High Precision Imaging of Microscopic Spread of Glioblastoma with a Targeted Ultrasensitive SERRS Molecular Imaging Probe.

The dismal prognosis of patients with malignant brain tumors such as glioblastoma multiforme (GBM) is attributed mostly to their diffuse growth pattern and early microscopic tumor spread to distant regions of the brain. Because the microscopic tumor foci cannot be visualized with current imaging modalities, it remains impossible to direct treatments optimally. Here we explored the ability of integrin-targeted surface-enhanced resonance Raman spectroscopy (SERRS) nanoparticles to depict the true tumor extent in a GBM mouse model that closely mimics the pathology in humans. The recently developed SERRS-nanoparticles have a sensitivity of detection in the femtomolar range. An RGD-peptide-conjugated version for integrin-targeting (RGD-SERRS) was compared directly to its non-targeted RAD-SERRS control in the same mice via Raman multiplexing. Pre-blocking with RGD peptide before injection of RGD-SERRS nanoparticles was used to verify the specificity of integrin-targeting. In contrast to the current belief that the enhanced permeability and retention (EPR) effect results in a baseline uptake of nanoparticles regardless of their surface chemistry, integrin-targeting was shown to be highly specific, with markedly lower accumulation after pre-blocking. While the non-targeted SERRS particles enabled delineation of the main tumor, the RGD-SERRS nanoparticles afforded a major improvement in visualization of the true extent and the diffuse margins of the main tumor. This included the detection of unexpected tumor areas distant to the main tumor, tracks of migrating cells of 2-3 cells in diameter, and even isolated distant tumor cell clusters of less than 5 cells. This Raman spectroscopy-based nanoparticle-imaging technology holds promise to allow high precision visualization of the true extent of malignant brain tumors.

Introduction

High grade gliomas, in particular GBMs (= grade IV astrocytoma), are still considered universally fatal 1, with an average mean survival of 12-15 months after diagnosis 2, 3. Surgery remains a mainstay in the treatment of GBMs. Tumor resections have shown to improve patient survival 4-6, and the extent of resection of both primary and recurrent GBMs directly correlates with the length of survival 5, 7, 8. Studies examining the patterns of GBM recurrence have shown 80-90% of recurrences to occur near the original resection margins, suggesting residual tumor cells due to incomplete resection as the cause 9, 10. In the other cases, recurrences occur at distant sites in the brain because of microscopic locoregional tumor spread. Extensive attempts with multiple different modalities have been made to determine the true extent of tumor spread. Several studies have suggested that magnetic resonance imaging (MRI) using FLAIR or diffusion-weighted sequences could enable imaging of microscopic tumor infiltrations 11, 12. However, the tumor borders as assessed by pre-operative MRI are often incongruent with those during surgery because of brain shift during the procedure 13, 14. Intra-operative MRI is limited by several factors such as the inability to achieve microscopic resolution, inaccuracies due to false-positive contrast enhancement 15, or the necessity for frequent administrations of gadolinium contrast agents 16. In addition, small molecule Gd-agents are known to disperse from the initial area of tumor enhancement into the peritumoral zone of edema over time, therefore causing inaccuracies in estimating the true tumor borders 17. Ultrasound has the advantage of providing real-time imaging and has been reported to improve patient outcome 18, 19 but cannot visualize microscopic cancer. Several optical methods have been explored, either based on intrinsic tissue properties 20-22 or by using exogenous contrast agents 13, 23-26. However, these optical techniques are often limited by small fields of view (microscopy), decreased specificity due to autofluorescence, or rapid photobleaching 27, 28.

Given these difficulties, neurosurgeons currently remain hindered in achieving complete resections that, in principle, could be possible in non-eloquent areas if a high precision detection method was available. Thus, there is an unmet need for an imaging method that can reveal the true microscopic extent of tumor spread.

We recently developed ultrasensitive nanoparticle imaging probes that rely on the effect of 'surface-enhanced resonance Raman spectroscopy' (SERRS) 29-31. In gliomas, the most common type of brain tumors, a subset of integrin receptors, such as αvβ3, is highly expressed on the tumor vasculature, angiogenic endothelial cells, and tumor cells 26, 32. Overexpression of integrin αvβ3 was demonstrated in the state-of-the-art RCAS-PDGF/N-tva transgenic mouse model of GBM 26, which presents with all the histopathological and imaging hallmarks of human high-grade tumors, such as infiltrating tumor margins, microvascular proliferation and pseudopalisading necrosis 33. In this study we investigated whether integrin-targeting of SERRS nanoparticles enables accurate delineation of the infiltrating tumor margins and subtle microscopic metastatic lesions of glioblastoma in the recapitulative RCAS-PDGF/N-tva mouse model.

Materials and Methods

Materials

Unless otherwise noted, all chemicals were obtained from Sigma Aldrich (St Louis, MO). The cyclic RGDyK and RADyK peptides were purchased from AnaSpec (Fremont, CA).

SERRS nanoparticle synthesis

SERRS nanoparticles were synthesized as previously described 29, 30. In brief, 60-nm gold nanospheres were synthesized by rapid addition of 7.5 mL 1% (w/v) sodium citrate tribasic dihydrate to a 1 L boiling aqueous gold chloride (HAuCl4; 0.25 mM) solution. Gold nanostars were synthesized by rapid addition of 10 ml 20 mM HAuCl4 to 1.0 L ice-cold 40 mM ascorbic acid. The as-synthesized gold nanoparticles were collected by centrifugation and dialyzed (MWCO 3.5 kDa). The dialyzed gold nanoparticle dispersion (360 μL; 2.0 nM) was added to 3.5 mL absolute ethanol containing 53 μL 28% (v/v) ammonium hydroxide, 106 μL 99.999% tetraethyl orthosilicate (TEOS), and 7 μL of 25 mM resonant Raman reporter in N,N-dimethylformamide, which constituted either IR792 perchlorate (IR792) or IR780 perchlorate (IR780) for the nanospheres and nanostars, respectively. After 25 minutes, the as-synthesized SERRS nanoparticles were washed with ethanol, and redispersed in 1.0 mL ethanol containing 5% (v/v) (3-mercaptopropyl)trimethoxysilane (MPTMS) and heated at 70 oC for 2 hours. The sulfhydryl-modified SERRS nanoparticles were collected by centrifugation and washed first with ethanol and then water.

Integrin-targeted SERRS nanoparticle synthesis

The heterobifunctional linker, poly(ethylene glycol) (N-hydroxy-succinimide 5-pentanoate) ether N'-(3-maleimidopropionyl)aminoethane (NHS-PEG4k-Mal) was was reacted for 30 minutes with cRGDyK or cRADyK in a 1:1 molar ratio after which the sulfhydryl-modified SERRS nanoparticles 'IR792' or sulfhydryl-modified SERRS nanoparticles 'IR780' were added to cRGDyK-PEG4k-Mal or cRADyK-PEG4k-Mal in a 10 mM 2-(N-morpholino)ethanesulfonic acid (MES buffer pH 7.1), respectively. After 2 hours, 0.5 μL 50% (v/v) hydroxylamine was added to quench any remaining unreacted N-hydroxy-succinimide functional groups. The integrin-targeted cyclic RGDyK-conjugated SERRS (RGD-SERRS) nanoparticles and non-targeted control cyclic RADyK-conjugated SERRS (RAD-SERRS) nanoparticles were washed with water and redispersed in 0.22-μm filter sterilized 10 mM MES buffer (pH 7.3) at a final concentration of 3.5 nM.

Integrin-targeted SERRS nanoparticle characterization

For characterization by electron microscopy, samples of the RGD-SERRS and RAD-SERRS nanoparticles were deposited on carbon film-coated copper grids (300 Mesh, Electron Microscopy Sciences) and air-dried. Images were typically acquired at magnification factors ranging from 25,000× to 500,000× using a JEOL 1200EX (JEOL USA, Inc.) transmission electron microscope operating at 80 kV. The concentration and size distribution of the SERRS nanoparticles were determined by nanoparticle tracking analysis (NTA; NanoSight NS500; Malvern Instruments Inc.; Westborough, MA). SERRS signal intensity of 1.0 nM SERRS nanoparticle dispersion was measured on a Renishaw inVia Raman microscopy system (Renishaw, Hoffman Estates, IL) at 50-μW laser power, 785 nm laser wavelength, 1.0-s acquisition time, and a 5× objective.

Animals

All animal experiments have been approved by the Institutional Animal Care and Use Committees of Memorial Sloan Kettering Cancer Center (#06-07-011).

The somatic gene transfer system RCAS/tv-a was used to generate the glioblastoma-bearing mice 34-37. Briefly, the 4-8 week-old Nestin-tv-a/Ink4a-arf-/-/Ptenfl/fl mice were stereotactically injected with 4×104 DF1 cells transfected with RCAS-Pdgfb and RCAS-Cre (1:1 mixture, 1 µL) into the brain, coordinates bregma 1.7 mm (anterior), 0.5 mm (right), and depth 2.5 mm from the dural surface. The genetic aberrations including overexpression of oncogene Pdgfb and loss of tumor suppressor genes (Ink4a-arf and Pten) led to near complete penetrance of glioblastoma within 2 months. Tumor incidence and size were determined by weekly MRI scans (Bruker Biospin Corp., Billerica, MA) starting four weeks after DF-1 cell injection 38.

Raman imaging and signal quantification

All Raman scans were performed on an InVia Raman microscope (Renishaw) equipped with a piezo-controlled stage for micron-resolved spatial mapping, a 300-mW 785-nm diode laser and a 1-inch CCD detector with a spectral resolution of 1.07 cm-1. The SERRS spectra were collected through a 5× or 20× objective (Leica). Laser output at the microscope objective was measured with a handheld laser power meter (Edmund Optics, Inc.) and determined to be 100 mW at the microscope objective. Typically, in vivo and ex vivo Raman scans were performed at 10-100 mW laser power, which is below the safety requirement for clinical application 39, 1.5 s acquisition time, using the StreamLineTM high-speed acquisition mode. All Raman images were acquired and analyzed under the same conditions, including the same laser power, Raman integration times (per pixel), focal plane (same objective lens), and a threshold setting of 0.1. Subsequently, Raman images were analyzed using Metamorph Microscopy Automation and Image Analysis software (Molecular Devices, Sunnyvale, CA) image processing software. Regions of interests were defined by histology and photographs of the brain tissue slices. SERRS intensities were then measured within all regions of the tissue, along with tumor area measurements to allow for normalization of intensity values between tissue samples. Intensity of SERRS signal in areas outside of the tissue regions were used to set thresholds, which were set at 0.1 such that intensity levels corresponding to areas surrounding the tissue were considered to be noise. The mean signal and standard deviations of different experimental groups were calculated. A Student's t-test (two-tailed, unpaired) was performed and the level of significance was set at p < 0.05.

Multiplexed Raman imaging in GBM-bearing mice

Prior to injection, 75 μL of 3.5 nM RGD-SERRS (IR792) nanoparticles and 75 μL of 3.5 nM RAD-SERRS (IR780) nanoparticles were mixed. This mixture was then injected intravenously via tail vein. After 18-24 hours, the GBM-bearing animals were sacrificed by CO2 asphyxiation and brains were harvested, fixed in 4% paraformaldehyde to prevent autolytic decomposition of the brain tissue, and kept at 4 °C overnight. Raman imaging was performed on the fixed brain and/or on paraffin-embedded coronal brain sections instead of in live mice in order to achieve the highest possible precision in correlating the Raman signal with the histological information. The Raman spectra of the RGD-SERRS and non-targeted RAD-SERRS nanoparticles (i.e. IR792 and IR780 Raman spectrum, respectively) were unmixed by a direct classical least squares (DCLS) algorithm that is embedded in the Wire 3.4 Raman imaging software (Renishaw). After deconvolution of the two Raman flavors, Raman images were generated using identical parameters for both RGD- and RAD-SERRS nanoparticles.

Histology

Intact GBM-bearing brains were sliced coronally (1 mm slice thickness) and embedded in paraffin. 5 μm-thick continuous sections were cut for hematoxylin and eosin (H&E) staining and immunohistochemistry (IHC) staining, followed by high-resolution Raman imaging on paraffin blocks. IHC staining was performed on the Discovery XT biomarker platform (Ventana, Tucson, AZ) as previously described 30. Antibodies for OLIG2 (1:300, AB9610, Millipore, Temecula, CA), polyethylene glycol (1:100, ab51257, Abcam, Cambridge, MA), ITGB3 (1:100, 13166, Cell Signaling, Danvers, MA), ITGAV (1:2000, ab76609, Abcam), HA-tag (1:200, 11867423001, Roche, San Francisco, CA), IBA1 (1:600, 019-19741, Wako, Richmond, VA), NOS3 (1:200, 610296, BD Biosciences, Franklin Lakes, NJ) and ACTA2 (1:350, M0851, DAKO, Carpinteria, CA) were used as the primary antibodies. The slides were digitally scanned with Pannoramic Flash (3DHistech, Hungary) and relevant tissue areas were exported into tiff format. Quantification of Olig2 was performed using ImageJ/FIJI (NIH). Color deconvolution algorithm was used to determine the area of positive signal, which was normalized to tissue area.

Results

Generation and characterization of integrin-targeted and non-targeted control SERRS nanoparticles

Integrin-targeted RGD-SERRS nanoparticles were synthesized by conjugating cyclic RGDyK (RGD) via the ε-amino group of the lysine to the sulfhydryl functional groups on silica-coated SERRS nanospheres via a heterobifunctional polyethylene glycol linker (4000 Da; PEG4k) (flavor 1; IR792). In a similar manner, non-targeted control RAD-SERRS nanoparticles were generated by conjugating cyclic RADyK (RAD) to silica-coated SERRS nanostars (flavor 2; IR780 perchlorate) (Figure ). Both RGD-SERRS and RAD-SERRS nanoparticles were characterized by Raman spectroscopy, nanoparticle tracking analysis and transmission electron microscopy (Figure ). Equimolar amounts of the RGD-SERRS and RAD-SERRS nanoparticles, which have similar limits of detection (i.e. 10-15 M, 29, 30) generated similar SERRS signal intensity (~20,000 cps) for the most intense peak of each dye , i.e. 950 cm-1 for the RAD-SERRS (green) and 1200 cm-1 for the RGD-SERRS (red). In addition, the distinct Raman fingerprints of the RAD- and RGD-SERRS nanoparticles allowed multiplexed detection (Figure ). Nanoparticle tracking analysis demonstrated a similar hydrodynamic radius of ~140 nm for both SERRS nanoparticle versions. RGD-SERRS and RAD-SERRS nanoparticle can also be distinguished by TEM due to the difference in their gold core geometry (Figure ).

Specificity of integrin-targeted SERRS nanoparticle-guided glioblastoma detection

To determine the targeting specificity of the integrin-targeted RGD-SERRS nanoparticles, an equimolar dose (13 fmol/g) of RGD-SERRS and non-targeted RAD-SERRS nanoparticles was co-injected via tail vein in RCAS-Pdgfb/N-tva GBM-bearing mice (n=7) 40. In order to preserve the brain tissue architecture and prevent autolytic decomposition, multiplexed Raman imaging was performed on paraformaldehyde-fixed coronal brain sections. Raman images of RGD-SERRS and RAD-SERRS from a representative GBM-bearing brain slice were shown in Figure . SERRS intensities of RGD-SERRS and RAD-SERRS from the healthy brains and tumors (normalized by the area of interest) were quantified, respectively. Significantly elevated RGD-SERRS signal intensity was noted in tumors (24.3±11.8 counts/pixel; n=7, p<0.01). Interestingly, RGD-SERRS nanoparticles, but not RAD-SERRS, were detected in an area separate from the bulk tumor in the contralateral hemisphere (white arrowheads in Figure ). Strong positive expression of integrin β3 (ITGB3) but weak expression of OLIG2 were observed in this separate tumor focus, indicating that overexpressed integrins, including integrin β3, were specifically targeted by the RGD-SERRS (black arrowheads in Figure ). The ratio of RGD-SERRS to RAD-SERRS in tumors was 188.8±58.4. Raman images of RGD-SERRS and RAD-SERRS from injected GBM-bearing brain slices were correlated with OLIG2-expression, a protein that is highly overexpressed in the tumor cells of the applied model (n=7, Figure ; 41). A strong positive Pearson correlation was found for RGD-SERRS and OLIG2 immunostaining (R=0.84, p<0.05) in the analyzed sections. Together, these results suggest that the RGD-SERRS nanoparticles preferentially accumulated due to the intratumoral overexpression of αvβ3 integrins.

To further confirm integrin-targeting of the RGD-SERRS nanoparticles, an in vivo blocking study was performed in which cRGDyK peptide (1.0 nmol/g) was injected into a separate cohort of GBM-bearing mice (n=5) 30 minutes prior to the RGD-SERRS nanoparticle injection (26 fmol/g). As shown in Figure , the mean value of RGD-SERRS signal intensity of tumor tissues in the blocking group was 3.6±3.3×10-3 counts/pixel, which was significantly lower than that in the non-blocking group (24.3±11.8 counts/pixel) (p<0.01) when identical imaging and data analysis conditions were applied.

Integrin-targeted SERRS nanoparticles enable detection of microscopic tumor invasion beyond the borders of the main tumor

GBMs are aggressive and locally invasive tumors. GBM cells preferentially invade along the preexisting tissue elements via surface, perineuronal, perivascular, and intrafascicular growth 42. In a representative GBM-bearing mouse that was injected intravenously with RGD-SERRS nanoparticles (26 fmol/g), strong SERRS signals were detected within the bulk tumor in the upper right cerebrum, but not in normal contralateral brain tissue, with a fine delineation of the actual tumor margin confirmed by H&E staining (Figure ). RGD-SERRS signal was also detected along the medial septum (arrowhead 1), which was infiltrated by GBM cells (OLIG2-positive) and other ITGB3-overexpressing cells. Anti-polyethylene glycol (PEG) immunostaining, which stains for the PEG-linker on the RGD-SERRS nanoparticle surface, confirmed the presence of RGD-SERRS nanoparticles in areas with high integrin β3 (ITGB3) expression (Figure ). Furthermore, in the “infiltration front” of GBM cells penetrating into normal brain, presence of RGD-SERRS nanoparticles (PEG-positive) was demonstrated adjacent to ITGB3- and OLIG2-positive cells (arrowhead in lower panels, Figure ). This ability to delineate the true tumor extent (diffuse borders, microscopic tumor extensions) was observed in all five mice.

Integrin-targeted SERRS nanoparticles enable identification of very small locoregional metastases with microscopic precision

As exemplified in a different animal (Figure ), RGD-SERRS nanoparticle-based Raman imaging enabled accurate delineation of the bulk tumor located in the right hemisphere (Figure ). However, more importantly, a microscopic SERRS-positive focus (Figure , upper panel, left) located at a distance of ~350 μm from the bulk tumor was identified by Raman imaging. The spectral analysis of this pixel proved the presence of the unique Raman fingerprint of RGD-SERRS nanoparticles (Figure , upper right). Corresponding adjacent sections from this tissue block were analyzed with immunohistochemistry staining of protein markers for tumor cells (α-HA tag), microglia cells (α-IBA1), endothelial cells (α-NOS3), and pericytes (α-ACTA2), which indeed identified this microscopic locoregional focus (~15 μm) as a small cluster of cells, including a GBM cell, a microglial cell, an endothelial cell and a pericyte (Figure , lower panels). This corresponds well with observations that GBM cells tend to invade individually or in small groups in between the dense network of neuronal and glial cell processes, and thus give rise to multifocal locoregional gliomas 43.

Collectively, integrin-targeted RGD-SERRS nanoparticle-based Raman imaging enabled accurate delineation of the actual extent of the tumor and the detection of locoregional satellite tumor foci, microscopic extensions invading across the entire brain, and even isolated GBM cells on their migratory paths.

Discussion

In this work, we used a targeted version of our recently developed SERRS-nanoprobes 30, which have femtomolar sensitivity and enable high precision imaging of a universally fatal brain tumor type. The SERRS-nanoprobes are orders-of-magnitude brighter than previous non-resonant SERS probes 13. To our knowledge, this is the first report of a targeted SERRS nanoprobe enabling brain tumor imaging in vivo. Our goal was to enable visualization of the true microscopic extent of GBMs. We used the genetically engineered RCAS-Pdgfb-driven/tv-a 34 murine model to test these probes in vivo, because this tumor model is considered to most closely recapitulate the human GBM development and biology. In order to prevent autolytic decomposition of the GBM-bearing mouse brains and to achieve the highest possible accuracy in correlating the SERRS data with the histological data, we performed ex vivo Raman imaging studies on paraformaldehyde-fixed specimens of GBM-bearing animals that were injected with the nanoprobes 18-24 hours prior to sacrificing and subsequent tissue processing.

We found that the SERRS images of the integrin-targeted SERRS nanoprobes correlate well with the actual extent of the microscopic spread of the GBM as indicated by immunohistochemistry correlation. This included the typical diffuse spread at the margins of the bulk tumor, and importantly, the microscopic extensions of migrating cells throughout the brain. Surprisingly, even isolated tumor cell clusters as small as a few cells could be imaged. The homing of the RGD-SERRS nanoparticles proved to be highly specific, as shown by the marked decrease of SERRS signal after pre-blocking with RGDyK peptide. This provides interesting insights into how to achieve high specificity of targeting with nanoparticles. To date, there is an ongoing debate on whether active targeting of nanoprobes by incorporating target-specific moieties such as antibodies, affibodies, peptides, or small molecules to the nanoprobe surfaces 44 is more effective than solely relying on the 'enhanced permeability and retention' (EPR) effect. In our previous work, using the U87MG and TS543 orthotopic models, we have observed that the EPR effect alone can be an effective mechanism for tumor-directed delivery of nanoparticles 30. However, in the genetically engineered tumor model employed in the current study, we observed that the addition of a targeting peptide increased the specific accumulation of the nanoprobe to the expressed target and significantly changed the intratumoral distribution compared to an otherwise identical non-targeted nanoparticle.

It is not intuitive that pre-blocking with cRGDyK peptide would decrease the accumulation of the RGD-targeted nanoprobe to the degree we have observed. One would expect that after blocking of its designated target, RGD-SERRS would still accumulate to a substantial degree in the tumor via the EPR effect. Interestingly, our results are consistent with several other reports that demonstrated that pre-blocking of integrin by free cRGDyK-peptide significantly decreased the accumulation of cRGDyK-targeted nanoparticles at the tumor site 45, 46. In a previous study we showed that integrin β3 was predominantly overexpressed on tumor endothelial cells, along with a lower and more heterogeneous expression on the tumor cells in the same glioblastoma mouse model 26. This might, at least in part, explain the major effect of RGD pre-blocking observed in this study. While cyclic RGD-peptides are known to inhibit avb3-integrin signaling, thereby displaying long-term anti-angiogenic effects and blood-vessel normalization, short-term effects of integrin inhibition on tumor blood vessel integrity and its effect on RGD-targeted nanoparticle accumulation are less well studied and remain to be explored.

While we have demonstrated that integrin-targeting significantly improved the accumulation of SERRS nanoparticles in GBM in the RCAS/tv-a model, this PDGF driven brain tumor is characteristic of approximately 25-30% of human GBMs. Therefore, further studies are needed to determine whether RGD-SERRS nanoparticles also enable accurate tumor delineation in other GBM-subtypes with distinct genetic backgrounds such as EGFR-amplifications or loss of NF1.

The detection of SE(R)RS nanoparticles by Raman spectroscopy-based imaging approaches has major advantages over other molecular contrast agent approaches. For example, as opposed to fluorescence imaging, our SERRS nanoprobes not only have superior sensitivity, they also are significantly more photostable than current fluorochromes 30. SERRS nanoparticles with different Raman reporters also emit complex spectra when excited with a single 785-nm excitation source 13, 30, 47. These complex, fingerprint-like spectra are unique to the SERRS-nanoparticles and are not produced by any endogenous biomolecule, thereby providing high signal specificity. This overcomes disadvantages of fluorescence imaging agents, where tissue autofluorescence can often be limiting by creating false-positive signals. The high tumor-to-background ratios produced by the RGD-SERRS nanoparticles prevents any unnecessary resection of healthy tissue and is therefore of particular importance in delicate surgical scenarios such as brain tumor resections 48. Whether a truly complete surgical removal of all tumor cells (and therefore cure) would be feasible with our RGD-SERRS nanoparticles in humans remains to be evaluated. In the case of GBMs, complete surgical resection is expected to be limited to cases without infiltration in eloquent brain areas. However, complete surgical resections should be feasible in other brain tumor types, such as e.g. low-grade gliomas, ependymomas or pilocytic astrocytomas 49-51, which are known to express integrin αvβ3 52, 53. In pediatric brain tumors, resection can be curative, but requires the highest precision in order to achieve removal of all tumor cells while sparing crucial cerebral structures. While the presented method could be especially impactful in such scenarios, in the current study we deliberately selected GBM for the purpose of testing our technique in order to evaluate how well it would perform in the most challenging scenario.

SERS nanoparticles of similar size, materials, and surface composition as those used here have already passed extensive cytotoxicity studies 54, therapeutic gold-silica nanoparticles are in advanced clinical trials 44, and wide-field Raman imaging devices are being developed 55, 56. We have also recently demonstrated that an already commercially available hand-held Raman scanner can be used to guide the resection of microscopic glioblastoma with high precision 38. Thus, this approach using RGD-conjugated SERRS nanoprobes should have a high potential for clinical translation.