EGFR Overexpressed in Colonic Neoplasia Can be Detected on Wide-Field Endoscopic Imaging

Colorectal cancer (CRC) is one of the most common causes of cancer-related mortality worldwide. Approximately 1,361,000 new cases were diagnosed globally in 2012, resulting in ~694,000 annual deaths.1 These numbers are expected to nearly double over the next 20 years as obesity grows at epidemic levels and more developing countries are adopting a Western diet.2 Greater emphasis on early detection of premalignant lesions is needed.3 Imaging with colonoscopy is widely accepted by patients and referring physicians, and it is the preferred method for screening of CRC.4 Currently, white-light illumination is used to detect adenomas based on structural changes. Unfortunately, a significant miss rate of >25% has been found on back-to-back exams for grossly visible adenomas.5, 6, 7 Moreover, premalignant lesions (dysplasia) that are flat can also give rise to cancer,8 and they have been found with a prevalence as high as 36%.9 Flat lesions may be more biologically aggressive than polyps,10 and five times more likely to harbor either in situ or submucosal adenocarcinoma in some patient populations.11 Sessile serrated adenomas (SSAs) can also be flat or slightly raised in appearance, and they can also progress to cancer.12 Thus, imaging methods with improved contrast and sensitivity to molecular rather than morphological properties may improve early detection and prevention of CRC. Epidermal growth factor receptor (EGFR) is a transmembrane tyrosine kinase that stimulates normal epithelial cell growth and differentiation.13 Ligand binding to the extracellular domains (ECD) 1 and 3 of EGFR results in receptor dimerization and autophosphorylation.14 This cell surface receptor has an important role in the development of a number of epithelial-derived cancers,15 and it is an important target for CRC therapy.16, 17 Overexpression of EGFR has been reported in as high as 97% of colonic adenocarcionomas, and it is a validated biomarker for CRC.18, 19 Adenomas with high-grade dysplasia and villous features on histology have been shown to exhibit increased expression of EGFR on immunohistochemistry.20 Furthermore, EGFR gene copy number has been found to increase with histological progression of disease.21, 22 In animals, EGFR signaling was required to form adenomas in azoxymethane-induced mouse models of CRC,23 and it was shown to promote flat lesions in aberrant crypt foci in rat colon.24 These findings support further development of EGFR as a promising imaging target for early detection of flat colonic lesions. Fluorescence provides images with high contrast to visualize molecular expression of neoplastic lesions in real time. Antibodies,25 enzyme-activated probes,26 and lectins27 are being developed as specific imaging agents to target premalignant lesions on endoscopy. We have recently shown that peptides are promising for use as clinical diagnostic imaging agents.28, 29 Peptides have low molecular weight and can achieve high specificity with binding affinities on the nanomolar scale. This probe platform has flexibility to be labeled with a broad range of fluorophores,30 and it is inexpensive to produce in large quantities. These features are well-suited to provide effective surveillance of large patient populations in procedure units that perform endoscopic procedures in high volume. We hypothesize that a peptide specific for EGFR can be developed with high specificity and rapid binding for detection of premalignant colonic lesions with topical administration. We measured the apparent association time constant for peptide binding to HT29 cells as an assessment of time scale. HT29 cells were grown to ~80% confluence in 10- cm dishes, and detached with PBS-based cell dissociation buffer (Invitrogen). Cells (~105) were incubated with 5 μm QRH*-Cy5.5 at 4 °C for various intervals ranging from 0 to 20 min. The cells were centrifuged, washed with cold PBS, and fixed with 4% PFA. Flow cytometric analysis was performed as described above, and the median fluorescence intensity (y) was taken as a ratio with that of HT29 cells without the addition of peptide at different time points (t) using the Flowjo software (LLC, Ashland, OR). The rate constant k was calculated by fitting the data to a first-order kinetics model, y(t)=Imax[1-exp(−kt)], where Imax=maximum value,35 using the Prism 5.0 software (GraphPad, La Jolla, CA). Imaging was performed using a small-animal endoscope (Karl Storz Veterinary Endoscopy, Goleta, CA).37 A xenon light source provides white-light illumination via a fluid light cable. A diode-pumped solid-state laser (TechnicaLaser, Orlando, FL) provides excitation at λex=671 nm. The laser beam is expanded to a diameter of 3 mm to fill the aperture of the light cable. The laser power at the tip of the endoscope is <2 mW. The white-light images reflect off a dichroic mirror (Semrock, Rochester, NY, #FF685-Di02–25 × 36, λc=685 nm), and they are focused by an achromatic doublet (Edmund Optics, Barrington, NJ, #32–323). They pass through a neutral density filter (optical density 1), and they are detected by a color camera (Point Gray Research, Richmond, British Columbia, Canada, #GX-FW-28S5C-C). The fluorescence images pass through a dichroic and band-pass filter (Semrock, #FF01–716/40–25, λC=716 nm, Δλ=40 nm), and are focused on a monochrome camera (Point Gray Research, #GX-FW-28S5M-C). All videos (1932 × 1452 resolution) are recorded at 15 frames per sec via a firewire connection. White-light and fluorescence videos were exported in avi format with 24 (RGB) and 8 (grayscale) bit digital resolution for white-light and fluorescence images, respectively. Streams that showed minimum motion artifact and absence of debris (stool, mucus) were selected for quantification. Individual frames were exported using the custom Matlab software. After completion of imaging, the animals were killed. The colon was excised and divided longitudinally. The gross specimen with the polyp and flat lesions from Figure 4 are shown, Figure 5a. Fluorescence imaging was used to guide a perpendicular section of the mucosal surface along the horizontal and vertical red lines. The flat lesions show mucosa with nonpolypoid morphology on histology (H&E), Figure 5b. Foci of dysplasia (red boxes) can be seen separated by regions of normal. The average fluorescence intensity from three regions of interest with dimensions of 25 × 25 pixels were picked at random from areas of “high” intensity and adjacent areas of “low” intensity. The target-to-background ratios were determined by taking ratios of the means of these results. We measured significantly greater fluorescence intensity from both polyps (n=15) and flat lesions (n=15) compared with that from the adjacent normal mucosa, target-to-background ratio 4.0±1.7 and 2.7±0.7, respectively. The difference in results between QRH*-Cy5.5 and PEH*-Cy5.5 was significant. A high-magnification view of histology from the flat regions (red boxes) show features of low-grade dysplasia (arrows), including collections of irregular crypts lined by epithelium with crowded, elongated, and hyperchromatic nuclei, Figure 5d–f. Histology of the polyp also shows identical features of dysplasia, Figure 5g. We then evaluated cell surface expression of EGFR in human specimens of flat adenomas excised from the proximal colon to assess future clinical relevance. These lesions are known to have unique and more aggressive biology.10 On immunohistochemistry with anti-EGFR, Supplementary Figure S8A,B shows no staining from representative sections of normal and hyperplastic polyps. Supplementary Figure S8C,D shows intense cell surface staining (arrows) for SSA and adenoma (dysplasia). Supplementary Figure S8E shows that EGFR is overexpressed in proximal lesions by greater than four-fold for dysplasia (n=26) and greater than three-fold for SSA (n=6) compared with normal (n=13) and hyperplastic polyps (n=12). Supplementary Figure S8F shows that EGFR can be used to distinguish premalignant (dysplasia and SSA) from benign (hyperplastic polyps and normal) colon in the proximal colon with 94% sensitivity and 92% specificity. We have identified the QRHKPRE peptide that binds to domain 2 of EGFR. This location has a β-hairpin that forms dimers rather than stimulate mitogenic activity, which occurs when EGF docks between domains 1 and 3.14 We confirmed lack of cytoplasmic signaling with peptide binding in HT29 cell lines, Supplementary Figure S4. The ECD of EGFR for human and mouse has 97.5% homology, and it is 100% conserved in domain 2.42 We used a structural model to optimize the sequence to achieve high binding affinity of kd=50 nm. In addition, binding occurs rapidly <2.5 min (k=0.406/min) with topical administration. This time scale is compatible with clinical use during colonoscopy. We demonstrated in vivo use of this peptide to detect flat and polypoid colonic adenomas that were diagnosed as low-grade dysplasia on pathology in a spontaneous mouse model of CRC. The flat lesions were not seen on white-light endoscopy. Furthermore, this peptide was found to bind human colonic dysplasia with 90% sensitivity and 93% specificity. These results were confirmed on immunohistochemistry using a known antibody, and they support the development of EGFR as an imaging target for early detection of premalignant colonic lesions that may otherwise go undetected on conventional white-light colonoscopy. With topical administration, peptides can be delivered in high concentrations to the mucosa at a risk of harboring disease to maximize binding interactions and to achieve high image contrast with little risk for toxicity. This approach results in a rapid binding with minimal background, and it avoids undesired biodistribution of the exogenous agent to other tissues, such as what occurs with the intravenous administration. Because of their small size, peptides have reduced immunogenicity and lower large-scale production costs. Peptides specific for EGFR have been previously developed for use as therapeutic agents with systemic administration for metastatic disease.43, 44, 45 For in vivo imaging, we used a mouse model that spontaneously develops colonic adenomas that may have either flat or polypoid architecture. We performed repetitive imaging using a near-infrared fluorescence endoscope to localize the premalignant lesions. Cy5.5 was used because this fluorophore emits in a spectral regime that is less sensitive to hemoglobin absorption and tissue scattering, minimizes background from tissue autofluorescence, and provides the maximum light penetration depth.40 We confirmed expression of EGFR in dysplastic mouse crypts on immunohistochemistry using two validated antibodies. Imaging of EGFR has been performed previously in a mouse orthotopic xenograft model of CRC. Human recombinant EGF was labeled with IRDye 800 CW (N-hydroxysuccinimide ester), and it was shown to bind to a mouse xenograft tumor that overexpressesed EGFR on whole-body fluorescence imaging. Peak signal was reached 2 days after injection.46 In vivo imaging has also been performed with a handheld confocal endomicroscope in the cecum of a xenograft mouse model using an FITC-labeled anti-EGFR antibody by exposing the tumor with an abdominal incision.47 For future clinical imaging, a wide-field endoscope that is sensitive to fluorescence can be used, and it may be able to distinguish sporadic and SSAs from hyperplastic polyps based on the EGFR expression level. We expect to find a much higher EGFR expression level in human adenomas, as shown in Figure 6. We have previously demonstrated a peptide VRPMPLQ that was identified using human biopsy specimens for selection with phage display. This peptide was labeled with FITC and used to detect human dysplastic colorectal polyps in vivo with confocal endomicroscopy.28 Because this peptide was selected empirically, the target is unknown and its clinical use may not be widely generalizable. By comparison, EGFR is a known target that is overexpressed by many cancers of epithelial origin, including lung,48 breast,49 pancreas,50 head and neck,51 and esophagus.52 This peptide is promising for in vivo use as an imaging agent to target premalignant lesions in the proximal colon that are flat in appearance and go undetected on conventional white-light colonoscopy that may otherwise lead to preventable cancers, and may also have broad use for early cancer detection in other imaging applications. Guarantor of the article: Thomas D. Wang, MD, PhD. Specific author contributions: JZ, BPJ, ERF, and TDW designed the research; JZ, BPJ, AP, XD, and TDW performed the research; JZ, BPJ, XD, and TDW contributed new reagents or analytic tools; JZ, BPJ, LZ, RK, SRO, and TDW analyzed the data; JZ, BPJ, and TDW wrote the manuscript. JZ, BPJ, and TDW are co-inventors on a patent disclosure submitted to the University of Michigan on the peptide presented in this manuscript. Financial support: None. This study was funded in part by National Institutes of Health U54 CA163059, R01 CA142750, U54 CA13642, and P50 CA93990 (TDW). Supplementary Information accompanies this paper on the Clinical and Translational Gastroenterology website