Introduction
Cancer response varies among individuals treated with molecular therapy targeted to the vascular endothelial growth factor (VEGF) receptor1, 2, 3, 4. As we expand our armamentarium of molecular targeted therapy for cancer, we can tailor treatment to individuals by measuring their cancer response to the therapeutic regimen. However, assessment of cancer response is typically delayed for months after initiation of therapy, thereby impeding the modification of the treatment regimen. The challenge in developing systems to detect cancer response to treatment is that any one of the multiple sites of metastasis can develop resistance to treatment. Thus, all sites must be simultaneously assessed during treatment, thereby necessitating noninvasive imaging systems. Current efforts to personalize effective therapy for people with cancer use serial imaging, tumor biopsies or both to assess response. These methods of monitoring cancer response are inefficient because tumor volume changes typically occur after the individuals are on therapy for prolonged time intervals. Biopsies are not easily performed in people with brain tumors, lung cancer or pancreatic cancer. Furthermore, biopsies can result in sampling error, resulting in an inaccurate assessment of the pharmacodynamic response to therapy.
Monitoring cancer response to treatment promises to improve our ability to tailor therapy specifically to an individual and rapidly evaluate new pharmaceuticals. In this regard, molecular imaging techniques using radiolabeled annexin V have provided some substantiation of the use of noninvasive imaging to assess cancer response to therapy5. Clinical evaluation of radiolabeled annexin V binding to tumors correlates with the number of tumor apoptotic cells derived from histological analysis6. Because of the slow blood clearance of annexin, however, only planar images of low image quality are obtained after radiotracer injection7. Nevertheless, radiolabeled peptides rapidly clear from the circulation and are therefore a useful alternative means for imaging specific molecular targets8, 9. Here we propose the use of recombinant peptides selected from phage-displayed peptide libraries to provide a means to noninvasively assess tumor vascular response to several VEGF receptor tyrosine kinase inhibitors (TKIs) combined with cytotoxic therapy. Screening of phage-displayed peptide libraries has been established as a way to discover peptide ligands that bind to tumor vasculature, cancer cells or specific molecular targets10, 11, 12. Phage display technology allows for the insertion of random DNA sequences into the bacteriophage genome, which, in turn, encodes the phage capsid proteins, thereby providing peptides expressed on the phage surface that bind to cell surface molecules13, 14.
Results
Selection of phage-displayed peptides that bind to treated tumors
The goal of this study was to determine whether recombinant peptides can be used to differentiate responding from nonresponding cancers immediately after initiation of treatment. In vivo selection of phage-displayed peptide libraries simultaneously provides positive and subtractive screens, because organs are spatially separated from tumors. Phages that bound to the vasculature of nonresponding tumors and normal tissues were discarded, whereas phages bound to responding tumors were enriched through serial rounds of biopanning. Lewis lung carcinoma tumors treated with combined radiation and 5-[5-Fluoro-2-oxo-1,2-dihydroindol-(3Z)-ylidenemethyl]-2,4-dimethyl-1H-pyrrole-3-carboxylic acid (2-diethylamino-ethyl)amide (SU11248), a VEGF receptor tyrosine kinase inhibitor, showed regression, whereas tumors treated with SU11248 alone or radiation alone showed attenuation of tumor growth at 2 weeks after therapy (Fig. 1a). TUNEL staining of tumor sections (Fig. 1b,c) showed a substantial increase in the number of apoptotic endothelial cells in the microvascular endothelium of tumors treated with both radiation and SU11248. In vivo biopanning was conducted with a T7 phage–based random peptide library. After six rounds of biopanning, the relative ratio of phages recovered from the treated tumors increased 90-fold (P < 0.01) when compared to those from other organs, suggesting that peptide sequences that bound to responding tumors had been successfully enriched in the biopanning process.
Figure 1: HVGGSSV phage binding to treated tumors.
Lewis lung carcinoma tumors were implanted into both hind limbs of nude mice. The mice were treated systemically with SU11248. The tumor in the left hind limb was irradiated with 3 Gy, whereas the right hind limb tumor served as an internal treatment control with drug alone. (a) Shown is the bar graph of fold increase in tumor volume at 2 weeks after treatment with vehicle alone (control), 3 Gy alone, SU11248 alone (SU) or combination of SU11248 and 3 Gy (SU + 3 Gy). Tumor volumes were measured by use of calipers. (b,c) Shown are microscopic photographs of TUNEL-stained tumor sections at 24 h after treatment with vehicle alone (b) or combination of SU11248 and 3 Gy (c). Apoptotic endothelial cells (brown) are indicated by arrows. Sections were counterstained with hematoxylin. Scale bars, 50 
m. (d,e) Cy7-labeled HVGGSSV phage was injected into the circulation through a jugular catheter. Shown are NIR images obtained 48 h after C7-labeled HVGGSSV phage injection into SU11248-treated mice bearing Lewis lung carcinoma tumors (d) or H460 tumors (e). The arrows indicate tumors treated with 3 Gy, whereas tumors in the opposite hind limb received 0 Gy (drug alone). (f) Radiance of tumors was correlated with tumor growth delay. Shown is the correlation between radiance from peptide bound within treated tumors and subsequent delay in the growth of tumor shown in a. R2 = 0.7938, *P < 0.05.
Fifty phage plaques were amplified with PCR, and the deduced peptide sequences with their relative abundance are listed in Table 1. We found that three phage peptides were dominantly enriched in the treated tumors. In order to prevent any bias from tumor models and biopanning protocols, we applied the same protocols with the original phage-displayed peptide library on another tumor model, murine GL261 glioblastoma xenografts, in which combination of SU11248 and radiation showed enhanced tumor growth control as compared to tumors treated with SU11248 alone15. The peptides displayed by the recovered phages after six rounds of biopanning in the GL261 model are listed in Table 1. The two independent screenings of a phage-displayed peptide library on Lewis lung carcinoma and GL261 tumors resulted in the isolation of a panel of peptides that share the common motif GSXV, where 'X' is a variable amino acid (Table 1). This amino acid sequence was not found in 100 clones that were randomly picked and sequenced from the unselected library (data not shown).
Although the HVGGSSV peptide sequence was the predominant phage-encoded peptide isolated from both Lewis lung carcinoma and GL261 tumors, the possibility remained that the dominance was the result of overgrowth or preferred enrichment of the phage over others16, 17. To test this possibility, the cloned phages from Table 1 were simultaneously injected and predilection for tumor binding was compared in the same mouse. Phages were equally mixed together for injection into the venous circulation of tumor-bearing mice. Mice were treated with either radiation alone, SU11248 alone, radiation and SU11248 or no treatment (control). Fifty phage plaques recovered from tumors in each group were randomly sequenced for identification. The specificity of peptide-bearing phages was evaluated by the frequency of each phage recovered from tumors treated with different regimens. The HVGGSSV-displaying phage (HVGGSSV phage) had the highest specificity among the phage clones that bound tumors treated with SU11248 alone or with the combination of radiation and SU11248 (P < 0.01). The HVGGSSV phage was not found in the phages recovered from the untreated tumors. Therefore, we selected the HVGGSSV peptide for further characterization as an indicator of responsiveness to molecular targeted therapy.
HVGGSSV phage binding to treated tumors
To evaluate differential binding of HVGGSSV to responding cancers, two Lewis lung carcinoma tumors were implanted in the hind limbs of C57BL/6 mice. The mice were treated systemically with SU11248. One tumor was treated with localized cytotoxic therapy (3 Gy), and the other received drug alone. To visualize the biodistribution of phage in mice, Cy7-labeled HVGGSSV phage was injected into the circulation through a jugular catheter, and Cy7-labeled phage was imaged with near-infrared imaging (NIR). HVGGSSV phage binding within treated tumors was detected within 24 h and persisted beyond 48 h. At 48 h after injection, the phage was cleared from the circulation, and the HVGGSSV phage was predominantly localized in the Lewis lung carcinoma tumors treated with combined SU11248 and radiation (Fig. 1d). Similarly, H460 tumors implanted into both hind limbs showed increased radiance from HVGGSSV phage within tumors treated with SU11248 followed by irradiation as compared to tumors treated with drug alone (Fig. 1e). The relative radiance of the HVGGSSV phage in the treated tumor (normalized to that of the whole body) was significantly increased (P < 0.05, n = 5, Student's t-test) over untreated tumors or the control phage in the treated tumors (Fig. 1f). To determine whether the level of radiance from peptide binding within tumors correlated with tumor growth delay, tumor volume and radiance were measured daily for 14 d. SU11248 alone and radiation alone produced minimal attenuation of tumor growth as compared to the combined treatment of radiation and SU11248 (Fig. 1e). The radiance of peptide binding within tumors correlated with the attenuation of tumor growth (R2 = 0.7938, P < 0.05).
To determine whether the HVGGSSV peptide serves as the functional motif during binding to treated tumors, the HVGGSSV peptide (GGGNHVGGSSV) and a control peptide with a scrambled sequence (GGGSGVSGHVN) were synthesized and labeled with Cy7 at the N-terminal amine group. Three glycine residues were placed at the amino terminus to separate the HVGGSSV peptide from Cy7. Labeled peptide was injected intravenously into tumor-bearing mice. NIR imaging was used to monitor the biodistribution of both peptides in tumor-bearing mice treated with SU11248 and radiation. NIR images revealed that the HVGGSSV peptide, but not the control scrambled sequence peptide, binds to Lewis lung carcinoma tumors that were treated with both radiation and SU11248 (Supplementary Fig. 1 online). We observed that HVGGSSV peptide also maintained binding to responding tumors when biotin was conjugated to the amino terminus (Supplementary Fig. 1), and thus we linked biotinylated HVGGSSV to Cy7-labeled streptavidin. Lewis lung carcinoma tumor–bearing mice were treated with SU11248 and radiation and labeled peptide was injected intravenously into tumor-bearing mice. Cy7-streptavidin–biotinylated HVGGSSV complex was detected within treated tumors (Supplementary Fig. 1). The tumor binding pattern of the HVGGSSV peptide alone (without the phage scaffold) indicates that HVGGSSV is the functional motif for target binding.
HVGGSSV differentiates responding from nonresponding tumors
To determine whether HVGGSSV peptide binding differentiates between responding and nonresponding tumors in the same subject, SU11248 and radiation were used to treat mice bearing responsive and nonresponsive tumors (Lewis lung carcinoma and B16F0, respectively). Both tumors were treated with SU11248 and 3 Gy for 5 consecutive d. B16F0 tumors continued to grow, whereas Lewis lung carcinoma tumors showed no growth (Fig. 2a). After the first treatment, peptide was injected into the circulation of tumor-bearing mice, and NIR imaging was used to study the biodistribution of Cy7-HVGGSSV peptide. Peptide was detected in the regressing Lewis lung carcinoma tumors (right hind limb) but not in the nonresponsive B16F0 tumors (left hind limb; Fig. 2b). Both tumors were removed and sectioned for imaging by use of the Xenogen IVIS imaging system. A marked reduction in NIR emission was detected in the nonresponding B16F0 tumors, as compared to a significant increase in radiance from the responsive Lewis lung carcinoma tumor (Fig. 2c, P < 0.05, n = 5). B16F0 and BxPC3 human pancreatic cancer tumors showed no tumor growth delay while they were treated daily with 40 mg/kg SU11248 (Fig. 2d). BxPC3 tumors had increased in volume by 45% by day 11 of treatment with SU11248, but had increased by 42% with vehicle control. We therefore studied radiance from labeled HVGGSSV in SU11248-treated BxPC3 tumors to confirm the specificity of peptide binding only to responding tumors. Before treatment, HVGGSSV peptide showed minimal radiance in BxPC3 tumors (Fig. 2e). Treated mice showed no increase in peptide radiance within BxPC3 tumors. Quantification of tumor radiance showed no significant change in treated tumors as compared to untreated tumors (Fig. 2e). Lewis lung carcinoma tumor volume was reduced by 60% after SU11248 treatment compared to untreated tumors (Fig. 2e). In contrast, BxPC3 tumors showed no volume reduction in tumors after SU11248 treatment compared to untreated tumors (Fig. 2e). Radiance from HVGGSSV peptide increased in the SU11248-treated Lewis lung carcinoma tumors, but not in the treated BxPC3 tumors (P < 0.001; Fig. 2e).
Figure 2: HVGGSSV peptide differentiates between responding and nonresponding tumors.
Lewis lung carcinoma tumors were implanted into the right hind limb and B16F0 tumors were implanted into the left hind limb of the same mouse. Both the Lewis lung carcinoma and B16F0 tumors were treated with SU11248 and 3 Gy for 5 consecutive d. (a) Tumor volumes were measured at day 0 and day 5. Four hours after the first treatment, Cy7-labeled HVGGSSV peptide was infused and NIR imaging was used to study peptide binding to treated tumors. (b) Shown are NIR images of mice obtained at indicated times after the first treatment. (c) At 74 h after treatment, tumors were removed, sectioned and imaged by NIR using the Xenogen IVIS imaging system. Shown are representative NIR images of sectioned B16F0 and Lewis lung carcinoma tumors. (d) BxPC3 pancreatic cancer cells were implanted into both hind limbs. Mice were then treated with SU11248 at 40 mg/kg. Cy7-labeled HVGGSSV peptide was administered intravenously 4 h after treatment. Shown are NIR images obtained at 72 h after injection. (e) Tumor volumes at day 11 of treatment were compared to volumes at day 0 for both Lewis lung carcinoma tumors and BxPC3 tumors. Shown is the percentage change in tumor volume in SU11248-treated tumors compared to untreated controls. Radiance (photons/cm2/s) was measured in both Lewis lung carcinoma and BxPC3 tumors. Shown are the mean and s.e.m. of the percentage increase in radiance from tumors treated with SU11248 compared to untreated control tumors. *P < 0.001.
Full size image (75 KB)To determine the histologic binding pattern of HVGGSSV peptide within tumors, we stained and photographed microscopic sections of treated tumors. Immunohistochemistry of tumors treated with SU11248 and radiation followed by injection of the control scrambled peptide is shown in Supplementary Fig. 2a online. We found no binding of the control peptide within the tumor microvasculature. Similarly, untreated control tumors showed no peptide binding within the tumor microvasculature (Supplementary Fig. 2b). Tumors treated with either SU11248 alone or SU11248 with radiation showed peptide binding to the microvascular endothelium and a few adjacent Lewis lung carcinoma tumor cells. This histologic pattern of peptide binding to tumor endothelium was similar in tumors treated with drug alone (Supplementary Fig. 2c) as compared to those treated with drug and radiation (Supplementary Fig. 2d). To determine the differential binding of peptide to responding tumors as compared to nonresponding tumors, we studied the microscopic sections of the Lewis lung carcinoma and the B16F0 tumors, respectively. The immunofluorescence microscopy of the Lewis lung carcinoma tumors treated with SU11248 and radiation (Supplementary Fig. 2e) shows HVGGSSV peptide binding within the tumor microvasculature. In contrast, the nonresponding B16F0 tumor showed no peptide binding within the tumor microvasculature (Supplementary Fig. 2f).
The use of HVGGSSV peptide to assess tumor response to other TKIs was studied in tumor-bearing mice treated with PTK787 (Fig. 3a), AEE788 (Fig. 3b) or SU5416 (Fig. 3c). SU5416 is a VEGF receptor–specific inhibitor, and PTK787 inhibits all known VEGF receptors18, 19, 20, 21, whereas AEE788 inhibits both the VEGF and the epidermal growth factor receptors22, 23. Each drug produced a >200% increase in HVGGSSV radiance within tumors (Fig. 3d). Cy7-labeled HVGGSSV peptide was injected 4 h after treatment with each of these TKIs, and NIR images were obtained daily thereafter. Labeled HVGGSSV bound responding tumors treated with all of the VEGFR TKIs that were studied. In contrast, the negative control scrambled peptide showed no binding within tumors after treatment with any of the TKIs (data not shown).
Figure 3: Peptide assessment of tumor response to all TKIs.
Mice bearing Lewis lung carcinoma tumors in the right hind limb were treated with 75 mg/kg PTK787 (a), or 60 mg/kg AEE788 (b). (c) Mice bearing both B16F0 (left hind limb) and Lewis lung carcinoma (right hind limb) tumors were treated with SU5416 (40 mg/kg). Mice were then imaged with control scrambled peptide (left mouse) or HVGGSSV peptide (right mouse). Cy7-labeled peptide was injected 4 h after treatment. Shown are NIR images obtained at 24 h after peptide injection. Arrows indicate the location of the tumors. (d) The bar graph shows the percentage 
s.e.m. increase in peptide radiance from tumors shown in a–c (*P < 0.05, comparing radiance in untreated to treated tumor).
Peptide binding detects response to therapy in all tumor models
To determine whether the HVGGSSV peptide can be used to distinguish treatment response within other tumor types, we studied orthotopic tumors within the lung, liver and brain. D54 human glioblastoma cells were injected into the cerebra of nude mice (Fig. 4a), and H460 lung cancer cells were injected by tail vein to form pulmonary tumor nodules (Fig. 4b). HT22 human colon carcinoma was injected into the spleens of nude mice, and liver metastases developed over the course of 2 weeks, whereas the injection site developed an intra-abdominal tumor (Fig. 4c). Tumor-bearing mice were treated with SU11248 followed by irradiation. To evaluate HVGGSSV peptide binding within these various orthotopic tumors, we studied Cy7-labeled peptide binding to tumors before and after treatment. HVGGSSV peptide binding was not detected in untreated tumors (data not shown). Tumors treated with SU11248 and radiation showed NIR imaging of peptide binding within 24 h of injection (Fig. 4a–c). PC3 human prostate cancer cells and MDA-MB-231 breast cancer cells injected into the hind limbs of nude mice developed into tumors that were treated with SU11248 followed by irradiation. Cy7-HVGGSSV peptide was injected 4 h after treatment. The NIR images show Cy7-labeled peptide binding within treated tumors but no binding to untreated control tumors (Fig. 4d,e). All tumor types showed a significant increase in radiance compared to untreated controls (P < 0.05, n = 3). The sensitivity of detection for responsive tumors was 40 out of 42 responding mouse tumors by HVGGSSV radiance. In comparison, 0 of 26 nonresponding or untreated tumors showed increased peptide radiance within tumors (P < 0.001).
Figure 4: Peptide binding detects response to therapy in all tumor models.
(a–e) Tumor development was induced by the following methods: D54 human glioblastoma cells were injected into the cerebra (a), H460 lung cancer cells were injected through the tail vein (to develop pulmonary metastases; b), HT22 human colon cancer cells were injected into the spleen (to develop liver metastases; c), and PC3 prostate cancer cells (d) and MDA-MB-231 breast cancer cells (e) were injected subcutaneously into the hind limbs of nude mice. The tumor-bearing mice were treated with SU11248 for 1 h before irradiation. Cy7-labeled HVGGSSV peptide was injected intravenously 4 h after treatment. Shown are NIR images obtained 48 h after peptide injection. (f) Shown is the linear correlation of peptide binding (relative radiance) to tumor response (tumor volume fold change) in human breast cancer MDA-MB-231 murine hind limb tumors. R2 = 0.868; P < 0.01. (g) H460 and HT29 tumor–bearing mice were treated with SU11248 doses of 0, 4 or 40 mg/kg daily for 14 d. Tumor volumes were measured by use of calipers. The bar graph (top) shows the fold increase in tumor volumes for H460 or HT29 tumors. These same H460 or HT29 tumors were imaged by NIR, and fold increase in radiance within tumors was compared. The bar graph (bottom) shows the fold increase in radiance in treated tumors as compared to untreated tumors. *P < 0.05.
Full size image (80 KB)To determine whether radiance from peptide binding in tumors correlated with tumor response, we studied the dose-dependent response to SU11248 in MDA-MB-231. MDA-MB-231 cells implanted into the hind limbs of mice were treated for 5 consecutive d. The combination of SU11248 with X-ray irradiation resulted in the most efficient tumor growth control, and tumors responded to high-dose (40 mg/kg), but not low-dose (4 mg/kg), treatment with SU11248 alone (Fig. 4f). After the first treatment, Cy7-labeled peptide was injected intravenously into tumor-bearing mice. NIR images were captured and the radiance from peptide binding in tumors was measured. A linear correlation (R2 = 0.8684, n = 5 for each group) was found for peptide binding (relative radiance) and tumor response to treatment (fold change in tumor volume) (Fig. 4g).
Although clinical studies evaluating the efficacy of SU11248 have shown that 1–4 mg/kg is tolerable and effective in people with cancer, preclinical assessment of the drug in mouse models used doses of 40–80 mg/kg15, 24. To determine whether TKIs induce a dose-dependent increase in peptide radiance, H460 and HT29 tumor–bearing mice were treated with 0, 4 or 40 mg/kg of SU11248 without irradiation. Tumor volumes and radiance were measured daily for 14 d. H460 tumor growth did not respond to 4 mg/kg SU11248, and there was no increase in peptide radiance from these tumors (Fig. 4g). Likewise, HT29 tumors showed no response to 4 mg/kg SU11248 and no increase in peptide radiance from tumors (Fig. 4g). In comparison, high-dose TKI therapy (40 mg/kg SU11248) did achieve tumor regression and a corresponding increase in peptide radiance from tumors (Fig. 4g). These data indicate that peptide radiance can be used to titrate the dosage of molecular targeted therapy to a level needed to achieve response within cancer.
Discussion
Phage-display technology promises to aid the discovery of useful imaging peptides that facilitate molecular imaging of cancer pharmacodynamic response to TKI therapy. To compare, in recent years annexin V has been proposed as an imaging probe to detect apoptotic cells. Radiolabeled annexin V has allowed for the noninvasive monitoring of cell death by use of single-photon emission computerized tomography (SPECT)5, 6, 7. Clinical evaluation suggested that tumor uptake of technetium-99m–labeled annexin V correlated with the number of tumor apoptotic cells derived from histological analysis6. Annexin V, however, has limitations that include the complex GMP production of this large protein and its slow renal clearance, which limits its use as an imaging agent, as it results in low-quality planar images 24 h after injection6, 7. Small peptides such as HVGGSSV are a more cost-effective alternative that show rapid clearance, resulting in efficient assessment of tumor response to therapy. Table 1 shows the peptides we recovered from tumors treated with SU11248 and radiation. Here we show that this technology can be used to select phage-displayed peptides that distinguish between responding and nonresponding cancers. These peptides can serve as 'molecular beacons' that bind to susceptible cancers, allowing for a more rapid assessment of cancer responsiveness to therapy. Rapid noninvasive assessment of the pharmacodynamic response of cancer promises to minimize the duration of ineffective treatment regimens in people with cancer and to speed drug development.
The sensitivity and specificity of molecular imaging agents are most crucial in evaluating small peptides as molecular markers for cancer response. In the present study, sensitivity was defined as the ability of the peptide to detect all tumors that respond to any VEGF receptor TKI. Therefore, we studied several tumor models in a variety of organs and also studied several TKIs and found that the HVGGSSV peptide detects response in all of these experimental models. Specificity was defined as the ability of the peptide to discern responding from nonresponding cancers. Although we have not found peptide binding to other pathological tissues, such as that resulting from inflammation (Supplementary Fig. 3), this nonspecific binding would be detected before therapy and again after beginning therapy. The increase in peptide binding during the response to therapy would distinguish responsive from nonresponsive cancers. We have shown a direct correlation between the amount of increased peptide binding and the responsiveness of the cancer.
We speculate that HVGGSSV peptide binds to a protein that is unveiled during endothelial response to therapy. The amino acid sequence GSXV is homologous to the carboxyl terminus of ligands that bind to the PDZ domain of several membrane proteins. Preliminary experiments show that one such protein, Tax interacting protein, binds to HVGGSSV (data not shown). The abundance of this protein also increases in membrane preparations of cells treated with TKIs and radiation. HVGGSSV shows better differentiation of responding and nonresponding cancers as compared to the other GSXV peptides listed in Table 1. We therefore plan clinical evaluation of the radiolabeled HVGGSSV peptide.
Methods
Tumor models.
We purchased cell lines from American Type Culture Collection (ATCC) and maintained them in DMEM medium supplemented with 10% FCS and 1% penicillin-streptomycin as recommended by ATCC. Tumor cell lines include Lewis lung carcinoma (ATCC CRL-1435), B16F0 melanoma (ATCC CRL-6322), human glioblastoma D54, human lung carcinoma H460 (ATCC HTB-177), human colon cancer HT22, human prostate cancer PC-3 (ATCC CRL-1435) and breast cancer MDA-MB-231 (ATCC HTB-26). BxpC3 human pancreatic cancer cells were also obtained from ATCC. We segregated monolayer cells with 80% confluence and suspended them in phosphate-buffered saline (PBS). We developed heterotopic models by subcutaneously inoculating cell suspensions (5 
105 cells or adjusted for different cell types) into nude mice. We implanted the tumors in both hind limbs of mice and used them for experiments when the tumor size reached 0.5 cm in diameter. We developed orthotopic brain cancer models by intracranially injecting D54 cancer cells. Lung tumors developed after tail vein injection of H460 cells. Liver metastases developed after splenic injection of HT22 cells.
Tyrosine kinase inhibitors.
SU11248 was synthesized in the Vanderbilt Institute of Chemical Biology using the five-step method previously described24, 25. We purchased SU5416 from Calbiochem and PTK787 and AEE788 from Novartis. We dissolved PTK787, SU11248 and SU5416 in DMSO and AEE788 and PTK787 in PBS. We administered the TKIs by intraperitoneal injection at the following doses: PTK787, 75 mg/kg; AEE 788, 60 mg/kg; SU5416, 40 mg/kg; SU11248, 4 or 40 mg/kg. We delivered radiation with a 
-irradiator. All protocols in animal experiments were reviewed and approved by the Vanderbilt University Institutional Animal Care & Use Committee.
Biopanning phage-displayed libraries.
We conducted in vivo biopanning as described14 with a T7 phage–based random peptide library. The phage-displayed peptide library represents 1 108 independent clones of phages expressing random nonamer peptides that are displayed on T7 phages as fusion proteins with the amino terminus of 10A capsid protein. After the tumor-bearing mice were treated, we administered phage libraries by intracardiac injection at 4 h after irradiation. We partially purified amplified phages by polyethylene glycol precipitation and resuspended them in PBS for the next round of biopanning. After six rounds of biopanning, we isolated single plaques from soft agar and amplified gene fragments encoding peptides by PCR, following standard protocols14, 16. The PCR primers include an upstream primer (5'-AGCGGACCAGATTATCGCTA-3') and a downstream primer (5'-AACCCTCAAGACCCGTTTA-3'). We performed the sequencing reaction with one primer and collected the sequence data in an ABI 377 sequencer. We deduced the peptide sequences from the decoded DNA information. Detailed protocol is available in Nature Protocols26.
Near-infrared imaging.
We labeled the PEG-precipitated phages, synthetic HVGGSSV peptide (Genemed Synthesis) or streptavidin (Sigma) with amine-reactive Cy7 dye (Amersham) by following the manufacturer's instructions. We injected labeled phages or the complex of biotinylated peptide and streptavidin-Cy7 conjugate into the circulation by tail vein or jugular vein catheter in tumor-bearing mice that had been treated with irradiation, TKIs, or both. We took NIR images with the IVIS imaging system (Xenogen) at various time points after injection. Radiance (photons/s/cm2) was measured in the region of interest (ROI) by using the program provided by Xenogen. When we correlated peptide binding (radiance) to tumor response (tumor growth), we normalized radiance from peptide within tumors to that of the whole body.
Tumor growth study.
We implanted tumors into the hind limbs of mice and started treatment when tumor size reached 0.5 cm in diameter. Treatment groups included irradiation alone (3 Gy), SU11248 alone (40 mg/kg), combined treatment with irradiation and SU11248, and untreated control. We administered SU11248 by intraperitoneal injection. All treatments were given once a day for 5 consecutive d. We measured tumor size every other day by use of calipers, and we calculated fold increase in tumor volume (compared to the tumor size on the first day of treatment) to show tumor responsiveness to the treatment. We included six mice in each group.
Human umbilical vein endothelial cell coculture with tumor cells.
We established human umbilical vein endothelial cell and Lewis lung carcinoma cell coculture in a chamber system (Corning) for 3 d. We then treated cells with SU11248 and 3 Gy. Immediately after treatment, we incubated cells with the biotinylated HVGGSSV peptide or control scrambled VSVGHGS peptide. We used FITC-conjugated streptavidin to detect peptide binding on the cell surface. We used fluorescent microscopy to photograph both treated and untreated control cells.
Statistical analyses.
We analyzed group comparisons with Student's t-test. We developed linear correlations of peptide binding and tumor response to treatment by use of correlation coefficient of tumor growth and radiance datasets (CORREL).
Note: Supplementary information is available on the Nature Medicine website.
Author contributions
Z.H. and A.F. discovered and prioritized peptides. A.F., R.D. and H.O. performed all mouse imaging. H.W. and L.G. performed cell assays microscopy and histology. D.E.H. is the principle investigator and oversaw all studies.