Background:
The rapid expansion of targeted therapies and cancer immunotherapies demands technologies capable of comprehensively characterizing the molecular composition of patient tumors and their tumor microenvironment (TME). Because the TME contains diverse immune and stromal cell populations that influence therapeutic response, highly multiplexed imaging with subcellular spatial resolution has become increasingly important for precision medicine. Surface-enhanced Raman spectroscopy (SERS) imaging offers a unique solution by combining nanoparticle(NP)-based contrast agents with exceptionally narrow spectral fingerprints, enabling multiplexing far beyond conventional fluorescence methods while maintaining high molecular specificity and sensitivity.1

Hypothesis/Objective:
In this work, we present the first comprehensive library of 65 spectrally distinct SERS nanoparticle probes, designed through density functional theory (DFT)-guided reporter selection and quantitative spectral orthogonality analysis. The library enables quantitative spectral unmixing of up to 65 molecular targets within a single imaging acquisition, providing a versatile platform for highly multiplexed molecular profiling. We further demonstrate multiplexed immune profiling of human formalin-fixed paraffin-embedded (FFPE) tissue sections and introduce a rational framework for selecting optimal probe combinations for future personalized imaging applications and surgical guidance. We hypothesized that rational design of Raman reporter molecules guided by DFT calculations and quantitative spectral orthogonality analysis would enable the creation of a substantially expanded library of spectrally distinct SERS nanoparticle probes with minimal spectral crosstalk. We further hypothesized that such a library would support accurate quantitative spectral unmixing, allowing simultaneous detection of dozens of molecular targets within a single Raman imaging acquisition. Accordingly, our objectives were to (i) synthesize and characterize a comprehensive library of 65 SERS nanoparticle flavors with unique spectral fingerprints, (ii) validate their multiplexing performance through quantitative spectral unmixing of all 65 probes in a single imaging experiment, (iii) demonstrate their utility for multiplexed immune spatial profiling of human FFPE tissue sections by targeting multiple immune biomarkers and comparing Raman imaging with gold-standard immunohistochemistry, (iv) evaluate spectral unmixing performance under complex topographical conditions using a multiplexed phantom, and (v) establish a hierarchical clustering framework for rational selection of spectrally orthogonal probe combinations for future high-dimensional imaging panels.

Methods:
Fabrication of SERS NPs: De-novo synthesized SERS NPs were prepared by: (1) creation of 60-nm spheric AuNPs; (2) labeling of AuNPs with unique Raman reporter; (3) silica coating; (4) thiolation; (5) conjugation with biotargeting species; (6) capping remaining thiols; (7) purification. SERS spectra simulation: Density functional theory (DFT) facilitated selection of Raman reporters which met the key criterion of high spectral orthogonality.2,3 Multiplexed imaging assessment: A computer-controlled x-y translation stage was used to raster scan various SERS NP mixtures and topography phantom. Human tissue imaging: Immune profiling was performed on 4-μm thick healthy human tonsil tissue sections (N = 10). Archived FFPE tissue sections were prepared using heat induced epitope retrieval (HIER) technique. Human tonsil tissue sections were stained with SERS NPs for CD3, CD20 and isotype control IgG. Each specimen was stained with experimentally established optimal concentration of SERS NPs: 250 μL of 50 pM NP staining solution for 20 min. Multiplexing robustness was further validated using a three-dimensional topography phantom containing spatially overlapping mixtures of SERS NPs.

Results:
We developed an expanded library of 65 spectrally unique SERS nanoparticle probes (Fig. 1A) with minimal spectral overlap, enabling accurate spectral deconvolution during a single Raman imaging acquisition. Simultaneous imaging of all 65 nanoparticle flavors demonstrated excellent spectral separation using NNLS spectral unmixing with negligible crosstalk (Fig. 1B). DFT-guided reporter selection produced highly orthogonal Raman signatures while preserving strong signal intensity. Multiplexed Raman imaging of human FFPE tonsil tissue accurately identified major immune populations, including T cells (CD3, CD4, CD8), B cells (CD20), macrophages (CD68), and proliferating cells (Ki-67), with spatial distributions closely matching corresponding gold-standard IHC staining (Fig. 1C). Topography phantom experiments further demonstrated accurate quantitative identification of multiple SERS nanoparticle mixtures occupying the same spatial locations after spectral unmixing, validating multiplexed imaging under complex imaging conditions (Fig. 1D). Finally, hierarchical clustering organized all 65 nanoparticle probes into four spectral families, providing a systematic strategy for selecting spectrally orthogonal probe combinations for future high-dimensional imaging panels (Fig. 1E).

Conclusion:
We present the first validated library of 65 spectrally distinct SERS nanoparticle probes for highly multiplexed Raman imaging. This expanded Raman probe library enables simultaneous molecular profiling with unprecedented multiplexing capacity while maintaining high targeting specificity and quantitative spectral unmixing. By combining rational nanoparticle design, computational reporter selection, multiplexed immune profiling, and topographical validation, this platform establishes a scalable framework for next-generation spatial biology and personalized medicine.

Image/Figure:
Click to View

Image/Figure Caption:
Figure 1. Unraveling the true multiplexing capabilities of Raman imaging in conjunction with SERS NPs to enable highly specific molecular expression and spatial profiling. (A) Architecture of SERS NPs consisting of a 60-nm gold core labeled with one of 65 Raman reporters and coated with a silica shell. Each SERS NP flavor emits the desirable unique spectral signature with narrow Raman bands, suitable for high degree of multiplexing. (B) Overlay of 65 SERS imaging channels corresponding to the presence of each SERS NP flavor, imaged in a single Raman mapping acquisition. The multiplexed images were generated through the non-negative least squares (NNLS) spectral unmixing of Raman spectra in each pixel. Arbitrary colors have been assigned to each SERS NP batch imaged. The NNLS spectral unmixing method was able to successfully separate all 65 SERS NPs into their respective imaging channels with minimal crosstalk. Scale bar represents 10 mm. (C) IHC stains of 4-μm thick human tonsil FFPE tissue section for CD3, CD20, CD4, CD68, Ki-67, and CD8, which represent spatial distribution of T and B cells. Scale bars for optical and SERS images represent 50 μm. Note: IHC stains were performed on six different tissue sections, while the Raman imaging channels were obtained from a single patient’s tissue specimen. Note: Raman images correlate well with the gold standard IHC stain, which demonstrated majority of T cells located outside the follicular zones and B cells observed primarily inside of follicles. (D) Multiplexed topography validation with SERS NPs that can be mixed within the same pixels and still be correctly recognized and interpreted after quantitative spectral unmixing. Topography phantom coated with mixtures of SERS NPs was used to demonstrate proof of concept for breast cancer resection. (E) Hierarchical clustering dendrogram of the library of 65 SERS NPs to demonstrate the selection of SERS NP flavor selection for multiplexed imaging applications. The clustering was performed on baseline-subtracted SERS spectra normalized to 2-unit variance. For this grid, hierarchical clustering has arranged the SERS flavors into three vibronic SERS Families labeled I, II, III, and IV. Spectral intensities along the horizontal axis have been arranged to highlight the spectral features that are shared among that vibronic family and indicate how that family differs from the others.

Author
Olga Eremina
University of Southern California