Intraoperative identification of carcinoma at lumpectomy margins would enable reduced re-excision rates, which are currently as high as 20% to 50%. Although imaging of disease-associated biomarkers can identify malignancies with high specificity, multiplexed imaging of such biomarkers is necessary to detect molecularly heterogeneous carcinomas with high sensitivity. We have developed a Raman-encoded molecular imaging (REMI) technique in which targeted nanoparticles are topically applied on excised tissues to enable rapid visualization of a multiplexed panel of cell surface biomarkers at surgical margin surfaces. A first-ever clinical study was performed in which 57 fresh specimens were imaged with REMI to simultaneously quantify the expression of four biomarkers HER2, ER, EGFR, and CD44. Combined detection of these biomarkers enabled REMI to achieve 89.3% sensitivity and 92.1% specificity for the detection of breast carcinoma. These results highlight the sensitivity and specificity of REMI to detect biomarkers in freshly resected tissue, which has the potential to reduce the rate of re-excision procedures in cancer patients. Cancer Res; 77(16); 4506–16. ©2017 AACR.

Approximately 200,000 patients are diagnosed with early-stage breast carcinoma each year in the United States, for which breast-conserving surgery (a.k.a. partial mastectomy or lumpectomy) is a standard intervention (1). Unfortunately, among various institutions, between 20% and 50% of these patients require additional surgery if postoperative pathology reveals that the resection margins are positive for carcinoma (2). Recent studies and consensus guidelines recommend that lumpectomy procedures be considered complete when a negative margin is observed for invasive carcinoma (3)—otherwise referred to as “no tumor on ink” (i.e., no carcinoma at the surgical margin surface). Whether invasive or noninvasive (in situ), there is no debate that the presence of carcinoma at the surgical margin surface would warrant re-excision at all institutions and that intraoperative identification of residual carcinoma at these surfaces would thereby greatly reduce the rates of re-excision surgeries.

Frozen section analysis has been explored for the intraoperative assessment of surgical margins. However, frozen sectioning is particularly difficult to perform for fatty breast tissues, requires additional surgical time, introduces freezing artifacts that compromise postoperative histologic diagnosis, and can yield significant false-negative rates since only a small number of thin tissue sections can be rapidly prepared and evaluated during surgery (e.g., low sensitivities of <70% have been reported; refs. 4, 5). A number of other ex vivo imaging techniques have been proposed to assess breast margins, such as confocal microscopy (6), nonlinear microscopy (95.4% sensitivity, 93.3% specificity; refs. 7, 8), optical coherence tomography (90%–100% sensitivity, 75%–82% specificity; refs. 9, 10), optical coherence microelastography (11), light reflectance spectroscopy (74% sensitivity, 86% specificity; refs. 12, 13), autofluorescence lifetime measurement (12), and intrinsic Raman spectroscopy (94%–100% sensitivity, 96%–100% specificity; refs. 14–16). Optical-sectioning microscopy techniques, such as confocal microscopy and nonlinear microscopy, can obtain high-resolution and high-contrast images of freshly resected tissues. However, applying these techniques for intraoperative imaging of large tissue surfaces is challenging due to the relatively slow speed of these point-by-point imaging techniques. In addition, these techniques suffer from an extremely limited depth of focus, necessitating elaborate tissue-flattening or volumetric imaging strategies to visualize a large tissue surface with topological irregularities (6, 7). Spectroscopic methods (e.g., reflectance, Raman, and autofluorescence lifetime) reveal the relative concentrations of chemical constituents such as hydrocarbons, lipids, nucleic acids, and/or the size distribution of photon-scattering objects in tissues, and have yielded promising results for tumor detection (12–16). However, the ability to image the expression of molecular biomarkers, which play a prominent role in recent advances in precision medicine (17, 18), would offer complementary information that could further improve the sensitivity and specificity of detection.

Over the past few decades, there has been increasing confidence in the presumption that molecular-imaging approaches—namely, the imaging of protein biomarkers of cancer—can identify tumors with a high degree of sensitivity and specificity (19). High specificity would be valuable for intraoperative cancer detection to minimize over-excision and to optimize patient cosmesis (a major goal of breast-conserving surgeries). However, in order to achieve high detection sensitivity, multiple biomarkers should be evaluated since the molecular profiles of most cancers, including breast carcinoma, vary greatly between patients as well as spatially and temporally within a single tumor mass (20, 21). Although quantum dots (QD) have been used for multiplexed imaging of three to five biomarkers in breast cancer cells and formalin-fixed tissue sections, (22), no clinical studies have been reported to evaluate the efficacy of QD-based imaging technologies for guiding lumpectomy. Recently, surface-enhanced Raman-scattering (SERS) nanoparticles (NP) have attracted wide interest due to their excellent multiplexing capabilities over QDs and fluorescent dyes (23, 24). These SERS NPs are available in many “flavors,” each of which emits a characteristic Raman fingerprint spectrum when illuminated at a common wavelength (25–27). By functionalizing various flavors of SERS NPs with different targeting molecules [e.g., antibodies (28), affibodies (29), etc.], the NPs can be multiplexed to simultaneously target and image a large panel of protein biomarkers.

We have recently demonstrated that a Raman-encoded molecular imaging (REMI) technique, which utilizes the topical application of multiplexed SERS NPs, enables the rapid visualization of multiple cell-surface biomarkers at the surfaces of fresh tissues (28, 30–32). In our REMI approach, toxicity and sterility concerns are circumvented by staining and imaging fresh surgical specimens ex vivo, with the potential to allow for expedited regulatory approval and rapid clinical translation.

A critical component of REMI is a ratiometric imaging method that enables accurate quantification of biomarker expression levels by utilizing one untargeted NP flavor to normalize for the nonspecific accumulation exhibited by all of the NPs (both the targeted and untargeted NPs), as for example due to off-target binding, uneven NP delivery and washout, and variations in tissue permeability and retention (33–39). The SERS NPs used for REMI are particularly well-suited for accurate ratiometric quantification due to the fact that they are all excited at a single wavelength (785 nm) and emit Raman spectra within the same narrow spectral range (approximately 850–900 nm), which ensures that all NPs are irradiated identically and are affected by the same tissue optical properties. This is in contrast to fluorescent agents, in which a large range of laser-illumination wavelengths and collection bands are typically needed for multiplexed imaging. As a result, each fluorescent agent is affected by different tissue-scattering and absorption properties (wavelength-dependent), which requires calibration with wavelength-dependent light propagation models and thus complicates ratiometric quantification (40). Our previous studies have verified that REMI provides a linear measurement of NP concentrations/ratios when applied on tissues (Supplementary Fig. S1A and S1B; refs. 28, 30, 31, 33, 41, 42) and, more importantly, provides a quantitative measure of biomarker expression levels that agrees with both flow cytometry and gold-standard IHC (Supplementary Fig. S2A–S2D; refs. 28, 30, 32).

Here, we assess the diagnostic accuracy of REMI for identifying carcinoma at the surfaces of freshly excised breast specimens through simultaneous quantification of four cell-surface biomarkers—HER2, membrane estrogen receptor (mER), EGFR, and CD44 (a brief description of these biomarkers is provided in the Supplementary Information). By imaging 57 fresh specimens from 29 patients, we validated and optimized the sensitivity and specificity of REMI to identify the elevated expression of the four biomarkers mentioned above, in comparison with IHC, which is the clinical gold-standard method for assessing protein expression in tissues. In addition, by combining all four biomarkers, we assessed the overall sensitivity and specificity of REMI for identifying breast carcinoma at surgical margin surfaces, in comparison with the diagnostic gold standard of hematoxylin and eosin (H&E) histopathology.

REMI system

A customized spectral-imaging system has been developed to measure the concentration and concentration ratio of SERS NPs that are topically applied on tissue specimens (30). The NP-stained surface (surgical margin) of the tissue specimen was raster scanned with a two-axis stage (Newmark systems Inc., ET-50-11) and imaged using a fixed spectral-imaging probe (FiberTech Optica Inc.). The imaging probe utilizes a multimode fiber (100-μm core, 0.10 NA) at the center of the probe for illumination, and 27 surrounding multimode fibers (200-μm core, 0.22 NA) for light collection (Raman, autofluorescence, and back-scattered laser light). A 785-nm diode laser (18 mW at the tissue) is used to illuminate the tissue, creating a laser spot with a diameter of 0.5 mm (imaging resolution). Light collected by the 27 multimode fibers are transmitted to a customized spectrometer (Andor Holospec), where they are filtered (to remove autofluorescence and back-scattered laser light) and then dispersed onto a cooled deep-depletion spectroscopic CCD (Andor, Newton DU920P-BR-DD). The detector integration time (i.e., the spectral acquisition rate, which equals the pixel rate) utilized in this study was 50 ms. A direct-classical-least-squares algorithm was employed to calculate the concentrations and ratios of various SERS NP flavors as described previously (30, 31) and as summarized in the Supplementary Information. Figure 1 provides an illustration of the REMI process.

Figure 1.

REMI for intraoperative guidance of lumpectomy. In a clinical implementation of REMI, freshly resected human breast tissues from lumpectomy procedures are immediately transferred to a pathology suite for intraoperative consultation. Each specimen is topically stained with a mixture of SERS NPs (multiple biomarker-targeted NPs and at least one untargeted control NP, step 1), followed by spectroscopic imaging of the surgical margin surface (step 2). The acquired SERS spectra are demultiplexed to determine the ratio of the targeted vs. untargeted NPs (step 3), which enables the quantification of various biomarker targets (e.g., EGFR, HER2, mER, and CD44 in this study, step 4). REMI images of the individual biomarkers are combined to detect the presence of residual tumors at the surgical margin surfaces of the specimens (step 5). The entire REMI procedure (staining, rinsing, imaging, spectral demultiplexing) was performed within 10 to 15 minutes depending upon the size of the specimen.

Figure 1.

REMI for intraoperative guidance of lumpectomy. In a clinical implementation of REMI, freshly resected human breast tissues from lumpectomy procedures are immediately transferred to a pathology suite for intraoperative consultation. Each specimen is topically stained with a mixture of SERS NPs (multiple biomarker-targeted NPs and at least one untargeted control NP, step 1), followed by spectroscopic imaging of the surgical margin surface (step 2). The acquired SERS spectra are demultiplexed to determine the ratio of the targeted vs. untargeted NPs (step 3), which enables the quantification of various biomarker targets (e.g., EGFR, HER2, mER, and CD44 in this study, step 4). REMI images of the individual biomarkers are combined to detect the presence of residual tumors at the surgical margin surfaces of the specimens (step 5). The entire REMI procedure (staining, rinsing, imaging, spectral demultiplexing) was performed within 10 to 15 minutes depending upon the size of the specimen.

Close modal

Tissue staining and imaging

Unstained specimens were first imaged to acquire a set of background spectra for the calculation of their principal components (to account for the tissue background during least-squares demultiplexing). The specimen surfaces (surgical margins) were then topically stained with the NP staining solution using a convection-enhanced staining method (28). Based on a previously optimized staining protocol (30), the staining solution consisted of a mixture of 5 flavors of SERS NPs (150 pmol/L per flavor) supplemented with 1% BSA to minimize nonspecific binding. After 5 minutes of staining, the tissue sample was rinsed in 50-mL PBS with gentle agitation for 10 seconds, followed by raster-scanned imaging of the entire stained tissue surface (>3 cm2/min), followed by spectral demultiplexing (∼1,000 spectra/s) and ratiometric mapping (<1 minute). The entire REMI procedure (including image processing) can be performed within 15 minutes.

SERS NPs and functionalization

The SERS NPs were purchased from BD (Becton, Dickinson and Company). These NPs consist of a 60-nm-diameter gold core, a layer of Raman reporters adsorbed onto the surface of the gold cores, and a 60-nm-thick silica coating (Supplementary Fig. S3A). The thick silica shell makes the SERS NP signals insensitive to the environment and immune from signal changes induced by aggregation of the NPs. Previous studies have demonstrated excellent linearity of NP measurements in a variety of animal and human tissues over a wide range of concentrations (25). Five “flavors” of NPs were used in this study, identified as S420, S421, S440, S481, and S493, each of which emits a characteristic Raman spectrum due to chemical differences in the Raman reporter layer (Supplementary Fig. S3B).

Using a previously described conjugation protocol (33), unique SERS NP flavors were functionalized with different mAbs targeting either EGFR, HER2, ER, or CD44. In addition, negative-control NPs were prepared by conjugating one NP flavor with an isotype control antibody (mouse IgG1). The NPs, which were functionalized with active thiols at their surface, were first reacted with a fluorophore, DyLight 650 Maleimide (Thermo Scientific, 62295), for the purposes of flow-cytometry characterization. The NPs were then conjugated with either an isotype control (Thermo Scientific, MA110407), an anti-EGFR mAb (Thermo Scientific, MS378PABX), an anti-HER2 mAb (Thermo Scientific, MS229PABX), an anti-ER mAb (Thermo Scientific, RM9101S0) or an anti-CD44 mAb (Abcam plc., ab6124) at 500 molar equivalents per NP. To ensure reproducible tissue imaging throughout the duration of this study, mAb-conjugated NPs were used for tissue imaging experiments only when they yielded flow-cytometry results that were similar (±20% error) to the ones shown in Supplementary Fig. S4A–S4D. The flow-cytometry methods are described in the Supplementary Information.

Acquisition and handling of human breast tissues

This study was approved by the University of Washington Institutional Review Board, and was conducted in accordance with the provisions of the Declaration of Helsinki. Deidentified human breast tissue specimens were obtained from consenting patients and imaged within 1 hour after lumpectomy or mastectomy at the University of Washington Medical Center (informed consent was obtained from all patients). Tissue collection was managed by the Northwest BioTrust under an Institutional Review Board exemption for these deidentified tissues. After imaging, the tissues were fixed with 10% formalin and submitted for histopathology (IHC and H&E staining). To facilitate the correlative analysis between REMI and histopathology images, the fixed tissues were flattened against the bottom of the tissue mold, and sectioned as close as possible to the imaged surface (within ∼200 μm of the surface). According to a pathologic modeling study (43), a tissue section at a depth of <250 μm is predicted to exhibit minimal variation in important pathologic parameters compared with the surface. The histopathology slides were scanned using a whole slide scanner (Hamamatsu Nanozoomer).

Pathology for validation

Histology slides from the breast tissue specimens were independently interpreted by two pathologists who were blinded to the REMI results. For each specimen, a slide-scanned composite H&E image was first analyzed to annotate the carcinomatous regions. The corresponding four IHC images (HER2, EGFR, ER, and CD44) were then scored according to standard-of-care scoring criteria (44–47). For IHC of HER2, a score of 0 or 1+ is considered negative, a score of 2+ is considered equivocal, and 3+ is considered positive (for a full discussion of HER2 scoring criteria, see 45). For IHC of ER, an Allred score of 0–2 is considered negative, and an Allred score of 3–8 is considered positive. For IHC of CD44 and EGFR, positive expression requires that >1% of carcinoma cells exhibit membranous staining.

Statistical analysis

Statistical analysis was performed in Matlab or Origin. Receiver operating characteristic (ROC) curves were plotted in MedCalc. Statistical significance was calculated by a Student t test (two-sample, unpaired), and the level of significance was set at P < 0.001. For all of the box plots, the bottom and top of the box represent the 1st and 3rd quartiles of the dataset, respectively, and the band inside the box represents the median (2nd quartile) of the data.

REMI workflow

We performed REMI of deidentified fresh tissue specimens obtained from lumpectomy and mastectomy procedures (57 specimens from 29 patients). Each specimen was topically stained with a mixture of 5 flavors of SERS NPs (4 targeted and 1 untargeted control, 150 pmol/L/flavor) using a convection-enhanced staining method (28), followed by raster-scanned imaging and spectral demultiplexing to simultaneously quantify the expression levels of 4 biomarkers—HER2, mER, EGFR, and CD44. The entire REMI procedure (staining, rinsing, imaging, spectral demultiplexing) was performed within 10 to 15 minutes depending upon the size of the specimen (see Fig. 1 for an illustration of the REMI process). This time frame is comparable with that of current intraoperative guidance techniques such as frozen-section analysis, which typically requires 15 to 30 minutes and suffers from sampling errors and freezing artifacts due to the high lipid content in breast tissues. After imaging, specimens were submitted for standard formalin-fixed paraffin-embedded (FFPE) histology with H&E staining, as well as IHC of the relevant protein targets. Additional details regarding the study design and biostatistics, the technical robustness of REMI, and the biomarkers assessed in this study are provided in the Supplementary Information.

REMI enables detection of various molecular subtypes of carcinoma

Representative REMI images and corresponding histology images (H&E and IHC) are shown in Fig. 2A–C from a specimen containing HER2-positive DCIS. Previously, we had demonstrated the ability to perform REMI of the HER2 receptor in fresh breast tissues by multiplexing two NP flavors (HER2-NPs and isotype-NPs; ref. 30). By improving the tissue staining protocol and spectral resolution of our system (28), we are now able to image four biomarkers (with five multiplexed NP flavors) in the same amount of time as the previous single-biomarker imaging procedure (<15 minutes). REMI of all four biomarkers exhibits good agreement with H&E and IHC data (Fig. 2B and C). REMI also enables the detection of double-positive (ER+ and HER2+) breast tumors (Supplementary Fig. S5A–S5C).

Figure 2.

REMI enables multiplexed detection of disease-associated biomarkers. Here, an example is shown of a patient with a HER2-positive neoplasm in which REMI successfully identifies the overexpression of this cell-surface biomarker. A, Photograph of a human breast specimen with DCIS. B, REMI results. Scale bars, 5 mm. The color bar indicates NP ratios. C, Validation data: H&E histology and IHC. H&E histology is the clinical gold-standard method for the detection of carcinoma, and IHC is a clinical gold-standard method for the assessment of protein expression. In this example, the specimen is positive for HER2 and negative for ER, EGFR, and CD44. Scale bars, 50 μm.

Figure 2.

REMI enables multiplexed detection of disease-associated biomarkers. Here, an example is shown of a patient with a HER2-positive neoplasm in which REMI successfully identifies the overexpression of this cell-surface biomarker. A, Photograph of a human breast specimen with DCIS. B, REMI results. Scale bars, 5 mm. The color bar indicates NP ratios. C, Validation data: H&E histology and IHC. H&E histology is the clinical gold-standard method for the detection of carcinoma, and IHC is a clinical gold-standard method for the assessment of protein expression. In this example, the specimen is positive for HER2 and negative for ER, EGFR, and CD44. Scale bars, 50 μm.

Close modal

Triple-negative breast carcinomas, which express negligible levels of ER, PR, and HER2, but which frequently overexpress EGFR and/or Ki-67 (47), are an aggressive breast carcinoma subtype with a high rate of recurrence. Supplementary Fig. S6A–S6C provides an example of REMI for a specimen with triple-negative invasive ductal carcinoma (IDC). The images in Supplementary Fig. S6B reveal elevated EGFR expression and low HER2 and ER expression, in agreement with IHC (Supplementary Fig. S6C). These results demonstrate the value of including EGFR in the REMI biomarker panel for guiding the complete removal of triple-negative breast tumors.

REMI enables discrimination between malignant and benign lesions

Intraoperative evaluation of margins is especially challenging when malignant lesions are mixed with benign lesions such as usual ductal hyperplasia (UDH). Figure 3A–C shows two specimens from a single patient, in which there is a benign lesion (UDH) in one specimen and a malignant lesion (IDC) in the other. Postoperative IHC validation data agree with the REMI results (Fig. 3B and C). REMI shows that the expressions of ER and CD44 are slightly elevated in the benign lobule and UDH regions (where moderate nuclear or cytoplasmic staining is shown in Fig. 3C), but are significantly elevated in the IDC regions (where strong nuclear or membranous staining is shown in Fig. 3C). These results show that the IDC regions and benign regions (UDH and benign lobules and stroma) can be well distinguished by setting an appropriate REMI threshold, as will be discussed later. A total of seven patient specimens were found to contain benign lesions, including two cases of UDH, one case of fibroadenoma (Supplementary Fig. S7A–S7C), and four cases of fibrocystic changes (see Supplementary Table S1 for a summary of these benign lesions).

Figure 3.

REMI enables discrimination between benign and malignant lesions. A, Photograph of two tissue specimens from a single patient. The specimen on the left presents regions of UDH (benign lesion), and the specimen on the right presents IDC (malignant lesion). B, REMI reveals that the overexpression of ER and CD44 is associated with IDC (specimen on right) but not UDH (specimen on left). Scale bars, 5 mm. The color bar indicates NP ratios. C, Validation data: H&E and IHC. The specimen with IDC is positive for ER and CD44, which is concordant with the REMI results. The specimen with UDH is negative for all four biomarkers. See text for details on how the IHC results are scored based on standard-of-care methods. Scale bars, 50 μm.

Figure 3.

REMI enables discrimination between benign and malignant lesions. A, Photograph of two tissue specimens from a single patient. The specimen on the left presents regions of UDH (benign lesion), and the specimen on the right presents IDC (malignant lesion). B, REMI reveals that the overexpression of ER and CD44 is associated with IDC (specimen on right) but not UDH (specimen on left). Scale bars, 5 mm. The color bar indicates NP ratios. C, Validation data: H&E and IHC. The specimen with IDC is positive for ER and CD44, which is concordant with the REMI results. The specimen with UDH is negative for all four biomarkers. See text for details on how the IHC results are scored based on standard-of-care methods. Scale bars, 50 μm.

Close modal

REMI enables detection of carcinoma in patients whose molecular phenotype varies over time

The molecular profile of breast carcinoma in a single patient often varies spatially and over time, such that a single targeted imaging agent may not detect the tumor at all locations within the tumor mass, or at all times (20). For example, several of the patients who were diagnosed with ER-positive carcinoma (by core needle biopsy) underwent endocrine therapy before lumpectomy. Resected specimens from these patients were imaged using REMI and submitted for histology, which showed that three of these patient's tumors no longer expressed ER (as confirmed through repeated rounds of IHC, Supplementary Table S2). The ability to detect alternative tumor biomarkers (CD44, in the case shown in Fig. 4A–C) is essential to detect carcinoma in these situations where a previously expressed biomarker is no longer present, either due to neoadjuvant treatment or natural disease progression.

Figure 4.

The multiplexed imaging of multiple biomarkers, enabled by REMI, improves detection sensitivity for malignancies with molecular phenotypes that are spatially and/or temporally varying. A, Photograph of a human breast specimen with IDC that was ER-positive on biopsy but ER-negative on lumpectomy, presumably as a result of neoadjuvant endocrine therapy. B, REMI results. Scale bars, 5 mm. The color bar indicates NP ratios. C, Validation data: H&E and IHC. The specimen is positive for CD44 and negative for HER2 (cytoplasmic staining only), ER, and EGFR. See text for details on how the IHC results are scored based on standard-of-care methods. Scale bars, 50 μm.

Figure 4.

The multiplexed imaging of multiple biomarkers, enabled by REMI, improves detection sensitivity for malignancies with molecular phenotypes that are spatially and/or temporally varying. A, Photograph of a human breast specimen with IDC that was ER-positive on biopsy but ER-negative on lumpectomy, presumably as a result of neoadjuvant endocrine therapy. B, REMI results. Scale bars, 5 mm. The color bar indicates NP ratios. C, Validation data: H&E and IHC. The specimen is positive for CD44 and negative for HER2 (cytoplasmic staining only), ER, and EGFR. See text for details on how the IHC results are scored based on standard-of-care methods. Scale bars, 50 μm.

Close modal

Sensitivity and specificity of REMI for the detection of individual biomarkers and carcinoma

Prior to assessing the overall sensitivity and specificity of REMI for carcinoma detection, an ROC analysis was first performed for the detection of biomarker overexpression (for each of the four biomarkers), with IHC as the gold standard. REMI images and corresponding IHC images were uniformly divided into larger regions of interest (ROI) of 2 mm × 2 mm (a 4 × 4-pixel bin) to improve the correlative analysis between REMI and IHC (Fig. 5A). This was done because it is technically challenging to cut an en face histology section that approximates a large tissue surface, which results in imperfect spatial coregistration between REMI images and corresponding IHC images on a pixel-to-pixel basis. Using ROIs rather than pixels, specimens, or patients is also a standard method for quantitative analysis of digitized H&E and IHC slides (48, 49). Imperfect coregistration was only found in 8 specimens that contained mixed tumor and benign regions (Fig. 5A). To ensure accurate ROC analyses, a total of 74 ROIs that contained a mixture of both tumor and benign tissues were excluded to minimize coregistration issues (i.e., only pure tumor or pure benign ROIs were used for ROC analyses). Each ROI was assigned a “gold-standard” diagnostic result of 0 (negative) or 1 (positive) for the overexpression of each cell-surface biomarker based upon the consensus interpretation of IHC slides by two pathologists, N.P. Reder and S.M. Dintzis (Fig. 5A). The average NP ratio from each ROI in an REMI image was calculated, and its value was compared with a threshold value to determine if the NP ratio was indicative of overexpression of a biomarker (ratio > threshold). By varying this threshold value, it was possible to construct an ROC plot that displays the tradeoff between sensitivity and specificity for detecting biomarker overexpression. Figure 5B shows the ROC curves for each biomarker (based on 2106 ROIs). REMI achieves a sensitivity of between 90.0% and 93.7% to detect the elevated expression of HER2, EGFR, or CD44, when a high specificity (>90%) is enforced. A post hoc power analysis of our ROC curve results showed that we had a power of 0.99 for detecting a significant difference from the null for HER2, EGFR, and CD44, and a power of 0.36 for ER at an alpha value of 0.05. The sensitivity for detecting mER is low (if >90% specificity is enforced) due to the limited expression of ER at cell surfaces, a topic that will be discussed in greater depth later.

Figure 5.

ROC analysis for the identification of biomarker overexpression by REMI. A, Coregistration of IHC and REMI data. B, ROC curves for the biomarkers EGFR, HER2, ER, and CD44. The ratio of targeted NPs vs. isotype-NPs from each of 2106 ROIs was used to generate the ROC curves, in which IHC was used as the gold standard for detection of overexpressed biomarkers. For the detection of EGFR overexpression, the AUC = 0.96 with a 95% confidence interval (CI) of 0.96–0.97. For the detection of HER2 overexpression, the AUC = 0.96 with a 95% CI of 0.95–0.97. For the detection of mER overexpression, the AUC = 0.67 with a 95% CI of 0.64–0.69. For the detection of CD44 overexpression, the AUC = 0.94 with a 95% CI of 0.93–0.95.

Figure 5.

ROC analysis for the identification of biomarker overexpression by REMI. A, Coregistration of IHC and REMI data. B, ROC curves for the biomarkers EGFR, HER2, ER, and CD44. The ratio of targeted NPs vs. isotype-NPs from each of 2106 ROIs was used to generate the ROC curves, in which IHC was used as the gold standard for detection of overexpressed biomarkers. For the detection of EGFR overexpression, the AUC = 0.96 with a 95% confidence interval (CI) of 0.96–0.97. For the detection of HER2 overexpression, the AUC = 0.96 with a 95% CI of 0.95–0.97. For the detection of mER overexpression, the AUC = 0.67 with a 95% CI of 0.64–0.69. For the detection of CD44 overexpression, the AUC = 0.94 with a 95% CI of 0.93–0.95.

Close modal

In this early-stage feasibility study, the overall sensitivity and specificity for the detection of breast carcinoma were calculated by assuming that a tissue region would be considered malignant if any one (or more) of the four candidate biomarkers was positively expressed in that region (Fig. 1). For tumor detection, H&E histology was used as a gold standard. Based on these assumptions, the overall sensitivity and specificity of REMI for breast carcinoma detection (based on ROIs) in this study were 89.3% and 92.1%, respectively (Supplementary Table S2). On a per-patient basis, REMI achieved 92.9% sensitivity (2 patients had ER+ tumors misdiagnosed as benign) and 89.7% specificity (3 patients had benign specimens misdiagnosed as HER2+ tumors). In the future, it may be possible to utilize machine-learning methods to more accurately identify tumor regions based on unique biomarker-expression signatures, rather than our simple algorithm that assumes that a tissue region is malignant if any of the biomarker targets is overexpressed.

REMI is less sensitive than histology to intracellular background staining

For IHC evaluation of cell-surface receptors such as HER2 and EGFR, only membranous staining is classified as positive, whereas cytoplasmic staining is ignored (50). This is a potential source of error in IHC, where pathologists must carefully differentiate between membranous and cytoplasmic expression. Interestingly, REMI of fresh tissue is intrinsically insensitive to cytoplasmic targets (Fig. 6A) because these fresh tissue surfaces are predominantly composed of intact cells in which cell membranes are preferentially exposed to the SERS NPs (Supplementary Fig. S8). Note that the large SERS NPs (120 nm) used in our REMI approach exhibit negligible internalization through cell membranes after 5 to 10 minutes of topical application (28, 31), and therefore do not have efficient access to intracellular proteins. This differs from thin tissue sections on a pathology slide, in which intracellular contents are preferentially exposed (and stained), but cell membranes are minimally sampled (Supplementary Fig. S8). The insensitivity of REMI to intracellular protein also explains the low sensitivity to detect ER expression (Fig. 5B), which is primarily a nuclear biomarker with weak cell-surface expression. Despite the low sensitivity of REMI to detect mER, the overall sensitivity for tumor detection is aided by the ability to image a multiplexed panel of four biomarkers.

Figure 6.

REMI can quantify the expression of cell-surface biomarkers but is intrinsically insensitive to intracellular (e.g., cytoplasmic and nuclear) expression. A challenging issue for pathologists, when assessing the expression of cell-surface biomarkers with IHC, is differentiating between intracellular and membranous staining and only using the latter as a basis for scoring the expression levels of cell-surface proteins. REMI stains and images fresh tissue surfaces that are composed of mostly intact cells, in which the large SERS NPs are not easily internalized. Therefore, REMI is intrinsically less sensitive to intracellular targets and provides reliable quantification of cell-surface biomarker expression, with good correlation to IHC scores (as determined by expert pathologists). A, Correlation between REMI and IHC results for cytoplasmic vs. cell-surface membrane targets (2106 ROIs in total). B, Examples of IHC images showing nonexistent or weak cytoplasmic staining (row 1), intermediate or strong cytoplasmic staining (row 2), and cell-membrane staining (rows 3 and 4). Scale bars, 50 μm. The box plots in A are all significantly different from each other (P < 0.001).

Figure 6.

REMI can quantify the expression of cell-surface biomarkers but is intrinsically insensitive to intracellular (e.g., cytoplasmic and nuclear) expression. A challenging issue for pathologists, when assessing the expression of cell-surface biomarkers with IHC, is differentiating between intracellular and membranous staining and only using the latter as a basis for scoring the expression levels of cell-surface proteins. REMI stains and images fresh tissue surfaces that are composed of mostly intact cells, in which the large SERS NPs are not easily internalized. Therefore, REMI is intrinsically less sensitive to intracellular targets and provides reliable quantification of cell-surface biomarker expression, with good correlation to IHC scores (as determined by expert pathologists). A, Correlation between REMI and IHC results for cytoplasmic vs. cell-surface membrane targets (2106 ROIs in total). B, Examples of IHC images showing nonexistent or weak cytoplasmic staining (row 1), intermediate or strong cytoplasmic staining (row 2), and cell-membrane staining (rows 3 and 4). Scale bars, 50 μm. The box plots in A are all significantly different from each other (P < 0.001).

Close modal

Through imaging experiments with fresh human breast tissues, we have demonstrated that REMI allows for the rapid (<15 minutes) quantification of multiple biomarkers for the purposes of detecting positive margins during lumpectomy procedures (Figs. 2–4; Supplementary Figs. S5–S7). Because the molecular profiles of breast carcinoma vary greatly between patients (e.g., Figs. 2–4; Supplementary Figs. S5 and S6) as well as within a single patient over time (e.g., Fig. 4), the ability to image various carcinoma biomarkers, enabled by REMI, is necessary to detect malignancies with high sensitivity (89.3%). In addition, REMI can achieve high detection specificity (92.1%), thereby allowing for the discrimination between benign and malignant lesions (e.g., UDH and IDC in Fig. 3). Note that the presence of hemoglobin (Figs. 3 and 4; Supplementary Fig. S6), surgical inks (Figs. 3 and 4; Supplementary Fig. S7), and/or fluorescent dyes do not interfere with the accurate measurement of SERS NPs due to the well-defined fingerprint spectra of SERS NPs and a robust demultiplexing algorithm (30).

As mentioned previously, a critical component of REMI is a ratiometric imaging strategy (33–35, 37, 38) that enables accurate and sensitive identification of biomarker overexpression without the confounding effects of nonspecific background signals (Supplementary Figs. S9A–S9E and S10A and S10B). As shown in Supplementary Figs. S9B and S10B, the imaging of the raw concentration (signal) of targeted NPs fails to differentiate between malignant and benign regions due to the misleading nonspecific accumulation of the NPs, a phenomenon that, to some degree, affects all molecular imaging approaches in which exogenous contrast agents are delivered either systemically or topically. For example, the nonspecific accumulation of NPs is heavily influenced by the mechanical properties of a tissue, such as porosity and interstitial pressure, which are often higher in benign tissues compared with dense tumors (Supplementary Fig. S9B; refs. 31, 33, 51). In addition, the imaging of raw NP concentrations (signals) suffers from other confounding effects such as variations in illumination power and detector working distance (for example, due to tissue-surface irregularities), whereas ratiometric imaging is insensitive to those effects (33). Several features make our REMI approach particularly well-suited for a ratiometric or “paired-agent” imaging strategy (35, 39). (i) All of the SERS NP flavors are identical in terms of geometry and surface properties (same silica coating), which allow them to exhibit identical nonspecific behavior. (ii) Multiplexed SERS NPs are excited at a single illumination wavelength (785 nm), thereby obviating wavelength-dependent effects and ensuring that all NP flavors are illuminated identically in terms of intensity and spot size (unlike multiplexed fluorescent agents, which are often excited at disparate wavelengths).

REMI possesses a number of advantages over conventional frozen-section and FFPE histopathology, but can also complement these methods because it is nondestructive (no sectioning is necessary) and has not been shown to interfere with downstream pathology (Figs. 2–4; Supplementary Figs. S5–S7; refs. 30, 31). By comprehensively imaging an entire surgical margin surface, REMI can overcome the sampling limitations associated with slide-based histology, in which tissues are sectioned in the vertical direction at periodic intervals with minimal sampling of the tissue surfaces (i.e., the inked margin). An interesting finding is that REMI is quite insensitive to intracellular (e.g., cytoplasmic and nuclear) targets (Fig. 6A), which can often confound the quantification of cell-surface receptors via microscopic visualization of IHC slides (Fig. 6B). This is due to the fact that intracellular proteins are minimally exposed at fresh (unsectioned) tissue surfaces as compared with thin tissue sections on histology slides (Supplementary Fig. S8), as well as the relatively large size of the SERS NPs (∼120 nm; Supplementary Fig. S3A), which are largely confined to the tissue surface and are not internalized by cells during the brief staining durations utilized in REMI.

There are several clear opportunities to further improve the REMI technique. Increasing the number of biomarker targets can potentially allow for improved sensitivity of REMI-based detection of molecularly heterogeneous tumors. For example, a cell-surface biomarker that could be considered in future REMI panels is MUC1, which has been reported to be expressed in up to 90% of breast tumors (52). Although a few studies have shown the feasibility of demultiplexing up to 10 flavors of untargeted SERS NPs (25, 53), the ability to accurately quantify larger panels of biomarkers (>5) with targeted NPs remains to be demonstrated. As an intraoperative imaging technique, the current implementation of REMI allows for accurate detection of positive margins with submillimeter spatial resolution under time-constrained intraoperative conditions (imaging requires <5 minutes). In the future, the imaging speed and resolution of REMI can be further improved by using brighter SERS NPs (e.g., resonant Raman NPs; ref. 54) and/or advanced detection schemes such as a spectral compression technique that our group has recently described (41). Finally, REMI may potentially be combined with wide-area fresh-tissue microscopy techniques, such as structured-illumination microscopy or light-sheet microscopy (55), to allow surgeons and pathologists to visualize biomarker expression patterns in the context of their tissue microenvironment (i.e., intraoperative wide-area immunofluorescence microscopy).

Translational impact and summary

At most medical institutions, at the final stages of lumpectomy, resected tissues or biopsy shavings at the edges of the resection cavity are marked (“inked”) at the surgical margin surfaces, and are submitted for histopathology, a process that takes days. In pathology, these tissue specimens are physically sectioned at approximately 0.5-cm intervals such that each histology slide displays a cross-section of the specimen from the inked surface down toward the direction of the original tumor core. Consequently, less than 1% of the total inked margin surface is sampled using this approach. A surface-imaging approach like REMI is ideally suited for guiding lumpectomy procedures, especially because it enables comprehensive nondestructive imaging of an entire surgical margin surface without the undersampling of conventional postoperative histology. For in situ breast carcinomas, close margins (e.g., within 1 to 2 mm of the inked margin) continue to be of concern for most institutions, and REMI should be supplemented with standard-of-care postoperative pathology to identify such cases. However, it should be noted that the historic preference to re-excise patients with close margins (especially for DCIS) may have arisen because of the sampling limitations of conventional pathology, and that this safety margin may prove to be unnecessary in future outcomes-based clinical studies in which a comprehensive surface-imaging approach like REMI is utilized.

In summary, multiplexed molecular imaging, as enabled by REMI, has value for the detection of tumors that exhibit significant interpatient and intrapatient heterogeneity. Through comprehensive molecular imaging of the surfaces of resected breast tissues, in which four biomarkers are simultaneously visualized and quantified, we have demonstrated that REMI has the feasibility to provide rapid (<15 minutes) detection of positive surgical margins with 89.3% sensitivity and 92.1% specificity. There is potential for REMI to guide a variety of other tumor-resection procedures in which large surface areas should be inspected for the presence of residual tumor, such as radical prostatectomies and the resection of head and neck carcinomas.

Jonathan T.C. Liu has an issued patent: “Raman imaging devices and methods of molecular imaging.” No potential conflicts of interest were disclosed by the other authors.

Conception and design: Y. “Winston” Wang, J.T.C. Liu

Development of methodology: Y. “Winston” Wang, M.A. Wall, J.T.C. Liu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. “Winston” Wang, J.T.C. Liu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. “Winston” Wang, N.P. Reder, S. Kang, J.T.C. Liu

Writing, review, and/or revision of the manuscript: Y. “Winston” Wang, N.P. Reder, A.K. Glaser, S.H. Javid, S.M. Dintzis, J.T.C. Liu

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. “Winston” Wang, Q. Yang, J.T.C. Liu

Study supervision: J.T.C. Liu

The authors acknowledge the technical assistance of the Histology and Imaging Core Facility at the University of Washington.

The authors acknowledge support from the NIH/NIBIBR21 EB015016 (J.T.C. Liu), the NIH/NCIR21 CA215561 (J.T.C. Liu), the Department of Mechanical Engineering at the University of Washington, and the department of education GAANN fellowship program (S. Kang). The Northwest BioTrust (NWBT) is supported, in part, by the NIH (P30-CA015704).

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.

1.
DeSantis
C
,
Ma
J
,
Bryan
L
,
Jemal
A
. 
Breast cancer statistics, 2013
.
CA Cancer J Clin
2014
;
64
:
52
62
.
2.
Jacobs
L
. 
Positive margins: the challenge continues for breast surgeons
.
Ann Surg Oncol
2008
;
15
:
1271
2
.
3.
Moran
MS
,
Schnitt
SJ
,
Giuliano
AE
,
Harris
JR
,
Khan
SA
,
Horton
J
, et al
Society of Surgical Oncology–American Society for radiation oncology consensus guideline on margins for breast-conserving surgery with whole-breast irradiation in stages I and II invasive breast cancer
.
J Clin Oncol
2014
;
32
:
1507
15
.
4.
Jorns
JM
,
Daignault
S
,
Sabel
MS
,
Wu
AJ
. 
Is intraoperative frozen section analysis of reexcision specimens of value in preventing reoperation in breast-conserving therapy?
Am J Clin Pathol
2014
;
142
:
601
8
.
5.
Cendan
JC
,
Coco
D
,
Copeland
EM
 3rd
. 
Accuracy of intraoperative frozen-section analysis of breast cancer lumpectomy-bed margins
.
J Am Coll Surg
2005
;
201
:
194
8
.
6.
Abeytunge
S
,
Larson
B
,
Peterson
G
,
Morrow
M
,
Rajadhyaksha
M
,
Murray
MP
. 
Evaluation of breast tissue with confocal strip-mosaicking microscopy: a test approach emulating pathology-like examination
.
J Biomed Opt
2017
;
22
:
34002
.
7.
Tao
YK
,
Shen
D
,
Sheikine
Y
,
Ahsen
OO
,
Wang
HH
,
Schmolze
DB
, et al
Assessment of breast pathologies using nonlinear microscopy
.
Proc Natl Acad Sci U S A
2014
;
111
:
15304
9
.
8.
Tu
H
,
Liu
Y
,
Turchinovich
D
,
Marjanovic
M
,
Lyngsø
JK
,
Lægsgaard
J
, et al
Stain-free histopathology by programmable supercontinuum pulses
.
Nat Photon
2016
;
10
:
534
40
.
9.
Nguyen
FT
,
Zysk
AM
,
Chaney
EJ
,
Kotynek
JG
,
Oliphant
UJ
,
Bellafiore
FJ
, et al
Intraoperative evaluation of breast tumor margins with optical coherence tomography
.
Cancer Res
2009
;
69
:
8790
6
.
10.
Assayag
O
,
Antoine
M
,
Sigal-Zafrani
B
,
Riben
M
,
Harms
F
,
Burcheri
A
, et al
Large field, high resolution full-field optical coherence tomography: a pre-clinical study of human breast tissue and cancer assessment
.
Technol Cancer Res Treat
2014
;
13
:
455
68
.
11.
Kennedy
BF
,
McLaughlin
RA
,
Kennedy
KM
,
Chin
L
,
Wijesinghe
P
,
Curatolo
A
, et al
Investigation of optical coherence microelastography as a method to visualize cancers in human breast tissue
.
Cancer Res
2015
;
75
:
3236
45
.
12.
Sharma
V
,
Shivalingaiah
S
,
Peng
Y
,
Euhus
D
,
Gryczynski
Z
,
Liu
H
. 
Auto-fluorescence lifetime and light reflectance spectroscopy for breast cancer diagnosis: potential tools for intraoperative margin detection
.
Biomed Opt Express
2012
;
3
:
1825
40
.
13.
Brown
JQ
,
Bydlon
TM
,
Kennedy
SA
,
Caldwell
ML
,
Gallagher
JE
,
Junker
M
, et al
Optical spectral surveillance of breast tissue landscapes for detection of residual disease in breast tumor margins
.
PLoS One
2013
;
8
:
e69906
.
14.
Keller
MD
,
Vargis
E
,
de Matos Granja
N
,
Wilson
RH
,
Mycek
MA
,
Kelley
MC
, et al
Development of a spatially offset Raman spectroscopy probe for breast tumor surgical margin evaluation
.
J Biomed Opt
2011
;
16
:
077006
.
15.
Liu
CH
,
Zhou
Y
,
Sun
Y
,
Li
JY
,
Zhou
LX
,
Boydston-White
S
, et al
Resonance Raman and Raman spectroscopy for breast cancer detection
.
Technol Cancer Res Treat
2013
;
12
:
371
82
.
16.
Kong
K
,
Rowlands
CJ
,
Varma
S
,
Perkins
W
,
Leach
IH
,
Koloydenko
AA
, et al
Diagnosis of tumors during tissue-conserving surgery with integrated autofluorescence and Raman scattering microscopy
.
Proc Natl Acad Sci U S A
2013
;
110
:
15189
94
.
17.
Sadeghipour
N
,
Davis
SC
,
Tichauer
KM
. 
Generalized paired-agent kinetic model for in vivo quantification of cancer cell-surface receptors under receptor saturation conditions
.
Phys Med Biol
2016
;
62
:
394
414
.
18.
Collins
FS
,
Varmus
H
. 
A new initiative on precision medicine
.
N Engl J Med
2015
;
372
:
793
5
.
19.
Weissleder
R
,
Ross
BD
,
Rehemtulla
A
,
Gambhir
SS
. 
Molecular imaging: principles and practice
.
Shelton, Connecticut: People's Medical Publishing House
, 
2010
.
20.
Zardavas
D
,
Irrthum
A
,
Swanton
C
,
Piccart
M
. 
Clinical management of breast cancer heterogeneity
.
Nat Rev Clin Oncol
2015
;
12
:
381
94
.
21.
Marusyk
A
,
Polyak
K
. 
Tumor heterogeneity: causes and consequences
.
Biochim Biophys Acta
2010
;
1805
:
105
17
.
22.
Yezhelyev
MV
,
Al-Hajj
A
,
Morris
C
,
Marcus
AI
,
Liu
T
,
Lewis
M
, et al
In situ molecular profiling of breast cancer biomarkers with multicolor quantum dots
.
Adv Mater
2007
;
19
:
3146
51
.
23.
Wang
Y
,
Yan
B
,
Chen
L
. 
SERS tags: novel optical nanoprobes for bioanalysis
.
Chem Rev
2013
;
113
:
1391
428
.
24.
McClatchy
DM
 3rd
,
Krishnaswamy
V
,
Kanick
SC
,
Elliott
JT
,
Wells
WA
,
Barth
RJ
 Jr
, et al
Molecular dyes used for surgical specimen margin orientation allow for intraoperative optical assessment during breast conserving surgery
.
J Biomed Opt
2015
;
20
:
040504
.
25.
Zavaleta
CL
,
Garai
E
,
Liu
JT
,
Sensarn
S
,
Mandella
MJ
,
Van de Sompel
D
, et al
A Raman-based endoscopic strategy for multiplexed molecular imaging
.
Proc Natl Acad Sci U S A
2013
;
110
:
E2288
97
.
26.
Jokerst
JV
,
Cole
AJ
,
Van de Sompel
D
,
Gambhir
SS
. 
Gold nanorods for ovarian cancer detection with photoacoustic imaging and resection guidance via Raman imaging in living mice
.
ACS Nano
2012
;
6
:
10366
77
.
27.
Bohndiek
SE
,
Wagadarikar
A
,
Zavaleta
CL
,
Van de Sompel
D
,
Garai
E
,
Jokerst
JV
, et al
A small animal Raman instrument for rapid, wide-area, spectroscopic imaging
.
Proc Natl Acad Sci U S A
2013
;
110
:
12408
13
.
28.
Wang
YW
,
Doerksen
JD
,
Kang
S
,
Walsh
D
,
Yang
Q
,
Hong
D
, et al
Multiplexed molecular imaging of fresh tissue surfaces enabled by convection-enhanced topical staining with SERS-coded nanoparticles
.
Small (Weinheim an der Bergstrasse, Germany)
2016
;
12
:
5612
21
.
29.
Jokerst
JV
,
Miao
Z
,
Zavaleta
C
,
Cheng
Z
,
Gambhir
SS
. 
Affibody-functionalized gold-silica nanoparticles for Raman molecular imaging of the epidermal growth factor receptor
.
Small (Weinheim an der Bergstrasse, Germany)
2011
;
7
:
625
33
.
30.
Wang
Y
,
Kang
S
,
Khan
A
,
Ruttner
G
,
Leigh
SY
,
Murray
M
, et al
Quantitative molecular phenotyping with topically applied SERS nanoparticles for intraoperative guidance of breast cancer lumpectomy
.
Sci Rep
2016
;
6
:
21242
.
31.
Wang
Y
,
Kang
S
,
Doerksen
J
,
Glaser
A
,
Liu
J
. 
Surgical guidance via multiplexed molecular imaging of fresh tissues labeled with SERS-coded nanoparticles
.
IEEE J Sel Top Quantum Electron
2016
;
22
:
154
64
.
32.
Wang
YW
,
Kang
S
,
Khan
A
,
Bao
PQ
,
Liu
JTC
. 
In vivo multiplexed molecular imaging of esophageal cancer via spectral endoscopy of topically applied SERS nanoparticles
.
Biomed Opt Express
2015
;
6
:
3714
23
.
33.
Wang
YW
,
Khan
A
,
Som
M
,
Wang
D
,
Chen
Y
,
Leigh
SY
, et al
Rapid ratiometric biomarker detection with topically applied SERS nanoparticles
.
Technology (Singap World Sci)
2014
;
2
:
118
32
.
34.
Liu
JTC
,
Helms
MW
,
Mandella
MJ
,
Crawford
JM
,
Kino
GS
,
Contag
CH
. 
Quantifying cell-surface biomarker expression in thick tissues with ratiometric three-dimensional microscopy
.
Biophys J
2009
;
96
:
2405
14
.
35.
Tichauer
KM
,
Samkoe
KS
,
Gunn
JR
,
Kanick
SC
,
Hoopes
PJ
,
Barth
RJ
, et al
Microscopic lymph node tumor burden quantified by macroscopic dual-tracer molecular imaging
.
Nat Med
2014
;
20
:
1348
53
.
36.
Wang
YW
,
Khan
A
,
Leigh
SY
,
Wang
D
,
Chen
Y
,
Meza
D
, et al
Comprehensive spectral endoscopy of topically applied SERS nanoparticles in the rat esophagus
.
Biomed Opt Express
2014
;
5
:
2883
95
.
37.
Mallia
RJ
,
McVeigh
PZ
,
Fisher
CJ
,
Veilleux
I
,
Wilson
BC
. 
Wide-field multiplexed imaging of EGFR-targeted cancers using topical application of NIR SERS nanoprobes
.
Nanomedicine (Lond)
2015
;
10
:
89
101
.
38.
Oseledchyk
A
,
Andreou
C
,
Wall
MA
,
Kircher
MF
. 
Folate-targeted surface-enhanced resonance raman scattering nanoprobe ratiometry for detection of microscopic ovarian cancer
.
ACS Nano
2016
;
11
:
1488
97
.
39.
Tichauer
KM
,
Wang
Y
,
Pogue
BW
,
Liu
JTC
. 
Quantitative in vivo cell-surface receptor imaging in oncology: kinetic modeling and paired-agent principles from nuclear medicine and optical imaging
.
Phys Med Biol
2015
;
60
:
R239
.
40.
Ntziachristos
V
. 
Going deeper than microscopy: the optical imaging frontier in biology
.
Nat Meth
2010
;
7
:
603
14
.
41.
Kang
S
,
Wang
Y
,
Reder
NP
,
Liu
JTC
. 
Multiplexed molecular imaging of biomarker-targeted SERS nanoparticles on fresh tissue specimens with channel-compressed spectrometry
.
PLoS One
2016
;
11
:
e0163473
.
42.
Leigh
SY
,
Som
M
,
Liu
JTC
. 
Method for assessing the reliability of molecular diagnostics based on multiplexed SERS-coded nanoparticles
.
PLoS One
2013
;
8
:
e62084
.
43.
Cserni
G
. 
A model for determining the optimum histology of sentinel lymph nodes in breast cancer
.
J Clin Pathol
2004
;
57
:
467
71
.
44.
Hammond
MEH
,
Hayes
DF
,
Dowsett
M
,
Allred
DC
,
Hagerty
KL
,
Badve
S
, et al
American Society of Clinical Oncology/College of American Pathologists guideline recommendations for immunohistochemical testing of estrogen and progesterone receptors in breast cancer
.
J Clin Oncol
2010
;
28
:
2784
95
.
45.
Wolff
AC
,
Hammond
MEH
,
Hicks
DG
,
Dowsett
M
,
McShane
LM
,
Allison
KH
, et al
Recommendations for human epidermal growth factor receptor 2 testing in breast cancer: American Society of Clinical Oncology/College of American Pathologists Clinical Practice Guideline Update
.
J Clin Oncol
2013
;
31
:
3997
4013
.
46.
Honeth
G
,
Bendahl
P-O
,
Ringnér
M
,
Saal
LH
,
Gruvberger-Saal
SK
,
Lövgren
K
, et al
The CD44+/CD24-phenotype is enriched in basal-like breast tumors
.
Breast Cancer Res
2008
;
10
:
1
12
.
47.
Nielsen
TO
,
Hsu
FD
,
Jensen
K
,
Cheang
M
,
Karaca
G
,
Hu
Z
, et al
Immunohistochemical and clinical characterization of the basal-like subtype of invasive breast carcinoma
.
Clin Cancer Res
2004
;
10
:
5367
74
.
48.
Kothari
S
,
Phan
JH
,
Stokes
TH
,
Wang
MD
. 
Pathology imaging informatics for quantitative analysis of whole-slide images
.
J Am Med Inform Assoc
2013
;
20
:
1099
108
.
49.
Metzger
GJ
,
Dankbar
SC
,
Henriksen
J
,
Rizzardi
AE
,
Rosener
NK
,
Schmechel
SC
. 
Development of multigene expression signature maps at the protein level from digitized immunohistochemistry slides
.
PLoS One
2012
;
7
:
e33520
.
50.
Taylor
SL
,
Platt-Higgins
A
,
Rudland
PS
,
Winstanley
JH
,
Barraclough
R
. 
Cytoplasmic staining of c-erbB-2 is not associated with the presence of detectable c-erbB-2 mRNA in breast cancer specimens
.
Int J Cancer
1998
;
76
:
459
63
.
51.
Tichauer
KM
,
Samkoe
KS
,
Sexton
KJ
,
Gunn
JR
,
Hasan
T
,
Pogue
BW
. 
Improved tumor contrast achieved by single time point dual-reporter fluorescence imaging
.
J Biomed Opt
2012
;
17
:
066001
.
52.
Regimbald
LH
,
Pilarski
LM
,
Longenecker
BM
,
Reddish
MA
,
Zimmermann
G
,
Hugh
JC
. 
The breast mucin MUC1 as a novel adhesion ligand for endothelial intercellular adhesion molecule 1 in breast cancer
.
Cancer Res
1996
;
56
:
4244
9
.
53.
Zavaleta
CL
,
Smith
BR
,
Walton
I
,
Doering
W
,
Davis
G
,
Shojaei
B
, et al
Multiplexed imaging of surface enhanced Raman scattering nanotags in living mice using noninvasive Raman spectroscopy
.
Proc Natl Acad Sci U S A
2009
;
106
:
13511
6
.
54.
Harmsen
S
,
Huang
R
,
Wall
MA
,
Karabeber
H
,
Samii
JM
,
Spaliviero
M
, et al
Surface-enhanced resonance raman scattering nanostars for high precision cancer imaging
.
Sci Transl Med
2015
;
7
:
271ra7
.
55.
Wang
M
,
Kimbrell
HZ
,
Sholl
AB
,
Tulman
DB
,
Elfer
KN
,
Schlichenmeyer
TC
, et al
High-resolution rapid diagnostic imaging of whole prostate biopsies using video-rate fluorescence structured illumination microscopy
.
Cancer Res
2015
;
75
:
4032
41
.

Supplementary data