In Vivo Molecular Imaging to Diagnose and Subtype Tumors through Receptor-Targeted Optically Labeled Monoclonal Antibodies^,

Abstract

Molecular imaging of cell surface receptors can potentially diagnose tumors based on their distinct expression profiles. Using multifilter spectrally resolved optical imaging with three fluorescently labeled antibodies, we simultaneously imaged three different cell surface receptors to distinguish tumor types noninvasively. We selected tumors overexpressing different subtypes of EGFR receptor: HER-1 (A431) and HER-2 (NIH3T3/HER2⁺), or interleukin-2 receptor α-subunit receptor (IL-2Rα; SP2/Tac). After tumor establishment, a cocktail of three fluorescently labeled monoclonal antibodies was injected: cetuximab-Cy5 (targetingHER-1), trastuzumab-Cy7(HER-2),anddaclizumab-AlexaFluor-700 (IL-2Ra). Optical fluorescence imaging was performed after 24 hours with both a red filter set and three successive filter sets (yellow, red, and deep red). Spectrally resolved imaging of 10 mice clearly distinguished A431, NIH3T3/HER2⁺, and SP2-Tac tumors based on their distinct optical spectra. Three-filter sets significantly increased the signal-to-background ratio compared to a single-filter set by reducing the background signal, thus significantly improving the differentiation of each of the receptors targeted (P < .022). In conclusion, following multifilter spectrally resolved imaging, different tumor types can be simultaneously distinguished and diagnosed in vivo. Multiple filter sets increase the signal-to-noise ratio by substantially reducing the background signal, and may allow more optical dyes to be resolved within the narrow limits of the near-infrared spectrum.

Introduction

As new imaging modalities are developed and existing imaging techniques continue to improve their sensitivity and spatial resolution, imaging at the cellular level has become a real possibility. It is hoped that these molecular imaging techniques will play a prominent role in the field of oncological imaging. Molecular imaging has a potential clinical application not only for detecting tumors but also for subtyping them based on their distinct receptor expression. As with immunohistochemistry, it is unlikely that a diagnosis can be made based on a single cell surface receptor and thus, multiple cell surface receptors must be labeled. To successfully achieve this in vivo immunohistochemistry, the imaging modality chosen needs to have sufficient sensitivity to detect the relatively low number of receptors expressed on the tumor surface as well as the ability to distinguish one receptor type from another. Of the current clinically available imaging techniques, only positron emission tomography and single-photon emission computed tomography have sufficient sensitivity; however, it is difficult to label and image each surface receptor with a characteristic radionuclide distinguishable by its emission energy with gamma scintigraphy. Optical imaging not only offers sufficient sensitivity, with the added advantages of increased spatial resolution and absence of ionizing radiation but also allows multicolor imaging to be performed based on the use of fluorophores with differing emission wavelengths. However, a limitation of optical imaging is its reduced depth penetration compared with scintigraphy and magnetic resonance imaging. Optical imaging can depict tumors anywhere in the body of mice and rats [1]; however, clinically, its application may be limited to the imaging of either relatively superficial tumors (e.g., breast cancer) or those accessible using the endoscope (e.g., lung, gastrointestinal tract, and abdominal cavity).

We have previously developed a spectrally resolved optical imaging technique with the ability to accurately diagnose and differentiate tumors based on either human epidermal growth factor receptor type 1 (HER-1) or human epidermal growth factor receptor type 2 (HER-2) expression [2]. This technique used a mixture of optical agents linked to monoclonal antibodies (mAbs) that directly targeted the epidermal growth factor receptors (EGFRs) expressed by different tumor cell lines. Spectrally resolved optical imaging with mAbs was shown to have certain advantages over radiolabeled antibodies imaged with scintigraphy; that is, the ability to identify tumors at earlier time points, simultaneously differentiate tumor types, partially reduce background signal, and avoid exposure to radiation.

The EGFR family consists of four subtypes of tyrosine kinase receptors. In normal circumstances, these receptors play a vital regulatory role in normal cellular signaling and development; however, mutations can lead to neoplasia. Overexpression of EGFRs has been shown to correlate with poor prognosis [3,4]. Antibodies to these receptors (cetuximab for HER-1 and trastuzumab for HER-2) have been used clinically in selected patients. Because these agents successfully target cancers, it is expected that, at substantially lower doses, they could be also be used for molecular imaging to select treatable patients with appropriate agents.

Whereas two different antibodies can be useful in characterizing some tumors, it is likely that most tumors will require more than two antibody types to be fully characterized. At the same time, it is important to maximize the depth penetration of the fluorophore to be able to detect a signal throughout a tumor. Maximal depth penetration is best achieved with near-infrared (NIR) dyes that emit in the light range of 650 to 900 nm. At this wavelength, hemoglobin, muscle, and fat are least efficient in absorbing light [5]; furthermore, the autoflourescence (background) signal from the skin is also minimal [6]. The need to use NIR dyes means that only a limited number of separate optical probes can be simultaneously resolved based on their individual spectra to avoid overlap. Organic dyes have distinct narrow excitation spectra close to their respective broad peak emission spectra, and successful resolution of more than three dyes is technically challenging. Only nonorganic quantum dots and the Keima series of fluorescent proteins yield multiple emission peaks after a single excitation and offer the potential to resolve greater numbers of optical probes. However, in vivo imaging with the Keima proteins requires gene transfection to induce expression using gene therapy with viral or nonviral vectors [7,8], and is yet to be successfully applied to clinical practice, and quantum dots bring with them the potential for significant toxicity based on their Cd-Se cores, which will make their translation into the clinic challenging [9,10].

In theory, the subtyping of selected receptors based on markers of tumor aggressiveness could be used to guide treatment and determine prognosis. Here, we describe the use of three optically labeled mAbs administered as an intravenous cocktail in mice implanted with three different xenografts. We selected tumor cell lines expressing two different subtypes of the same family of receptors, EGFRs, i.e., HER-1 (A431 cells) and HER-2 (NIH3T3/HER2⁺ cells), and a tumor cell line positive for a separate family of receptors, i.e., the SP2/Tac cell line, expressing the interleukin-2 receptor alpha subunit (IL-2Rα: Tac) receptor. Imaging was performed using the commercially available, Food and Drug Administration (FDA)-approved mAbs cetuximab, trastuzumab, and daclizumab, which specifically target the HER-1, HER-2, and IL-2Rα (Tac) receptors, respectively.

Materials and Methods

Tumor Cells

Four established cell lines were used: A431, NIH3T3, SP2, and LS174T. A431 is a human epidermoid carcinoma cell line overexpressing the HER-1 receptor, but minimally expressing the HER-2 receptor. NIH3T3 are murine fibroblast-like cells that express neither HER-1 nor HER-2 receptors. The NIH3T3 cells were transfected with HER-2 genes (NIH3T3/HER2⁺) to overexpress HER-2 receptors [5]. SP2 are murine myeloma cells. These cells were transfected with the IL-2Rα (Tac) gene and thus are Tac-(IL-2Rα) positive (SP2/Tac). LS174T is a human colon carcinoma cell line that is weakly positive for HER-2 and serves as a control cell line.

A431 cell lines were cultured in Dulbecco's modified Eagle's medium (Gibco, Gaithersburg, MD) containing 10% fetal bovine serum (Gibco). NIH3T3/HER2⁺ cells were grown in RPMI 1640 medium (Gibco) containing 10% fetal bovine serum, 0.03% l-glutamine at 37°C, 100 U/ml penicillin, and 100 µg/ml streptomycin in 5% CO₂.

Synthesis of Antibody-Conjugated NIR Fluorescence Dyes Antibodies and fluorescent dyes.

Erbitux, an FDA-approved chimeric human-murine antibody specific for HER-1 receptors, generically known as cetuximab, was purchased from Bristol-Myers Squibb Company (Princeton, NJ). Herceptin, an FDA-approved humanized anti-HER-2 antibody, generically known as trastuzumab, which has a complimentary determination region (CDR) against HER-2 grafted on a human IgG1 framework, was purchased from Genentech Inc. (South San Francisco, CA). Zenapax, a murine-human chimerized mAb to the IL-2Rα (Tac) receptor of T-cells, generically known as daclizumab, was purchased from Hoffmann-La Roche Inc. (Nutley, NJ). AlexaFluor700 NHS ester, used for the imaging study, and Rhodamine Green (RhodG) NHS ester, used in the initial flow cytometry and fluorescent microscopy experiments, were purchased from Invitrogen Corporation (Carlsbad, CA). Cy5 NHS ester and Cy7 NHS ester were purchased from GE Healthcare (Piscataway, NJ).

Conjugation process.

At room temperature, 500 µg (3.3 nmol) of cetuximab, trastuzumab, or daclizumab in Na₂HPO₄ was incubated with 20 nmol of RhodG, 40 nmol of Cy5 NHS ester, 30 nmol of Cy7 NHS ester, or 30 nmol of AlexaFluor700 NHS ester dissolved in 5 mM DMSO, respectively, at pH 8.5 for 15 minutes at room temperature. The mixture was purified with a Sephadex G50 column (PD-10; GE Healthcare). Cetuximab-conjugated RhodG, trastuzumab-conjugated RhodG, cetuximab-conjugated Cy5, daclizumab-conjugated AlexaFluor700, and trastuzumab-conjugated Cy7 samples (cetuximab-RhodG, trastuzumab-RhodG, cetuximab-Cy5 and trastuzumab-Cy7, respectively) were kept at 4°C in the refrigerator as stock solutions. The protein concentrations of cetuximab-RhodG, trastuzumab-RhodG, cetuximab-Cy5 and trastuzumab-Cy7 samples were determined with Coomassie Plus protein assay kit (Pierce Biotechnology, Rockford, IL) by measuring the absorption at 595 nm with a UV-Vis system (8453 Value UV-Visible Value System; Agilent Technologies, Santa Clara, CA) using standard solutions of known concentrations of cetuximab, trastuzumab, or daclizumab (100, 200, and 400 µg/ml). The concentrations of RhodG, Cy5, and Cy7 were then measured by absorption at 503, 675, and 747 nm, respectively, with the UV-Vis system to confirm the number of fluorophore molecules conjugated with each cetuximab, trastuzumab, or daclizumab molecule. The number of fluorophore molecules per cetuximab, trastuzumab, or daclizumab was adjusted to approximately 2.

Flow cytometry study.

One-color flow cytometry studies were performed to assess the specificity of daclizumab-Rhodamine Green to both SP2/Tac and SP2/0 cells. Rhodamine Green was used as a fluorescent dye because flow cytometry is not efficient in the NIR range. In total, 1 x 10⁵ cells were placed on a 12-chamber well and incubated for 12 hours. Daclizumab-Rhodamine Green was added to the medium (1 µg/ml) and the cells were incubated for 72 hours. At this time point, flow cytometry was performed, employing the argon ion 488 nm laser for excitation. Signals from cells were collected using a 530/30 nm band-pass filter. Cells were analyzed in a FACScan cytometer (Becton Dickinson, Franklin Lakes, NJ) and all data were analyzed using CellQuest software (Becton Dickinson). The fluorescence capability of each fluorophore was referred to as the mean fluorescence index. Fluorescence-assisted flow cytometry studies following all other combinations of A431 and NIH3T3/HER2⁺ cells with cetuximab-Rhodamine Green, trastuzumab-Rhodamine Green, and daclizumab-Rhodamine Green have previously been reported [2,11].

Fluorescence microscopy.

In total, 1 x 10⁴ SP2/TAC cells and SP2/0 cells (negative control cells, not expressing the IL-2Rα receptor) were plated on a coverglass-bottom culture well and incubated for 16 hours. A conjugate of the antibody daclizumab linked to Rhodamine Green was added to each of the two cultured media (30 µg/ml). The cells were incubated and removed at the following time points: 1, 4, 8, 24, 48, and 72 hours. Following removal, cells were washed once with PBS and fluorescence microscopy was immediately performed using an Olympus BX61 microscope (Olympus America Inc., Melville, NY) equipped with the following filter settings: excitation wavelength = 470 to 490 nm and emission wavelength = 515 to 550 nm. Transmitted light differential interference contrast images were also acquired. Fluorescence microscopy studies following incubation of A431 cells with cetuximab-Cy5.5 and NIH3T3/HER2⁺ cells with trastuzumab-Cy7 have previously been demonstrated to be efficacious [2]. Images were taken with a 2-second exposure.

Animal Model

All procedures were carried out in compliance with the Guide for the Care and Use of Laboratory Animal Resources (1996), National Research Council, and approved by the local Animal Care and Use Committee. One from each of the tumors A431, NIH3T3/HER2⁺, SP2/TAC, and LS174T was established. For each cell line, 2 million (200 µl) cells, suspended in PBS, were injected subcutaneously onto the back of the mice, following intraperitoneal administration of ketamine (90 mg/kg; Ketaset; Aveco, Inc., Fort Dodge, IA) with xylazine (9 mg/kg; AnaSed; Lloyd, Inc., Shenandoah, IA) as a general anesthetic. A431 cells were implanted on the right flank, NIH3T3/HER2⁺ cells on the left flank, SP2/TAC on the right shoulder, and LS174T on the left shoulder regions (Figure W1). Tumors were monitored for growth and imaging was undertaken when tumors reached an appropriate size (10–14 days after implantation). In total, 20 mice were injected with all three cell lines and, from this group, 10 mice that grew tumors of comparable (∼ 5 mm) sizes were selected for use in the optical imaging study.

In Vivo Spectral Fluorescence Imaging Study

Ten female nude mice (National Cancer Institute Animal Production Facility, Frederick, MD) were implanted with tumors (as described above). A 200-µl mixture containing 50 µl (concentration of 1 µg/µl) of each of the three antibody-optical agent conjugates and 50 µl of PBS was prepared. The antibody cocktail was injected intravenously through the mouse tail vein 24 hours before optical imaging. The dosing strategy and the decision to image at 24 hours were based on the success of a previous work [2]. Before imaging, mice were anesthetized with intraperitoneally administered 10% pentobarbital sodium (Nembutal; Abbott Laboratories, Abbott Park, IL) with 0.1% scopolamine butylbromide (Buscopan Injection; Nippon Boehlringer Ingelheim Co., Tokyo, Japan). Spectral fluorescence images were obtained using the Maestro In Vivo Imaging System (CRi Inc., Woburn, MA).

In all cases, optical image sets were acquired with both a red filter set and three successive filter sets, i.e., yellow, red, and deep red, for acquisition of one complete image cube. For the yellow light filter, a band-pass filter from 575 to 605 nm and a long-pass filter of 645 nm were used for excitation and emission light, respectively. For the red filter set, these values were from 615 to 665 nm and 700 nm, respectively; for the deep red filter set, the values were from 671 to 705 nm and 750 nm, respectively (Figure 1). The tunable filter was automatically stepped in 10-nm increments from 650 to 950 nm for the combined filter sets, whereas the camera captured images at each wavelength interval with constant exposure. Spectral fluorescence images based on autofluorescence, Cy5, Cy7, and AlexaFluor700 spectra were obtained.

Figure 1.

Schematic representation of the multiple-filter acquisition technique. The optical dyes Cy5, AlexaFluor700, and Cy7 have peak emissions of 694, 719, and 776 nm, respectively. Three filter sets (yellow, red, and deep red) were used to acquire a single image cube over these wavelengths. For the yellow light filter, a band-pass filter from 575 to 605 nm (590/30) and a longpass filter of 645 nm were used for excitation and emission light, respectively. For the red filter set, these values were 615 to 665 (640/50) and 700 nm, respectively; for the deep red filter set, the values were 671 to 705 (688/34) and 750 nm, respectively.

Spectral libraries for Cy5 and Cy7 were imported and the respective spectra unmixed, using a commercial software (Maestro software; CRi Inc., Woburn, MA). Specifically, blood pool images from non-tumor-containing regions were unmixed from tumor-containing regions to reduce background contamination from the unbound optically labeled antibodies. Mice were sacrificed with carbon dioxide immediately after completion of imaging. Surgery was then performed to resect the tumors and enable ex vivo optical imaging, using the same Maestro settings.

Results

In Vitro Analysis

Flow cytometry studies were performed after 72 hours of incubation of daclizumab-Rhodamine Green and either SP2/Tac or SP2/0 cells. Daclizumab-Rhodamine Green showed a strong binding affinity to the IL-2Rα-overexpressing SP2/Tac cell, several orders of magnitude higher than the SP2/0 (IL-2Rα-negative) control cells (Figure 2). The SP2/0 cells showed only a minimal, nonspecific binding of the daclizumab-Rhodamine Green complex.

Figure 2.

FACS flow cytometry: Results for (A) SP2/Tac and (B) SP2/0 cell lines 72 hours after incubation with antibody linked to a fluorescent dye. (A) About 2 µg/ml daclizumab (antibody to IL-2Rα receptor) linked to Rhodamine Green, shows strong binding affinity to IL-2Rα-overexpressing SP2/Tac cells. (B) SP2/0 cells (negative control, not expressing IL-2Rα receptors) show only minimal, nonspecific binding of the daclizumab-Rhodamine Green.

Fluorescence microscopy was performed for SP2/Tac cells or for SP2/0 cells following their incubation with daclizumab-Rhodamine Green. SP2/Tac cells start to show binding of daclizumab-RhodG to the IL-2Rα receptors expressed on their cell surface as early as 1 hour post-incubation. With time, the complexes are gradually internalized into the cell. This is initially seen by 24 hours and becomes more apparent by the 48-hour time point when the daclizumab-Rhodamine Green complex can clearly be seen in the perinuclear region (Figure 3B). Control SP2/0 cells (which do not express the IL-2Rα receptor) do not show uptake of daclizumab-Rhodamine Green at any of the time points (Figure 3D). As with previous reports [2,11], all antibodies showed specific binding only to their respective receptor-positive cells on in vitro analysis, except for trastuzumab-Rhodamine Green that showed weak positive signals from A431 cells probably because of minimal but positive expression of HER-2 receptor on these cells.

Figure 3.

Fluorescence microscopy studies. (A and B) The amounts 1 x 10⁴ SP2/Tac cells and (C and D) 1 x 10⁴ SP2/0 cells incubated with daclizumab-Rhodamine Green (RhodG) and viewed under fluorescence filter after 1, 4, 8, 24, 48, and 72 hours of incubation. In each case, differential interference contrast images are shown in the left column and fluorescent light images in the right column. SP2/Tac cells start to show binding of daclizumab-RhodG to the IL-2Rα receptors expressed on their cell surface as early as 1 hour postincubation. With time, the complexes are gradually internalized into the cell with gradual perinuclear localization. Control SP2/0 cells (which do not express the IL-2Rα receptor) do not show uptake of daclizumab-RhodG at any of the time points. Experiments were performed on an Olympus BX61 microscope (magnification, x 20) using a 2-second exposure.

In Vivo Imaging

To quantify our results, pixel intensity values were obtained from each of the respective spectral images (Cy5, Cy7, and AlexaFluor700) for each of the three tumors present on the mice. Regions of interest (ROIs) were drawn around each tumor and also on the midline region (to measure the background signal). Average pixel intensity from the tumor ROI was divided by the average pixel intensity from the background ROI to derive a signal-to-background ratio (SBR) for each of the three tumor types in each of the spectral ranges. This process was repeated in each mouse for the data set collected by the single-filter acquisition process (Table W1) and the multifilter acquisition process (Table W2).

It is expected that NIH3T3/HER2⁺ tumors will bind trastuzumab-Cy7, having a high SBR for Cy7 spectra, and a low ratio (close to background) for the Cy5 and AlexaFluor700 spectra. Conversely, A431 tumors (which are targeted by cetuximab-Cy5) should have high Cy5 SBR and low Cy7 and AlexaFluor700 values, and SP2/Tac tumors should have high AlexaFluor700 spectra SBR, but low Cy5 and Cy7 ratios. Negative control LS174 tumors (not expressing HER-1, HER-2, or IL-2Ra receptors) would be expected to emit a signal consistent with background autofluorescence.

In all 10 mice, spectral fluorescence imaging was able to differentiate A431, NIH3T3/HER2⁺, and SP2/Tac tumors 24 hours after injection of the antibody cocktail both in vivo (Figure 4) and ex vivo, postresection (Figure 5). Following the unmixing process, A431 can be seen as red, NIH3T3/HER2⁺ as green, and SP2/Tac as blue. As expected, the control LS174 tumors produced spectra consistent with the background signal, which can be subtracted by the unmixing process, producing a colorless appearance for these tumors. The multiexcitation filter (yellow, red, and deep red filter) image acquisitions (Figure 4B) generally produced much clearer images than those obtained by a single (red) filter (Figure 4A). This is reflected in the quantitative results, where the multifilter acquisition data consistently produced higher signal-to-noise ratios. Interestingly, the ex vivo-resected tumor images do not obviously reflect the in vivo findings. The ex vivo tumor images acquired by a single-filter process (Figure 5A) are as clearly differentiated from one another as with the multifilter-acquired images (Figure 5B). This can be explained by the fact that the multifilter process substantially reduces the background signal primarily from the skin and blood vessels similar to that which can be achieved by removing the tumor from the body. This can be demonstrated in Table 1 by comparing background signals from the one-filter and the three-filter acquisition-here, results are only displayed for each tumor type that is expected to give the highest signal in each of the respective spectra (i.e., A431 for Cy5, NIH3T3/HER2⁺ for Cy7, and SP2/Tac for AlexaFluor700). In 29 of 30 cases (exception: mouse 1, A431 in Cy5 spectra), the background signal was decreased by using the three-filter acquisition. For ex vivo images, the background signal, due either to the absence of autofluorescence produced by the skin and/or to the absence of the blood pool effect in the skin, is not present, thus producing images with a single filter comparable to those obtained in vivo with a multifilter set.

Figure 4.

Multitumor in vivo spectral fluorescence imaging. Unmixed in vivo optical imaging following acquisition by (A) a single (red) filter or by (B) three (yellow, red, and deep red) filters in the same mouse. Bold arrow, A431 tumors; arrowhead, NIH3T3/HER2⁺ tumors; curved arrow, SP2/Tac tumors; clear arrow, LS174T tumors. Columns left to right demonstrate composite, Cy5 spectral, AlexaFluor700 spectral, and Cy7 spectral images, respectively. Cy5 spectral image shows strong uptake of cetuximab-Cy5 by A431 tumors, AlexaFluor700 spectra shows high uptake by SP2/Tac tumors, and Cy7 spectral image shows increased uptake of trastuzumab-Cy7 by NIH3T3/HER2⁺ tumors. The composite image (unmixing) allows differentiation of the tumors: red, A431; green, NIH3T3/HER2⁺; blue, SP2/Tac. Multifilter acquisition and unmixing more clearly differentiates the tumor type, significantly increasing the signal-to-background ratios by reducing the signal derived from autofluorescence and the blood pool effect (background).

Figure 5.

Unmixed ex vivo optical imaging following acquisition by (A) a single (red) filter or by (B) three (yellow, red, and deep red) filters in the same mouse. Bold arrow, A431 tumors; arrowhead, NIH3T3/HER2⁺ tumors; curved arrow, SP2/Tac tumors; clear arrow, LS174T tumors. Columns left to right demonstrate white light, AlexaFluor700, composite, Cy5, Cy7 spectral images, and composite images, respectively. As with in vivo imaging, the composite image (unmixing) allows differentiation of the tumors: red, A431; green, NIH3T3/HER2⁺; blue, SP2/Tac. Unmixing following both single- and multifilter acquisition produces similar results, implying that multifilter acquisition is most useful in reducing the background (noise) signal.

Table 1.

Statistical Comparison of One-Filter and Three-Filter Data Results.

Mouse No.	One-Filter BG	Three-Filter BG	One-Filter SBR	Three-Filter SBR	Statistics
Cy5 Spectra: A431 Tumors
1	50.48	0.66	8.09	7.48	N = 10
2	36.16	0.42	16.68	465.00	Paired comparison:
3	21.3	1.51	27.53	225.41	Gap of means = 192.8121236629
4	8.86	1.78	51.08	162.33	P = 2/512 = 0.0039
5	9.52	3.08	47.18	84.26	Wilcoxon signed rank test:
6	4.84	0.38	91.58	711.26	R = 1
7	29.5	4.12	16.53	73.42	P = 2/512 = 0.0039
8	6.16	1.09	72.05	364.30	z = 2.65, P = .0080
9	4.17	1.25	16.03	52.31
10	2.98	0.97	135.23	264.32	Sign test:P = .0215
Cy7 Spectra: NIH3T3/HER2+ Tumors
1	54.41	0.21	3.82	27.98	N = 10
2	69.31	0.45	3.64	386.76	Paired comparison:
3	56.1	2.71	3.28	70.09	Gap of means = 121.219256261032
4	30.14	1.44	3.31	57.13	P = 0/512 = 0.00e + 0
5	37.19	1.76	2.64	59.37	Wilcoxon signed rank test:
6	43.38	0.81	2.28	103.77	R = 0
7	61.93	0.61	2.64	253.07	P = 1/512 = 0.0020
8	41.29	0.86	3.48	166.41	z = 2.75, P = .0059
9	26.75	1.66	2.25	47.17
10	50.03	2.16	2.84	70.63	Sign test:P = .0020
AlexaFluor700 Spectra: SP2/Tac Tumors
1	103.1	0.28	1.17	455.75	N = 10
2	93.71	0.53	1.97	118.66	Paired comparison:
3	47.02	1.59	2.56	68.74	Gap of means = 132.667392120434
4	20.18	1.31	6.59	71.86	P = 1/512 = 0.0020
5	24.29	2.37	7.09	312.24	Wilcoxon signed rank test:
6	11.7	0.42	10.46	113.67	R = 1
7	78.85	1.11	1.82	0.95	P = 2/512 = 0.0039
8	15.77	1.16	4.14	45.14	z = 2.65, P = .0080
9	7.23	0.99	7.09	37.53
10	8.79	0.73	16.15	161.16	Sign test:P = .0215

Signal-to-noise ratios are calculated by dividing tumor signal by the background signal. These values are derived from ROIs placed over the respective tumors and midline regions (background). Results are given only for the tumor type that is expected to give the highest signal in each of the respective spectra (i.e., A431 for Cy5, NIH3T3/HER2⁺ for Cy7, and SP2/Tac for AlexaFluor700).

Statistics were performed comparing the three-filter-derived SBRs to one-filter-acquired SBRs using the Wilcoxon signed rank test for nonparametric data.

BG, background signal; SBR, signal-to-background ratio.

We applied statistics to the signal-to-noise ratios derived for both the single-filter and the multifilter acquisition data sets to determine whether this apparent increase in SBR was significant (Table 1). Using a paired nonparametric analysis with the Wilcoxon signed rank test and comparing each tumor-to-background signal ratio, we can demonstrate that the use of multifilter acquisition significantly improves the SBR for tumors. For the 10 mice imaged: three-filter acquisition improved detection of tumors over background, significantly increasing the signal-to-noise ratio for A431 tumors in the Cy5 spectra, average SBR increased from 48.2 (one filter) to 241.0 (three filters), P = .0215; NIH3T3/HER2⁺ tumors in the Cy7 spectra, mean SBR increase from 3.0 to 124.2 (P = .0020); and SP2/Tac tumors in the AlexaFluor700 spectral range, SBR average increase from 5.9 to 138.6 (P = .0215). Thus, by reducing the background (noise), the signal of each tumor becomes more apparent, allowing a clear distinction between the three tumor subtypes.

Discussion

This work establishes the feasibility of using three separate optical probes to simultaneously image receptors that are expressed by tumor cells in vivo. Furthermore, we have shown that the use of three different filter sets significantly improves the signal-to-noise ratio and improves the differentiation of each of the receptors targeted compared with a single excitation. To maximize the depth penetration, NIR wavelength fluorophores were employed on all antibodies. Within the NIR range, we are further limited by the broad signal peaks of the organic dyes used and the background, autofluorescence signal, which is more prominent at the lower end of the NIR range. Thus, the ability to image fluorophores at this range is the key to increasing the number of dyes that can be simultaneously distinguished. To this end, we selected Cy5 (peak emission = 675 nm), AlexaFluor700 (peak emission = 723 nm), and Cy7 (peak emission = 779 nm). The cyanine dye Cy5.5, used previously, has a peak emission of 696 nm, which is closer to that of AlexaFluor700 than Cy5, and leads to difficulties in their spectral separation (data not shown). To obtain a more constant emission signal from each one of these dyes, various parameters including quantum yields of distinct dyes, efficiency of the excitation filter, the pharmacokinetics of each antibody, and the sensitivity of a charge-coupled device camera should be taken into account. Although the dose of each antibody was fixed at 50 µg to simplify the experiment, we were still able to successfully distinguish distinct tumor types. The increased signal-to-noise ratios achieved and ability to better separate the dyes by using multiple filter sets may allow us to extend the number of organic dyes that we can image in vivo to four or possibly five. Another way to overcome the restrictions of the NIR range is to use nonorganic dyes which have multiple narrow peak emission bandwidths and require only a single excitation light [12]; however, such dyes are not without concerns over their toxicity profile, which may hinder their transfer to the clinic.

The development of in vivo optical imaging with fluorophores parallels the development of in vitro histopathologic methods. Histology was originally based on simple staining methods to identify architectural features of the specimen. Subsequently, pathologists developed techniques to locate molecular epitopes expressed on cells (particularly tumor cells) using antibodies, leading to the field of immunohistochemistry [13]. Similarly, the use of fluorophore-labeled antibodies enables immunofluorescence imaging; both techniques are already performed in vitro with the traditional brightfield (light) microscope. Double or triple stains can be used to gain more information from specimens. However, multiple staining often cannot be performed simultaneously and, therefore, serial sections may need to be separately prepared which is disadvantageous [14]. Furthermore, with immunofluorescence, background signal from autofluorescence may compromise the quality of the image [15]. Histologic techniques using multispectral imaging have recently been developed to try and overcome some of these inherent disadvantages, by resolving multiple dyes within a single specimen, and correcting for autofluorescence [16]. Such techniques use similar principles to those of our in vivo optical imaging technique: the use of targeted fluorophores, acquisition of an image cube consisting of a stack of images obtained with different band-pass filters through the visible and NIR spectra, and subsequent spectral resolution based on the peak emission of each dye. These methods can be performed using either the brightfield microscopes or the newly developed fluorescence microscopes [14]. Multicolor immunohistochemistry methods allow exploration of molecular interactions at a cellular level, a process that has been termed molecular morphology [16]. Histology has the advantage of dealing with thinly sliced ex vivo tissue, thus depth penetration and toxicity are not issues, and nonorganic dyes (e.g., quantum dots), with their inherent advantages, can be used. The success of such ex vivo histologic multispectral imaging shows the way forward for our in vivo molecular optical imaging technique and offers insights into how best to use such methods.

The difficulty with in vivo imaging, is the prominent background signal arising from a combination of autofluorescence from normal (nontumor) tissues and circulating, unbound antibody-optical agent complexes within the blood pool. When dealing with in vitro histologic samples, autofluorescence from the normal tissue is not an issue and unbound signaling agents can be washed out during preparation. When imaging in an in vivo setting, autofluorescence is almost completely eliminated with a simple light-emitting diode flashlight and proper narrow-bandwidth filter enabling high-resolution, multicolor imaging of tumors in mice [17]. In contrast, the unbound injected agents, which remain in the circulation, need to be cleared either by physiological processes which may take considerable time or, alternatively, by the extracorporeal circulation, i.e., immunoadsorption [18], which introduces additional complexity and risks. Therefore, the target-to-background signal ratio can be a significant problem for in vivo imaging. The spectral imaging technique using a cocktail of multiple color reagents is able to define a distinct shape of the blood pool (background) spectrum produced by the circulating mixed unbound agents. The specific signal from agents that are successfully bound to their targets can then be resolved from this background spectrum. Therefore, high background signal, produced by unbound injected agents, is much less of a problem for this method and enables us to use intact IgG molecules (mAbs) for targeting, despite their acknowledged prolonged blood clearance.

A limitation of this work is that subcutaneous tumor implants were employed. It is generally believed that orthotopic implantation or genetically engineered transgenic animals provide more realistic models [19]. However, our experiments were used as a proof of principle of the targeted imaging technique in vivo, and the selected cell lines are known to express their respective antigens in both subcutaneous and orthotopic xenografts and so the simpler approach was desirable. Additionally, tumors may overexpress more than one antigen and the use of models in which only one receptor is highly overexpressed may not be representative. Nevertheless, the technique needs to be established before it can be refined, and further tested for ability to detect smaller copies of targets and to quantify the extent of overexpression in mixed models which will be the subject of future studies.

In conclusion, tumors were simultaneously distinguished and diagnosed in vivo, using a cocktail of three optically labeled antibodies each labeled with a different fluorophore. Multiple filter sets increased the signal-to-noise ratio by substantially reducing the background signal, and may allow more than three optical dyes to be resolved within the narrow limits of the NIR range. The reduced depth penetration of the technique would realistically limit future clinical applications to imaging either superficial tumors, or those that are endoscopically accessible. Our technique produces highresolution polychromatic images and offers potential for multicolor, multiparametric diagnosis and typing of tumors in the clinic. Furthermore, optical imaging has the advantages of sufficient sensitivity to provide semiquantitative assessment of the number of receptors expressed on cellular surfaces and sufficient resolution to detect subsurface, submillimeter tumors without requiring ionizing radiation.

Abbreviations

HER-1

human epidermal growth factor receptor type 1

HER-2

human epidermal growth factor receptor type 2

EGFR

epidermal growth factor receptor

IL-2Rα

interleukin-2 receptor α subunit

SBR

signal-to-background ratio

ROI

region of interest

NIR

near-infrared

mAb

monoclonal antibody