Intravital imaging of mouse colonic adenoma using MMP-based molecular probes with multi-channel fluorescence endoscopy

: Intravital imaging has provided molecular, cellular and anatomical insight into the study of tumor. Early detection and treatment of gastrointestinal (GI) diseases can be enhanced with specific molecular markers and endoscopic imaging modalities. We present a wide-field multi-channel fluorescence endoscope to screen GI tract for colon cancer using multiple molecular probes targeting matrix metalloproteinases (MMP) conjugated with quantum dots (QD) in AOM/DSS mouse model. MMP9 and MMP14 antibody (Ab)-QD conjugates demonstrate specific binding to colonic adenoma. The average target-to-background (T/B) ratios are 2.10 ± 0.28 and 1.78 ± 0.18 for MMP14 Ab-QD and MMP9 Ab-QD, respectively. The overlap between the two molecular probes is 67.7 ± 8.4%. The presence of false negative indicates that even more number of


Introduction
Early detection and the removal of precancerous lesions have been shown to reduce mortality rate associated with colon cancer and this could be aided by specific markers and endoscopic imaging modalities. An endoscope can transmit light deep into the body through optical fibers and visualize lesions that are typically inaccessible by other means [1]. However, it has several limitations in detecting small lesions or identifying microscopic pathological features. For example, conventional colonoscopic diagnosis with white light can miss up to 20% of adenomas [2,3]. Routine colonoscopy procedures cannot provide histopathologic information by itself, and sometimes requires unnecessary biopsy or removal of suspected lesions for diagnosis, often requiring patients to revisit for additional biopsies or examinations. Thus, several methods to improve its accuracy to identify dysplasia or cancer have been proposed including narrow-band imaging, autofluorescence imaging and chromoendsocopy [4][5][6]. High-resolution endoscopy with preclinical animal models has revealed microscopic details in disease progression [7][8][9]. However, these modalities have fluctuating results for detecting colon cancer due to limited molecular specificity.
Recently, molecular imaging has emerged in gastroenterology by use of fluorescently labeled probes to specifically highlight neoplastic lesions on the basis of their molecular signatures [10][11][12]. Neoplastic lesions often overexpress certain receptors or enzymes in transformed mucosa. These contrasts between normal and abnormal tissue provide an effective method of detection with imaging techniques. Proteases are potential candidates of molecular targets as proteolytic enzymes play an important role in cell proliferation, invasion, angiogenesis and metastasis. These enzymes provide important means for the detection and diagnosis of cancer in the digestive tract, for example, colon cancer [13,14]. Matrix metalloproteinase (MMP) enable malignant cells to breach basement membranes and subsequently invade neighboring tissues. Several MMPs are expressed in cancers at much higher levels than in normal tissue. The extent of expression is known to be related to tumor stage [15][16][17][18].
With specific molecular targets labeled with distinct fluorescent molecules such as quantum dots, wide-field fluorescence endoscope enables rapid imaging of large mucosa in GI tract to screen potentially cancerous lesions. Together with white-light imaging, fluorescence endoscopic imaging can be realized with emission filters and excitation light sources. This approach can provide a red flag imaging technique for the detection of suspicious lesions. In preclinical study, this imaging system could be used for characterization and visualization of molecular probes that are supposed to bind to specific targets associated with tumor. As molecular expression of tumors is known to be heterogeneous, multiple molecular targeting strategies could increase the detection specificity of tumor screening [19,20].
Here we show the endoscopic imaging of colonic adenoma in a colon cancer mouse model with multiple MMP antibody-quantum dot (MMP Ab-QD) conjugates with a home-modified wide-field multi-channel fluorescence endoscope.

AOM/DSS colorectal cancer model
4-week-old female BALB/c mice (DBL, Eumseong, South Korea) were used for this study. After acclimatization for seven days with basal diet and standard water feeding, single intraperitoneal (i.p.) injection of AOM (10 mg/kg body weight; Sigma-Aldrich, MO, USA) was done for all mice. One week later, drinking water containing 2% DSS (MP biomedicals, Illkirch, France) was administered for 7 days. Then the drinking water switches back to standard drinking water for 2 weeks. This administration method of DSS in the drinking water was repeated. All mice were weighed twice per week. To check tumor formation in the colonic wall, colonoscopic examination was performed. All animal experiments were performed with protocols approved by the Institutional Animal Care and Use Committee (IACUC) of the Gwangju Institute of Science and Technology.

Preparation of targeted fluorescence molecular probes
MMP9 and MMP14 Ab were conjugated with red and yellow light emitting QD (MMP9 Ab-Red QD and MMP14 Ab-Yellow QD, respectively). Two different color emitting CdSe/CdS/ZnS(Core/Shell/Shell) QDs were synthesized using procedures previously published (See Appendix Section) [21]. The absorption and fluorescence profiles of probes are shown in Fig. 1, where the emission peaks can be found at 585 and 630 nm, respectively. The QD surface was co-decorated with a zwitterionic ligand and a ligand bearing carboxylic acid. The carboxylic acids were used for conjugation with MMP9 Ab and MMP14 Ab [21].

Multi-channel fluorescence and white light imaging endoscope setup
To image molecular fluorescence from mouse colonic mucosa, we constructed multi-channel fluorescence endoscopic system using a commercial endoscope probe (HOPKINS II Telescope 27301AA, Karl Storz, Tuttlingen, Germany) by adding detection and excitation light paths (Fig. 2). The detection part composes of endoscopic probe, emission filter wheel with various filters, an achromatic lens (f = 50 mm, Thorlabs, NJ, USA), sensitive color charge coupled device (CCD) camera (QIClick, QImaging, Surrey, BC, Canada), an inflating pump for mildly inflating the colon during imaging and a white light source (LS-100W, Light Solution, Cheonan, Korea). The excitation light is generated from a laser source with wavelength of473nm (Shaghai Dream Lasers Technology, SDL-473-300T,Shanghai, China) that excites corresponding fluorescent probes after coupling into two coherent fiber bundles (FIGH-10-500N, Fujikura, Tokyo, Japan). Endoscopic images were obtained via the imaging scope with white light and emitted fluorescence light after the excitation light exits the distal end of the fiber bundles to illuminated specimen. The reflected light by the white light source was displayed to form a white light image onto the color CCD camera and fluorescence passes through emission filters (FF02-628/40-25 and FF01-579/34-25, Semrock, NY, USA) prior to detection with another CCD camera. The imaging scope and the achromatic lens combination form a magnified image on the sensor of the CCD camera, which is visualized on the computer monitor with commercial software (QCapture, QImaging, Surrey, BC, Canada). The exposure time of the CCD camera was adjusted from 100 μs to 1sdepending on the sample brightness. The system achieves the lateral resolution of 44 μm and field of view of 7.5 mm at a working distance of 5 mm using USAF 1951 resolution target (R3L2S1P, Thorlabs, NJ, USA). We performed a controlled experiment in vitro to determine the detection limit of our fluorescence endoscope by imaging Red QD samples with known concentrations of 1.01×10 11 , 2.00×10 10 and 4.00×10 9 (QDs/mm 2 ) as well as the background with typical operating condition (exposure time of 500 ms). We obtained the detection of limit of our fluorescence endoscope as 8.60×10 9 (QDs/mm 2 ) based on the Rose criterion of SNR of five. However determination of the detection limit in vivo is non-trivial as the actual concentration of the molecular probes at the tumor lesion depends on each tumor and varies over time even with controlled amount of injection.

In vivo molecular fluorescence imaging of colonic tumors
For evaluation of each molecular probe, four mice were used for MMP9 Ab-Red QD and three mice were used for MMP14 Ab-Yellow QD. Prior to imaging inside mouse colon, the colon was washed with phosphate buffered saline (PBS) to remove debris. The endoscopic imaging with white light was performed to identify the presence of adenomas. Animals bearing tumors were injected intravenously via tail vein with each molecular probe (concentration: 100 nM, volume: 100 μL). The mice were gas anesthetized with isoflurane. Many molecularly targeting imaging relies on labeling target-specific molecules after intravenous (i.v.) injection when enough probes bind to their targets. The method has been particularly useful for imaging receptors and cell-surface-expressed molecular targets. After i.v. injection of MMP Ab-QD conjugates, we measured fluorescence signal from tumor of AOM/DSS mouse model at designated time points using 579/34 nm and 628/40 nm emission filter with 473 nm excitation and compared the signal value and target-to-background (T/B) ratio. Typical imaging area was located in 0.5 cm to 3.0 cm depth from the anus of the mouse. By 12 hours after i.v. injection of MMP14 Ab-Yellow QD, the fluorescence intensity was increased. However, the peak of fluorescence intensity was increased by 24 hours after i.v. injection of MMP9 Ab-Red QD. Although the peak times were different, each T/B ratio at 12hours after injection of the probes was larger than at 24 hours (Fig. 8).
In addition, we concomitantly injected two molecular probes intravenously with the same concentration (100 nM) and volume (100 μL) and imaged 12 hours after injection. In order to determine the fluorescence T/B ratio, snap shot images were used for quantifying fluorescence intensities. All fluorescence images were analyzed with the Image J software (National Institutes of Health, USA). Circular ROIs were place in the tumor (S t ) and in adjacent normal mucosa (S n ). All ROI intensities were recorded in pixel intensity values between 0 and 255. After obtaining mean signal intensities, the T/B ratio was determined as S t /S n [22]. Data are presented as means ± standard deviation of the mean, and each signal intensity value represents either tumor or normal mucosa signal intensity subtracted by background signal intensity (intrinsic camera noise). The significance of differences between signal intensity values was determined by using Mann-Whitney U test. P<0.05 was considered to indicate significant difference. Analysis was performed with the statistical software package for social science (SPSS, version 19.0.0; Chicago, IL, USA).

Ex vivo colon tissue fluorescence imaging by molecular probes
Two and three AOM/DSS mice model (age over 12 weeks old) were used for ex vivo fluorescence imaging with control IgG-QD and MMP Ab-QDs, respectively. For ex-vivo mouse colon tissue imaging, tumor-bearing mice were sacrificed and the colons were longitudinally excised. The excised colon tissues were incubated in 100 nM MMP9-Ab Red QD and MMP14-Ab Yellow QD (with IgG-yellow QD, IgG-red QD, respectively) for 30 min at the same time. These tissues were subsequently rinsed three times with 1X PBS. The fluorescence signal, expressed at the colon tissue (Fig. 6F-6H), was obtained using the IVIS spectrum system (Perkin Elmer, MA, USA). Images were acquired with excitation at 430 nm and emission at 500-800 nm using auto acquisition, a binning factor of 8 and field of view of 13.4 cm. Final fluorescence imaging was obtained with spectral unmixing using the Living Image® 4.3.1 software (Perkin Elmer, MA, USA) and was quantified radiant efficiency.

Histologic analysis of excised colon tumor
Twenty-four hours after molecular probes were injected, mice were sacrificed and the colons were surgically excised. The colon tissues were examined ex vivo using optical imaging system followed by hematoxylin and eosin (H&E) staining. For histopathological evaluation, normal and tumoral lesions were excised, fixed for 24 hours in 4% (w/v) paraformaldehyde prior to H&E stain. Each specimen was examined by an experienced pathologist using standard published criteria [23,24].

Quantitative fluorescence image correction based on reflection image
We analyzed experimental images based on previous quantitative fluorescence molecular imaging [25]. As the integration time affect the amount of signal, we kept the same integration time for quantitative comparisons between experimental groups that maximizes the use of dynamic range of our CCD detector. Corrected fluorescence signal intensity was calculated by dividing the fluorescence image by a sequentially acquired white light image on a pixel-by-pixel basis. Prior to analysis, all pixels outside the circular endoscopic field of view (FOV) were set to zero on the corrected image. Afterwards we set the threshold value as the average of maximum value and minimum value of the images. We compared several threshold settings such as 25%, 50%, and 75% of the min-max value of the fluorescence image, and found 50% threshold gave better correlation with tumor boundary based on whitelight endoscopic imaging. Although this simple approach works reasonable in our AOM/DSS tumor model and QD conjugated molecular probes, we may need to use more sophisticated threshold techniques for other situations such as depressed tumor and serrated tumor.

Intravital examination of MMP-bound colon tumors with wide-field fluorescence endoscope
The specific binding activity of the MMP targeted probe for colonic adenomas was assessed in vivo in 2 groups of mice. First, we randomly assigned three mice per group from those which developed grossly visible tumors by performing white light endoscopy. Twelve hours after MMP targeted probes were injected intravenously, fluorescence imaging was performed for two groups: (1) MMP14 Ab-Yellow QD (six tumors in 3 animals) and (2) MMP9 Ab-Red QD (seven tumors in 4 animals). The images from white light endoscopy of a representative adenoma are shown in Fig. 3(A) and 3(D). Specific binding of each MMP targeted probe to corresponding colonic tumor is demonstrated by comparing the fluorescence images collected with the white light endoscopy (Fig. 3(B) and Fig. 3(E)). In Fig. 3(B), MMP14-Yellow QD demonstrated obvious contrast that enabled detection of flat lesions compared with white light image in Fig. 3(A). From histological inspection, the presence of adenoma was confirmed ( Fig. 3(C)). In addition MMP9 Ab-Red QD probe was used to detect suspicious tumor lesion in different animal. The T/B ratio arising from fluorescent lesions was significantly higher than unity compared with surrounding non-fluorescent lesions. The fluorescence images of colonic adenoma collected with MMP14 Ab-Yellow QD and MMP9 Ab-Red QD shows increased signal over adjacent surface with a T/B ratio of 2.10 ± 0.28 (average ± standard deviation) and 1.78 ± 0.18, respectively. H&E of the adenoma is shown in Fig. 3(C) and 3(F).

Early detection of flat colonic tumor with MMP14-based molecular probe
Use of MMP14 Ab-Yellow QD achieved specific contrast enhancement that enabled detection of suspicious flat lesions that could have been missed with white light endoscopy. A suspicious region was identified with white light imaging (Fig. 4(A)) and subsequent fluorescence imaging based on MMP14 Ab-Yellow QD probe revealed the presence of flat tumor (In Fig. 4(B), white arrow) that was confirmed later as adenoma from histological inspection( Fig. 4(C)).

Multi-channel tumor targeted fluorescence images from multiple molecular probes
To increase the sensitivity of detection for screening purpose, the simultaneous examination of multiple molecular targets with distinct fluorescent QD labels were tried. Twelve hours prior to endoscopic imaging we sequentially injected MMP14 Ab-Yellow QD and MMP9 Ab-Red QD probes intravenously both with the concentration of 100 nM and the volume of 100 μL. In a representative white light image (Fig. 5(A)), three suspicious colonic adenomas from a representative AOM/DSS treated animal is indicated with white arrows. The lesions with different sizes and shapes were identifiable in both channels. Using two separate fluorescent emission channels, each binding patterns to colonic tumors can be displayed (Fig.  5(F) and 5(G)), and the overlapping area was shown as a merged image in Fig. 5(H) based on the fluorescence correction method described in the Materials and Methods. There was a substantial overlap between two distinguished areas highlighted with each molecular probe used (overlap ratio = 83.8 ± 5.4% (MMP14 Ab-Yellow QD), 77.8 ± 6.7% (MMP9 Ab-Red QD) and 67.7 ± 8.4% (both MMPs) with n = 7 lesions in three animals). After fixation, the lesion was confirmed as high-grade adenoma from the inspection of H&E staining. These results suggest that both MMP-based molecular probes bind to similar locations within the AOM/DSS colon tumors.
While the combination of MMP9 and MMP14-based probes effectively showed the detection of common tumor lesion, the two molecular targeting probes might not be sufficient. As an example, we also found lesions where two clearly identifiable tumors with white light endoscopy did not give signals in both molecular channels. While both MMP-QD probes were overexpressed in one tumor (white arrow), no signal was detected in a neighboring tumor (red arrow) (Fig. 6). We prepared another set of AOM/DSS-treated mice. The excised colons were imaged by IVIS imaging system after stained by control IgG-QD conjugate (n = 2) and MMP Ab-QD (MMP14 Ab-yellow QD and MMP9 Ab-Red QD, n = 3). While control IgG-QD probe showed minimum non-specific signal among tumor site, MMP9 and MMP14 Ab-QD probes showed noticeable signals throughout tumor sites. As the in vivo observation showed in Fig.  5, some relatively large tumors did not show signal with molecular fluorescence channels as exemplified in the yellow inset of Fig. 6(G) and 6(H). These tumors may have less availability for the molecular probes to infiltrate (indicated by red arrow in Fig. 6(G) and 6(H)). MMP-based molecular probes were bound to similar suspicious tumor lesion both in vivo and ex vivo image.

Discussions and conclusion
We present in vivo wide-field multi-channel fluorescence endoscopic imaging adapted from a commercial rodent colonoscope with AOM/DSS mouse colon cancer model. The system is designed with a rigid endoscope coupled to a laser source and filter wheel, and consolidated into a single camera that provides both white light and molecular information. The general color CCD has limitation based on Bayer filter which would affect the ability of the CCD to perform accurate and efficient fluorescent imaging in the visible band. To reduce this limitation, we used scientific grade color CCD camera for detecting fluorescence signal.
For molecular targeting, we synthesized multiple MMP antibody-QD probes for endoscopic observation towards screening purpose. Among MMPs, MMP9 and MMP14 are proteloytic enzymes and are known to play a significant role in cancer transformation of mucosa in the gastrointestinal tract. While various mouse models have been used for study of colon cancer, we used azoxymethane (AOM)/dextran sulfate sodium (DSS) model that has gained popularity due to its reproducibility to form sporadic tumors. MMP9 and MMP14 antibody-QD conjugates were used both individually and simultaneously to screen precancerous lesions over large mucosal surface area. Based on pharmacokinetic data with multiple MMP Ab-QD probes, we identified 12 hours after i.v injection to maximize the accumulation of the probes in tumor region while the unbound or nonspecific probes were cleared. The non-specific binding of QD probes are supposed to be minimal as they were designed with anti-fouling zwitterionic QD surface coating. The average T/B ratios of MMP14 Ab-QD and MMP9 Ab-QD (Fig. 7(B)) that were bound to colonic adenomas identified from endoscopic imaging were not significantly different.
Tumor-specific fluorescence signal by two molecular probes can be found at suspected lesions and these molecular probes are expressed similarly in both tumor lesions (Fig. 7). However, these (MMP9 and MMP14) may not be enough to detect all. With MMP9 Ab-Red QD probe, twelve out of fourteen tumors were detectable with seven AOM/DSS mice while eleven out of thirteen tumors were detectable with six AOM/DSS mice with MMP14 Ab-Yellow QD. Potential reasons could include the dissimilar delivery of molecular probes into one tumor that might not be perfused well or fundamental absence of corresponding molecular targets in the tumor. Either way could be explained as tumor heterogeneity [19,20,26]. Another potential reason for this phenomenon might be linked to inflammatory process as we used colitis-induced colon cancer mouse model [27]. In the later stage inflammatory process, the MMP expression rate of MMP was lower [28]. In addition, another inflammatory biomarker, COX-2, also showed similar tendency [29]. Therefore, MMP expressions of tumors could be heterogeneous with different inflammatory stage [30]. Treatment of mice with AOM and DSS rapidly and effectively models human colitis-associated cancer. While this model is highly popular in the study of colon tumor development in the setting of inflammation, limitations still exist. Like most other colon cancer models, AOM/DSS tumors rarely progressed to produce metastasis. Currently we are developing other colon cancer models with metastatic potential to see if our finding with MMP-based detection also applies in similar fashion. Yoon et al. used MMP-activatable probe in ex vivo colon tissue with AOM/DSS mouse model [31]. In their study, the sensitivity was 82% that is comparable to our sensitivity of 85.7% (MMP9) and 84.6% (MMP14). Their probe showed relatively higher T/B ratio of 5.70 ± 2.30 than our T/B ratio of 1.78 ± 0.18 (MMP9) and 2.10 ± 0.28 (MMP14) as their activatable probe could minimize background signal though there exists potential toxicity issue [32]. Lately, a topically applied enzymatically activatable probe targeting cancer associated enzyme showed rapid detection and high target-to-background in murine colon cancer model [33]. The use of QDs have several unique advantages allowing multiple colors in visible spectrum range and can be extended to more number of channels in the near-infrared window with only single excitation source due to relatively narrow and symmetric emission spectrum. Until 12 hours after i.v. injection of MMP14 Ab-Yellow QD, the fluorescence intensity was increased. However, the signal became weak afterwards. Otherwise, the peak of fluorescence intensity was increased by 24 hours after i.v. injection of MMP9 Ab-Red QD and the signal became weak after 24 hours. Although the peak times appear different, each T/B ratio at 12 hours after injection of the probes was larger than at 24 hours. From these data we choose 12 hours after i.v. injection of our molecular probes as the optimal imaging time, and used this throughout our in vivo imaging experiments.