1.

Introduction

Advances in in vivo imaging of the molecular expression of cancer have provided access to understanding the tumor microenvironment and how variations in expression of proteins and signaling molecules affect cancer development and treatment.¹ The ability to visualize the molecular information of tumors has the potential to aid in surgical removal of lesions,^2,3 provide a route for personalized treatment and treatment monitoring,^4,5 and provide a method of early detection of a lesion.^6,7 Colorectal cancer is the third most commonly diagnosed and third most deadly cancer in the United States, accounting for 8% of cases in both categories, with mouse models being critical tools in its study.⁸ Currently, many studies use xenograft models of cancer, allowing for the selection of cell types (and thus molecular expression) present and for the placement of the tumor in an area accessible to imaging.^9–13 Xenograft models, however, have many deficiencies, including limited relevance to spontaneous carcinogenesis, or to the natural tumor microenvironment.^14,15 The carcinogen azoxymethane (AOM) has been shown to cause the sporadic growth of colorectal tumors in the distal colon of mice, exhibiting many of the morphological and pathological features associated with sporadic colorectal cancer in humans.¹⁶ This model can be used to study the tumors of the colon in vivo, providing a physiologically relevant method for studying colorectal cancer and its detection.

Targeted contrast agents have great potential to detect cancer specific biochemicals in vivo, allowing for earlier detection of the disease rather than morphological changes alone. Quantum dots are nanoscopic particles of semiconductors with fluorescence emission affected by the size of the particle. They have many desirable fluorescence qualities for biological imaging, such as high quantum efficiency, high photostability, a wide range of excitation wavelengths, a narrow emission band, and they can be produced to emit over a wide range of wavelengths.^17–21 Many studies have used quantum dots as fluorescent markers for cancer in vitro^22–27 and in animal models in vivo,^28–40 with their use in vivo being mostly limited to nontargeted^29,34,37 or xenograft^{28,30–33,38–40} labeling. While numerous molecular targets of carcinogenesis exist, it has been shown that vascular endothelial growth factor receptor 2 (VEGFR2) is upregulated in many cancers, including some colorectal cancers, as it is important in tumor angiogenesis.^41–45 Also, VEGFR2 is currently considered a predictor for clinical outcome and, in some instances, is used in targeted therapy with antiangiogenic drugs.^46–48 Recently, our group demonstrated that QDot655 targeted to VEGFR2 (QD655-VEGFR2) can be applied to the colon of carcinogen-treated mice in vivo and provides significantly increased contrast between diseased and undiseased tissue ex vivo (explanted) with high sensitivity and specificity compared to a control contrast agent.⁴⁹ This quantum dot is readily available, easy to conjugate, emits in a region of light that has less endogenous emission interference, and is compatible with all commonly used optical coherence tomography (OCT) systems. Thus, evidence exists that quantum dots bioconjugated to VEGFR2 antibodies can provide a mechanism for evaluating the molecular changes of colorectal tumors in vivo.

Optical imaging of ex vivo and xenograft tissue, as well as in vitro models, allows for high resolution and sensitivity to targeted biomolecules; however, these simplified systems fail to incorporate the complex environment of both the diseased tissue and surrounding undiseased tissue. Utilization of endoscopic imaging techniques provides a direct pathway for nondestructive and minimally invasive optical imaging of tissue in vivo. Our group has developed and described in detail a minimally invasive microendoscopic dual-modality imaging system combining OCT and laser induced fluorescence spectroscopy (OCT/LIF) to be used in conjunction with the AOM mouse model of colorectal cancer.^50–54 This technology allows for simultaneous longitudinal tracking of morphological (OCT) and biochemical (fluorescence) changes as colorectal cancer develops and, in combination with the AOM mouse model, provides a rare opportunity to study cancer development both in vivo and in situ. Also, this technology allows for the superficial imaging of tissue through minimally invasive access. Because of this, lavage, or surface labeling, of this tissue using the contrast agent can be performed, avoiding intravenous injections of high concentrations of potentially toxic contrast agents and providing maximal labeling of the available receptors on the mucosal cell surface. The relatively large size of the contrast agent and topical application also limit the ability of the contrast agent to move into the tissue, further reducing exposure. While studies suggest that accumulated quantum dots do not have cytotoxic effects in vivo,^55–57 it is still important to minimize this exposure for future use in both animal studies and human clinical imaging.

In this study, QD655-VEGFR2 was used in a longitudinal in vivo study on the ability to visualize VEGFR2 expression using OCT/LIF dual-modality imaging. The goal of this study was to determine if QD655-VEGFR2 could be used in conjunction with OCT/LIF dual-modality imaging to visualize VEGFR2 expression in colorectal tissue. We have previously shown that QD655-VEGFR2 can be used to label adenoma expressing VEGFR2 through in vivo labeling and ex vivo fluorescence imaging with high sensitivity and specificity, and also using the OCT/LIF microendoscope ex vivo.⁴⁹ Quantum dot fluorescence in the colon should coordinate with VEGFR2 expressing adenoma. OCT provided the gold standard for the presence of adenoma with immunohistochemistry (IHC) as a gold standard for VEGFR2 expression. Proper targeting of the contrast agent should yield a stronger fluorescence signal from the diseased regions of the colon than from the surrounding undiseased regions. QD655-VEGFR2 was able to target VGEFR2 expressing diseased areas of colon; however, challenges in fully flushing the unbound contrast agent from the colon before imaging arise when moving from ex vivo imaging to in vivo imaging.

2.

Methods

2.4.

Image/Data Processing

OCT images were used to create a binary map of the presence of adenoma in a two-dimensional (2-D) format with the length of the colon on the $x$ axis and the rotation on the $y$ axis. Adenomas were identified according to previously published criteria.⁵⁸ Briefly, each longitudinal OCT image slice was segmented into 36 segments along the 30 mm length. Each segment was identified as either diseased (adenoma) or undiseased based on the previously published criteria (thickened regional mucosa and/or moderate to marked protrusion of mucosa to more than twice the average regional thickness, moderate to marked signal attenuation and faint or obscured tissue boundaries). Results from all eight longitudinal image slices from a colon were stacked together to create an $8\times 36$ matrix tumor map (Fig. 1). This procedure was performed for all mice and time points.

Fig. 1

(a) Optical coherence tomography (OCT) image at one circumferential position. OCT images were graded based on a previously published method for adenoma detection. This method grades segments of the colon as either adenoma or undiseased. (b) All eight radial positions for one mouse and (c) the binary grading of the OCT images from the same mouse to be used in mask formation and analysis. The white area shows how an adenoma transfers from the original OCT images to the tumor map. Then, the adenoma is dilated in all directions and the dilation as well as the original adenoma were considered the diseased area for mask formation. Then the diseased area was dilated again in all directions, and this dilation minus the diseased area is considered the undiseased area for mask formation.

Fluorescence emission intensities collected by the LIF subsystem contained spectral information originating from the tissue autofluorescence and the autofluorescence of the system components, as well as the desired signal from the quantum dots [Fig. 2(a)]. Analysis of the spectra identified five separate spectral components, one belonging to QD655 and the others originating endogenously from the tissue or system [Fig. 2(b)]. It should be noted that in time point 5 of the experiment, a spectrum with a central wavelength of 605 nm was detected instead of the expected 655 nm from QD655. The manufacturer later sent a letter to inform of a mislabeled sample of QD605 being sent to us instead of QD655. For this reason, the spectrum for QD605 was isolated for time point 5. Utilizing a nonlinear least squares curve fit, the quantum dot signal was separated from the collected spectrum for all collected spectra. A 2-D map of the quantum dot fluorescence intensity was formed with the length of the colon on the $x$ axis and the rotation on the $y$ axis, similar to the tumor maps from OCT [Fig. 2(c)]. Unless stated otherwise, these fluorescence intensity maps were downsampled to 36 elements using bicubic interpolation to match the dimensions of the OCT tumor maps for each of the analytical methods described below. The distal 2 mm were removed from the OCT and LIF maps, as the quantum dots tended to aggregate in the fur and squamous epithelium around the anus. This processing was performed for all mice and time points.

Fig. 2

(a) A surface map of the full fluorescence spectrum from one circumferential position along the length of a colon before the signals are separated. (b) One data point along the length of the colon demonstrating the five distinct spectra that were identified and separated out of the measured spectrum. (c) All eight circumferential positions of the QDot655 (QD655) signal after spectral separation from the measured signal stacked allowing for visualization of the colon as if it were cut along the length between circumferential positions 1 and 8 and splayed open. (d) The position of the eight circumferential image acquisition locations.

3.

Results/Discussion

Table 1 includes a count of all mice that started and completed the study, the adenoma count at the final time point, and the number of diseased/undiseased region pairs used in the analysis (which includes the sum of all region pairs at all time points, including in some cases mice that did not survive to the final time point). Three mice died as a result of complications from anesthesia. The other five mice were euthanized due to excessive tumor burden/bleeding and intestinal impactions. No saline-treated (control) mice developed adenoma during the study. Adenoma appeared in four AOM-treated mice by time point 3, in six AOM-treated mice by time point 4, and in all but one AOM-treated mouse by time point 5. In the AOM-treated group, the QD655-VEGFR2 labeled group had 14 measured adenomas by the final time point, the QD655-IC labeled group had 9 measured adenomas by the final time point, and the saline labeled group had 7 measured adenomas by the final time point.

Table 1

This table includes the total number of mice, the number of mice surviving at time point 5, the total number of adenoma that developed by time point 5, and the total number of diseased and undiseased regions measured in each contrast agent/treatment group combination. The number of diseased and undiseased regions measured includes each adenoma at each time point, meaning an adenoma at time point 4 and the same adenoma at time point 5 were considered as two separate adenomas. Also, if the diseased regions of two adenomas in one colon overlapped after dilation, they were considered one diseased area. There was one undiseased region measured for every diseased region measured.

Using the tumor maps generated through OCT, the average fluorescence intensity from the diseased tissue and the surrounding undiseased tissue was measured for each contrast agent/treatment group combination. As seen in Fig. 3, the fluorescence intensity for the diseased regions of QD655-VEGFR2 labeled, AOM-treated colons was 3.2 times stronger than that of the surrounding undiseased tissue suggesting that the contrast agent is correctly labeling the diseased areas. Intensities were not significantly different, however, due to a high standard deviation in both data sets ($p=0.12$). The average fluorescence of the diseased regions of QD655-VEGFR2 labeled colons was 4.8 times stronger than the diseased regions of QD655-IC labeled colons, suggesting that the binding was specific. Again, this difference was not significant due to large standard deviations ($p=0.09$). There was no significant difference in the average intensity of diseased and undiseased regions of QD655-IC labeled colons, suggesting that this nonspecific agent was passively accumulating in the colon. The average intensities for all other contrast agent/treatment group combinations were low. The minimal fluorescence intensity registered from both AOM-treated and control colons lavaged with only saline represents the low level of tissue autofluorescence. Low levels of fluorescence from control colons lavaged with QD655-VEGFR2 and QD655-IC likely represent passive accumulation of the contrast agent in the colon. A significant difference in average fluorescence intensity was seen between diseased areas of QD655-VEGFR2 labeled, AOM-treated colons and diseased areas of saline labeled, AOM-treated colons (factor of 48, $p=0.049$). Also, a significant difference was seen between undiseased regions of QD655-VEGFR2 labeled AOM-treated colons and QD655-VEGFR2 labeled control colons (factor of 5.6, $p=0.032$), indicating that there is significantly more contrast agent accumulating in the undiseased tissue of AOM-treated colon than control colons.

Fig. 3

Fluorescence intensity measurements of each diseased region and the surrounding undiseased regions for each colon were taken. All diseased regions or all undiseased regions for all colons of mice labeled with a particular contrast agent and in a particular treatment group were averaged together for a representative signal intensity. All tumors from all time points were combined for each group. Standard error is displayed, as well as $p$ values from the Student’s $t$ test. $N$ values can be taken from Table 1 as the number of diseased/undiseased regions measured.

Comparison of data from this in vivo study and our previously published ex vivo study⁴⁹ shows strikingly similar trends, as seen in Fig. 4, in which the average fluorescence intensities from both the in vivo study and the ex vivo study are plotted together. The ex vivo study results demonstrated statistically significant differences between groups, which in this study failed to achieve significance due to large standard deviations. The difference between our current study and the previous study lies in the in vivo imaging and data processing. In the previous ex vivo study, colons were excised and splayed open for imaging, allowing thorough rinsing of the tissue and greatly reducing the unbound contrast agent present in the colon before imaging was conducted. Although most imaging was performed utilizing a fluorescence microscope, an example ex vivo imaging using the OCT/LIF microendoscope employed in this study showed strong QD655-VEGFR2 signal in the tumor, but nearly an undetectable signal in undiseased tissue, suggesting that differences in the significance in the data are likely attributed to the differences in the ability to fully rinse the unbound contrast agents from the colons when imaging in vivo, rather than being attributable to the endoscope. In vivo rinsing of the mouse colon poses challenges in removing unbound contrast agent from the mucosa due to the small diameter of the colon and the tendency of fluid to accumulate in the more proximal colon, particularly as adenoma develops and the lumen of the colon becomes blocked.

Fig. 4

The fluorescence intensity results from the current in vivo study compared to those from the previous ex vivo study. Both the diseased and undiseased tissues for each contrast agent/treatment group are represented. Error bars represent standard error.

As seen in Fig. 4, the difference in the average fluorescence intensity between the diseased and undiseased regions of QD655-VEGFR2 labeled, AOM-treated colons was similar, as was the difference in the average intensity of the QD655-VEGFR2 labeled, AOM-treated diseased regions and the QD655-IC labeled, AOM-treated diseased regions for the in vivo and ex vivo studies. In the in vivo study, the average fluorescence intensity of the undiseased areas of the colons in the QD655-VEGFR2 and QD655-IC labeled, AOM-treated colons and control colons was higher than these regions in the ex vivo study, suggesting more passive accumulation. It is interesting to note that the ex vivo study captured more autofluorescence from the tissue (see in saline labeled tissue), since the ex vivo study did not utilize spectral separation of the quantum dot and autofluorescence signals.

Classification accuracy in the current study data for QD655-VEGFR2 labeled colons revealed a low diagnostic performance, as might be expected given the inability to achieve statistical significance between diseased and undiseased regions. Maximum classification performance ($\text{sensitivity}+\text{specificity}$) was 140% (perfect performance would be at 200%). A typical sensitivity was 40% with 80 to 85% specificity. In the previous ex vivo study, a sensitivity of 85.7% at a specificity of 91.3% was obtained. Given the similarities in average trends between in vivo and ex vivo studies, but the large standard deviations in the in vivo study, it was speculated that challenges in fully flushing the unbound contrast agent from the colons in vivo account for the differences in classification accuracy.

The distribution of the fluorescence signal along the length of the colon and at each circumferential image location for each time point and contrast agent supports this supposition. The fluorescence intensity is typically higher in the proximal end of the colon, as well as around the anus (Fig. 5, near 30 and 0 mm). The nonspecific fluorescence at the anus was mostly removed by excluding the distal 2 mm, which was contaminated with fluorescence from dyed fur and squamous epithelium. Data from QD655-IC labeled colons had a similar trend in fluorescence intensity distribution without correlation to tumor locations. Circumferential distributions revealed that in several time points, there is an accumulation of fluorescence signal at a ventral position (in the direction of gravity, data not shown). The inability to completely remove unbound contrast agent from the colon is most likely the key contributor to this nonspecific fluorescence signal distribution along the length of the colon and in the direction of gravity, causing a large signal variation.

Fig. 5

Fluorescence intensity at each longitudinal position was summed circumferentially for each mouse to demonstrate the fluorescence distribution along the length of the colon. This was performed for all living mice at each time point. Each treatment group [QD-V = QD655-vascular endothelial growth factor receptor 2 (VEGFR2), QD-IC = QD655-isotype control] is indicated on the $y$ axis, with the small line in the middle of each group indicating where the azoxymethane (AOM) treated mice stop and the control mice start. A value of 0.5 is set as the maximum display value, causing all values $>0.5$ to be saturated.

While evidence suggests that contrast agent accumulation occurs due to the difficulty in flushing the small mouse colon, it is important to also note that some of the accumulation of signal in the nondiseased regions may be attributed to actual changes in VEGFR2 expression throughout AOM-treated colons. AOM-treated undiseased areas are not normal compared to saline-treated control colons. They have been shown to have abnormal crypt structure⁵⁹ and have an increased expression of VEGFR2 at a level between that of adenoma and control tissue (Fig. 6). There are also some changes in the morphology of the undiseased tissue in AOM-treated colon and control colons (e.g., thicker mucosa), which may contribute to the increased accumulation of contrast agent in the undiseased regions. Also, it is for this reason that the control colons (no AOM injections) were used to calculate the specificity of the contrast agent, demonstrating the true performance in completely normal tissue. Because the average relative fluorescence signal from the QD655-VEGFR2 labeled undiseased regions in AOM-treated mice was about two times lower in the ex vivo study than the in vivo study, the elevation in VEGFR2 expression in AOM-treated colons was likely not the major contributor to the fluorescence signal found in undiseased tissue in this study.

Fig. 6

Immunohistochemistry for VEGFR2 was performed on all colons after time point 5 imaging. Adenomas from AOM mice (a) are different in both VEGFR2 expression and morphology than undiseased tissue from control mice (c). This difference is less notable for the undiseased tissue in the same AOM treated mice (b), potentially contributing to nonspecific accumulation of contrast agent in undiseased tissue. Scale bar: $100\mu \mathrm{m}$.

This study provides further evidence that QD655-VEGFR2 contrast agent can selectively target VEGFR2 expressing adenoma and that the OCT/LIF imaging system is capable of detecting this signal due to the narrow emission spectrum of the quantum dots allowing the ease of separation from competing signals. In future studies, more thorough flushing of the unbound contrast agent is expected to result in a reduction in nonspecific signal as well as standard deviation of measured fluorescence intensity values, leading to increased significance between contrast agent/treatment groups and improved classification of diseased and undiseased regions. Use of a miniature balloon catheter that blocked the proximal colon and the constrained contrast agent and flushing solutions to the distal colon only could greatly aid complete removal of unbound contrast agent from the colon.

4.

Conclusions

Through this study, and the previously published ex vivo study, the ability of the QD655-VEGFR2 contrast agent to target VEGFR2 expressing adenoma in an AOM-treated mouse colon has been demonstrated. In this in vivo imaging study, the average intensity of the diseased regions of colons labeled with QD655-VEGFR2 was 3.2 times higher than the intensity from surrounding undiseased tissue in the same colons and 4.8 times higher than the intensity from the diseased regions of colons labeled with QD655-IC. This is similar to the results obtained in the ex vivo study. The main challenge of moving from ex vivo imaging to in vivo imaging in this application is the nonspecific accumulation of the contrast agent in the colon, providing a variable background that increased standard deviations and precluded statistical significance in the fluorescence intensity data. It is notable that while the small diameter of the mouse colon provides a great challenge in flushing of the colon, the larger diameter of the human colon would mitigate this concern. While improvements in flushing of the unbound contrast agent will need to be investigated, this study demonstrates the ability of the OCT/LIF system to detect the contrast agent in vivo, providing a minimally invasive in vivo imaging technique for visualizing molecular expression of colon adenoma.