIR dyes match 785 nm – the NIR laser irradiation for Raman excitation wavelength, most frequently used and most suitable for bio-imaging applications.⁴² Such overlap provides an additional gain of SERRS and improves the absolute Raman signal intensities of the contrast agent.^6,29,43 However, on the other hand, after efficiently absorbing light at 785 nm, the NIR dyes can also emit light, which might explain the presence of extra-background for SERRS NPs spectra as compared to SERS NPs, as shown in Fig. 2. As a rule, the height of the fluorescence background changes non-linearly with an analyte or contrast agent concentration. Therefore, fluorescence background removal is one of the foremost challenges for quantitative analysis of Raman spectra in many samples.⁴⁴ However, the background subtraction for multi-component mixture is not straightforward. We also observed crowded peaks in the spectral fingerprint region associated with the Raman spectra of the NIR resonant dyes as opposed to the non-resonant dyes (Fig. 2). This spectral crowding is concerning as it will ultimately limit the potential to achieve higher plexities comparable with commercial systems already capable of achieving 50-plex imaging.^45,46 Another challenge for achieving higher plexities with SERRS NPs, comes from the limited number of NIR dyes available. So far, the largest number of NIR dyes used for multiplexed Raman imaging is 5,²³ while there are at least 31 small organic molecules which can serve as effective Raman reporters for SERS NPs,⁴⁷ with even more SERS barcodes waiting to be discovered.²⁴ No doubt, more NIR dyes can be synthesized for screening if they can serve as more sensitive reporters for Raman imaging.^29,48,49 But NIR dyes will still have complex spectral barcodes crowded in the fingerprint region, with interference of fluorescence background, and thus a significant degree of overlap for the main Raman bands.

Multiplexing potential of SERRS and SERS nanoparticles

Since its discovery, Raman scattering has been a powerful tool for analyzing the chemical composition of samples since each band’s appearance is connected to the vibrational modes of specific chemical bonds. Thus, Raman spectroscopy can be successfully utilized for intrinsic imaging of biological tissues.^50,51 One of the major advantages of using Raman-labeled contrast agents is the ability to design a NP with the desired features (i.e., spectral fingerprint) for the intended biomedical application. Although NIR dyes offer an added sensitivity boost in Raman imaging, they typically possess rather similar chemical structures.²¹ The identical chemical moiety is indicated with blue color in Fig. 2. In this case, one may expect a significant spectral overlap of not only the fluorescence background, but Raman bands themselves (see ESI Fig. S13^†), based on the relatively minor modifications in the structural moieties of the NIR reporters.

For quantitative multiplexed Raman imaging, it is essential to reliably spectrally separate each Raman-labeled NP reference spectrum from their mixed spectra. As we increase the number of NP flavors, which we intend to use simultaneously in one sample, the probability for spectral overlap also increases, thus making the unmixing of the multiplexed spectra more challenging and/or less accurate.⁵² The bulkier chemical structures of NIR dyes result in more crowded Raman spectra as compared to non-resonant smaller reporters (see ESI Fig. S14^†). To estimate the potential impact, we first calculated the correlation as a spectral inner product³¹ between all possible 2-plex combinations of the Raman reporters among each group of five SERS NP (Fig. 3a) and five SERRS NP flavors (Fig. 3b). The correlation matrix was built from the normalized SERS spectrum of each NP flavor in the range of 200–2000 cm⁻¹. In these heat maps, color indicates the value of the spectral correlation, where yellow stands for a higher degree of overlap, and dark blue color – for a lower overlap. The two types of Raman reporters, non-resonant and resonant NIR dyes, indeed demonstrate different levels of spectral overlap with one another, as evidenced by the yellow off-diagonal elements of the matrix (Fig. 3b). For instance, IR-797 shows a significant overlap with DTTC, IR-775, IR-780, and IR-792 with correlation values close to 1.0 (see ESI, Table S7^†). Among 26 SERS NP flavors we had recently reported,¹¹ some pairs demonstrated spectral correlation below 10%. For this work, we chose the most challenging subsets for an accurate spectral unmixing combination of five non-resonant Raman reporters: BPE, BMMBP, PODT, PTT, and 44DP. This subset of SERS flavors demonstrates the highest spectral similarity possible within the 26-membered library of SERS NPs.¹¹ However, in contrast, non-resonant smaller reporters still possess simpler spectra with more sparse Raman peaks resulting in less spectral overlap across flavors, as indicated by the blue off-diagonal elements of the matrix (Fig. 3a).

Fig. 3.

Multiplexing compatibility of SERS and SERRS NPs. (a) Correlation matrices built from the spectra of SERS NPs (a) and SERRS NPs, (b) demonstrating similarity among each type of flavors. The color bar indicates the level of fitting signals, where 1 (yellow) means 100% spectral overlap of two flavors and 0 (dark blue) means 50% spectral overlap of two flavors. (c) The lowest (well-conditioned subsets) and the highest (ill-conditioned subsets) condition numbers for different plexities of SERS and SERRS NP flavors (see ESI Table S8^†). A lower condition number is preferred to achieve easier unmixing of the NP subsets and higher plexity imaging. Notice how both the ill- and well-conditioned subsets of NP mixtures maintain a low condition number for the SERS NPs as opposed to the SERRS NPs. For deriving the condition number for each combination, we used the normalized reference spectra of SERS and SERRS NPs without baseline subtraction.

Next, considering higher plexities which are outside of the 2D covariance matrix, we calculated condition numbers for all possible subsets of the SERS and SERRS flavors (Fig. 3c). A condition number $(\kappa )$ of a function indicates how much deviation in the input will be amplified to deviation in the output with a factor of $\kappa$.²³ In the case of spectral unmixing, the condition number represents the amplification factor in the spectral deconvolution error because of the fluorescence background and the overlap of Raman peaks. Thus, we call the combination that provides the lowest condition number as a well-conditioned subset of flavors and will give the lowest unmixing error, and the combination with the highest condition number as an ill-conditioned subset of flavors, which will show the highest unmixing error (see ESI Table S8^†).

Calculating the lowest and the highest condition numbers, we compared SERS and SERRS NP flavors with respect to the number of required plexity. Logically, the condition numbers increase with the increasing plexity (Fig. 3c). Notably, for the NIR dyes, the condition numbers show a much steeper increase with higher plexities compared to non-resonant Raman reporters, clearly indicating more potential complications for multiplexing with SERRS NPs. For instance, a given combination of flavors with the condition number of 10 is expected to result in a 10-fold increase in the spectral unmixing error upon multiplexing. The calculated condition numbers indicate that one observes a 10-fold increase in the spectral unmixing error for performing either 3-plex with SERRS NPs or 22-plex with SERS NPs (see ESI Fig. S16^†). It should be noted that utilizing Raman-labeled NPs with higher unmixing errors can still be acceptable depending on the particular imaging application and accuracy requirements. For instance, some investigators may only need to interrogate a single feature with greater sensitivity, in which the overall Raman signal intensity may be a priority. More advanced spectral unmixing algorithms may also be used to reduce the error caused by a high background and significant spectral overlap.²²

Sensitivity of SERRS and SERS nanoparticles for Raman imaging

In order to assess sensitivity for Raman imaging, we used circular pieces of coffee filter paper as a substrate for applying the NPs to avoid “coffee ring” effects (Fig. 4, see ESI Fig. S17^†). In the dilution series, the highest concentration was 150 pM, which was previously utilized for successful cancerous tissue staining,⁵³ and, with 2× dilution factor, we decreased the NP concentration down to 4.6 fM. For spectral unmixing, we used the reference spectra of the pure components: 37.5 pM SERS and SERRS NPs. The resulting Raman imaging channels for each flavor are shown in Fig. 4. The limits of detection (LODs) were calculated from the extrapolation of the linear plots of weight values, obtained after spectral unmixing with NNLS, to the points, which intersect (3.3×) the standard deviation of weight values from a blank paper substrate. Choosing a LOD with a signal-to-background ratio equal to 3.3 allows one to reliably detect each NP flavor in the analyzed sample with over 95% confidence. Spectral standard deviations for each NP concentration indicate the level of fluctuations present in the spectra (shown as the purple shaded regions in Fig. 4a and b).

Fig. 4.

Sensitivity comparison of SERRS versus SERS NPs in Raman imaging experiments. The average Raman spectra with standard deviation (shaded area, n = 500) for (a) 37.5 pM SERRS NPs (labeled with DTTC, IR-775, IR-780, IR-792, or IR-797) and (b) 37.5 pM SERS NPs (labeled with BPE, BMMBP, PODT, PTT, or 44DP). Raman imaging channels for the dilution series of SERRS NPs (a) and SERS NPs (b) starting from 150 pM with 2× dilution steps. Scale bars represent 5 mm. Scheme of nanoparticle concentrations applied onto the coffee filter paper substrate (c). Limits of detection for SERS and SERRS NPs (d).

We observed that the overall peak intensities of SERRS NPs were often higher than those of SERS NPs due to their added resonance effect. However, it is important to note that this alone does not indicate an ability of NPs to be better detected using Raman imaging. Although the overall signal intensity is generally higher for SERRS, the higher intrinsic fluorescence increases the overall background and thus decreases the ability for NPs to be detected at lower concentrations, especially in their complex mixtures. Historically, in biomedical imaging, the ability to detect a signal above the background is what gives clinicians image contrast and confidence that the NPs are in fact present and detectable. Notably, we observed that some SERS NPs perform on the similar level with the SERRS NPs with regard to their LODs, especially when it comes to spectral unmixing results, that are more practically relevant than the absolute signal intensities when it comes to overall detection sensitivity. Lower spectral background levels as seen in the SERS NP batches (Fig. 4) provide better reproducibility of spectral unmixing and allow smaller concentrations to be accurately measured. In single-plex mixtures, we observed comparable LODs for SERS and SERRS NP types (Fig. 4c) based on the NNLS-unmixed weight values. However, all SERRS NPs demonstrated much broader spectral standard deviations compared to SERS NPs (Fig. 4a and b).

Importantly, both SERS and SERRS NPs have been useful for improving the depth of penetration imaging using spatially offset Raman spectroscopy (SORS).^39,54,55 The surface enhancement that these NPs offer have made it possible to image a maximal depth of 7.1 cm (ref. 56) and up to 14 cm in biological tissues⁵⁷ and detect biomolecules through the skull.⁵⁸ The surface-enhanced spatially offset Raman spectroscopy (SESORS) approach has great potential for overcoming the limited depth of penetration properties that are characteristic of most optical imaging techniques. Imaging at increased depths offers researchers a whole new opportunity to explore new clinical applications for Raman imaging with NPs and could accelerate its clinical translation.⁵⁹

Mixing and spectral unmixing of SERRS and SERS nanoparticles

Next, to assess quantitative multiplexing, we imaged mixtures of SERS and SERRS NPs. We used Raman spectra of each SERS solution and SERRS NP solution as the reference spectra for the spectral unmixing with NNLS. We evaluated the accuracy of spectral unmixing using the five known NP flavors tested above for the respective SERS and SERRS NP batches. The molar ratios of the NPs within the respective mixtures were varied concurrently as 0 : 1 : 2 : 3 : 4 with the fifth flavor being absent, representing a negative control. As the mixture spectra in each imaging pixel can be fitted to the linear combination of the five NP reference spectra, we could calculate the weight values and, thus, the ratio of the reference NPs in the mixture from the fitting. Importantly, all SERS NP mixed ratios linearly correlated with the estimated ratios. However, the SERRS NPs had a higher spectral deconvolution error of 28.1(±10.3)% with R² = 0.902 as opposed to 4.8(±1.3)% with R² = 0.999 for SERS NPs (Fig. 5a and b), which were defined as the average difference between the mixed ratio and the estimated ratio of the NPs.

Fig. 5.

Spectral unmixing accuracy. The estimated ratios of the five-color SERS (a) and SERRS NPs (b) from the spectral unmixing, which were normalized to the average concentration of the BPE-labeled and DTTC NPs, respectively, and the average Raman spectra of the nanoparticle mixtures with standard deviation (shaded area, n = 500). The SERS NPs were mixed with a 0 : 1: 2 : 3 : 4 ratio of PODT, BPE, 44DP, PTT, and BMMBP-labeled NPs, respectively. The SERRS NPs were mixed with a 0 : 1 : 2 : 3 : 4 ratio of IR-780, IR-792, DTTC, IR-775, and IR797-labeled NPs, respectively. (c) Raman imaging channels for the dilution series of DTTC-labeled NPs in the presence of 37.5 pM IR-775, IR-792, IR-780, and IR-797-labeled NPs, and BPE-labeled NPs in the presence of 37.5 pM BMMBP, PODT, PTT, and 44DP-labeled NPs starting from 150 pM with 2× dilution steps. Scale bars represent 5 mm. The average Raman spectra of the nanoparticle mixtures with standard deviation (shaded area, n = 500) for BPE- and DTTC-labeled NPs, both at 37.5 pM concentration. (c) Unmixed weight values for BPE-labeled NPs in the presence of 37.5 pM BMMBP, PODT, PTT, and 44DP-labeled NPs and DTTC-labeled NPs in the presence of 37.5 pM IR-775, IR-792, IR-780, and IR-797-labeled NPs, along with the LOD concentration for each NP flavor in 1-plex, which can be used as a ground rule for assessing the impact of multiplexing.

We also evaluated the sensitivity of the five-plex mixtures for the detection of a low-abundance NP, the concentrations of four NP flavors were kept constant, and the fifth nanoparticle concentration was serially diluted (Fig. 5c). Notably, we were able to detect the change in the concentrations of each studied non-resonant SERS NP type within the mixtures and their unmixed weight values linearly correlated with the input concentration (Pearson’s correlation coefficient r ≥ 0.995 in all cases). However, with the resonant SERRS NPs, we observed significant positive errors throughout the diluted mixture series that did not linearly correlate with the input concentration shown in Fig. 5c. The resonant reporter DTTC was still falsely detectable beyond the 1-plex LOD in the mixed samples (Fig. 5c). This suggests that the large spectral overlap among the resonant SERRS reporters is reasonable for misleading the unmixing algorithm to report the presence of DTTC-labeled SERRS NPs when it is, in fact, not present. Contrarily, the non-resonant reporter BPE-labeled NPs demonstrated linear results in 5-plex mixtures consistent with the model 1-plex solution. The unmixing algorithm had no problem identifying the presence and correct quantity of the BPE-labeled NPs in the 5-plex mixture. This supports the idea that non-resonant reporters have unique spectral signatures with little spectral overlap. In summary, we observed that the resonant SERRS NPs with heptamethine cyanine dyes possess higher fluorescence backgrounds and more crowded Raman signatures that limit their ability to simultaneously maintain high sensitivity and accuracy when attempting to unmix multiple spectral signatures.

Sensitive multiplex imaging methods are becoming an essential tool in biomedical research as they can offer quantitative information about numerous cell types and proteins while providing spatially intact data about their relationships. To help improve personalized medicine, it is of vital importance to understand a patient’s unique molecular expression profile. Providing researchers and clinicians with detailed molecular maps of a patient’s tumor can offer important insights about the tumor microenvironment and the heterogeneity that exists both within the tumor and across patients. Utilizing a quantitative highly multiplexed imaging platform could aid in prognosis, predicting treatment response, and improving the overall patient outcome. SERS-based Raman imaging has the potential to offer quantitative, non-destructive, high-plexed spatial information in a single imaging acquisition and should be further developed as a multiplexed imaging platform.

Conclusions

In this work, we demonstrated the performance of non-resonant SERS and resonant SERRS NPs, which have been actively utilized in biomedical applications, for multiplexed Raman imaging. We assessed the multiplexing capabilities of the imaging nano-probes using the same acquisition settings and identical gold nano-cores of non-resonant Raman reporters: BPE, 44DP, PTT, PODT, and BMMBP and resonant with 785 nm laser wavelength conventional heptamethine cyanine NIR dyes: IR-775, DTTC, IR-780, IR-792, and IR-797. Although the SERRS NPs demonstrated higher absolute Raman intensities, the NIR-resonant Raman reporters exhibited a higher background when excited at 785 nm and more crowded Raman spectra due to their bulkier molecular structures. This significantly decreased the overall signal-to-background ratios associated with these SERRS NPs and thus their ability to be accurately detected in their complex mixtures, as demonstrated in this study. These complex SERRS NP spectra also demonstrated more spectral overlapping than the SERS spectra, resulting in a lower accuracy of biomarker sensing with increased plexity. Our study revealed that the increased spectral overlap among resonant Raman reporters has the potential to mislead the unmixing algorithms to falsely detect the presence of a NP reporter probe when it is, in fact, not present. This could ultimately lead to higher false-positive results in a patient sample after the spectral deconvolution. Thus, we showcased that these factors are important to consider when developing new nano-based Raman imaging strategies for various multiplexing biomedical applications. In conclusion, the non-resonant SERS-based Raman imaging approach studied here provided superior multiplexing capabilities with the potential to offer researchers and clinicians crucial quantitative molecular insights to improve patient care and outcome.