Sentinel Lymph Node Biopsy Revisited: Ultrasound-Guided Photoacoustic Detection of Micrometastases Using Molecularly Targeted Plasmonic Nanosensors

Accurate detection of regional lymph node (LN) metastases is a critical step in staging, prognosis, and development of a treatment plan for patients with cancer (1–3). Lymphatic mapping with sentinel LN (SLN) biopsy has been introduced as an alternative to elective LN dissection and has gained rapid acceptance because it allows improved accuracy and decreased patient morbidity for regional cancer staging (4, 5). However, SLN biopsy still has major limitations. The procedure requires two or more injections of radionuclide tracers followed by an invasive surgical procedure associated with removal of one or more LNs identified as SLNs with associated risks of short- and long-term morbidity and up to 2 weeks of waiting for full histopathology evaluation (6, 7). In addition, the detection limits for pathologic identification of metastatic cells remain subject to the skill and patience of the pathologist. In fact, LN metastases go undetected in up to 9% of patients with melanoma, allowing the disease to spread untreated (8). Therefore, there is a definite and urgent clinical need for an imaging technique that is widely available, is noninvasive and simple to perform, is safe, and can reliably detect and adequately diagnose LN micrometastases in real time.

A host of imaging modalities have been tested in animals and patients in attempts to improve the accuracy and safety of SLN biopsy (9–14). For example, ultrasound imaging has been used to detect changes in LN size, shape, and blood flow, which are associated with metastasis (10). In addition, ultrasound imaging of microbubbles and fluorescence imaging of dye have been proposed as an approach to guide SLN biopsy (14, 15). Positron emission tomography (PET) has showed limited sensitivity in detection of metastatic deposits and has been able to reliably detect metastases only with sizes greater than 80 mm³ (11). Magnetic resonance imaging (MRI) of systemically injected superparamagnetic iron oxide nanoparticles has been shown to indirectly indicate potential presence of metastasis by detecting a disrupted drainage in metastatic LNs (16). Therefore, although these imaging approaches have shown potential in providing useful morphologic and functional information, to date, none have demonstrated sufficient sensitivity and specificity to replace the current invasive SLN procedure (11, 12).

Photoacoustic imaging has recently been introduced as a modality that can improve the detection threshold and sensitivity of radiographic imaging modalities (17–22). It is an emerging hybrid imaging technique that combines the high contrast and sensitivity of optical imaging with the excellent depth resolution of ultrasound imaging. Images are acquired by detecting broadband ultrasound waves that propagate from sites of optical absorption after the tissue is irradiated with a nanosecond pulsed laser. Previous studies have shown that photoacoustic imaging can accurately track intradermally injected dyes or nanoparticles for localization of LNs with high sensitivity (23, 24). However, even though these approaches eliminate the use of radioactive compounds, they still require surgical removal and histopathologic evaluation of the SLNs and they do not provide information about the metastatic state of the SLN.

To address the need for molecular-specific detection of micrometastasis with enhanced specificity and sensitivity, we developed a novel method based on spectroscopic photoacoustic (sPA) imaging with molecularly activated plasmonic nanosensors (MAPS). The MAPS (Fig. 1A) consisted of 40-nm spherical gold nanoparticles (AuNPs) targeted to the epidermal growth factor receptor (EGFR), a molecular target associated with carcinogenesis in many cancers, including lung, oral cavity, and cervix (25), through directional conjugation with anti-EGFR monoclonal antibodies and polyethylene glycol (PEG; refs. 26, 27). Here, we show that MAPS can be used in in vivo applications where nonspecific signal due to delivery of a contrast agent hinders the ability to visualize the underlying molecular expressions. In an orthotopic nude mouse model of squamous cell carcinoma of the oral cavity (SCCOC) that develops LN micrometastases (28), we show that a single peritumoral injection of MAPS followed by sPA imaging is sufficient to detect micrometastases as small as 50 μm in size. Molecular-specific interactions between the EGFR-targeted MAPS and tumor cells lead to a dramatic change in the spectroscopic signal of MAPS in sPA imaging that enables highly sensitive detection of micrometastasis. Thus, the combination of traditional ultrasound imaging to provide guidance and anatomic information with sPA MAPS-based detection of cancer cells provides an integrated technique for the detection of small micrometastases with high sensitivity and specificity. Our method has a great potential for clinical translation because it is based on a combination of a well-established ultrasound imaging and emerging clinically translatable sPA imaging and MAPS. This powerful combination can provide dramatic improvement in the clinical staging, prognosis, and therapeutic planning for patients with cancer with metastatic disease.

Molecular-targeted AuNPs were synthesized as previously described (26). First, 40-nm spherical AuNPs were prepared by heating 100 mL of a 0.01% (w/v) water solution of chloroauric acid (HAuCl₄; Sigma-Aldrich) to boiling, and rapidly adding 4 mL of a 1% (w/v) solution of sodium citrate. Then, anti-EGFR or anti-RG16 monoclonal antibodies (clone C225 or RG16; Sigma-Aldrich) were attached to the AuNPs using directional conjugation chemistry that uses a carbohydrate chain on the Fc portion of the antibody leaving antigen binding sites on the Fab moiety available for targeting (26). Briefly, antibodies at 1 mg/mL in 40 mmol/L HEPES buffer, pH 7.5 and 100 mmol/L sodium periodate (NaIO₄; Sigma-Aldrich) were mixed at 10:1 (v/v) ratio, respectively, and were incubated for 30 minutes in dark at room temperature to oxidize hydroxyl groups of the carbohydrate chain in the Fc antibody regions. Then, approximately 100-fold molar excess of a heterofunctional hydrazide- -PEG–dithiol linker (dithiolaromatic–PEG6k–CONHNH₂; SensoPath Technologies) was added to the oxidized antibodies for 1 hour. The hydrazide portion of the linker reacts with aldehydes on antibody molecules to form a stable linkage. The antibody–linker complexes were purified using a 100,000 Da molecular weight cutoff centrifugal filter (Millipore). After purification, the modified antibodies were resuspended in 40 mmol/L HEPES (pH 7.4) at concentration of 0.1 mg/mL and were mixed with prepared AuNPs at 1:10 antibody:nanoparticle (v/v) ratio for 20 minutes at room temperature. During this step, a stable bond is formed between the gold surface and the linker's thiol groups. Subsequently, 10⁻⁵ mol/L 1× phosphate-buffered saline (PBS) solution of methoxy-poly(ethylene-glycol)thiol (mPEG-SH; 14 kD; Shearwater Polymers) was added at ca. 1:10 mPEG-SH to nanoparticle ratio (v/v) for 20 minutes to passivate the surface of nanoparticle conjugates that is not covered by antibodies (Fig. 1A). The final conjugates were washed using centrifugation (4,000 rpm, 30 minutes) and were resuspended in 1× PBS at concentration of ca. 1 × 10¹² nanoparticles/mL.

All animal studies were approved by the Institutional Animal Care and Use Committee at The University of Texas at Austin (Austin, TX). Balb/c nude mice of 2 months of age were used in this study (Charles River Laboratories). The primary tumors were initiated with a submucosal injection of 300,000 FaDu-Luciferase cells suspended in 30 μL of DMEM media into the tongue. The FaDu cells (ATCC, 2009) are a primary human squamous cell carcinoma of the pharynx cell line that were validated at the Fragment Analysis Facility at Johns Hopkins University (Baltimore, MD) via short-tandem repeat profiling on January 20, 2011. The cells were passaged for fewer than 6 months before these studies. The tumors were allowed to grow to 3 to 4 mm in diameter before the imaging experiments were performed. At this point, 82% of the inoculated mice developed micrometastases in the cervical LNs.

During ultrasound and photoacoustic imaging, the mice were anesthetized with isoflurane (1.5%, 0.5 L/min O₂). Heart rate, respiration rate, and body temperature were monitored using a heated electrocardiogram pad. The mice were imaged before the injection of nanoparticles. Forty microliters of sterile-filtered nanoparticle solution was injected peritumorally immediately after the first imaging session while the mice were still under anesthesia. The total injected dose was 1.6 pmol of nanoparticles that is equivalent to 40 μg of gold. Imaging was performed continuously for 2 hours following the injection of nanoparticles.

Following the ultrasound and photoacoustic imaging, the bioluminescent cancer cells in the mice were imaged using an IVIS Spectrum (PerkinElmer). Once anesthetized with isoflurane, the mice were injected with 100 μL of RediJect d-Luciferin (PerkinElmer). Bioluminescence imaging was performed for 10 to 15 minutes following the injection. Then, mice were euthanized by an overdose of isoflurane and cervical dislocation. Bioluminescence imaging was performed to guide the resection of the primary tumor and metastases. The primary tumor and cervical LNs were fixed in 10% formalin for 24 hours and, then, were transferred to 70% ethanol. Samples were then embedded in paraffin for histologic analysis.

A Vevo LAZR high frequency ultrasound and photoacoustic imaging system (VisualSonics) equipped with a linear array transducer (LZ-550, 40 MHz center frequency) was used to acquire all photoacoustic and ultrasound images. The co-registered spatial dimensions of the collected ultrasound and photoacoustic images were 14 mm (width) by 15 mm (depth). Ultrasound and photoacoustic images at each optical wavelength covering three-dimensional (3D) volumes surrounding the cervical LNs were acquired by scanning the ultrasound transducer in the elevational direction with a step size of 76 μm. The laser was tuned to optical wavelengths of 680, 700, 720, 740, 760, 780, 800, 820, 840, and 860 nm. Laser fluences, measured by a Nova II power meter with a PE50BB sensor (Ophir), were 10 to 20 mJ/cm²; all laser energies were below the American National Standards Institute (ANSI) safe exposure level for human skin. Photoacoustic images were averaged eight times, thus suppressing uncorrelated noise. Each photoacoustic image was normalized by the measured fluence to correct for the pulse-to-pulse laser energy variations.

To resolve different types of optical absorbers, we developed an ultrasound-guided sPA imaging algorithm. We restricted our analysis to three predominant absorbers: oxygenated hemoglobin (⁠|${\rm Hb}_{{\rm O}2}$|⁠), deoxygenated hemoglobin (Hb), and activated MAPS. The ultrasound image was used to automatically segment the mouse to remove photoacoustic imaging artifacts on the surface of the skin. The relative contributions of |${\rm Hb}_{{\rm O}2}$|⁠, Hb, and MAPS to the overall photoacoustic signal in each pixel was determined using a previously developed linear least squares method (29). Pixels containing negative concentrations (an artifact arising from noise or the presence of absorbers other than |${\rm Hb}_{{\rm O}2}$|⁠, Hb, and MAPS) were not displayed. The oxygen saturation of hemoglobin (SO₂) was computed as the ratio of |${\rm Hb}_{{\rm O}2}$| concentration to total hemoglobin concentration. The sPA_Hb signal was defined as the total hemoglobin concentration while the sPA_MAPS signal was defined as the concentration of activated MAPS. sPA images were displayed using yellow color for activated sPA_MAPS and blue/red colors for sPA_Hb with SO₂ varying from 60% (blue) to 100% (red). The range from 100% to 60% was chosen because it represents normal physiologic extremes of blood oxygen saturation (e.g., arterial and venous blood). Pixels below 60% SO₂ were colored blue. The intensity of the sPA signal was determined from the average photoacoustic signal.

The SLNs were identified and manually segmented using the 3-D visualization of ultrasound data. Only the signal inside or within 100 μm of the LNs was considered. The estimated concentrations were filtered with a 5 × 5 × 5 voxel median filter to suppress physiologic motion. The ratio between the residual of the least squares approximation and the average concentration was used as a threshold to discard pixels whose spectra did not correlate strongly with the analyzed absorbers. In general, most pixels were retained during this operation. We noticed that this step mainly led to a suppression of the signal generated at the surface of the skin (because melanin was not included in spectral analysis). Finally, the total signal stemming from the activated MAPS was summed throughout each node to be used as an indicator of metastasis.

Formalin-fixed, paraffin-embedded samples were sliced in 100-μm levels. The samples were stained using hematoxylin and eosin (H&E) stain to show tissue morphology and silver stain to detect gold and anti-EGFR rabbit polyclonal antibodies (Sigma HPA018530) for molecular-specific identification of EGFR(+) metastatic cells. A Leica DMI 3000B microscope coupled to a DFC 290 camera was used to record color images.

Dark-field and hyperspectral imaging were performed on a Leica DM 6000 microscope with a 20× objective and xenon lamp. Hyperspectral imaging was carried out with PARISS system (Lightform, Inc.) attached to the Leica DM 6000 microscope. The system was calibrated using a multi-ion discharge lamp (MIDL) with known emission spectrum. To normalize scattering spectra with the incident lamp spectrum, a spectrum of the excitation xenon lamp was acquired using a Labsphere Spectralon calibrated scattering substrate.

A Lilliefors test was used to determine that the sPA_MAPS signal in the LNs did not follow a normal distribution. A nonparametric Mann–Whitney U test was used to test the null hypothesis that the sPA_MAPS signal in metastatic nodes was not significantly greater than the controls. All LNs from mice injected with EGFR-targeted MAPS were used to evaluate the sensitivity and specificity of the imaging technique. A receive operator characteristic (ROC) curve constructed from the sPA_MAPS signal was used to determine the threshold for sensitivity and specificity.

Three distinct states can be encountered in vivo after injection of MAPS (Fig 1A): (i) cells with no MAPS present at the site of interest, (ii) no interactions between cells and delivered MAPS due to the lack of a molecular biomarker of interest, and (iii) molecular-specific interactions between cells and targeted MAPS. Hyperspectral and dark-field optical microscopy of cell cultures simulating these three states (Fig. 1) show that delivery of MAPS increases optical absorption and scattering in the wavelength range near nanoparticle plasmon resonance peak (Fig. 1B and D), while molecular-specific interactions between MAPS and EGFR-expressing cancer cells are associated with plasmon resonance coupling, resulting in striking optical changes including red spectral shift in excess of 100 nm and broadening of nanoparticle extinction spectra (Fig. 1B and E). This sensing method is unique among optical activatable imaging agents, which typically rely on enzymatic cleavage for activation and, therefore, are mostly limited to detection of various proteinases (13).

To establish the utility of MAPS in vivo, we performed sPA imaging with MAPS for noninvasive detection of micrometastases in the lymphatics of the orthotopic nude mouse model of SCCOC (28). Either EGFR-targeted MAPS or control AuNPs conjugated with a nonspecific RG16 monoclonal antibody were injected peritumorally and were allowed to drain to the cervical LNs of tumor-bearing mice. In clinical practice, a similar injection of radionuclide tracers is routinely performed to identify the location of LNs; however, it does not allow for the detection of cancer cells. The RG16 antibody acts as a nonspecific control because it targets heavy chains of rabbit IgG/IgM/IgA, which are not present in mouse or FaDu cells. The injected dose was approximately 1.6 pmol, which is significantly less than a dose of AuNPs with any reported cytotoxicity in vivo (30). The cervical LNs, the site of micrometastatic foci in this mouse model (28), were identified using ultrasound imaging.

To differentiate cancer cells labeled with MAPS from the background, we implemented an image analysis algorithm to spectroscopically resolve the two major contributors to photoacoustic signal in our studies—hemoglobin (sPA_Hb) and MAPS-labeled cancer cells (sPA_MAPS)—in the red-near-infrared (NIR) wavelength region (Fig. 2). The relative contributions of sPA_Hb and sPA_MAPS to the overall photoacoustic signal were determined via a linear least squares method (29). The SO₂, as calculated by the ratio between the concentration of oxygenated hemoglobin and total hemoglobin, was used to assign a color to the sPA_Hb signal. No significant changes in the red-NIR photoacoustic signal intensity from LNs were evident in control groups for several hours following injection of MAPS due to low absorption of noninteracting MAPS in this spectral region and no apparent changes in blood flow. In contrast, a strong increase in the photoacoustic signal was observed in LNs harboring metastases less than 2 hours after administration of MAPS (Fig. 2).

Overlaid ultrasound and sPA images of cervical LNs (Fig. 3C–E) depicting the anatomy (gray) and contributions from sPA_Hb (blue/red) and sPA_MAPS (yellow) 2 hours after the peritumoral injection of MAPS show the ability of our method to detect micrometastases in the lymphatics with high spatial resolution. LNs are identifiable by a dark bean-shaped region in the ultrasound images (outlined using white dashed line); the hypoechoic contrast is typical for LNs in ultrasound imaging (see an example in Supplementary Fig. S1; ref. 31). The injection of EGFR-targeted MAPS leads to a strong increase in the sPA_MAPS signal from the LNs with micrometastases (Fig. 3C). A 3D reconstruction of the ultrasound/sPA images shows that the sPA_MAPS signal is constrained to a small region on the edge of the LN (Fig. 3G), which is consistent with the location of metastases observed in this mouse model. Furthermore, there was excellent correlation between the spatial location of the sPA_MAPS signal and the in situ bioluminescence signal from metastatic cells (Fig. 4).

Overall, tumor-bearing mice injected with EGFR-targeted MAPS exhibit a statistically significant increase in sPA_MAPS signal from metastatic LNs (N = 7 nodes) over LNs of control mice injected with RG16-conjugated AuNPs (N = 7 nodes; P = 0.006) and either LNs of normal mice (N = 5 nodes) or healthy LNs in tumor-bearing mice (N = 3 nodes) injected with EGFR-targeted MAPS (total N = 8 nodes; P = 0.005; Fig. 3F). Statistical analysis of the data results in a sensitivity of 100% and a specificity of 87.5% of our method in detection of LN micrometastasis. The results show that our method is able to detect metastases as small as 50 μm (which was determined by measuring the longest diameter of the metastasis across all histology slices), which corresponds to approximately 30 cells (if we assume a spherical metastasis and cell diameter of 15 μm). The single false-positive came from a LN in a tumor-bearing mouse with a confirmed metastasis in the contralateral cervical LN. Given the fact that bilateral LN metastases developed in a large portion of mice in this study (21% of metastatic mice), it is possible that MAPS are more sensitive than the validation using histology (the current gold standard) and this false-positive represents an occult metastasis.

Continuous sPA imaging of cervical LNs immediately following MAPS injection shows progressive labeling of cancer cells over the course of 2 hours, indicating quick delivery and uptake of the nanoparticles (Fig. 5). A steady increase in sPA_MAPS signal amplitude and volume is seen in a small localized region on the border of the LN. The vasculature that is visible throughout the rest of the image experiences little variation over the same timescale. The contrast generated by MAPS generally tended to plateau after 2 to 3 hours. This indicates that the optimal time point to perform imaging occurs in the first few hours after the injection. Furthermore, the 2-hour timescale is consistent with nanoparticle drainage through the lymphatics observed in vivo (32, 33) and cellular uptake of MAPS observed in vitro (34). Taken together, these observations provide additional support to the conclusion that the highly localized signal is due to tumor cell–mediated activation of MAPS.

H&E staining (Fig. 6A–F) confirmed the presence of the metastases that were indicated by the bioluminescence imaging and were detected by sPA imaging of MAPS. The metastases had a tendency to form in the subcapsular region of the node near the afferent lymphatic vessel or within the vessel itself. This trend is in excellent correlation with the regions of elevated sPA_MAPS and bioluminescence signal (Figs. 3, 4, and 6). Immunohistochemistry confirmed the overexpression of EGFR in the micrometastases (Fig. 6G–I). Furthermore, hyperspectral dark-field reflectance optical microscopy of the excised tissue slices showed strong labeling of tumor cells and plasmon resonance coupling between MAPS at the location of the micrometastases (Fig. 7). No such signal was observed in the metastatic mice injected with RG16-conjugated AuNPs. These results further indicate that activation of MAPS is specific to EGFR overexpressing cells.

Molecular-specific imaging of small cellular clusters in vivo at substantial depth remains a great challenge in modern biology and medicine (35). Attempts have been made to extend virtually every biomedical imaging modality, including PET (36), MRI (37), ultrasound imaging (38, 39), optical imaging (40), and photoacoustic imaging (21, 22), to localize molecular expressions in vivo. Nevertheless, accurate measurements of molecular expressions in a tissue remain a difficult task. Indeed, it is commonly assumed that the presence of a molecularly targeted contrast agent correlates to the level of expression of the targeted molecule. What this assumption fails to consider is that delivery of a contrast agent to a site of interest and interaction between the agent and its target are two independent events. Therefore, passive accumulation of a contrast agent in a tissue (e.g., via the enhanced permeability and retention effect in tumors; ref. 41) leads to decreased specificity in molecular sensing (35, 42).

Advanced optical imaging techniques, such as Förster Resonance Energy Transfer (43) or fluorescent imaging using activatable probes (13), are able to provide a much more specific visualization of molecular expressions. However, these methods suffer from low penetration depth (less than one millimeter in the ballistic regime of photon transport) or poor resolution and sensitivity at greater depths (40). Furthermore, fluorescent probes are prone to photobleaching. All of these factors severely limit clinical applicability of optical techniques for molecular detection of diseases, including micrometastases. Photoacoustic imaging addresses these limitations by combining the high contrast of optical imaging with the excellent resolution of ultrasound imaging at depths up to 5 cm (20–22, 44). In addition, plasmonic nanoparticles characterized by enhanced optical absorption and photostability provide strong contrast in photoacoustic imaging (19, 44, 45). However, previous efforts to combine plasmonic nanoparticles with in vivo photoacoustic imaging have not addressed the fundamental challenge in specificity that most other molecular imaging modalities suffer from: nanoparticles that have interacted with a molecule of interest cannot be differentiated from those that have simply been delivered to the region.

Our approach based on nanoscale interactions between cells and MAPS provides a straightforward mechanism to decouple nanoparticle delivery from molecular interactions (Supplementary Fig. S2). Our results show an excellent ability to detect small colonies of cancerous cells. Indeed, the 100% sensitivity and 87.5% specificity of sPA imaging of MAPS are significantly better than the results obtained using other noninvasive imaging modalities (46). For example, PET, while more sensitive than many other radiographic modalities, has 50% sensitivity and 87% specificity in detection of impalpable cervical metastases in human patients with oral cancer (47). Demonstrated here is the capacity to detect metastatic tumors consisting of just few tens of malignant cells deep inside tissue.

The translation of this technique to the clinic will require both a clinical imaging system and a clinically approved contrast agent. The clinical imaging system must be developed to image tissues at 3 to 5 cm depth. This could be achieved by combining a clinically relevant lower frequency ultrasound transducer (e.g., 10 MHz center frequency) with a tunable laser source in the red-NIR spectral region that provides sufficient pulse power at clinically save irradiation levels; this combination will achieve light penetration in tissue and propagation of acoustic waves that are appropriate for LN imaging in multiple anatomic locations including LNs associated with oral and breast cancers. In general, we do not expect the approach to depend on the location of the primary tumor or its proximity to the SLN. What is important is that the primary tumor needs to be accessible for an injection of the MAPS and the LNs must be accessible for photoacoustic imaging (i.e., they must be within a few centimeters of a handheld or endoscopic imaging probe). We expect that the sensitivity of our imaging method in the clinic will be primarily limited by the ability to deliver light deep in tissue. In addition, while the dose of gold used in this study is well below any previously reported toxicity threshold (30), gold nanoparticles are not yet accepted for widespread clinical use. One of the concerns associated with gold nanoparticles is potential long-term toxicity because particles larger than 5 to 6 nm are not easily cleared from the body. Previous studies have shown that nanoparticles with sizes below ca. 6 nm can undergo efficient clearance through urine and feces (48–50). Therefore, scaling down the MAPS to 5 nm is a viable path toward clinical acceptance. Because the MAPS are delivered through the lymphatics, the short half-life of small particles in the bloodstream will not affect our technology as opposed to applications requiring systemic delivery where small particles can undergo quick renal clearance. Overall, the clinical translation of the developed method is possible and anticipated.

S.Y. Emelianov and K.V. Sokolov have ownership interest in and are consultants/advisory board members for NanoHybrids, Inc. No potential conflicts of interest were disclosed by the other authors.

Conception and design: G.P. Luke, J.N. Myers, S.Y. Emelianov, K.V. Sokolov

Development of methodology: G.P. Luke, S.Y. Emelianov, K.V. Sokolov

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G.P. Luke

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G.P. Luke, J.N. Myers, S.Y. Emelianov, K.V. Sokolov

Writing, review, and/or revision of the manuscript: G.P. Luke, J.N. Myers, S.Y. Emelianov, K.V. Sokolov

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S.Y. Emelianov, K.V. Sokolov

Study supervision: S.Y. Emelianov, K.V. Sokolov

The authors thank Dr. Aristarchos Papagiannaros of the University of Texas MD Anderson Cancer Center for his help with nanoparticle conjugation, tumor inoculation, and histology in preliminary studies, Dr. Pratixa Joshi, Dr. Justina Tam, and Mr. Chun-Hsien Wu of The University of Texas at Austin for their assistance with nanosphere synthesis and antibody conjugation, Dr. Timothy Larson at The University of Texas at Austin for his help with optical microscopy and hyperspectral image processing of cell cultures, Ms. Sally Amen of the University of Texas at Austin for consulting on the statistical analysis, and Mrs. Nancy Wick Otto of the University of Texas MD Anderson Cancer Center Science Park for her expertise on immunohistochemistry optimization and processing.

The work was supported, in part, by grants from the NIH (R01EB008101 and F31CA168168).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.